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

This paper presents an inversion model based on deep learning to solve electromagnetic inverse scattering (EMIS) problems. The traditional methods have poor accuracy and high computational cost in solving EMIS. In order to solve these issues, a high-precision imaging model is proposed based on U-net structure. This model includes two parts: the down-sampling area and the up-sampling area. The up-sampling area first extracts the corresponding feature fragments from the measured scattering field, and compresses these features. Then, these features are retrieved and expanded by the up-sampling area. Finally, the dielectric parameters of unknown scatterers are reconstructed. The numerical test results show that this method is effective and feasible for solving EMIS problems.