SAR Image Despeckling Based on U-shaped Transformer from a Single Noisy Image
Chushi Yu, Yoan Shin · 2022 13th International Conference on Information and Communication Technology Convergence (ICTC) · 2022
Remote sensing (RS) plays an important role in recent earth observation research and synthetic aperture radar (SAR) is widely applied as active remote sensing. Due to the coherent characteristics of radar illumination, the SAR images are polluted by speckle noise, which makes it difficult to understand and widely used. In this paper, we propose an improved SAR image despeckling approach based on a self-supervised strategy. The proposed method performs the transformer block and residual block as the main core of the architecture. We use the mean squared error with a regularization loss as the loss function. Experiments have been conducted on both simulated and real SAR datasets. The results show that the proposed method can preserve details and reduce smoothing better than several despeckling methods. The speckle noises were suppressed significantly, and the information of reconstructed images were well maintained.