Deep Learning for Radar and Communications Automatic Target Recognition

"This exciting resource identifies technical challenges, benefits, and directions of Deep Learning (DL) based object classification using radar data (i.e., Synthetic Aperture Radar / SAR and High range resolution Radar / HRR data). An overview of machine learning (ML) theory to include a history, background primer, and example and performance of ML algorithm (i.e., DL method) on video imagery is provided. Radar data with issues of collection, application, and examples for SAR/HRR data and communication signals analysis is also discussed. Practical considerations of deploying such techniques, including performance evaluation, hardware issues, and the future unresolved issues are presented."--Amazon.com Artech House Radar Series Deep Learning for Radar and Communications Automatic Target Recognition 1 Contents 6 Foreword 12 Preface 14 CHAPTER 1 Machine Learning and Radio Frequency: Past, Present, and Future 20 1.1 Introduction 20 1.1.1 Radio Frequency Signals 20 1.1.2 Radio Frequency Applications 23 1.1.3 Radar Data Collection and Imaging 26 1.2 ATR Analysis 33 1.2.1 ATR History 33 1.2.2 ATR from SAR 34 1.3 Radar Object Classification: Past Approach 34 1.3.1 Template-Based ATR 34 1.3.2 Model-Based ATR 36 1.4 Radar Object Classification: Current Approach 38 1.5 Radar Object Classification: Future Approach 39 1.5.1 Data Science 40 1.5.2 Artificial Intelligence 41 1.6 Book Organization 42 1.7 Summary 43 References 43 CHAPTER 2 Mathematical Foundations for Machine Learning 48 2.1 Linear Algebra 48 2.1.1 Vector Addition, Multiplication, and Transpose 48 2.1.2 Matrix Multiplication 49 2.1.3 Matrix Inversion 50 2.1.4 Principal Components Analysis 50 2.1.5 Convolution 53 2.2 Multivariate Calculus for Optimization 53 2.2.1 Vector Calculus 54 2.2.2 Gradient Descent Algorithm 55 2.3 Backpropagation 58 2.4 Statistics and Probability Theory 62 2.4.1 Basic Probability 63 2.4.2 Probability Density Functions 63 2.4.3 Maximum Likelihood Estimation 65 2.4.4 Bayes’ Theorem 66 2.5 Summary 68 References 68 CHAPTER 3 Review of Machine Learning Algorithms 70 3.1 Introduction 70 3.1.1 ML Process 71 3.1.2 Machine Learning Methods 73 3.2 Supervised Learning 78 3.2.1 Linear Classifier 79 3.2.2 Nonlinear Classifier 89 3.3 Unsupervised Learning 101 3.3.1 K-Means Clustering 101 3.3.2 K-Medoid Clustering 103 3.3.3 Random Forest 104 3.3.4 Gaussian Mixture Models 105 3.4 Semisupervised Learning 107 3.4.1 Generative Approaches 107 3.4.2 Graph-Based Methods 108 3.5 Summary 112 References 113 CHAPTER 4 A Review of Deep Learning Algorithms 116 4.1 Introduction 116 4.1.1 Deep Neural Networks 117 4.1.2 Autoencoder 119 4.2 Neural Networks 124 4.2.1 Feed Forward Neural Networks 124 4.2.2 Sequential Neural Networks 133 4.2.3 Stochastic Neural Networks 138 4.3 Reward-Based Learning 142 4.3.1 Reinforcement Learning 142 4.3.2 Active Learning 145 4.3.3 Transfer Learning 145 4.4 Generative Adversarial Networks 149 4.5 Summary 155 References 156 CHAPTER 5 Radio Frequency Data for ML Research 160 5.1 Introduction 160 5.2 Big Data 160 5.2.1 Data at Rest versus Data in Motion 161 5.2.2 Data in Open versus Data of Importance 162 5.2.3 Data in Collection versus Data from Simulation 165 5.2.4 Data in Use versus Data as Manipulated 167 5.3 Synthetic Aperture Radar Data 169 5.4 Public Release SAR Data for ML Research 170 5.4.1 MSTAR: Moving and Stationary Target Acquisition and Recognition Data Set 170 5.4.2 CVDome 172 5.4.3 SAMPLE 173 5.5 Communication Signals Data 175 5.5.1 RF Signal Data Library 176 5.5.2 Northeastern University Data Set RF Fingerprinting 177 5.6 Challenge Problems with RF Data 177 5.7 Summary 179 References 181 CHAPTER 6 Deep Learning for Single-Target Classification in SAR Imagery 184 6.1 Introduction 184 6.1.1 Machine Learning SAR Image Classification 185 