Modeling and Optimization of Signals Using Machine Learning Techniques E-Book

• Herausgeber: John Wiley & Sons
• Kategorie: Wissenschaft und neue Technologien
• Sprache: Englisch

Seitenzahl: 536

Veröffentlichungsjahr: 2024

Beschreibung

Explore the power of machine learning to revolutionize signal processing and optimization with cutting-edge techniques and practical insights in this outstanding new volume from Scrivener Publishing.

Modeling and Optimization of Signals using Machine Learning Techniques is designed for researchers from academia, industries, and R&D organizations worldwide who are passionate about advancing machine learning methods, signal processing theory, data mining, artificial intelligence, and optimization. This book addresses the role of machine learning in transforming vast signal databases from sensor networks, internet services, and communication systems into actionable decision systems. It explores the development of computational solutions and novel models to handle complex real-world signals such as speech, music, biomedical data, and multimedia.

Through comprehensive coverage of cutting-edge techniques, this book equips readers with the tools to automate signal processing and analysis, ultimately enhancing the retrieval of valuable information from extensive data storage systems. By providing both theoretical insights and practical guidance, the book serves as a comprehensive resource for researchers, engineers, and practitioners aiming to harness the power of machine learning in signal processing.

Whether for the veteran engineer, scientist in the lab, student, or faculty, this groundbreaking new volume is a valuable resource for researchers and other industry professionals interested in the intersection of technology and agriculture.

Leseprobe

Table of Contents

Cover

Table of Contents

Series Page

Title Page

Copyright Page

Preface

1 Land Use and Land Cover Mapping of Remotely Sensed Data Using Fuzzy Set Theory-Related Algorithm

1.1 Introduction

1.2 Image Classification

1.3 Unsupervised Classification

1.4 Supervised Classification

1.5 Overview of Fuzzy Sets

1.6 Methodology

1.7 Results and Discussion

1.8 Conclusion

References

2 Role of AI in Mortality Prediction in Intensive Care Unit Patients

2.1 Introduction

2.2 Background

2.3 Objectives

2.4 Machine Learning and Mortality Prediction

2.5 Discussions

2.6 Conclusion

2.7 Future Work

2.8 Acknowledgments

2.9 Funding

2.10 Competing Interest

References

3 A Survey on Malware Detection Using Machine Learning

3.1 Background

3.2 Introduction

3.3 Literature Survey

3.4 Discussion

3.5 Conclusion

References

4 EEG Data Analysis for IQ Test Using Machine Learning Approaches: A Survey

Introduction

4.1 Related Work

4.2 Equations

4.3 Classification

4.4 Data Set

4.5 Information Obtained by EEG Signals

4.6 Discussion

4.7 Conclusion

References

5 Machine Learning Methods in Radio Frequency and Microwave Domain

5.1 Introduction

5.2 Background on Machine Learning

5.3 ML in RF Circuit Modeling and Synthesis

5.4 Conclusion

References

6 A Survey: Emotion Detection Using Facial Reorganization Using Convolutional Neural Network (CNN) and Viola–Jones Algorithm

6.1 Introduction

6.2 Review of Literature

6.3 Report on Present Investigation

6.4 Algorithms

6.5 Viola–Jones Algorithm

6.6 Diagram

6.7 Results and Discussion

6.8 Limitations and Future Scope

6.9 Summary and Conclusion

References

7 Power Quality Events Classification Using Digital Signal Processing and Machine Learning Techniques

7.1 Introduction

7.2 Methodology for the Identification of PQ Events

7.3 Power Quality Problems Arising in the Modern Power System

7.4 Digital Signal Processing-Based Feature Extraction of PQ Events

7.5 Feature Selection and Optimization

7.6 Machine Learning-Based Classification of PQ Disturbances

7.7 Summary and Conclusion

References

8 Hybridization of Artificial Neural Network with Spotted Hyena Optimization (SHO) Algorithm for Heart Disease Detection

8.1 Introduction

8.2 Literature Survey

8.3 Proposed Methodology

8.4 Artificial Neural Network

8.5 Software Implementation Requirements

8.6 Conclusion

References

9 The Role of Artificial Intelligence, Machine Learning, and Deep Learning to Combat the Socio-Economic Impact of the Global COVID-19 Pandemic

