Research on Deep Learning-Based SAR Image Denoising Algorithm

Synthetic Aperture Radar (SAR) has the advantages of high penetration, high resolution, all-weather, and all-time, but due to the special imaging mechanism of SAR images, there are a large number of coherent speckle noises in the images. If not processed, it will greatly limit the subsequent use of data. In order to improve the use of SAR images, noise removal processing is needed. In order to choose the appropriate noise removal algorithm, this paper compares the performance of three noise removal algorithms: multiplication model, enhanced Lee-sigma filter, and CNN trained on SAR images (SAR-CNN). The results show that the multiplication model has better noise removal ability and ability to extract effective information.