ICDSMLA 2021: Proceedings of the 3rd International Conference on Data Science, Machine Learning and Applications 9811959358, 9789811959356

Table of contents :
Preface
Contents
Road Accident Detection and Indication System
1 Introduction
2 Existing Models
3 Proposed Model
4 The Design of Vehicle Crash Alert System
5 Conclusion
References
Attendance System Based on Face Recognition Using Haar Cascade and LBPH Algorithm
1 Introduction
2 Problem Definition
3 Proposed System
4 Dataset and Methodology
5 Results and Outcome
6 Conclusion
References
Different Thresholding Techniques in Image Processing : A Review
1 Introduction
2 Thresholding
2.1 Global Thresholding
2.2 Multi-thresholding
2.3 Adaptive Thresholding
3 Choosing a Thresholding Method
4 Conclusion
References
Dynamic Weighting Selection for Predictive Torque and Flux Control of Industrial Drives
1 Introduction
2 Predictive Torque and Flux Control
3 Proposed Dynamic Weighting Factor Selection
4 Simulation Results
5 Conclusion
References
Population Index and Analysis Based on Different Geographies; Using Distance Measurement, Social Distancing, and Deep Learning
1 Introduction
1.1 How is Social Distancing Efficient?
1.2 Social Distancing
2 Research Background and Relevant Work
2.1 Pedestrian Detection and Distance Measurement
2.2 Object Detection
2.3 Social Distancing for COVID-19
3 Object Detection and Distance Measurement
3.1 Object Detection
3.2 Distance Measurement Between Detected Objects
4 Proposed Methodology
4.1 Stage I: Detecting Violators
4.2 Stage II: Finding Busy Hours
5 Experiment and Result Analysis
5.1 YOLO Outputs
5.2 Database of Violations
5.3 Busy Hours
5.4 Discussion
6 Conclusion
7 Future Scope
References
On the Discriminability of Samples Using Binarized ReLU Activations
1 Introduction
1.1 Contributions
2 Metrics Used in the Experimental Setup
3 Experiments
3.1 Influence of Number of Layers
3.2 Influence of Sample Size and Number of Training Labels
4 Conclusion and Future Work
References
Supervised and Unsupervised Machine Learning Approaches—A Survey
1 Introduction
2 Different Kinds of Learning
2.1 Supervised Machine Learning
2.2 Unsupervised Machine Learning Algorithm
3 Conclusion
References
Skin Cancer Classification Using Deep Learning
1 Introduction
2 Literature Survey
3 Objectives
4 Existing System
5 Proposed Approach
6 Conclusion
References
Crop Yield Prediction Using Deep Learning
1 Introduction
2 Related Work
3 Proposed Idea
4 Data Flow Diagram
5 Training Specifications
6 Performance Evaluation
7 Experimental Results
8 Discussion and Conclusions
References
Real-Time Tweets Streaming and Comparison Using Naïve Bayes Classifier
1 Introduction
2 Problem Definition
3 Literature Survey
4 Proposed Work
5 Conclusion and Future Enhancement
References
Smart Shopping Trolley for Billing System
1 Introduction
2 Related Work
3 Objective
4 Problem Statement
5 Proposed Architecture Design
6 Implementation and Results
7 Conclusion and Future Work
References
A Survey on IoT Protocol in Real-Time Applications and Its Architectures
1 Introduction
2 Functional Overview of IOT
2.1 Sensing
2.2 Communication
2.3 Computation
2.4 Analysis
3 Standards
3.1 Application Protocols
3.2 Constrained Application Protocol (CoAP)
3.3 Message Queuing Telemetry Transport (MQTT)
3.4 Advanced Message Queuing Protocol (AMQP)
4 AMQP and MQTT Comparative Analysis
5 Conclusion
References
Safe Characteristic Signature Systems with Different Jurisdiction Using Blockchain in E-Health Records
1 Introduction
2 Existing System
3 Proposed System
4 System Design
5 Flow Chart Diagram
6 Class Diagram
7 Modules Description
7.1 Admin
7.2 Admin Module
7.3 User Module
7.4 Auditor
8 Conclusion
References
Web-Based Trash Segregation Using Deep Learning Algorithm
1 Introduction
2 Proposed System
3 Implementation
4 Analysis and Benefits of Proposed System
5 System Design and Flowcharts
6 Conclusion
7 Future Enhancement
References
Home Automation Using Face Recognition for Wireless Security
1 Introduction
2 Related Work
2.1 Bluetooth-Based IoT Home Automation System
2.2 Voice Recognition-Based IoT Home Automation
2.3 GSM-Based Home Automation System
3 Proposed Work
3.1 Hardware Components
3.2 Software Requirements
4 Implementation Details
5 Result
6 Conclusion
Bibliography
Hybrid-Network Intrusion Detection (H-NID) Model Using Machine Learning Techniques (MLTs)
1 Introduction
2 Related Work
3 Methodology and Proposed Framework
3.1 Classification Algorithms Used
4 Experiments and Results
5 Conclusions
References
Impact of Using Partial Gait Energy Images for Human Recognition by Gait Analysis
1 Introduction
2 Literature Review
3 Methodology
3.1 Dataset
3.2 Gait Energy Image (GEI)
3.3 Principal Component Analysis
3.4 Experimental Setup
4 Results
5 Conclusion
References
Several Routing Protocols, Features and Limitations for Wireless Mesh Network (WMN): A Review
1 Introduction
2 Literature Review
3 Limitations and Features of WMNs
3.1 Limitations of WMNs
3.2 Features of WMNs
4 Preliminaries on Routing Protocols for WMNs
4.1 Hop Counting Routing Protocols
4.2 Link Level Routing Protocols
4.3 End-to-End QoS Routing Protocols
5 Conclusion and Future Scope
References
A Deep Meta-model for Environmental Sound Recognition
1 Introduction
2 Related Work
3 Methodology
3.1 Base Models
3.2 The Proposed Deep Meta-model
4 Experimental Evaluation
4.1 Dataset Used
4.2 Performance Measures Used for Evaluation
4.3 Results and Discussion
5 Conclusion
References
Spatial Computing: Next Big Thing of Physical and Digital World
1 Introduction
2 Role of Spatial Computing in Various Fields
2.1 Industry
2.2 Manufacturing
2.3 Automotive Industry
3 Benefits of Spatial Computing
4 Difference Between Spatial Computing, Augmented Reality (AR) and Mixed Reality (MR)
5 Spatial Computing as a Digital Era
6 SAR Technology
7 Spatial Computing With IoT
8 Spatial Web Technology
9 Conclusion
References
Cloud Accessing Based on IOT Oriented WSNs for Optimal Water Conservation in Farming
1 Introduction
2 Literature Survey
3 Proposed System for Irrigation with IoT
4 Results and Discussions
5 Conclusion and Future Work
References
S-Extension Patch: A Simple and Efficient Way to Extend an Object Detection Model
1 Introduction
2 Related Works
3 Methodology
3.1 Similarity Threshold Check (Compatibility Check)
3.2 Compatible Classes Selection
3.3 Dual-Parallel Inference Technique
3.4 Technique with Trackers
3.5 Model Architectures
3.6 Training Environment
4 Experiments and Discussions
4.1 Dual-Parallel Inference Technique
4.2 Technique with Trackers
5 Future Works and Llimitations
6 Conclusion
References
Roles and Impact of ASHA Workers in Combating COVID-19: Case Study Bhubaneswar
1 Introduction
1.1 Fund Allocation to Public Healthcare and Human Development Index
1.2 Health Status and Global Hunger Index of India
1.3 Role of Anganwadis and ASHA Workers
2 Literature Review
3 Research Gap
4 Methodology
5 Conclusion
References
Challenges and Requirements for Integrating Renewable Energy Systems with the Grid
1 Introduction
2 Grid Interconnection Stages
3 Requirements for Compliance with Grid Code
4 Analysis for Integration Challenges
5 Conclusion
References
Design of Progressive Monitoring Overhead Water Tank
1 Introduction
2 Architectural Details of Proposed System
3 Proposed System
3.1 Raspberry Pi Zero W
3.2 Arduino Uno
3.3 Ultrasonic Sensor
3.4 Flow Sensor
3.5 Turbidity Sensor
3.6 TDS Sensor
3.7 PH Sensor
3.8 Motor and Motor Driver
3.9 Setup
4 Methodology
5 Results
6 Conclusion
References
An Anchor-Based Fuzzy Rough Feature Selection for Text Categorization
1 Introduction
2 Brief Review of Concepts
2.1 Anchor Graph
3 Fuzzy Rough Feature Selection
4 Proposed Hybrid Anchor-Based Fuzzy Rough Feature Selection (ABFRFS)
5 Experimental Results
5.1 Dataset
6 Evaluation Metrics
7 Experiments and Analysis
8 Conclusions
References
Fabric Variation and Visualization Using Light Dependent Factor
1 Introduction
2 Literature Survey
3 Documentation of LDR Sensor After Analysis
4 Procedure
5 Experimental Section
6 Block Diagram
7 Algorithm
8 Results
9 Conclusion
10 Future Scope
References
Pulse Rate Estimation with a Smartphone Camera Using Image Processing Algorithm
1 Introduction
2 Methodology
2.1 Threshold Calculation
2.2 Centroid Point Computation
2.3 Computation and Estimation of Pulse Rate
3 Simulated Results
4 Conclusion
References
Multilayer Perceptron Based Early On-Site Estimation of PGA During an Earthquake
1 Introduction
2 Methodology
2.1 Dataset
3 Data Munging
3.1 Feature Window Population
3.2 MLP-Based Back Propagation Neural Network (MLP-BPNN)
4 Results and Discussion
4.1 Effect of Delay from P-onset on Predictor Performance
5 Conclusion
References
IoT-Equipped Smart Campus Using LoRa Technology
1 Introduction
2 Literature Survey
3 Methodology
3.1 Overview
4 Results and Conclusions
5 Summary
References
A Novel Approach for Visualizing Medical Big Data Using Variational Autoencoders
1 Introduction
2 Related Work
3 Proposed Methodology
3.1 Variational Autoencoder
4 Performance Metrics
4.1 Accuracy
4.2 Precision
4.3 Recall
5 Results and Discussion
6 Conclusion
References
An Efficient Cybersecurity Framework for Detecting Network Attacks Using Deep Learning
1 Introduction
2 Related Work
3 Dataset
4 Architecture
5 Experiments and Result
6 Conclusion and Future Work
References
Evaluation of Network Parameters in Cloud Environment
1 Introduction
2 Literature Review
3 Design Methodology
4 Results and Discussions
5 Conclusion
6 Future Scope
References
Multiple DG Placement in Distribution Network with Reconfiguration Process for Active Power Loss Minimization
1 Introduction
2 LFA and NR Algorithms
2.1 LFA
2.2 Voltage Stability Index (VSI)
2.3 Network Reconfiguration Algorithms (NRAs)
2.4 Flowcharts for NRAs
2.5 Placement of a DG
3 Simulated Results and Comparison
3.1 Power Loss of 33-Bus RDS with Base Configuration
3.2 Comparison of Network Reconfiguration Algorithms
3.3 Planning of DG at Optimal Location
4 Conclusion
References
Using Dynamic Models to Showcase Pandemic Prevention Empirical Covid-19
1 Introduction
2 Literature Review
3 Methodology
4 Conclusion
References
Mobile Application for Predicting Diseases
1 Introduction
2 Literature Survey
2.1 Mobile Application for Health Prediction Using Data Mining—Humne (2017)
2.2 Disease Prediction Application Based on Symptoms—Ramandeep Singh Sethi (2019)
3 Methodology
3.1 Dataset and Data Cleaning
3.2 Mobile Application
3.3 Flask Application
3.4 User Authentication API
3.5 Reading Input from Mobile Application
3.6 Associating Symptoms with Disease
3.7 Mapping the Symptoms
3.8 Disease Prediction
4 Algorithms and Frameworks
4.1 Machine Learning Algorithm
4.2 Web Application Framework
4.3 Mobile Application
4.4 Cloud Deployment
5 Results
6 Conclusion
References
IoT-Based Smart Classroom Environment
1 Introduction
2 Literature Survey
3 Proposed Methodology
4 Result Analysis
5 Conclusions
6 Future Work
References
Fuzzy Logic Control of Liquid Level in a Single Tank with IoT-Based Monitoring System
1 Introduction
2 Pictorial Representation of the Single Tank Process
3 Conventional Control Strategy
4 Fuzzy Gain Scheduled Pi Controller
5 Results
6 Monitoring and Transmitting Liquid Level Through IoT
7 Conclusion
References
Real-Time Traffic Management System Using Machine Learning and Image Processing
1 Introduction
2 Problem Statement
3 Proposed Architecture Design
4 Implementation and Results
5 Conclusion
References
Wireless Animatronic Hand Using Infrared Sensor
1 Introduction
2 Literature Survey
3 Important Terminology
3.1 Infrared Sensor
3.2 CC2500 Module
4 Working
5 Result
6 Conclusion
7 Future Scopes
References
A Study on Core Challenges in Coffee Plant Leave Disease Segmentation and Identification on Various Factors
1 Introduction
2 Extrinsic Factors
2.1 Background Image
2.2 Image Capture Conditions
3 Intrinsic Factors
3.1 Symptom Segmentation
3.2 Symptom Variations
3.3 Multiple Simultaneous Disorders
3.4 Various Disorders with Like Symptoms
4 Other Challenges
5 Future Projections and Probable Solutions for Present Restrictions
6 Research Objectives
7 Conclusions
References
Emotional AI-Based Analysis of Social Media Posts
1 Introduction
2 Literature Review
3 Proposed System
4 Results and Discussions
4.1 Word-Level Sentiment Analysis
4.2 Querying Sentiment Words
5 Conclusion
References
Efficient Brain Tumor Detection Method Using Feature Optimization and Machine Learning Algorithm
1 Introduction
2 Related Work
3 Proposed Methodology
4 Experimental Results
5 Conclusion and Future Scope
References
AI Voice-Assisted Fitness Coach with Body Pose Recognition
1 Introduction
2 Literature Review
3 Architecture
4 Methodology
5 Data
5.1 ML Techniques Used
5.2 ML Pipeline
6 Results
6.1 Workout Catalog
6.2 Statistics of Per-session Workouts
6.3 Results Screenshots
7 Future Discussions
8 Conclusion
References
Voting Ensemble Learning Technique with Improved Accuracy for the CAD Diagnosis
1 Introduction
2 Literature Survey
3 Methodology and Evaluation in Voting Ensemble Learning Techniques
3.1 Dataset Description and Pre-processing
3.2 Feature Selection
3.3 Voting Ensemble Learning
3.4 Mathematical Intuitions of Soft Voting and Hard Voting Ensemble Technique
4 Simulation Results
5 Conclusions
References
Performance Review of Various Classification Methods in Machine Learning for IVF Dataset Using Python
1 Introduction
2 Literature Survey
3 Classification Algorithms Used
4 Methodology
4.1 Data Collection
4.2 Data Cleaning Step
4.3 Data Transformation Step
4.4 Data Reduction Step
4.5 Data Mining Step
4.6 Data Interpretation Step
5 Results of Classification Algorithms
6 Comparison of Classifiers
7 Conclusion
References
Privacy Preserving Mechanism for Medical Data Stored in Cloud During Health Emergencies
1 Introduction
2 Literature Survey
3 System Architecture
3.1 Sequence Diagram
3.2 Data Flow Diagram
3.3 System Design
4 Results
5 Conclusion
References
Deep Learning Applications in Sentiment Analysis
1 Introduction
2 Literature Review
3 Proposed Model
4 Result
5 Conclusion
References
Providing Medical Awareness to Rural Community for Health Emergencies
1 Introduction
2 Literature Survey
3 System Architecture
4 Results and Discussions
5 Conclusion
References
Goodness Ratio and Throughput Improvement Using Multi-criteria LEACH Method in Group Sensing Device Network
1 Introduction
1.1 Motivation
2 Literature Survey
3 Proposed Multi-criteria LEACH
3.1 Single SDGN Network
3.2 Multiple SDGN Network
3.3 SSD Selection Method
3.4 Path Formation Using Multi-criteria LEACH
3.5 Mobile Sink Data Collection
4 Simulation Results
4.1 LEACH
4.2 ELEACH
4.3 Route Discovery Using Multi-criteria LEACH
4.4 Comparison of Multi-criteria LEACH with LEACH and E-LEACH
5 Conclusion
References
Spectral Clustering to Detect Malignant Prostate Using Multimodal Images
1 Introduction
2 Proposed Methodology
2.1 Statement of Ethics
2.2 Data Acquisition and Description
2.3 Noise Delination
2.4 Spectral Clustering Based Prostate Area Segmentation
2.5 Validation and Performance Evaluation
3 Experimental Result and Discussion
4 Conclusion
References
Content-Based Retrieval Using Autoencoder and Transfer Learning
1 Introduction
2 Literature Survey and Related Work
3 Proposed Work
3.1 Methodology
4 Comparison of Methods
4.1 Transfer Learning Methods
4.2 Training Autoencoder
5 Conclusion
References
Facial Recognition-Based Attendance and Smart COVID-19 Norms Monitor
1 Introduction
2 Problem Statement
3 Related Work
3.1 Haar Cascade
3.2 Local Binary Pattern Histogram
3.3 Existing Attendance Systems
4 Proposed Work
5 Conclusion
References
Sensitivity Analysis of Regularization Techniques in Convolution Neural Networks with Tensorflow
1 Introduction to Neural Networks
2 Motivations
3 Literature Survey
4 Convolution Neural Network
5 Regularization in Deep Learning
5.1 Parameter Norm Penalties
5.2 Early Stopping
5.3 Dropout
5.4 Image Data Augmentation
6 Results and Discussion
7 Conclusion
References
Organic Mart: E Commerce Web Site for Agriculture
1 Introduction
2 Problem Statement
3 Literature Survey
4 Methodology
4.1 System Design
4.2 System Architecture
5 Result Analysis
6 Conclusion
References
Multi Station Approximation and Noise Mitigation Process to OFDM Systems Using Successive JCI
1 Introduction
2 System Model
3 Proposed Approaches
4 Experimental Results
5 Conclusion
References
State Budget’s Allocation Management Platform
1 Introduction
2 Literature Survey
3 Implementation
4 Idealogy
5 Results
6 Conclusion
References
An Overview of Data Aggregation Techniques with Special Sensing Intelligent Device Selection Approaches
1 Introduction
2 Tree-Based Methods
3 Group-Based Methods
4 Multi Route Methods
5 Hybrid Methods
6 Special Spin Head Selection Methods
7 Conclusion
References
Mining Health Dataset for Risk Identification
1 Introduction
2 Proposed Methodology
2.1 Graph Development
2.2 Standardized Weights
2.3 Semi-supervised Learning Method Applied on HeteroHER Graph
2.4 Notations
2.5 Objective Function
2.6 Convexity
2.7 Optimization Procedure
2.8 Iterative Solution
2.9 Convergence
3 Conclusion
References
Predictive Control Techniques for Induction Motor Drive for Industrial Applications
1 Introduction
2 Modeling of IM and VSI
3 Model Predictive Control Techniques for an IM Drive
3.1 PCC
3.2 PTC
3.3 PFC
4 Results
5 Conclusion
References
Transliteration from English to Telugu Using Phrase-Based Machine Translation for General Domain English Words
1 Introduction
2 Literature Review
3 Challenges Faced in Transliteration from English to Telugu
3.1 Difference in No. of Alphabets
3.2 Presence of Silent Letter in the Source Language
3.3 Diversity in Pronunciation Associated with a Letter or a Set of Letters (Grapheme) in Different Words
3.4 Pronunciation of Letters on Their Position in a Word
4 Database Creation
5 Segmentation and Training of the Transliteration System
5.1 SegM1
5.2 SegM2
5.3 SegM3
5.4 Settings Used in Training the Transliteration System
5.5 Extraction of Best Transliteration by Combining the Top Three Transliterations Generated When the Data Set Was Segmented by SegM1, SegM2, and SegM3
5.6 A Modified Edit Distance for Accuracy Check Using the Pronunciations of the Target Language
6 Result and Conclusion
References
A Hybrid Artificial Bee Colony Algorithm for the Degree-Constrained Minimum Spanning Tree Problem
1 Introduction
2 hABC for the dc-MST Problem
2.1 Generation of Each Initial Solution of the Population
2.2 Selection Method
2.3 Neighborhood Operators
2.4 Local Search
2.5 Scout Bee Phase
3 Computational Results
3.1 Comparison of hABC with State-of-the-Art Approaches
4 Conclusions
References
Torque Ripple Reduction Control Strategies of Sensor and Sensorless BLDC Motor: A Review
1 Introduction
2 Mathematical Modeling of BLDC Motor
3 Torque Ripple Reduction of Brushless Direct Current Motor
3.1 Commutation Torque Ripple Analysis
3.2 Condition for Minimum Torque Ripple
3.3 Duty Cycle Expression During Conduction
4 Control Strategy for Sensor BLDC Motor for Torque Ripple Reduction
5 Control Strategy for Sensorless BLDC Motor for Torque Ripple Reduction
6 Results and Discussion
7 Conclusion
References
Secured Communication Using Hash Function and Steganography
1 Introduction
2 Literature Review
3 Proposed System
4 Result and Discussion
5 Conclusion
References
Swarm Intelligence and Its Impact on Data Mining and Knowledge Discovery
1 Introduction
1.1 Biological Behavior
1.2 Swarms and Artificial Life
1.3 Data Mining
1.4 Steps of Knowledge Discovery
2 Knowledge Discovery and Swarm Intelligence
3 Ant Colony Optimization and Data Mining
4 Conclusion
References
Design and Implementation of Deep Learning Based Illicit Drug Supplier Detection System
1 Introduction
2 Design of Convolutional Neural Network (CNN)
3 Drug Distributor Detection System Using CNN
3.1 Logging Metrics of CNN in Keras
4 Hardware Implementation of Proposed System
5 Results
5.1 Output of Face Detected from Video Stream
5.2 Face Recognition
6 Conclusion
References
Malaria Detection with Flask Using Deep Learning Model
1 Introduction
2 Related Work
3 Proposed Methodology
3.1 Flask
3.2 Proposed Architecture
4 Experimental Results
5 Conclusion
References
A Survey on AGPA Nature-Inspired Techniques in Vehicular Ad-Hoc Networks
1 Introduction
2 Related Works
3 Discussion
4 Conclusion
References
Reader for Blind Using the Raspberry Pi
1 Introduction
2 Literature Survey
3 Proposed Design
3.1 Methodology
3.2 Block Diagram of Proposed Design
4 Results
5 Conclusion
References
Design and Development of Real-Time Evaluation System for Cognitive Assessment of Students
1 Introduction
2 Materials and Methods
3 Experimental Procedure
4 Results and Discussion
5 Conclusions and Outlook
References
An Effective CNN Method Using Multi-SVM Process for Brain Tumor Segmentation and Detection from MR Images
1 Introduction
2 Methodology and Experimentation Results
3 CNN and SVM Classification
4 Conclusion and Summary
References
Underwater Image Enhancement Using Color Balance and Image Fusion via Gamma Correction
1 Introduction
1.1 Need and Significance
2 Process Flow and Methodology
3 Results and Analysis
4 Conclusion
References
Multimodal Medical Image Fusion Approach Using PCNN Model and Shearlet Transforms via Max Flat FIR Filter
1 Introduction
2 Literature Review
3 Methodology
4 Design and Process Flow of Implementation
5 Experimental Investigations
6 Conclusion
References
A Novel Wideband Millimeter-Wave-Based OFDM Uplink System to Analyze Spectral Efficiency
1 Introduction
2 Literature Review
3 Methodology and Process Flow
3.1 Process Flow
4 Results and Analysis
5 Conclusion
References
Design of QCA-Based 2 to 1 Multiplexer
1 Introduction
2 Basics of QCA
2.1 QCA Basic Cell
3 Proposed Design of 2 to 1 Mux
4 Simulation Results
5 Conclusion
References
Design of QCA-Based XOR/XNOR Structures
1 Introduction
1.1 QCA Basic Cell
1.2 Existing XOR QCA Cell
2 Proposed QCA XOR Cell
3 Simulation Results
4 Conclusion
References
Design of QCA-Based 1-Bit Magnitude Comparator
1 Introduction
2 Literature Review
3 Proposed Architecture of QCA-Based 1-Bit Magnitude Comparator
4 Simulation Results
5 Conclusion
References
A Novel Multimodal Anatomical Medical Image Fusion Using Structure Extraction
1 Introduction
2 Proposed Methodology
3 Process Flow and Algorithm
4 Results and Analysis
5 Conclusion
References
Parametric Analysis for Channel Estimation in Massive MIMO Systems with 1-Bits ADCs
1 Introduction
2 Aim and Significance of the Work
3 Proposed Methodology and Algorithms
4 Experimental Results and Analysis:
5 Conclusion
References
Image Dehazing Using Improved Dark Channel and Vanherk Model
1 Introduction
1.1 Background
1.2 Objectives
1.3 Need and Importance
2 Literature Review
2.1 Suggestions
3 Methodology
3.1 Existing Methods
3.2 Proposed Method
4 Implementation Tool
5 Results
6 Conclusion
References
A Novel Bayesian Fusion Model for IR and Visible Images
1 Introduction
1.1 Background
1.2 Objectives
1.3 Need and Importance
2 Literature Review
2.1 Suggestions
2.2 Advantages
2.3 Limitations
3 Methodology
3.1 Existing Methods
3.2 Proposed Method
4 Implementation Tool
5 Results
6 Conclusion
References
Retinal Boundary Segmentation in OCT Images Using Active Contour Model
1 Introduction
2 Significance of the Proposed Work
3 Proposed Methodology
4 Experimental Investigations:
5 Conclusion
References
Spectral Efficiency for Multi-bit and Blind Medium Estimation of DCO-OFDM Used Vehicular Visible Light Communication
1 Introduction
2 Literature Review
3 Existing Methodology
4 Proposed Methodology
5 Results
6 Conclusion
References
An Efficient Retinal Layer Segmentation Based on Deep Learning Regression Technique for Early Diagnosis of Retinal Diseases in OCT and FUNDUS Images
1 Introduction
2 Need and Significance of the Proposed Work
3 Proposed Methodology
4 Conclusion
References
Design of QCA-Based BCD Adder
1 Introduction
2 Basics of QCA
2.1 QCA Basic Cell
3 Existing Architecture
4 Proposed System
5 Results and Discussion
6 Conclusion
References

Citation preview

Lecture Notes in Electrical Engineering 947

Amit Kumar Sabrina Senatore Vinit Kumar Gunjan   Editors

ICDSMLA 2021 Proceedings of the 3rd International Conference on Data Science, Machine Learning and Applications

Lecture Notes in Electrical Engineering Volume 947

Series Editors Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli Federico II, Naples, Italy Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán, Mexico Bijaya Ketan Panigrahi, Department of Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore Rüdiger Dillmann, Humanoids and Intelligent Systems Laboratory, Karlsruhe Institute for Technology, Karlsruhe, Germany Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China Gianluigi Ferrari, Università di Parma, Parma, Italy Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid, Madrid, Spain Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität München, Munich, Germany Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA, USA Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt Torsten Kroeger, Stanford University, Stanford, CA, USA Yong Li, Hunan University, Changsha, Hunan, China Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA Ferran Martín, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra, Barcelona, Spain Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA Sebastian Möller, Quality and Usability Laboratory, TU Berlin, Berlin, Germany Subhas Mukhopadhyay, School of Engineering and Advanced Technology, Massey University, Palmerston North, Manawatu-Wanganui, New Zealand Cun-Zheng Ning, Department of Electrical Engineering, Arizona State University, Tempe, AZ, USA Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan Luca Oneto, Department of Informatics, Bioengineering, Robotics and Systems Engineering, University of Genova, Genova, Genova, Italy Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Gan Woon Seng, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, Singapore Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Germany Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China Walter Zamboni, DIEM—Università degli studi di Salerno, Fisciano, Salerno, Italy Junjie James Zhang, Charlotte, NC, USA

The book series Lecture Notes in Electrical Engineering (LNEE) publishes the latest developments in Electrical Engineering—quickly, informally and in high quality. While original research reported in proceedings and monographs has traditionally formed the core of LNEE, we also encourage authors to submit books devoted to supporting student education and professional training in the various fields and applications areas of electrical engineering. The series cover classical and emerging topics concerning: • • • • • • • • • • • •

Communication Engineering, Information Theory and Networks Electronics Engineering and Microelectronics Signal, Image and Speech Processing Wireless and Mobile Communication Circuits and Systems Energy Systems, Power Electronics and Electrical Machines Electro-optical Engineering Instrumentation Engineering Avionics Engineering Control Systems Internet-of-Things and Cybersecurity Biomedical Devices, MEMS and NEMS

For general information about this book series, comments or suggestions, please contact [email protected]. To submit a proposal or request further information, please contact the Publishing Editor in your country: China Jasmine Dou, Editor ([email protected]) India, Japan, Rest of Asia Swati Meherishi, Editorial Director ([email protected]) Southeast Asia, Australia, New Zealand Ramesh Nath Premnath, Editor ([email protected]) USA, Canada Michael Luby, Senior Editor ([email protected]) All other Countries Leontina Di Cecco, Senior Editor ([email protected]) ** This series is indexed by EI Compendex and Scopus databases. **

Amit Kumar · Sabrina Senatore · Vinit Kumar Gunjan Editors

ICDSMLA 2021 Proceedings of the 3rd International Conference on Data Science, Machine Learning and Applications

Editors Amit Kumar BioAxis DNA Research Centre Private Ltd. Hyderabad, Telangana, India Vinit Kumar Gunjan Department of Computer Science and Engineering CMR Institute of Technology Hyderabad, Telangana, India

Sabrina Senatore Department of Computer Engineering, Electrical Engineering and Applied Mathematics University of Salerno Fisciano, Salerno, Italy

ISSN 1876-1100 ISSN 1876-1119 (electronic) Lecture Notes in Electrical Engineering ISBN 978-981-19-5935-6 ISBN 978-981-19-5936-3 (eBook) https://doi.org/10.1007/978-981-19-5936-3 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

The rising technologies of artificial intelligence (AI), data science, and machine learning (ML) are advancing businesses more quickly than ever. Success in this era of digital transformation depends on leveraging analytics to uncover vast amount of data with detailed insights. In the past, these insights were uncovered manually through extensive data analysis and data complexity which continue to rise. The most recent technologies for data scientists are AI and ML, which allow them to quickly customize the data to verify its usefulness. With the advancements in computing power, organizations are now interested in analyzing both internal and external data to identify previously unknown insights that will drive operational efficiency, boost sales, and provide a competitive marketplace advantage. AI is a technology that continues to advance rapidly, and the discourse on AI ethics and governance is also evolving. Globally, a number of different sets of ‘AI ethics principles’ have been put forward by multilateral organizations, private sector entities, and several nation states. The future of AI is determined by a diverse group of stakeholders, including researchers, private organizations, government, standardsetting bodies, regulators, and general citizens. Around the world, many countries and organizations have defined principles to guide responsible management of AI for stakeholders. Though automated solutions are expected to introduce objectivity to decision making, recent cases globally have shown that AI solutions have the potential to be ‘biased’ against specific sections of society. This can lead to inconsistent output across a diverse demography. Selected papers from the International Conference on Data Science, Machine Learning and Applications, 2021 are placed in this volume. The chapters in this book provide an overview of the key concepts and theories underlying the technologies and applications deliberated during the 2021 ICDSMLA Conference, with a focus on data science, machine learning, face recognition, evolutionary algorithms, like genetic algorithms, automotive applications, automation devices using artificial neural networks, business management systems, the Internet of Things (IoT), and contemporary speech processing systems. Additionally, the topics covered in this book are contemporary developments in sensor networks, VLSI systems, precision v

vi

Preface

agriculture, and medical diagnostic systems. Wherever appropriate, a discussion of learning and software modules in artificial intelligence, soft computing, and deep learning algorithms is included. In a nutshell, this book sheds light on the data science and ML-based societal innovations that are useful for the social good. Hyderabad, India Fisciano, Italy Hyderabad, India

Amit Kumar Sabrina Senatore Vinit Kumar Gunjan

Contents

Road Accident Detection and Indication System . . . . . . . . . . . . . . . . . . . . . . Y. Lavanya, P. BhagyaSri, P. BhuvanaSri, and K. Noha Namratha Attendance System Based on Face Recognition Using Haar Cascade and LBPH Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Akshat Kumar Rai, A. Akash, G. Kavyashree, and Thaseen Taj Different Thresholding Techniques in Image Processing : A Review . . . . Radha Seelaboyina and Rajeev Vishwakarma Dynamic Weighting Selection for Predictive Torque and Flux Control of Industrial Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Vishnu Prasad Muddineni, Anil Kumar Bonala, and Thanuja Penthala Population Index and Analysis Based on Different Geographies; Using Distance Measurement, Social Distancing, and Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bhushan Chougule, Samiksha Baral, Minal Tayde, and Kaustubh Sakhare On the Discriminability of Samples Using Binarized ReLU Activations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Michał Lewandowski, Werner Zellinger, Hamid Eghbal-zadeh, Natalia Shepeleva, and Bernhard A. Moser Supervised and Unsupervised Machine Learning Approaches—A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Esther Varma and Puja S. Prasad

1

9 23

31

45

65

73

Skin Cancer Classification Using Deep Learning . . . . . . . . . . . . . . . . . . . . . D. K. Yashaswini, Pratheeksha C. Dhanpal, and S. A. Bhoomika

83

Crop Yield Prediction Using Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . K. Mamatha, Shantideepa Samantha, and Kundan Kumar Prasad

93

vii

viii

Contents

Real-Time Tweets Streaming and Comparison Using Naïve Bayes Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 S. R. Shankara Gowda, Rose King, and M. R. Pavan Kumar Smart Shopping Trolley for Billing System . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 R. Kishor Kumar, V. Ashwitha, S. Jeevitha, P. Pranusri, and D. Rakshitha A Survey on IoT Protocol in Real-Time Applications and Its Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 M. L. Umashankar, S. Mallikarjunaswamy, N. Sharmila, D. Mahesh Kumar, and K. R. Nataraj Safe Characteristic Signature Systems with Different Jurisdiction Using Blockchain in E-Health Records . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Shivakumar Dalali, B. K. Pramod, Ranjith Kumar, and M. J. Thejas Jain Web-Based Trash Segregation Using Deep Learning Algorithm . . . . . . . . 139 S. Sheeba, Akshay Mohan, Ashish Kumar Jha, Bikash Agarwal, and Priya Singh Home Automation Using Face Recognition for Wireless Security . . . . . . . 149 B. S. Umashankar, Mandalia Vishal Shailesh, Md Shaghil Z. Ansari, and Rahul Markandey Hybrid-Network Intrusion Detection (H-NID) Model Using Machine Learning Techniques (MLTs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 K. R. Pradeep, Arjun S. Gowda, and M. Dakshayini Impact of Using Partial Gait Energy Images for Human Recognition by Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Devanshi Singh and K. T. Thomas Several Routing Protocols, Features and Limitations for Wireless Mesh Network (WMN): A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Jasleen Kaur and Hardeep Singh A Deep Meta-model for Environmental Sound Recognition . . . . . . . . . . . . 201 K. S. Arun Spatial Computing: Next Big Thing of Physical and Digital World . . . . . 211 Dweepna Garg, Bhavika Patel, Radhika Patel, and Ritika Jani Cloud Accessing Based on IOT Oriented WSNs for Optimal Water Conservation in Farming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 K. Raju, Y. Lavanya, J. Prasanth Kumar, and Jagan Mohan Rao S-Extension Patch: A Simple and Efficient Way to Extend an Object Detection Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Dishant Parikh

Contents

ix

Roles and Impact of ASHA Workers in Combating COVID-19: Case Study Bhubaneswar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Manjusha Pandey, S. N. Misra, Abhipsa Ray, and S. S. Rautaray Challenges and Requirements for Integrating Renewable Energy Systems with the Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Komal Bai, Vikas Sindhu, Ahteshamul Haque, and V. S. Bharath Kurukuru Design of Progressive Monitoring Overhead Water Tank . . . . . . . . . . . . . . 267 N. Alivelu Manga, Surya Teja Manupati, N. S. C. Viswanadh, P. Sriram, and D. V. S. G. Varun An Anchor-Based Fuzzy Rough Feature Selection for Text Categorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 Ananya Gupta and Shahin Ara Begum Fabric Variation and Visualization Using Light Dependent Factor . . . . . 293 Gorsa Lakshmi Niharika, Shahana Bano, Kondapaneni Charan Sai, Kavuri Rohith, and Dasaradh Gutta Pulse Rate Estimation with a Smartphone Camera Using Image Processing Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 E. C. Sowmiya, K. Nirmala, and L. Suganthi Multilayer Perceptron Based Early On-Site Estimation of PGA During an Earthquake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Siddhartha Sarkar, Satish Kumar, Anubrata Roy, and Bhargab Das IoT-Equipped Smart Campus Using LoRa Technology . . . . . . . . . . . . . . . . 327 D. Annapurna, D. Tejus, Girish Narayan, Shrushti Hegde, and Parth PratimMishra A Novel Approach for Visualizing Medical Big Data Using Variational Autoencoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 G. Madhukar Rao and Dharavath Ramesh An Efficient Cybersecurity Framework for Detecting Network Attacks Using Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 K. R. Nataraj, Manasa, and M. Chandana Evaluation of Network Parameters in Cloud Environment . . . . . . . . . . . . . 355 S. R. Ahrthi, G. Sinchana, A. Trisha, and B. Sahana Multiple DG Placement in Distribution Network with Reconfiguration Process for Active Power Loss Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363 B. Devulal, M. Siva, D. Ravi Kumar, A. Supriya, and P. Sushma Devi

x

Contents

Using Dynamic Models to Showcase Pandemic Prevention Empirical Covid-19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 G. S. Gowramma, Swaraj, B. Chandra Kiran, and O. Manoj Kumar Mobile Application for Predicting Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . 385 G. S. Gowramma, S. P. Vaishnav Bharadwaj, and P. Rahul IoT-Based Smart Classroom Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 D. Annapurna, Bhavan Naik, Akshaya Visvanathan, Akhil S. Kumar, and Atharva Moghe Fuzzy Logic Control of Liquid Level in a Single Tank with IoT-Based Monitoring System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401 K. Thirupura Sundari, R. Giri, M. G. Umamaheswari, S. Durgadevi, and C. Komathi Real-Time Traffic Management System Using Machine Learning and Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415 S. Sheela, K. R. Nataraj, and K. R. Rekha Wireless Animatronic Hand Using Infrared Sensor . . . . . . . . . . . . . . . . . . . 425 M. Balakarthikeyan, D. Rajesh, M. Sai Jagadeesh, and G. Santhosh Kumar A Study on Core Challenges in Coffee Plant Leave Disease Segmentation and Identification on Various Factors . . . . . . . . . . . . . . . . . . 433 S. Santhosh Kumar, B. K. Raghavendra, S. Ashoka, and Siddaraju Emotional AI-Based Analysis of Social Media Posts . . . . . . . . . . . . . . . . . . . 447 Geleta Negasa Binegde Efficient Brain Tumor Detection Method Using Feature Optimization and Machine Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . 455 Ashish Bhatt and Vineeta Saxena Nigam AI Voice-Assisted Fitness Coach with Body Pose Recognition . . . . . . . . . . 467 S. Moulya, T. R. Pragathi, and Pandurang S. Kambali Voting Ensemble Learning Technique with Improved Accuracy for the CAD Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477 Geetha Pratyusha Miriyala and Arun Kumar Sinha Performance Review of Various Classification Methods in Machine Learning for IVF Dataset Using Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489 G. S. Gowramma, Shantharam Nayak, and K. Rakshitha Privacy Preserving Mechanism for Medical Data Stored in Cloud During Health Emergencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503 R. Bhaskar, M. Komala, B. M. Lekhana, and J. Pooja

Contents

xi

Deep Learning Applications in Sentiment Analysis . . . . . . . . . . . . . . . . . . . 513 Abhilash Shukla, Dhatri Raval, Jaimin Undavia, Nilay Vaidya, Krishna Kant, Sohil Pandya, and Atul Patel Providing Medical Awareness to Rural Community for Health Emergencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521 R. Bhaskar, A. Tharunsai, K. M. Sneha, and N. Tejaswini Goodness Ratio and Throughput Improvement Using Multi-criteria LEACH Method in Group Sensing Device Network . . . . . . 527 V. S. Rekha and Siddaraju Spectral Clustering to Detect Malignant Prostate Using Multimodal Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549 Kiran Ingale, Pratibha Shingare, and Mangal Mahajan Content-Based Retrieval Using Autoencoder and Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559 Poornima Raikar and S. M. Joshi Facial Recognition-Based Attendance and Smart COVID-19 Norms Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571 B. S. Umashankar, S. Lakshmi Narayan, M. Ruthvik, and Prajwal Deshpande Sensitivity Analysis of Regularization Techniques in Convolution Neural Networks with Tensorflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581 Shivakumar Dalali, B. E. ManjunathSwamy, Giridhar Gowda, and N. S. Girish Rao Salanke Organic Mart: E Commerce Web Site for Agriculture . . . . . . . . . . . . . . . . 593 Shivakumar Dalali, C. J. Adarsh, B. K. Abhishek, and K. Akshay Multi Station Approximation and Noise Mitigation Process to OFDM Systems Using Successive JCI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603 N. P. Sarada Devi and M. L. Ravi Chandra State Budget’s Allocation Management Platform . . . . . . . . . . . . . . . . . . . . . 613 Shivakumar Dalali, G. Prem Kumar, S. Nishanth, and M. N. Kumar Raja An Overview of Data Aggregation Techniques with Special Sensing Intelligent Device Selection Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621 G. Nirmala and C. D. Guruprakash Mining Health Dataset for Risk Identification . . . . . . . . . . . . . . . . . . . . . . . . 635 Swapna Gangone, Bhoomeshwar Bala, and G. J. Bharat Kumar Predictive Control Techniques for Induction Motor Drive for Industrial Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643 Thanuja Penthala, Saravanan Kaliyaperumal, Vishnu Prasad Muddineni, and Anil Kumar Bonala

xii

Contents

Transliteration from English to Telugu Using Phrase-Based Machine Translation for General Domain English Words . . . . . . . . . . . . . 657 Radha Mogla, Chellapilla Vasantha Lakshmi, and Niladri Chatterjee A Hybrid Artificial Bee Colony Algorithm for the Degree-Constrained Minimum Spanning Tree Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671 Sudishna Ghoshal and Shyam Sundar Torque Ripple Reduction Control Strategies of Sensor and Sensorless BLDC Motor: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 681 M. Karthika and K. C. R. Nisha Secured Communication Using Hash Function and Steganography . . . . . 693 P. Shanmuga Priya, T. Manikandan, R. Sathya, T. Helan Vidhya, D. Sasirekha, and M. Tamilarasi Swarm Intelligence and Its Impact on Data Mining and Knowledge Discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699 Mantripragada Yaswanth Bhanu Murthy, Katragadda Prasanthi, Mrudula Kilaru, and Sandhya Rani Kakarla Design and Implementation of Deep Learning Based Illicit Drug Supplier Detection System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707 M. Arulmozhi, Nandini G. Iyer, C. Amutha, S. Jeny Sophia, P. Sivakumar, and S. B. Nivethitha Malaria Detection with Flask Using Deep Learning Model . . . . . . . . . . . . 721 Deshmukh Sushant and Parag Bhalchandra A Survey on AGPA Nature-Inspired Techniques in Vehicular Ad-Hoc Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 729 Smita Rani Sahu and Biswajit Tripathy Reader for Blind Using the Raspberry Pi . . . . . . . . . . . . . . . . . . . . . . . . . . . . 741 Y. Pavan Kumar Reddy, G. Hemadri, U. Jaya Nithya, D. S. Haneef Basha, and Y. Hari Krishna Design and Development of Real-Time Evaluation System for Cognitive Assessment of Students . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 749 S. Lokesh and T. Sreenivasulu Reddy An Effective CNN Method Using Multi-SVM Process for Brain Tumor Segmentation and Detection from MR Images . . . . . . . . . . . . . . . . . 759 M. Ravi Kishore, V. Dinesh Kumar, J. Kiranmai, G. Bhuvaneshwar, E. Koteshwara Goud, and Delampady Suresh