9.1 Introduction

9.2 Discussions on the Coronavirus

9.3 Bad Impacts of the Coronavirus

9.4 Benefits Due to the Impact of COVID-19

9.5 Role of Technology to Combat the Global Pandemic COVID-19

9.6 The Role of Artificial Intelligence, Machine Learning, and Deep Learning in COVID-19

9.7 Related Studies

9.8 Conclusion

References

10 A Review on Smart Bin Management Systems

10.1 Introduction

10.2 Related Work

10.3 Challenges, Solution, and Issues

10.4 Advantages

Conclusion

References

11 Unlocking Machine Learning: 10 Innovative Avenues to Grasp Complex Concepts

11.1 Regression

11.2 Classification

11.3 Clustering

11.4 Clustering (k-means)

11.5 Reduction of Dimensionality

11.6 The Ensemble Method

11.7 Transfer of Learning

11.8 Learning Through Reinforcement

11.9 Processing of Natural Languages

11.10 Word Embeddings

11.11 Conclusion

References

12 Recognition Attendance System Ensuring COVID-19 Security

12.1 Introduction

12.2 Literature Survey

12.3 Software Requirements

12.4 Hardware Requirements

12.5 Methodology

12.6 Building the Database

12.7 Pi Camera for Extracting Face Features

12.8 Real-Time Testing on Raspberry Pi

12.9 Contactless Body Temperature Monitoring

12.10 Raspberry-Pi Setting Up an SMTP Email

12.11 Uploading to the Database

12.12 Updating the Website

12.13 Report Generation

12.14 Result

12.15 Discussion

12.16 Conclusion

References

13 Real-Time Industrial Noise Cancellation for the Extraction of Human Voice

13.1 Introduction

13.2 Literature Survey

13.3 Methodology

13.4 Experimental Results

13.5 Conclusion

References

14 Machine Learning-Based Water Monitoring System Using IoT

14.1 Introduction

14.2 Smart Water Monitoring System

14.3 Sensors and Hardware

14.4 PowerBI Reports

14.5 Conclusion

References

15 Design and Modelling of an Automated Driving Inspector Powered by Arduino and Raspberry Pi

15.1 Introduction

15.2 Literature Survey

15.3 Results

15.4 Conclusion

References

16 Kalman Filter-Based Seizure Prediction Using Concatenated Serial-Parallel Block Technique

16.1 Introduction

16.2 Prior Work

16.3 Proposed Method

16.4 Serial-Parallel Block Concatenation Approach

16.5 Algorithm

16.6 Kalman Filter

16.7 Results and Discussion

16.8 Conclusion

References

17 Current Advancements in Steganography: A Review

17.1 Introduction

17.2 Evaluation Parameters

17.3 Types of Steganography

17.4 Traditional Steganographic Techniques

17.5 CNN-Based Steganographic Techniques

17.6 GAN-Based Steganographic Techniques

17.7 Steganalysis

17.8 Applications

17.9 Dataset Used for Steganography

17.10 Conclusion

References

18 Human Emotion Recognition Intelligence System Using Machine Learning

18.1 Introduction

18.2 Literature Review

18.3 Problem Statement

18.4 Methodology

18.5 Results

18.6 Applications

18.7 Conclusion

18.8 Future Work

References

19 Computing in Cognitive Science Using Ensemble Learning

19.1 Introduction

19.2 Recognition of Human Activities

19.3 Methodology

19.4 Applying the Boosting-Based Ensemble Learning

19.5 Human Activity Features Computability

19.6 Conclusion

References

About the Editors

Index

Also of Interest

End User License Agreement

List of Tables

Chapter 1

Table 1.1 Confusion matrix for conventional FCM.

Chapter 2

Table 2.1 Inclusion criteria.

Table 2.2 Exclusion criteria.

Table 2.3 Most important features.

Chapter 4

Table 4.1 Comparison between IQ levels with different methods.

Chapter 5

Table 5.1 Summary of machine learning techniques used in different papers.

Chapter 7

Table 7.1 Performance parameters of the ML classifiers.

Table 7.2 A comparison of performance parameters of the different ML classifie...

Chapter 9

Table 9.1 Comparative month-wise coronavirus statistics for the whole world [1...

Table 9.2 Coronavirus vaccines for emergency use in different countries.

Table 9.3 Result of the psychological impact.

Table 9.4 Result of the mental health impact.

Table 9.5 Internet traffic growth in various places.

Table 9.6 Percentage increase in various usages of the Internet.

Table 9.7 A list of dashboards available in India.

Table 9.8 A list of worldwide dashboards.

Table 9.9 Role of AI in the COVID-19 pandemic.

Chapter 11

Table 11.1 Costs of used cars: Regression example data.

Table 11.2 A classification problem (example data).

Table 11.3 Example of k-means algorithm data.

Chapter 16

Table 16.1 Data sets and recognition rate.

Chapter 17

Table 17.1 Summary of traditional steganographic techniques.

Table 17.2 Summary of CNN-based steganographic techniques.

Table 17.3 Summary of the GAN-based steganographic methods.

Table 17.4 Details of the datasets used in literature for steganography.

Chapter 19

Table 19.1 Movements captured in a predetermined period of time.

Table 19.2 Protocols.

List of Illustrations

Chapter 1

Figure 1.1 Illustration of an unsupervised classification [15].

Figure 1.2 Illustration of a supervised classification [19].

Figure 1.3 Flowchart of the modified FCM.

Figure 1.4 Plot of the pixel mean values v/s band numbers.

Figure 1.5 Gaussian membership function.

Figure 1.6 Membership functions defined for band 6.

Figure 1.7 Membership functions defined for band 5.

Figure 1.8 Membership functions defined for band 7.

Figure 1.9 Rule viewer showing the connection between input and output.

Figure 1.10 Study area used for the research.

Figure 1.11 FCM classified map of the study area for band combination 574.

Figure 1.12 Modified FCM classified map for the study area.

Figure 1.13 Classification output of the FIS method.

Figure 1.14 Classification of the K-means method.

Chapter 2

Figure 2.1 Different ML algorithms.

Figure 2.2 Model selection techniques.

Chapter 3

Figure 3.1 A simple example of SVM.

Figure 3.2 A simple example of neural network technology.

Figure 3.3 A simple example of a decision tree.

Figure 3.4 A simple example of Android behavior-based malware detection.

Figure 3.5 A simple example of naive Bayes.

Figure 3.6 A simple example of K-nearest neighbor.

Chapter 4

Figure 4.1 Learning style classifications.

Figure 4.2 Connection between tweak of theta power and intellectual abilities.

Figure 4.3 (a) Flowchart of IQ level order utilizing the EEG-based insight cla...

Figure 4.4(a) The entire producer for IQ estimation [8].

Figure 4.4(b) One example to show before and after preparing [8].

Figure 4.5 Flowchart of EEG test [9].

Chapter 5

Figure 5.1 ML design flowchart.

Figure 5.2 (a) Linear regression.

Figure 5.2 (b) Polynomial regression.

Figure 5.2 (c) K-means.

Figure 5.3 The proposed framework [20].

Figure 5.4 Volterra-ANN device model [21].

Figure 5.5 The framework of the EMLDE method [22].

Figure 5.6 ANN array methodology as a block diagram [23].

Figure 5.7 Flowchart of the suggested improved optimizer [24].

Chapter 6

Figure 6.1 Application flow diagram.

Figure 6.2 Confusion matrix.

Figure 6.3 Image input feature learning and classification.

Figure 6.4 Working diagram for the system.

Figure 6.5 The application’s use case diagram.

Figure 6.6 Log in page.

Figure 6.7 Emotion detection of the user.

Figure 6.8 Music suggestions for the user.

Figure 6.9 User expression detection.

Figure 6.10 Music is played based on the emotion.

Chapter 7

Figure 7.1 Methodology of the proposed work.

Figure 7.2 RMS waveform of the voltage sag event.

Figure 7.3 RMS waveform for the voltage swell event.

Figure 7.4 RMS waveform of the overvoltage event.

Figure 7.5 RMS waveform of the undervoltage event.

Figure 7.6 Impulsive transient overvoltage due to lightning.

Figure 7.7 Low-frequency oscillatory transient voltage.

Figure 7.8 Harmonics and the resultant waveform.

Figure 7.9 Multiresolution decomposition of the signal.

Figure 7.10 Plot of DWT coefficients of a pure sine wave.

Figure 7.11 Plot of DWT coefficients of a harmonics waveform.

Figure 7.12 Wavelet energy distribution level of a voltage sag.

Figure 7.13 Wavelet energy distribution level of voltage harmonics.

Figure 7.14 GA-based feature optimization and selection for SVM classifier.

Figure 7.15 The SVM workflow (the hyperplane maximizes the separation of two d...

Figure 7.16 Artificial neural network with three hidden layers.

