Super-Resolution Neural Networks for High-Contrast Electromagnetic Scattering Problems

This letter proposes a super-resolution (SR) neural network model for high-contrast electromagnetic scattering problems. The model is designed to predict fine-grid field distributions based on low-cost coarse-grid simulations. By integrating a spatial channel attention mechanism, the model enhances accuracy in capturing field discontinuities induced by strong scatterers. Additionally, a residual-in-residual architecture is incorporated to provide the network with sufficient depth for effective correction of dispersion errors. The efficiency and accuracy of the proposed model have been validated through numerical experiments. Comparative evaluations with a recently proposed electromagnetic SR network, supplemented by rigorous ablation studies, further demonstrate the superior performance of our approach in high-contrast scenarios.