An Edge Feature Extraction Method for Remote Sensing Image Edge Based on Generative Adversarial Strategy

Affected by changes in illumination, weather, and surface activities, multi-temporal remote sensing image data changes slowly all the time. How to extract edge feature is a key common problem that needs to be resolved in the field of remote sensing image intelligent processing. This paper proposes an edge feature extraction method for remote sensing image edges based on a generative confrontation strategy. This method first designs a CycleGAN network based on the Smooth L1 loss function to meet the requirements of robust feature extraction for edges with semantic information in remote sensing images. The network transfers the style of remote sensing images, and finally extracts edge feature for the images after the style transfer, which is used as the description of the robust features on the surface. Experimental results show that the edge extraction model designed in this paper improves the accuracy of robust feature extraction of remote sensing images by 8.1%. In addition, the method is less affected by environmental factors such as illumination, and can realize the intelligent extraction of robust features of the remote sensing image. The relevant code of this article has been published on the research group's home page.