Muutke küpsiste eelistusi

Deep Neural Network Design for Radar Applications [Kõva köide]

Edited by (University of Alabama, USA)
  • Formaat: Hardback, 420 pages, kõrgus x laius: 234x156 mm
  • Sari: Radar, Sonar and Navigation
  • Ilmumisaeg: 04-Feb-2021
  • Kirjastus: Institution of Engineering and Technology
  • ISBN-10: 1785618520
  • ISBN-13: 9781785618529
Teised raamatud teemal:
Deep Neural Network Design for Radar ApplicationsSuurem pilt
  • Formaat: Hardback, 420 pages, kõrgus x laius: 234x156 mm
  • Sari: Radar, Sonar and Navigation
  • Ilmumisaeg: 04-Feb-2021
  • Kirjastus: Institution of Engineering and Technology
  • ISBN-10: 1785618520
  • ISBN-13: 9781785618529
Teised raamatud teemal:
Püsilink: https://www.kriso.ee/db/9781785618529.html
Märksõnad:

Novel deep learning models are achieving state-of-the-art accuracy in the area of radar target recognition, sometimes exceeding human-level performance. The book provides an introduction to the key aspects of machine learning that any radar engineer seeking to apply deep learning to radar signal processing ought to be aware of.



Novel deep learning approaches are achieving state-of-the-art accuracy in the area of radar target recognition, enabling applications beyond the scope of human-level performance. This book provides an introduction to the unique aspects of machine learning for radar signal processing that any scientist or engineer seeking to apply these technologies ought to be aware of.

The book begins with three introductory chapters on radar systems and phenomenology, machine learning principles, and optimization for training common deep neural network (DNN) architectures. Subsequently, the book summarizes radar-specific issues relating to the different domain representations in which radar data may be presented to DNNs and synthetic data generation for training dataset augmentation. Further chapters focus on specific radar applications, which relate to DNN design for micro-Doppler analysis, SAR-based automatic target recognition, radar remote sensing, and emerging fields, such as data fusion and image reconstruction.

Edited by an acknowledged expert, and with contributions from an international team of authors, this book provides a solid introduction to the fundamentals of radar and machine learning, and then goes on to explore a range of technologies, applications and challenges in this developing field. This book is also a valuable resource for both radar engineers seeking to learn more about deep learning, as well as computer scientists who are seeking to explore novel applications of machine learning.

In an era where the applications of RF sensing are multiplying by the day, this book serves as an easily accessible primer on the nuances of deep learning for radar applications.