Contents

xiii

Underwater Image Enhancement Using Color Balance and Image Fusion via Gamma Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767 K. Giridhar Chaitanya, B. Chandana, S. Jyoshna Devi, P. Gowthami, and B. Varshith Reddy Multimodal Medical Image Fusion Approach Using PCNN Model and Shearlet Transforms via Max Flat FIR Filter . . . . . . . . . . . . . . . . . . . . . 773 Y. Pavan Kumar Reddy, A. Vaishnavi, M. Sudheeshnavi Devi, M. Siva Prasad, and B. Sreenadh Reddy A Novel Wideband Millimeter-Wave-Based OFDM Uplink System to Analyze Spectral Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 781 C. H. Nagaraju, Manoj Kumar Patil, C. Maheswari, U. K. Rahul, and D. Rajesh Design of QCA-Based 2 to 1 Multiplexer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791 M. Ravi Kishore, B. Amaravathy, V. Siva Nagendra Prasad, M. Surya Prakash Reddy, P. Sudarshan, and N. Bala Dastagiri Design of QCA-Based XOR/XNOR Structures . . . . . . . . . . . . . . . . . . . . . . . 799 Gunda Sudha Kiran, U. Dinesh Kumar, K. Chandra Sekhar, L. Deepika, and Y. Krishna Vamsi Design of QCA-Based 1-Bit Magnitude Comparator . . . . . . . . . . . . . . . . . . 809 P. Syamala Devi, K. Vaniha, K. Vidya Sagar, P. Vinitha, and K. Sumanth Kumar A Novel Multimodal Anatomical Medical Image Fusion Using Structure Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 817 G. Obulesu, K. Aparna, S. Afrin, G. Abhinav Kumar Reddy, and A. Kavitha Parametric Analysis for Channel Estimation in Massive MIMO Systems with 1-Bits ADCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827 CH. Nagaraju, S. Arshia Shajarin, V. Bhaskar Reddy, V. Bhaskar Reddy, and C. Anil Kumar Reddy Image Dehazing Using Improved Dark Channel and Vanherk Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 837 S. Fahimuddin, D. Lavanya, T. Manasa, S. Maruthi Praveen, and M. Raveendra Babu A Novel Bayesian Fusion Model for IR and Visible Images . . . . . . . . . . . . 851 S. Fahimuddin, A. Sree Keerthana Reddy, B. Rajitha, K. Sai Prasanth, and U. Sai Retinal Boundary Segmentation in OCT Images Using Active Contour Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863 Shaik Fahimuddin, T. Subbarayudu, M. Vinay Kumar Reddy, G. Venkata Sudharshan, and G. Sudharshan Reddy

xiv

Contents

Spectral Efficiency for Multi-bit and Blind Medium Estimation of DCO-OFDM Used Vehicular Visible Light Communication . . . . . . . . . 873 Shaik Karimullah, E. Sai Sumanth Goud, and K. Lava Kumar Reddy An Efficient Retinal Layer Segmentation Based on Deep Learning Regression Technique for Early Diagnosis of Retinal Diseases in OCT and FUNDUS Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 887 L. Siva Yamini, S. Shylu, G. Viveka, J. Sai Dheeraj, and N. Srihari Design of QCA-Based BCD Adder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893 S. Javeed Basha, B. Shilpa, A. Vyshnavi, Y. Soma Sundar Reddy, and C. Sudharshan

Road Accident Detection and Indication System Y. Lavanya, P. BhagyaSri, P. BhuvanaSri, and K. Noha Namratha

Abstract This main objective of this paper is to develop a detection system using Arduino, GPS, GSM and MEMS when a vehicle met with an accident. Here, MEMS technology detects the sudden change in the axes of vehicle and GSM module will send a message with the location of the accident to nearby hospital and relatives to alert them. GPS module is used to share the location of accident in the form of Google map link, derived from the latitude and longitude form. The alert message received by other side also contains the speed of vehicle in kmph. The proposed idea can also be used to track a vehicle continuously, by just making few changes in both software and hardware. The alert system has been developed in real time and performance is discussed. Keywords Arduino UNO · MEMS · GSM module · GPS module · Accident detection system

1 Introduction Nowadays, the usage of own transportation, mainly two wheelers, has been increased due to many reasons than the public transportation. Apart from that it increases more vehicles to move on roads resulting in the increase of road traffic and also the risk of road accidents. Because of this reason, human beings are losing their lives as there is no immediate medical support and transportation facility. The accidents cannot be avoided all the time but to some extent human lives can be saved by reducing the occurrence of accidents and providing them medical support at the earliest. The proposed system provides scope for instant medical facilities as soon as the incident is happened by sending a message to the nearest hospitals and to the registered known people. The system which is proposed here can also be used to track and find the stolen vehicles and lost traveling luggage, etc. The proposed system mainly constitutes Y. Lavanya (B) · P. BhagyaSri · P. BhuvanaSri · K. Noha Namratha Department of ECE, Ramachandra College of Engineering, Eluru, Andhra Pradesh, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_1

1

2

Y. Lavanya et al.

a one-board embedded system that has GPS and GSM modules connected to an Arduino UNO. Additionally, it consists of a vibration sensor that works on the principle of piezoelectric effect which identifies the condition of the vehicle based on vibration produced at the receiver end. The vibration signal produced is then compared with the vibration signal values that a vehicle generates normally. Based on the difference, it is confirmed that the vehicle has undergone an accident. And information regarding that incident and the spot where it exactly was happened are then informed to the registered people and the nearby hospitals with the help of GSM module. As we know that global system for mobiles (GSM) technology is used to establish cellular connection and works with the help of a SIM card inserted in it. GPS is used to trace the position of the vehicle. This paper also addresses an intelligent emergency alert and safe system. This paper is framed as a discussion on existing alert systems, proposed model, real-time expansion and working on vehicle accident vigilance system.

2 Existing Models In [1], the litterateur tells about a system which makes use of a smart phone that helps to be vigilant when accident is occurred. In order to locate and share the accident spot, GPS technology is used. The alert message, which is generated, will be sent to the nearby people who are being present within the accident zone so that the prominence of the message will increase. Based on that, it will be sent to emergency services which are being close to the collision area. Due to privacy problem that are being faced by the people in the society, “block-chain methodology” is suggested in this first existing system to provide privacy to the users. The block-chain technology is an interlinked systematic chain of blocks that contains transaction history and other user data. It works under the principle of decentralized distributed digital ledger. The other system prescribed by the litterateur in [2] provides a system which is limited only to detect and avoid crashing of two wheelers and to notify the nearby medical facilities and to request for medical assistance. A vibrator has been attached to the vehicle, and a heartbeat sensor is placed to the human body. So with the help of that sensor, we can sense the pulse rate of the user, and hence, the seriousness of the accident can be understood. As Smart phone user consists of the android application will helps to intimate the user’s family and friends about the crash happened for back up. Due to presence of another application like goggle mapping helps to share the location of the accident that has been occurred which will come very handy to rescue the patient. In reference paper [3], road accidents are mainly detected by using a novel technology called the deep learning methodology, in which a convolution neural network is used. A convolutional neural network (ConvNet/CNN) is a deep learning algorithm which can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other. The camera module of the system is deployed in accident-prone areas. Whenever

Road Accident Detection and Indication System

3

an accident occurs, it will detect the accident and immediately report about it to the nearby control room. Hence, an adequate data is sent to the desired emergency services, and if this system is utilized properly, many human lives can be saved. In paper [4], the writer gives an idea on a practical model in order to reduce two wheelers crashing and helps to reduce the various possibilities that are sentenced to death. In this system, it is also ensured that the chances of two wheelers accident can be slowly be demolished strongly to some extent (approximately of 9%). In paper [5], that are mentioned in reference, the writer specifies that an automatic caution system has been designed to confirm or to provide the safety of the people by providing a rescue team to immediately respond to the incident around and to provide necessary medical help. Here, author uses a Zig-Bee technology to transmit the alert message to rescue team. In paper [6], similar to paper [3], a deep learning-based technique is preferred mainly for detection purpose. As we know that this method helps to identify the accurate spot where the collision has been taken place. Also it sends a message for medical emergency. Deep learning technology is been used to create a web-based platform which consists of sensors which can detect the collision. We already know that over speed is the basic reason for occurrence of road accidents. In [7], the author proposes a system which interprets and monitors the velocity of the vehicle using GPRS system, and GPS system detects the accident spot by using the collected data and sends it to alert service center. In service centers, using microcontroller, the new speed is compared with the standard or restricted speed limit, and if the new speed exceeds the limit value, then it understood that the situation is abnormal. In [8], the author proposes a system in which different types of sensors are used to sense the chances of occurrence of an accident that might occur by the vehicles, using some specific sensors that are inserted to the vehicles. So with the help of this system, caution can be given for instantaneous back up or for alerting. In reference paper [9], similar to paper [7], the pen provides a system which can sense the speed of a vehicle and notifies the passengers who are travelling or present within the vehicle, just before the occurrence of an accident, which may occur. The data collection can be done with the help different sensors that are placed within the vehicle. Later, if accident occurs, it will be intimated to the concerned members through messaging system. But this method is more suitable for large vehicles. In [10], the litterateur suggests a system to analyze physical parameters like pressure, acceleration, stress or strain, etc., to identify the status of a vehicle when it is in motion. By continuous monitoring and analyzing the system’s data, the success rate of detection can be increased. Finally, in [11], the pen proposes a system to reduce the false alarm that is wrong information has been occurred by setting up an application which can detect an object’s all kinds of physical motions and to intimate accident alert to the respective departments to provide required facilities. The limitations to be noted from the recent caution systems are: 1. Cost of the system is more.

4

Y. Lavanya et al.

2. If there is any problem, the mobile, like network connectivity, charging etc., the spot cannot be mapped. 3. In case of emergency, when the person who met with an accident, is unconscious, he cannot confirm the assistance, which has to be done.

3 Proposed Model The proposed system, which deals with the accident monitoring alert system shown in above Fig. 1, the main element here, is Arduino UNO; the main application of this is to communicate with other elements easily. If there occurs an accident, the system gets alerted and starts initiated with accelerometer, by communicating with Arduino UNO, it finds the changes which are happened suddenly in the motion, and eventually vibration sensor also entails to identify the accident. If the collision occurs, the Arduino UNO identifies it, starts communicating to GPS (NEO-6M) and procures the location coordinates. It calculates the latitude and longitude of the received location and that is the essential thing of the project. Now, the GPS receives the data and it is given to the GSM for eventual developments in the system. The vehicle gets fixed by installing the system, and the end user needs to register with their mobile number, and then the GSM alerts the user with a message to that registered number if collision occurs. Once the emergency situation occurs like accident, the hospital nearby receives the information through the GSM from the accident location. The message consists of the accident location through the google maps with respect to the message emergency. Fig. 1 Schematic of the proposed model

Road Accident Detection and Indication System

5

Nowadays, the accident number has been increased; the accident occurred location will be provided by the system and later the information will be send to the GPS and GSM system authorities wirelessly. The medical help will be given within a short time after detecting the location of the accident; the location will be detected by the vehicle crash alert system. The sensors get activated immediately after the vehicle undergoes an accident and so there the ambulance takes minimum(less) time to reach the location of the accident occurring spot. The main sensor of the system is vibration sensor which is very essential in detecting the vehicle crash and the coordinates will be found with the help of GPS and the immediate medical help will be initiated from the authority once the information received from the GSM module. Ultimately, the men life will be saved in an intelligent way using the system.

4 The Design of Vehicle Crash Alert System The proposed system of VCAS figure shows in Fig. 2. The accident detection system circuit is given in Fig. 2, and the components and the connections are given in the following board of Arduino UNO. The transmission pin of the GPS’s is connected to the Arduino UNO 10th pin.

Fig. 2 Schematic diagram of vehicle crash alert system

6

Y. Lavanya et al.

The D2 and D3 pins are connected to the RX and TX pins of the GSM module and also to the Arduino board. The GPS which has RX will be kept as it is. The 12v power supply is connected to GSM separately and also to ground. The Arduino UNO and LCD are also interfaced and in parallel LCD pins of D4, D5, D6 and D7 are connected to the Arduino UNO pins of the 6,7,8,9 and the Arduino number 4 and 5 are fixed to the LCD of EN pin and connected to the RW pin as well to the ground. The brightness of the LCD screen is controlled by the potentiometer which is associated to it. The Arduino UNO board pins of A1, A2 and A3 and affixed to the X, Y, Z pins of the accelerometer. All the components are grounded as well to it the accelerometer is also grounded. The Arduino UNO board A4 pin is connected to the Vout of the vibration sensors and then grounded. This is the way the Arduino UNO board is interfaced with all the other components. The scooter is used to test the proposed method. The below given are the merits of the proposed system, • The threats and hazards will be monitored. • The nearby hospitals and police stations are alerted by sending the information (message). • The system is very economical. • All types of vehicle can use this system. • Automatically, the message will notify the accident. • This technique is used for a social cause. • The manual functioning was not needed for the system. The prototype in Fig. 3 ensures compactness and an easy installable system in a vehicle and can be further reduced. The importance of saving life depends on time and now this can make in advance.

5 Conclusion All the required data is gathered with the help of sensors to detect the vehicle crash in the proposed system, and the Arduino microcontroller helps to processes it. And in order to know the current location of the accident, the GPS module is used. For transmitting the messages to different types of devices here, the Arduino plays as transmitter. The registered mobile number will receive the message once the GSM module receives the alert from the vibration sensor which is activated along with the accelerometer when the accident takes place. After the accident takes place, the coordinates of that geographical area will be located using the GPS. With the help of this system, we can probably reduce the accident number and also the system can prevent the deaths. The system is very economical and very user friendly.

Road Accident Detection and Indication System

7

Fig. 3 Prototype of accident detection and alert system

References 1. Praba Devi GS, Miraclin Joyce Pamila JC (2019) Accident alert system application using a privacy-preserving blockchain-based incentive mechanism. In: 5th International conference on advanced computing & communication systems (ICACCS), Coimbatore, India, pp 390–394 2. Mishra B, Singh N, Singh R (2014) Master-slave group based model for co-ordinator selection, an improvement of bully algorithm. In: 2014 International conference on parallel, distributed and grid computing. IEEE, pp 457–460 3. Rajesh G, Benny AR, Harikrishnan A, Jacob Abraham J, John NP (2020) A deep learning based accident detection system In: International conference on communication and signal processing (ICCSP), Chennai, India, pp 1322–1325 4. Gunjan VK, Prasad PS, Mukherjee S (2020) Biometric template protection scheme-cancelable biometrics. In: ICCCE 2019. Springer, Singapore, pp 405–411 5. Dhanya S, Ameenudeen PE, Vasudev A, Benny A, Joy S (2018) Automated accident alert In: International conference on emerging trends and innovations in engineering and technological research (ICETIETR), Ernakulam, pp 1–6 6. Singh N, Ahuja NJ (2019) Empirical analysis of explicating the tacit knowledge background, challenges and experimental findings. Int J Innov Technol Explor Eng (IJITEE) 8(10):4559– 4568 7. SyedulAmin M, Jalil J, Reaz MBI (2012) Accident detection and reporting system using GPS. In: GPRS and GSM technology international conference on informatics, electronics & vision (ICIEV), Dhaka, pp 640–643 8. Kodali RK, Sahu S (2017) MQTT based vehicle accident detection and alert system. In: 3rd International conference on applied and theoretical computing and communication technology (iCATccT), Tumkur, pp 186–189

8

Y. Lavanya et al.

9. Shaik F, Sharma AK, Ahmed SM, Gunjan VK, Naik C (2016) An improved model for analysis of Diabetic Retinopathy related imagery. Indian J Sci Technol 9:44 10. Basheer FB, Alias JJ, Favas CM, Navas V, Farhan NK, Raghu CV (2013) Design of accident detection and alert system for motorcycles. In: IEEE global humanitarian technology conference: South Asia Satellite (GHTC-SAS), Trivandrum, pp 85–89 11. Faiz AB, Imteaj A, Chowdhury M (2015) Smart vehicle accident detection and alarming system using a smart phone. In: International conference on computer and information engineering (ICCIE), Rajshahi, pp 66–69

Attendance System Based on Face Recognition Using Haar Cascade and LBPH Algorithm Akshat Kumar Rai, A. Akash, G. Kavyashree, and Thaseen Taj

Abstract To manage attendance of each student is much more complex when it’s done manually, and it has to be done for each class. To overcome this, we can use face recognition application where it takes less time comparatively, and it is stored in the database; this avoids confusions, proxies, and error of storing data when done manually. The students need to fill their data along with parent’s details, so when the students are absent, the message will be passed to them. After taking attendance through the application, mails and message of the data entry of attendance will be sent to the respective teachers. In advance, the details need to be entered beforehand, so when the students come in front of the camera, their face is recognized by comparing them from database containing faces. When it is unable to recognize, the faces will be stored in unknown, so when there is a mishap of being absent, it can be checked later on. This method of attendance is much more successful. Compare to other algorithms, haar cascade and local binary pattern histogram is best due to their robustness and less false rate. Keywords Haar cascade · Attendance system · Face recognition · Face detection · Gray scale · LPBH

1 Introduction Traditionally, the attendance of students is taken by calling out each student names and marking it out on the student’s attendance registers which is provided for the faculty members. This method may sound easy and less complicated, but the problem arises when the number of students are more it may result with errors in the registers; it is common for humans to make mistakes, while being in a rush to avoid this, we can use face recognition application where the data of each students are done when they come in front of the camera and click recognize, and it will be stored in the data

A. K. Rai · A. Akash · G. Kavyashree · T. Taj (B) Department of CSE, Don Bosco Institute of Technology, Bangalore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_2

9

10

A. K. Rai et al.

automatically by recognizing them from the faces which is been already stored in the data with faces. This method has many advantages like its time efficient and storing of data doesn’t cause any mishaps. One more advantage is there won’t be any proxies. When the process is completed, the teacher must click on mail where the database of attendance will be sent to the faculty, and the students who are absent the respective parent will receive the message.

2 Problem Definition Though many systems are developed for face recognition, but still, it has many challenges like in computer vision and recognition of pattern researches like features are computed slowly, it needs auxiliary information and for broad practical applications. The problem of face recognition can be stated as follows: face recognition human facial features like the mouth, nose, and eyes in a full-frontal face image.

3 Proposed System The process is done by two algorithms which is haar cascade, used for face detection and lPBH, used for face recognition. The application is created using Python module in graphical user interface. Tkinter is the package used which makes it easier in labeling and other things. The process follows totally 3 steps first by taking image of the students along with their details and storing them in the database and training those images in the database finally testing them, so when the student comes in front of the camera, it tries to recognize them by comparing them from the database which as the faces already trained and tested and enter them into the attendance sheet. If the person doesn’t belong to the database, it will be stored in the unknown image file. The stages are clearly explained in the further sections.

4 Dataset and Methodology Face recognition-based attendance system: The graphical user interphase of the system has been shown in Fig. 1. The system has been divided into 3 sections: creation of database, dataset training and dataset testing, sending alert messages in an extension. (1) Creation of Database • Initialize the camera and to get the attention of student’s alert message is set

Attendance System Based on Face Recognition Using …

11

Fig. 1 System GUI

• • • (2) • • • (3)

Input of user id Conversion of the image from RGB to gray scale, face detection and It will be stored in the database by labeling them up to 100 Dataset Training Recognizer of LBPH face will be initialized Next, faces and ID from the folder are taken to train the LBPH recognizer The trained data will be saved as xml or yml file Dataset Testing

Loading from xml and yml of haar classifier, LBPH face recognizer and data that has been trained is done. • • • •

the image will be captured from the camera, next convert them into gray scale, the face will be detected from it, Finally, prediction of the face is done by using above recognizer.

The system in this uses haar cascade algorithm in detection of face which in turn uses modified haar cascades in detection. We can use any camera like USB Webcam to capture photos. The system will access console of SSH in laptop. Most important of all, it will require a lot of positive and negative images in training the haar cascades. In which, the positive images are pseudocode for the attendance system. Input: Real-time video of the student’s face will be taken output: Excel sheet of the student’s attendance. (1) Each frame of the image will be transferred from RGB to gray scale

12

A. K. Rai et al.

(2) Next, apply haar cascade classifier for face detection and get the region of interest (3) Now, we need to apply the LBPH algorithm on the region of interest to get the features (4) If it is for enrollment, then features will be stored in the database else if it is for verification then do the post-processing Haar Cascade Algorithm Haar cascade is a machine learning algorithm used for detection of objects in an image or video. It is a process where a cascade function is used to train a lot of positive and negative images. It is used for detection of objects in images. It will also be able to identify any kind of object. As per Fig. 2, the first step is to collect all the haar features. It selects those features which are adjacent to the rectangular regions at a specific location in the detection window. Then, it sums up all the pixel intensities in each of the region and calculates the difference between these sums. Loading the dataset We will plot the first image in our dataset and check its size. By default, the shape of every image in the dataset is 141 × 141, so we will not need to check the shape of all the images. Sometimes, it depends on the size of the face and distance from the camera. Data preprocessing We need to have a target variable. That means the column will get created for each output, and variable will be assigned to each of them; the first layer will be also be taken in an input shape. This is the shape of each input image, 141, 141, 1 as seen earlier on, with the 1 signifying that the images are gray scale.

Fig. 2 Haar cascade algorithm

Attendance System Based on Face Recognition Using …

13

Training the model Now, it will train the model. For training, we will use the ‘fit()’ function in the model with the following parameters: Data training will be taken as x; target of data will be taken as y and validation data, and all the number of epochs. For the data validation, we going to use the test set provided in the dataset, which we have split into x and y. Now, this will be using the model to make predictions. Local binary patterns histogram The goal of face detection is to detect and locate faces from the image, to extract and to use the in other areas. Also, nowadays, there are many different algorithms to accomplish this face detection and recognition, such as Eigen faces, fisher faces, scale-invariant feature transform and speed up robust features. In this section, we going to use LBPH-based face detection algorithm as depicted in Fig. 3. LBP is an easy but powerful way to extract all features and label those pixels from the image. The algorithm used to follow this step Initializing them temp = 0 is the first step where I those are the training images where H = 0, then initialize them to the pattern histogram The model of the label of LPBH is calculated Keep adding the corresponding bin by one Next step is to get the greatest LBP feature during each of the face image and later merging them into the unique vector (7) Then, it is time to compare all those features. (8) Last step, finally, when it resembles with the image that is stored in the database, the image will be recognized (1) (2) (3) (4) (5) (6)

5 Results and Outcome In the experiment, first, we need to create a dataset of each student which will have a distinct ID number with an image of a face in the database, and then, those features of the face from all the image will be extracted At the end, it gets segregated, and recognition of those face of each student along with information will be done (Figs. 4 and 5). Now, it is required to compare the input image of the face which has been detected and extract all the facial features and compare them with the face images in the database; if failed to recognize, they will be stored in the unknown images as shown in Fig. 6. Using the algorithms, the details of all the faces of each student known and unknown will be compared that have been trained in the system. In this research, three major tasks have been performed those are face detection, train, and finally recognize the face of each student from the real-time video camera. A. Face detecting and preprocessing

14

A. K. Rai et al.

Fig. 3 LBPH algorithm flowchart

In this step, the system will detect the face as image as the input with real-time camera with less resolution or high resolution depending taken individually or in group. Firstly, we need to convert all those frames from RGB to gray scale to do the detection of the faces. We will apply haar cascade function that has been trained and detect those features in other images. In the system, we will use haar features like edges, lines, and four rectangles that are nose, mouth, and eyes. For those large or huge image of a variable size of an image, it will take or need a lot of computations, and all those features most of the time will be irrelevant. But, by using AdaBoost,

Attendance System Based on Face Recognition Using …

15

Fig. 4 Data flow diagram of face recognition

Fig. 5 Sequence diagram of face recognition

we will select the best among those. Then, region of interest like which will contain faces is extracted and will be sent to the next stage as shown in Fig. 7. B. Images of face is trained First, the image is captured, and we will be preprocessing them (Fig. 8); later, we need to train those images in the dataset. In the training phase, those images of recognition will be applied that will get stored of the values of histogram of images of the face (Fig. 9). C. Post-processing and recognizing image of the face

16

Fig. 6 Unknown person

Fig. 7 Detection of face

A. K. Rai et al.

Attendance System Based on Face Recognition Using …

17

Fig. 8 Extraction and preprocessed faces of those students stored in the dataset

Fig. 9 Dataset training

The last and final task is to recognize the image of the face. The cascaded haar classifier and training them to recognize and training recognition will be used for the recognition of the face. Then, at the end, classifier will be comparing those stored images of face with the input images of face, and if all the features of input images of face get matched with in the images stored in the database, the recognition of face result gets displayed with real-time video camera of the screen with name and university seat number (Figs. 10 and 11).

18

A. K. Rai et al.

Fig. 10 Recognizing face images

Fig. 11 Attendance sheet absent list of students

We can also get an absent list on teacher’s mobile number as well as on parent’s mail ID by choosing following options (Figs. 12 and 13). Teacher can also get an absent students list on their mobile number for that same day as well as previous day.

Attendance System Based on Face Recognition Using …

Fig. 12 Absent list for parents

Fig. 13 Absent for teachers

19

20

A. K. Rai et al.

6 Conclusion Compared to all other algorithms, this one is much faster and has less false rate. This application is very much useful in schools and colleges since it saves time of teacher which will be wasted in talking attendance manually; thus, even preventing from proxies and even parents can have a track of their child attending each classes or not. Since messages will be automatically generated and passed to parents if the child doesn’t come under the database of that class and teachers or faculty will have the complete details of students who are absent with a Excel sheet to their mails which has been stored in the database already. The great advantage of this application is, it will be able to recognize the face of the students even when there is a slight changes like sometimes students wear glasses or arrive without them and if they have beard or not also it can recognize them easily. When the students come in front of the camera, they will be recognized with the images that have been trained; even when it fails to recognize them, it will be stored in unknown images which later can be used to check for errors and managed attendance according to it.

References 1. Joseph J, Zacharia KP (2013) Automatic attendance management system using face recognition. Int J Sci Res (IJSR) 2. Chiagozie OG, Nwaji OG (2012) Radio- frequency identification (RFID) based attendance system with automatic door unit. Acad Res Int 3. Shoewu O, Idowu O.A (2012) Development of attendance management system using biometrics. Pac J Sci Technol 4. Khatun A, Haque AKMF, Ahmed S, Rahman MM (2015) Design and implementation of iris recognition based attendance management system. In: 2015 International conference on electrical engineering and information communication technology (ICEEICT), Dhaka, 2015, pp 1–6 5. Ahonen T, Hadid A, Pietik¨ainen M (2004) Face recognition with local binary patterns. In: Proceedings of the advances in visual computing; springer science and business media LLC, vol 3021. Berlin, Germany, pp 469–481 6. O¨ zdil, O¨ zbilen MM (2014) A survey on comparison of face recognition algorithms. In: 2014 IEEE 8th international conference on application of information and communication technologies (AICT), Astana, 2014, pp 1–3 7. Ahonen T, Hadid A, Pietikainen M (2006) Face description with local binary patterns: application to face recognition. IEEE Trans Pattern Anal Mach Intell 28(12):2037–2041 8. Viola P, Jones M (2001) Rapid object detection using a boosted cascade of simple features. In: Proceedings of the 2001 IEEE computer society conference on computer vision and pattern recognition. CVPR 2001, Kauai, HI, USA, 2001, pp I–I 9. Will Berger, Deep Learning Haar Cascade Explained, WILL BERGER. http://www.willberger. org/cascade-haar-explained 10. Kelvin Salton do Prado, Face Recognition: Understanding LBPH Algorithm, towards data science. https://towardsdatascience.com/facerecognition-how-lbph-works-90ec258c3d6b 11. Samet R, Tanriverdi M (2017) Face recognition- based mobile automatic classroom attendance management system. In: 2017 international conference on cyberworlds (CW), Chester, 2017, pp 253–256

Attendance System Based on Face Recognition Using …

21

12. Seifedine K, Mohamad S (2010) Wireless attendance management system based on iris recognition. Sci Res Essays 5:1428–1435 13. Ahmed A, Guo J, Ali F, Deeba F, Ahmed A (2018) LBPH based improved face recognition at low resolution. In: 2018 International conference on artificial intelligence and big data (ICAIBD), 2018: IEEE, pp 144–147

Different Thresholding Techniques in Image Processing : A Review Radha Seelaboyina and Rajeev Vishwakarma

Abstract Document data is captured through optical scanning or digital video, resulting in a file of picture elements, or pixels, which serves as the raw input for document analysis. These pixels are samples of intensity values taken in a grid pattern throughout the document page, with intensity values ranging from OFF (0) to ON (1) for binary pictures, 0–255 for gray-scale images, and 3 channels of 0–255 color values for color images. The initial stage in document analysis is to process this image so that it may be analyzed further. Thresholding is used to convert a gray-scale or color image to a binary image, and noise reduction is used to remove superfluous data. The goal of this paper is to summarize some thresholding technique for image processing. Keywords Thresholding · Global thresholding · Binarizations · Adaptive thresholding · Intensity histogram

1 Introduction In this treatment of document processing, we deal with images containing text and graphics of binary information [1]. That is, these images contain a single foreground level that is the text and graphics of interest, and a single background level upon which the foreground contrasts. We will also call the foreground: objects, regions of interest, or components. The documents may also contain true gray-scale (or color) information, such as in photographic figures; however, besides recognizing the presence of a gray-scale picture in a document, we leave the analysis of pictures to the more general fields of image analysis and machine vision. Though the information is binary, the data—in the form of pixels with intensity values—are not likely to have only two levels, but instead a range of intensities [2]. This may be due to non-uniform printing or non-uniform reflectance from the page, or a result of intensity transitions at the region edges that are located between foreground and background regions. The R. Seelaboyina (B) · R. Vishwakarma Dr. A. P. J. Abdul Kalam University, Indore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_3

23

24

R. Seelaboyina and R. Vishwakarma

objective in binarization is to mark pixels that belong to true foreground regions with a single intensity (ON) and background regions with a different intensity (OFF). Figure illustrates the results of binarizing a document image at different threshold values. The ON-values are shown in black in our figures, and the OFF values are white [3]. For documents with a good contrast of components against a uniform background, binary scanners are available that combine digitization with thresholding to yield binary data. However, for the many documents that have a wide range of background and object intensities, this fixed threshold level often does not yield images with clear separation between the foreground components and background. For instance, when a document is printed on differently colored paper or when the foreground components are faded due to photocopying, or when different scanners have different light levels, the best threshold value will also be different. For these cases, there are two alternatives [4]. One is to empirically determine the best binarization setting on the scanner (most binary scanners provide this adjustment), and to do this, each time an image is poorly binarized. The other alternative is to start with gray-scale images (having a range of intensities, usually from 0 to 255) from the digitization stage, then use methods for automatic threshold determination to better perform binarization. While the latter alternative requires more input data and processing, the advantage is that a good threshold level can be found automatically, ensuring consistently good images, and precluding the need for time-consuming manual adjustment and repeated digitization.

2 Thresholding The following discussion presumes initial digitization to gray-scale images. If the pixel values of the components and those of the background are fairly consistent in their respective values over the entire image, then a single threshold value can be found for the image [5]. This use of a single threshold for all image pixels is called global thresholding. Processing methods are described below that automatically determine the best global threshold value for different images. For many documents, however, a single global threshold value cannot be used even for a single image due to non-uniformities within foreground and background regions [6]. For example, for a document containing white background areas as well as highlighted areas of a different background color, the best thresholds will change by area. For this type of image, different threshold values are required for different local areas; this is adaptive thresholding.

2.1 Global Thresholding The most straightforward way to automatically select a global threshold is by use of a histogram of the pixel intensities in the image. The intensity histogram plots the number of pixels with values at each intensity level. For a histogram of a document image. For an image with well-differentiated foreground and background intensities,

Different Thresholding Techniques in Image Processing : A Review

25

the histogram will have two distinct peaks [7]. The valley between these peaks can be found as the minimum between two maxima and the intensity value there is chosen as the threshold that best separates the two peaks.

2.1.1

Drawbacks

There are a number of drawbacks to global threshold selection based on the shape of the intensity distribution. The first is that images do not always contain welldifferentiated foreground and background intensities due to poor contrast and noise. A second is that, especially for an image of sparse foreground components, such as for most graphics images, the peak representing the foreground will be much smaller than the peak of the background intensities. This often makes it difficult to find the valley between the two peaks. In addition, reliable peak and valley detection are separate problems unto themselves.

2.1.2

Solution

One way to improve this approach is to compile a histogram of pixel intensities that are weighted by the inverse of their edge strength values. Region pixels with low edge values will be weighted more highly than boundary and noise pixels with higher edge values, thus sharpening the histogram peaks due to these regions and facilitating threshold detection between them [8]. Second, an analogous technique is to highly weight intensities of pixels with high edge values, then choose the threshold at the peak of this histogram, corresponding to the transition between regions. This requires peak detection of a single maximum, and this is often easier than valley detection between two peaks. This approach also reduces the problem of large size discrepancy between foreground and background region peaks because edge pixels are accumulated on the histogram instead of region pixels; the difference between a small and large size area is a linear quantity for edges versus a much larger squared quantity for regions. A third method uses a Laplacian weighting. The Laplacian is the second derivative operator, which highly weights transitions from regions into edges (the first derivative highly weights edges). This will highly weight the border pixels of both foreground regions and their surrounding backgrounds, and because of this, the histogram will have two peaks of similar area. Though these histogram shape techniques offer the advantage that peak and valley detection are intuitive, still peak detection is susceptible to error due to noise and poorly separated regions. Furthermore, when the foreground or background region consists of many narrow regions, such as for text, edge, and Laplacian measurement may be poor due to very abrupt transitions (narrow edges) between foreground and background.

26

R. Seelaboyina and R. Vishwakarma

2.2 Multi-thresholding A number of techniques determine classes by formal pattern recognition techniques that optimize some measure of separation. One approach is minimum error thresholding. Here, the foreground and background intensity distributions are modeled as normal (Gaussian or bell-shaped) probability density functions. For each intensity value (from 0 to 255, or a smaller range if the threshold is known to be limited to it), the means and variances are calculated for the foreground and background classes, and the threshold is chosen such that the misclassification error between the two classes is minimized [9]. This latter method is classified as a parametric technique because of the assumption that the gray-scale distribution can be modeled as a probability density function. This is a popular method for many computer vision applications, but some experiments indicate that documents do not adhere well to this model, and thus results with this method are poorer than nonparametric approaches. One nonparametric approach is Otsu’s method. Calculations are first made of the ratio of between-class variance to within-class variance for each potential threshold value. The classes here are the foreground and background pixels, and the purpose is to find the threshold that maximizes the variance of intensities between the two classes, and minimizes them within each class. This ratio is calculated for all potential threshold levels and the level at which the ratio is maximum is the chosen threshold. A similar approach to Otsu’s employs an information theory measure, entropy, which is a measure of the information in the image expressed as the average number of bits required to represent the information. Here, the entropy for the two classes is calculated for each potential threshold, and the threshold where the sum of the two entropies is largest is chosen as the best threshold. Another thresholding approach is by moment preservation. This is less popular than the methods above, however, we have found it to be more effective in binarizing document images containing text. For this method, a threshold is chosen that best preserves moment statistics in the resulting binary image as compared with the initial gray-scale image. These moments are calculated from the intensity histogram—the first four moments are required for binarization. Many images have more than just two levels. For instance, magazines often employ boxes to highlight text where the background of the box has a different color than the white background of the page. In this case, the image has three levels: background, foreground text, and background of highlight box. To properly threshold an image of this type, multi-thresholding must be performed. There are many fewer multi-thresholding methods than binarization methods. Most require that the number of levels is known. For the cases where the number of levels is not known beforehand, one method will determine the number of levels automatically and perform appropriate thresholding. This added level of flexibility may sometimes lead to unexpected results. For instance, a magazine cover with three intensity levels may be thresholded to four levels instead due to the presence of an address label that is thresholded at a separate level.

Different Thresholding Techniques in Image Processing : A Review

27

2.3 Adaptive Thresholding A common way to perform adaptive thresholding is by analyzing gray-level intensities within local windows across the image to determine local thresholds. White and Rohrer describe an adaptive thresholding algorithm for separating characters from background. The threshold is continuously changed through the image by estimating the background level as a two-dimensional running-average of local pixel values taken for all pixels in the image [8]. Mitchell and Gillies describe a similar thresholding method where background white level normalization is first done by estimating the white level and subtracting this level from the raw image. Then, segmentation of characters is accomplished by applying a range of thresholds and selecting the resulting image with the least noise content. Noise content is measured as the sum of areas occupied by components that are smaller and thinner than empirically determined parameters. Looking back at the results of binarization for different thresholds in it can be seen that the best threshold selection yields the least visible noise [10]. The main problem with any adaptive binarization technique is the choice of window size. The chosen window size should be large enough to guarantee that a large enough number of background pixels are included to obtain a good estimate of average value, but not so large as to average over non-uniform background intensities. However, often the features in the image vary in size such that there are problems with fixed window size. To remedy this, domain dependent information can be used to check that the results of binarization give the expected features (a large blob of an ONvalued region is not expected in a page of smaller symbols, for instance). If the result is unexpected, then the window size can be modified and binarization applied again.

3 Choosing a Thresholding Method Whether global or adaptive thresholding methods are used for binarization, one can never expect perfect results. Depending on the quality of the original, there may be gaps in lines, ragged edges on region boundaries, and extraneous pixel regions of ON and OFF values [11]. This fact that processing results will not be perfect is generally true with other document processing methods, and indeed image processing in general. The recommended procedure is to process as well as possible at each step of processing, but to defer decisions that do not have to be made until later steps to avoid making irreparable errors. In later steps, there is more information as a result of processing to that point, and this provides greater context and higher level descriptions to aid in making correct decisions, and ultimately recognition. Deferment, when possible, is a principle appropriate for all stages of document analysis. A number of different thresholding methods have been presented in this section. It is the case that no single method is best for all image types and applications. For simpler problems where the image characteristics do not vary much within the image or across different images, then the simpler methods will suffice [7]. For more difficult problems of

28

R. Seelaboyina and R. Vishwakarma

noise or varying image characteristics, more complex (and time-consuming) methods will usually be required. Commercial products vary in their thresholding capabilities. Today’s scanners usually perform binarization with respect to a fixed threshold. More sophisticated document systems provide manual or automatic histogram-based techniques for global thresholding [12]. The most common use of adaptive thresholding is in special purpose systems used by banks to image checks. The best way to choose a method at this time is first by narrowing the choices by the method descriptions, then just experimenting with the different methods and examining their results. Because there is no “best” thresholding method, there is still room for future research here. One problem that requires more work is to identify thresholding methods or approaches that best work on documents with particular characteristics [13]. Many of the methods described above were not formulated in particular for documents, and their performance on them is not well known. Documents have characteristics, such as very thin lines that will favor one method above another.

4 Conclusion In this paper, we discussed different techniques of thresholding as well as how best to quantify the results of thresholding. For text, one way is to perform optical character recognition on the binarized results and measure the recognition rate for different thresholds. We also discussed problem that requires further work is that of multithresholding. While multi-thresholding capabilities have been claimed for some of the methods discussed above, not much dedicated work has been focused on this problem. For other reviews and more complete comparisons of thresholding methods on global and multi-thresholding techniques, and on adaptive techniques, we suggest just manually setting a threshold when the documents are similar and testing is performed beforehand. For automatic, global threshold determination, we have found that the moment-preserving method works well on documents. For adaptive thresholding, the method of is a good choice. This paper also gives background and comparison on these adaptive methods. For multi-thresholding, the method is appropriate if the number of thresholds is known, and the method if not.

