A physical Scattering-Guided Prior-Enhanced Mamba model for high-fidelity SAR ship generation in complex backgrounds

Synthetic Aperture Radar (SAR) image synthesis remains challenging in Earth observation (EO) due to unique speckle noise and the high dynamic range of backscatter coefficients. Existing models typically rely on direct pixel-wise synthesis, ignoring physical scattering mechanisms. This causes textural distortions that hinder downstream geospatial interpretation. To address these bottlenecks, we propose a physically-guided, coarse-to-fine generation framework, Scattering-Guided Prior-Enhanced Mamba (SGPMamba). Specifically, a Scattering-Guided Prior (SGP) model is introduced to transform multiplicative speckle into an additive formulation, constraining the network to generate physically reliable textures. We then design a Mamba refinement architecture for efficient long-range spatial modeling, which efficiently refines the SGP-generated priors through long-range spatial modeling. Additionally, we introduce the Clutter-Aware Attention Gate (CAG) and Residual Refinement Head (RRH) in the Mamba refinement architecture, which can restore the complex textures and high-frequency scattering details. Experiments on SSDD and HRSID datasets provide systematic quantitative and qualitative evidence for the effectiveness of the proposed framework. These results suggest that incorporating scattering-aware physical priors into state-space refinement can provide a useful methodological direction for physically consistent SAR image synthesis. An interactive GUI is also provided to illustrate the controllable synthesis process. Video: https://youtu.be/V31SkO271Ho . Code and datasets: https://github.com/Pluto-cc/SGPMamba .