About the editor xv
Acknowledgements xvii
Prologue: perspectives on deep learning of RF data 1(8)
Sevgi Zubeyde Gurbuz
Eric S. Mason
P.1 The need for novel DNN architectures
3(2)
P.2 Physics-aware ML that exploits RF data richness
5(2)
P.3 RF sensing problems that can benefit from DL
7(1)
P.4 Overview of this book
8(1)
Part I Fundamentals
9(88)
1 Radar systems, signals, and phenomenology
11(32)
Sevgi Zubeyde Gurbuz
Shunqiao Sun
David Tahmoush
1.1 Physics of electromagnetic scattering
11(7)
1.1.1 Effects of propagation medium
14(1)
1.1.2 Radar cross section
14(2)
1.1.3 Radar range equation
16(2)
1.2 Basic radar measurements and waveforms
18(7)
1.2.1 Continuous wave radar
18(2)
1.2.2 Pulsed CW radar
20(1)
1.2.3 Range and Doppler resolution
21(1)
1.2.4 Pulsed-Doppler radar
22(1)
1.2.5 Frequency-modulated continuous wave (FMCW) radar
22(2)
1.2.6 Ambiguity functions and range-Doppler coupling
24(1)
1.3 Real and synthetic aperture radar processing
25(6)
1.3.1 Moving target indication
26(1)
1.3.2 Array processing techniques
26(2)
1.3.3 The micro-Doppler effect and time-frequency analysis
28(1)
1.3.4 Synthetic aperture radar
29(2)
1.4 Radar data denoising for machine learning
31(1)
1.5 Radar data representations for machine learning
32(7)
1.5.1 State-of-the-art automotive radar
33(1)
1.5.2 Automotive radar with MIMO radar technology
34(2)
1.5.3 High-resolution automotive radar and point clouds
36(3)
1.6 Additional reading
39(1)
References
40(3)
2 Basic principles of machine learning
43(26)
Ali Cafer Gurbuz
Fauzia Ahmad
2.1 Learning from data
44(3)
2.1.1 Supervised learning
44(2)
2.1.2 Unsupervised learning
46(1)
2.2 Ingredients of an ML algorithm
47(1)
2.3 Basic techniques of supervised and unsupervised learning
48(6)
2.3.1 Supervised learning approaches
49(4)
2.3.2 Unsupervised learning approaches
53(1)
2.4 Evaluation of a machine learning algorithm
54(9)
2.4.1 General workflow of an ML algorithm
54(1)
2.4.2 Performance metrics
55(4)
2.4.3 Measuring generalization: training, validation, and test sets
59(1)
2.4.4 Overfitting and underfitting
60(3)
2.5 Conclusion
63(1)
References
63(6)
3 Theoretical foundations of deep learning
69(28)
Stefan Bruggenwirth
Simon Wagner
3.1 Introduction
69(2)
3.2 Perceptron
71(1)
3.3 Sigmoid perceptron
72(1)
3.4 Multilayer perceptron
72(2)
3.5 Gradient descent
74(1)
3.6 Backpropagation
75(3)
3.7 Improvements
78(4)
3.7.1 Activation functions
78(2)
3.7.2 Momentum
80(1)
3.7.3 Stochastic gradient descent
80(1)
3.7.4 Cross-entropy cost function
81(1)
3.7.5 Regularization
81(1)
3.8 The short history of deep learning
82(1)
3.9 Convolutional neural networks
82(5)
3.9.1 Structure of a convolutional neural network
83(2)
3.9.2 Training of a convolutional neural network
85(2)
3.10 Autoencoders
87(6)
3.10.1 Data compression with autoencoders
88(3)
3.10.2 Other applications of autoencoders
91(2)
3.11 Advanced training techniques and further applications of deep learning
93(2)
References
95(2)
Part II Special topics
97(110)
4 Radar data representation for classification of activities of daily living
99(24)
Baris Erol
Moeness G. Amin
4.1 Introduction
99(2)
4.2 Radar signal model and domain suitability
101(5)
4.3 Multilinear subspace learning
106(2)
4.3.1 Multilinear algebra basics and notations
106(1)
4.3.2 Multilinear principal component analysis (MPCA)
107(1)
4.3.3 Multilinear discriminant analysis
108(1)
4.4 Optimization considerations for multidimensional methods
108(6)
4.4.1 Iterative projections
108(1)
4.4.2 Initialization
109(1)
4.4.3 Termination criteria and convergence
110(1)
4.4.4 Number of projections
111(3)
4.5 Boosting the MPCA
114(1)
4.6 Experimental results
115(2)
4.7 Conclusion
117(1)
References
118(5)
5 Challenges in training DNNs for classification of radar micro-DoppIer signatures
123(40)
Sevgi Z. Gurbuz
Moeness G. Amin
Mehmet S. Seyfioglu
