PolSAR Super-Resolution by Fusion of LR-PolSAR and HR-SinSAR Based on Diffusion Model

Fully polarimetric synthetic aperture radar (PolSAR) provides rich scattering information but is often limited by low spatial resolution. To address this problem, we propose a novel diffusion model-based fusion framework to reconstruct high-resolution PolSAR (HR-PolSAR) images. Our method extends the image restoration stochastic differential equation (IR-SDE) framework by incorporating both low-resolution PolSAR (LR-PolSAR) and co-registered high-resolution single-polarization SAR (HR-SinSAR) as inputs. A dual-source architecture with cross-attention mechanism is introduced to effectively fuse spatial and polarimetric features. Additionally, we employ a training-free hybrid guided sampling strategy, which simultaneously enforces structural consistency derived from HR-SinSAR and maintains polarimetric fidelity from LR-PolSAR. We conduct extensive experiments on both simulated and real-world datasets of vehicles. The experimental results show that our approach achieves state-of-the-art performance, especially in recovering fine structural details of man-made targets and preserving polarimetric characteristics.