MFDBANet: Multi Feature Dual Branch Attention Network for SAR Image Denoising

To address the prevalent speckle noise issue in Synthetic Aperture Radar (SAR) images, we propose a Multi Feature Dual Branch Attention Network (MFDBANet). The network primarily consists of two branches and four modules: upper branch, lower branch, Feature Enhancement Block, Residual Dual Attention Block, Channel Attention Block, and a Pixel Attention Block. The network utilizes a feature enhancement module to amplify the texture information of the input image. After obtaining the feature enhanced image, the upper branch extracts image features, which are then connected with the features extracted by the lower branch to obtain fused features. The extracted fused features are then mapped to the Residual Dual Attention Block to extract deep features from the SAR image. Finally, the denoised image is generated through global residual learning. Experimental results demonstrate that the proposed algorithm significantly suppresses speckle noise while effectively preserving edge and other detailed information in the image.