Baris Erol
5.1 Theory of training complex models
124(3)
5.1.1 Bias-variance trade-off
124(1)
5.1.2 Requirements on training sample support
125(1)
5.1.3 Machine learning theory versus practice: empirical studies
126(1)
5.2 Training with small amounts of real data
127(2)
5.2.1 Unsupervised pretraining
128(1)
5.3 Cross-frequency training using data from other radars
129(5)
5.4 Transfer learning using pretrained networks
134(2)
5.5 Training with synthetic data from kinematic models
136(5)
5.5.1 Modeling radar return with MOCAP data
136(1)
5.5.2 Diversification of synthetic micro-Doppler signatures
137(1)
5.5.3 Residual learning
138(3)
5.6 Training with synthetic data generated by adversarial neural networks
141(15)
5.6.1 Wasserstein GAN
144(1)
5.6.2 Conditional variational autoencoder
145(1)
5.6.3 Auxiliary conditional GAN
146(1)
5.6.4 Analysis of kinematic fidelity and diversity -'
147(2)
5.6.5 Boosting performance with PCA-based kinematic sifting
149(3)
5.6.6 Integrating kinematics into GAN architecture
152(4)
5.7 Conclusion
156(1)
References
156(7)
6 Machine learning techniques for SAR data augmentation
163(44)
Benjamin Lewis
Theresa Scarnati
Michael Levy
John Nehrbass
Edmund Zelnio
Elizabeth Sudkamp
6.1 Introduction
163(2)
6.2 Data generation and the SAMPLE dataset
165(11)
6.2.1 Electromagnetic data generation
166(5)
6.2.2 Model truthing
171(3)
6.2.3 Image formation
174(2)
6.3 Deep learning evaluation and baseline
176(4)
6.4 Addressing the synthetic/measurement gap with deep neural networks
180(22)
6.4.1 SAR data augmentation
180(1)
6.4.2 Mathematical despeckling
181(3)
6.4.3 Layer freezing and layerwise training
184(3)
6.4.4 Autoencoders
187(6)
6.4.5 Generative adversarial networks
193(3)
6.4.6 Siamese and triplet networks
196(6)
6.5 Conclusion
202(1)
Acknowledgments
202(1)
References
202(5)
Part III Applications
207(182)
7 Classifying micro-Doppler signatures using deep convolutional neural networks
209(34)
Youngwook Kim
7.1 Micro-Doppler and its representation
209(3)
7.1.1 Micro-Doppler theory
209(1)
7.1.2 Representing micro-Doppler signatures
210(1)
7.1.3 Conventional approaches to classify micro-Doppler signatures
211(1)
7.2 Micro-Doppler classification using a deep convolutional neural network
212(15)
7.2.1 Detecting and classifying human activities
213(2)
7.2.2 Using simulated data to classify human activities
215(2)
7.2.3 Classifying human gaits in multiple targets using DCNN
217(2)
7.2.4 Recognizing hand gestures
219(3)
7.2.5 Recognizing human voices
222(2)
7.2.6 Classifying drones using DCNN with merged Doppler images
224(3)
7.3 Classification of micro-Doppler signatures using transfer learning
227(10)
7.3.1 Classifying human aquatic activities using transfer learning
228(3)
7.3.2 Recognizing human voices using transfer learning
231(1)
7.3.3 Classifying moving targets using transfer learning with DCNN and micro-Doppler spectrogram
232(1)
7.3.4 Comparing transfer learning to data augmentation using generative adversarial networks
232(1)
7.3.5 GANs for augmentation of simulated human data
233(1)
7.3.6 Use of GANs for open-ended applications
234(1)
7.3.7 Human activity classification with the aid of GANs
235(2)
7.4 Conclusion
237(1)
References
238(5)
8 Deep neural network design for SAR/ISAR-based automatic target recognition
243(36)
Simon Wagner
Stefan Briiggenwirth
8.1 Introduction
243(2)
8.2 Deep learning methods used for target recognition
245(3)
8.3 Datasets
248(5)
8.3.1 MSTAR SAR dataset
248(1)
8.3.2 TIRA ISAR dataset
249(1)
8.3.3 Artificial training data
250(3)
8.4 Classification system
253(8)
8.4.1 CNN structure
254(2)
8.4.2 Training algorithm
256(3)
8.4.3 Regularization methods
259(2)
8.5 Experimental results
261(13)
8.5.1 CNN structure
261(5)
8.5.2 Training method and cost function
266(2)
8.5.3 Regularization methods
268(2)
8.5.4 Artificial training data
