Abstract

This paper addresses the challenge of diminished accuracy in electromagnetic inverse scattering (EMIS) when both dielectric and plasma coexist within a domain of interest (DOI). It proposes the Electromagnetic Inverse Scattering Network (EIS-Net) based on deep learning. The model integrates attention mechanisms and residual structures, enabling accurate reconstruction of physical parameters of scatterers in scenarios with limited training samples or the presence of diverse types of scatterers within the region. Finally, simulation results substantiate the satisfactory performance of using EIS-Net for the inverse scattering of mixed targets.