Figure 7.17 Back propagation neural network.

Figure 7.18 Process flow of back-propagation training algorithm.

Figure 7.19 Probabilistic neural network.

Figure 7.20 A comparison of the classifier’s precision with a different number...

Figure 7.21 A comparison of the accuracy of the different classifiers with var...

Figure 7.22 Specificity comparison of different classifiers.

Figure 7.23 Sensitivity comparison of different classifiers.

Chapter 8

Figure 8.1 Block diagram of the proposed ANN-SHO approach.

Figure 8.2 Categories of machine learning.

Figure 8.3 Artificial neuron.

Figure 8.4 Different layers of an ANN.

Figure 8.5 ANN architecture.

Figure 8.6 Structure of ANN modeling with multiple hidden layers.

Figure 8.7 Prey searching.

Figure 8.8 Prey attacking.

Figure 8.9 Flow chart of SHO.

Figure 8.10 Overall GUI modules.

Figure 8.11 Training dataset with target.

Figure 8.12 Normalized training dataset with target.

Figure 8.13 Test dataset with target.

Figure 8.14 Normalized test dataset with target.

Figure 8.15 Hidden and output layer trained weights.

Figure 8.16 Error plot for the ANN-SHO.

Figure 8.17 Convergence plot for the proposed ANN-SHO.

Figure 8.18 Hidden layer random weights and hidden layer trained weights.

Figure 8.19 Convergence comparison of the proposed method.

Figure 8.20 Prediction accuracy.

Chapter 9

Figure 9.1 Different variants of the coronavirus.

Figure 9.2 Symptoms-wise rate of infection.

Figure 9.3 The comparison of mortality rates in the first and second waves. (a...

Figure 9.4 Classification of the overall impact due to COVID-19.

Figure 9.5 Returns of various sectors on the world stock market.

Figure 9.6 The reduced concentration of PM2.5 in different cities of the world...

Figure 9.7 The reduced concentration of PM10 in different cities of the world.

Figure 9.8 The reduced concentration of different cities of the world.

Figure 9.9 The reduced concentration of

O

3

and

NO

2

in different cities of the ...

Figure 9.10 Technological roles during COVID-19.

Figure 9.11 Various methods of AI.

Figure 9.12 Overview of machine learning.

Chapter 10

Figure 10.1 Different kinds of waste.

Figure 10.2 Smart waste management system.

Figure 10.3 Full message intimation.

Chapter 11

Figure 11.1 Representation of data in graphical form.

Figure 11.2 Data scatter plot.

Figure 11.3 Example of a random forest by way of majority voting.

Figure 11.4 A neural network (NN) with a hidden layer (source: Wikipedia).

Figure 11.5 Shallow NN.

Figure 11.6 DNN model with three hidden layers.

Figure 11.7 A neuron (in Wikipedia, The Free Encyclopedia).

Figure 11.8 Flow of signals in a biological neuron.

Chapter 12

Figure 12.1 Visual Studio Code as the code editor software.

Figure 12.2 MLX90614 IR temperature sensor connected to the Raspberry Pi.

Figure 12.3 Pi camera with Raspberry Pi.

Figure 12.4 Raspberry Pi 4 model B.

Figure 12.5 OLED display (where a message and a temperature are shown).

Figure 12.6 Configuration of facial recognition and storing data in the databa...

Figure 12.7 Dataset of images with folders corresponding to each person.

Figure 12.8 Image in the dataset/Ramya for training.

Figure 12.9 Image in the dataset/Saksha for training.

Figure 12.10 Capturing a face in real time.

Figure 12.11 Acquiring the MLX90614 package/library.

Figure 12.12 Connecting the MLX90614 sensor to the Raspberry Pi.

Figure 12.13 Screenshot of a sample datasheet.

Figure 12.14 Reading the temperature values from the sensor.

Figure 12.15 Receiving an email from Raspberry Pi.

Figure 12.16 Connecting the Pi camera to the Raspberry Pi.

Figure 12.17 Uploading data to the database.

Figure 12.18 Date-wise attendance details updated on the website.

Figure 12.19 Visual of the website.

Figure 12.20 Sample report.

Figure 12.21 Block diagram of a facial recognition-based attendance system.

Figure 12.22 Rendered digital design of Covid Secure Recat.

Figure 12.23 Output prediction of Ramya based on trained data.

Figure 12.24 Output prediction of Saksha based on trained data.

Figure 12.25 Output prediction of no recognition.

Figure 12.26 Displaying ‘Hello’ message on the OLED.

Figure 12.27 Email received from the Raspberry Pi for a person with high tempe...

Figure 12.28 Final circuitry consisting of the Raspberry Pi, Pi camera, and ML...

Figure 12.29 Final product displaying the name and temperature of the person.

Figure 12.30 Admin page on the website with access to details.

Figure 12.31 Date-wise attendance details.

Figure 12.32 User page on the website with access to details.

Chapter 13

Figure 13.1 The basic flow of the processing system.

Figure 13.2 Design of the system at the machine end.

Figure 13.3 Design of the system at the user end.

Figure 13.4 Time domain analysis of the signals.

Figure 13.5 Frequency domain analysis of the signals.

Chapter 14

Figure 14.1 Block diagram of a smart water monitoring system.

Figure 14.2 Azure IoT hub.

Figure 14.3 Azure SQL database.

Figure 14.4 PowerBI report generation.

Figure 14.5 Water from the resources is monitored, distributed, and used for v...

Chapter 15

Figure 15.1 Architecture diagram of the sensors installed in the car.

Figure 15.2 A block diagram depicting the process flow.

Figure 15.3 Images collected to form the dataset.

Figure 15.4 The device to be fitted inside the car.

Figure 15.5 Fixed acceleration mount with ultrasonic sensor.

Figure 15.6 Gear mount placed around the gear lever.

Figure 15.7 The result of the sensors attached to the accelerator, brake, and ...

Figure 15.8 The result of the sensors attached to the gear and the values from...

Figure 15.9 Steering stability map of the driver during the test.

Figure 15.10 Comparisons of driving skills with the dataset using MobileNet.

Chapter 16

Figure 16.1 Standard adaptive filter.

Figure 16.2 Adaptive filter block processing.

Figure 16.3 Serial-parallel block concatenation approach.

Figure 16.4 ROC result.

Chapter 17

Figure 17.1 Common terms used in steganography.

Figure 17.2 Types of steganography based on the host.

Figure 17.3 Types of steganography based on the domain.

Figure 17.4 Images used for traditional steganography: (a) Baboon, (b) Lena, (...

