electromagnetics inverse-scattering using physics-informed generative adverserial neworks

Electromagnetic inverse scattering problems (isps) aim to reconstruct unknown scatterer properties from scattered field data but face challenges from nonlinearity and ill-posedness. traditional methods suffer from either poor accuracy (non-iterative) or high computational cost (iterative), while existing deep learning approaches lack generalization or incorporate physics only superficially. this paper proposes a physics-informed generative adversarial network (pigan) for full-wave isps. the method uses a generator to refine coarse back-propagation images into realistic scatterer profiles, guided by a discriminator. key innovations include an enhanced loss function combining pixel-wise and perceptual adversarial losses for multi-level feature matching, and integration of additional physical quantities as network inputs. testing on synthetic data demonstrates that pigan significantly outperforms conventional methods in reconstruction accuracy and computational efficiency, validating its effectiveness for electromagnetic inverse scattering applications.