Applications of Deep Learning in Electromagnetics: Teaching Maxwell's equations to machines

Description

Deep learning has started to be applied to solving many electromagnetic problems, including the development of fast modelling solvers, accurate imaging algorithms, efficient design tools for antennas, as well as tools for wireless links/channels characterization. The contents of this book represent pioneer applications of deep learning techniques to electromagnetic engineering, where physical principles described by the Maxwell's equations dominate. With the development of deep learning techniques, improvement in learning capacity and generalization ability may allow machines to "learn" from properly collected data and "master" the physical laws in certain controlled boundary conditions. In the long run, a hybridization of fundamental physical principles with knowledge from training data could unleash numerous possibilities in electromagnetic theory and engineering that used to be impossible due to the limit of data information and ability of computation. Electromagnetic applications of deep learning covered in the book include electromagnetic forward modeling, free-space inverse scattering, non-destructive testing and evaluation, subsurface imaging, biomedical imaging, direction of arrival estimation, remote sensing, digital satellite communications, imaging and gesture recognition, metamaterials and metasurfaces design, as well as microwave circuit modeling. Applications of Deep Learning in Electromagnetics contains valuable information for researchers looking for new tools to solve Maxwell's equations, students of electromagnetic theory, and researchers in the field of deep learning with an interest in novel applications. Cover Contents About the editors Foreword Acknowledgment 1 An introduction to deep learning for electromagnetics 1.1 Introduction 1.2 Basic concepts and taxonomy 1.2.1 What is deep learning? 1.2.2 Classification of deep learning techniques 1.3 Popular DL architectures 1.3.1 Convolutional neural networks 1.3.2 Recurrent neural networks 1.3.3 Generative adversarial networks 1.3.4 Autoencoders 1.4 Conclusions Acknowledgments References 2 Deep learning techniques for electromagnetic forward modeling 2.1 Introduction 2.2 DL and ordinary/partial differential equations 2.3 Fully data-driven forward modeling 2.4 DL-assisted forward modeling 2.5 Physics-inspired forward modeling 2.6 Summary and outlook References 3 Deep learning techniques for free-space inverse scattering 3.1 Inverse scattering challenges 3.2 Traditional approaches 3.2.1 Traditional approximate solutions 3.2.2 Traditional iterative methods 3.3 Artificial neural networks applied to inverse scattering 3.4 Shallow network architectures 3.5 Black-box approaches 3.5.1 Approaches for phaseless data 3.5.2 Application in electrical impedance and capacitance tomography 3.6 Learning-augmented iterative methods 3.7 Non-iterative learning methods 3.8 Closing remarks References 4 Deep learning techniques for non-destructive testing and evaluation 4.1 Introduction 4.2 Principles of electromagnetic NDT&E modeling 4.2.1 Field solution for the flawless piece and calculationof the signal geometry Z ( p) TR 4.2.2 Defect response: calculation of the flaw signal Z ( d) TR 4.2.3 Examples 4.2.4 Inverse problems by means of optimization and machine learning approaches 4.3 Applications of deep learning approaches for forward and inverse problems in NDT&E 4.3.1 Most relevant deep learning architecture in NDT&E 4.4 Application of deep learning to electromagnetic NDT&E 4.4.1 Deep learning in electromagnetic NDT&E applied to the energy sector 4.4.2 Applications to the transportation and civil infrastructures sectors 4.4.3 Applications to the manufacturing and agri-food sectors 4.5 Applications to higher frequency NDT&E methods 4.5.1 Infrared thermography testing and terahertz wave testing 4.5.2 Radiographic testing 4.6 Future trends and open issues for deep learning algorithms as applied to electromagnetic NDT&E 4.7 Conclusion and remarks 4.8 Acknowledgments References 5 Deep learning techniques for subsurface imaging 5.1 Introduction 5.2 Purely data-driven approach 5.2.1 Convolutional neural network 5.2.2 Recurrent neural network 5.2.3 Generative adversarial network 5.3 Physics embedded data-driven approach 5.3.1 Supervised descent method 5.3.2 Physics embedded deep neural network 5.4 Learning-assisted physics-driven approach 5.5 Deep learning in seismic data inversion 5.5.1 Inversion with unsupervised RNN 5.5.2 Low-frequency data prediction 5.5.3 Physically realistic dataset construction 5.5.4 Learning the optimization 5.5.5 Deep learning constrained traveltime tomography 5.6 Deep learning in multi-physics joint inversion 5.7 Construction of the training dataset 5.8 Conclusions and outlooks