References 1. Kaur N, Kaur R (2011) A review on various methods of image thresholding. Int J Comput Sci Eng 3(10):3441 2. Kaur L, Gupta S, Chauhan RC (2002) Image denoising using wavelet thresholding. In: ICVGIP, vol 2, pp 16–18 3. Gnanadurai D, Sadasivam V (2008) An efficient adaptive thresholding technique for wavelet based image denoising. Int J Electron Commun Eng 2(8):1703–1708 4. Jansen M, Bultheel A (2001) Empirical Bayes approach to improve wavelet thresholding for image noise reduction. J Am Stat Assoc 96(454):629–639

Different Thresholding Techniques in Image Processing : A Review

29

5. Nakagawa Y, Rosenfeld A (1979) Some experiments on variable thresholding. Pattern Recogn 11(3):191–204 6. Roy P, Dutta S, Dey N, Dey G, Chakraborty S, Ray R (2014) Adaptive thresholding: a comparative study. In: 2014 International conference on control, instrumentation, communication and computational technologies (ICCICCT). IEEE, pp 1182–1186 7. Imocha Singh O, Sinam T, James O, Romen Singh T (2012) Local contrast and mean thresholding in image binarization. Int J Comput Appl 51(6) 8. Wellner PD (1993) Adaptive thresholding for the digitaldesk. Xerox, EPC1993-110, pp 1–19 9. Zhong S, Cherkassky V (2000) Image denoising using wavelet thresholding and model selection. In: Proceedings 2000 International conference on image processing (Cat. No. 00CH37101), vol 3. IEEE, pp 262–265 10. Bataineh B, Sheikh Abdullah SNH, Omar K (2011) An adaptive local binarization method for document images based on a novel thresholding method and dynamic windows. Pattern Recogn Lett, 32(14):1805–1813 11. Su B, Lu S, Tan CW (2011) Combination of document image binarization techniques. In: 2011 International conference on document analysis and recognition. IEEE, pp 22–26 12. Ramesh N, Yoo J-H, Sethi IK (1995) Thresholding based on histogram approximation. IEE Proc Vision Image Signal Process 142(5):271–279 13. Nacereddine N, Hamami L, Tridi M, Oucief N (2005) Non-parametric histogram-based thresholding methods for weld defect detection in radiography. In: Transactions on Enformatika, systems sciences and engineering. Citeseer

Dynamic Weighting Selection for Predictive Torque and Flux Control of Industrial Drives Vishnu Prasad Muddineni, Anil Kumar Bonala, and Thanuja Penthala

Abstract Finite Set-Model Predictive Control (FS-MPC) based torque and flux control (popularly known as PTC) is one of the advanced control techniques used for the adjustable speed operation of industrial drives. This control technique offers fast dynamic response and retains the decoupled nature present in the conventional counterpart. One of the main challenges of this control approach is to select the suitable weighting factor (WF) for the respective objectives in cost function. In general, a heuristically selected WF is assigned to the flux objective in the cost function. The selected WF is constant irrespective of the drive operating conditions. In this paper, dynamically weighted objectives with modified cost function are introduced to overcome the above problem. To demonstrate the efficacy of proposed dynamic weighting approach, simulation results are illustrated for an induction motor (IM) drive with different operating conditions. Keywords Speed control · Induction motor · Predictive control · Torque and flux control · Industrial drive · Weighting factor

1 Introduction PTC is the model-based predictive control technique with finite number of control actions [1, 2]. The basic idea of this control technique is based on the well-established conventional DTC technique used for adjustable speed operation in industrial drive applications. The nonlinear hysteresis controllers with predefined state selection in DTC are replaced with model predictive approach in PTC. The major advantages of PTC over the conventional DTC are; all the admissible voltage vectors are evaluated for the selection of suitable control action and flexibility of adding additional control objectives and nonlinearities into the cost function [2]. The control objectives in the PTC are stator flux and torque and all the voltage vectors corresponding to the 2-L VSI are considered as feasible control actions. The V. P. Muddineni (B) · A. K. Bonala · T. Penthala Department of Electrical and Electronics Engineering, Vaagdevi College of Engineering, Telangana 506005, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_4

31

32

V. P. Muddineni et al.

control objectives are predicted for the subsequent sampling period for all the admissible voltage vectors. The cost function is defined with predicted control variables and the corresponding references. The voltage vector for the subsequent sampling period will be selected by optimizing the cost-function [2]. A WF is assigned to flux objective to maintain relative importance between the control objectives. Flux WF is the only parameter to be tuned for the implementation of PTC. However, the tuning of this WF is not straight forward. Several methods are presented for selection of WFs [2–6]. In [2], equal importance is given to both the control objectives in cost function. This selection is considered as conventional PTC in this paper. In [5], WF is designed using artificial neural network approach for FCS-MPC of power electronic applications. In [3, 4], multi-criteria decision-making methods, viz., TOPSIS and GRA methods are employed to simplify selection of WFs in cost-function of PTC for IM drives. However, the selected WFs are constant irrespective of drive operating conditions. To overcome these problems a simple method for dynamic WF selection is presented in this paper.

2 Predictive Torque and Flux Control Conventional PTC control for IM drive is presented in this section and the corresponding block diagram is shown in Fig. 1. The required mathematical models of the converter and IM for estimation and prediction of the control technique are given below. The output voltage vector of a 2-level voltage source inverter can be expressed as u=

Fig. 1 PTC for IM drive [7]

  2 VDC Sa + aSb + a 2 Sc 3

(1)

Dynamic Weighting Selection for Predictive Torque and Flux …

33

Fig. 2 Voltage vector location of a 2-level VSI

where a = e2π/3 , a total of eight voltage vectors are available with 2-level voltage source inverter. Out of which, six are active vectors and two are null vectors. The voltage vectors arrangement in complex plane and their switching states are shown in Fig. 2. The model of an IM with Ψ s and is as state variables can be expressed by using following relations X˙ = AX + BU

(2)

where U = [u s 0]T  A=

0 −Rs λ(Rr − j L r ωr ) −λ(Rs L r + Rr L s ) + j ωr



T  B = 1 λL r λ=

1 L s L r − L 2m

Electromagnetic torque and rotor mechanical speed of the IM can be expressed as Te = J

  3 pIm ψs∗ i s 2

dωr = Te − Tl dt

(3) (4)

34

V. P. Muddineni et al.

Initially, stator flux is estimated by using the measured quantities and discrete time model of the drive as given in the Eq. (5).   ψsk = ψsk−1 + Ts −Rs i sk−1 + u k−1 s

(5)

The predictions of Ψ s , is and T e for the next sampling period can be performed with all the available voltage vectors of the inverter. 

ψsk+1

 m

    = ψsk + Ts A12 i sk + Ts B11 u ks m

    fixed term

variable term

     k+1  i s m = i sk + Ts A21 ψsk + A22 i sk + Ts B21 u ks m

    fixed term



 k+1

Te

m

(6)

(7)

variable term

  3 = pIm ψsk+1∗ i sk+1 m 2

(8)

The terms A12 , A21 , A22 , B11 , and B12 in the above representation are taken from Eq. (2), and m represents the set of all admissible voltage vectors. Finally, cost function is optimized to select the control action for the subsequent sampling period and they are given as   Jm = Teref − Tek+1 m + λ ψsref − ψsk+1 m u opt = arg

min

(u 0 ,u 1 ,...u 7 )

J

(9) (10)

To maintain the trade-off between T e and Ψ s in the cost-function, a WF λ is employed to the stator flux term. In this paper, equal importance is given to both the control objectives by selecting the λ value by using the relation given in Eq. (11) and this is referred as conventional PTC in this paper. T nom λ = enom ψ

(11)

s

3 Proposed Dynamic Weighting Factor Selection The WF obtained in Eq. (11) is constant for all the drive operating conditions. To address this problem, a dynamic WF is obtained in every sampling period by using the mean error of individual control objectives. The implementation flowchart of proposed method is depicted in Fig. 3. The following procedure is adopted to obtain the appropriate WFs in each sampling period.

Dynamic Weighting Selection for Predictive Torque and Flux … Fig. 3 Implementation flow chart for proposed method

35

36

V. P. Muddineni et al.

Initially, individual cost function will be defined for each control objective and they are given as   (JT )m = Teref − Tek+1 m 

Jψ

 m

= ψsref − ψsk+1 m

(12) (13)

The obtained error values for all the admissible control actions will be arranged as a dataset and it given as ⎡

JT 1 ⎢ JT 2 ⎢ Xi j = ⎢ . ⎣ ..

⎤ Jψ1 Jψ2 ⎥ ⎥ .. ⎥ . ⎦

(14)

JT 7 Jψ7 In the above relation, i represents the control objectives and j represents the number of admissible control actions. The above dataset is to be normalized to a scale of 0 to 1 by using the below fuzzy normalization Yi T =

JTmax − Ji T JTmax − JTmin

Yiψ =

Jψmax − Ji ψ Jψmax − Jψmin

(15) (16)

where J T max and J T min are the maximum and minimum values of torque, J y max and J y min are the maximum and minimum values stator flux obtained from the dataset given in Eq. (14). The mean value of the normalized control objectives is obtained from the following relation σT =

m  Yi T m i=1,2..

(17)

σψ =

m  Yiψ m i=1,2..

(18)

Now, the WFs will be obtained for both control objectives by the following relations wT =

σT σT + σψ

(19)

wψ =

σψ σT + σψ

(20)

Dynamic Weighting Selection for Predictive Torque and Flux …

37

Finally, a single performance index is formulated which is similar to cost function in the conventional approach and it is given as Ci = wT Yi T + wψ Yi ψ

(21)

The optimal vector for next sampling period is obtained by using the following relation u opt = arg

max Ci

(i=1,2...m)

(22)

4 Simulation Results Simulation results for the speed control of IM drive with different operating conditions are being tested for both the conventional and proposed control approaches are presented in this section. Results are illustrated for both the steady and dynamic operating conditions of the drive. The steady-state operations of drives with conventional PTC for different WFs are shown in Figs. 4, 5, 6, 7 and 8. In these results, drive is operating with 8 N-m load at rated speed. For the results presented in Fig. 4, WF is selected by using Eq. (11) by giving the equal importance to both torque and flux. From this result, it is evident that more deviations in the stator flux and distorted current. To improve the response of stator flux, different WFs are assigned to flux error term. Figure 5 shows the results for λ = 25, Fig. 6 shows the results for λ = 50, Fig. 7 shows results for λ = 75 and Fig. 8 shows results for λ = 90. From all these results, WF λ = 75 gives relatively better torque, flux, and current profiles. Figure 9 shows the results with proposed method for same operating conditions. The selection of dynamic weights for torque and flux is given in Fig. 10. Based on the torque and flux errors, weights are dynamically calculated in each sampling period to provide the better possible torque and flux responses. Also, it can be observed the combined weight is always equal to one. The proposed method offers similar response to PTC with λ = 75. Figures 11 and 12 show the dynamic load response of the drive with PTC for λ = 75 and proposed method, respectively. Figures 13 and 14 show the speed reversal of the drive with PTC for λ = 75 and proposed method, respectively. From these results, it is evident that both the cases offer similar dynamic response.

38 Fig. 4 Steady state of the drive with PTC for λ = 14.66

Fig. 5 Steady state of the drive with PTC for λ = 25

V. P. Muddineni et al.

Dynamic Weighting Selection for Predictive Torque and Flux … Fig. 6 Steady state of the drive with PTC for λ = 50

Fig. 7 Steady state of the drive with PTC for λ = 75

39

40 Fig. 8 Steady state of the drive with PTC for λ = 90

Fig. 9 Steady state of the drive with proposed PTC

V. P. Muddineni et al.

Dynamic Weighting Selection for Predictive Torque and Flux … Fig. 10 WF selection with proposed PTC under steady state operation

Fig. 11 Dynamic response of PTC with λ = 75

41

42 Fig. 12 Dynamic response of proposed PTC

Fig. 13 Speed reversal of PTC with λ = 75

V. P. Muddineni et al.

Dynamic Weighting Selection for Predictive Torque and Flux …

43

Fig. 14 Speed reversal of proposed PTC

5 Conclusion This paper presents a dynamically selected WF method for PTC of IM drive. The proposed method eliminates the heuristically selected WF and updates the weights in every sampling period dynamically depending on the operating conditions of the drive. The proposed method offers similar response under different operating conditions with respect to the heuristically selected WF. However, the major problem with proposed method is the additional computational burden added to the control algorithm. This problem can be easily overcome using modern day processors. Finally, this method can be extended to other FS-MPC algorithms as a generalized approach for WF selection.

References 1. Miranda H, Cortés P, Yuz JI, Rodríguez J (2009) Predictive torque control of induction machines based on state-space models. IEEE Trans Ind Electron 56:1916–1924 2. Rodríguez J, Kennel RM, Espinoza JR, Trincado M, Silva CA, Rojas CA (2012) Highperformance control strategies for electrical drives: an experimental assessment. IEEE Trans Ind Electron 59:812–820 3. Muddineni VP, Sandepudi SR, Bonala AK (2017) Finite control set predictive torque control for induction motor drive with simplified weighting factor selection using TOPSIS method. IET Electr Power Appl 11:749–760 4. Muddineni VP, Bonala AK, Sandepudi SR (2021) Grey relational analysis-based objective function optimization for predictive torque control of induction machine. IEEE Trans Ind Appl 57:835–844 5. Dragiˇcevi´c T, Novak M (2019) Weighting Factor Design in model predictive control of power electronic converters: an artificial neural network approach. IEEE Trans Ind Electron 66:8870– 8880

44

V. P. Muddineni et al.

6. Arshad MH, Abido MA, Salem A, Elsayed AH (2019) Weighting factors optimization of model predictive torque control of induction motor using NSGA-II with TOPSIS decision making. IEEE Access. 7:177595–177606 7. Rodriguez J, Cortes P (2012) Predictive control of power converters and electrical drives. Wiley

Population Index and Analysis Based on Different Geographies; Using Distance Measurement, Social Distancing, and Deep Learning Bhushan Chougule, Samiksha Baral, Minal Tayde, and Kaustubh Sakhare

Abstract Nowadays, every individual is familiar with the COVID-19 pandemic which has caused great turmoil in everyone’s life. Also, they are aware that there is no medicine or drug to cure COVID immediately, and people are at the risk of losing their lives. Lack of vaccines or delay in vaccine production for mass results social distancing being the only measure to tackle this pandemic. As a result, social distancing has proven to be a very reliable and efficient way to diminish the growth of this disease; the reason why lockdowns are imposed, and people are asked to keep some distance from each other, for their safety as there will be minimal physical contact. Machine learning and artificial intelligence come into the picture in every solution to a generic problem the community faces nowadays like in medical, supply chain management, face detection, etc. Using the power of AI algorithms, the paper aims to develop a robust system to monitor and analyze social distance measurement protocols at public places during the COVID-19 pandemic with the help of CCTV feed and check whether they abide by the safety protocols or not by measuring the distance between them. The proposed approach is implemented to enumerate the number of violations at a popular public place to prevent massive crowds at particular periods. The proposed method is suitable to construct a scrutiny system at a public place to alert people and eschew mass gatherings that can be concluded using achieved results. The paper also has an analysis of the performance of different models of R-CNN, Fast R-CNN, and YOLO. YOLO architectures are validated based on object detection and object tracking rate in real time. Keywords R-CNN · Fast R-CNN · YOLO · Social distancing · COVID-19 · Distance measurement · Object detection · Object tracking

B. Chougule (B) · S. Baral · M. Tayde · K. Sakhare SCTR’s Pune Institute of Computer Technology, Dhankawadi, Pune, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_5

45

46

B. Chougule et al.

1 Introduction To control the spread of COVID, “social distancing” is without a doubt the best measure to do unless each person gets fully vaccinated. Social distancing is observed as an unprecedented measure by all the health agencies. That is the reason, countries across the world had to bring nationwide lockdowns for months which surely slowed and diminished the growth of the virus in public. Taking the example of the first lockdown in China’s Hubei province in January 2020, the confirmed cases were doubling up each day in February by forcing a compulsive lockdown, and then for 5 consecutive days in March, they did not get a single new confirmed case. This proves that social distancing when enacted in China reduced the cases; hence, this measure was adopted worldwide.

1.1 How is Social Distancing Efficient? The COVID-19 virus is spreading easily with physical contact among people. According to the World Health Organization (WHO), “COVID-19 is transmitted via droplets and fomites during close unprotected contact between an infector and infectee”. Fomite is the concept of an object being probable to transmit the virus due to contact with it. Fomite may include utensils, furniture, food, clothes, etc. Thus, keeping a social distance from people and fomites by avoiding proximity is the main step to avoid COVID-19.

1.2 Social Distancing A minimum of 2 m distance from another individual is recommended by experts. Social distancing is a way that does not use any medicines or antibiotics and is still the strongest way to counter COVID-19. The long-term goal is to bring back normality in our way of life and enjoy our freedom. Social distancing protocol can be ensured in public places and the workplace. To do that, the research paper implements human detection using advanced AI algorithms and calculates the distance between these people in real time from live camera footage to check whether people are abiding by safety protocols or not. The work carried can be used at various places like workstations, offices, schools, academies, hospitals, restaurants, etc., thus building a safe environment by managing efficiently. The different ways in which social distancing can be achieved has been illustrated in Fig. 1.

Population Index and Analysis Based on Different …

47

Fig. 1 Various ways to achieve social distancing

2 Research Background and Relevant Work The research paper presents solutions to avoid social gatherings for saving the lives of people at public places by utilizing real-time pedestrian detection and distance measurement methods to calculate the number of violations at the time. The authors are willing to suggest that the reason behind increasing the infection and spread of the virus among people is avoiding social distancing norms. Figure 2 consolidates the need of pedestrian detection and distance measurement, object detection, and social distancing as the main factors of the model needed to avoid the spread of COVID-19.

2.1 Pedestrian Detection and Distance Measurement The deep learning-based methodology includes a video frame from the camera as an input and the pre-trained open-source object detection to assess the distance between pedestrians for determining the social distancing measures in a video to assuage the impact of the COVID pandemic [1]. Punn et al. [2] An obtained result from the suggested framework is compared with faster region-based convolution neural network (CNN) and single-shot detector (SSD) in terms of object classification and localization which are loss values, mean average precision (mAP), and frames per second (FPS). The result shows that YOLO v3 displayed superior outcomes with a balanced mAP and FPS score to monitor the social distancing in real time. Nguyen et al. [3] proposes technologies like machine learning, computer vision, ultrasound, thermal, etc. It presented and reviewed various technologies with an overview, inspected the state of the art, and considered how it can be utilized to uplift and promote social distancing measures and facing open issues in it with latent solutions to address it as per Fig. 3.

48 Fig. 2 Need of the model

Fig. 3 Need of pedestrian detection and distance measurement

B. Chougule et al.

Population Index and Analysis Based on Different …

49

Fig. 4 Need of object detection

2.2 Object Detection Detecting instances of perceptible objects of a certain class in digital images (such as plants, humans, animals, birds, or cars) in object detection is an important computer vision task. It can be grouped into two subtopics: cognition and “general object detection” include methods to simulate the human vision under a unified framework for detecting different types of objects, and “detection applications” include specific application scenarios for detection, such as face detection, text detection, and pedestrian detection [4, 5] as per Fig. 4.

2.3 Social Distancing for COVID-19 The paper presents a scientific study of busy hours at public places to avoid the glut of people. The economic impact of social distancing measures targets decreasing infections among the people. Studies have proved that social distancing measures fall off the aggregate economic loss during the lockdown [6]. Many scientists and researchers put their best foot forward to find a solution to stop the spread of COVID, and many articles have been published related to severe acute respiratory syndrome (SARS),

50

B. Chougule et al.

Fig. 5 Need of social distancing

Middle East respiratory syndrome (MERS), and COVID-19 [7]. Automating diagnosis of COVID-19 consisting of computed tomography (CT) and chest X-ray (CXR) imaging and using a deep learning approach to get better results is commendable and can be found in [8] and Fig. 5.

3 Object Detection and Distance Measurement 3.1 Object Detection Object detection is used to detect objects in the given frame and to place a bounding box around each detected object in that frame. Deep learning is a powerful method for object detection. Algorithm Selection:

3.1.1

R-CNN

Region-based CNN is a method based on selective search. The main principle used behind this method is to extract limited regions from the image. Thus, the research works on these limited regions. The regions which are predicted can be overlapping with each other and may be vary in size as well. So, based on intersection over union (IOU) score: Where there is minimum suppression that region is used to ignore the bounding boxes.

3.1.2

Fast R-CNN

The fast R-CNN method follows the R-CNN algorithm. Feature extraction in the original image takes place only once. Then based on the location of the region proposals, the relevant ROI features are chosen.

Population Index and Analysis Based on Different …

3.1.3

51

Haar Cascade

The “Haar wavelet” is the origin behind the Haar cascade classifier. To analyze pixels into squares by function in the image, this technique is used, which uses “integral image” ideas to figure out the “features” detected from an image. The algorithm used by Haar cascades includes the selection of a small number of essential features from a wide dataset that is already provided and the result of classifiers, i.e., well-organized additionally detection of face using various cascading techniques in an image.

3.1.4

YOLO

The you only look once (YOLO) algorithm, shown in Fig. 6, is among the finest algorithms for object detection in the current era and works in a completely different manner than what traditional algorithms used to have. As the name suggests (you only look once), in a single view, the algorithm is able to predict and give results. The bounding boxes are created using the method of probabilities and thresholds. This algorithm is used for real-time processing for object detection. R-CNN and fast R-CNN are selective search algorithms, and they slow down the performance and are time consuming algorithms, while YOLO predicts with single network evaluation, in contrast to others which require substantial amounts for a single image. As per Fig. 7, in object tracking first, a new ID is given to every detected person, and then, a bounding box is drawn around each detected person and has to measure centroid for each bounding box. In the next frame, the purple one is the old centroid, and the yellow one is the new centroid. To find out which person has moved, where or which two persons are the same person and then calculate Euclidean distance between every old centroid to every new centroid, and the close pairs will be detected as the same person. When the centroid disappears from the frame, if any is not found for consecutively 50 frames, erase the ID from our array.

Fig. 6 YOLO for object detection

52

B. Chougule et al.

Fig. 7 Object tracking

3.2 Distance Measurement Between Detected Objects 1. F = (P × D)/W 2. D’ = (W × F)/P After a deep understanding and discussion of all these methods, the conclusion is that the paper is best suited by using the YOLO algorithm. The Figs. 7 and 8 clearly describe how YOLO works and how it suits the vision of this project to bring it to fruition.

4 Proposed Methodology The paper proposes a solution-efficient proposal to use the YOLO algorithm to detect the crowd density at an area chosen by the user, which follows COVID-19 safety protocols by maintaining social distancing norms. The ML model also finds violations over some time and presents it to the user for a gasket of pre-verified data

Fig. 8 Distance measurement formula

Population Index and Analysis Based on Different …

53

Fig. 9 Overall working methodology

of population index for a particular area for them to plan their future visits to that particular area with utmost protection as per Fig. 9.

4.1 Stage I: Detecting Violators In Fig. 10, the stage mainly focuses on detecting pedestrians in the frame and then calculating the distance between every pedestrian who is detected in the frame, and then, it will show how many people are at high, low, and not at risk. It contains the following features: 1. To get a camera frame or video feed. 2. Deep learning is firstly used to detect and then to localize the pedestrians in the frame. 3. Find the distance between pedestrians: To plot pedestrians equidistantly using the Euclidean distance formula and thus calculate the approximate distance between them. 4. Find and detect violators who do not follow the minimum distance rule. The respective flowchart of object detection to detect if it is at safe or unsafe distance has been illustrated in Fig. 11.

Fig. 10 Detection of violations

54

B. Chougule et al.

Fig. 11 Stage 1—flowchart of object detection

4.2 Stage II: Finding Busy Hours Figure 12 shows a stage mainly focused on findings of busy hours based upon the number of violations at a particular time to avoid overcrowding at public places. It contains the following features: 1. On a particular day and time, using a camera frame, detection of people can be done by using the YOLO algorithm. 2. After detection, distance is measured between two humans, and if that distance is less than the safest distance, those humans have violated social distancing norms. 3. If the number of violators is more in a particular region, then that region is occupied by a large number of people. 4. Compared to the capacity of the region, people are more so they are not following social distancing. One could avoid going to these particular regions where there are more violators, or they can choose different timing to visit that place.

Population Index and Analysis Based on Different …

55

Fig. 12 Busy hours in a day

The respective flowchart of detection of busy hours to further detect if it is at safe or unsafe to visit the place has been illustrated in Fig. 13. Fig. 13 Stage 2—flowchart of busy hours

56

B. Chougule et al.

Fig. 14 Ideal case of social distancing

5 Experiment and Result Analysis 5.1 YOLO Outputs 5.1.1

Ideal Case

In this Fig. 14, humans are detected in the video as being sufficiently far away from each other. That is why a green bounding box is created to showcase that they are socially distant.

5.1.2

Intermediate Case

In Fig. 15, the paper shows an observation that there are a few violations showcased by the red bounding boxes. Our project keeps a record of these violations with their timestamp.

5.1.3

Worst Case

In this Fig. 16, all people detected in the frame are violating the social distancing norms.

Population Index and Analysis Based on Different …

57

Fig. 15 Intermediate case of social distancing

Fig. 16 Worst case of social distancing

5.2 Database of Violations Using a MySQL and Python connector, database is created which is used to store the number of violations at a particular time (Fig. 17). It also helps to give an idea of how many people are maintaining social distancing norms and able to monitor COVID-19 social distancing violations more efficiently.

58

B. Chougule et al.

Fig. 17 Database of violations

5.3 Busy Hours The proposed method has been implemented at a famous cafe in Pune, India. The following graph shows the busy hours of that cafe in terms of people visiting from Monday to Friday. It also shows an average of all the days for a better observation. It is observed that in a pandemic, people are gathering more at public places. To avoid this, one could use these visualizations and analysis and avoid going to that particular place at that specific time and choose a less crowded time as per their convenience. At this place, human detection is carried out via video frames for a while from Monday to Friday during working hours; and after detection, distance is measured between the two humans. If the distance is less than the safest distance, a violation occurs. If more violations are observed, that means the number of people present at that place is more. So to follow the social distancing norms, one could avoid such situations via these visualizations and find a better and suitable time according to their comfort and convenience. The research was conducted at a place having an area equal to 5293 m2 where FSI is calculated to around 2.5. Considering the rules and regulations given by the government and the capacity of the restaurant, 150 people can be accommodated at a particular time under normal conditions. But in case of a pandemic according to the government’s guidelines, restaurants should operate with 50% of their full capacity [9]. So, let us consider two pedestrians that are 6 feet apart, 50 people approx. can visit the place at the same time (with 0 violations). As per the

Population Index and Analysis Based on Different …

59

Figs. 18, 19, 20, 21, 22 and 23, as the number of people visiting goes on increasing, number of violations also increases because the distance between two pedestrians will not be safe. Fig. 18 Busy hours of Monday

Fig. 19 Busy hours of Tuesday

60

B. Chougule et al.

Fig. 20 Busy hours of Wednesday

Fig. 21 Busy hours of Thursday

5.4 Discussion Strength: In the research, use of YOLO is a great strength because the YOLO algorithm is one of the finest object detection algorithms, and compared to other

Population Index and Analysis Based on Different …

61

Fig. 22 Busy hours of Friday

Fig. 23 Weekly average busy hours

object detection algorithms, YOLO is the fastest. Using YOLO, real-time detection is achieved. If a person wants to go out in a public region, he could use such a module to get information about that region before stepping out and could avoid going to crowded regions.

62

B. Chougule et al.

Challenges: For human or pedestrian detection, which algorithm should be used was a difficult task. Before going for YOLO, the Haar cascade algorithm was used but it did not have the required result and contained some inaccuracies. Getting a high-resolution video frame was quite challenging.

6 Conclusion The paper aims to be effective for making social distancing between people at all times, making things more systematic and safer not only during COVID-19 but after that also. With the help of artificial intelligence (AI), the same technology that is the foundation of self-driving Tesla and Netflix recommendations, combined with edge computing, the technology that is reshaping the Internet of things (IoT), the paper aims to practice social distancing with least disturbance to our daily lives. Using algorithms like R-CNN and YOLO, the research performed real-time human detection for fast and accurate responses enabling us to monitor social distancing at public places like malls, companies, institutions, hospitals, etc. With efficient monitoring using database management skills, our tool helps in bringing about discipline among people with its robust and specific approach to achieve maximum safety for all people. The paper monitors a particular area and predicts the number of people visiting that place in a week, giving us weekly insights of the traffic seen per day, which in turn gives insights on the following week’s traffic, and then alerts the general public accordingly for their safety.

7 Future Scope In addition to the research, real-time analyzes of the violators could be implemented along with an application of face detection. Immediate action can be taken, for example, alerting the violator by sending them a message. To make this project more effective, the inclusion of mask detection along with social distancing could be added. Another future scope could also be analyzing the previous week’s data and using an ML model to predict the following week’s population traffic in a particular area. The paper aims to create a surveillance app soon which would have all of this, and as soon as the user puts in the desired location of where they want to visit, it entails predictive analysis of last week data, gives insights on this week as per the day of week and time of day, and suggests which time is best suited for the user instead of randomly visiting at any place and finding a busy traffic present.

Population Index and Analysis Based on Different …

63

References 1. Agarwal S, Punn NS, Sonbhadra SK, Nagabhushan P, Pandian K, Saxena P (2020) Unleashing the power of disruptive and emerging technologies amid covid 2019: A detailed review. arXiv preprint arXiv:2005.11507 2. Punn NS, Sonbhadra SK, Agarwal S (2020) Monitoring covid-19 social distancing with person detection and tracking via fine- tuned YOLO v3 and deep sort techniques. arXiv preprint arXiv:2005.01385 3. Nguyen CT, Saputra YM, Van Huynh N, Nguyen N-T, Khoa TV, Tuan BM, Nguyen DN, Hoang DT, Vu TX, Dutkiewicz E et al (2020) Enabling and emerging technologies for social distancing: acomprehensive survey, arXiv preprint arXiv:2005.02816 4. Zou Z, Shi Z, Guo Y, Ye J, Object Detection in 20 Years: A Survey. https://arxiv.org/pdf/1905. 05055.pdf 5. Georgievski B (2020) Object detection and tracking in 2020. Medium. https://blog.netcetera. com/object-detection-and-tracking-in-2020-f10fb6ff9af3 6. Kahale N (2020) On the economic impact of social distancing measures. SSRN Electron J. https://doi.org/10.2139/ssrn.3578415 7. Sonbhadraa SK, Agarwala S, Nagabhushan P, Target specific mining of COVID-19 scholarly articles using one-class approach. https://arxiv.org/pdf/2004.11706.pdf 8. Punn NS, Agarwal S, Automated diagnosis of COVID-19 with limited posteroanterior chest X-ray images using fine-tuned deep neural networks. https://arxiv.org/pdf/2004.11676.pdf 9. Guidelines for effective control of COVID-19 [As per Ministry of Home Affairs (MHA) Order NO. 40-3/2020-DM-I(A) dated 23rd March 2021]. https://www.restaurantindia.in/news/restau rants-in-maharashtra-to-operate-with-50-capacity-govt.n19720 10. Advice for the public on COVID-19—World Health Organization. WHO www.who.int/eme rgencies/diseases/novel-coronavirus-2019/advice-for-public 11. Worldometer for the daily covid cases. www.worldometers.info/coronavirus/ 12. Berglund J, Tracking COVID-19: There’s an App for That. https://www.embs.org/pulse/art icles/tracking-covid-19-theres-an-app-for-that/ 13. Robakowska M, Tyranska-Fobke A, Nowak J, Slezak D, Zuratynski P, Robakowski P, Nadolny K, Ładny JR, The use of drones during mass events. https://journals.viamedica.pl/disaster_ and_emergency_medicine/article/view/55211 14. Harvey J, LaPlace A (2019) Megapixels.cc: Origins, ethics, and privacy implications of publicly available face recognition image datasets. https://megapixels.cc/ 15. Zhao Z-Q, Zheng P, Xu S-T, Wu X (2019) Object detection with deep learning: a review. IEEE Trans. Neural Networks Learn. Syst. 30(11):3212–3232 16. Kajabad EN, Ivanov SV (2019) People detection and finding attractive areas by the use of movement detection analysis and deep learning approach. Procedia Comput Sci 156:327–337. (ISSN 1877–0509) 17. Jiang H, Lim WY, Ng JS, Chie Teng HZ, Yu H, Xiong Z, Niyato D, Miao C, AI-Empowered decision support for COVID-19 social distancing. https://www.aaai.org/AAAI21Papers/DEMO227.JiangH.pdf 18. Alahi A, Bierlaire M, Vandergheynst P (2014) Robust real-time pedestrians’ detection in urban environments with low-resolution cameras. Transp Res Part C Emerg Technol 39:113–128

On the Discriminability of Samples Using Binarized ReLU Activations Michał Lewandowski, Werner Zellinger, Hamid Eghbal-zadeh, Natalia Shepeleva, and Bernhard A. Moser

Abstract Binarized ReLU activations are considered as a metric space equipped with the Hamming distance. While for two-layer ReLU networks with random Gaussian weights it can be shown theoretically that local metric properties are approximately preserved, we experimentally study the discrimination capability in this Hamming space for deeper ReLU networks and look also at the non-local behavior. It turns out that the discrimination capability is approximately preserved as expected. Keywords Discriminability · ReLU networks · Reduced information

1 Introduction In this paper, we concentrate on deep ReLU networks with binarized activations and outputs. ReLU networks perform particularly well in many practical tasks, such as generative adversarial networks [1, 2], domain adaptation methods [3–5], and two sample tests based on neural networks [6, 7]. Particularly, for embedded systems, the reduction of precision in the inference is interesting from the point of view of keeping computational efforts and power consumption low [8]. But, there is also a more theoretical motivation that comes from metric embedding by looking at neural networks as metric preserving mappings in some appropriate spaces [9–14]. For example, in [14], two-layer ReLU networks with random Gaussian weights and binarized activations and outputs, respectively, are considered. Interestingly, this setting guarantees approximately isometric embedding into the Hamming space. This analysis shows that each standard DNN layer (with random Gaussian weights) performs a stable embedding of the data from one layer to the next by preserving local structures in the manifold. For deeper networks, the analysis becomes much M. Lewandowski (B) · W. Zellinger · N. Shepeleva · B. A. Moser Software Competence Center Hagenberg GmbH, Hagenberg 4232, Austria e-mail: [email protected] H. Eghbal-zadeh LIT AI Laboratory and Institute of Computational Perception, Johannes Kepler University, Linz 4040, Austria © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_6

65

66

M. Lewandowski et al.

more complicated due to the nested composition of nonlinear functions. In contrast to shallow networks, deep networks allow representing restricted Boltzmann machines with a number of parameters exponentially greater than the number of the network parameters [15]. Looking at the preimage in the layer below induced by the layer above for some specific output, for example 0, shows that a ReLU network leads to a tree of nested polyhedral cells which become smaller in size the deeper the network. By constructing specific ReLU networks [16] show exemplary that deep networks divide the input space into an exponential number of (polyhedral) sets, which is not possible with a single layer with the same number of parameters. This way, deep neural networks are more expressive than shallow ones. Reference [17] provides an upper and lower bound on the number of polyhedral cells and consider the influence of width versus depth of ReLU network on the number of created linear regions, finding out that wider ReLU networks result in a finer tessellation than deeper ones. Though such partial results, it is not yet fully understood how the network architecture influences the geometry and distribution of the induced cells, Reference [18] points out that these induced polyhedral cells Ci are actually equivalence classes [x] (up to the border) resulting from the equivalence relation, x ∼ y, that two sample points in the input space are considered equivalent if they show the same binarized activation profile. We take up this view by introducing a metric in the tessellation space of cells, T = {[x] |, x ∈ Rd }, as editing distance of adjacent cells. This means, the number of minimal adjacent cells sharing a common face connecting two cells in this tessellation introduces a notion of distance, dT . In our analysis, we exploit the relation that the Hamming distance, d H , on the binary activation states and the distance in the tessellation have the same distinguishability behavior. This means, the Hamming distance is able to distinguish between points in different cells, meaning d H (a(x1 ), a(x2 )) > 0 ⇐⇒ dT (x1 , x2 ) > 0, where a(x) = (β1,1 (x), . . . , β1,n 1 (x); . . . ; β L ,1 (x), . . . , β L ,n L (x)) denotes the vector of binarized activations βk,ik in the kth layer of neuron i k ∈ {1, . . . , n k }: βk,ik (x)=1 if ak,ik (x)>0, and βk,ik (x) = 0 otherwise, with a total number of L layers, where ak,ik denotes the activation in node i k of layer k.

1.1 Contributions While for two-layer ReLU networks with random Gaussian weights, it can be shown theoretically that local metric properties are approximately preserved; we experimentally study the discrimination capability in this Hamming space for deeper ReLU networks and look also at the non-local behavior. We show experimentally that the discrimination capability is approximately preserved locally also for deeper networks. In this context, we give synthetic examples which indicate that binarized activation values contain enough of information to distinguish between points localized differently in the data space.

On the Discriminability of Samples Using Binarized ...

67

The rest of the paper is organized into three main sections. Section 2 describes investigated metrics, Sect. 3 describes experiments, and Sect. 4 concludes our work and points at its extensions.

2 Metrics Used in the Experimental Setup For samples S1 = {x1 , . . . , xm } ∼ μ and S2 = {y1 . . . , yn } ∼ ν, xi , y j ∈ Rd , we start our analysis with Hamming distance between samples containing one observation. Next, we extend our analysis to samples containing more observations, where Hamming distance is not unambiguous anymore, and use MMD distance with an adequate kernel, the exponentiated Hamming, instead. Finally, we benchmark the proposed metrics with the Wasserstein-2 distance, following the authors of [19], who show that it correlates with a visual distinguishability of images, and thus is applicable in real-world scenarios. 1. Hamming. For one point samples | S1 |=| S2 |= 1, we study the behavior of the Hamming distance d H (a(x1 ), a(y1 )) =| {a(x1 )i /= a(y1 )i , i = 1, . . . , d} |. 2. Maximum Mean Discrepancy. The unbiased estimator of MMD is [20] MMD2u [F, X, Y ] =

1 m(m−1)

m Σ m ) ( Σ k xi , x j +

i=1 j/=i

1 n(n−1)

n m Σ ) ( Σ −2mn k xi , y j ,

n Σ n ) ( Σ k yi , y j

i=1 j/=i

i=1 j=1

where F is a class of functions f : X → R and k(xi , y j ) = exp {−d H (xi , y j )} is the exponentianted Hamming kernel [21]. 3. Wasserstein-p. We study W p (μ, ν) with dH as the base distance [22] ∮ W pp (μ, ν) =

inf

γ ∈┌(μ,ν) M×M

d H (a(S1 ), a(S2 )) p dγ (a(S1 ), a(S2 )),

where ┌(μ, ν) is a collection of all measures on Rd × Rd with marginals μ, ν. Wasserstein- p distances enjoy a geometrical interpretation, what makes it particularly well-suited to work with once geometry of data is involved. In our experiments, we estimate it using Sinkhorn algorithm [23]. 4. Fréchet. Assuming μ = N (μ1 , Σ1 ) and ν = N (μ2 , Σ2 ), Wasserstein-2 distance can be calculated analytically as [24]: ⎛ d 2 (μ, ν) = ∥μ1 − μ2 ∥22 + Tr ⎝

Σ 1

+

Σ 2

)1/2 ⎞ ΣΣ ⎠. −2 (

1

2

68

(a) Training data.

M. Lewandowski et al.

(b) Tessellation and test data.

(c) Metric analysis.

Fig. 1 Three steps of our experimental setup for distance analysis. From the left: the training setup of a 15:15:15 (3 hidden layers, 15 neurons each) ReLU network, the testing setup: how do we compute metrics’ values, results

3 Experiments We create a number of multivariate normal distributions,1 with means spaced evenly on n-dim space, and random covariance matrices (Fig. 1a). To verify discriminability, we compare behavior of proposed metrics to the Fréchet distance (abbreviated fd in figures), computed on data and activation spaces. We train a ReLU network to distinguish the indicated number of groups with a high accuracy (Fig. 1a, groups are distinguished by colors), and store the network’s parameters. Next, we create samples consisting of one or more points on a diagonal of the cube of our training setup (Fig. 1b), and propagate them through the pretrained ReLU network to get their activation values. Finally, we compare the behavior of metrics.

3.1 Influence of Number of Layers We present results in Fig. 2. Note different metrics’ behavior for lower numbers of layers compared to higher numbers. This reflects higher concentration of cells in the tessellation in a close neighborhood to the input data and low concentration of cells further away, and will be amplified by a higher number of cells in consistency with the analysis of [16]. We point attention of the reader that we do not expect binary distances between samples to behave as the Euclidean distance between means. In fact, a convex behavior of distances might in some cases be more desirable, allowing to distinguish between very similar samples faster than relying on their Euclidean differences, especially in a high variance scenario.

1

For normal distributions Wasserstein-2 distance has a closed analytical form.

On the Discriminability of Samples Using Binarized ...

(a) 128:128:128

(b) .128 : ... . . : 128. 5 times

69

(c) .256 : ... . . : 256. 10 times

Fig. 2 Behavior of metrics with varying number of layers. Captions describe used architecture

(a) 5 labels, 100 data (b) 10 labels, 100 data (c) 20 labels, 100 data points points points

(d) 5 labels, 10 data (e) 10 labels, 10 data (f) 20 labels, 10 data points points points Fig. 3 Behavior of metrics in function of varying number of data points in each distinctly centered Gaussian distribution, and number of groups we group points into. We used 50:50:50 ReLU network

3.2 Influence of Sample Size and Number of Training Labels In the following, we show that our results extend to larger sample sizes, different number of labels and also consider MMD distance (Fig. 3).

70

M. Lewandowski et al.

4 Conclusion and Future Work Though we definitely lose some information comparing to real-valued activation values, binarized activations preserve clear distinguishability capabilities. The analysis could be refined by taking the geometry of the cells into account. This means, by checking which of the activation states βk,ik refer to hyperplanes that touch the cell. This way we expect to establish an isometry embedding, what is left for an upcoming paper. In this context, we will also check applications, e.g., by constructing a two sample test statistics using binarized activation values of some sample. Moreover, we will check the effects where data are sparse compared to the number of dimensions. Acknowledgements The research reported in this paper has been funded by the Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation and Technology (BMK), the Federal Ministry for Digital and Economic Affairs (BMDW), and the Province of Upper Austria in the frame of the COMET–Competence Centers for Excellent Technologies Program and the COMET Module S3AI managed by the Austrian Research Promotion Agency FFG. The LIT AI Lab is financed by the Federal State of Upper Austria.

References 1. Heusel M, Ramsauer H, Unterthiner T, Nessler B, Hochreiter S (2017) GANs trained by a two time-scale update rule converge to a local Nash equilibrium 2. Salimans T, Goodfellow I, Zaremba W, Cheung V, Radford A, Chen X (2016) Improved techniques for training gans. In: Proceedings of the 30th International conference on neural information processing systems. NIPS’16, Red Hook, NY, USA, pp 2234–2242 3. Ganin Y, Ustinova E, Ajakan H, Germain P, Larochelle H, Laviolette F, Marchand M, Lempitsky V (2016) Domain-adversarial training of neural networks. J Mach Learn Res 17(1):2096–2130 4. Long M, Zhu H, Wang J, Jordan MI (2017) Deep transfer learning with joint adaptation networks. In: Precup D, Teh YW (eds) Proceedings of the 34th International conference on machine learning. Proceedings of machine learning research, vol 70, pp 2208–2217. International Convention Centre, Sydney, Australia 5. Zellinger W, Moser BA, Saminger-Platz S (2021) On generalization in moment-based domain adaptation. Ann Math Artif Intell 333–369 6. Lopez-Paz D, Oquab M (2017) Revisiting classifier two-sample tests. In: International conference on learning representations. Toulon, France 7. Kirchler M, Khorasani S, Kloft M, Lippert C (2020) Two-sample testing using deep learning. In: Chiappa S, Calandra R (eds) Proceedings of the twenty third international conference on artificial intelligence and statistics. Proceedings of machine learning research, vol 108, pp 1387–1398 8. Conti F, Schiavone PD, Benini L (2018) Xnor neural engine: a hardware accelerator ip for 21.6 fj/op binary neural network inference. IEEE Trans Comput-Aided Des Integr Circuits Syst 37(11):2940–2951 9. Indyk P, Matousek J (2004) Low-distortion embeddings of finite metric spaces. In: Handbook of discrete and computational geometry, pp 177–196 10. van der Maaten L, Hinton G (2008) Visualizing data using t-sne. J Mach Learn Res 9(86):2579– 2605

On the Discriminability of Samples Using Binarized ...

