A SAR image self-supervised denoising method and system based on blind spot network

Abstract

The application provides a SAR image self-supervision denoising method and system based on a blind spot network, wherein the method comprises the following steps: acquiring a SAR image, obtaining a homomorphic noise image according to the SAR image; constructing a mask mapper, obtaining a denoising mask image and a denoising SAR image according to the homomorphic noise image and the mask mapper; constructing a global mapper, obtaining a denoising blind spot image according to the denoising mask image and the global mapper; constructing a multi-task reproduction loss function, repeatedly performing loss value calculation until the loss function converges, and obtaining a target denoising network; acquiring a SAR image to be denoised, inputting the SAR image to be denoised into the target denoising network, and obtaining a denoising image. The application does not need to train the denoising network under the condition of a clean SAR image, can be migrated to an existing denoising network, ensures that each pixel point of the image is fully utilized, avoids information loss, better retains image texture, edge and other information, and solves the problems that a current network cannot be trained and the denoising effect is poor.

Images