Chapter 18

Figure 18.1 Schematic representation of the proposed solution.

Figure 18.2 Schema of the MFCC extraction.

Figure 18.3 Evaluation of results.

Figure 18.4 The confusion matrix.

Chapter 19

Figure 19.1 Weak classifiers.

Figure 19.2 Human activity recognition.

Guide

Cover Page

Table of Contents

Series Page

Title Page

Copyright Page

Preface

Begin Reading

About the Editors

Index

Also of Interest

WILEY END USER LICENSE AGREEMENT

Pages

ii

iii

iv

xix

xx

xxi

xxii

xxiii

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

271

272

273

274

275

276

277

278

279

280

281

283

284

285

286

287

288

289

290

291

292

293

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

313

314

315

316

317

318

319

320

321

322

323

324

325

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

383

384

385

386

387

389

390

391

392

393

Scrivener Publishing100 Cummings Center, Suite 541JBeverly, MA 01915-6106

Publishers at ScrivenerMartin Scrivener ([email protected])Phillip Carmical ([email protected])

Modeling and Optimization of Signals Using Machine Learning Techniques

Edited by

Chandra Singh

Rathishchandra R. Gatti

K.V.S.S.S.S. Sairam

Manjunatha Badiger

Naveen Kumar S.

and

Varun Saxena

This edition first published 2024 by John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA and Scrivener Publishing LLC, 100 Cummings Center, Suite 541J, Beverly, MA 01915, USA© 2024 Scrivener Publishing LLCFor more information about Scrivener publications please visit www.scrivenerpublishing.com.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

Wiley Global Headquarters111 River Street, Hoboken, NJ 07030, USA

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Limit of Liability/Disclaimer of WarrantyWhile the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchant-ability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials, or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read.

Library of Congress Cataloging-in-Publication Data

ISBN 978-1-119-84768-7

Front cover images supplied by Pixabay.comCover design by Russell Richardson

Preface

Organization of the Book

The chapters are organized as concepts of modeling and optimization of signal processing in machine learning.

Signal processing captures, interprets, describes, and manipulates physical phenomena. Mathematics, statistics, probability, and stochastic processes are among the signal processing languages we use to interpret real-world phenomena, model them, and extract useful information. This book comprises chapters on key problems in machine learning (ML) and signal processing arenas. This book also presents the different kinds of signals that humans and machines use to communicate, and their treatments and applications. It discusses various mathematical methods involved in signal processing and ML, thereby enabling students to design their own models and optimize them efficiently. The book focuses on mathematical principles, and contains coding-based assignments for implementation. Prior exposure to ML is not required. The book also focuses on applications in signal processing and communication. This book is intended for advanced undergraduate and postgraduate students, researchers, and practitioners engaged with signal processing and its applications.

Chapter 1 Land Use and Land Cover Mapping of Remotely Sensed Data Using Fuzzy Set Theory Related Algorithm

This chapter focuses on remote sensing theory, differences between practical and ideal remote sensing systems, Landsat 8 data characteristics, and concepts of electromagnetic radiation. Remote sensing can be broadly defined as the science of obtaining useful information about physical objects and/or environment, through recording and measuring energy patterns from sensor systems.

Chapter 2 Role of AI in Mortality Prediction in Intensive Care Unit Patients

This chapter focuses on intensive care unit mortality rate and risk prediction, both of which are major concerns in any hospital. Prediction is a term highly dependent on the correct analysis of available data.

Chapter 3 A Survey on Malware Detection Using Machine Learning

This chapter presents ML structures that have been tried, investigated, and created to portray and order malware into their malware families, utilizing attributes separated and obtained from static and live examination of noxious programming.

Chapter 4 EEG Data Analysis for IQ Test Using Machine Learning Approaches: A Survey

This chapter presents an analysis of various approaches utilized for IQ test as a study. From this, the intention is to understand the various methodologies used for intelligence level estimation and use them to distinguish the downsides of existing framework.

Chapter 5 Machine Learning Methods in Radio Frequency and Microwave Domain

This chapter presents a comprehensive review on recent research advancements and ML techniques used for the optimization of RF circuits.

Chapter 6 A Survey: Emotion Detection Using Facial Reorganization Using Convolutional Neural Network (CNN) and Viola-Jones Algorithm

This chapter includes mental health and related problems. Keeping this in mind, we came up with an idea where we create an app that detects user’s emotions and suggests a binaural beats playlist for them to help calm their emotions.

Chapter 7 Power Quality Events Classification Using Digital Signal Processing and Machine Learning Techniques

This chapter discusses feature extraction based on wavelet transformation in detail. Feature selection and optimization play a vital role in optimizing datasets to the classifier after feature extraction. We use a genetic algorithm (GA) for this purpose. Then, machine learning models like probabilistic neural network (PNN), back propagation neural (BPN) network, and support vector machine (SVM) are used as classifiers to classify various PQ events.

Chapter 8 Hybridization of Artificial Neural Network with Spotted Hyena Optimization (SHO) Algorithm for Heart Disease Detection

In this chapter, we focus on the proposed method’s (SHO algorithm) effectiveness verified by different patients considering 13 constraints as the dataset. These constraints are evaluated for training and testing each data in the dataset. The efficiency of the proposed approach is shown in comparison with other methods, SLO, PSO, GDA, and GA.

Chapter 9 The Role of Artificial Intelligence, Machine Learning, and Deep Learning to Combat the Socio-Economic Impact of the Global COVID-19 Pandemic

This chapter discusses the severe impacts of COVID-19 and the role of technology particularly artificial intelligence, machine learning, and deep learning in combating the virus. Various statistical data for this survey are also presented.

Chapter 10 A Review on Smart Bin Management Systems

This chapter provides a review of various approaches used in smart bin management systems.

Chapter 11 Unlocking Machine Learning: 10 Innovative Avenues to Grasp Complex Concepts

This chapter provides a mathematical statement that expresses facts in the context of a problem, a model often known as a machine-learning algorithm. We look at 10 alternative approaches for demystifying ML and offering a knowledge route used by individuals unfamiliar with fundamental ideas, each with clear descriptions, images, and examples.

Chapter 12 Recognition Attendance System Ensuring COVID-19 Security

This chapter focuses on a face detecting, temperature sensing, report generating, and all-in-one attendance system that is low cost and easily integrated into the current system to re-innovate an existing project methodology and make it available at a low cost.

Chapter 13 Real-Time Industrial Noise Cancellation for the Extraction of Human Voice

This chapter on processing and playing uses the concept of multiprocessing in processors with each core of the processor performing a different function. The user is protected from surrounding noises by an additional passive attenuator.