71

11. Suárez-Díaz JL, García S, Herrera F (2018) A tutorial on distance metric learning: mathematical foundations, algorithms, experimental analysis, prospects and challenges (with appendices on mathematical background and detailed algorithms explanation). arXiv:1812-05944 12. Xiao C, Zhong P, Zheng C (2018) Bourgan: generative networks with metric embeddings. Adv Neural Inf Process Syst 2269–2280 13. Courty N, Flamary R, Ducoffe M (2018) Learning Wasserstein embeddings. In: ICLR 2018— 6th International conference on learning representations. Vancouver, Canada, pp 1–13 14. Giryes R, Sapiro G, Bronstein A (2016) Deep neural networks with random gaussian weights: a universal classification strategy? IEEE Trans Signal Process 64:3444–3457 15. Montúfar G, Morton J (2015) When does a mixture of products contain a product of mixtures? SIAM J Discrete Math 29(1):321–347 16. Montúfar GF, Pascanu R, Cho K, Bengio Y (2014) On the number of linear regions of deep neural networks. In: Advances in neural information processing systems, vol 27 17. Serra T, Tjandraatmadja C, Ramalingam S (2018) Bounding and counting linear regions of deep neural networks. In: International conference on machine learning, pp 4558–4566 18. Shepeleva N, Zellinger W, Lewandowski M, Moser B (2020) Relu code space: a basis for rating network quality besides accuracy 19. Heusel M, Ramsauer H, Unterthiner T, Nessler B, Hochreiter S (2017) Gans trained by a two time-scale update rule converge to a local Nash equilibrium. In: Proceedings of the 31st International conference on neural information processing systems. NIPS’17, Red Hook, NY, USA, pp 6629–6640 20. Gretton A, Borgwardt KM, Rasch MJ, Schölkopf B, Smola A (2012) A kernel two-sample test. J Mach Learn Res 13:723–773 21. Yang J, Liu Q, Rao V, Neville J (2018) Goodness-of-fit testing for discrete distributions via stein discrepancy. In: Dy J, Krause A (eds) Proceedings of the 35th International conference on machine learning. Proceedings of machine learning research, vol 80, pp 5561–5570. Stockholmsmässan, Stockholm Sweden 22. Peyré G, Cuturi M et al (2019) Computational optimal transport: with applications to data science. Found Trends Mach Learn 11:355–607 23. Cuturi M (2013) Sinkhorn distances: lightspeed computation of optimal transport. In: NIPS, vol 2, p 4 24. Fréchet MM (1957) Sur la distance de deux lois de probabilité. CR Acad Sci Paris 244:689–692

Supervised and Unsupervised Machine Learning Approaches—A Survey C. Esther Varma and Puja S. Prasad

Abstract Machine learning task is broadly divided into supervised and unsupervised approaches. In supervised learning, output is already known and we have to train the model by giving lot of data called labeled dataset to train our model. The main goal is to predict the outcome. It includes regression and classification problem. In unsupervised learning, no output mapping with input as well as it is independent in nature. The dataset used in unsupervised machine learning is unlabeled. The main focus of this paper is to give detailed understanding of supervised and unsupervised machine learning algorithm with pseudocodes. Keywords Supervised · Unsupervised · Decision tree · K-means · PCA

1 Introduction Machine learning is utilized to show machines how to deal with the information all the more productively. In some cases subsequent to review the information, we cannot decipher the example or concentrate data from the information. All things considered, we apply AI [1]. With the plenitude of datasets accessible, the interest for AI is in ascent. Numerous businesses from medication to military apply AI to remove pertinent data. The motivation behind AI is to gain from the information [2]. Many investigations have been done on the most proficient method to cause machines to learn without help from anyone else [3]. Numerous mathematicians and software engineers apply a few ways to deal with find the arrangement of this issue [3]. Some of them are exhibited in [4, 5]. The different types of machine learning algorithms have been depicted in Fig. 1.

C. Esther Varma (B) · P. S. Prasad GCET, Hyderabad, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_7

73

74

C. Esther Varma and P. S. Prasad

Fig. 1 Machine learning algorithms

2 Different Kinds of Learning The learning directed artificial algorithm is those calculations which need outside help [6]. The info dataset is partitioned into train and test dataset. The train dataset has yield variable which should be anticipated or grouped [2]. All calculations take in some sort of examples from the preparation data set and apply them to the test dataset for expectation or grouping [7, 8].

2.1 Supervised Machine Learning The flowchart of supervised machine learning algorithm has been illustrated in Fig. 2. Three most well-known directed artificial algorithm has been talked about here. 1. Decision tree 2. Naive Bayes

Fig. 2 Supervised machine learning algorithm

Supervised and Unsupervised Machine Learning Approaches—A Survey

75

Fig. 3 Decision Tree

3. Support vector machine Decision Tree: Decision trees are those kind of trees which gatherings credits by arranging them dependent on their qualities [9]. Choice tree is utilized primarily for grouping reason. Each tree comprises hubs and branches [10]. Every hubs addresses credits in a gathering that will be arranged and each branch addresses a worth that the hub can take [11]. A model of decision tree has been shown in Fig. 3. There are two types of decision tree and are based totally at the type of goal variable we have. It may be of two sorts: Categorical Variable Decision Tree: Decision tree which has a specific target variable is called a categorical variable decision tree [12]. Continuous Variable Decision Tree: Decision tree which has a continuous target variable is called continuous variable decision tree. Decision bushes classify the examples through sorting them down the tree from the basis to some leaf node, with the leaf node supplying the class to the instance. Each node in the tree acts as a test case for a few attribute, and every edge descending from that node corresponds to one of the feasible answers to the check case. This process is recursive in nature and is repeated for each subtree rooted at the brand new nodes [13]. Decision tree is simple to apprehend, interpret and visualize. Decision trees implicitly perform variable screening or function choice. It can manage both numerical and express information. It can also take care of multi-output problems. Decision trees require incredibly little attempt from customers for statistics coaching. Nonlinear relationships among parameters do no longer have an effect on tree performance. The pseudocode for decision tree is portrayed below, where S, A and y are preparing set, input quality and target characteristic separately [14].

76

C. Esther Varma and P. S. Prasad

Pseudocode for Decision Tree procedure DTInducer(S,A,y) 1: T = TreeGrowing(S,A, y) 2: Return TreePruning(S,T) procedure TreeGrowing(S,A, y) 1: Create a tree T 2: if one of the Stopping Criteria is fulfilled then 3: Mark the root node in T as a leaf with the most common value of y in S as the class. 4: else 5: Find a discrete function f(A) of the input attributes values such that splitting S according to f(A)’s outcomes (v1 ,...,vn ) gains the best splitting metric. 6: if best splitting metric ≥ threshold then 7: Label the root node in T as f(A) 8: for each outcome v i of f(A) do 9: Subtreei = TreeGrowing (σ f(A)=vi S, A, y). 10: Connect the root node of T to Subtree, with an edge that is labeled as v i 11: end for 12: else 13: Mark the root node in T as a leaf with the most Common value of y in S as the class. 14: end if 15: end if 16: Return T procedure TreePruning(S,T, y) 1: repeat 2: Select a node t in T such that pruning it maximally improve some evaluation criteria 3: if t /= ø then 4: T = pruned (T,t) 5: end if 6: until t /= ø 7: Return T

Naive Bayes: Two important styles of Naive Bayes algorithms are: Gaussian Naive Bayes: Gaussian Naive Bayes is a variation of Naive Bayes that follows Gaussian normal distribution and supports continuous facts. Naive Bayes is a collection of supervised gadget getting to know category algorithms based on the Bayes theorem. It is a simple class approach, but has excessive functionality. Multinomial Naive Bayes: The Gaussian assumption just described is in no way the most effective easy assumption that might be used to specify the generative distribution for every label. Another useful instance is multinomial naive Bayes, where the features are assumed to be generated from a simple multinomial distribution [15]. The multinomial distribution describes the possibility of watching counts among some of categories, and accordingly multinomial Naive Bayes is most appropriate for features that represent counts or count number rates. Mostly focuses on the text order industry [16]. It is primarily utilized for bunching and order reason [17]. The fundamental engineering of Bayes relies upon the restrictive likelihood. It makes trees dependent on their likelihood of occurring. These trees are otherwise called Bayesian network.

Supervised and Unsupervised Machine Learning Approaches—A Survey

77

Pseudocode of Naive Bayes INPUT: training set T, hold-out set H, initial number of components k 0 , and convergence threshold δ EM and δ Add Initialize M with one component. k←k 0 repeat Add k new mixture components to M, initialized using k random examples from T. Remove the k initialization examples from T. repeat E-step: Fractionally assign examples in T to mixture components, using M. M-step: Compute maximum likelihood parameters for M, using the filled in data. If log P(H I M) is best so far, save M in M beat . Every 5 cycles, prune low-weight components of M. until log P(H I M) fails to improve by ratio δ Add . Execute E-step and M-step twice more on M beat using examples from both H and T. Return M beat.

Support Vector Machine: Another most generally utilized best in class AI procedure is support vector machine (SVM). It is mostly utilized for characterization. SVM deals with the rule of edge computation [18]. It fundamentally draws edges between the classes. The edges are attracted such a design that the distance between the margin and the classes is maximum and hence minimizing the classification error [14] The SVM kernel is a feature that takes low-dimensional enter space and transforms it into better-dimensional area, i.e., it converts no longer separable trouble to separable trouble. It is generally beneficial in nonlinear separation issues. Simply placed the kernel, it does some extraordinarily complicated data alterations then finds out the procedure to separate the facts primarily based on the labels or outputs described. Support vector machine has several advantages—very effective in excessive dimensional cases. Its reminiscence green because it makes use of a subset of training points in the decision function referred to as guide vectors. Different kernel capabilities may be specified for the selection capabilities and its possible to specify custom kernels.

2.2 Unsupervised Machine Learning Algorithm It is also sometimes called unaided learning. In this algorithm learns not many components from the information [19]. At the point when new information is presented, it utilizes the recently scholarly components to perceive the class of the information. It is mostly utilized for bunching and size reduction [20]. Two main clustering algorithms are:

78

C. Esther Varma and P. S. Prasad

1. K-means clustering bunching or gathering is a sort of solo learning method that when starts, makes bunches naturally. The things which has comparable attributes are placed in a similar group [18]. This calculation is called k-implies on the grounds that it makes k particular groups. The following are the drawbacks of the algorithm(a) The learning set of rules calls for a priori specification of the number of cluster facilities. (b) The use of exclusive assignment—If there are two incredibly overlapping information, then okay means will no longer be capable of solve that there are two clusters. (c) The studying algorithm is not invariant to nonlinear ameliorations, i.e., with distinctive illustration of records, we get exclusive outcomes (information represented in form of Cartesian coordinates and polar coordinates will supply special effects). (d) Euclidean distance measures can unequally weight underlying elements. (e) The gaining knowledge of set of rules offers the neighborhood optima of the squared errors feature. (f) Randomly choosing of the cluster center cannot lead us to the fruitful end result. Pl. Refer Fig. (g) Applicable best when imply is described, i.e., fails for express information. (h) Unable to address noisy data and outliers. (i) Algorithm fails for nonlinear facts set. The mean of the qualities in a specific group is the focal point of that group [21]. Pseudocode of k-means Clustering function Direct-k-means() Initialize k prototypes (w 1 ,…..,w k ) such that w j =il , j m {1,…..,k}, l m {1,…..,n} Each cluster C j is associated with a prototype w j Repeat for each input vector il , where l m {1,…..,n}, do Assign il to the cluster C j with nearest prototype w j. (i.e, I il - w j* I ≤ I il - w j I, , j m {1,…..,k}) for each cluster C j, where , j m {1,…..,k}, do Update the protocurrently type w j to be the centroid of all samples Σ in C j so that w j = il m Cj il / I C j I Compute the error function:

E=

Σ j=1

Σ il∈C j

| | |il − w j |2

Until E does not change significantly or cluster membership no longer changes

Supervised and Unsupervised Machine Learning Approaches—A Survey

79

2. Principal Component Analysis(PCA) In principal component analysis or PCA, the element of the information is decreased to make the calculations quicker and simpler. To see how PCA functions, we should take an illustration of 2D information. Principal element analysis of a records matrix extracts the dominant styles in the matrix in phrases of a complementary set of rating and loading plots. It is the obligation of the facts analyst to formulate the clinical issue handy in terms of PC projections, PLS regressions and so forth. Ask yourself, or the investigator, why the facts matrix became accrued, and for what purpose the experiments and measurements had been made. Specify before the analysis what varieties of patterns you would count on and what you will find interesting. At the point when the information is being plot in a chart, it will take up two tomahawks [17]. PCA is applied on the information, the information then, at that point, will be 1D. Pseudocode of PCA R←X for(k = 0,……,K-1) do { λ=0 T(k) ← R(k) for(j=0,……, J ) do { P(k) ← R(k) T(k) P(k) ← P(k) || P(k) ||-1 T(k) ← R P(k) λ’ ← || T(k) || if( |λ’ – λ| ≤ ε ) then break λ ← λ’ } R ← R - T(k) (P(k) )T } Return T, P, R

3 Conclusion In this paper, we have discussed about different machine learning algorithm. Decision tree, SVM and Naive Bayes are supervised machine learning Algorithm. In machine learning, the input is result as well as data and the output is rules contrary to

80

C. Esther Varma and P. S. Prasad

traditional programming languages. This paper gives idea about supervised as well as unsupervised machine learning algorithm as well as their types. Decision tree is classifier and it can be used for classification as well as regression purposes both. But mostly, it is used for classification purposes. SVM is support vector machine and its main aim is separate the classes using hyper-plane. In unsupervised machine learning, machine only looks for the pattern as data has no labels. Training starts with huge data that form a feature vectors which using an algorithm is converted into predictive model that is tested with new set of data supervised machine learning is less complex, conducts offline analysis and gives comparatively more accurate result than unsupervised learning that is more complex and performs real-time analysis.

References 1. 2. 3. 4. 5. 6.

7. 8. 9.

10. 11. 12. 13. 14.

15. 16. 17. 18. 19.

Bonaccorso G (2017) Machine learning algorithms. Packt Publishing Ltd. Goodfellow I, Bengio Y, Courville A (2016) Machine learning basics. Deep Learn 1(7):98–164 Dietterich TG (1997) Machine-learning research. AI magazine 18(4):97–97 El Naqa I, Murphy MJ (2015) What is machine learning? In: Machine learning in radiation oncology. Springer, pp 3–11 K¨ording KP, K¨onig P (2001) Supervised and unsupervised learning with two sites of synaptic integration. J Comput Neurosci 11(3):207–215 Arunraj NS, Hable R, Fernandes M, Leidl K, Heigl M (2017) Comparison of super- vised, semi-supervised and unsupervised learning methods in network intrusion detection system (nids) application. Anwendungen und Konzepte der Wirtschaftsinformatik 6 Chen L, Zhai Y, He Q, Wang W, Deng M (2020) Integrating deep supervised, self- supervised and unsupervised learning for single-cell RNA-seq clustering and annotation. Genes 11(7):792 ButlerKT, Davies DW, Cartwright H, Isayev O, Walsh A (2018) Machine learning for molecular and materials science. Nature 559(7715):547–555 Liu W, Chawla S, Cieslak DA, Chawla NV (2010) A robust decision tree algorithm for imbalanced data sets. In: Proceedings of the 2010 SIAM international conference on data mining. SIAM, pp 766–777 Jordan MI, Mitchell TM (2015) Machine learning: trends, perspectives, and prospects. Science 349(6245):255–260 Manwani N, Sastry PS (2011) Geometric decision tree. IEEE Trans Syst Man Cybern Part B Cybern 42(1):181–192 Ayodele TO (2010) Types of machine learning algorithms. New Adv Mach Learn 3:19–48 Wei J, Chu X, Sun X-Y, Kun Xu, Deng H-X, Chen J, Wei Z, Lei M (2019) Machine learning in materials science. InfoMat 1(3):338–358 Lee K, Caverlee J, Webb S (2010) Uncovering social spammers: social honeypots+ machine learning. In: Proceedings of the 33rd international ACM SIGIR conference on research and development in information retrieval, pp 435–442 Witten IH, Frank E, Hall MA, Pal CJ (2005) Mining data: Practical machine learning tools and techniques. In: Data Mining 2, p 4 Tom M Mitchell. Does machine learning really work? AI magazine, 18(3):11–11, 1997. Mohri M, Rostamizadeh A, Talwalkar A (2018) Foundations of machine learning. MIT press Raschka S (2015) Python machine learning. Packt publishing Ltd. Zhou Z-H (2016) Learnware: on the future of machine learning. Front Comput Sci 10(4):589– 590

Supervised and Unsupervised Machine Learning Approaches—A Survey

81

20. Hilas CS, Mastorocostas PA (2008) An application of supervised and unsupervised learning approaches to telecommunications fraud detection. Knowl Based Syst 21(7):721–726 21. Oral M, Oral EL, Aydın A (2012) Supervised versus unsupervised learning for construction crew productivity prediction. Autom Constr 22:271–276

Skin Cancer Classification Using Deep Learning D. K. Yashaswini, Pratheeksha C. Dhanpal, and S. A. Bhoomika

Abstract In the past 10-years, from 2008 to 2018, the annual sort of skin cancer cases has raised by fifty-three percent due to increased ultraviolet exposure. Though skin cancer is one of the foremost deadly variants of malignant neoplastic disease, a faster identification can cause a very high chance of survival. The primary step of diagnosis of a lesion by a specialist is visual examination of the suspicious skin lesion. It is found that a specialized doctor who treats skin typically carries out a sequence of phases, initial from eye examination of distrusted injuries, followed by dermoscopy (magnifying injuries microscopically) and later with a diagnostic test such as biopsy. This process is time consuming and the patient might progress to future stages. What is more correct designation is subjective; most effective skin doctor has associative accuracy of eightieth in properly diagnosing the carcinoma type. Adding to those difficulties, there do not seem to be several masterful dermatologists out there for public aid. In association, correct diagnosis is important, adding to the similarities of some lesion types; what is more important is that the diagnostic accuracy correlates powerfully with the masterful experience of the medical. An increased help to the skin doctor is delivered through the emerging technologies of deep learning. The basic goal of this method is to train a model to solve the problem by reviewing cancer images. The model will be constructed without any programming skills, which is a unique quality of the presentation. Convolutional neural network (CNN) primarily based classifiers became the most effective selection for cancer detection within the recent era. The analysis has indicated that classifiers that supported CNN classify carcinoma pictures similar to dermatologists that has allowed a fast and life-saving diagnosis. Keywords Skin cancer · Deep learning · EDA · Analyzing · Convolutional neural networks (CNNs) · Precession · Accuracy

D. K. Yashaswini (B) · P. C. Dhanpal · S. A. Bhoomika Don Bosco Institute of Technology, Bangalore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_8

83

84

D. K. Yashaswini et al.

1 Introduction Skin carcinoma is known to be one of the most dangerous and widespread illness [1] annually increased by fifty-three percent. In USA, 40 lakh new cases of carcinoma were recorded. The overall statistics has measures that are undoubtedly fearful [2, 3]. Reports of the recent times have shown that from 2008 to 2018 period, a fifty percent rise in new skin cancer cases has been identified annually [1, 4]. The death rate caused from the illness is predicted to increase within the coming10 years range. The chance of survival is very less [5, 6] if diagnosed in the last few stages. However, the survival rate is 97 if the carcinoma can be noticed in its initial stages [2]. This calls the first detection of carcinoma. This paper tackles the problem of initial diagnosing, providing a better accuracy. A qualified specialist has been observed to follow a series of procedures, commencing with an eye inspection of suspicious areas, next dermoscopy (microscopically magnifying lesions), and lastly a biopsy. This can be time consuming and also the patient’s condition may progress to further stages. Furthermore, correct diagnosing is required, counting on the practitioner’s talent. It has been seen that the simplest specialist has a precision which is less than eightieth in properly diagnosing the carcinoma [7]. In addition to such complications, there are not several skillful dermatologists available globally for public health care. With the purpose of diagnosing carcinoma quickly at the initial stages and to solve a number of issues, there are a few in depth research solutions by establishing a model with image analysis algorithms [7]. The bulk of those algorithmic explanations is consistent, meaning they require knowledge to be usually distributed. Since the nature of information cannot be managed, these ways can be inadequate to exactly diagnose the illness. Nonparametric explanations, on the other hand, cannot rely on the requirement that the data is of distribution type. By utilizing deep learning, a CNN model is implemented in this work to provide additional help to the specialist. The benefit of this method is that it allows us to train a computer to solve the problem by analyzing cancer images. The uniqueness of the performance is that the model will be designed while not having much acquaintance in the programming side. The proficiency of diagnosis achieved by this model is around 97.6% and sometimes it is found to be 100%. Studies recommend that within the space of carcinoma discovery and image taxonomy, there lies an inordinateness of analysis. A survey providing the detailed information of those ways is on the market in Refs. [1, 5. Each of those papers makes use of the obtainable state-of-the-art ways and claims enhancements on its performance.

Skin Cancer Classification Using Deep Learning

85

2 Literature Survey Muhammad Qasim Khan et al. [1] In this analysis paper, they have provided an effective smart system for the classification of melanoma skin carcinoma and nevus. It was discovered that lesion detection and segmentation were the major downside that caused the misclassification. The K-means cluster algorithm using centroid selection was accustomed to obtain the ROI from the lesion precisely and competently. Textural and pigment type extraction methods were accustomed to acquire the best-suited types for taxonomy. For feature collection, GLCM and LBP types were united with the color features to attain a high-level classification accurateness of about 96%. With this, their suggested technique has been ready to classify carcinoma pictures into melanoma skin cancer and nevus with accurateness and resourcefulness. The efficiency and functioning of the projected approach were valid on dermis image dataset. Mohammad Ali Kadampu et al. [8] The DLS is available in both desktop and cloud versions. It retains multi-GPU training with up to four GPUs in its open edition, while its corporate edition supports extra GPUs. They are using a cloud version in conjunction with a server GPU-XEON-E5-8 GB for this project. The DLS’ architecture includes capabilities for project setup, transmitting data, newer versions, model training, model testing, and code formation. Drag the relevant dashboard panel to pick different deep learning algorithms. DLS makes it easy to create deep learning models quickly and easily. Jour et al. [2] The efforts to the pc-based system were digital pictures taken by ELM, with a possibility to feature alternative learning system like ultrasound or confocal microscopy. Within the 1st stage, preprocessing of those pictures was carried out that enables reduced ill-effects and numerous artifacts like hair that might exist within the lesion segments. It was then followed by the detection of the lesion by image segmentation method. Once the affected region was localized, completely different chromatic as well as morphological types may well be quantified and utilized for classification. Haenssle et al. [4] This dataset represented a spectrum of melanocytic lesions routinely observed in everyday clinical practice because nearly two-thirds of the benign naevi were non-excised lesions approved by continuous exams. The testset-300 photos were recovered from the University of Heidelberg’s Department of Dermatology’s high-quality valid image database. For image acquisition, a variety of dermoscopic sequences were used. There was no overlap between the training/validation and testing datasets. Maron et al. [3] Each participant was asked to review a hundred thirty pigmented skin lesions. Diagnostic test choices of dermatologists with no MelaFind versus MelaFind and dermatologists with MelaFind versus dermatologists with no MelaFind were matched. The MelaFind info, once merged into the ultimate diagnostic test, will improve diagnostic test sensitivity with modest impact on diagnostic test specificity. Dermatologists with no MelaFind had an average sensitivity to malignant melanoma of sixty nine percent and a normal specificity of fifty five percent. MelaFind had

86

D. K. Yashaswini et al.

bigger sensitivity than dermatologists only (96.9% versus 69.5%, one-sided p < 0.00001) and lesser specificity (9.2% versus 55.9%, one-sided p < 0.00001). Zalaubek Iet al. [7] Short-term SDDI was accustomed to watch the melanocytic lesions. As a result of a lentigo or ephelis sometimes may be enclosed among the medical diagnosis of lentigo maligna, a small number of patients with lentigo (n = 10) and ephelis (n = 1) experienced surgery and were included in the study. Despite a decent age distribution and a small sample size of individuals with these lesions, it was found to be effective. The age categories of 0–18 years had n = 0, 19–35 years had n = 5, 36–50 years had n = 3, and above 65 years had n = 1, suggesting that age had no bearing on the research. One in every of North American country has antecedently revealed that the history of a patient regarding the change in the nevus at preliminary appearance failed to associate with subsequent variations spotted through short-term SDDI. Hence, the result of short-term SDDI of benign melanocytic lacerations is freelance of gender, standard account of amendment in lesion, lesion distance, and substantial site. Merelyage impacts the result, having larger amendment taking place in youth fulera and adolescence (0–18 years, specificity of seventy five percent) and adulthood (>65 years, specificity of 77%). Erkol et al. [9] People were employed from the pigmented skin cancer clinic in urban center during 2000 and 2001. People with a minimum of ten melanocytic naevi were elect consecutively till a complete of ten people in every of 5 age teams was found. Age teams were taken as 0–15 years, 16–30 years, 31–45 years, 46– 60 years, and above 60 years. Digitized pictures of no inheritable melanocytic naevi, outlined as benign melanocytic productions showing a diameter of a minimum of five millimeter with a macular element and that were not apparent among the initial year of life, were valued by dermoscopic standards. The relations of dermoscopic features as a purpose of patient age were studied. Absolute numbers and frequencies were determined, provided as percentages, likewise as predominance of the dermoscopic styles of naevi within the various age teams.

3 Objectives The primary objectives are • Classification of the carcinoma images to denote the type of cancer with an improved degree of accuracy by the exploitation of deep learning algorithms. • To establish the recent analysis trends, challenges, and opportunities within the field of skin cancer treatment. • Investigate the present solutions for the diagnosis of cancer types and provide a scientific review of those solutions on the basis of similarities and variations.

Skin Cancer Classification Using Deep Learning

87

4 Existing System The first step convoluted in the diagnosis of a laceration by a specialist is visual examination of the suspicious skin house. It is found that a talented and the most experienced specialist sometimes carries out a sequence of known actions, beginning from pictorial inspection of symptoms suggestive, followed by dermoscopy (microscopically magnifying lesions), and finally a clinical examination. The manual review from dermoscopy descriptions created by dermatologists is typically long, erring, and subjective (even well trained dermatologists could turn out wide variable diagnostic results). During this regard, machine-controlled recognition approaches square measure extremely demanded. The start of dermoscopy has assisted an intense boost in clinical diagnosis with the purpose that malignant melanoma is spotted within the health center at the earliest of its stages. The worldwide acceptance of this technique has acceptable a huge collection of dermoscopy pictures of different carcinoma lesions that include both benign and malignant variants valid by histopathology. The event of advanced technologies within the areas of image processing and machine learning has given us the flexibility to permit a vibrant modification of various skin cancer variants that need no diagnostic test. These latest technologies ought to permit not solely on early detection of carcinoma, but also additional reduction of the massive range of unnecessary and expensive diagnostic test. Disadvantages of existing system • Time intense. • Costlier diagnostic test procedures. • Not enough well trained dermatologists to detect carcinoma accurately.

5 Proposed Approach An early diagnostic algorithmic program targeted on deep convolutional neural networks that with efficiency differentiates between different skin cancers variants. Its architecture has been shown in Fig. 1. Apart from image classification into various skin cancer types with the aid of CNN and data augmentation, we would prefer to convert the model into a web app. The essential style of the web app would enable the dermatologist to upload a picture of the carcinoma kind, and therefore, the model designed would classify the image into acceptable cancer kind. The complete pipeline about the image classification is as follows • Our input could be a training dataset that consists of N pictures, every labeled with one among K completely different categories. • We train a classifier using this training set to be told what all of the categories seems like.

88

D. K. Yashaswini et al.

Fig. 1 Architecture of CNN

• On the top, we tend to measure the degree of the classifier by making it to predict labels for a brand new set of pictures that it has never met previous. We would then match the truth labels of those pictures to those foreseen by the classifier. A. Pseudocode Pseudocode is an off-the-cuff sophisticated description of the operational norm of a bug or alternative algorithmic rule. It uses the structural practices of a programing language, however, it is meant for human understanding instead of machine analysis. Pseudocode generally overlooks facts that do not seem to be vital for human intelligence of the algorithmic rule, like variable declarations, system-specific code, and a few subroutines. The programing language is increased with tongue description details, wherever suitable, or with solid mathematical notations. The aim of pseudocode is that it is easier for the society to know than a standard programing language code. B. Data preparation and building a deep learning classifier The practice of tidying and reworking data before processing and analysis is data preparation. It is considered as one of the vital steps before data processing and sometimes includes data reformatting and making improvements to data and also the dataset combining to balance the data that is needed. For data experts or business users, data preparation can be a lengthy process, but it is necessary to set data in context in order to turn it into meaningful insights and eliminate bias caused by poor data quality. A neural network with several hidden layers is known as a deep learning neural network. The following are some characteristics of building a neural network model: • • • • •

Number of layers Types of those layers Number of units (neurons) in every layer Activation functions of every layer Output layers images are resized to a particular size for the model. This is mainly done for the model to work efficiently.

Skin Cancer Classification Using Deep Learning

89

C. Exploratory Data Analysis Approach exploratory data analysis (EDA) is a nurturing approach/viewpoint for data analysis that employs a wide range of methods (mostly graphical) to maximize insights into an data set, and reveal the primary structure, extract the required variables, identify outliers and anomalies, test the underlying expectations, build penurious models; and determine optimum subject settings. EDA conjointly supports various stakeholders by approving they are enquiring the proper queries. EDA will facilitate us to answer such questions about customary deviations, categorical variables, and confidence intervals. Once EDA is finished and insights area unit drawn, its options will then be used for a lot of refined knowledge analysis or modeling, as well as machine learning. Some of the exploratory data analysis that was performed on our dataset, to obtain the graphical representations of insights of the data shown in Fig. 2 and Table 1. 250 200 150 100 50 0 Infectious disease

Benign disease 82

Age realated skin disease 31.7

others

43

Fig. 2 Disease distribution over gender

Table 1 Location of diseases over gender Disease gender

Total No. (259)

%

Male No.

%

Female No.

%

P value

Squamous cell carcinoma

82

31.7

43

31.6

39

32

NS

Infectious disease

120

46.3

59

43.1

61

50

NS

Benign disease

177

68.3

95

69.3

82

67.2

NS

Age-related skin disease

222

85.7

118

86.1

104

85.3

NS

Others

164

63.3

82

59.9

31

67.2

NS

90

D. K. Yashaswini et al.

D. Tuning the Model and checking Results Convolutional neural network involves building multiple blocks, like convolution layers, pooling layers, and absolutely connected layers and is intended to mechanically and adaptively learn special hierarchies of options through a backpropagation rule. Deploying the Model Once the model was built and trained to predict appropriate skin cancer variant using CNN and data augmentation, the model was later converted into a simple web app. The essential style of the web app would permit a dermatologist to upload an image of the unknown or doubtful carcinoma type and the model designed would classify the image into applicable cancer variant. Expected outcome It is observed from the results that the trained system is effectively utilized by physicians to diagnose the carcinoma with a greater degree of accuracy in the initial stages of cancer (Fig. 3). Since the tool is developed to be lot user friendly for image classification, it will serve as an automatic medicine of the skin cancer detection. The model trained was found to have an accuracy of up to 0.9679, i.e., 96.79%. The loss in the model was summed to be 0.1686, i.e., 16.86%. • The basic design of web app would allow dermatologist to upload images and the model built would classify images accordingly (Fig. 4). • Henceforth, the app was able to fulfill the requirements of the dermatologists. Fig. 3 Analyzing the accuracy of the model

Skin Cancer Classification Using Deep Learning

91

Fig. 4 Classified dermal image

6 Conclusion Outcome of this project, we have mentioned varied ways to identify and classify skin carcinoma lesions like convolutional neural network and data augmentation. It may be complete from the results that the trained model may be effectively utilized by physicians to diagnose the cancer type a lot more accurately. This tool would be a lot more helpful in the areas, where the specialists within the medical field might not be able to detect appropriately. Since the tool is designed to be user friendly for image classification, it will serve as an automatic medical specialty of the carcinoma Apart from image classification into variants of skin cancer via CNN and data augmentation, we have further modified the model into a web app. The essential style of the web app would permit a doctor particularly a dermatologist to upload an image of the carcinoma variant and the model engineered would classify the image into applicable cancer type.

References 1. Muhammad Qasim Khan AH (2019) Classification of melanoma nevus in digital images for diagnosis of skin cancer. IEEE Access 7:86–91, 90132–90144

92

D. K. Yashaswini et al.

2. Jour (2013) Computer aided diagnostic support system for skin cancer: a review of techniques and algorithms. Int J Biomed Imaging 13:1–22 3. Maron RC (2014) To excise or not: Impact of MelaFind on German dermatologists’ decisions on biopsy atypical lesions. J Ger Soc Dermatol 12(7):606–614 4. Haenssle HA (2018) Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. J Ann Oncol 29(8):1836–1842 5. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S (2017) Dermatologist level classification of skin cancer with deep neural networks. Nature 115. https://doi.org/10. 1038/nature21056 6. Gehrke J (2009) Classification and regression trees. IGI global; 2009, encyclopedia of datawarehousing and mining, 2nd ed. https://doi.org/10.4018/978-1-60566-010-3.ch031 7. Menzies SW (2011) Variables predicting change in benign melanocytic nevi undergoing shortterm dermoscopic imaging 147(6):655–659 8. Mohammad Ali Kadampu SR (2020) Skin cancer detection: applying a deep learning based model driven architecture in the cloud for classification of dermal cell images. J Inf Med Unlocked 19:1–6 9. Zalaubek I (2006) Age-related prevalence of dermoscopy patterns in acquired melanocytic Naevi. Natl Libr Med Sci 154(2):299–304. https://doi.org/10.1111/j.1365-2133.2005.06973 10. Zorman MMM, Kokol SP, Malcic I (1997) The limitations of decision trees and automatic learning in real world medical decision making. J Med Syst:403–15 11. Larsen K (2005) Generalized neaıvebayes classifier. In: ACM SIGKDD explorations newsletter, 7 of 1. USA: ACM, pp 76–81. T

Crop Yield Prediction Using Deep Learning K. Mamatha, Shantideepa Samantha, and Kundan Kumar Prasad

Abstract Agriculture provides a living for around 58% of India’s population. Agriculture, forestry, and fisheries were expected to generate |19.48 lakh crore in FY20. Given the significance of agriculture in India, farmers might benefit from early forecasting of agricultural yields. The study focuses on predicting agricultural yield, for Karnataka state using the regression with neural network model. The final constructed dataset takes parameters like agricultural area, crop, taluka, year, season, district wise annual rainfall (mm), district wise maximum and minimum temperature (°C) and harvest or yield for the time period of 1997–2017. The underlying model is built utilizing a Multilayer Perceptron Neural Network, a ReLu Activation function, an Adam Optimizer, and 50 epochs with a batch size of 200. The end of the training gained 96.43% accuracy on test data. Several additional well-known regression algorithms such as Multinomial Linear Regression, Random Forest Regression, and Support Vector Machine are also constructed and trained using the same dataset so as to compare their performance to the base model. From the final comparison results it was found that neural network model has outperformed classic machine models for crop yield prediction in terms of both mean absolute error and accuracy. Keywords Neural network · Support vector regression · Random forest regression · Linear regression · ReLu · Adam optimizer

1 Introduction Agriculture in India stretches back to the Indus Valley Civilization Era, and maybe much earlier in some regions of Southern India. In terms of agriculture output, India stands second in the world. While agriculture’s proportion of the Indian economy has gradually decreased according to the fast development of the industrial and service sectors, to less than 15%, the sector’s relevance in India’s social and economic fabric

K. Mamatha (B) · S. Samantha · K. K. Prasad Department of ISE, Don Bosco Institute of Technology, Bengaluru, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_9

93

94

K. Mamatha et al.

stretches much beyond this statistic. The reason for this deterioration in the agriculture industry is because farmers are not empowered, and there is a lack of application of information technology in the farming sector. Farmers are less knowledgeable about the crops they cultivate. We typically overcome this challenge by utilizing appropriate deep learning algorithms to forecast crop output and name based on a variety of parameters like as temperature, rainfall, season, and location. Based on the data source supplied by the Indian government, this study presents a Neural Network system to forecast agricultural production and crop success rate. The main challenge encountered when assembling the work was the lack of a single source dataset to train the suggested model on. To address these issues, all dispersed data is collected and relevant feature engineering and data preprocessing steps are employed. The data source is massive, comprising records for all the areas of India that were sieved and processed to acquire records for Karnataka state, resulting in 12,961 entries. The cycle of the crop data for summer, Kharif, Rabi, fall, and the entire year is used. To obtain records for the state of Karnataka, the dataset is pre-processed using Pandas and profiling tools of pandas in Python. The crop yield forecast model employs an artificial neural network’s back propagation technique. The technology of multilayer perceptrons is employed. The proposed work has a wide range of applications in improving real-world farming conditions. Every year, a large amount of crop is damaged owing to a lack of understanding of weather patterns such as temperature, rainfall, and so on, which have a significant impact on crop output. This initiative not only aids in forecasting these characteristics throughout the year, but it also aids in projecting agricultural yields in various seasons based on historical trends. As a result, it enables farmers to select the best crop to plant in order to incur the fewest losses. Different regression models are also constructed using machine learning, and their efficiency and accuracy are compared to the Neural Network model in order to provide some tangible results.

2 Related Work This study examines the rice productivity and sustainability producing regions that are reliant on acceptable climatic conditions by deducing experimental findings acquired by using an SMO classifier employing the WEKA tool to a dataset of 27 districts in the state of Maharashtra [1]. The suggested work uses soil and PH samples as input and predicts crops that are suited for the soil and fertilizer that can be utilized as a solution in the form of a webpage. So, soil information is gathered using sensors, and the data is transferred from the Arduino through Zigbee and Wireless Sensor Network (WSN) to MATLAB, where it is analyzed and processed using Artificial Neural Network (ANN), and crop suggestions are made using SVM [2]. The major aim for this research is to look at the linkages between the climate and the yield of a crop and the historical trends and changes that has an impact in climate-related factors on finger millets, rice, wheat and maize production in Nepal’s Morang region [3]. This study intended to evaluate these novel data mining tools and

Crop Yield Prediction Using Deep Learning

95

finger millet finger millet finger millet tools and mechanisms and enroll them to the database’s numerous characteristics to see whether relevant correlations could be discovered [4]. The major goal of this research was to test a strategy for incorporating satellite imaging characteristics into A crop development methodology was built to approximate the wheat yields during spring season at the macro level and computed separately [5]. This study forecasts the suitable crop using crop yield prediction algorithms that detect distinct soil characteristics and meteorological condition factors. It illustrates the artificial neural network algorithm’s capacity to monitor and predict agricultural production in remote and rural locations [6]. The goal of this study is to conduct a comprehensive assessment of the current literature on DL methods in BDA. The findings might aid researchers in developing double-armed prognostic studies that properly evaluate the potential of the major elements of Deep Learning techniques, study the effectiveness of Deep Learning approaches in increasing Big Data Analysis, and investigate the benefits and drawbacks of Deep Learning approaches [7]. In this study, researchers used a Multi-Layer Perceptron, back propagation based on feed forward deep neural network to forecast wheat production (ANN) [8]. The suggested model is an example of a more sophisticated model that uses a highly complex set of variables to forecast wheat production. ECOSYS and SIRIUS, like CERES, are sophisticated models that integrate a multitude of factors and heavily rely on computer modeling to anticipate wheat growth [9].

3 Proposed Idea Figure 1 depicts the modules constructed in the proposed work. It is made up of three major components. One module is dedicated to early yield forecast, utilizing characteristics such as crop area, yearly rainfall, temperature data, and Karnataka state output history from 1998 to 2014. The second module is the fertilizer recommendation system, which takes into account the quantity of three major nutrients in the soil, namely, nitrogen, phosphorus, and potassium, as well as the crop to be sown, and suggests the fertilizer that may be utilized to increase crop output. The third module is a crop-specific WPI trends indicator that depicts the whole index price over the following 12 months graphically. A. Dataset The data for this study was obtained from the Indian government’s website. The datasets are freely accessible for research and scholarly purposes. The collection contains information spanning the years 1997 to 2017. All the required datasets are collected and pre-processed to obtain a final dataset which will be used to train the model. For the experiment in this investigation, the following parameters are used. Crop State

Rice, Wheat, Castor seeds, Bajra, Arhar, Season, groundnut, cottonseed, tur, and other crops are included in the dataset. Karnataka.

96

K. Mamatha et al.

Yield Predictor

• Takes Annual Rainfall, Maximum Temperature, Minimum Temperature, District in Karnataka, Crop Season, and crop to be sown as input and predicts the yield of the crop

Fertiliser Recommendation System

• Takes in the three main nutrients present in the soil, i.e. N, P, K values and the crop to be sown as inputs and suggests a suitable fertilizer that can be used to enhance production

WPI trends Indicator

• Provides graphical representation of whole price index trends of a commodity for the following 12 months.

Fig. 1 System modules

District

‘Bagalkot’, ‘Bengaluru Rural’, ‘Bellary’, ‘Belgaum’, ‘Chikmagalur’,‘Chithadurga’, ‘Dharwad’, etc. Season Autumn, Summer (Kharif), Winter (Rabi), and Yearly. Year 1997 to 2017. Rainfall Monthly rainfall data (mm) for each district of Karnataka State, whose sum is taken to evaluate annual rainfall and concatenated to the final dataset. Temperature District wise maximum and minimum temperature (°C), who’s mean is calculated and appended to the final dataset. Production It is expressed as tons per hector in million. Fertilizers Describes the amount of N, P, K required in the soil in order to grow a specific crop in a region. The data in the government dataset has been checked for outliers and anomalies. The parameters were also translated to numerical and category formats to meet the model’s requirements. Figure 2 displays a bar graph from 1998 to 2014 that correlates the season and produce of all Karnataka districts. According to the graph below, the bulk of the crops cultivated in Karnataka are year-round crops. B. Methodology To integrate all of the data sets obtained for this investigation, Microsoft Office Excel was employed. Step 1: Acquiring monthly average records from Indian Government databases for each parameter (rainfall, minimum, median, maximum temperature, and reference crop transpiration) from 1997 to 2017. Step 2: Calculate the total precipitation for each region during the summer (Kharif) season each year (July to November). Step 3: Computing the average temperature for each region all throughout summer (Kharif) season for the minimum, average, and maximum temperatures annually.

Crop Yield Prediction Using Deep Learning

97

Fig. 2 Year-specific bar graph to depict the correlation of season versus production

Step 4: During the Kharif season, calculate the average reference crop evapotranspiration for each year for each district in Karnataka state. Step 5: Obtaining information on each district’s acreage, output, and rice crop yield from 1997 to 2017, based on publicly accessible Indian Government sources and databases. Step 6: The raw collected data was then compiled in Excel Spreadsheet into a single document with the following columns: name of the region, year, rainfall, minimal temperature, average temperature, maximum temperature, area, production, and yield. Step 7: Because specific climatic characteristics of a year or harvest quantity statistics were not accessible for some of the districts, those records were removed. The data from that specific year was not included in the current study. Each record now has a record number. Step 8: Unrequited columns were deleted from the data set in order to prepare it for use with the multilayer perception method. They were name of the district, and year. Step 9: All the records was then arranged according to area. The current study did not take into account areas less than 100 hectares. As a result, those records were eliminated. Step 10: Using sklearn, the attribute values in the label is transformed to encoding. Step 11: The entire data source was then sorted based on the date of production. Step 12: The study then evaluates harvest yield as an output parameter, as well as characteristics such as type of crop, cultivation area, taluk, and season. Step 13: The final dataset was then sorted and compiled into a.csv file for iterative use in Python Tensor Flow with the multilayer perceptron method.

