Synthetic aperture radar image despeckling with a residual learning of convolutional neural network

Abstract SAR images despeckling based on deep learning attracted more researchers’ attention because of its performance and computational efficiency. In order to embody recent improvements for SAR images despeckling models, we proposed a model base on deep learning, called Synthetic Aperture Radar (SAR) image despeckling with Convolutional Neural Networks (SID-CNN), which is based on residual learning and discriminative training. In this model, the multiplicative noise in the SAR image was removed by residual learning and discriminative training in the form of additive noise. Whilst we illustrated the capability of the proposed SID-CNN model on synthetic images and real SAR images, respectively. Finally, compared with the-state-of-art SAR image despeckling methods, extensive experiments demonstrated that the proposed model had a better capacity to remove SAR images speckle with high restoration quality and computational efficiency.