MFGB-S3W Denoiser: Multi-scale Fuzzy Granular-Ball and Sequential Three-Way Decision for Remote Sensing Image Denoising

Existing denoising algorithms for remote sensing images (RSIs) often suffer from high noise sensitivity, insufficient retention of local details, and loss of information due to non-discriminatory denoising. Therefore, this paper proposes a denoising algorithm based on multi-scale fuzzy granular ball (MFGB) and sequential three-way decision (S3WD), denoted as the MFGB-S3W denoiser. First, a noise outlier assessment method is designed based on fuzzy rough set theory to quantify the sample noise outlier score to distinguish the noisy image from the normal image; second, the MFGB space is constructed to capture the global structural correlation and the local detail differences of the image through the multi-granularity feature fusion. Finally, combined with the S3WD mechanism, the image is segmented layer by layer to realize the precise localization of the noisy image. Experimental results demonstrate that the proposed model yields superior performance in terms of \(mAP_{50}\) , PSNR, SSIM, and ERGAS on the NWPU VHR-10 and RSOD remote sensing datasets, outperforming the Blind2Unblind method without MFGB and S3WD fusion, as well as the DnCNN and BM3D approaches. The proposed method effectively suppresses noise while preserving noise-free images, offering a promising solution for denoising complex RSIs.