98

K. Mamatha et al.

Fig. 3 Dataflow diagram of the proposed system

User Input

Data Preprocessing

Encoding Categorical Data

Prediction output

Feeding Data into the Neural Network Model

Step 14: Model is trained in this step. Using linear regression with a neural network with three layers and the Adam optimizer. Step 15: Using an 80:20 split, the data source was segregated into test and train sets.

4 Data Flow Diagram The diagram below depicted in Fig. 3 shows the flow of data through the system. The flow of all modules stays constant, with the only variation being the final result. Inputs for the relevant modules, such as yearly rainfall, temperature, district, crop name, season, and fertilizer data, are obtained via a web-based application by the user. A JSON data object is returned, which has been scaled with the sklearn package. The categorical data, such as district, season, and crop name, is again one hot encoded, and the data object is ultimately transformed to a numpy array. This information is subsequently put into the Neural Network model. Artificial Neural Network—An Overview An Artificial Neuron is essentially a biological neuron engineering technique. It has a gadget with several inputs and just one output. As shown in Fig. 4, the ANN is made up of a huge number of basic processing units that are linked and stacked [10]. An ANN starts with a phase of training in which It begins to identify patterns in data, whether they are visual, audio, or linguistic. During this supervised phase, the network compares its expected results to what it was supposed to produce—the projected output. Back propagation is employed to reconcile the disparity between the two findings. This implies that the system operates backward, from the output unit to the input units, adjusting the weight of its links between the modules until the gap between the estimated and planned results produces the smallest feasible error.

5 Training Specifications The following are the important parameters that were evaluated for testing:

Crop Yield Prediction Using Deep Learning

99

Fig. 4 Layers and connections of ANN model

• • • • • • • • •

Layer: 3 Neuron at each layer: Layer 1, Layer 2 = 20 Layer 3 = 1 Batch_size = 100, Activation: ReLu Oprimizer = Adam Bpochs = 50 Kernel_initializer = ‘uniform lr rate: 0.01.

6 Performance Evaluation We utilize scatter plots to compare the actual test data output to the predictions generated by the model on the test data. The graph below illustrated in Fig. 5 shows a linear connection between the actual and predicted results. A positive slope with a strong correlation between the actual and anticipated results indicates a greater success rate. It can also be stated that for the majority of the test inputs, the model was able to forecast a yield with extremely low error to the actual yield output. The algorithm’s performance is measured using the two metrics listed below. (A) Mean Absolute Error MAE is a weighted average magnitude of absolute differences between N estimated vectors S = x1, x2,…, xN and S_ = y1, y2,…, yN, with the associated loss function calculated as: N ||xi − yi || where || ・ || denotes L1 norm [11]. L MAE (S, S ∗ ) = N1 i=1 (B) R-Squared

100

K. Mamatha et al.

Fig. 5 Linear correlation between actual and predicted results

R-squared (R2 ) is a quantitative measure that reflects the percentage of a dependent variable’s variance explained by an independent variable or variables in a regression model. Whereas correlation shows the significance of the association between an independent and dependent variable, R-squared represents how well the fluctuation of one variable explains the variation of the other. The fraction of a fund’s or security’s movements that can be accounted by changes in a standard deviation is typically described as R-squared. The R-Squared score for the proposed work was able to reach approximately 0.9645 within 50 epochs. The score can be calculated using the formula below: R2 = 1 −

RSS TSS

where R2 is coefficient of determination RSS is Sum of Squares of residual TSS is total sum of squares The graph of Fig. 6 depicts the metric scores discussed above for our neural network model.

7 Experimental Results In the current study, we collected multiple datasets and performed appropriate feature engineering to build a single source of data that accounts for all of the essential features to help model correctness. To provide a comparison study of our neural network model’s performance, we utilize the same dataset to train three additional regression models, namely, the Multinomial Regression model, the Random Forest

Crop Yield Prediction Using Deep Learning

101

Fig. 6 Performance graphs of the model

regression model, and the support vector regression model. All three models’ performance was evaluated using the same two measures described above: mean absolute error and R-Squared score. The bar graph of Fig. 7 compares the performance of all three models, as well as our targeted model, the neural network model. It is clearly shown that the Artificial Neural Network outperforms the other regression methods in terms of accuracy and error rate.

Fig. 7 Comparative analysis of the performance of all the regression models

102

K. Mamatha et al.

8 Discussion and Conclusions A non-linear way of interpreting the connection is necessary to demonstrate the interactions between the factors impacting crop production. Because of the complexities of the elements influencing crop output, a linear approach such as linear regression was judged insufficient to depict the interconnections between the factors and crop yield. For forecasting agricultural yield, ANN was thought to be a viable alternative to standard regression methods. A neural network not only predicts non-linear correlation successfully, but it can also recognize complicated patterns in data and train appropriately, something most traditional approaches fail to do.

References 1. Gandhi N, Armstrong LJ, Petkar O, Tripathi A (2016) Rice crop yield prediction in India using support vector machines. In: 2016 13th international joint conference on computer science and software engineering (JCSSE) 2. Preethi G, Rathi Priya V, Sanjula SM, Lalitha SD, Vijaya Bindhu B (2020) Agro based crop and fertilizer recommendation system using machine learning. Eu J Mol Clin Med 7(4), ISSN: 2515–8260 3. Khanal B (2015) Correlation of climatic factors with cereal crops yield: a study from historical data of Morang District, Nepal. J Agric Environ 16 4. Raorane AA, Kulkarni RV (2013) Review—role of data mining in agriculture. (IJCSIT) Int J Comput Sci Inf Technol 4(2):270–272, ISSN: 0975-964 5. Doraiswamy PC, Moulin S, Cook PW, Stern A (2003) Crop yield assessment from remote sensing. Photogramm Eng Remote Sens 69(6):665–674 6. Priyanka T, Soni P, Malathy C (2018) Agricultural crop yield prediction using artificial intelligence and satellite imagery. Eurasian J Anal Chem 13(SP), 6–12, ISSN: 1306-3057 7. Hordri NF, Samar A, Yuhaniz SS, Shamsuddin SM (2017) A systematic literature review on features of deep learning in big data analytics. Int J Adv Soft Comput Appl 9(1), ISSN: 2074-8523 8. Kadir MKA, Ayob MZ, Miniappan N (2014) Wheat yield prediction: artificial neural network based approach. In: 2014 4th International Conference on Engineering Technology and Technopreneuship (ICE2T) 9. Brooks RJ et al (2001) Simplifying Sirius: sensitivity analysis and development of a meta-model for wheat yield prediction. Eur J Agron 14:43–60 10. Mishra M, Srivastava M (2014) A view of artificial neural network. IEEE Int Conf Adv Eng Technol Res (ICAETR—2014) 11. Qi J, Du J, Siniscalchi SM, Ma X, Lee CH (2020) On mean absolute error for deep neural network based vector-to-vector regression. IEEE Signal Process Lett 27

Real-Time Tweets Streaming and Comparison Using Naïve Bayes Classifier S. R. Shankara Gowda, Rose King, and M. R. Pavan Kumar

Abstract The rapid growth of social media sites in recent times has introduced a special environment for researching human actions. Micro blogging website (Twitter) which also allows users to access and communicate their opinions on various of topics, occurrences, brands, and service providers. Tweets have been classified into different subgroups relating to the subject of the research. Various machine learning algorithms, such as baseline, Naive Bayes classifier, support vector machine (SVM), and many others, are presently used this to categorize posts on Twitter into favourable and unfavourable groups based on their views and opinions. This paper describes a solution to improve comparison between Twitter accounts besides trying to implement Naive Bayes using perception vibrant data for training from of the Twitter database. SentiWordNet, in combined effect with Naïve Bayes classifier, can help increase tweet accuracy rate by providing hope and optimism, hatefulness, and integrity ratings for phrases in Twitter messages. Tweepy, which is a Python package and Python-Twitter APIs are used in the actual introduction of the new system. The main work of this paper is to perform real-time streaming, collect the data set while performing real time and make comparison between them using Naïve Bayes. Keywords Biometric verification · Twitter · Tweets · Naïve Bayes

1 Introduction The most fundamental requirement for improved activity and resource security is user identification. The typical approaches of username and password are PINs, credentials, and identification cards. The majority of “strong” credentials or prongs are now easily detected, so according reports on the security of traditional diagnostic techniques [1–3]. Login credentials, prongs, and identity cards are indeed corrected immediately, stolen, and misplaced. Individual physiological characteristics can be used in multifactor authentication user identification systems to counteract these S. R. S. Gowda (B) · R. King · M. R. P. Kumar Department of Information Science and Engineering, Don Bosco Institute of Technology Kumbalgodu, Bengaluru 560074, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_10

103

104

S. R. Shankara Gowda et al.

drawbacks. Face, fingerprints, iris, and voice, and also some gait, serial number, and other behavioural features, cannot be published, hijacked, or copied. When compared to physiological biometrics, user recognition is more amenable to life changes and has the advantages of being flexible, relatively inexpensive, and efficiency [1, 2]. Because of their rapid evolution, activity recognition is difficult to tackle and replicate. It is far easier to generate a synthetic fingerprint than it is to try to emulate a human’s trying to walk style or posture. By analysing behavioural patterns in online user interactions and measuring their performance for user identification, this work reinforces the concept. Because of their rapid evolution, activity recognition is simple to manipulate but also reproduce. Recreating a user’s trying to walk aesthetic or posture is far more difficult than cultivating a duplicate fingerprint. By defining behavioural patterns in online user interactions and evaluating their performance for user identification, this paper validates the concept. Twitter is a popular social media platform that enables us to communicate with each other at all and. According to a national bureau of statistics, Twitter has approximately 5.6 billion users worldwide and sends out 3.1 million tweets per day at a rate of 9,500 tweets per second. In addition, every day, Twitter tends to attract 2.5 million new users. As a result, the paper’s biggest purpose is to use graded user behavioural features to construct a tool to identify users by matching such ways of thinking from interactive Twitter data. We would like to continue our research to examine the effect of feature matching over time to discover how durable such behavioural factors are.

2 Problem Definition We proposed a framework in this paper for identifying instead of dynamic descriptive statistics. People on Twitter are ranked based on their flexible interactions. The steps in our proposed methodology are as follows: 1. Obtain tweets from a set of user profiles. 2. Extract interactive data from tweets like replies, retweets, hashtags, and URLs. 3. Start investigating social behavioural biometric features by analysing acquired user data. 4. For person authentication, use machine learning techniques on the training and test sets.

3 Literature Survey Applying user recognition on Twitter is not an easy task. The problem in user recognition is classifying the polarity of the given text document, sentence, or feature/aspect level. For performing data extraction, we require Twitter data containing tweets

Real-Time Tweets Streaming and Comparison …

105

pertaining to a particular keyword or query term. We have used Tweeter application programming interface, which really is free to the public, to gather data and messages. Guo s’est cetera [4] began to look into whether a person’s “handshaking” style could be used to recognize them. Shaking hands is a succession of humankind that should be carried out in order to unlock a user’s phone screen [4] geographic website, steady, and differentiated eccentric trends in users’ shaking hands routines that they used to verify mobile phone users. Movement of trying to pick up a virtual device was mentioned by Feng comme et al. in [5] and Jiang alors que et al. [6] proposed a leisure activity biometric as a different type of actions facial recognition. The above study is a serious examination into leisure action forecast period. A few research on text classification on various online information collected, especially Twitter, have also been undertaken, with a few significant elements a certain assist in the finding of online trust or opinions in many circumstances when epidemics attack across the universe. A list of valuable possessions that have been used as allusions could be discovered in just this segment. Researchers utilized data from Twitter to see if they would anticipate seasonal flu as well as other contagious diseases in real-time basis [1]. By reviewing Twitter message search terms and monitoring infection prevalence rate and risk reduction in real-time basis, researchers were able to determine the geographical factors in a 2009 infectious disease outbreak [2]. It is not easy to implement user recognition on social media. The complication in user recognition is determining the polarity of a text document, sentence, or functionality level. During the 2014 Swine flu outbreak, social media users shared vital health information from news organizations, to peak social media activities organized in 24 h of current events. [3]. Kaur and Sharma [7] investigated how people were feeling well about COVID19 coronavirus epidemic, investigating how various individuals felt about it. As a result, useful coronavirus tweets were collected using the Twitter data, which would then be evaluated utilizing machine learning approach related to positive, negative, and neutral emotions. The NLTK library has also been used to preprocess the fetched tweets by both the authors. The impact of COVID-19 on the world has been investigated and envisioned in [8] by monitoring favourable and unfavourable intensions from individuals around the globe using machine learning techniques and computational methods in object recognition on the Training data. Over a threemonth period of January 6 to January 30, 2020, the person who wrote this article of [9] published a collection of COVID-19-related snapchat filters to look for relevant Twitter posts. The platform is used to obtain tweets in plain text file, which would then be rescued. Samuel et al. [10] Exemplify assumptions in to the COVID-19 trying to reach its highest degree in the Americas using text-based advanced statistics and suitable textual data visualization. In the research area of trend analysis, the article gives a theoretical knowledge of two important machine learning techniques methods and compares their effectiveness in order to classify coronavirus Twitter messages of variable sizes. In [11], the authors have proposed an effectual platform for collection

106

S. R. Shankara Gowda et al.

of data, warehousing, strategic planning, mineral extraction, as well as other activities. De Choudhury et al. [12] Study examined at just how people’s feelings evolved over time using Twitter messages. Recognizing the emotions associated with both the messages being analysed is among the most interesting results obtained from text-based data analysis [13]. Algorithms can use additional data from across all training samples to anticipate each part of the reason component.

4 Proposed Work Machine learning is the best technique in research field due to its accuracy in results. In the user recognition, mainly supervised technique and natural language processing are used. It has three stages: exploratory data analysis, data preprocessing and future engineering, vectorization, model validation and results. The below mentioned steps describe our proposed methodology (Fig. 1). First Step: Data analysis—In this stage, we will do some data analysis of the machine learning problem to get some insights.

Fig. 1 Block diagram of tweet extraction

Real-Time Tweets Streaming and Comparison …

107

Second Step: Data preprocessing—The data set has a much of the unstructured tweets (Fig. 2) which has to be preprocessed to make the natural language processing model. We implemented the following techniques to preprocess the raw data. 1. Removal of punctuations 2. Removal of stopwords 3. Normalization of words Third Step: Vectorization and model selection—Two techniques for vectorization of data are counter vectorization and Tf-Idf transformation. Counter vectorization mainly generates a sparse matrix with all the words in the document Tf-Idf-term frequency that defines the number of occurrences in term t and document d (Fig. 3 and Fig. 4). Fourth Step: Model validation—Since this is a natural language processing, we have validated the existing data set, we have implemented conventional split technique

Fig. 2 Unstructured tweets with hashtags, URL, emoji, etc.

Fig. 3 Real-time tweet streaming after removing hashtags, URL, emoji, etc.

108

S. R. Shankara Gowda et al.

Fig. 4 Real-time graph using Matplotlib

Fig. 5 Confusion matrix using Naïve Bayes

and pipeline model to validate data sets accuracy is measured using confusion matrix and classification (Fig. 5). Fifth Step: Results—In the result section, we are comparing the user based on their tweets and showing them through bar plot graph as shown in Fig. 6. System Design For the system design, we need to have a Twitter application and a Twitter developer account. We are making use of Tweepy, which is an easy approach to use Python library package (Fig. 7). Tokenization: All unnecessary symbols and numeric values are removed and returned a pure list of words. Normalization: Abbreviations are replaced with actual words. Next is the classification algorithm, we have used Naïve Bayes classifier to classify the tweets.

Real-Time Tweets Streaming and Comparison …

109

Fig. 6 Comparison between two users using Naïve Bayes

Fig. 7 Flow chart of user recognition

Input Text (keyword)

Result in Pie-chart

Tweets Retrieval

Classified Tweets

Data Preprocessing

Naïve Bayes Algorithm

5 Conclusion and Future Enhancement Machine learning approaches have been so far good in getting the perfect results than the other approaches that are discussed. Our classification approach provides a good improvement in perfection by using the simple feature and data set. However, there are still a number of things that we would like to consider as future work. The task of extraction and comparison, in the domain of machine learning, is still in the developing stage and far from complete. We have extracted or streamed live tweets from the Twitter user using keys provided by the Twitter developer account and authenticated them by enabling the Oauth authentication by providing our organization detail.

References 1. Choriano PK, Talves K (2016) Flu track.org: open-source and linked data for epidemiology. J Health Inf 2. Signorini A, Segre AM, Polgreen PM (2011) The using twitter to track disease activity and public concern in the united states during the H1N1 influenza pandemic 2011. PLoS ONE

110

S. R. Shankara Gowda et al.

3. Househ M (2016) Communicating Ebola through social media and electronic news media outlets: a cross-sectional study. Health Inf J 22:470–478 4. Guo Y, Yang L, Ding X, Han J, Liu Y (2013) OpenSesame: Unlocking smart phones through handshaking biometrics. In: Proceedings of IEEE IN FOCOM. IEEE, pp 365–369 5. Feng T, Zhao X, Shi W (2013) Investigating mobile device picking-up motion as a novel biometric modality. In: The IEEE 6th global forum on authentication has published its proceedings: theory, applications and systems (BTAS), pp 1–6 6. Jiang W, Xiang J, Liu L, Zha D, Wang L (2013) From mini house game to hobby-driven behavioral biometrics-based password. In: Proceedings of the IEEE global forum on confidence is in its 12th year, security and privacy in computing and communications (TrustCom). IEEE, pp 712–719 7. Kaur C, Sharma A (2020) Twitter sentiment analysis on coronavirus using Textblob. EasyChair, pp 2516–2314 8. Ra M, Ab B, Kc S, COVID 19 Outbreak: Tweet based Analysis and Visualization towards the Influence of Coronavirus in the World 9. Medford RJ, Saleh SN, Sumarsono A, Perl TM, Lehmann CUJm (2020) An ”Infodemic”: leveraging high volume twitter data to understand public sentiment for the COVID 19 outbreak 10. Samuel J, Ali GG, Rahman M, Esawi E, Samuel Y (2020) COVID-19 public sentiment insights and machine learning for tweet classification corona virus public sentiment insights and machine learning for tweet classification documentation. In: COVID-19 public subjectivity natural language processing and perspectives for hashtag categorization 11. Carvalho JP, Rosa H, Brogueira G, Batista F (2017) MISNIS: an intelligent platform for twitter topic mining. Expert Syst:374–388 12. De Choudhury M, Counts S, Horvitz E (2013) Predicting postpartum changes in emotion and behavior via social media. In: Proceedings of the SIGCHI conference on human factors in computing systems, held in Paris, France, pp 3267–3276 13. Samuel J, Holowczak R, Benbunan-Fich R (2014) Automating dominance discovery in synchronous computer-mediated communication. In: Proceedings of the 2014 IEEE Hawaii international conference on system sciences, pp 1804–1812 14. Jorgensen Z, Yu T (2011) “On mouse dynamics as a behavioural biometric for authentication”, Sixth Session Crit. ACM Symp Inf Comput Commun Secur 2011:476–482 15. Bours P (2012) Continuous keystroke dynamics: a different perspective towards biometric evaluation Inf Secur Tech Rep 17(1):36–43 16. Frank M, Biedert R, Ma E, Martinovic I, Song D (2013) Touchalytics: On the applicability of touchscreen input as a behavioural biometric for continuous authentication. IEEE Trans Inf Forensics Secur 8(1):136–148

Smart Shopping Trolley for Billing System R. Kishor Kumar, V. Ashwitha, S. Jeevitha, P. Pranusri, and D. Rakshitha

Abstract As the technology is developing day by day, shopping malls should be capable of handling the crowd smartly. The various items are purchased in shopping mall or markets with the aid of shopping trolley. After purchasing, the customer needs to pay the bill in the counter. In that case, they have to wait for long time for the person in billing section to scan each and every product. This is why there has been an inevitable demand for quick and easy payment of bills in shopping places. This project is developed for the main purpose of saving the customers time and makes shopping much easier. This is based on Raspberry with a LCD and QR or barcode scanner and a wireless technology called Wi-Fi. The LCD used is a 20 × 4 and WiFi modules make the wireless network to work easily between a certain ranges. A brief description about the project is it produces a method wherein the customer can scan any product using the barcode scanner installed in trolley and the information is displayed on the LCD screen. The system is provided with four options such as add, multiple, delete, and stop button which can be used in the different case. The customer can scan any number of products, the cost is added to the final bill, and the customer gets the bill to their mail id given. Keywords Raspberry pi · Barcode scanner · LCD · Wi-Fi module · Shopping malls

1 Introduction Development and innovations are increasing in every part. These innovations and developments play significant role in modernization and encouraging life to be more rapid and stress-free. Human life has become one of the struggles. Currently, human lifestyle has been changed completely. People do not have time to go for a shopping and waste their time, so most of the people prefer to go for shopping mall where they can buy all the products at the same place. Shopping in malls actually not only saves R. Kishor Kumar (B) · V. Ashwitha · S. Jeevitha · P. Pranusri · D. Rakshitha Don Bosco Institute of Technology, Bangalore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_11

111

112

R. Kishor Kumar et al.

time but also saves roaming for different shops. These days human is dependent on technology which satisfies their needs accordingly. Human wants to decrease the stress and work by depending on technology. Approximately, a normal human spends 1–1.5 h for the shopping. This scenario can be of two types of shopping: shopping physically and shopping online. Shopping online does not require the customer he himself should be present in the shopping mall. It can be teleshopping, online shopping, and so on. Shopping physically includes customer has to go for the shopping mall and buy his needs, stand in a queue, and finish his shopping. Currently, in traditional method, people go for shopping mall and take a cart along with them. The customer puts all the products he wanted to buy into the basket and goes to billing section. So, at the exit point there will be many customers for the bill, this creates a long queue, and people have to stand in a queue and finish his shopping which is very time consuming so in order to avoid this scenario, the project is developed “Smart shopping trolley for billing system.” Here, in this project it reduces the customer time for standing in the queue and finishes the shopping as early as possible. Technology is increasing very rapidly so it has been wireless communication. So here, we are using Wi-Fi module because it uses less power and is cost effective. This project mainly addresses the issues like customer standing in a long queue for billing and avoiding the man power. The goal of smart shopping trolley for billing system is when customer enters into the shopping mall and takes a basket. As every product will be having an RFID tag, the customer he himself can scan the product by using RFID scanner which contains camera in it. A load sensor is designed to sense the load. When the product is scanned and put into the basket, the load cell senses the weight and matches with the product weight which is stored in the database. If the weight does not match with the actual weight, then it is intercepted in the case of discrepancy. The basket will be provided with add, delete, and multiple buttons. So the customer can use those buttons accordingly. After the customer finishes shopping, he/she can pay the bill as assisted and come out of the shopping mall.

2 Related Work Shopping nowadays is like purchasing the product and standing in a long queue for the product to be scanned and then checkout. This results in time delay for customers, and the technology currently utilized in checkouts is barcode scanning. Nowadays, barcodes are there in almost every product. Paper [1] describes implementation of electronic shopping trolley based on the android technology which was proposed. In this paper, an android technology was used to scan the barcode of the product. Paper [2] explains that at the billing session customer may experience many problems, they need to wait in long queues, and even if they exceed budget limit or do not have sufficient money in order to overcome these problems they implemented

Smart Shopping Trolley for Billing System

113

barcode design scanners and touch screen display which are used along with raspberry pi. Touch screen display is used to display the information of products such as product weight, product cost, and product name. Constant power supply is also needed in this case; here, switch buttons are used to add the item to the shopping trolley as well as delete the item from the trolley, and finally, the customer makes a payment through manually or online mode like PhonePe, Gpay, Paytm, etc. This will reduce the time while shopping and is customer friendly. Paper [3] describes an implementation of Radio Frequency Identification reader that is fitted into the trolley. In this paper, RFID tags need to be inserted in each and every product that has to be scanned. Here, they are using liquid crystal display; along with this, matrix 4 × 4 keypad is used to set the budget limit which is to alert the customer if he/she crosses the limit. Later, the customer can make the payment by its choice. Paper [4] explains that how the theft prevention can be controlled by barcode scanners and Wi-Fi module and load cell sensors are used to control the theft prevention. Load cell is fixed to the bottom of the trolley. It is used to synchronizing input from the load cell; here, the load cell matches the weight of the product with the weight stored in database at base station; if it not matches, it warns the customer by buzzer.

3 Objective The main objective of the project is to eliminate the long queues in the shopping place, to design a barcode scanner for the scanning of the products using raspberry pi, to automatically generate the bill, to prevent the theft in the shopping malls or supermarkets, and automatic scanning of items which requires less time and effort.

4 Problem Statement The problem in today’s system is that there is a huge crowd in shopping malls which results in long queues in billing system; especially during some special occasions or weekends, the place will be more crowded and the person at the billing system should scan all the products one by one and generate the bill which also requires more time and increases the waiting time of customer which makes them irritated. When customer goes for shopping, he may have planned for the shopping of some budget and he may exceed it in the shopping process.

114

R. Kishor Kumar et al.

5 Proposed Architecture Design To eliminate the problems in today’s system, the proposed system uses a Raspberry pi which is interfaced with the camera module, load cell, and other components to enable customers to self-scan the products in the trolley and then generate the bill which reduces time and also the budget setting option is provided so that limit of the customer will not be crossed. The system is fitted a load cell down the trolley for the matching of the scanned items which prevents the theft. Figure 1 shows the block diagram of the overall system. Raspberry pi is an in-built processor. To scan QR code of every product, we use camera that we have connected to the one end of the raspberry pi. The purpose of using switch module is to add or delete the products that we are connected to the raspberry pi. After scanning every product, the details of that particular product that is cost, weight, etc., are displayed on the LCD which is connected to the other end of the raspberry pi; the purpose of using load cell is to identify the weight of the every product which is also connected to the raspberry pi. The Raspberry we use is 3 model B with the operating voltage of 5 V, which consists of 40 pins but we use 28 GPIO pins, RAM, and the storage capacity of SD card is 16–32 gb. The camera used is with the processing frequency of 1.2 Ghz and 5 megapixel camera with the operating voltage of 5 V and 20 * 4 LCD display. Here, we use load cell with the operating voltage of 5 V and H * 711 load cell driver to read analog values. The function of using switch module is used for selecting the option from the cart where four buttons are provided as add, delete, multiple, and stop button. Figure 2 shows the flowchart of the system. To start the billing process, we first initialize the smart trolley by turning on the start button. Initially, the cart will be empty. Whenever the consumer makes purchase, they will search for barcode; if it Fig. 1 Block diagram

Smart Shopping Trolley for Billing System

115

is present, they have to scan that particular product. The related data of that product will be displayed on LCD which in turn will add the number of products to the billing system; as the product increases, amount will be incremented accordingly; then, upload key is pressed; data is sent to the counter; and now, customer can collect bill in the counter when customer is done with their shopping, and they can turn off smart shopping trolley.

Fig. 2 Process flow

116

R. Kishor Kumar et al.

6 Implementation and Results The pictures of the project model are shown in above Figs. 3, 4, and 5. The raspberry pi and the system are connected through Wi-Fi, and then, the execution starts and there are several cases in the project model where the customers want to buy only one item in the shopping mall, or customer wants the same product multiple items, or all different products; all the cases can be shown in working model. Case 1: Adding an item to the cart—customer picks the product from the mart, scans it through the camera interfaced, puts into the cart, and sets the stop button after which the shopping is ended, and the product name, total cost, and the bill are generated. Case 2: Adding multiple items to the cart—when customer needs the same product multiple items, he scans the product and sets multiple button and should enter the Fig. 3 Project model

Fig. 4 Barcode scanning

Smart Shopping Trolley for Billing System

117

Fig. 5 Electronic bill generated

count of the product so that the system takes the count and calculates the cost of the product. Case 3: Deleting an item from the cart—when customers scan the product and then they feel that they do not need that, then they can remove the item using the delete button. Case 4: Mismatch of the weight from the cart—when the customer scans one product and tries to put different item into the cart, the mismatch of the weight happens as the load sensor senses it. Case 5: Crossing the budget limit—the customer will be setting the budget limit at the start of the shopping; if he exceeds the shopping, an alert message is given to customer. Case 6: Adding different items to the cart—customer can add different items by scanning and adding it to the cart.

7 Conclusion and Future Work Overall, the proposed system is very economical and easy to use, and it does not require any special skills for its usage. As a whole, the system is becoming smart and consuming less time which is very advantageous for the customers. This model provides solutions for many existing problems, and customer can get the services better. This model was mainly designed for the self-checkout system which consumes less time. It is fully synchronized; that is, the weight of the product should be matched with the weight provided in the database which is very important in theft prevention. This is very simple to use and can be implemented easily. Further many improvements can be made by using different techniques like AI and ML.

118

R. Kishor Kumar et al.

References 1. Wankhede SS, Radke DP, Tiwari S, Nikose A, Khadse DB, Jamthe DV () 2018 Electronic shopping trolley for shopping mall using android. IEEE 2. Viswanadha V, Pavan Kumar P, Chiranjeevi Reddy S (2018) Smart shopping cart. IEEE 3. AgarwalIsha, Sanjay Chawandke (2017) RFID based supermarket shopping system. IEEE 4. Chaure B, Jain P (2016) Development of e-shopping cart with theft control mechanism: no queue ICETT 5. Chadha R, Kakkar S, Aggarwal G (2019) Automated shopping and billing system using radiofrequency identification. IEEE 6. Chen YL, Lee JS (2016) Development of smart shopping carts with customer oriented service. Syst Sci and Eng (ICSSE), ISSN: 2325-0925 7. Chandrasekhar P, Sangeetha T (2014) Smart shopping cart with automatic billing system through RFID and ZigBee. Inf Commun Embed Syst (ICICES), ISBN: 978-1-4799-3698-4 8. Li R, Tianyi S, Capurso N, Yu J (2017) IoT applications on secure smart shopping system. IEEE Internet of Things J 4(6), ISSN: 2327-4662 9. Akshay Kumar, Abhinav Gupta, Balamurugan S, Balaji S, Marimuthu R (2017) Smart shopping cart. Microelectron Devices Circ Syst (ICMDCS), ISBN: 978-1-5386-1717-5 10. Ali Z, Sonkusare R (2014) RFID based Smart Shopping: an overview. Adv Commun Comput Technol (ICACACT), ISBN:978-1-4799-7318-7

A Survey on IoT Protocol in Real-Time Applications and Its Architectures M. L. Umashankar, S. Mallikarjunaswamy, N. Sharmila, D. Mahesh Kumar, and K. R. Nataraj

Abstract In recent days, activities are completely automated and are more flexible because of Internet of Things [IoT]. Due to IoT devices, effective integrated communication between all types of embedded devices is possible. The devices should have communication in stipulated range through the support of Internet. This significant functionality is built by the IoT protocols. In the digital world, internet-connected devices has enabled the IoT protocols to be one of the significant fields in computing, and it acts as a bridge between the physical and the cyber-world. This paper gives a brief overview of Internet of Things (IoT) and standards used to it. Important IoT protocols and their review are presented. The similarities and dissimilarities of Advanced Message Queuing Protocol (AMQP) and Message Queue Telemetry Transport (MQTT) protocols are also discussed. The main aim of this review papers is to provide a functional view of IoT architecture, standards and protocols. Keywords Advanced message queuing protocol · Message queuing telemetry transport · Constrained application protocol · Internet of things

M. L. Umashankar Department of Artificial Intelligence and Machine Learning, BMS Institute of Technology and Management, Bangalore, Karnataka 560064, India S. Mallikarjunaswamy (B) · K. R. Nataraj Department of Electronics and Communication Engineering, JSS Academy of Technical Education, Bangalore, Karnataka 560060, India e-mail: [email protected] N. Sharmila Department of Electrical and Electronics Engineering, JSS Science and Technology University, Mysore, Karnataka 570006, India D. Mahesh Kumar Department of Electronics and Communication Engineering, JSS Academy of Technical Education, Bangalore, Karnataka 560060, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_12

119

120

M. L. Umashankar et al.

1 Introduction The increase in the use of Internet on the everyday task and rapid growth in technology are revolutionizing and automatizing the world scenario to improve the quality of human lives as shown in Fig. 1. IoT enables the Internet-connected devices to connect together for different application. It is network of different combination of domain and heterogeneous devices. These devices act as an actuators to sense the inputs and respond immediately by quickly analyzed. The physical devices of the network are realized by using these sensing devices and can be used in several application such as communication, transportation, agriculture, home automation, healthcare, industrial automation and emergency services. Raspberry Pi and Arduino microcontrollers are enable the devices to measure sensor’s data and send it over the Internet. Car, refrigerator and any other electronic devices are the example of such devices. Nowadays use of transmission, collection, consolidation and displaying of sensor’s data are most common practice. In other words, data between physical devices each other are being controlled by IoT physical devices. In upcoming days, billions and trillions of devices will get involved in building a smarter world [1]. In addition to this, IoT application is extended to smart domain such as e-health, transportation, home and industry automation as described by Said and Albagory [2]. IoT is a model of large number of heterogeneous devices, so data interoperability is a challenge in data mining. To overcome this issue, most popular methods are presented in Madhu et al. [3]. In IoT, many researchers are contributing in design and development of protocols for different fields of IoT. The secure authentication is provided for different nodes or devices in a hierarchical network of IoT. Pooja et al. [4] presented protocol for secure communication among the nodes which is formulated using three way authentications. Sadiya Thazeen et al. [5] presented comparison of IoT and Cyber-Physical Systems (CPS). Several aspects of IoT and CPS are consider during comparison such as differences between CPS and IoT, overview of the CPS, security and privacy in IoT, challenges of each layer of architectures and various architectures of IoT. Li et al. [6] presented a survey on security and privacy in IoT. In this paper, different attack on IoT device, limitation of the IoT device, IoT security at different layers, authentication and access to IoT device have been addressed. Hence, this helps in to understand various aspects of security. The rest of this paper is organized as follows; Internet of things functional overview is presented in Sect. 2, the protocol standards are discussed and highlighted in Sect. 3, comparative analysis of AMQP and MQTT is presented in Sect. 4 and Conclusion is presented in Sect. 5.

2 Functional Overview of IOT Sensing, analysis, communication and computation are the four functioning modules of Internet of Things. IoT implementation using different functional modules is

A Survey on IoT Protocol in Real-Time Applications …

121

Fig. 1 IoT applications

shown in Fig. 2 [7]. These functional modules are used to establish the functional environment for different applications ranging from industries to healthcare.

2.1 Sensing Different types of sensors are used to collect different types of data such as nonprogrammable peripheral interface devices, actuators, etc. [8]. The sensing devices collect the data in raw form and store it in temporary storage without processing. In the constrained environment such as the uses of deployment and low power, these devices are used.

2.2 Communication Several protocols such as Wi-Fi, Bluetooth, LTE-Advanced, NFC, UWB, RFID and communication links are used to store the data in storage devices. To implement IoT standards such as MQTT, CoAP, AMQP, Data Distribution Service (DDS), Extensible Messaging and Presence Protocol (XMPP), etc. [9], above protocols are used since it support for the IoT platform. Device usability and their environment need to

122

M. L. Umashankar et al.

Fig. 2 IoT functional view

be keeping in the mind when developing these protocols. As the growth of Internet of Things is increased rapidly, the network will be connected with large number of heterogeneous devices. In IoT communication due to this various issues communication between the devices is challenging such as memory usage, Quality of Service (QoS), addressing and identification.

2.3 Computation Structures of data verification make them available for processing which is done by protocols during computation. The separate hardware and software level of processing is done. Cloud computing and Fog or Edge are used on software level. Embedded devices such as sensors, actuators and other devices are connected in IoT. The above devices generates large amount of data, and to extract knowledge from this, a complex computations is required. In Lenka et al. [10] presented various IoT platforms for hardware utilization such as the, Raspberry Pi, Intel Galileo, BeagleBone, Cubieboard, Arduino and WiSense. Authors also highlighted use of smart devices such as laptop and mobile. Different software and operating systems are used by

A Survey on IoT Protocol in Real-Time Applications …

123

these smart devices such as C, C++, JAVA and TinyOS, LiteOS, RiotOS, Android, Contiki etc. The Fog or Edge computing and cloud computing which provide smart computing ability to IoT [11]. A cloud is a platform which stores the data and managed the data in real time, and it can be processed further for intelligent analysis.

2.4 Analysis Analysis is extracting the vision from information for further processing in IoT. Through sensors data are collected in the form of batches during analysis. These data are in large amount, and it is called Big Data. Analysis of data helps in understanding disposal of the data, with interest toward avoiding failures minimizing maintenance and improving operations. • Analysis of Intelligent: Cognitive technologies driven the analysis of data. The data analysis process varies in the form of voice, information, vision and usability with the ability of technological advancement. For real-time data streams, analysis is very much required. Within an IoT managing, the real-time data and ability to analyze unstructured data in data model are the open challenging issues in data analytics. • Intelligent Actions: Machine to Human (M2H) and Machine to Machine (M2M) interfaces can be expressed for intelligent actions. In improving machine functionality and minimizing the machine prices, an intelligent agent will play a vital role in IoT [12]. In unpredictable situations, machine actions will become more challenging to privacy, security, machine interoperability and mean-reverting human behaviors which slow adoption of new technologies in IoT as described by Shin et al. [13], Taher et al. [14] IoT functions, elements and levels are illustrated in Table1.

3 Standards Internet of Things has covered scope of different areas and fields. To provide the features of IoT and to support in various distinct fields, protocols and standards are used by many groups. Many organizations such as Internet Engineering Task Force (IETF), Wide Web Consortium (W3C), Institute of Electrical and Electronics Engineers (IEEE) are proposed several standards to support of Internet of Things. Protocols standardization is process is fully done by the Internet Engineering Task Force Ghahramani [12]. Different level of standards is highlighted in Fig. 3. Different domains such as infrastructures, influential and application above standards are utilized [15].

124

M. L. Umashankar et al.

Table 1 IoT levels, functions and elements Level

Function

Elements

Sensing

Raw data sense and measure

Actuators, smart sensors, RFID tags and embedded sensors

Communication At application and infrastructure level, Application level a gateway protocols are used for HTTP-REST, CoAP, MQTT, AMQP, linking and providing DDS and XMPP Infrastructure level IEEE 802.15.4, LTE-A, Z-wave, RPL, EPC-global and IPV4/IPV6 6LowPAN Computational

Processing and storage ability provided by software and hardware level

Software: LiteOS, TinyOS, Android, RiotOS, Cloud Computing and Contiki Hardware: Microprocessors, Microcontrollers, Arduino and Raspberry Pi

Analysis

For specific purpose normalize and concluding the data

For services related identity purpose, ubiquitous services, collaborative aware services and information aggregation

Fig. 3 IoT standards

3.1 Application Protocols The data presentation and formatting is the responsibility of the application layer. HTTP protocol has been considered as reference protocol form long time in communication as shown in Fig. 4. HTTP especially supported for Internet applications

A Survey on IoT Protocol in Real-Time Applications …

125

Fig. 4 HTTP protocol formation

but HTTP is not suitable for some of the constrained environment. Hence to find solution, several other protocols are developed such as MQTT, CoAP, MQTT-SN, XMPP, DDS, AMQP etc. The protocols MQTT and AMQP are most popular and widely used protocols [13]. Some of the popular protocols discussed in this section are Shin et al., Taher et al.

3.2 Constrained Application Protocol (CoAP) IETF is proposed the CoAP for managing and retrieving information for devices and sensors. For IoT, application layer protocol and literature with functionalities are presented of CoAP), Municio et al. [16]. Fulfilling the needs of resource-constrained devices which is the primary aim of CoAP protocol. Two-layered approach is used by CoAP; request–response models and processing the features of messaging. In general, the messaging model deals with interchanging the messages asynchronously through the user datagram protocol Stute et al. [17]. For IoT applications, CoAP is bound to UDP to make it more suitable Moosavi [18]. Hence, it is not limited to TCP and DTLS implementation if required. It uses protocols such as Representational State Transfer (REST) protocol to share the communication model with HTTP Mahendra et al. [19]. Confirmable, Non-confirmable, Acknowledge and Reset are four types of messages which are represented by two bits such as 00, 01, 10 and11, respectively. It is well suited for constrained oriented environments, and it uses four methods supported by HTTP Get, Post, Put and Delete Mallikarjunaswamy et al. [20].

126

M. L. Umashankar et al.

3.3 Message Queuing Telemetry Transport (MQTT) MQTT is designed based on binary lightweight protocol as bandwidth efficient and consumes very low power. It is called as reliable protocol since it uses acknowledgment scheme in all formats as shown in Fig. 5. It uses the concept of asynchronous message queuing, hence it is referred as open protocol. For machine-tomachine communication, it uses subscribe architecture or publish in an environment of low bandwidth and works on transmission control protocol. Connection semantics, routing and endpoint are the three elements specification provided by MQTT Shivaji et al. [21]. Publisher, subscriber and broker are the three component consists in the protocol. To enable the messages be pushed to the clients, the publish and subscribe are event-driven. MQTT broker is with the central control. Between the sender and right receivers, the broker is responsible for dispatching all messages is given in Eq. 1. MQTT packet length = control header + length + protocol level + connect flags + payload

(1)

QoS level of delivery assurance is defined by protocol level. Maximum 4 bytes is the packet length and control header is fixed 1 byte. MQTT protocols are developed in two versions, for TCP/IP protocol, MQTT is designed and for UDP and ZigBee protocols, MQTT-SN is designed.

3.4 Advanced Message Queuing Protocol (AMQP) For the message-oriented environment, AMQP is an open standard application layer protocol is addressed in Mallikarjunaswamy [22]. AMQP is an open standard application layer protocol for the message-oriented environment which is presented in Chaitra et al. [23]. Its main objective is to enhance interoperability by making to operate in different systems and applications to work together. It is also open source and asynchronous protocol. AMQP has a same architectural scheme as MQTT like publishing, broker and subscriber, but it includes message exchange mechanism like separate queues for the respective subscriber as shown in Fig. 6. According to the predefined criteria as shown in Fig. 6, the exchange model receives the messages from the publisher and route them to queues. To examine the message and route, it uses a routine and instances in proper queue by using key, which is called as a virtual address.

A Survey on IoT Protocol in Real-Time Applications …

127

Fig. 5 MQTT communication protocol

Fig. 6 Mechanism of AMQP protocol

4 AMQP and MQTT Comparative Analysis In the literature, most popular protocols used are AMQP and MQTT. Different parameters are consider to make comparison between MQTT and AMQT such as frame structure as in Table 2, use of the protocol, response and transaction. The similarities and dissimilarities are highlighted in Tables 3 and 4 of MQTT, CoAP and AMQT protocols [24–27].