Chapter 14 Machine Learning-Based Water Monitoring System Using IoT

This chapter aims to provide a machine learning-based water monitoring system using multiple sensors. The system can collect a variety of data from water, including temperature and pH, and then, water clarity is measured.

Chapter 15 Design and Modeling of an Automated Driving Inspector Powered by Arduino and Raspberry Pi

This chapter focuses on the effort made to secure way of inspecting test drive during COVID-19 pandemic using new technologies for the societal benefits. The Secured and Automated Driving Inspector powered by Arduino and Raspberry Pi is a new technology, which can be embedded in the test cars for producing eligible drivers by assisting the inspector during driver’s test.

Chapter 16 Kalman Filter-Based Seizure Prediction Using Concatenated Serial-Parallel Block Technique

This chapter focuses on a method that combines two distinct techniques, the Kalman filter and a concatenated serial-parallel block technique, to predict epileptic seizures. The Kalman filter is a mathematical tool often used for estimating and predicting the state of dynamic systems. In the context of seizure prediction, it is employed to analyze and model the dynamics of brain activity data.

Chapter 17 Current Advancements in Steganography: A review

This chapter focuses on the details of the datasets used for steganography and also contains some details of steganography’s counterpart called steganalysis.

Chapter 18 Human Emotion Recognition Intelligence System Using Machine Learning

This chapter investigates a speech emotion detection system that outperforms previous systems on the basis of data, feature selection, and methodology.

Chapter 19 Computing in Cognitive Science Using Ensemble Learning

This chapter focuses on human–computer interaction to find a technique to identify human activity and apply ensemble learning from an individual level to produce a more accurate prediction of a whole that fully mirrors social-cognitive science in a very efficient computational approach.

Overall, editing this book was an incredible opportunity. We are indebted to many individuals for their help and support, including our families, friends, and fellow members. We are especially grateful to the chairman, principal, faculty, and fraternity of the Sahyadri College of Engineering and Management, Mangaluru and NMAM Institute of Technology, Nitte.

Chandra Singh

Rathishchandra R. Gatti

K.V.S.S.S.S. Sairam

Manjunatha Badiger

Naveen Kumar S.

Varun Saxena

1Land Use and Land Cover Mapping of Remotely Sensed Data Using Fuzzy Set Theory-Related Algorithm

Adithya Kumar* and Shivakumar B.R.

NMAM Institute of Technology, NITTE, Udupi, Karnataka, India

Abstract

All bodies, planets, living beings, and inanimate objects emit electromagnetic radiation, and the amount and type of radiation emitted depend largely on their temperatures. Electromagnetic radiation may be emitted by an object or may come from another body and could be reflected by it. There are operational satellite systems that sample virtually every region of the electromagnetic spectrum, with spatial resolutions from 0.5 to 5,000 m. The scientific community’s interest in spatiotemporal studies of global change, environmental monitoring, and human impacts involves the use of remote sensing data. Remote sensing systems, particularly those located on satellites, provide a repetitive and synoptic vision of Earth, which is of great interest in monitoring and analyzing human activities and their impacts. The activities include the evaluation and monitoring of the environment like urban growth and hazardous waste; detection and monitoring of global changes, deforestation, global warming, and exploration of non-renewable resources and their land use; civil engineering;, the acquisition of satellite imagery from relevant sources; the preprocessing of satellite imagery as per requirement; and the development and classification of satellite data using fuzzy C-means (FCM), modified FCM, K-means, and fuzzy inference system (FIS) techniques.

Keywords: FCM, modified FCM, K-means, FIS techniques

1.1 Introduction

In this chapter, an introduction to remote sensing theory, the differences between practical and ideal remote sensing systems and the Landsat 8 data characteristics and concepts of electromagnetic radiation are presented [1, 2]. Remote sensing can be broadly defined as the science of obtaining useful information about physical objects and/or the environment through recording and measuring energy patterns from the sensor systems. The information about distant objects and the environment is obtained through the electromagnetic radiations emitted or reflected by the Earth’s surface features, which are of interest. Multiple data users that make use of the ideal remote sensing system would be able to acquire the required data at a higher speed with less or no expense of any required area of interest [3, 4]. Given this information, multiple data users would be able to make decisions on managing and observing the Earth’s surface features. A real remote sensing system encounters a lot of problems in each stage of the remote sensing system. The following section explains the limitations of all ideal remote sensing systems. The spectral distribution of reflected and/or emitted electromagnetic energy is not uniform, and it varies from time and place on different Earth features. Passive remote sensing systems entirely depend on the energy from the sun. Solar energy is non-uniform with respect to time and place. The sources used in the remote sensing systems are also non- uniform with respect to wavelength and time. Therefore, there is always a need to calibrate these sources depending on the mission, time, and location. In a real remote sensing system, there is always a need for atmospheric and radiometric corrections due to atmospheric errors. The atmosphere always modifies the energy emitted and reflected by the Earth’s surface features. Eliminating the atmospheric effects is important in those applications involving observations of the same geographical area. Different materials do not reflect and emit energy uniquely. There is spectral overlapping in the recorded data. The spectral response plays an important role in detecting, identifying, and analyzing the Earth’s surface materials. A good understanding of energy interactions is required to obtain information about the required features. From the above points, it is clear that super sensors do not exist. The sensors cannot be sensitive to all possible wavelengths and have limits with respect to sensitivity. The sensors record the objects according to their spatial resolution. The ability of the sensors to differentiate the smallest object and separate it from its surroundings is called spatial resolution. The spatial resolution depends on the heterogeneity of the ground area being sensed. The digital images obtained consist of mixed pixels causing problems in identifying the land cover classes. Depending on the application, the choice of sensors varies from airborne vehicles to space stations. Photographic systems produce images of higher spatial resolution, but spectral sensitivity is absent in those sensors. Therefore, to obtain information on all the land cover classes, sensors with higher spectral resolution are required rather than sensors with higher spatial resolution. However, sensors having higher spectral resolution are expensive. Proper image processing techniques are required to visualize if the sensor data are of low resolution. Therefore, there are always tradeoffs between cost and resolution. Remote sensing systems generally are of two types: passive and active. An active remote sensing system possesses an illumination system or source of energy. It directs this energy to the objects to measure the energy. Some examples of active sensor systems are radars and laser scanners. The passive remote sensing system utilizes the sun’s energy to record the emitted and/or reflected values of the surface features. Landsat series sensors are an example for passive remote sensing systems. The sensors that generate artificial radiation to the Earth’s surface features are called active sensors. Radio detection and ranging or radar is used to detect the Earth’s surface features using the pulses of electromagnetic radiation in microwave ranges. These radiations are not affected by weather conditions such as clouds, fog, and wind. The reflected signal intensity is used to extract the information. These types of sensors using self-generated radiation are useful for specific satellite missions and operations [5, 6].

