Dilated Residual Shrinkage Network for SAR Image Despeckling
Lin Nie, Chen Gao, Qingfeng Zhou, Chanzi Liu · 2021 IEEE 6th International Conference on Signal and Image Processing (ICSIP) · 2021
This paper proposes a dilated residual shrinkage network for SAR image despeckling (SAR-DRSN). The proposed SAR-DRSN consists of one main network and one shrinkage subnetwork. The main network is based on the dilated convolution and residual learning, and combined with the soft thresholding. The soft thresholding and shrinkage subnetwork are embedded in our network. In our SAR-DRSN, different thresholds are set for each feature adaptively by the shrinkage network. The feature map is calculated by the dilated convolution and shrinkage network, which can enhance the information of useful feature channels and make the target information more important. Experimental results show that, compare with the existing deep learning network for SAR image despeckling, the proposed SAR-DRSN has better despeckling performance according to both subjective visual assessment of image quality and objective evaluation.