A New Approach for Solving Inverse Scattering Problems Based on Physics-informed Supervised Residual Learning

In this paper, we propose a new approach for solving inverse scattering problems (ISPs) by applying the physics-informed supervised residual learning (PhiSRL) to embody the Born iterative method (BIM). Stemming from the mathematical link between the fixed-point iteration method and residual neural network (ResNet), PhiSRL fulfills the alternate iteration process of BIM by predicting the modifications of the candidate solutions regarding the calculated residuals. Thus, the proposed approach can perform the inversions of both data and models at the same time. The effectiveness of the proposed approach is further validated by synthetic data. This paper provides new insights for designing the deep learning (DL) based methods with the knowledge of traditional computational electromagnetic (EM) algorithms.