DSNUNet: An Improved Forest Change Detection Network by Combining Sentinel-1 and Sentinel-2 Images

The use of remote sensing images to detect forest changes is great significance for resource management. With the development and implementation deep learning algorithms in change detection, a large number models have been designed multi-phase images. Although synthetic aperture radar (SAR) data strong potential application detection tasks, most existing learning-based optical imagery. Therefore, effectively combine SAR this paper proposes double Siamese branch-based network called DSNUNet. DSNUNet uses two sets feature branches extract features from dual-phase employs shared weights into groups. In proposed DSNUNet, different extraction branch widths were used compensate difference amount information between was validated by experiments on manually annotated dataset. According obtained results, method outperformed other methods, achieving an F1-score 76.40%. addition, combinations width analyzed study. results revealed optimal performance model at initial channel numbers imaging image 32 8, respectively. prediction demonstrated effectiveness accurately predicting suppressing cloud interferences some extent.