Physics-Driven Electromagnetic Inverse Scattering With Limited-Aperture Measurements

Le résumé fourni par la source

Electromagnetic inverse scattering problems (ISPs) in engineering applications are typically constructed based on limited-aperture measurement data, which further exacerbates the inherent nonlinearity and ill-posedness. To alleviate these issues, this article proposes a novel inverse scattering network architecture, the limited aperture network (LA-Net), which can be flexibly applied to arbitrary apertures (90°–360°) without retraining. The proposed LA-Net integrates two core operators, the wavelet coefficient learning operator (WCLO) and the induced current reconstruction operator (ICRO). Specifically, WCLO is defined as a wavelet-compressed module that compresses the input deterministic current components dimension from 2N$\times 2$N to N$\times $N. This operator significantly reduced computational burden. ICRO is an unrolling-network module to predict the final accurate induced current from a preliminary restored induced current by WCLO. Finally, a physics-informed loss function is used to constrain the model. The dataset contains scatterers illuminated by varying numbers of antennas, while the network could independently process data from a single incidence. This architecture enables the proposed LA-Net to handle arbitrarily limited apertures without requiring additional training. Under the joint constraints of WCLO and ICRO with multiphysics loss functions, the unrolling network is capable of reconstructing high-quality images of scatterers even with a small random number of antennas.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.