Scattering center guided mono-static radar cross section prediction

Radar cross section (RCS) is a fundamental electromagnetic property that characterizes how objects scatter incident radar signals. Recently, learning-based methods have shown great potential in near-field electromagnetic computation and RCS estimation. However, most existing approaches predominantly focus on capturing an object's global structure, often overlooking the impact of fine surface details. Since both global geometry and surface details influence RCS, this limitation hinders accurate modeling. To address this problem, we propose a novel method that incorporates scattering center theory to explicitly decouple geometric features into structural shape and fine surface detail representations, enhancing the network's ability to capture multi-scale information. Our network, Scattering Center guided RCS Network (SC-RNet), is specifically designed to extract and fuse shallow and deep features corresponding to local textures and overall structures, respectively. Experimental results demonstrate that SC-RNet provides improved modeling of the influence of local surface variations on RCS and achieves consistent and reliable prediction performance in both qualitative and quantitative evaluations. Furthermore, ablation studies confirm the effectiveness of each proposed module.