128

M. L. Umashankar et al.

Table 2 CoAP and MQTT protocol comparison CoAP

MQTT

Works over UDP

Works over UDP

4-bytes size of packet header

2-bytes size of packet header

Avoid duplication of messages

Using message-ID and DUP flag duplication of messages are avoided

Messages on the original device and convention named topics are used to manage their sources

On the device itself, resources are managed

Table 3 AMQP and MQTT protocol similarities AMQP

MQTT

Similarities

(1) Queuing scheme of message (2) Nature of asynchronous (3) Cloud computing supports (4) Minimal set of configuration (5) TCP/IP developed

Table 4 AMQP and MQTT protocol dissimilarities AMQP

MQTT

Protocol use

For any bandwidth network and for any device, AMQP protocol can be used

In low bandwidth, networks intended to design for small and dump devices MQTT protocol is preferred

Optimization of frame

Fragmentation use buffered oriented approach is used by AMQP protocol

• Writing of frames is easy for low memory devices using stream oriented approach of MQTT protocol • No fragmentation

Transaction

Across message queues, it supports transactions

Not support for transaction

Response

Different acknowledgment is supported with use cases

Basic acknowledgments are supported

The comparison results MQTT and AMQP helps in selection of protocols in different IoT application in a real-life scenario.

A Survey on IoT Protocol in Real-Time Applications …

129

5 Conclusion Internet of Things has a huge contribution for the development of technology and application design along with the artificial intelligence and machine learning. Data is a major component in all the applications. Secure data transfer is of utmost importance via networking interfaces which has to be authenticated. Protocols and computational algorithms are necessary for processing the senses data through various IoT devices to predict results. In this paper, we present the overview and characteristics of IoT protocols. Apart from this, an extensive analysis has been carried out on the IoT application layer protocol such as CoAP, AMQP and MQTT used in different IoT applications. Furthermore, we have highlighted similarities, dissimilarities and comparison of application layer protocol. However, this review helps the several researchers to identify and utilize in appropriate protocol for different real-time IoT applications. Acknowledgements The authors would like to thank BMS Institute of Technology and Management, Bengaluru, JSS Academy of Technical Education, Bengaluru, Don Bosco Institution of Technology, JSS Science and Technology, Mysore, SJB Institute of Technology, Bengaluru, Visvesvaraya Technological University (VTU), Belagavi , Vision Group on Science and Technology (VGST) Karnataka Fund for Infrastructure strengthening in Science and Technology Level—2 and Center For Interdisciplinary Research (CIR) JSSATEB for all the support and encouragement provided by them to take up this research work and publish this paper.

References 1. Umashankar ML et al (2020) Design of high speed reconfigurable distributed life time efficient routing algorithm in wireless sensor network. J Comput Theor Nanosci 17:3860–3866 2. Said O, Albagory Y, Nofal M, Al Raddady F (2017) IoT-RTP and IoT-RTCP: adaptive protocols for multimedia transmission over internet of things environments. IEEE Access 5:16757–16773 3. Madhu TA et al (2020) Design of fuzzy logic controlled hybrid model for the control of voltage and frequency in microgrid. Indian J Sci Technol 13(35):3612–3629 4. Pooja S et al (2021) Adaptive sparsity through hybrid regularization for effective image deblurring. Indian J Sci Technol 14(24):2051–2068 5. Thazeen S et al (2021) Conventional and subspace algorithms for mobile source detection and radiation formation. Traitement Signal 38:135–145 6. Li P, Su J, Wang X (2020) iTLS: lightweight transport-layer security protocol for IoT with minimal latency and perfect forward secrecy. IEEE Internet Things J 7(8):6828–6841 7. Swamy SN, Kota SR (2020) An empirical study on system level aspects of internet of things (IoT). IEEE Access 8:188082–188134 8. Wazid M, Das AK, Odelu V, Kumar N, Conti M, Jo M (2018) Design of secure user authenticated key management protocol for generic IoT networks. IEEE Internet Things J 5(1):269–282 9. Liu A, Alqazzaz A, Ming H, Dharmalingam B (2021) Iotverif: automatic verification of SSL/TLS certificate for IoT applications. IEEE Access 9:27038–27050 10. Lenka RK, Rath AK, Sharma S (2019) Building reliable routing infrastructure for Green IoT network. IEEE Access 7:129892–129909

130

M. L. Umashankar et al.

11. Ma Y, Yan L, Huang X, Ma M, Li D (2020) DTLShps: SDN-based DTLS handshake protocol simplification for IoT. IEEE Internet Things J 7(4):3349–3362 12. Ghahramani M, Javidan R, Shojafar M, Taheri R, Alazab M, Tafazolli R (2021) RSS: an energy-efficient approach for securing IoT service protocols against the DoS attack. IEEE Internet Things J 8(5):3619–3635 13. Shin D, Yun K, Kim J, Astillo PV, Kim J, You I (2019) A security protocol for route optimization in DMM-based smart home IoT networks. IEEE Access 7:142531–142550 14. Taher BH, Jiang S, Yassin AA, Lu H (2019) Low-overhead remote user authentication protocol for IoT based on a fuzzy extractor and feature extraction. IEEE Access 7:148950–148966 15. Al-Janabi TA, Al-Raweshidy HS (2018) A centralized routing protocol with a scheduled mobile sink-based AI for large scale I-IoT. IEEE Sensors J 18(24):10248–10261 16. Municio E, Latré S, Marquez-Barja JM (2021) Extending network programmability to the things overlay using distributed industrial IoT protocols. IEEE Trans Ind Inf 17(1):251–259 17. Stute M, Agarwal P, Kumar A, Asadi A, Hollick M (2020) LIDOR: a lightweight DoS-resilient communication protocol for safety-critical IoT systems. IEEE Internet Things J 7(8):6802– 6816 18. Moosavi SR (2015) SEA: a secure and efficient authentication and authorization architecture for IoT-based healthcare using smart gateways. Procedia Comput Sci 52:452–459 19. Mahendra HN et al (2019) Evolution of real-time onboard processing and classification of remotely sensed data. Int J Eng Adv Technol 9:7153–7158 20. Mallikarjunaswamy S et al (2020) Implementation of an effective hybrid model for islanded microgrid energy management. Indian J Sci Technol 13:2733–2746 21. Shivaji R et al (2020) Design and implementation of reconfigurable DCT based adaptive PST techniques in OFDM communication system using interleaver encoder. Indian J Sci Technol 13:2108–2120 22. Mallikarjunaswamy S et al (2014) Design of high-speed reconfigurable coprocessor for nextgeneration communication platform. Emerg Res Electron Comput Sci Technol 57–67 23. Chaitra S et al (2021) A comprehensive review of parallel concatenation of LDPC code techniques. Indian J Sci Technol 14:527–539 24. Satish P et al (2020) A comprehensive review of blind deconvolution techniques for image deblurring. Traitement Signal 37:135–145 25. Manjunath TN, Mallikarjunaswamy S, Komala M, Sharmila N, Manu KS (2021) An efficient hybrid reconfigurable wind gas turbine power management system using MPPT algorithm. Int J Power Electron Drive Syst (IJPEDS) 12(4):2501–2501 26. Mallikarjunaswamy S, Sharmila N (2021) A novel architecture for cluster based false data injection attack detection and location identification in smart grid. Adv Thermofluids Renew Energy 599–611 27. Shivaji R, Nataraj KR (2021) Implementation of an effective hybrid partial transmit sequence model for peak to average power ratio in MIMO OFDM system. In: 2nd international conference on data science, machine learning and applications, pp 1343–1353

Safe Characteristic Signature Systems with Different Jurisdiction Using Blockchain in E-Health Records Shivakumar Dalali, B. K. Pramod, Ranjith Kumar, and M. J. Thejas Jain

Abstract Nowadays users can store the information remotely in the cloud and whenever they want, they can access information at any place with distributed storage system. For the information stored in remote places, information respectability reviewing is used to ensure truthfulness of the information put away in the virtual storage. In distributed virtual storage, for example, the Bank Transaction Records system, the virtual storage may consist of important information. The important information is not to be displayed to the unknown persons while the virtual records are accessed. Accessed documents are encoded and that can understand the sensitive data carefully and neatly. The most effective method to acknowledge information offering to sensitive data are carefully handled in remote places, data reviewing not been systematically examined till today. So, to address this issue, we are bringing the far-off information examine to ensure the idea that acknowledges information directed to sensitive data handled carefully here. In this plan, a tool is brought to clean the information squares comparing to most wanted data of the record and changes these information squares’ marks to essential one for the cleaned document. These marks are carried out to check the completeness of the cleaned document in the span of reviewing. Afterward our scheme uses the record stored in the virtual storage that is accessed and used by needy ones based on the precondition that the important content is secured. The distant information rectitude inspect is ready to be competently executed. Suggested plan is based on personality-based-cryptography. Keywords Signature system · Health · Blockchain

1 Introduction With virtual storage arrangement, anyone can distantly keep their information to the virtual storage and get to know that the data sharing with others in the cloud maintains the integrity of the data with the help of remote data integrity as shown in Fig. 1. In S. Dalali (B) · B. K. Pramod · R. Kumar · M. J. Thejas Jain Department of Computer Science Engineering, Don Bosco Institute of Technology, Bengaluru, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_13

131

132

S. Dalali et al.

Fig. 1 EHRs systems in the present and future

virtual storage like the patient information Records system or the bank transaction system, the virtual storage file might have important content that important content should not be used by anyone when the virtual storage file is accessed [1] if we do not want to share that information encrypt the accessed file that can make the important content complicated, but the accessed file is cannot be use by needy ones [2]. We can resolve this query with the use of remote data integrity auditing. Here, a tool is brought up to clean the bunch of data related to the important content of the file and send these bunch of data signature into right person for cleaned file. These are adjusted to check the completeness of cleaned file in time of composed auditing. By the outcome this plan allows storing of content in the virtual storage and that can be shared and accessed by anyone, but the important content complicated by encrypting it. The patient should have right to access his EHRs for managing and sharing them independently. The protection inspection and ability of checking shows that our suggested idea gives protection and works in a proper manner.

2 Existing System In the existing system there is an unmanageable complicated hindrance for customers to place the data local disk. Every organizations and each person like to place their content in the virtual storage. Anyhow there is no guarantee for the data stored in the virtual storage. The content in the virtual storage may be lost because of unavoidable predetermined software errors or hardware problems in clouds. In this type of systems their may be a chance of patient may lose authority over the content and patient cannot able to add or edit the content present in the virtual storage, because provided as fully controlled over the data present in the cloud. The accessibility of the data is very limited for the patients or customer.

Safe Characteristic Signature Systems with Different …

133

3 Proposed System To solve the query mentioned in existing system, we have to encipher the accessed content after passing it to the virtual storage with the use of various encryption algorithms and store the signature used to check the completeness of this encrypted content [3]. Then we have to pass the encrypted content and its related signature to the virtual storage. This approach assures that data integrity is successfully achieved. Since only owner of the content can decrypt the content with the help of decryption algorithm [4]. It makes difficult to access the data for other users who does not have the password or signature [5]. For example, encrypting file in electronic health record system of patients can maintain the privacy of patient personal data and hospitals. So there is only one way to access the file in the cloud by using decrypting keys [6].

4 System Design Figure 2 shows framework architecture for configuring and recognizing the general information retrieval schema for the Web Application. Architecture design is bind to the aims to establish for a Web Application, the resources brought in front of a user who visit the Web Application, and the directional procedure that has been created. Content architecture preoccupied about the way in which resource content and formed for demonstration and direction. WebApp systems brought forward in the way in which the application is formed to control user communication to handle internal running operations, effect direction, and present idea. Web Application architecture is confined within the arena of the development atmosphere in which the application has to be set up.

5 Flow Chart Diagram It is principal to perfect the entire given task and fulfill time requirement. First, we have to give the block id as input. That block id is the address of the required block of data. The block content is encrypted using some cryptographic algorithms. We have to unzip that block of content using some keys that stored in the database. Then get root hash of the required block of content and get the details of the next block. Using that details extract the next block content and find the previous block hash (PBH) using some algorithm and keys. If that root hash of required block and previous block hash of next block is same then integrity is preserved. Now we can display the message to the user like block is corrupted if the root hash and previous block hash is not same else display message like block is not corrupted as shown in Fig. 3.

134

S. Dalali et al.

Fig. 2 Framework architecture for configuring and recognizing the general information retrieval schema for the Web Application

6 Class Diagram This section gives creation of class diagram for admin (Fig. 4). This section gives creation of class diagram for file download at user (Fig. 5).

7 Modules Description Modules used in our projects are explained in this section.

7.1 Admin Admin have privileges to engender the utilizer, during utilizer engenderment we will send utilizer id and password to their email id and additionally he will maintain the cloud servers’ configurations.

Safe Characteristic Signature Systems with Different …

Fig. 3 Flow chart of cryptographic algorithms

7.2 Admin Module • • • • • • • • •

Login. Profile (View and Edit Only). Admin can view and edit his profile data in database. User List (Integrate, Edit, Expunge). Admin will be the responsible to integrate the incipient Member. Auditors List (Integrate, Edit, Expunge). Admin can integrate, edit and expunge the Auditor. Transaction Details. Admin can check the transaction details of each individual User’s.

135

136

Fig. 4 Class diagram for admin

Fig. 5 Class diagram for user-file download

S. Dalali et al.

Safe Characteristic Signature Systems with Different …

137

• Change Password.

7.3 User Module • • • • • • • • • • • • • • • • • • • •

Login. User Register. Fill the utilizer details to get the utilizer name and password. Profile (View and Edit Only). User can view and edit his profile data in database. File Upload. File Cull from the User’s Local system. Click uploads and get the file. The file has to read the data and to abstract the special symbol and nugatory words like is, was utilizing preprocessing technique etc. And extract the keywords and check with sensitive dataset. If matching then that sensitive keywords engender the hashtag. Transaction Prosperously message will be send to the Utilizer. Retrieve the particular content from the Cloud Storage. After downloading if the file is encrypted then it will get decrypt utilizing AES Algorithm. Transaction record will be recorded in the table. Verification. Select the file from listed files and verify the files. Transaction Details. View the transaction of logged utilizer. Change Password.

7.4 Auditor • • • • • • • • •

Login. Profile (View and Edit Only). User can view and edit his profile data in database. Verification (View). View all users’ request for the verification. Audit the File. Select the content from the list. Press the Verify (Audit) Button. Transfer the request of SHA1 Key to the Controlling server for uploading the content. • For the uploaded content generate SHA2 key. • Transfer the SHA2 key to Auditor.

138

• • • •

S. Dalali et al.

Get the Encrypted SHA1 key and decrypt the SHA1 key from User’s DB. Compare both SHA1 with SHA2 key. Display the outcome. Change password.

8 Conclusion Targeting safeguarding persistent security in an EHRs framework on square chain, different ascendant elements are brought into ABS and set forward a MA-ABS plot, which matches the imperative parameters of the blockchain, just like guaranteeing the obscurity and changelessness of the data. PRF seeds are required among ascendant elements and the patient private keys should be developed, adulterated ascendant elements cannot thrive in plot assaults. Definitively, the security of the convention is demonstrated under the CBDH proposition as far as unforgetability and flawless protection. The correlation investigation exhibits the presentation and the expense of these convention increments directly with the quantity of ascendant substances and patient properties also.

References 1. Ren K, Wang C, Wang Q (2012) Security challenges for the public cloud. IEEE Internet Comput 16(1):69–73 2. Ateniese G, Burns R, Curtmola R, Herring J, Kissner L, Peterson Z, Song D (2007) Provable data possession at untrusted stores. In: Proceedings of the 14th ACM conference on computer and communications security, series CCS ’07, pp 598–609 3. Juels, Kaliski BS (2007) Pors: Proofs of retrievability for large files. In: Proceedings of the 14th ACM conference on computer and communications security, series CCS ’07, pp 584–597 4. Shacham H, Waters B (2013) Compact proofs of retrievability. J. Cryptology 26(3):442–483 5. Wang C, Chow SSM, Wang Q, Ren K, Lou W (2013) Privacy-preserving public auditing for secure cloud storage. IEEE Trans Comput 62(2):362–375 6. Worku SG, Xu C, Zhao J, He X (2014) Secure and efficient privacy preserving public auditing scheme for cloud storage. Comput Electr Eng 40(5):1703–1713

Web-Based Trash Segregation Using Deep Learning Algorithm S. Sheeba, Akshay Mohan, Ashish Kumar Jha, Bikash Agarwal, and Priya Singh

Abstract This study proposes to classify waste as recyclable and helps them to improve the sorting process of compact waste collected from the public. The proposed system is an advance hybrid multi-layered deep learning system which uses an algorithm of CNN for capturing the image of waste and sorting the compact waste which can be further recyclable to a useful product. It reduces the man efforts and helps the mankind to automate the system of garbage separation. This projected system is using the CNN model which is trained and assessed by niche technology to collect and sort the waste. It is proven that the speed of detection the waste is above 90%, which is quit precision relying on image only inputs. Keywords Public solid waste (PSW) · National Green Tribunal (NGT) · Convolutional neural network (CNN)

1 Introduction As per the global report of compact waste management, annual total compact collection is predicted to reach more than 200 crore tons by 2025. This entire compact waste will cost us $37,500 crore in compact waste management. Inappropriate compact waste management can lead the whole world toward multiple crises such as economy, public health care, and natural imbalance of the globe. As per the survey, the recycling of public solid waste (PSW) has been known as one of the best nature friendly plan to deal with the urban waste by the NGT in India. Actual waste recycling can help the world to find the answer of several health and wealth problems. It can find the solution of issues related to uncooked resource, protective energy, reducing emission of greenhouse gases, aquatic contamination, reducing soil pollution, reducing the land fillings, etc. In developing country like

S. Sheeba (B) · A. Mohan · A. K. Jha · B. Agarwal · P. Singh Department of Computer Science and Engineering, Don Bosco Institute of Technology, Bengaluru, Karnataka, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_14

139

140

S. Sheeba et al.

India, Brazil, etc., PSW recycling depends on household waste separation by scavengers and collectors who makes profit of this waste. However, the developing countries have vast budding to progress the waste recycling. The public waste recycling is bit slow in the country like India and Brazil but still these countries are accreting themselves for improving the percentage of recycling. Our system can help these countries boost the growth percentage of recycling waste. In India, PSW recycling fully depends on scroungers and collectors who collect the waste for their own profit. Even there is huge development of garbage management in the countries which are in developing stage. As per the CPCB report, in the year of 2014–15, total waste collected was 91% from that 27% was treated and rest all get dumped in the landfills. A recent studies states that India requires a city like Bangalore to dump its whole waste. The approximate size of Bangalore is 709 km2 . For the rapid improvement of the waste recycling, we proposed our multitasking waste management system to improve the performance and technique of sorting and classifying the waste collected from the public. Our system reduces the man efforts along with that it sorts the waste automatically which can be recyclable. The system proposed has the accuracy of approx. 90%, which signifies the performance of our proposed system. Obstacles which degrade the waste recycling process are as follows (1) Lack in implementation of government plans and issues in budget allocation: inadequate government directive to enforce the plans of government and budget for PSW and public health management. (2) Basic garbage management education and awareness among the public regarding waste management: homes are unconscious of the importance of self-waste reprocessing and separating the dry and wet waste. (3) Lack of technological advancement in the waste management: lack of effective recycling technology. (4) Lack of an organization who manages the expense of recycling: the high cost of manual waste classification.

2 Proposed System The objective of the work is to replace the existing system of waste management and propose an easier way to collect the waste from households, hospitals, etc., to the recycling centers and aware people regarding waste segregation (dry and wet). The utmost significant aim for appropriate waste management is to guard the atmosphere and for the well-being and protection of the populace. Decrease the capacity of the leftover stream through the implementation of waste reduction techniques and recycling programs (Fig. 1). It will help the traditional waste collection system to get digitalized, by which locals will get encouraged and motivated to segregate and store the waste to get rewarded with reward points to their respective wallets. Indirectly, our project

Web-Based Trash Segregation Using Deep Learning Algorithm

141

Fig. 1 Solid waste recycling

enhances the purity of soil, water, and air which was earlier contaminated by land fillings. In this modern era, waste management has ongoing challenges due to weak establishments, and quick development. All of these tasks, sideways with the absence of sympathetic of diverse issues that underwrite to the grading of waste managing, affect the treatment of waste which can be resolved by our undergoing project. Our system includes from the very activities of the door to door collection of segregated waste that would have helped the user to make their surrounding clean. This system helps the recycling chain and cause employment upliftment. Recycling increases the consumption of natural resources, reduces environmental pollution, develops the economy, and creates additional jobs.

3 Implementation A. Web Framework B. (1) Django It is a free and open source frame work used for web application. This framework is getting implemented in Python. Framework is defined as the collection of different

142

S. Sheeba et al.

modules which makes the development easier. These modules get grouped together to create an application or website availed to an existing source. C. Front End Implementation (1) HTML and CSS A the HyperText Markup Language (HTML) and cascading style sheets (CSS) are two of the essential machineries for edifice web pages. HTML offers the edifice of the page, CSS the (visual and aural) design, for a variability of devices along with visuals and scripting. CSS is the language for relating the performance of web pages, counting ensigns, design, and letterings. It lets one to familiarize the exhibition to dissimilar types of strategies, such as big screens, minor screens, or laser printer. CSS is self-governing of HTML and can be used with any XML-created profitable language. (2) Node.js It helps the user to run the JavaScript code beyond any specific browser. It is a JavaScript runtime environment which includes an open source and cross platform. It represents the JavaScript everywhere, helps to unify the development requirements of a web application in a single programming language rather than different. D. Back End Implementation (1) Python A Python is a HLL, construed, collaborative, and object concerned with scripting language. It is intended to be extremely legible. It uses English language keywords often, whereas further languages use many grammar rules, and it has fewer syntactical buildings than other languages. Server-side web applications can be shaped by it. (2) Convolutional Neural Network (CNN) Artificial neural network performs very well while the implementation of machine learning. For image, audio, words classification task, artificial neural network gets in use (Fig. 2). Input Layers: In neuronal nets, input layers perform some calculations through its neurons the quantity of neurons in an input layer be contingent on the form of our exercise facts. Hidden Layers: The hidden layers make the neuronal nets as larger to ML algorithms. There should be zero or additional than zero hidden layers in the neuronal networks. Output Layers: The output layers are accountable for creating the final output results. The output layer receives the efforts which are approved on or after the layers beforehand it, and performs the controls finished its neurons, and then, the output is computed.

Web-Based Trash Segregation Using Deep Learning Algorithm

143

Fig. 2 Layers of regular neural network

(3) Support Vector Machine Regression (SVM) Support vector machines (SVMs) are commanding hitherto supple overseen ML algorithms which are cast-off both for cataloging and deterioration. But usually, they are used in cataloging glitches. In 1960s, SVMs were primary presented but far along they got advanced in 1990. These are zero for all points that are exclusive the group. In SVM reversion, the input x is first charted onto a m-dimensional feature space using some fixed (not straight) charting, and then, a lined structure is built in this feature planetary. Some mathematical notations are as follows: f (x, ω) =

m ∑

ω j g j (x) + bg j (x),

j = 1, . . . , m

j−1

where x signifies a set of nonlinear transformations, and b is the “bias” term. Frequently the information is expected to be zero nasty (this can be attained by preprocessing), so the bias term is released. (4) MongoDB MongoDB chains arena, variety question, and systematic appearance explorations. Queries can reappearance precise arenas of brochures and also contain handler-welldefined JS roles. Queries can also be arranged to reoccurrence a chance model of grades of an assumed size, MongoDB is across-stage file-concerned with database program confidential as a NoSQL database program, MongoDB uses JSON-like brochures with scheme. MongoDB is advanced by MongoDB Inc. and licensed under the server-side public licenses (SSPL). MongoDB is planned to chance the strains of modern apps with a technology substance that allows you through (Fig. 3): • The file data structure—donating you the finest method to toil with data. • A dispersed schemes enterprise—agreeing you to logically put data where you want it.

144

S. Sheeba et al.

Fig. 3 Intelligent operational data platform

• A united involvement that gives you the liberty to run anywhere. • Letting you to future-proof your work and eradicate seller bolt-in.

4 Analysis and Benefits of Proposed System Waste transmission sites are amenities where public solid waste is unpacked from gathering automobiles and fleetingly held while it is refilled on top of bigger extended-distance carriage automobiles for carriage to landfills or other handling or discarding services. In the transportation of surplus waste, automobiles or ampoules used for the gathering such as the bounces, roll off/on, adjacent loaders, and tipper automobiles in confident zones. Waste should not be observable to the community, nor visible to open atmosphere avoiding their sprinkling. The principles also designated that stowing amenities set up by public establishments shall be daily joined to for defrayal of waste. The boxes or ampules wherever positioned shall be vacant already they jump abundant (Fig. 4). Municipal solid waste usually contains many types of waste. By reducing the polluted water, by reducing the organic oxygen demand (OOD) value and segregation of compact waste is indispensable to make the collected municipal waste more environment approachable and also to reduce the appreciated energy used in its recycling. Solid waste is easy to recycle compare to the liquid waste. The dormant energy contemporary in the organic waste can be used for profitable operation by using the efficient waste dispensation and recycling technologies. The recyclable wastes also offer a few additional benefits as follows: • The fall of 60% to over 90% in the total quantity of waste gets reduce, depending upon the waste composition and the adopted technology been used.

Web-Based Trash Segregation Using Deep Learning Algorithm

145

Fig. 4 Waste estimation analysis

• It also reduces the demands of land in the cities, which is already a big challenge in cities. • It also reduces the price of carriage of waste to far-missing landfill sites in the cities. It also indirectly contributes in reduction of air pollution. • With the above implementation, the overall environmental pollution will get reduced.

5 System Design and Flowcharts This unplanned modernization redirects us toward a weak waste management, which results in several health issues. All of the challenges can be resolved by the better understanding of waste management that understanding of waste treatment undergoes in our project. Our system includes from the very activities of the door to door collection of segregated waste that would have helped the user to make their surrounding clean. This system helps the recycling chain and cause employment upliftment. Recycling decreases the extra burden from the nature; it will also reduce the pollution from environment and directly or indirectly creates more employment. Architectural implementation of the system design (Fig. 5): i. ii. iii. iv. v. vi.

Request generation by user. Acknowledgement by owner. Request generation by owner to the co-worker. Request acknowledgement by co-worker. Action performed by co-worker as per owner’s instruction. Delivery of collected waste from user to factory through co-worker and acknowledgement to the owner regarding waste delivery. Simultaneously, reward will be given to the user as per the quality of waste. vii. Factory will give statistic acknowledgement to the owner.

146

S. Sheeba et al.

Fig. 5 System architecture

6 Conclusion It is a system of involuntary organization grounded on deep learning and CNN and is planned to categorize discarding and recycling of compact waste in the built-up municipal area with the end solution of waste. This system improves the humanoid sensual and the intellect procedure scheme by arranging a medium- determination photographic camera together with some basic visual sensors. The study and surveys made in Latvia depicts the full-fledged implementation of the proposed system with deep learning and CNN in refining solid waste cataloging’s competence and efficiency in the real time. It also motivates the people to learn the importance of waste separation. In considering the current situation is demanding of an environmentally friendly waste processing and management system, and the proposed project of waste management can improve the health and wealth of the world.

7 Future Enhancement It motivates the people to learn the importance of waste separation. As the volume of waste is rapidly increasing globally and we need a full proof system to manage and resolve the waste processing system, the proposed project of waste management is beneficial in either way economically and morally. Proposed system can

Web-Based Trash Segregation Using Deep Learning Algorithm

147

be implemented as E-commerce system, on which the items will be made from the collected waste material. Our proposed system will also motivate the people to buy the product from our E-commerce system and earn the reward benefits. It will reduce the dumping of garbage, which pollutes the environment.

References 1. 2. 3. 4. 5.

6. 7.

8.

9.

https://www.hindawi.com/journls/cin/2018/5060857/ http://kerwwnelsvm.tripod.com/ https://www.geeksforgeekes.org/image-classifier-using-cnn/ Hoornwaeg D, Bhaada-Tata P (2012) What a waste: a global review of solid waste management. World Bank, Washington, DC, USA Sanderson RE (1993) Environmental Protection Agency Office of Federal Activities’ guidance on incorporating EPA’s pollution prevention strategy into the environmental review process. EPA, Washington, DC, USA Williams PT (2005) Waste treatment and disposal. Wiley, West Sussex, UK Chhristensen TH, Geentil E, Booldrin A, Laarsen AW, Weaidema BP, Haaiuschild M (2009) C balances, carbon dioxides emissiones and globals warmings potential in L.C.A-modellings of wastes managements system. Waste Manage Res 273(08):7017–7115 USA EP (2018) Fact and figure about material, wastes and recyclings. EP, Washington, DC, US. https://www.eppa.gov/fact-and-/figure-/about-materials-waste-and-/recyclings/advanc ings-sustainables-materials-managements-01 Kofoworolas OF Recoveries and recyclings practice in municipal compact waste management in Lagos, Nigeriaa. Waste

Home Automation Using Face Recognition for Wireless Security B. S. Umashankar, Mandalia Vishal Shailesh, Md Shaghil Z. Ansari, and Rahul Markandey

Abstract The aim is to develop a detailed and precise face recognition-based home automation system which is based on IoT using modules and microcontroller for wireless security. This project mainly highlights and focuses on controlling certain devices such as home appliance by integrating it to Internet and making it an IoTbased system, and building an accurate and smart wireless home security system using face recognition system and Wi-Fi as communication protocol and certain network protocols. Looking at the current scenario, we find an existing system which is assimilated with home automation system. The paper is being focused more on reducing the limitation with respect to being more widespread in terms of usage. As far as the study area of this paper is considered, NodeMCU microcontroller unit along with relays module is used to control electrical appliances. Face recognition module will provide smart security wherein a captured image is sent through an E-mail/MMS to the owner using IoT-based cloud service such as Internet when a face is detected. User can be authorized to control security controllers using certain application after authenticating. Hence, resulting in being useful to people with physically challenging disabilities. Keywords Home security · Home automation · Face recognition · NodeMCU · Microcontroller · IoT · Smart devices · Network protocols

1 Introduction Home automation refers to microcontroller-based technology for controlling home appliances. Automation is a necessity now since it provides security, easy control and efficiency. In this, the IoT sensors transmit the status of appliances and provide updates to the user with the help of a Wi-Fi module. The user being away from home can access and modify the status of electric appliances, i.e. switching it ON/OFF. B. S. Umashankar (B) · M. V. Shailesh · M. S. Z. Ansari · R. Markandey Department of Computer Science and Engineering, Don Bosco Institute of Technology, Bengaluru 560074, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_15

149

150

B. S. Umashankar et al.

This paper is developed to describe the automated approach for controlling home appliances and security using the face recognition module. We always dream about having all the work done automatically for us just at the tip of the finger. The home automation system is one of the systems that fit well in this scenario. The dual aspects of this project are wireless security and wireless home automation. The currently working prototype of the system is based on a notified system. Where the authorized person is sent alerts when a particular event occurs. The notification is sent based on a network protocol and the main mode of application being SMS using the Internet. The system is trained in a specific course of action that if any sort of unusual activity is recorded or observed, the authorized person then and then is notified using the SMS service and further circumstances can be avoided. Various other means of events can be handled using the ideology of home automation which are counted essential. For example, electronic appliances can be turned on using voice command. Of the multiple advantages, a useful one is that even when Wi-Fi is not available, we can go over by using our mobile cellular network services. This feature is lacking in existing prototype methods. Overcoming the drawbacks of the existing systems, we have implemented this system which is designed to provide security at comfort. Thus, the aim of this work is to reduce power consumption and also better time management. Further, there is need to provide the safety and wireless security of the various appliances at home. As we know technology is upgrading each and every minute so, it is very much necessary for us to implement the latest technology. Most of the work in the field of IoT-based automation and security is in progress and there is tremendous scope for improvements in final field implementation. Further, this paper is organized in sections. Section 2 highlights the basics of current existing systems and their comparison. Sections 3 and 4 explain the design and implementation of our work, respectively. Finally, Sect. 5 discusses results, and 6 conclusion and further enhancements.

2 Related Work In this section of the paper, the existing work of a wireless home automation system using IoT is described.

2.1 Bluetooth-Based IoT Home Automation System It is a system which can be integrated with Bluetooth for the definite purpose of controlling and monitoring home automation. The system can make use of smart phones which are portable and reliable. The Bluetooth system basically works on a server-client procedure. The PC/Smart phone acts as a receiving device. It has a higher rate of communication and is available at an affordable cost. Bluetooth has

Home Automation Using Face Recognition for Wireless Security

151

undergone several revisions in its technology. Bluetooth Class 1 operating range is up to 100 m and Class 2 range is up to 30 m. Most smart phones operate within Class 2 range. Bluetooth’s disadvantage is its limited range. Outside the range, the system will not be able to control the automated home appliances. The effective range is further decreased by Wi-Fi interference.

2.2 Voice Recognition-Based IoT Home Automation Voice recognition mode of communication can turn out to be very reliable. Since voice commands have shown consistency and accuracy since few years due to enhance development in the system, it is a prime reliable and efficient system to be implemented. The communication that takes place between the smart phone and the Arduino Uno is categorized as a wireless communication and is carried using the Bluetooth technology. This aids people who are specially challenged and want to control using their voice.

2.3 GSM-Based Home Automation System A vital system that is utilized by the smart home automation is the global system for mobile communication, also called as GSM. In the addressed system, the GSM system communicates between the main module and appliances using the mod of text messages. One of the drawbacks of the GSM-based home automation system that can be brought to notice is that it lacks the guarantee whether the text messages will be delivered to the designated system. Hence, it is particularly not reliable under various circumstances. These automation systems were particularly developed to reduce the human efforts by managing the increasing number of home appliances like television sets, music systems, refrigerators, dish washers, water heaters, air-conditioners, etc. The existing IoT-based home automation systems use core technologies such as Wi-Fi, Zigbee, Bluetooth, Arduino Board, GSM connectivity, etc. Each and every technology has some advantages and drawbacks. There is immense scope for further research to address the technology issues and remove their potential drawbacks.

3 Proposed Work To overcome the drawbacks, we are implementing “Home Automation using Face Recognition for Wireless Security”.

152

B. S. Umashankar et al.

3.1 Hardware Components A. Raspberry Pi 3 B + Board: Raspberry Pi is a remarkable fully functional digital computer which is inexpensive and compact. B. Node MCU: The IoT platform with respect to NodeMCU is an open source which is made available to everybody. The NodeMCU obviously will include a particular firmware, the firmware in return successfully runs on the ESP8266 Wi-Fi SOC. C. Relay Board: A relay is used to switch-ON or switch-OFF appliances. Relay selection is based on what needs to be switched ON and OFF. D. MQ135 Gas Sensor/Air Quality Sensor: This sensor ionizes the gases which come in its contact and varies the resistance of a small electro-chemical sensing material. E. DHT11 Temperature–Humidity Sensor: It is a low-cost sensor for sensing temperature and humidity of the environment. It can be easily interfaced with Raspberry Pi. The sensor consists of a capacitive humidity sensing element and a thermistor for sensing temperature. F. Infrared (IR) Sensor: It is an electronic device which senses some aspects of its surrounding environment by emitting and/or detecting infrared radiation. G. Camera Module: The camera module is powered from a single + 3.3 V power supply. It can be used to capture the images as well as videos. H. LCD Touch Screen: A liquid crystal display (LCD) touch screen has additional layers over the display element to provide touch functionality. I. Mobile/WEB Controller: The Mobile/WEB controller is the control module using which the user can interact with the system from a remote location as well as send commands to the system. J. GSM Module: A global system for mobile communication (GSM) module is used to establish communication between a mobile device and a GSM or GPRS system.

3.2 Software Requirements A. Raspbian OS: It is a Debian GNU/Linux-based operating system for Raspberry Pi. It is now called Raspberry Pi OS. B. Open CV Library: Open Source Computer Vision (OpenCV) library is an open source computer vision and machine learning software library. It has more than 2500 optimized algorithms. It focuses on real-time applications. It is written in C++ and its primary interface is also in C++. Its binding is in Python, Java and MATLAB. OpenCV runs on a variety of platform, i.e. Windows, Linux, macOS, OpenBSD in desktop and Android, IOS and Blackberry on mobile. C. Twilio Services: Twilio is a cloud communications platform as a service (CPaaS) company. Developers can use Twilio Web service APIs to programmatically

Home Automation Using Face Recognition for Wireless Security

153

make and receive phone calls. They can also send and receive text messages and perform other communication functions. Before getting into the details of the proposed system, we must understand the importance and the necessity for this work. A robust user interface application or portal that can be used to operate the entire home automation system without any technical guidance is lacking in most of the current automation systems. The designed user interface application should be easy to use, it should be reliable and accessible from anywhere in the world. It should not limit the user to operate the system infrastructure from just a particular area or from home. The process of upgrading the system has always been a challenge for the adaptation of the system. If a person wants to install a system, there is a need of the technical person to guide throughout the process of installation, and every device needs to be integrated into the control panel. After that, if a user wishes to upgrade the system by integrating new devices /appliances, it’s very difficult as the user needs to go for the whole process again as there is a need to reconfigure the system to integrate the new devices and would again need the assistance of technical guidance. Every system is expected to fail in one or the other scenario let it be a power failure, hardware failure or software failure. Moreover if all the systems are integrated at a single point, the failure of that single point would make the whole system down. Such difficulties faced by a user are addressed and this work proposes the following solutions and improvements. The Web interface of the proposed home automation system is used to operate and monitor the state of various appliances and user commands that can be easily modified to support additional hardware interface modules (Fig. 1). The advantage of this home automation system is that it is accessible from the smart phones with Internet connectivity. Also they can be accessed from a variety of Web/mobile controller devices. The speed, security, mobility and flexibility of the overall system is enhanced.

Fig. 1 Classification of application and technology

154

B. S. Umashankar et al.

4 Implementation Details The presented prototype can be implemented for three proposed systems (Fig. 2). (1) A Complete smart Home Automation and Wireless Security In this prototype, we will be able to control all electrical appliances from long distances through any Web/Mobile controller. In this project, we are integrating Web cam to recognize the person and if the person is authorized to enter the home premises. The automated system will trigger the door lock system, if the person is not recognized, the image of the person will be sent to the admin via Email/MMS to intimate about the arrival of unauthorized person and if the admin wants to allow the person, he can then using the Web/Mobile controller trigger the door lock system. This prototype is time and storage efficient as it will trigger the recording only when the sensors give any input, for example: imagine there is a movement detected by the IR sensor, then the microcontroller will start the recording using the Web cam, and hence, it becomes easy for the admin to check for that particular footage rather than wasting time in checking the whole footage. To make automation simpler, we can control lights and fans or any other appliance connected to a relay board through the Internet. Even though for some reason Wi-Fi is unavailable, we have the option to go to 3G or 4G services. This will help us in operating our home appliances from a distance. This will be of great convenience to the differently abled and senior people to control home appliances (Fig. 3).

Fig. 2 Block diagram

Home Automation Using Face Recognition for Wireless Security

155

Fig. 3 Complete smart home automation and wireless security

(2) Face Recognition Based Security System In this prototype, we are integrating Web cam to recognize the person and if the person is authorized to enter the home premises. The automated system will trigger the door lock system, if the person is not recognized, the image of the person will be sent to the admin via Email/MMS to intimate about the arrival of unauthorized person and if the admin wants to allow the person, then he can then using the Web/Mobile controller trigger the door lock system. We can also integrate sensors, for example, placement of an IR sensor in the premises. This sensor detects the motion of the object/human. This detected signal will in turn become the input the microcontroller. The owner, who may/may not be present in that premises, will be instantly alerted by the message along with the captured image that is sent as an E-mail/MMS stating “There is an intruder movement in the premises”. Further, to trigger the security modules, so that the unauthorized person will be warned, the owner can press ‘Alert’ button in the user interface of the mobile device application. If situation demands, the owner will be able to send an SMS to the concerned security person or to the police department stating the intruder situation with the captured image (Fig. 4). (3) Smart Home Automation In this prototype, we can control all the appliances connected to the module using a Web/Mobile-based controller. We can integrate various sensors to make automation efficient and smart. We can include an IR sensor so that whenever there is a movement, the lights of the staircase would turn ON automatically, we can also integrate the gas sensor and fire sensor so that the user can be intimated if there is a gas leakage or fire in the premises. We can also integrate a temperature sensor so that the user can maintain the room temperature as per his need well before entering the room. This automation will not only make the whole system efficient but also safe. This controller can be of great help to the differently abled people and senior people, as it can be used conveniently (Fig. 5).

156

B. S. Umashankar et al.

Fig. 4 Face recognition-based security system

Fig. 5 Smart home automation

5 Result After successful implementation of the system, following are the results of the tested software and hardware prototype (Fig. 6).

Home Automation Using Face Recognition for Wireless Security

157

Fig. 6 Intimation of updated status to the user

Using Twilio service, user was greeted and intimated about the updated status of the premises. The images are captured using the Web cam and stored in the database and send to the user to intimate about the intruder. Users can also forward these details to the police authorities (Figs. 7, 8 and 9). Advantages of the System (1) (2)

Time management is the key factor considered in this system. Specific surveillance using an IR sensor.

Fig. 7 Captured and stored images

158

B. S. Umashankar et al.

Fig. 8 Relay board connection

Fig. 9 Web/Mobile controller module

(3)

The system provides alert only in the presence of intruders using the GSM module. (4) Face recognition for enhanced security. (5) Memory management is concentrated. (6) This low cost system. (7) Great help for differently abled people. (8) Devices can be easily controlled and managed from a remote location. (9) Highly secured and time-saving. (10) Remote surveillance over Web service.

6 Conclusion In this work, a real-time implementation of home automation using face recognition is demonstrated. The system using Raspberry Pi has made the system compact. So it is more reliable than the PC-based system. It has all the advantages of open source software. Also, it sends security alerts to the authorized user. The power bank can

Home Automation Using Face Recognition for Wireless Security

159

be used to provide the power backup to charge the Raspberry Pi so that down time due to power failure is very minimal. Also it is power-efficient and provides enough flexibility to suit the different requirements of people. Hence, the system is low-cost, fast and highly reliable. The automation makes the daily life of the humans very easy and secures as there is no need to have an overhead regarding the control of devices. The overall system is implemented keeping in mind the ease and safety of the user. All the sensors are integrated in such a way that users can have the complete automated feel of the premises. As future possibilities, we can integrate an OTP-based system along with face recognition at the door lock system to deal with the current COVID-19 scenario where the users are wearing masks, and face recognition is challenging. Also, a log system can be integrated to keep track of the run time of the particular appliance.

