Abstract:The electromagnetic inverse scattering problem is nonlinear and ill-posed. Traditional solving methods often face the double pressures of inversion accuracy and calculation efficiency. In recent years, with the successful application of machine learning technology in various engineering fields, electromagnetic imaging technologies assisted by artificial neural networks are expected to break through the bottleneck of traditional inversion algorithms and make large-scale real-time imaging possible. When using neural networks to solve electromagnetic inverse scattering problems, the most direct way is to use scattering data as the network input and unknown scattering targets as the output. This type of direct method requires the lowest labor cost, and the user’s participation is low. However, the training of such neural networks is difficult, and the performance may not be satisfying. In fact, such neural networks are inefficient because known information including physical formulas, imaging model, and other a-priori knowledge is not utilized. And they would have to spend unnecessary effort on learning these known information. Based on this, a novel approach is proposed, which replaces the original scattering data with the results of diffraction tomography as the input of the neural network. This procedure could greatly reduce the learning difficulty of the neural network. Diffraction tomography is excellent in real-time imaging problems by virtue of the fast Fourier transform. Due to the limitation of the Born approximation, it often suffers from low reconstruction accuracy issue. The combination of diffraction tomography and neural networks is able to better utilize the advantages of traditional inversion algorithms and machine learning techniques. Numerical simulations are included to demonstrate the feasibility and efficiency of the new method.

Enhanced Deep Learning Approach Based on the Deep Convolutional Encoder–Decoder Architecture for Electromagnetic Inverse Scattering Problems

Solving Electromagnetic Inverse Scattering Problems in Inhomogeneous Media by Deep Convolutional Encoder–Decoder Structure

Enhanced Two-Step Deep-Learning Approach for Electromagnetic-Inverse-Scattering Problems: Frequency Extrapolation and Scatterer Reconstruction

Enhanced Supervised Descent Learning Technique for Electromagnetic Inverse Scattering Problems by the Deep Convolutional Neural Networks

An End-to-End Neural Network for Complex Electromagnetic Simulations

Deep Learning Electromagnetic Inversion Solver Based on Two-Step Framework for High-Contrast and Heterogeneous Scatterers

A Physics-Induced Deep Learning Scheme for Electromagnetic Inverse Scattering

Deep Learning: A Powerful Framework for the Real-Time Solution of Inverse Scattering Problems

Multi-model fusion approach for electromagnetic inverse scattering problems

Towards Intelligent Electromagnetic Inverse Scattering Using Deep Learning Techniques and Information Metasurfaces

Electromagnetic Inverse Scattering of Mixed Targets Based on Deep Learning

DeepNIS: Deep Neural Network for Nonlinear Electromagnetic Inverse Scattering

Deep Learning with Attention Mechanism for Electromagnetic Inverse Scattering

Two-Dimensional Electromagnetic Solver Based on Deep Learning Technique

Complex-Valued Deep Convolutional Networks for Nonlinear Electromagnetic Inverse Scattering

Performance Analysis and Dynamic Evolution of Deep Convolutional Neural Network for Electromagnetic Inverse Scattering

An Improved Deep Learning Scheme for Solving 2-D and 3-D Inverse Scattering Problems

An Early Fusion Deep Learning Framework for Solving Electromagnetic Inverse Scattering Problems

Sophisticated Electromagnetic Scattering Solver Based on Deep Learning

A Novel Electromagnetic Reconstruction Algorithm for Dielectric Objects Using Neural Networks

A Reconstruction Method of Electromagnetic Scattering Target Based on Diffusion Model