Physics-Driven Neural Network for Solving Electromagnetic Inverse Scattering Problems

In recent years, deep learning-based methods have been introduced for solving inverse scattering problems (ISPs), but most of them heavily rely on large training datasets and suffer from limited generalization capability. In this paper, a new solving scheme is proposed where the solution is iteratively updated through a physics-driven neural network (PDNN), the hyperparameters of which are optimized by minimizing the loss function which incorporates the constraints from measured scattered fields and the prior information about scatterers. Unlike data-driven neural network solvers, PDNN does not require an offline training dataset. Its weights are iteratively updated based solely on the measured incident and scattered fields, similar in philosophy to conventional inverse algorithms but enhanced by neural-network flexibility. Thus, the generalization issue is eliminated. Moreover, to accelerate the imaging, a subregion enclosing the scatterers is automatically identified and used to restrict the computational domain. Numerical and experimental results demonstrate that the proposed scheme achieves high reconstruction accuracy and strong stability, even for complex and lossy scatterers.