Push the Generalization Limitation of Learning Approaches by Multidomain Weight-Sharing for Full-Wave Inverse Scattering

Recently, deep-learning (DL) approaches have shown their advantages in solving scientific problems including full-wave nonlinear inverse problems. However, these data-driven methods face severe problems, especially with the low generalization ability, which means that the trained models only work for scenarios with similar training data and physical setups. In this work, we propose a multidomain weight-sharing method (MDWS) for inverse scattering problems (ISPs), which increases the generalization ability of learning approaches for both data- and physical-based out-of-range tests. Specifically, the proposed MDWS utilizes a physical layer of Green's function to transform between the induced current domain and the electrical field domain, where weight-sharing blocks having the same weights in different incidences and stages are used to decouple the network structure from measurement setups. It is shown by intensive numerical and experimental tests both qualitatively and quantitatively that the proposed MDWS apparently outperforms the benchmarked method. Furthermore, the proposed weight-sharing architecture also provides an efficient way to build a large model in an electromagnetic society with much less memory and computational cost.