3D Targets Reconstruction from SAR Images Via 2D Gaussian Splatting

Reconstructing precise 3D target structures from Synthetic Aperture Radar (SAR) images is challenging due to limited viewing angles and complex scattering mechanisms. To address these challenges, we propose 2DGS-SAR, a novel reconstruction method that utilizes 2D Gaussians for scene representation. We construct a rendering pipeline consistent with SAR imaging mechanisms, encompassing a complete forward rendering and backward optimization process. Our method explicitly models the SAR imaging geometry and optimizes the electromagnetic scattering properties of 2D Gaussian surfels via differentiable rendering. To ensure geometric fidelity, we introduce a depth distortion loss to cluster primitives along rays and a geometric constraint loss to suppress artifacts and enforce scene boundaries. Experimental results on both simulated and measured datasets demonstrate that 2DGS-SAR achieves superior reconstruction accuracy and robustness compared to existing methods, effectively recovering detailed 3D structures under sparse view conditions.