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

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.