6.1.2 Deep Learning SAR Image Classification 186 6.2 SAR Data Preprocessing for Classification 187 6.3 SAR Data Sets 188 6.3.1 MSTAR SAR Data Set 188 6.3.2 CVDome SAR Data Set 190 6.4 Deep CNN Learning 191 6.4.1 DNN Model Design 191 6.4.2 Experimentation: Training and Verification 192 6.4.3 Evaluation: Testing and Validation 193 6.4.4 Confusion Matrix Analysis 194 6.5 Summary 200 References 202 CHAPTER 7 Deep Learning for Multiple Target Classification in SAR Imagery 206 7.1 Introduction 206 7.2 Challenges with Multiple-Target Classification 207 7.2.1 Constant False Alarm Rate Detector 208 7.2.2 R-CNNs 209 7.2.3 You Only Look Once 209 7.2.4 R-CNN Implementation 210 7.3 Multiple-Target Classification 212 7.3.1 Preprocessing 213 7.3.2 Two-Dimensional Discrete Wavelet Transforms for Noise Reduction 213 7.3.3 Noisy SAR Imagery Preprocessing by L1-Norm Minimization 215 7.3.4 Wavelet-Based Preprocessing and Target Detection 216 7.4 Target Classification 218 7.5 Multiple-Target Classification: Results and Analysis 219 7.6 Summary 221 References 221 CHAPTER 8 RF Signal Classification 224 8.1 Introduction 224 8.2 RF Communications Systems 226 8.2.1 RF Signals Analysis 227 8.2.2 RF Analog Signals Modulation 230 8.2.3 RF Digital Signals Modulation 231 8.2.4 RF Shift Keying 232 8.2.5 RF WiFi 234 8.2.6 RF Signal Detection 236 8.3 DL-Based RF Signal Classification 239 8.3.1 DEEP Learning for Communications 239 8.3.2 DEEP Learning for I/Q systems 239 8.3.3 DEEP Learning for RF-EO Fusion Systems 241 8.4 DL Communications Research Discussion 243 8.5 Summary 246 References 247 CHAPTER 9 Radio Frequency ATR Performance Evaluation 250 9.1 Introduction 250 9.2 Information Fusion 250 9.3 Test and Evaluation 254 9.3.1 Experiment Design 256 9.3.2 System Development 257 9.3.3 Systems Analysis 258 9.4 ATR Performance Evaluation 258 9.4.1 Confusion Matrix 260 9.4.2 Object Assessment from Confusion Matrix 262 9.4.3 Threat Assessment from Confusion Matrix 264 9.5 Receiver Operating Characteristic Curve 265 9.5.1 Receiver Operating Characteristic Curve from Confusion Matrix 265 9.5.2 Precision-Recall from Confusion Matrix 269 9.5.3 Confusion Matrix Fusion 271 9.6 Metric Presentation 272 9.6.1 National Imagery Interpretability Rating Scale 272 9.6.2 Display of Results 275 9.7 Conclusions 275 References 276 CHAPTER 10 Recent Topics in Machine Learning for Radio Frequency ATR 282 10.1 Introduction 282 10.2 Adversarial Machine Learning 283 10.2.1 AML for SAR ATR 283 10.2.2 AML for SAR Training 284 10.3 Transfer Learning 289 10.4 Energy-Efficient Computing for AI/ML 291 10.4.1 BM’s TrueNorth Neurosynaptic Processor 293 10.4.2 Energy-Efficient Deep Networks 294 10.4.3 MSTAR SAR Image Classification with TrueNorth 294 10.5 Near-Real-Time Training Algorithms 294 10.6 Summary 296 References 297 About the Authors 300 Index 302 Radar;,Deep,learning;,Target,recognition;,Machine,learning;,978-1-63081-637-7;,Artech,House Radar,Deep learning,Target recognition,Machine learning,978-1-63081-637-7,Artech House This authoritative resource presents a comprehensive illustration of modern Artificial Intelligence / Machine Learning (AI/ML) technology for radio frequency (RF) data exploitation. It identifies technical challenges, benefits, and directions of deep learning (DL) based object classification using radar data, including synthetic aperture radar (SAR) and high range resolution (HRR) radar. The performance of AI/ML algorithms is provided from an overview of machine learning (ML) theory that includes history, background primer, and examples. Radar data issues of collection, application, and examples for SAR/HRR data and communication signals analysis are discussed. In addition, this book presents practical considerations of deploying such techniques, including performance evaluation, energy-efficient computing, and the future unresolved issues.