References 6 Deep learning techniques for biomedical imaging 6.1 Introduction 6.2 Physics of medical imaging 6.2.1 Maxwell's equations 6.2.2 Formulations of EIT 6.2.3 Formulations of MWI 6.2.4 Inverse methods for EIT and MWI 6.3 Deep-learning in medical imaging 6.3.1 Machine learning 6.3.2 Deep learning neural networks 6.3.3 DNN in medical imaging 6.4 Hybrid physics-based learning-assisted medical imaging: example studies 6.4.1 Example 1: EIT-based SDL-assisted imaging 6.4.2 Example 2: MWI(CSI)-based UNet-assisted imaging 6.4.3 Example 3: MWI(BIM)-based CNN-assisted imaging 6.5 Summary References 7 Deep learning techniques for direction of arrival estimation 7.1 Introduction 7.2 Problem formulation 7.2.1 Conventional observation model 7.2.2 Overcomplete formulation of array outputs 7.2.3 Array imperfections 7.3 Deep learning framework for DOA estimation 7.3.1 Data pre-processing 7.3.2 Deep learning model 7.3.3 Post-processing for DOA refinement 7.4 A hybrid DNN architecture for DOA estimation 7.4.1 The hierarchical DNN structure 7.4.2 Training strategy of the hybrid DNN model 7.4.3 Simulations and analyses 7.5 Concluding remarks and future trends References 8 Deep learning techniques for remote sensing 8.1 Target recognition 8.1.1 Ship detection 8.1.2 Aircraft recognition 8.1.3 Footprint extraction 8.1.4 Few-shot recognition of SAR targets 8.2 Land use and land classification 8.2.1 Local climate zone classification 8.2.2 Crop-type classification 8.2.3 SAR-optical fusion for land segmentation 8.3 Disaster monitoring 8.3.1 Flood detection 8.3.2 Storm nowcasting 8.3.3 Lightning nowcasting 8.4 Forest applications 8.4.1 Tree species classification 8.4.2 Deforestation mapping 8.4.3 Fire monitoring 8.5 Conclusions References 9 Deep learning techniques for digital satellite communications 9.1 Introduction 9.2 Machine learning for SatCom 9.2.1 Deep learning 9.3 Digital satellite communication systems 9.3.1 Uplink segment 9.3.2 Space segment 9.3.3 Downlink segment 9.4 SatCom systems modelling 9.4.1 High-power amplifier modelling 9.5 SNR estimation 9.5.1 Autoencoders 9.5.2 SNR estimation methodology 9.5.3 Metrics 9.5.4 Application example 9.5.5 Metrics tuning and consistency analysis 9.5.6 Results and discussion 9.6 Input back-off estimation 9.6.1 Deep learning model for IBO estimation 9.6.2 Performance metric 9.6.3 Data generation 9.6.4 Results and discussion 9.7 Conclusion References 10 Deep learning techniques for imaging and gesture recognition 10.1 Introduction 10.2 Design of reprogrammable metasurface 10.3 Intelligent metasurface imager 10.3.1 System configuration 10.3.2 Results 10.4 VAE-based intelligent integrated metasurface sensor 10.4.1 System configuration 10.4.2 Variational auto-encoder (VAE) principle 10.4.3 Results 10.5 Free-energy-based intelligent integrated metasurface sensor 10.5.1 System configuration 10.5.2 Free-energy minimization principle 10.5.3 Results References 11 Deep learning techniques for metamaterials and metasurfaces design 11.1 Introduction 11.2 Discriminative learning approach 11.3 Generative learning approach 11.4 Reinforcement learning approach 11.5 Deep learning and optimization hybrid approach 11.6 Summary References 12 Deep learning techniques for microwave circuit modeling 12.1 Introduction 12.2 Feedforward deep neural network for microwave circuit modeling 12.2.1 Feedforward deep neural network and the vanishing gradient problem 12.2.2 A hybrid feedforward deep neural network 12.3 Recurrent neural networks for microwave circuit modeling 12.3.1 Global-feedback recurrent neural network 12.3.2 Adjoint recurrent neural network 12.3.3 Global-feedback deep recurrent neural network 12.3.4 Local-feedback deep recurrent neural network 12.3.5 Long short-term memory neural network 12.4 Application examples of deep neural network for microwave modeling 12.4.1 High-dimensional parameter-extraction modeling using the hybrid feedforward deep neural network 12.4.2 Macromodeling of audio amplifier using long short-term memory neural network 12.5 Discussion 12.6 Conclusion References 13 Concluding remarks, open challenges, and future trends 13.1 Introduction 13.2 Pros and cons of DL 13.2.1 High computational efficiency and accuracy 13.2.2 Bypassing feature engineering 13.2.3 Large amounts of training data 13.2.4 High computational burden 13.2.5 Deep architectures, not learning 13.2.6 Lack of transparency 13.3 Open challenges 13.3.1 The need for less data and higher efficiency 13.3.2 Handling data outside the training distribution 13.3.3 Improving flexibility and enabling multi-tasking 13.3.4 Counteracting over-fitting 13.4 Future trends 13.4.1 Few shot, one shot, and zero shot learning 13.4.2 Foundation models 13.4.3 Attention schemes and transformers 13.4.4 Deep symbolic reinforcement learning 13.5 Conclusions References Index Back Cover