Abstract: Due to the influence of scattering characteristics and imaging geometry, single-polarization SAR images suffer from severe speckle noise and significant image degradation, which makes them more prone to texture loss and structural distortion in super-resolution reconstruction. To address this problem, a diffusion-based super-resolution method that integrates frequency-domain processing and structure awareness is proposed. The method uses a latent diffusion model as the backbone, and introduces a wavelet-conditioned guidance module and a direction-aware enhancement module to improve performance.The wavelet-conditioned guidance module performs multi-scale modulation on high-frequency sub-bands through spatially adaptive normalization, thereby dynamically enhancing texture representation and high-frequency reconstruction capability according to the diffusion timestep. The direction-aware enhancement module incorporates multiple adaptive convolution kernels and is embedded into the deep residual structures of the encoder to strengthen the sensitivity of the compressed feature maps to structural information. In the experiments, a realistic degradation model combining speckle noise and blur kernels is also constructed to better approximate the real imaging chain.Experimental results show that the proposed method achieves significant improvements over existing frameworks on multiple datasets, with an average improvement of 8.12% compared with the best competing method, demonstrating its effectiveness and advancement in SAR image super-resolution.

Key words: SAR image super-resolution, remote sensing, diffusion models, wavelet decomposition, channel attention