Sensors that use natural light and the thermal radiation of the Earth’s features to extract the information are called passive. This information is greatly affected by the seasons, time of capture, and height of the satellite. The Landsat satellite is equipped with spectrometers that measure the signals at a spectrum of bands. Therefore, multispectral channels are obtained, which provide greater information. Satellite imagery consisting of pixel values are the measures of the intensity of the electromagnetic radiation of the Earth’s surface features. The values of brightness differ in the image based on time, type of area, and sensors. Certain properties of imagery like resolution should be known. The individual elements of the digital image are called pixels. The size of the pixel and the type of sensor determine the spatial resolution. The length of the edges of the pixels is used to measure resolution. The number of spectral channels is defined as the spectral resolution of the image. The reflection of Earth surface features occurs in different wavelengths. The human eyes can recognize only the visible spectrum of radiations. The land area, which has high reflections in particular wavelengths, is sensed by the satellite sensors. Most of the Earth observation satellites have between three and eight bands. The linear imaging self-scanning sensor III (LISS-III) has four spectral bands. The higher spectral resolution allows larger features to be extracted.

The time interval the satellite takes to reach the same area is called temporal resolution. The temporal resolution is determined by the orbit of the satellite, the altitude, and the sensor’s characteristics. Generally, Earth observation satellites have a repetition rate of 14–16 days. Landsat 7 has 16 days. The highest temporal resolution is for meteorological satellites, which is 15 minutes [7, 8].

The surface reflectance characteristics of three main Earth surface features are discussed here:

Vegetation:

The chlorophyll pigment in leaves absorbs radiation in red and blue wavelengths but reflects in the green wavelength. The reflectance characteristics vary with wavelength [

9

,

10

,

16

,

18

].

Water:

The water absorbs most of the radiation in the longer wavelengths. The water reflects in shorter wavelengths and hence blue in appearance. In the near-infrared wavelengths, the water is darker in color. The reflectance varies due to the depth of the water, the materials in the water, and the type of water

[12]

.

Soil:

The nature of the soil determines the reflection and emission. The presence of moisture decreases the reflectance. The reflectance of soil is present in a large range of wavelengths depending upon the soil

[13]

.

1.1.1 Overview on Landsat 8

Landsat 8 is an American Earth observation satellite launched in 2013. It is the seventh satellite to reach orbit successfully. The operational land imager (OLI) sensor consists of nine capturing devices, and thermal infrared sensor (TIS) has two capturing devices. The thermal bands measure the surface temperature and lighter pixels represent hot temperatures. Thermal bands provide a lot of information regarding heat units present in urban regions. The Landsat 8 orbit height is 705 km above the Earth’s surface. It collects 400 scenes of 185-km wide regions in one day, and an archive is made available on the United States Geographical Survey (USGS) website within 24 hours of acquisition. Generally, the first three band combinations of the initial nine bands are used as the inputs to the algorithms. Other band combinations can also be used as per the analyst’s choice. These band combinations are useful to extract certain land cover classes more precisely [14].

This conveys information on the various stages of remote sensing systems and the techniques used to capture remotely sensed (RS) data. The choice of the sensor was discussed, which is an important decision in obtaining proper RS data with good-quality images. In the next section, an introduction to the image classification process will be discussed.

1.2 Image Classification

Image classification, in remote sensing perspective, is said to be a collection of three processes:

Image preprocessing,

Classification, and

Image post-processing.

In the upcoming sections, a detailed explanation of each of the abovementioned processes is presented.

Preprocessing steps are applied to compensate for systematic errors and enhance the image for a higher level of interpretability, and are not necessarily implemented in all classification procedures. The most basic types of corrections used for preprocessing are radiometric and geometric corrections. Similarly, the most basic enhancement techniques are categorized into linear and non-linear enhancement techniques. Image corrections may be employed to remove the errors caused by the atmosphere, seasonal variations, and imperfections in scanning instruments. In the following sections, an explanation of the two types of corrections is presented. The sun’s azimuth, elevation, and atmospheric conditions cause radiometric errors. The digital numbers are error-prone and must be corrected. Radiometric correction is important when data are obtained at different time intervals. The pixel value obtained not only includes the emitted and/or reflected radiation of the Earth’s surface, but also the radiation scattered by the atmosphere. Because the interest is in the actual surface values, the corrections are required. The sensor records the magnitude of the electromagnetic radiations as a digital number (DN). These numbers can be transformed into radiance and reflectance. Metadata information is required to perform this conversion. This process improves the quality and interpretability of RS data. The geometric distortions are caused by the variations in terrain elevation and the Earth’s spinning motion. Since the Earth’s rotation on its axis is from west to east and satellite orbits occur from the Earth’s pole to pole, the errors in elevation occur. The correction of geometric errors can be done through a process called rubber sheeting. It involves the warping and stretching of the image to georegister control points. Image enhancement deals with converting raw satellite data into better, more interpretable images. The commonly used image enhancement techniques are discussed here. Image enhancement techniques are broadly classified into two ways: global enhancement and spatial enhancement. Examples for global enhancement are linear contrast stretch, histogram equalization, and piece-wise contrast stretch. Spatial enhancement techniques are concerned with smoothing and sharpening images. The contrast enhancement is done to distinguish the finer differences of the features. The higher DN values and additive effects cause a reduction in contrast to the blue and the infrared range of the electromagnetic spectrum. This effect is called a haze, and contrast enhancements are used to minimize the effects. The back scattering to space by the molecules in the atmosphere causes path radiance. This radiance is detected by the sensor, which must be eliminated. In the histogram plot, the low-end DN values are assigned to extreme black, and high-end DN values are assigned to extreme white. The pixels in the medium range are distributed linearly. The drawback of linear stretch rarely occurs. DN values are assigned many levels as it does to frequently occurring DN values. The linear contrast in most cases produces a dull image. The single land cover class can be discriminated properly using this method. The detailed differences within the regions of interest are done, and remaining areas are assigned a single gray tone. Histogram equalization allows areas of lower contrast to obtain higher contrast. This method is largely used to enhance the image. The image intensities are better distributed to perform equalization. The frequently repeating intensity values are spread efficiently, and therefore increase global contrast. Image classification is an important part in digital image processing since it is necessary to have an image displaying a magnitude of colors representing different land cover features of the Earth’s surface. To obtain this imagery, many classification algorithms are presented in the literature. Image classification can be categorized based on multiple factors. The most basic categorization is based on the requirement of training sites. The results of this categorization are supervised and unsupervised classification techniques. The first step in the RS image classification stage is the acquisition of imagery. The satellite imagery obtained must be visually interpreted for corrections. There will be a requirement for both the removal of noises and the enhancement techniques. These are the preprocessing stages before classification. The preprocessing stages like noise removal and enhancement techniques are done if the input data are found to have errors; otherwise, according to requirement, only necessary preprocessing steps can be followed such as radiometric or geometric corrections. After all these operations, the classification process begins. The classification procedure has two main categories, unsupervised and supervised.