Bibliography 1. Raghunandan M, Raghav P, Aradhya HVR (2018) Object detection algorithms for video surveillance applications. In: 2018 International conference on communication and signal processing (ICCSP), Chennai, pp 0563–0568 2. Ransing RS, Rajput M (2015) Smart home for elderly care, based on wireless sensor network. In: 2015 International conference on nascent technologies in the engineering field (ICNTE), Navi Mumbai 3. Brundha SM, Lakshmi P, Santhanalakshmi S Home automation in client-server approach with user notification along with efficient security alerting system 4. Vikram N, Harish KS, Nihaal MS, Umesh R, Shetty A, Kumar A A low-cost home automation system using Wi-Fi based wireless sensor network incorporating internet of things (IoT) 5. Culjak I, Abram D, Pribanic T, Dzapo H, Cifrek M A brief introduction to OpenCV 6. Ziegler S, Nikoletsea S, Krco S, Rolim J, Fernandes J Internet of things and crowd-sourcing—a paradigm change for the research on the internet of things 7. Voas J, Agresti B, Laplante PA (2018) A closer look at IoT ’s things. IT Professional 20(3):11–14 8. Hassan QF (2018) Introduction to the internet of things. In: Internet of Things A to Z: technologies and applications. IEEE 9. Shinoy DS (2017) Implementing the internet of things for renewable energy. In: Internet of Things A to Z: technologies and applications. IEEE 10. Y. Kung, S. Liou, G. Qiu, B. Zu, Z. Wang and G. Jong, “Home monitoring system based internet of things.” 11. Brundha SM, Lakshmi P, Santhanalakshmi S (2018) Home automation in client-server approach with user notification along with efficient security alerting system. In: 2018 International conference on smart technologies for smart core (SmartTechCore), Bangalore, pp 756–901 12. Benderius CB, Malmsten Lundgren V (2018) The best rated human machine interface design for autonomous vehicles in the 2016 grand cooperative driving challenge. IEEE Trans Intell Transp Syst 19(4):1302–1307

Hybrid-Network Intrusion Detection (H-NID) Model Using Machine Learning Techniques (MLTs) K. R. Pradeep, Arjun S. Gowda, and M. Dakshayini

Abstract Providing computer security is one of a significant challenge. Software and its mechanisms have been established to provide the security to avoid intrusion, which includes Intrusion Detection Systems (IDS). IDS helps to detect the exertions to outbreak a network and detect anomalous actions and its activities. It includes the details of uncertainty in probing of different kinds of attacks. IDS demands the need for combination of Machine Learning (ML) methods integrated into a hybrid model. In this paper, the Hybrid Machine Learning (HML) model is proposed that predicts the attacks against the network provided with better performance. The proposed system includes an innovative IDS having good network act helping to perceive the strange attacks, achieved by ML algorithms such as Decision Tree (DT), Naive Bayes (NB), Random Forest (FR), K-Means, and SVM. The algorithms with the good results are used to build a hybrid model. The proposed HML method improves the accurateness and efficiency for detecting the attacks by the IDS system. This research work recommends a system for access within a hybrid model, i.e., HybridNetwork Intrusion Detection (H-NID) which is a hybrid network combination of DT, K-NN, NB, SVM, and RF to extract temporary and local data of network traffic which advances the accurateness of IDS. The training phase in H-NID uses stage wise approaches to scale out the model. This technique decreases this consequence quantity of unparalleled trials of various attacks on training of model execution. It progresses the strength of training and prediction. Lastly, test of H-NID is done for several types of network traffic from the CICIDS-2017 database as it works on a real network traffic dataset that simulates real-world conditions. The predicted results demonstrate that H-NID got 98.50% of accuracy, and the accuracy for each type of attack remained more than 94.65%, which achieved excellent results in all models. Various rules and restrictions do not work well. Keywords Intrusion detection system · Decision tree · Naive Bayes · K-NN · SVM · Random forest K. R. Pradeep (B) · A. S. Gowda · M. Dakshayini Deparment of Information Science and Engineering, Don Bosco Institute of Technology, VTU, Bengaluru, Karnataka, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_16

161

162

K. R. Pradeep et al.

1 Introduction With the swift development of refined attacks and the susceptibilities on computer networks (inferred as intrusion), recognition of the intrusions has developed an extent of extreme importance. IDS supports observing the computer network and detects outbreaks, illegal events, or any malicious behavior. In turn, they are supporting the security of computers and networks. The basic architecture of an IDS is shown in Fig. 1. Cyber-attacks have become more sophisticated, reachable, and targeted than ever. So, safety strategies need to be developed continuously. In terms of discovery strategies, two types of IDS are identified such as anomaly detection and misdiagnosis [5]. Common problems under traditional methods of the uncommon acquisition include removing an incorrect network traffic feature and Fig. 1 Proposed framework

Hybrid-Network Intrusion Detection (H-NID) Model …

163

strain in creating attack detection systems, leading to a higher level of false alarm when mediating the traffic attack. It is hard for network security people in detecting anonymous threats, forcing them to do none. So, traditional methods do not apply to today’s Internet in terms of its large amount of data. The common methods are IDS based on ML. The ML-based approach is highly distinguished and notices network traffic by examining hand-drawn network traffic features. At the same time, the hybrid-based system will not only explore manual-generated features but also inevitably abstract features from real traffic. Therefore, hybrid-based techniques can prevent manual expulsion problems and improve detection accuracy compared to standard systems designed for ML. Hybrid-based access methods require many training details to achieve high accuracy for dissimilar kinds of attack on traffic data. In real-world environments and present databases [KDD99, CICIDS-2017], attack traffic remains relatively small compared to conventional traffic. In addition, some kinds of attack on network traffic are challenging to get and emulate, as the amount of data required for training is minimal for the existing model. These problems severely limit the accuracy of the hybrid-based approach, producing we be challenging to evaluate convinced types of attacks.

2 Related Work In growing proliferation of network security in the Internet, has become an severe problem. Rigid network behaviors such as DDos and aggressive force outbreaks are often ‘linked’ to wrong thoroughfare. Security investigators want to successfully analyze malignant traffic on the provided network which detects potential attacks and halt them immediately. Custom Discovery System: Traditional access methods mostly comprise mathematical analysis approach, signature analysis approach, and limit analysis approach. These approaches reflect traffic congestion; yet, they necessitate security investigators to have information associated to their own understanding; to do it, access program is based on the deep neural network (DNN). Accomplishment of ML algorithms often relies on the depiction of the data [6]. Depiction reading, also called as feature reading, a process of DN Communication, can be used to study descriptive elements of variance late data. They syndicate visual integration with DNN algorithms to find entry methods and use Deep Belief Networks (DBN) [7] to build an effective and formable login structure. Then, these research approaches develop their models that help to study presentations in hand-crafted traffic features without fully utilizing the capabilities of deep neural networks. It has been shown that the highest recognition percentage and the lowest accuracy level have an incorrect alarm which can be achieved by a set of advanced traffic features. Features of direct learning from underdone data may stand possible. This study has recognized a malicious traffic study mode created on an hybrid model to abstract and inspect unstructured network data for short-term and short-term data. In this work, training and testing are done

164

K. R. Pradeep et al.

on the CICIDS-2017 dataset. Sequence of tests is performed to demonstrate that the projected model eases the analysis of the active flow. Dataset IDS is an essential defense tool in intricate and massive network attacks, but in the absence of existing public data currently hampers the continuation. Several researchers use isolated data from an establishment to conduct statistics group test IDS lists that disturbs the reliability of the results. In this study, an effort is made to make better reflected real-time traffic patterns on real-time networks and new attack methods. The selected CICIDS-2017 (Canadian Institute for Cybersecurity) [8] database contains traffic congestion along with shared time-related attacks representing real network traffic. This data creates a specific characteristic of 25 handlers depending up on FTP, HTTP, SSH, HTTPS, and e-mail protocols that exactly imitate the actual network situation. Collected outbreak traffic attacks include eight types of attacks: login, SSH-Patator, Web Attack, Heartbleed, DDoS, DoS, Botnet, and FTP-Patator. As shown in Table 1, attacks have been taken place on morning and afternoon on these days Tuesday, Wednesday, Thursday, and Friday. Normal traffic remained during the whole day on Mondays and at non-stop time from Tuesday to Friday. The data type used for this database is pcap file. Later obtaining the data, analysis is done with the unique data and seven sorts of data intended for the next test in terms of data value and sound quality were selected: Common, PortScan, DoS, SSH-Patator, Infiltration, FTP-Patator, and Heartbleed. From Table 2, it is observed that dataset attack data for five days of traffic data is finest for strategy-based multiclass recognition model. Still, it would be pointed that the top detected model is able to detect the attacks. Consequently, to plan a distinctive IDS, traffic data of entire days must be united to a method sole dataset of IDS. By inclusion of files accessible in Table 2, the entire form of a dataset was created which comprises 31, 19, and 345 occurrences and 79 features comprising 15 Table 1 Dataset CICIDS-2017

Day

Type of traffic

Monday

Benign (Normal)

Tuesday

Benign, FTP-Patator, SSH-Patator

Wednesday

Benign, Dos Golden Eye, Dos Hulk, DoS Slow http test, Dos slowloris, Heartbleed

Thursday

Benign, Web Attack—Brute Force, Web Attack—SQL Injection, Web Attack—XSS, Infiltration

Friday

Benign, Bot, PortScan, DDos

Hybrid-Network Intrusion Detection (H-NID) Model …

165

Table 2 Explanation of files for CICIDS-2017 dataset Name of files

Day activity

Attacks found

Monday-Working Hours.pcap_ISCX.csv Tuesday-WorkingHours.pcap_ISCX.csv

Monday Tuesday

Benign (Normal human activities) Benign, FTP-Patator, SSH-Patator

Wednesday-workingHours.pcap_ISCX.csv

Wednesday

Benign, DoS Golden Eye, DoS Hulk, DoS Slowhttptest, DoS Slowloris, Heartbleed

Thursday-WorkingHousrs-Morning-Web Attacks.pcap_ISCX.csv

Thursday

Benign, Web Attack—Brute Force, Web Attack—Sql Injection, Web Attack—XSS

Thursday-WorkingHoursAfternoon-Infiltration.pcap_ISCX.csv

Thursday

Benign, Infiltration

Friday-working hoursMorning.pcap_ISCX.csv Friday-WorkingHours-Afternoon-PortScan.pcap_ISCX.csv

Friday Friday

Benign, Bot Benign, PortScan

Friday-Working Hours-Afternoon-Dos.pcap_ISCX.csv

Friday

Benign, DDoS

class. Additionally, inspecting the cases of the collective files, the dataset comprises missing class label of 2, 88, and 602 instances and 203 models consisting missing info. Removing such instances, the combined dataset of CICIDS-2017 is comprising of about 2,83,0540 cases along with no superfluous instances were detected. The features of the collective dataset along with class incidence yields are shown in Tables 3, 4, 5 and 6 correspondingly. Additional thing is observed that the dataset fulfills all the measures for precise IDS dataset, which includes whole network arrangement, traffic, labeled dataset, interface, available protocols, seizure, metadata attack diversity, and feature set. The CICIDS-2017 dataset format is single pcap file per day, which contains a lot of particulars, which is not appropriate for machine training so it is to split constant pcap files into many different components based on specific consistency. This involves six types to reduce network traffic by TCP, session, connection, category of service, network flow, and capture.

166 Table 3 CICIDS-2017 dataset records

K. R. Pradeep et al. Type of attack

Day

Total records

Benign Brute Force Attack

Monday Tuesday

529,918 445,909

Heartbleed Attack/DoS Wednesday Attack

Table 4 CICIDS-2017 dataset overall characteristics

Table 5 CICIDS-2017 dataset classwi.se instance occurrences

Table 6 Confusion matrix

692,703

Web Attack Infiltration Attack

Thursday (Morning) 170,366 Thursday (Afternoon) 288,602

Botnet Attack

Friday (Morning)

191,033

PortScan Attack DDoS Attack

Friday (Afternoon) Friday (Afternoon)

286,467 225,745

Dataset name

CICIDS2018

Dataset type

Multi-class

Year of release

2017

Total number of district instances (After removing missing instances) Number of features

2,830,540 79

Number of district classes

15

Class labels

Number of instances

BENIGN

2,359,087

DoS Hulk

231,072

PortScan

158,930

DDoS DoS GoldenEye

41,835 10,293

FTP-Patator

7938

SSH-Patator DoS slowloris

5897 5796

DoS Slowhttptest

5499

Botnet Web Attack—Brute Force

1966 1507

Web Attack—XSS

652

Infiltration Web Attack—Sql Injection

36 21

Heartbleed

11

Type of class

Predicted class Benign

Attack

Benign

True Positive

False Negative

Attack

False Positive

True Negative

Hybrid-Network Intrusion Detection (H-NID) Model …

167

3 Methodology and Proposed Framework This main aim of the projected H-NID model works on the CICIDS-2017 dataset, which is evaluated based on ML algorithms. The main strategy is on data pre-processing, feature selection, and CICIDS-2017 dataset. The following are the stages that are involved in the projected model as shown in Fig. 1. • Data Pre-processing: In this stage, it normalizes the initial data format, i.e., CSV file for the preparation of the data in a precise format which is appropriate for base of the data for data mining algorithms. In pre-processing, rarely the data might not be continually extant in a single place [9]. So, in that situation data needs to be grouped from diverse places and changed into a single and appropriate format, trailed by data cleaning and normalization. • Feature Selection: Here, the optimal feature is selected for classification. This shared data is processed on a feature selection algorithm that handles linearly and nonlinearly dependent data features. Finally, its efficacy is examined in occurrences of network intrusion detection. • Analyze the Results: Analysis is done on the dataset based on with normalization and without normalization. Normalization is a process of execution, which operates the dataset properties for some types of ML algorithms [10]. The main benefit of discretization is that some classification methods work only on notified properties, and it increases the order of correctness that relies upon observed information. The performance of ML algorithms is enhanced by discretization which alters constant data attribute values to a finite set with slight loss of info. In the next stage, apply the ML algorithms on CICIDS-2017 dataset and measure the performance dataset. • Analyze Trials: Analysis is done as per actions and procedures, reflected to the classification algorithms for feature selection which involves the explorations of choosing ideal features from the feature set nevertheless the kind of relationship among them. • Evaluation: In this final stage, evaluation of model is done based on performance which includes evaluation metrics (accuracy, time taken, recall, precision and F-measure).

3.1 Classification Algorithms Used A. Decision Tree The decision-making process (DT) is the most widely used ML methods in binary categories. It can predict based on training data after making quick choices by creating a small tree. Trees have nodes, and attributes mean test, as its name implies. The branch represents the test result, and the leaves are labeled for the separation of each last or last item. The utmost significant step in this process is to choose the best

168

K. R. Pradeep et al.

material. Ross Quinlan developed the tree algorithm ID3 [11]. Data mining and data mining were its main applications. It is now used to process natural language and Machine Learning. In this paper, the proposed model used the ID3 algorithm to determine whether the website was legitimate or criminal for sensitive information. To determine the result of a split algorithm, the following steps are taken: (1) Start by reviewing the training dataset. Give it the letter ’S,’ and it must be structured and separated. (2) Find out which datasets have the best quality. (3) Divide the ’S,’ each with several excellent features. (4) Create a node of the Decision Tree that contains the best attribute. (5) Repeat step 3 repeatedly to create a new solution tree unless you can no longer split it. As a result of the section, it represents the area behind the leaf. Entropy can be used to obtain information. Different attributes with numerical values are selected using divisive and profit-denominated data. B. Naïve Bayes Algorithm It is a conditional planning algorithm. Bayes theory governs its operation. It is optional because it is easy to use and requires little training. However, it is not desirable for various reasons, including low limitations for a few details, assuming the features are independent. XG Boost is a tree-based algorithm that makes decisions that come first in terms of speed and performance. Its purpose is to minimize damage to the old tree by creating a new tree regularly [12]. This approach, however, can be time-consuming. It is also easier than balance. C. Random Forest The traditional forest approach of arbitrary decision-making was first proposed by Ho in 1995. Ho established that tree forests separated from wet planes could gain accuracy as they grow without suffering the excessive increase, as long as the forests are periodically barred from being sensitive only to the magnitude of selected features. The following work on the same line concludes that other separation methods behave in the same way, as long as they are randomly forced to be resistant to the magnitude of a particular factor. Note that this perception of complex segregation (more extensive forest) to obtain more precisely is almost independent of the general belief that the severity of the divider can grow to a certain degree of precision before being severely damaged. An explanation of the forest’s approach to overcrowding can be found in Kleinberg’s stochastic prejudice. • From the training set, let the sum of cases be N, and if N cases are taken as random—by changing the unique data and later this sample is used for the training. • Since any variation in the input of M, the number m 0.05: compatible_classes[extension_class_list[i]] ← [base_class_list[ j]] j ← j + 1 i ← i + 1 return compatible_classes

4 Experiments and Discussions The S-Extension patch outperforms the previous techniques in terms of the following criteria: 1. Speed and ease of adaptation 2. Less need of data at the instance 1

All code can be accessed at https://github.com/Dishant-P/S-Extension-patch-Official-researchmodule.git.

S-Extension Patch: A Simple and Efficient Way to Extend ...

239

Algorithm 2 Dual-parallel inference algorithm INPUT: video: The inference video OUTPUT: Prediction for length of video pr ediction ← detector ( f rame) lock(shar ed_memor y) update shar ed_memor y with prediction and frame r elease() if there are any compatible_classes in pr ediction.classes lock(shar ed_memor y) classi f ier (compatible_classes_ regions ) update (shar ed_memor y) render the frame (and/or) save the results else lock(shar ed_memor y) render the frame (and/or) save the results r elease() Table 2 Comparison of S-Extension patch with earlier techniques Technique Need for data Speed of Time for Performance preprocessing adaptation training on old tasks Learning without forgetting Fine tuning Joint training S-Extension patch

Performance on extension tasks

Low

Slow

Medium

Best

Best

Medium High None

Fast Fast Fast

Medium High Low

Good Best Best

Fair Best Best

3. Adaptability to any network or system 4. Computational resources required. Additionally, the S-Extension patch can maintain the accuracy of the overall model as well. The S-Extension patch breaks down the speed of training and computational resources required by at least 75%. In the fine-tuning technique, the model forgets the previous knowledge and hence is not able to perform the previous tasks properly. The joint training technique works well on both old and new tasks but is not able to train quickly and requires a lot of data and annotations for the same. The technique of Learning without forgetting closes off both the previous limitations for the need of data and performance, but it still takes the same time for training as a regular object detection model and has the need for annotating the new data as well. Also, all three techniques require significant computational power to train the model. The S-Extension patch addresses all these limitations by leveraging the knowledge of the older tasks. I do concede there is some need for old tasks data but from an industry point of view, the old data is not a problem but the preprocessing time

240

D. Parikh

Table 3 Similarity matrix for three classes from COCO dataset and an extension class Class Bus Car Truck Van Bus Car Truck Van

0 0.977 0.0314 0.0468

0.0977 0 0.0685 0.0378

0.0314 0.0685 0 0.0292

0.0468 0.0378 0.0292 0

and cost of training is. As the S-Extension patch only involves training a regular classifier, it does not involve many preprocessing steps, especially annotation. And it also doesn’t need the computational resources or time for training. Table 2 sums up the entire argument on the criteria. Note: all techniques have relatively similar inference timings.

4.1 Dual-Parallel Inference Technique In the experiments, I trained a YOLOv5x model over the COCO dataset using a pre-trained set. The model gave a test mMAP (0.5:0.95) score of 50.4. The extension class: Van, was appended to the 80 base classes of the COCO dataset. In the joint training method, it took about 120 hours for my system to train the model and achieve the overall test mMAP (0.5:0.95) score of 49.8 (with extension class appended). In the S-Extension patch method, however, it took 2.4 hours for a ResNet152 classifier model to reach mMAP (0.5:0.95) score of 50.7 for the four similar classes namely, Car, Bus, Truck, and Van (extension class), identified with the similarity threshold, as shown in Table 3. The increase in the mMAP score for the S-Extension patch is not only because of the higher accuracy of extension classes but also an increase in the classification accuracy of the other three base classes. As seen in Table 3, not only Van but even Truck class is highly similar with Bus. It means that by training the classifier for highly similar classes, the model gets an additional chance to correct the classification done by the detection model and increase the overall accuracy. It is one of the most crucial advantages of the S-Extension patch as it allows us to use this technique not only in need of extension classes but also when a particular class gets a lower score or while working with a dataset with highly similar classes.

4.2 Technique with Trackers In the tracker-based algorithm technique, I used the DeepSort algorithm with class labels extraction. Although usually, the DeepSort tracking algorithm is not used with

S-Extension Patch: A Simple and Efficient Way to Extend ...

241

class labels, it was important to extract those to use it with the S-Extension patch. The tracker algorithm-based approach has one crucial advantage about the overall accuracy. Not only does it help in increasing the overall accuracy like the one explained in Sect. 4.1, but also in reducing the additional computational resources required and the inference timings. The technique with trackers achieves the same final mMAP score (0.5:0.95) of 50.7 as with the Dual-parallel inference technique but with 12% lower inference timings and 37% decrease in the overall computational usage.

5 Future Works and Llimitations Although the S-Extension patch covers most of the limitations, it is a relatively new technique with a lot of scope still to be explored. The technique can be further improved by using other similarity measures or appending the patch directly to the model while training and keeping a unified architecture. The S-Extension patch can be made more scalable by using a two-step detection model and using the patch for just the classification step, and fine-tuning the model there. It can significantly improve the speed and the time to implement the inference strategy. The strategy is also being experimented on on different visual models including classifiers and graphics rendering algorithms.

6 Conclusion The paper shows how we can leverage the knowledge of the previous data and performance to quickly extend the object detection model and implementation variations of the technique. I give considerable evidence to prove how the technique works better than other defined strategies on various criteria. The S-Extension patch is an elegant technique that can be used with any object detection model fulfilling the similarity threshold condition. S-Extension patch is easy to adapt to any system, can be trained quickly, and maintains (and in some cases increases) the accuracy as well the inference speed.

References 1. Datta R, Li J, Wang JZ (2005) Content-based image retrieval: approaches and trends of the new age. In: Proceedings of the 7th ACM SIGMM international workshop on multimedia information retrieval, pp 253–262 2. Ding K, Ma K, Wang S, Simoncelli EP (2020) Image quality assessment: unifying structure and texture similarity. arXiv:2004.07728 (2020) 3. Eakins J, Graham M (1999) Content-based image retrieval

242

D. Parikh

4. Gudivada VN, Raghavan VV (1995) Content based image retrieval systems. Computer 28(9):18–22 5. Hinton G, Vinyals O, Dean J (2015) Distilling the knowledge in a neural network. arXiv:1503.02531 6. Jung H, Ju J, Jung M, Kim J (2016) Less-forgetting learning in deep neural networks. arXiv:1607.00122 7. Jung H, Lee S, Yim J, Park S, Kim J (2015) Joint fine-tuning in deep neural networks for facial expression recognition. In: Proceedings of the IEEE international conference on computer vision, pp 2983–2991 8. Li Z, Hoiem D (2017) Learning without forgetting. IEEE Trans Pattern Anal Mach Intell 40(12):2935–2947 9. Plummer BA, Vasileva MI, Petsiuk V, Saenko K, Forsyth D (2020) Why do these match? explaining the behavior of image similarity models. In: Computer vision–ECCV 2020: 16th European conference, Glasgow, UK, 23–28 Aug 2020, Proceedings, Part XI 16. Springer, pp 652–669 10. Ragkhitwetsagul C, Krinke J, Marnette B (2018) A picture is worth a thousand words: code clone detection based on image similarity. In: 2018 IEEE 12th International workshop on software clones (IWSC). IEEE, pp 44–50 11. Shin HC, Roth HR, Gao M, Lu L, Xu Z, Nogues I, Yao J, Mollura D, Summers RM (2016) Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning. IEEE Trans Med Imaging 35(5):1285–1298 12. Shnain NA, Hussain ZM, Lu SF (2017) A feature-based structural measure: an image similarity measure for face recognition. Applied Sci 7(8):786 13. Soltau H, Saon G, Sainath TN (2014) Joint training of convolutional and non-convolutional neural networks. In: 2014 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, pp 5572–5576 14. Wang L, Rajan D (2020) An image similarity descriptor for classification tasks. J Vis Commun Image Represent 71:102847 15. Wang YX, Ramanan D, Hebert M (2017) Growing a brain: fine-tuning by increasing model capacity. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 2471–2480 16. Zhang C, Xie F, Yu H, Zhang J, Zhu L, Li Y (2021) PPIS-JOIN: a novel privacy-preserving image similarity join method. Neural Process Lett 1–19

Roles and Impact of ASHA Workers in Combating COVID-19: Case Study Bhubaneswar Manjusha Pandey, S. N. Misra, Abhipsa Ray, and S. S. Rautaray

Abstract The present COVID attack has significantly accentuated vulnerability of infections at community level where their basic healthcare and immunization program are being implemented by the ICDS scheme launched in 1975 through a slew of AWCs scattered all over the country. The AWC acts as a primary health center that provides supplementary nutrition to children (between 0 and 6 years of age) and pregnant and lactating mothers besides providing preschool education to children in the age group of 4–6 years. The ASHA workers associated with AWC under then NRHM are the first hand health workers available to them at the community level who act as a bridge between the dispensaries and the community members. With the outbreak of COVID-19, the role of ASHA workers has assumed increased salience as the governments are relying on them for community level combating of this outbreak. This paper takes a close look at fund allocations to the public healthcare sector among the developing and developed countries and also the interstate allocations and allocations for major schemes and the resultant impact on HDI. Keywords AWC · ASHA · SERVQUAL · HDI · COVID

1 Introduction Amidst combating the unprecedented second wave of COVID-19 pandemic, the Supreme Court of India made an observation that the health infrastructure inherited over past seventy years is not adequate. The COVID-19 outbreak wreaked havoc on the Indian healthcare system and caused 4 lakh deaths till July, 2021. India trails behind in almost all health indicators laid down by World Health Organization. This unsatisfactory healthcare system is because India has one doctor to for 1456 populations as against the WHO standard of 1:1000.This ratio drastically varies in M. Pandey · S. S. Rautaray School of Computer Engineering, KIIT University, Bhubaneswar, India S. N. Misra · A. Ray (B) School of Management, KIIT University, Bhubaneswar, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_23

243

244

M. Pandey et al.

the rural India (1:10,000) which accounts for 65% of the Indian population. 75% of the public healthcare infrastructure is concentrated in urban area whereas there is just one public hospital per 90,000 people in the rural areas.1 The rural healthcare sector is often characterized by understaffed facilities, inadequate infrastructure, and a lack of pharmaceuticals. There are around 1.58 lakh Subcenters (SCs); 25,743 Primary Health Centers (PHCs), and 5624 Community Health Centers (CHC) to facilitate healthcare services to the rural population.

1.1 Fund Allocation to Public Healthcare and Human Development Index It has been demonstrated conclusively that the HDI of a country is positively correlated with public health allocation and quality foundational education. India spends mere 3.54% of GDP (2018), whereas the developed countries [USA (16.9%), Germany (11.4%), Japan (10.9%)] spend much larger portions of their GDP. Among the BRICS countries, India spends the least: Brazil (9.5%), South Africa (8.2%), Russia (5.3%), and China (5.3%).2 The allocations reflect on the country’s HDI. In the Human Development Index, India stands at 131st place out of 189 countries, with a score of 0.645 while the HDI of developed countries and developing economies like China and Sri Lanka shows a healthier picture. The HDI of major developed and developing countries has been given in the Table 1. India stands at 131st place in the UNDP Human Development Index (2020) with 0.64 points. The HDI has taken the planetary pressure (per capita carbon dioxide emission and carbon footprint) in to consideration with the three basic dimensions: Life expectancy at birth, Years of Schooling, Gross National Income per capita in 2020 and released the adjusted HDI. The life expectancy at birth in India is 69.7 years, little below the South Asian average (69.9 years), and it has increased by 11.8 years in the past thirty years. The expected years of schooling are 12.2 years which has increased by 4.5 years in the past three decades.

1.1.1

Comparison of Healthcare Expenditure of Indian States

Health being a state subject in India, the public health expenditure is shared by the center and state. According to National Health Profile 2019 [1], India spends only 1.28% of GDP on public healthcare with the center bearing 37% of the cost and the states bearing 63%. However, the amount of money spent by each state 1

Chakraborty, I. (2018). Rural Healthcare Sector: A challenge yet to be resolved. Business Economics. Retrieved 20 October 2021, from https://businesseconomics.in/rural-healthcare-sectorchallenge-yet-be-resolved. 2 Current health expenditure (% of GDP). Data. Data.worldbank.org. Retrieved 17 October 2021, from https://data.worldbank.org/indicator/SH.XPD.CHEX.GD.ZS.

Roles and Impact of ASHA Workers …

245

Table 1 HDI ranking of developing and developed countries Country

HDI value (2019) Global ranking Public health expenditure (% of GDP) (2018)

Norway

0.957

1

10.05

Germany

0.947

6

11.43

Canada

0.929

16

10.79

USA

0.926

17

16.89

Israel

0.919

19

7.52

Japan

0.919

19

10.95

Russia

0.824

52

5.32

Sri Lanka

0.782

72

3.76

Brazil

0.765

84

9.51

China

0.761

85

5.35

South Africa 0.709

114

8.25

India

0.645

131

3.54

Bangladesh

0.632

133

2.34

Pakistan

0.557

154

3.20

Source UNDP HDI report (2020)

varies substantially. Community Health Centers (CHCs) and Primary Health Centers (PHCs) are established by state governments, and they get financial and technical assistance for the same under the National health Mission from the center. The allocations to public health expenditure by some key Indian state have been given in Fig. 1. Assam spends the highest percentage (7.1%) of its GDP on health and family welfare as against the average state allocations (5.5%). Odisha, West Bengal, UP, Bihar, Kerala, and Tamil Nadu spend higher than the average state allocations, whereas Punjab Maharashtra spends 4% of the state GDP. The National Health Policy 2017 calls on the states to spend at least 8% of their GDP on health, two third of which should on primary healthcare. In terms of available health professionals per 10,000 populations, Kerala does remarkably well with the largest number of health professionals (42 doctors and 23 nurses per 10,000 people), whereas states like Bihar, Chhatishgarh, and Odisha lag behind. The comparison of overall allocations to the department of health and family welfare and major schemes is discussed in the Table 2. The National Health Mission alone has been allocated almost half of the total allocation in budget estimate (Rs. 36,576 crore) in 2021–22. The National Rural Health Mission (NHRM), a component of NHM, has been granted Rs. 30,100 crore that is 6% more than 2020–21 revised estimates. There is also an increase of 5.2% in the allocation for the National Urban Health Mission (NUHM) in 2021–22 over the previous year revised estimates. The other item under the NHM head includes funds for medical education has quite surprisingly declined by almost 11% in 2021–22 over the previous year revised estimates. Just like the previous year, the BE for 2021–22 has received more than 100% increase as compared to the RE of 2020–21 under

246

M. Pandey et al. Allocations to Health and Family Welfare and percentage of State GDP of key Indian states FY 2021-22 (BE) Budgetary Allocation (‘000) (Rs. Crore) 32.01 19.06

18.63 13.01

12.23

10.35

Percentage of SGDP (%)

16.57

11.3 4.66

6.5

6.1

5.3

6.3

6.3

5.7

4.4

9.34

7.31 4

5.8

6.7

6.4

7.39 7.1

Fig. 1 Allocations to Health and Family Welfare and percentage of State GDP of key Indian states FY 2021–22 (BE). Source Compiled by authors from the state budget 2021–22. Note Average state allocation to health and family welfare: 5.5%

the head Ayushman Bharat: PMJAY which is a public health assurance scheme that provides healthcare assistance up to Rs. 5 Lakh per family per year to the vulnerable class. Currently, 10.7 crore families are covered under this scheme. It is to be noted that although BE for 2020–21 increased by 100% over RE of 2019–20, and the RE for 2020–21 had actually declined over even the utilized fund (RE) of 2019–20. The allocation for the PMSSY, which is a program to address regional disparities in the availability of affordable tertiary healthcare services, has also declined by almost 7% over the RE of 2020–21. Apart from these major allocations, 13,857 crore rupees and Rs. 360 crore rupees have been allocated to COVID-19 Emergency Response and Health System Preparedness Package and COVID-19 vaccination for healthcare workers and frontline workers. Figure 2 shows that the budgetary estimates have sharply increased from 2016–17, and to combat against the pandemic situation, the allocations to health infrastructure have further increased significantly. Up to 2014–15, there has been a trend of underutilization of funds, but from 2015–16 to 2020–21, the fund utilization has been more than 100% which implies that the various schemes run by the department has good fund absorption capacity. The highest percentage of actual expenditure (121%) can be seen in the year 2020–21 which is clearly because of COVID-19 pandemic.

1.2 Health Status and Global Hunger Index of India Over the past decades, India has made significant progress in terms of improving average life expectancy (from 35 years in 1950s to 68 years), reducing IMR (28 deaths per 1000 live births) and MMR (113 per 100,000 live births), eliminating polio,

Roles and Impact of ASHA Workers …

247

Table 2 Comparison of budget allocations to the Ministry of Health and Family Welfare and other schemes (2019–20 Actuals to 2021–22 BE) (in Rs. crore) Major heads

2019–20 (Actuals)

2020–21 (BE)

2020–21 (RE)

2021–22 (BE)

% change (2019–20 actuals to 2021–22 BE) (%)

Department of Health and Family Welfare

62,397

65,012

78,866

71,269

7

Department of Health Research

1861

2100

4062

2663

National Health Mission

34,660

33,400

35,144

36,576

• NRHMa

29,987

27,039

28,367

30,100

• NUHMb

850

950

950

1000

8

• Tertiary Care Programs

241

550

312

501

44

• Strengthening of State Drug Regulatory System

206

175

130

175

− 8%

• Human Resources for Health and Medical Education

3376

4686

5386

4800

19

PMSSYc

4683

6020

7517

7000

22

PMJAYd

41

20 3 0.2

3200

6400

3100

6400

Family Welfare Schemes

489

600

496

387

− 11

Infrastructure Development for Health Research

148

170

169

177

9

64,258

67,112

82,928

73,932

7

Total

Source Demand for Grants, Ministry of Health and Family Welfare, Union Budget, 2021–22; PRS [2] a National Rural Health Mission, b National Urban Health Mission, c Pradhan Mantri Swasthya Suraksha Yojana, d Pradhan Mantri Jan Arogya Yojana

reducing mortality caused by malaria and tuberculosis (TB), Kala-Azar. Despite the progress made in addressing the various health concerns, a large section of the rural population still die due to preventable diseases, suffer from malnutrition and complications during pregnancy and child birth. The rural population largely dependent on the public healthcare system as the private healthcare system is highly expensive that makes it unaffordable for the poor vulnerable class.

248

M. Pandey et al.

100000 80000 60000

BE

40000 20000 0

Actuals 97% 91%

82% 82% 87% 112% 102% 109% 100% 100% 121%

% Utilisation

Fig. 2 Budget allocation and expenditure of the Department of Health and Family Welfare from 2010–11 to 2020–21. Source Compiled by the authors (Data taken from Union Budgets, 2010–21)

In the recent Global Hunger Report (2021), India has a score of 27.5 and ranks 101st among 116 countries, which is quite alarming. Undernourishment, wasting of children, stunting of children, and child mortality are the four parameters on the basis of which the Global Hunger Index is calculated. According to the National Family Health Status (NFHS) (2019), 53.4% children and 53% of women of the age group 15–49 years are severely anemic. Malnourishment, poor antenatal care, poor hygiene, and sanitation are prevalent and ever existing. People in rural India die mostly from preventable diseases and lack of awareness about the health and sanitation. The Global Nutrition Report (2020) identifies malnutrition to be one of the biggest challenges of India and expresses concern that the COVID-19 pandemic has the potential to undo the progress achieved so far in terms of reducing malnutrition and hunger.

1.3 Role of Anganwadis and ASHA Workers The Anganwadi centers are at the fulcrum of early child development program [3] which includes provision of supplementary nutrition, immunization of all children, provide nutrition and health education, and basic health checkups [4]. As an integrated program for Children 0–6 years and expecting and lactating mothers, the program includes pre-natal care and post-natal care and care of new born babies. The Accredited Social Health Activists (ASHA) program was unveiled in 2005 as a part of the NRHM scheme to complement the work of the Anganwadi workers and ANMs and act as a bridge between the community and district health center besides providing basic healthcare facilities to babies and their mothers [5–7]. They are the voluntary health activists residing in the community being vigilant of the community health and facilitate the door step healthcare service to the beneficiaries. One ASHA worker is selected for every community of one thousand populations. The ASHA is primarily a female worker aged between 24–45 years having at least 8 years of formal education residing in the same community [8]. She is trained to

Roles and Impact of ASHA Workers …

249

deliver primary healthcare services at the grass root level by visiting the households in her locality. Her work is majorly focused on maternal and women’s healthcare for newborns and children. Motivating women to give birth in hospitals instead of through midwife, bringing children to immunization clinics, and encouraging family planning among the rural population comes under purview of her service [Ministry of Women and Child Development, (2012–15)]. She is also responsible for preventions of common infections, health nutrition, hygiene and sanitation, and motivating people to use the existing healthcare services. She receives incentives based on her performance. The roles and responsibilities of ASHA workers have evolved over the period. Along with Auxiliary Nurse Midwife (ANM) and Anganwadi workers (AWW), the ASHA has become an important element of the NRHM program in delivering the primary healthcare services in rural areas.

1.3.1

The Enhanced Role of ASHAs During COVID-19 Pandemic

The WHO declared COVID-19 as a public health emergency of international concern and declared it as a pandemic on 11th March 2020. Consequently, a nationwide lockdown was imposed in India necessitating the Anganwadi centers to close down accordingly the vital role that the Anganwadi centers, and ASHA workers were performing in terms of nutritional care for children and expecting mother and basic healthcare support to them has come to a grinding halt. India witnessed a reverse migration, laborers returning back to their native places due the lock down which heightened the possibility of the virus spreading at the community level. The ASHAs were included in the frontline workers attempting to prevent the spread of virus at the community level, in addition to their usual tasks of maternity and child healthcare and nutrition. In an extremely welcomed initiative, the ministry has launched a training program for the ASHA workers for supporting community surveillance process strengthening community linkage with public health services on preparedness, prevention, and control, enhancing control of public health measures and protection of healthcare workers against COVID-19. The ASHA workers’ expanded duty as frontline warriors put them under a lot of stress, challenges, hardships, and constant fear of getting infected themselves as well as their families. Recent survey conducted by OXFAM India in four states (Uttar Pradesh, Odisha, Bihar, and Chattisgarh) cites the dreadful working condition of ASHAs where only 23% of ASHAs provided with protective bodysuits, only 75% had masks, and 62% had gloves to perform their duties and many of them reported not receiving any training on how to use the PPEs on the field [9].

250

M. Pandey et al.

2 Literature Review (1) International The dependency and efficient usability of health infrastructures is totally dependent on the health workers across countries; hence, there is an acute requirement of trained workers and the same surges in the crisis times of pandemic and requires inclusion of even semi-trained lot of these health workers as mentioned by Wang et al. [10] in their research work report. Medical teams containing around 41,600 healthcare workers from thirty province and municipalities across China have been dispatched so far by the National Health Commission of the People’s Republic of China (NHCPRC) to Wuhan and Hubei to support the ongoing medical treatment to contain the Covid19 outbreak in Mainland China (New.qq.com.; 2020).The report raised concern for spreading of the virus through this increased number of semi-skilled workforce due to lack of information. According to Wang et al. [10], frontline healthcare workers (excluding infectious disease physicians) got insufficient IPC training, leaving them unaware of IPC for respiratory transmitted infectious illnesses. Healthcare staffs have not had enough time for comprehensive training and practice since the start of emergency responses. There was a dearth of professional supervision and direction, as well as monitoring methods. The potential threat of infection for healthcare professionals was heightened as a result of this condition. Also the outbreak of epidemics has a major impact on the quality of work life for health workers which otherwise also is not very suitable as studied by Almalki et al. [11] in their cross-sectional study as an attempt to analyze the quality of work life of health workers in the Jazan region, Saudi Arabia. Intra-hospital infection and transmission of pandemics such as COVID-19 are another major component that makes health personnel more likely to be super-spreaders of epidemics. The strategies and measures to protect healthcare workers in an acute tertiary hospital are described along with the domains of technologies and tools, work environmental factors, work task, and organizational conditions according to Gan [12], who has emphasized preventing intra-hospital transmission of communicable disease and also has suggested the use of the Systems Engineering Initiative for Patient Safety model. In the wake of a pandemic, the idea of zero occupational infection continues to be an attainable aim that all healthcare systems must strive for. (2) National Adhering to the guidelines received from WHO, Dr. Prashanth N Srinivas and his team [13] prepared a checklist of preparedness of primary health centers for combating COVID-19 and has emphasized the role of ASHAs workers as frontline workers for screenings and referrals in preparedness of combating COVID-19 at the community level. Ministry of Health and Family Welfare of the Government of India (2020) stated ASHAs as frontline workers for screening and referrals of COVID-19 infections at the community level as well as their roles and responsibilities in disseminating important information in the community regarding the preventative measures. The roles and responsibilities of the ASHAs specified by the ministry are to take active measures for early detection and referral of suspected COVID-19 cases, follow

Roles and Impact of ASHA Workers …

251

up patients in quarantine, checking on the high risk categories, informing the PHC worker/ANM/doctor, providing detailed information on personal protection (social distancing, hand hygiene, usage of mask), distributing mask to high-risk category among all the households in her locality (Ministry of Health and Family Welfare 2021). According to Desai et al. [14], 33% of Anganwadi staff felt burdened on their fundamental Anganwadi tasks as a result of their participation in other national health programmers and other activities. Anganwadi workers, who are underpaid and overworked, are the true suppliers of many fundamental services for impoverished Indian people; the study reveals that the AWWs are overworked and unable to rationalize their monotonous labor [15]. The government health officials and other authorities must remember that under the National Rural Health Mission, and a second comparable cadre called Accredited Social Health Activists (ASHA) has been created in each village that must be effectively utilized. As a result, the AWWs will be able to serve the community as per its needs. Saxena et al. [16] have detailed about the model followed in Uttrakhand and how the same model can be helpful pan India. Deena and Sivanesan [17] in their study on socio-economic status of Anganwadi employees with reference to Kanyakumari district have detailed about the problems faced by ICDS scheme implementers and their requirements.

3 Research Gap Even though there are several studies addressing the workload, stress and the petty incentives in comparison the numerous work that ASHA workers do; very few literatures were found regarding the training and its effectiveness as well as technological intervention to ease off the work load of frontline workers such as ASHAs and AWWs. The current study would try to find out how technology can be a potent weapon in the hands of the ASHA workers in tracking their beneficiaries against the stated objectives of Ministry of Health and Family welfare to effectively combat against any future pandemics. The proposal would also include suggestions to improve the healthcare support system in Anganwadis with a view to making them more resilient and effective in future times of crisis.

