Abstract:Electromagnetic inverse scattering is a challenging problem in many areas of science and engineering, including radar imaging, medical imaging, and non-destructive testing. The goal of inverse scattering is to recover the properties of an object from the scattered electromagnetic waves that are generated when the object is illuminated with incident waves. The inverse scattering problem is inherently difficult because the properties of the object cannot be measured directly, and only the scattered waves can be observed. In recent years, convolutional neural networks (CNNs) have shown great promise in solving inverse scattering problems. The U-Net model is a popular CNN architecture that has been used to solve a wide range of image processing and recognition tasks. However, the U-Net model has limitations in dealing with complex inverse scattering problems due to the limited information available in the scattered wave data. To address this limitation, we propose an improved U-Net model called M-Net, which incorporates multi-scale features and a mean output layer to improve the accuracy and stability of the reconstruction. The M-Net model consists of a multi-scale input layer, a U-shape convolutional neural network, and a multi-scale mean output layer. Direct prediction methods take scattering field data as network input, which can greatly reduce the manual calculation workload, but this method does not make full use of known physical a priori information, resulting in a waste of computing resources. Therefore, we use diffraction tomography (DT) images based on Born approximation as the network input, which can ensure imaging accuracy and improve computational efficiency. In order to verify the effectiveness of the proposed method, a simulation experiment is carried out with a target medium as the reconstruction target. The results show that the M-Net model combined with the tomographic diffraction algorithm is superior to the U-Net model and other existing direct-solving methods in terms of accuracy and efficiency in solving the electromagnetic inverse scattering problems. The error analysis further proves the superior performance of the M-Net model combined with the tomographic diffraction algorithm in the complex inverse scattering problem.

Improved Electromagnetic Inverse Scattering with M-Net Model Incorporating Diffraction Tomography

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

An Electromagnetic Inverse Scattering Imaging Method Based on U-net Structure

A Novel Electromagnetic Reconstruction Algorithm for Dielectric Objects Using Neural Networks

Deep Learning with Attention Mechanism for Electromagnetic Inverse Scattering

A Reconstruction Method of Electromagnetic Scattering Target Based on Diffusion Model

Nonlinear Electromagnetic Inverse Scattering Using DeepNIS

Convolutional Neural Networks for Multifrequency Electromagnetic Inverse Problems

MB-DECTNet: A Model-Based Unrolled Network for Accurate 3D DECT Reconstruction

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

On an artificial neural network for inverse scattering problems

Hybrid Dilated Convolutional Neural Network for Solving Electromagnetic Inverse Scattering Problems

Physics Embedded Deep Neural Network for Solving Full-Wave Inverse Scattering Problems

Multi-frequency Neural Born Iterative Method for Solving 2-D Inverse Scattering Problems

Two-Step Contrast Source Learning Method for Electromagnetic Inverse Scattering Problems

DeepNIS: Deep Neural Network for Nonlinear Electromagnetic Inverse Scattering

Two Steps Electromagnetic Quantitative Inversion Imaging Based on Convolutional Neural Network

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

Multi-Frequency Progressive Refinement for Learned Inverse Scattering

A More General Electromagnetic Inverse Scattering Method Based on Physics-Informed Neural Network

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