1.3 Unsupervised Classification

The ability of the algorithm to statistically group the data without training data is called unsupervised classification. The unsupervised classifier performs clustering or division of datasets with the help of similarity and dissimilarity features [11]. The classes that result after grouping are called spectral classes. The pixels closer to each other are more likely to belong to the same class. Therefore, as shown in Figure 1.1, the pixels are grouped by evaluating their features using the algorithm. The main aim of classification in remote sensing is to represent the different features of the Earth’s surface in a different magnitude of colors for environmental monitoring, change detection, and urban planning.

Figure 1.1 Illustration of an unsupervised classification [15].

1.4 Supervised Classification

Supervised classification algorithms use reference classes as training data. This additional information must be provided by the user. The pixels extracted from the different channels are provided as inputs to the trained classifier as shown in Figure 1.2.

The supervised classification techniques require time to collect the training data and to train the classifier. In remote sensing, the land cover classes already known are used as training data. The features of the training data are used as references. The verification of the classifier can also be done using training data. The post-processing involves an accuracy assessment of the obtained classification output. It involves the collection of ground truth information from trusted sources and correlates the points on the satellite imagery. This ground truth information is compared with the predicted output of the classifier to obtain the overall classification accuracy [17].

Figure 1.2 Illustration of a supervised classification [19].

1.5 Overview of Fuzzy Sets

In the previous section, the image classification procedure and preprocessing steps were explained. This section explains the concept of fuzzy sets and their applicability in RS data.

Lotfi A. Zadeh initiated the fuzzy logic at the University of California, Berkeley in 1965. Boolean logic or crisp logic is the opposite of fuzzy logic. The element of a set belongs to only one set in Boolean logic, but fuzzy logic evaluates the probability of the element belonging to every set. Notions that are rather small or very slow can be written mathematically and processed by the algorithms in more human-like thinking.

The membership functions are used to give the grades to these input variables. The membership functions are of different types such as Gaussian, triangular, and trapezoidal. They are chosen based on the requirements of the application. Due to the presence of spectrally similar land cover classes, the traditional maximum likelihood classifier does not properly differentiate within the classes. The misclassifications occur between the building and the road classes. The spatial features can be properly utilized by fuzzy pixel-based classifiers. The classifier uses the initial classes by the maximum likelihood classifier to perform classification. The fuzzy rules are then used to perform the decision on which class the pixels must be assigned. The process of converting a crisp quantity to fuzzy is called fuzzification. Many quantities like speed and temperature may be deterministic, but they have uncertainty. The uncertainty in the variable causes unclear information. This can be represented better using fuzzy sets. The membership functions are used to represent this information. Similarly, the pixels are also not crisp or deterministic in satellite imagery. In this study, fuzzy-related algorithms were used to perform classification. The classification process involves the assignment of pixels to a land cover class. The categorization of pixels in the image is performed by classification algorithms. The process of converting a fuzzy quantity to a crisp one is called defuzzification. The output obtained through the operation of fuzzy sets is fuzzy. Therefore, there is a need to convert it to crisp logic. The result of the land cover classification has to be converted back to crisp value to finally assign the pixel to one particular class. After all the pixels are converted to crisp logic, it is displayed as a thematic map. This provides a brief overview of fuzzy theory and its concepts. Properties and operations of fuzzy sets are also presented with relevant mathematical equations. A brief introduction to fuzzy classification is also presented. In the next section, the classification techniques used in the study are discussed in detail.

In the previous section, the fuzzy theory and its properties were discussed. In this section, the implementation of fuzzy C-means (FCM), K-means, and the proposed FCM are presented and discussed. This section also includes the application of fuzzy inference systems (FIS) for land cover classification.

1.5.1 Fuzzy C-Means Clustering

RS data can be partitioned by initializing the number of partitions or clusters required, which is c. For example, if three classes need to be classified, then c is three. Then three cluster centers are initialized randomly with the number in the range of 0 to 65,535. For example, c1 = 20,000, c2 = 40,000, and c3 = 50,000. These cluster centers are used to calculate the membership grades (uij), which lie in the range of 0 to 1.

Then, the cluster centers are recalculated through an iterative process that attempts to minimize the following objective function.

(1.1)

1.5.2 Algorithm of Fuzzy C-Means

The fuzzy C-means algorithm consists of the following steps:

Step 1: Convert the input image into a one-dimensional array.

Step 2: Select the number of clusters, c, and initialize randomly.

Step 3: Compute the membership matrix for pixel i and class j.

(1.2)

Step 4: Recalculate the c cluster centers.

(1.3)

Step 5: Calculate the objective function as in Section 1.5.1.

Step 6: Repeat steps 3 to 5 until the objective function is minimized. Otherwise, exit.

1.6 Methodology

Landsat 8 satellite imagery of a region in Kumta Taluk was obtained from the USGS Earth Explorer website. The preprocessing of the satellite imagery was performed by employing histogram equalization and radiometric corrections. The implementation of FCM, modified FCM, K-means, and FIS was done in MATLAB® R2015a. Performing land use land cover (LULC) mapping of the considered study area by using the above techniques was done to classify the required satellite image by feeding the inputs required. It is learned that the manual feature extraction of a 1 km2 area using high-resolution data like IKONOS is very expensive. Particularly, urban areas are highly dynamic, which are strongly involved in inter-urban land use pattern changes.