270(1)
8.5.5 Use of SVMs as classifier
271(3)
8.6 Summary and conclusion
274(1)
References
274(5)
9 Deep learning for passive synthetic aperture radar imaging
279(32)
Samia Kazemi
Eric Mason
Bariscan Yonel
Birsen Yazici
9.1 Introduction
279(3)
9.1.1 Passive radar
279(2)
9.1.2 DL and its application to SAR
281(1)
9.1.3 Organization of the chapter
282(1)
9.2 DL for inverse problem
282(2)
9.2.1 DL for forward modeling
283(1)
9.2.2 DL for inversion
284(1)
9.3 Problem statement?
284(1)
9.4 Bayesian and optimization-inspired DL for radar imaging
285(3)
9.4.1 Parameter learning via backpropagation
288(1)
9.5 Passive SAR imaging for unknown transmission waveform
288(2)
9.6 Passive SAR imaging for unknown transmission waveform and transmitter location
290(5)
9.6.1 Network design for imaging
291(2)
9.6.2 Network training and partial derivatives
293(2)
9.7 Numerical results
295(5)
9.8 Conclusion
300(1)
Acknowledgment
300(1)
Appendix
301(1)
A.1 Partial derivatives when transmission direction is known
301(2)
B.1 Partial derivatives when transmission direction is unknown
303(1)
References
304(7)
10 Fusion of deep representations in multistatic radar networks
311(44)
Jarez Satish Patel
Francesco Fioranelli
Matthew Ritchie
Hugh Griffiths
10.1 Introduction
311(2)
10.2 Experimental multistatic radar system
313(2)
10.2.1 Setup A
313(2)
10.2.2 Setup B
315(1)
10.3 Data fusion
315(19)
10.3.1 Significance in multistatic radar systems
315(1)
10.3.2 Data fusion methods
316(8)
10.3.3 Data fusion architectures
324(2)
10.3.4 Voting system implementations
326(2)
10.3.5 Weighted ensemble
328(1)
10.3.6 Match and rank level fusion implementations
329(4)
10.3.7 Summary
333(1)
10.4 Deep neural network implementations
334(7)
10.4.1 Spatially diverse human micro-Doppler
334(1)
10.4.2 Micro-drone payload classification
335(3)
10.4.3 Presence of jamming in human micro-Doppler
338(2)
10.4.4 Summary
340(1)
10.5 Jamming effects in multistatic radar
341(8)
10.5.1 Spectrogram degradation
341(1)
10.5.2 Jamming resilient systems
341(7)
10.5.3 Summary
348(1)
10.6 Conclusion
349(1)
References
350(5)
11 Application of deep learning to radar remote sensing
355(34)
John Rogers
Lucas Cagle
John E. Ball
Mehmet Kurum
Sevgi Z. Gurbuz
11.1 Open questions in DL for radar remote sensing
355(8)
11.1.1 What is the best way to exploit multimodal sensing data?
355(1)
11.1.2 How to satisfy the large data needs for training a DL system?
356(1)
11.1.3 How can prior knowledge on sensor quality be incorporated into joint domain models?
356(1)
11.1.4 What is the best way to leverage data, models, and prior knowledge?
357(1)
11.1.5 How can DL aide in solving multi-temporal processing challenges?
358(2)
11.1.6 How can the big data challenge presented by remote sensing data be addressed?
360(1)
11.1.7 How can DL be used to leverage nontraditional data sources?
361(1)
11.1.8 How can DL be used in automotive autonomy?
362(1)
11.2 Selected applications
363(11)
11.2.1 Land use and land cover (LULC) classification
364(1)
11.2.2 Change detection
365(2)
11.2.3 Ship detection
367(2)
11.2.4 Geophysical parameter estimation
369(2)
11.2.5 Radar aeroecology
371(3)
11.3 Additional resources
374(1)
11.4 Concluding remarks
375(1)
References
376(13)
Epilogue: looking toward the future 389(2)
Sevgi Zubeyde Gurbuz
Index 391
Sevgi Zubeyde Gurbuz is an assistant professor of electrical and computer engineering at the University of Alabama, USA. She received the SPIE Defense and Commercial Sensing Rising Researcher Award in 2020. Her research interests are in radar signal processing and machine learning for applications ranging from human activity and gait analysis for remote health monitoring in biomedical engineering to American Sign Language and gesture recognition for human-computer interaction, and multimodal remote sensing for earth sciences.