SAR Image Denoising Based on Denoising Diffusion Probabilistic Models

Synthetic Aperture Radar (SAR) images are widely applied in remote sensing research; however, their quality is often severely affected by noise interference during the imaging process. Speckle noise represents the most common noise type in SAR images, characterized by noise patterns similar to image content and manifested as random grayscale variations, which significantly impacts image analysis and application. Denoising Diffusion Probabilistic Models (DDPM) have become a research focus due to their exceptional image generation capabilities and multi-task robustness. This paper proposes a SAR image denoising method based on the DDPM model, which trains deep models by adding noise in the forward process and utilizing the backward process for recovery. Experimental results demonstrate that this method effectively removes speckle noise from SAR images while preserving edge details and enhancing the quality of denoised images.