In this letter, to break the limit of the traditional linear models for SARimage despeckling, we propose a novel deep learning approach by learning anon-linear end-to-end mapping between the noisy and clean SAR images with adilated residual network (SAR-DRN). SAR-DRN is based on dilated convolutions,which can both enlarge the receptive field and maintain the filter size andlayer depth with a lightweight structure. In addition, skip connections areadded to the despeckling model to reduce the vanishing gradient problem.Compared with the traditional despeckling methods, the proposed method showssuperior performance over the state-of-the-art methods on both quantitative andvisual assessments, especially for strong speckle noise.
展开▼