1.6.1 Modified Fuzzy C-Means Technique

The advantage of the proposed algorithm over conventional FCM is that it uses all the input bands of a multispectral imagery. This eliminates the effects of spectral correlation on the dataset. Figure 1.4 shows the mean values of each LULC class plotted against a band number. It is observed from Figure 1.4 that using the multispectral imagery at the input of FCM suffers from the effects of spectral overlapping, band correlation, and mixed pixel issues. Hence, it is advantageous to classify each multispectral band separately using FCM algorithm for an efficient classification process. The flowchart of modified FCM is depicted in Figure 1.3. When a three-band combination image is used, the pixel values are very close to each other, which is inaccurate. The plot of pixel means against bands as in Figure 1.4 shows that the Evergreen Forest, Water Body, and Vegetation have good separability in band 7 compared to other bands. It also shows that band 1 and band 2 together are of no use because of the high correlation in their pixel values.

1.6.2 Construction of a Fuzzy Inference System

The Gaussian function utilizes the mean (xc) and the standard deviation (σ) to calculate the membership grade (μx) as shown in Figure 1.5. The training areas are used to obtain the mean and the standard deviation from which the membership function of the land cover class is built. For example, the membership function of water is obtained from the training areas by calculating the mean and the standard deviation. Similarly, all the membership functions of the required land cover classes are obtained as shown in Figure 1.6 for band 6. By calculating the mean and the standard deviation from the required bands, the database of the membership functions is built as shown in Figures 1.6, 1.7 and 1.8. The rule base contains a number of fuzzy if–then rules.

Figure 1.3 Flowchart of the modified FCM.

Figure 1.4 Plot of the pixel mean values v/s band numbers.

Figure 1.5 Gaussian membership function.

Figure 1.6 Membership functions defined for band 6.

Figure 1.7 Membership functions defined for band 5.

Figure 1.8 Membership functions defined for band 7.

After the membership functions are prepared by using the Gaussian membership function, the rule base is prepared by using IF–THEN rules. These rules connect the membership functions using AND and OR operators as shown in Figure 1.9. In the rule viewer, the AND stands for minimum fuzzy set operation and the OR stands for maximum fuzzy set operation.

Figure 1.9 Rule viewer showing the connection between input and output.

The inputs provided to the rule base are fuzzified using the Gaussian membership function, and by applying the minimum and maximum rules, the fuzzy output is obtained. This fuzzy output is converted to a crisp output by applying the defuzzification operation.

1.6.3 K-Means Algorithm

The K-means algorithm is an unsupervised algorithm used for grouping pixels in a simple and easy way. The procedure is to assign centroids for each cluster and calculate the distance of the pixels with all the cluster centers. New cluster centers are calculated at the end of each iteration. Suppose that there are n samples: x1, x2, x3,…,xn. The objective is to develop k clusters and k should be less than n. The procedure for developing k clusters is as follows:

Step 1: Select k random points in the signal space; these points represent the initial cluster centers.

Step 2: Assign each pixel to the cluster to which it has the least distance.

Step 3: When all the pixels are assigned, recalculate the cluster centers to produce new centers.

Step 4: Repeat steps 2 and 3 until the centers become constant or until the difference in old and new cluster centers is negligible. After the centers are optimized, the algorithm stops and exits.

The objective function used to terminate the algorithm from its iterations is given by:

(1.4)

where ║xi(j) − cj║ is the term that calculates the distances of the pixels with the cluster center cj. The optimized cluster centers are then used to assign once again all the pixels to specific classes and display a classified image. Classification techniques used in the study are qualitatively and quantitatively presented. Three unsupervised clustering algorithms are discussed: FCM, K-means, and modified FCM. The supervised technique presented in the study is the FIS. In the next section, the results obtained by the classification techniques are discussed.

1.7 Results and Discussion

A detailed discussion of the classification techniques considered for the study has been provided. In this section, the results of the classification of Landsat 8 RS data using FCM, modified FCM, FIS, and K-means techniques are presented. The FCM technique has been implemented with two different approaches. In the first approach, a three-band multispectral image was provided as the input. This approach was not able to classify the land covers when diverse classes were present because of highly correlated pixel values. The second approach involved the use of all input data bands at the FCM input and showed promising classification results. Results for FIS are also presented in the chapter. The Landsat 8 data used in the study were acquired from the USGS website (www.earthexplorer.usgs.gov).

1.7.1 FCM Technique Results

Figures 1.10 and 1.11 show the classified output map of the study area. For example, 13 pixels belonging to Crop Land and Kharif have been misclassified to Thin Vegetation. Similarly, 418 pixels belonging to Evergreen Forest have been misclassified to Crop Land and Kharif. The overall accuracy of this classification method is 54.01%. The overall accuracy was obtained using Equation 1.5:

Figure 1.10 Study area used for the research.

Figure 1.11 FCM classified map of the study area for band combination 574.

(1.5)

1.7.2 Modified FCM Technique Results

The second dataset uses individual bands at the input of FCM successively. Table 1.1 shows the confusion matrix obtained for FCM using this method. The output classified map is shown in Figure 1.12. The overall classification accuracy of this classification is 97.83%. The land cover classes such as the Evergreen Forest, Scrub Land, Thin Vegetation, Kharif, Crop Land, and Water Body have been classified. A single-band classification properly classifies the land covers because each band is used based on its reflectance of the land features. The single band is classified to obtain a certain land cover class. Similarly, all the land cover classes are classified. This method avoids the pixel mixing caused by the three-band combination. The FIS is implemented by building a classifier. The process of building a classifier is by obtaining training areas of the required land cover classes for the specific channel. The membership functions of the FIS are built using the mean and the standard deviation of the land cover classes. The values obtained from the training areas build the Gaussian membership functions. These are obtained from the specific land cover classes. In FIS rules, the AND operator corresponds to the minimum value and the OR operator corresponds to the maximum value. All rules are connected using the OR operator, which takes the maximum value from the fuzzified inputs. The classified output of the FIS method is shown in Figure 1.13. The land cover classes have been well separated by the FIS system. The overall classification accuracy is found to be 91.19%. It is observed that the accuracy of the FIS classifier depends on the training values provided.

Table 1.1 Confusion matrix for conventional FCM.

P

Water body and scrub land

Crop land and Kharif

Evergreen forest

Thin vegetation

A

Water Body and Scrub Land

86

0

0

0

Crop Land and Kharif

0

28

0

13

Evergreen Forest

0

418

5355

4227

Thin Vegetation

0

12

0

29

Overall Classification Accuracy = 54.01%

A: Actual, P: Predicted

Figure 1.12 Modified FCM classified map for the study area.

Figure 1.13 Classification output of the FIS method.

The advantage of the FIS is the ability to implement the classification using simple IF-THEN rules. This system allows adding any number of input bands. It helps to make use of all the Landsat 8 bands, which contain more information than the conventional three-band combination input. The classification output of the FIS method is depicted in