4 Methodology The project undertakes primary study of ASHA workers and their beneficiaries in the Anganwadi centers in Bhubaneswar to make a realistic assessment of the efficacy of health training imparted to ASHA workers, and long-term improvement in the healthcare structure to ensure that NRHM scheme is both robust and resilient. This project intends to make an in-depth study in both urban, semi urban, and rural areas of the Kordha district. Sample of 450 (250 ASHAs and 200 beneficiaries) will be

252

M. Pandey et al.

Fig. 3 Extended SERVQUAL model

taken by using stratified random sampling, and area and cluster sampling would be done to collect samples from these areas that have predominantly poor population who are engaged as agricultural laborers, daily wages workers, and house hold helps. Questionnaire prepared to ascertain details of usefulness, difficulties, and suggestion for ASHA workers to be utilized as healthcare workforce would be evaluated based on an extended SERVQUAL model. Figure 3 depicts the conceptual framework, i.e., an extended SERQUAL model for the study. SERVQUAL is the most applied service quality model in fields like healthcare, education, banking and insurances, hotels, transports, e-government services, logistics, etc. The gap model introduced by Parasuraman et al. [18, 19] is the basis for the SERVQUAL model. SERVQUAL is more humanistic approach to evaluate the quality of service [20]. Grönroos [21] proposed two types of service qualities aspects such as technical quality (“what” service is being provided) and functional quality (“how” the service is provided). Technical quality comprises receiving of the service whereas functional quality comprises the manner of the service received. Application of different service quality models (SERVPERF, DINESERVE, etc.) can be found in every type of commercial services where it is being used to assess the expected and perceived service quality. The ASHAs being the enablers of primary healthcare the basic nature of their job come under the purview of service. In the above Fig. 3, the original SERVQUAL model [19] is added with social connect and impact of ASHA workers in community, readiness of AHSA workers as

Roles and Impact of ASHA Workers …

253

healthcare facilitators, and impact of technology as enabler for the ASHA workers along with following original ones: Reliability of ASHA workers to perform their services Assurance of knowledge and courtesy of ASHA workers tangibility is the infrastructure of AWC and quality of health infrastructures. Empathy entails personalized attention and care by ASHA workers. Responsiveness is promptness of ASHA workers. Convenience is contribution to convenience of beneficiaries by the ASHA workers social connect, and impact is connection of beneficiaries with the ASHA workers, and technology as enabler for the ASHA workers is inclusion of technology to enhancing capabilities of the ASHA workers. The validation would be of following hypothesis developed to analyze scheme comprehensively. H1: ASHA workers are providing the efficient healthcare facility at community level among children and expecting mothers, H2: AWC provides effective healthcare facilities to BPL families at community level, H3: Usefulness, difficulties, and suggestion for ASHA workers to be utilized as healthcare workforce have been accessed and achieved for contributing at the community level, H4: Technology would be effective enabler for training and inclusion of ASHA workers in healthcare workforce.

5 Conclusion The project is currently in the data collection phase. Samples are being collected from places across the city of Bhubaneswar, Khordha. The city has been divided into 3 zones (North, South-East, South-West) and 67 wards. The project aims to study the roles and impacts of ASHA workers in combating Covid-19 in urban, semi urban, and rural areas of the Bhubaneswar. The samples are being obtained geographically in such a manner that they proportionately reflect the whole research region by collecting samples from each of the 67 wards spread in three distinct zones. Two sets of questionnaires have been developed for the study; one for the ASHA workers and other for the beneficiaries. The questionnaire for ASHAs aims to study the real-time issues and challenges they faced on the field. The 1st segment is based on the demographic profile of the respondents in both. The next segment of questionnaire focuses on the changed work environment and pressure. It is to measure the vulnerability, changed roles and responsibilities and struggles they went through in combating the COVID-19 pandemic. The next segment is to measure the efficacy of the training imparted to the ASHAs. Further, the questionnaire attempts to assess current issues and challenges in delivering the primary healthcare facilities to the BPL families. The last two segments are to assess the technological intervention to ease the work of the ASHAs in for monitoring the health conditions of their beneficiaries’ babies and expecting mothers and how ASHA workers can be an important tool to combat any such unforeseen situation like COVID-19 in future. The questionnaire is framed with mixed scales: Likert, dichotomous questions (Yes/No), and opened ended questions to assess various aspects of the study. The 2nd set of questionnaire aims to assess the beneficiaries’ responses toward the services rendered by

254

M. Pandey et al.

the ASHAs. Appropriate machine learning techniques will be applied to analyze the data after the completion of the data collection. The overall objective would be to assess the role of ASHA workers as the interface between children and pregnant women affected by COVID and healthcare infrastructure available in the city of Bhubaneswar and the role of technology to bolster their capability in providing healthcare in such pandemic times. Acknowledgements This research is being financially supported by Indian Council of Social Science Research (ICSSR) that promotes research in social sciences in India.

References 1. Central Bureau of Health Intelligence (2019) National health profile 2019. Ministry of Health & Family Welfare, Government of India. Retrieved from http://www.cbhidghs.nic.in/showfile. php?lid=1147 2. PRS Legislative Research (2021) Demand for grants 2021–22 analysis health and family welfare. Retrieved from https://prsindia.org/files/budget/budget_parliament/2021/DFG%20A nalysis%202021-22_Health%20and%20Family%20Welfare.pdf 3. Tandon BN (1983) Hunter, health, and society: integrated child development services (ICDS) in India. Food Nutr Bull 5(3):1–4 4. Adhikari SK, Bredenkamp C (2009) Monitoring for nutrition results in ICDS: translating vision into action. IDS Bull 40(4):70–77 5. Government of India (2011) Integrated child development services (ICDS) scheme framework for development of the state annual programme implementation plans (APIPs). Ministry of Women and Child Development. Government of India, New Delhi 6. Government of India (2011b) Evaluation study on integrated child development schemes (ICDS). Programme Evaluation Organisation, Planning Commission. Government of India, New Delhi 7. Government of India (2011c) Mid-term appraisal eleventh five year plan 2007–2012. Planning Commission. Government of India, New Delhi 8. Sarin E, Lunsford SS, Sooden A, Rai S, Livesley N (2016) The mixed nature of incentives for community health workers: lessons from a qualitative study in two districts in India. Front Public Health 4:38 9. Singh A, Deedwania P, Vinay K, Chowdhury AR, Khanna P (2020) Is India’s health care infrastructure sufficient for handling COVID 19 pandemic. Int Arch Public Health Community Med 4:041 10. Wang J, Zhou M, Liu F (2020) Reasons for healthcare workers becoming infected with novel coronavirus disease 2019 (COVID-19) in China. J Hosp infect 105(1) 11. Almalki MJ, FitzGerald G, Clark M (2012) Quality of work life among primary health care nurses in the Jazan region, Saudi Arabia: a cross-sectional study. Hum Resour Health 10(1):1– 13 12. Gan WH, Lim JW, Koh D (2020) Preventing intra-hospital infection and transmission of coronavirus disease 2019 in health-care workers. Saf Health Work 11(2):241–243 13. Srinivas P (2020) COVID-19 preparedness checklist for rural primary health care & community settings. Retrieved from https://phfi.org/wp-content/uploads/2020/04/covid-19-preparednessguidance_checklist.pdf 14. Desai G, Pandit N, Sharma D (2012) Changing role of Anganwadi workers, a study conducted in Vadodara district. Healthline 3(1):41–44

Roles and Impact of ASHA Workers …

255

15. Ambast S (2021) Why do ASHA workers in India earn so little? [Blog]. Retrieved 29 Oct 2021 from https://www.cbgaindia.org/blog/why-do-asha-workers-in-india-earn-so-little/ 16. Saxena V, Kumari R, Kumar P, Nath B, Pal R (2015) Planning and preparation of VHND through convergence: Sharing experiences from Uttarakhand. Clin Epidemiol Glob Health 3(3):125–131 17. Deena JRJ, Sivanesan R (2019) A study on the problems involved in anganwadi centres. Think India J 22(19):294–303 18. Zeithaml VA, Parasuraman A, Berry LL (1985) Problems and strategies in services marketing. J Mark 49(2):33–46 19. Parasuraman A, Zeithaml VA, Berry L (1988) SERVQUAL: a multiple-item scale for measuring consumer perceptions of service quality. 64(1):12–40 20. Cavana RY, Corbett LM, Lo YG (2007) Developing zones of tolerance for managing passenger rail service quality. Int J Qual Reliab Manag 21. Grönroos C (1982) An applied service marketing theory. Europ J Mark

Challenges and Requirements for Integrating Renewable Energy Systems with the Grid Komal Bai, Vikas Sindhu, Ahteshamul Haque, and V. S. Bharath Kurukuru

Abstract To ensure stable operation of the grid-connected system even under high renewable energy penetration, this paper analyzes the phenomenon of fault ride through (FRT) using reactive power injection approach. The proposed approach is developed by limiting the current overshoot during the transient and time variant variations. This maintains a constant average active power and stabilizes the system to operate according to the grid standards. Further, numerical simulations and experiment are carried out by evaluating a symmetrical fault on a 16 kW three-phase grid-connected PV system to illustrate the performance of proposed method. The developed system regulates the DC link voltage, limits the maximum inverter current, and improves the voltage profile by injecting required reactive power during a fault and achieves dynamic grid support requirement. Keywords Renewable energy · Grid integration · Solar photovoltaic systems · Grid standards

1 Introduction Variable generation can be defined as generating technology whose power output cannot be reasonably controlled and varies over time. Different variable generation sources include wind, solar, ocean, and some hydro, and all these sources are renewable-based sources. In case of variable renewable energy (VRE) has two main characteristics that distinguish it from conventional form of energy generation which are as (1) its variable nature and (2) high degree of uncertainty. These two attributes may affect its planning and operations of bulk power system. K. Bai (B) · V. Sindhu Department Electronics and Communication Engineering, University Institute of Engineering & Technology (UIET MDU) M. D. University, Rohtak, India e-mail: [email protected] A. Haque · V. S. Bharath Kurukuru Department of Electrical Engineering, Advance Power Electronics Research Lab, Jamia Millia Islamia, New Delhi, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_24

257

258

K. Bai et al.

Day by day the size of global renewable market it’s increasing. More penetration of VRE has its own benefits, i.e., reduction of CO2 emission. Installed capacity of solar photovoltaic technology is 2.799 GW at the end of 2020 [1]. Solar energy has become an increasingly important energy installed capacity is 743 GW till 2020. Year 2020 was the best year in the history for the global wind industry with 93 GW of new capacity installed a 53% year on year [2]. Power generation through RES faced different issues during the initial stage. In general, the solar power can be used in these types (1) solar photovoltaic (PV) system—for generation of electricity, (2) solar thermal power plants—heat and electricity obtained, and (3) solar thermal energy system—thermal energy generated [3]. Large PV plants are generally installed on large area on free lands or agriculture land and it result in huge initial costs. Periodic maintenance required by PV modules, i.e., washing the absorptive surfaces, and it pollutes the water [3]. Up to a certain limit harmonic distortion exists when power a quality measurement is performed by connecting a PV plant. Harmonic distortion is because of power electronics (inverter) and various nonlinear loads are used in the house hold [3]. Hence, to overcome these drawbacks, the Ministry of New and Renewable Energy (MNRE), the central authority for all rules, regulations, polices and approves relating to renewable energy; international trading of electricity and national grid standards is managed by Central Electricity Regulatory Commission (CERC) and state electricity regulatory commission (SERC) deals with transmission and distribution of power [4]. As per Indian grid code, the day ahead scheduling on 15 min time black basis for all inter-regional trading is done [5]. Central Electricity Regulation Commission in sections 79 and 178 of electricity act, 2003 provides the standards guidelines and rules for stable and reliable operation of the power system, and this document is known as grid code. It is mandatory for reliable and secure power [6]. Further, sections of the paper are organized as follows: Sect. 2 deals with the grid integration stages. Section 3 identifies the requirements for compliance with the grid code, and the analysis for integration challenges is mentioned in Sect. 4. The research is concluded in Sect. 5.

2 Grid Interconnection Stages Power system planning requires sufficient resources and delivery capacity to interconnect the new supply. A well-founded and structured manner for the scope of demand requirement should be met. In case of consideration of growing share of VRE, the current method of power system planning is needed to be adjusted with the power system behavior with mainly dispatch able sources. A layout of grid-connected distributed generation system has been shown in Fig. 1. For the facility of extra variable generation into bulk power system, well-founded equipment standards are being developed. The same set of interconnection standards and procedures applicable to each generation resources which are interconnecting to the power grid for a reliable system. In stage 2 on the bases of future planning

Challenges and Requirements for Integrating Renewable Energy Systems …

259

Fig. 1 Layout of grid-connected distributed generation system

case, models are prepared by the electrical utility or market operator. At this point, network model is prepared with the new network operations model and new resource node and this reflect changes from stage 1. A polled settlement meter communication is also establishing for the gathering of real-time data for settlements by the market operator. On the completion of above steps, actual plant is connected to the grid. Now, for this to occur transmission infrastructure is need to be developed. Maximum leading and lagging reactive capability is needed for the acceptance of test result, and this also required for approval of commercial operation. There are few countries exist where grid codes are not complete or they may be under development, in that case it is very crucial point to perform a comprehensive FIS (in stage 1). With the help of this analysis, any possible adjustments can be done to accommodate the successful integration. For all generation resources, following three studies are to be included: (1) steady-state analysis, (2) short-circuit study, and (3) dynamic and transient stability analysis.

3 Requirements for Compliance with Grid Code Grid Code Reactive Power Compliance Requirements for REGS With the span of time penetration of renewable energy in transmission, networks will increase. Here, strong requirement of strict codes arises from grid operators, i.e., (1) transmission system operators (TSO) and (2) distribution system operators (DSO) [7, 8]. Steady state of network and recovery of voltage is impacted by reactive power. In case of renewable energy integration reactive power capabilities, grid codes define steady state as well as transient state. Grid code specification for FRT and voltage control is also affected by static and dynamic reactive power (Fig. 2). Wind generators and PEC interfaced generators (solar PV); grid code requirement set is discussed in Table 1.

260

K. Bai et al.

Fig. 2 Active reactive power control requirements for fault ride through and grid management

Table 1 Year wise grid code amendments [9] Year Description 2000 In March 99, utility of central transmission made this and this rule approved in Jan 2000 and it will be effective in Feb 2000 2002 Revision of Clause1.5 is “reporting of non-compliance” Replacements of “pool account” with “UI settlement system” incorporated with reactive power. Effective from march, 2002 2006 From the w.e.f. April 1, 2006, it is decided that regional load dispatch shall keep accounts of energy/power flow from the grid Grid codes should be specified, and the revision is according to CERC standards not by IEGC 2010 Concept of contractual area integration of RE to the grid is from May 3, 2010 Accountability of the states and distribution utilities 2011 Renewable energy scheduling was implemented 2012 Capacity of 10 MW or above at 33 kV of wind energy. Solar energy forecast is performed on the basis of weather condition and solar isolation

Reactive Power Requirement for Wind Generators All the grid codes discussed here for reactive power requirements for wind generators. Similar power specification is demanded in other grid codes for wind generators. According to the National Electricity Rules (NER), wind generators should have some characteristics, i.e., at the full operating range along with at full output at the point of connection (POC), reactive power control capability of ±0.93 power factor, throughout the full operating range of active power, and ±10% of nominal voltage [7]. Grid Code Specifications for PEC Integrated Energy System For different type of PEC interfaced energy system, grid operators still use strict grid code specifications. In the above discussed PEC system, wind generation excluded here. Barpanda, S. S. Saxena, S.C., Rathore, H., Dey, K. and Kumar, K. P., 2015 asserted that small

Challenges and Requirements for Integrating Renewable Energy Systems …

261

scale solar-PV system contains only such as, “AS/NZS 4777.2:2015 standard for four possible voltage ranges, namely V1 , V2 , V3 and V4 having Australian default voltage values of 207, 220, 244, and 255 V, should have 30% leading power factor capability for V1 , 30% lagging power factor capability for V4, and no regulation (i.e., 0%) is required for V2 and V3 ” [10]. On another side when we consider the case of German grid code and the French grid code which are separately point out that the any reactive power at its entire operating range should not be absorbed by low voltage solar-PV systems should [11]. Dynamic Reactive Power Requirement for FRT At the point of fault decrement in the voltage is noticed, and this voltage drop traveled across the network area until the fault is cleared [12]. Asynchronous wind generators in the fault duration need more reactive power.

4 Analysis for Integration Challenges The power system should adopt the new generation when it is connected to the electricity grid. On the other hand, are connected to the medium-voltage distribution grid or regional transmission grid should be connected to the utility-scale solar PV and wind turbines. Four phases of VRE integration have been figure out in response to the recent growth of renewable energy (IEA 2017). Different set of power system interconnection and operational challenges are described in Table 2 [13]. Following are the main components (1) a high-level control subsystem and (2) a low-level control subsystem and these are with power stage, and these are equipped with measurements included. The high-level control subsystem main parts are: control block for PQ–regulator is used for controlling the active and reactive power. Output reference current can be calculated by the help of control block which uses the transformer (grid side) power and reference power. Input of the low-level current logic is the reference current. Current block will be operational during following operation: Ramping Operation There should be a mechanism for limitation in rate of change (ROC) of the active power reference, voltage reference, frequency reference, and reactive power reference. Input signal is the reference signal for defining the rate of change. Main State Machine Controls and checks overall functionality of the plant. Operational plant states are described in main state machine states table. Low level control subsystem main parts. Control Block of Currents For the control mechanism of inner loop, there is regulator. Input current signal acts as reference signal for control block of currents and measurements is done before the transformer (converter side).

262

K. Bai et al.

Table 2 Attributes through each phase Phase 1

Phase 2

Phase 3

Phase 4

Characterization from a system perspective

VRE capacity is not relevant at the all system level

VRE capacity becomes noticeable to the system operator

In case of larger deflection in terms of supply/demand, balance flexibility is demanded

At some instant of time VRE, capacity is nearly 100%

How generator fleet is affected

Load and net Uncertainty and load not affected variability of net load remain unaffected,

Large difference is seen in operating patterns

At the clock rate, no power plants are running

How grid is affected

At the point of interconnection noticed, local grid condition

Condition of Different congestion occurs weather at transmission conditions result in power flow patterns across the grid

Recovery from disturbances and maintain stability

Main factor for challenges

Local conditions Match between Availability of System’s capacity in the grid demand and VRE flexible resources to withstand output disturbances

Fault State Machine An alarm message should be sent after checking all measurements, signals the fault state. When some fault detected, then the main circuit breaker (MCB) should immediately opened. Simulations Results For a grid-connected system, when the line voltage va , vb , and vc drops below 100 V, the corresponding line current i a , i b , and i c increases as shown in Fig. 3. These two conditions resulted due to change in direct and quadrature voltage/ current component vd and i d decreases and increases, respectively, at the time of fault. At the time instant of transient (when voltage decreases), the corresponding transient impact can be identified in the active and reactive power of the system. Further, when line voltage va , vb , and vc remains constant, and the corresponding line current i a , i b , and i c remains constant as shown in Fig. 4. These two conditions resulted due to change in direct and quadrature voltage/current component vd and id remains constant and iq increases at the time of fault. At the time instant of transient (when frequency changes), the corresponding transient impact can be identified in the (decrement) in active and (increment) in reactive power of the system. Similarly, when the line voltage va , vb , vc increases above 180 V, the corresponding line current i a , i b , i c decreases as shown in Fig. 5. These two conditions resulted due to change in direct and quadrature voltage/current component vd and id increases and decreases, respectively, at the time of fault with more transients. At the time instant of transient (when voltage increases), the corresponding transient impact can

Challenges and Requirements for Integrating Renewable Energy Systems …

263

Fig. 3 LVRT in weak grid without harmonic

Fig. 4 LFRT in weak grid without harmonic

be identified in the active and reactive power of the system. For a grid-connected system, when the line voltage va , vb , vc are constant, the corresponding line current i a , i b , and i c also remains constant as shown in Fig. 6. These two conditions resulted due to change in direct and quadrature voltage/current component vd and id also remains constant. At the time instant of transient (when frequency changes), the corresponding active and reactive power remains constant with less frequency of the system.

264

K. Bai et al.

Fig. 5 HVRT in weak grid without harmonics

Fig. 6 HFRT in weak grid without harmonics

5 Conclusion When variable energy sources more penetrate into the grid, it creates more uncertainties and technical challenges in the grid operation. In this paper, various challenges with integrating renewable energy system with the grid. Various aspects of integration of renewable energy system with the grid, i.e., grid interconnection stages (1) steady-state analysis, (2) short-circuit analysis), and (3) dynamic and transient stability, grid integration standards, requirements for compliance with grid code (A) general requirements, (B)system operation requirements and integration challenges are discussed.

Challenges and Requirements for Integrating Renewable Energy Systems …

265

References 1. World adds record New renewable energy capacity. http://www.irena.org 2. Lee J, Zhao F (2021) Global wind report 2021. Global Wind Energy Council, p 75, 2021. http:// www.gwec.net/global-figures/wind-energy-global-status/ 3. Kumar JCR, Majid MA (2020) Renewable energy for sustainable development in India: current status, future prospects, challenges, employment, and investment opportunities. Energy Sustain Soc 10:2. https://doi.org/10.1186/s13705-019-0232-1 4. Kulkarni SH, Anil TR, Gowdar RD (2016) Wind energy development in india and a methodology for evaluating performance of wind farm clusters. J Renew Energy 2016:6769405. https:// doi.org/10.1155/2016/6769405 5. Díaz-González F, Hau M, Sumper A, Gomis-Bellmunt O (2014) Participation of wind power plants in system frequency control: review of grid code requirements and control methods. Renew Sustain Energy Rev 34:551–564. https://doi.org/10.1016/j.rser.2014.03.040 6. Rohrig K et al (2019) Powering the 21st century by wind energy—Options, facts, figures. Appl Phys Rev 6(3):031303. https://doi.org/10.1063/1.5089877 7. Jain P, Wijayatunga P (2016) Grid integration of wind power. Best practices for emerging markets. ADB Sustainable Development Working Paper Series No. 43, April. Manila, Philippines 8. Impram S, Varbak Nese S, Oral B (2020) Challenges of renewable energy penetration on power system flexibility: a survey. Energy Strat Rev 31:100539. https://doi.org/10.1016/j.esr. 2020.100539 9. Barpanda SS, Saxena SC, Rathour H, Dey K Pawan Kumar KVN (2015) Renewable energy integration in Indian electricity market. In: 2015 IEEE PES Asia-Pacific power and energy engineering conference (APPEEC), 2015, pp 1–5. https://doi.org/10.1109/APPEEC.2015.738 1034 10. Shukla UK, Thampy A (2011) Analysis of competition and market power in the wholesale electricity market in India. Energy Policy 39(5):2699–2710. https://doi.org/10.1016/j.enpol. 2011.02.039 11. Sarkar MNI, Meegahapola LG, Datta M (2018) Reactive power management in renewable rich power grids: a review of grid-codes, renewable generators, support devices, control strategies and optimization algorithms. IEEE Access 6:41458–41489. https://doi.org/10.1109/ACCESS. 2018.2838563 12. Forum Netztechnik Netzbetrieb (2012) VDE-AR-N 4105:2011-08, Power generation systems connected to the low-voltage distribution network: technical minimum requirements for the connection to and parallel operation with low-voltage distribution networks 2011. In: Amaris H, Alonso M, Ortega CA (eds) Reactive power management of power networks with wind generation, vol 5. Springer, London, UK 13. ESMAP (Energy Sector Management Assistance Program) (2019) Grid integration requirements for variable renewable energy. ESMAP Technical Guide, World Bank, Washington, DC

Design of Progressive Monitoring Overhead Water Tank N. Alivelu Manga, Surya Teja Manupati, N. S. C. Viswanadh, P. Sriram, and D. V. S. G. Varun

Abstract In the era of smart homes, providing a smarter solution to water quality management is crucial, as water quality is being deteriorated over time. The proposed system uses Raspberry pi, Arduino Uno along with various sensors like turbidity sensor, pH sensor, TDS sensor to measure the quality of water. An ultrasonic sensor is used to measure the water level in the tank. A tank cleaning mechanism is proposed to clean the interior walls of the tank, thus preventing the growth of algae and bacteria. The system’s algorithm works efficiently to reduce power consumption, without hurting the functionality, so that the system can run continuously. Internet of Things (IoT) is deployed in the system for the user to control and interact with the system remotely using a mobile application. Keywords Smart homes · Raspberry pi · Arduino Uno · Ultrasonic sensor · Turbidity sensor · pH sensor · TDS sensor · Tank cleaning · Efficient algorithm · IoT

1 Introduction The value of ‘water’ is not known until the well runs dry. Humans depend on the water daily for various activities either domestic or commercial purposes. According to a survey, water scarcity is expected to affect more than 1.8 billion people around the world. Due to wastage of water, even the environment gets affected as some areas may face soil erosion, or heavy rainfalls or low rainfalls, etc. Even though three-fourths of the earth is covered with water, humans cannot use all the water as water quality is the deciding factor to tell whether a certain amount of water is used or not for certain activities. The system described here can be used to reduce wastage of water due to the running of overhead water tanks. The intent of using the ultrasonic sensor is that to detect the water level, which is mounted at top of the tank to calibrate the water N. Alivelu Manga (B) · S. T. Manupati · N. S. C. Viswanadh · P. Sriram · D. V. S. G. Varun Department of ECE, Chaitanya Bharathi Institute of Technology, Hyderabad, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Kumar et al. (eds.), ICDSMLA 2021, Lecture Notes in Electrical Engineering 947, https://doi.org/10.1007/978-981-19-5936-3_25

267

268

N. Alivelu Manga et al.

level in the overhead tank also by using Arduino. The concept of ECHOS is used to measure the level of the water, and further actions are performed. Usually, the height of the water in the water tank is measured continuously, and the water pump motor is switched on when a certain minimum threshold value is reached [1–3]. It gives a detailed picture of the quantity of water consumed. Another method is to keep two sensors, one at the top of tank, and other at bottom of tank [4, 5]. The flow sensor is attached to the inlet pipe of the water tank which measures the rate of flow of water into the tank. The readings are taken at discrete intervals of time to reduce the power consumption. Water quality is assessed based on physical appearance, chemical, and biological characteristics. By the utilization of tainted water, diseases like cholera, typhoid, rashes, fluorosis, etc. could affect health. The system uses different sensors like pH sensor, TDS sensor, and turbidity sensor to measure the chemical composition of water. If the water is contaminated, the system analyzes the chemical composition of water and sends data to the cloud and the enduser via IoT [6–8]. The user can have access to the data and controls of the system via the Internet. The quality of the water stored is measured and reported to the user along with the height of the water in the tank. In vis-à-vis to the stored water, the storage (i.e., water tanks) also plays a crucial role. In due course, the sealed tank creates pressure inside, consequently, silts and scum accrue on the walls, ceiling, and floor of the water tank. Due to this, there are high chances of Pseudomonas and Legionella bacteria forming on the interior wall of the water tank. To remove sediment scales, manual scrubbing is done using chemicals which are harmful. The cleaning process in the proposed system uses a brush and motor alignment, which cleans the water tank when operated. Every time the water tank is filled, the quality of water is tested and compared with standard values. If the water quality deteriorates, it indicates that the water tank is not pure and a notification is sent to the user via IoT and gives the user an option to clean the tank. The cleaning method for the overhead tank is mentioned using a detergent, brushes [9, 10]. The system follows an efficient algorithm to achieve the main purpose, i.e., to reduce wastage of water due to running of overhead water tanks without using much power compared to traditional methods. All the techniques used reduce the impact on the environment, as well as reduce human effort and stress. The paper is divided into 5 sections: 1. Introduction: This section gives a brief introduction to the paper and describes the outline of the system’s functionality along with past papers related to this paper. 2. Architectural Details of Proposed System: This section briefs about the physical structure of the system and shows models of the proposed system. 3. Proposed system: This section describes the sensors, actuators, and other parts used. 4. Methodology: This section describes the working of the system and the algorithm used in the system. 5. Results: This section summarizes the paper with the final conclusions reached.

Design of Progressive Monitoring Overhead Water Tank

269

2 Architectural Details of Proposed System The proposed system uses a single board computer, Raspberry Pi Zero W for processing the data, communicating with the cloud, and collecting sensor data from the ultrasonic sensor and flow sensor. The Raspberry Pi Zero W is a BCM2835 ARMv7 processor-based system-on-chip (SOC) with Wi-Fi and Bluetooth connectivity. The Raspberry Pi Zero W lacks multiple analog pins, for better and easier interfacing with extended processing support, we use an Arduino Uno to get the sensor readings from the pH sensor and turbidity sensor. This Arduino Uno is programmed in Arduino IDE to collect the data and send it to Raspberry Pi through serial communication over USB, which will be sent to the Raspberry PI Zero W. The Raspberry Pi Zero W is programmed to collect and process all the sensor data using Python language. All the obtained sensor readings are then passed over various functions to calculate the overall purity of the water. The Raspberry Pi Zero W is linked with the firebase real-time database using the application program interface (API) key provided by the cloud. Once the cloud is linked to the Raspberry Pi, then the water level and the purity of water are sent to the firebase. Unlike other approaches, the proposed system does not display any chemical compositions or graphs. All the data processing is done within the Raspberry Pi, and only the purity information and water level are updated to the cloud. The system is designed based on the end-user’s ease of using the system, and the final user has no use with the chemical composition. This is the specific reason for not displaying the chemical composition but to display only pureness and level of water. Figure 1 provides the basic working of the proposed system.

Fig. 1 Block diagram of proposed system

270

N. Alivelu Manga et al.

The data that is sent to the firebase is then projected over the android application that is built over the popular application making platform MIT APP inventor which is designed not only to get the data from the firebase and display in the android application but also to instruct the main processor to perform actions on the requirement. When the water quality is in the specified portable range, then on the android application, a purity notifying statement is displayed along with a slider indicating the water level. If the water quality is deviating from the specified portable range, then on the android application, an alert notification on the purity is displayed along with a slider indicating the water level and a tank cleaning option. When the user chooses to clean the tank, this instruction is sent back to the Raspberry Pi Zero W over the firebase and a cleaning action is initiated through a motor setup to drive away all the scaling sediments, algae which might be the reason for the impure water. If the water is impure even after the cleaning process is done, it indicates that water itself is not clean and the user can use filtering or any other chemical methods to purify water. The system is designed to have an automatic water level maintaining system. This is achieved by using an ultrasonic sensor when the ultrasonic sensor reading is above a permissive value. The Raspberry Pi switches off the water pump by using a relay module, which is an electrical switch that is operated through the signal from the Raspberry Pi sent to control the output AC water pump. The physical design plays an important role in terms of efficient measurement and durability of the system. The structure of the system as shown in Fig. 2 has an ultrasonic sensor at one end separately to measure the water level. The red-colored box consists of the water quality sensors required to test the purity of water. The blue-colored box is the motor for cleaning the tank. The vertical bars are brushes to clean the tank. All the other sensor setup is safely stored in a box like enclosure which is placed at the bottom outlet of the tank into which little amount of water is pumped using an electronic regulating valve to test the quality of the water before filling the water tank. Once the quality test is passed, the inlet water from pump will be allowed to fill the tank. The system is unique with its cleaning feature, which is achieved by using a motor and brush setup as seen in Fig. 3. Once the user chooses to clean the tank, the processing unit controls the cleaning motor to clean the walls of the tank to remove the debris and scum algae present on the walls. Fig. 2 3D model of proposed system

Design of Progressive Monitoring Overhead Water Tank

271

Fig. 3 3D model of cleaning mechanism

3 Proposed System In the process of designing the sensing part of the system, various sensors like ultrasonic sensor, pH sensor, turbidity sensor, and flow sensor were used. To process the sensor data obtained from sensors and to transmit it to the cloud, Raspberry Pi Zero W and Arduino were used with the help of firebase platform.

3.1 Raspberry Pi Zero W Raspberry Pi Zero W is a 1GHz, single core CPU with 512 MB RAM with inbuilt WiFi module and Bluetooth module. Flow sensor and ultrasonic sensor are connected to Raspberry Pi Zero W, as the data connected from them are very critical in the working of the whole system, when compared with other sensors.

3.2 Arduino Uno Arduino Uno is a micro-processor used to record the analog inputs from analog sensors like pH sensor, turbidity sensor, and TDS sensor. The Raspberry Pi Zero W is connected to Arduino UNO via USB interfacing. The analog sensors connected to Uno are taken for a certain time to avoid sensory errors. ADC module can be used as a substitute for Uno.

3.3 Ultrasonic Sensor The ultrasonic sensor consists of a TRIG pin, ECHO pin, and a control circuit. The sensor generates a frequency wave of 40 kHz, when there is an obstacle in between, then the wave bounces back. Using this principle, it calculates the distance of that

272

N. Alivelu Manga et al.

obstacle by knowing the time taken to send and receive signal. Ultrasonic sensor is used for detecting the water level in water tank.

3.4 Flow Sensor Flow sensor is used to observe the rate of flow of water into the tank. It constitutes of water rotor and hall effect sensor. When there is a flow of water through the valve, it results in the rotation of rotor. The output is a pulse signal which is recorded with the help of hall sensor.

3.5 Turbidity Sensor Turbidity sensor measures the amount of turbidity/insoluble particles present in the water tank. It sends a light beam into the water and usually a light detector is placed at 90° to the light source. The reflected light indicates the presence of insoluble particles. If more light is detected, it indicates that more particles are present in water. Turbidity sensor is an important sensor in this system, as the color of water gives us a broad idea of how pure the water stored might be. The equation for sensor reading into NTU is −1120.4 ∗ readings2 + readings ∗ 5742.3 − 4353.8, where readings are the sensor value and the output is in NTU.

3.6 TDS Sensor TDS sensor is used to find the total dissolved solids sin a solution and is measured in terms of PPM. TDS measures the conductivity of a solution. If the total dissolved solids increases, so does the conductivity, so does the ppm. TDS sensor tells us if the dissolved solids are more than certain threshold values.

3.7 PH Sensor pH sensor is used to determine the pH of a solution. The number of concentrated H+ ions determines the acidic or alkaline nature of the solution. A pH of 7 indicates neutral solution. pH sensor can tell us if there are any chemicals in the water stored which may change the nature of water.

Design of Progressive Monitoring Overhead Water Tank

273

3.8 Motor and Motor Driver A 12 V DC motor is used for cleaning purpose, which is mounted at the top of the water tank. The motor is connected to Arduino via a L298N motor driver. Raspberry Pi Zero W gives the main instruction to the Arduino, so that the cleaning process can be activated.

3.9 Setup The system provides the user with the water level of the tank, and the water quality on the Android application. This is achieved by using firebase database which is a real-time cloud hosted database, and this is linked with Raspberry Pi Zero W and Android application which are synced continuously. The firebase database provides the user with an API which is linked with the Android application designed using MIT app inventor. The Android displays the water level and quality of water as shown in Fig. 4. If the water is impure, the user will be notified about the quality and the user will be given a choice to clean the overhead water tank, as shown in Fig. 5. The water quality sensors are packed in a closed container, where only the sensing part of the sensor will be in contact with water. This ensures that the water quality sensors can perform their function properly without comprise in the functionality and life of sensor. Fig. 4 Experimental observation-1

274

N. Alivelu Manga et al.

Fig. 5 Experimental observation-2

4 Methodology The system is designed in a way that components in the process have less chance of getting damaged and consumes less energy, by taking sensor readings at discrete intervals of time without compromising the main functions of the system. Initially, the system checks whether there is any water left in the tank with the help of an ultrasonic sensor. If some water is present, then the water quality sensors analyze the purity of water and compare it with threshold values. The water quality sensors consist of a turbidity sensor, pH sensor, and TDS sensor. The water is initially checked with a turbidity sensor, as it observes the turbidity of water which is a key factor in the quality of water. If the water has turbidity within permissible range (= 4

IESD

4

The classification of SD is done into bad SDs and good SDs as depicted in the Fig. 2. Figure 2 shows the SDs labelled as good SD and bad SD. The good SDs are the SDs which have their remaining energy higher than or equal to four times the initial value. Goodness ratio value exhibits better performance. The goodness ratio (GR) is defined using Eq. 3. GR =

Good SD Count Bad SD Count

(3)

1.1 Motivation The SSD selection of LEACH will take only one criteria which is not enough and also suffers from massive energy consumption, delay, and bad goodness ratio [7]. For the case of E-LEACH, only variation is selection of SSD in which SSD is elected based on residual energy and amplification loss [8]. The disadvantage of both the methods can be overcome based on fuzzy terminology. The chance maximization can be done on the basis of fuzzy-based system which takes multiple criteria namely distance with respect reference point, remaining power, and moment SD computation. The various downgrades of existing system are. • SSD is elected based on randomized probability which can have lower energy. • Lot of communication analogies for link formation of SD with BS and vice versa. • Goodness ratio of the existing LEACH and E-LEACH is bad.

2 Literature Survey The method which uses the BSs along with SDs and SSDs has a longer path with multiple reasons, selection of SSD is based on single criteria which is bad in nature, and the packets send across multiple SDN do not perform good load balancing

530

V. S. Rekha and Siddaraju

[9]. The SSDs are responsible for load balancing, data cracking in the SDN. The collection or tracking repeats for every T duration, and if the total energy of all SDs falls below threshold value, a distress signal is communicated to base station [10]. The packet maunder is done based on energy levels across SDN. The method used here is based on energy and distance combination for selection of SSD [11]. The metric of various parameters, namely good SD count, bad SD count, and goodness factor, is used for studying the performance of the algorithms [12]. The buffer levels of SDs are measured and then during the route formation, the SD which has lower number of packets in buffer is chosen as the next forward SD. The improvement in percentage is done based on delivery ratio but will have lower delay [13]. The route formation is done from SSD to DSD which is done with maximization of goodness ratio and reduces the count of bad SDs [14]. The hop reduction is done by balancing the distance computation during same energy level SDs [15]. The SDs have limited battery power and also need to handle good amount of packet loads which will lead to more power consumption. The GSDN will sense data and send important data to control centre and amount of data transfer which is redundant will be reduced using fusion [16]. Multi-parameters are used in the selection SSD, namely energy parameter and distance parameter. The packets transfer follows a three-layer approach. In the first layer, transfer of packets will happen from SDs to SSDs. The second layer will transfer data to base station. The third layer will transfer data from base station to control centre with help of destination SDs and destination SSD. The amount of goodness ratio is increased [17]. The goodness factor can be improved by creating dynamic sets of SDs. The selection of SSD is done with the help of fuzzy values. The multi-level fuzzy makes use of distance and remaining energy for computation of criteria levels for selection of SSD [18]. The SDGN has the sense of independent SDs. Each SDGN will elect a SSD responsible for communication between independent sets. The SDGN suffers from power factor, memory slice size. The handoff-based system is responsible for communication between two SDGN when the two SDs are far away and will also improve the goodness ratio along with delivery rate [19]. The power generation can be optimized with the help of farms having wind turbines. The transmission of power to different areas and monitoring of region of interest for a long duration should be done in SDGN. There are three steps with first layer having random location formation for SDGN, the second is measuring of remaining battery level, and the last step is to send the data to destination control centre [20]. The balance of energy levels along with changing the location of SDN is an important matter. A model must be created based on energy level computation values in which a way that energy consumption value is minimized along with count of SDs during route formation. The method is responsible for improvement of goodness ratio [21]. The performance of SDGN is limited due to energy factor. The combination of data transmission rate along with residual energy is responsible for selection of SSD. The goodness ratio improvement is possible due to combination of data, flexibility of values along with reliability. The balance of energy along with stability helps in better maintained of volumes of data [22]. The performance of SDGN is limited due to energy factor. The combination of data transmission rate along with residual energy

Goodness Ratio and Throughput Improvement Using Multi-criteria …

531

is responsible for selection of SSD [23]. The limited amount of resources along with battery life and count of SDs is very important. The regions are divided into multiple subregions to form SDGN. The threshold is computed based on energy levels of SDs. The SSD will be selected based on chance factor which depends on energy. The chance factor is sorted so that first value corresponds to primary SSD and second value for secondary SSD [24]. Various applications, namely agriculture applications, forest-based system along with other industry application, are monitored with the help of SSD. The data is then send towards destination sink. Goodness ratio is increased by measuring energy efficiency for obtaining primary and secondary SSD. The number of bad SDs can be reduced with the help of this method [25].

3 Proposed Multi-criteria LEACH The value of SSD is found by making use of fuzzy technique with the help of three input variables, namely distance measure, energy measure, and mobility [24]. The SSD and DSD will form link if both of them are within same SDGN. If the SSD and DSD are in different SDGN, then communication between two independent SDGNs is done with the help of SSDs. The data collection is done with the help of moving main data collector (MDC) to collect the data from the SDs and then data once collected will be processed to get understandable results in the control centre. This section first describes the formation of SDGN followed by selection of SSD and then route formation between ISD to DSD.

3.1 Single SDGN Network The formation of SDGN is done by measuring parameters which form a set of three labels with label1 being unique sensing device (USD) number, first dimension of sensing device, and second dimension of sensing device [26]. The SD will have their own unique dimensional pair which is unique to other SDs. Each of the SDs are bounded with the end point limits of first dimension (FDstart, FDend) and second dimension (SDstart, SDend). FDstart is the first dimension starting limit for SDGN, and FDend is the first dimension end limit for SDGN. SDstart is the second dimension starting limit for SDGN, and SDend is the end limit for SDGN [23]. The output will be a matrix with Nsds and three parameters column. Nsds are the total number of SDs in the SDGN. ⎡

1 ⎢2 ⎢ SDGN = ⎢ . ⎣ .. NSDs

(fD1 , SD1 ) ( f D2 , SD2 ) (fDn , SDn )

⎤ ⎥ ⎥ ⎥ ⎦

(4)

532

V. S. Rekha and Siddaraju

Algorithm I Single SDGN Formation • Input: NSDs , FDstart , FDend , SDstart , SDend • Output: The formation of matrix with position and unique id information SDGNM • Description: it = 1 it : 1 → NSDs • Find the first dimension of SD in a random fashion between FDstart and FDend FDpi = FDpv FDstart ≤ FDpv ≤ FDstop and FDpv /= FDph . any FDspv which satisfies • Find the second dimension of SD in a random fashion between SDstart and SDend SDpi = SDpv SDpvv /= SDpvH . any SDpv which satisfies and SDstart ≤ SDpv ≤ SDend FDpH − first dimension for position of SD in history . SDpH − second dimension for position of SD in history • Form a value with dimensionality and id uniqueness (it, FDpn , SDpn ) • Save the itth data for SDGNM where,

Node

Position

it

(FDpn , SDpn )

it = it + 1 Note—Each of the SP has its own unique representation value in the SDGNM matrix.

3.2 Multiple SDGN Network Multiple SDGN formation will form independent SDGNs. The combination of SDs will form a SDGN. The outcome of this process is SDGN with each SDs having their own representation number. Multiple SDGN will be formed using the method described in algorithm II.

Goodness Ratio and Throughput Improvement Using Multi-criteria …

533

Algorithm II Multiple SDGN Formation • Input: NSDGN − number of SDGN Nsdlist − List of number of SDs required in each SDGN FDcordinates − set of end points of first dimension for SDGN SDcordinates − set of end points of second dimension for SDGN • Output: The SDGN having SDGNM along with unique SDGN id • Description: First value for SDGN is given a number 1 vsdgn = 1. Check for the condition vsdgn