Research on Change Detection Algorithms for Image Interaction and Edge Enhancement

Most of the existing detection methods ignore the spatial and temporal characteristics of the two images and global dependence of the interconnections, resulting in many irrelevant changes (such as illumination, the atmosphere, seasonal changes in vegetation) to produce the phenomenon of misdetection; at the same time, the existing methods of change detection. Meanwhile, the existing change detection methods are not easy to obtain structural details in high-resolution remote sensing images and the detection accuracy needs to be improved. Based on the global information interaction and edge detection method of dual-time-phase image, a network model is proposed, which consists of three parts and is divided into three modules, namely, edge structure and image feature extraction, image transformer and change region discrimination. Edge detection is performed on the dual-temporal phase image to obtain the edge map, edge detection can effectively improve the edge details of the detection results, the edge map and the original image are fused for feature extraction, and the extracted dual-temporal phase features are sent to the image transformer BIT for global spatial-temporal modeling to establish the connection, modeling the remote context in the dual-temporal image in an efficient and effective way, and finally through the simple region of change discriminative module for discrimination obtaining the detection results. The algorithm achieves 96.2% F1 on CDD dataset and 93.8% F1 on BCDD dataset, which is superior with better stability and robustness compared with some known algorithms.