MDGF-CD: Land-cover change detection with multi-level DiffFormer feature grouping fusion for VHR remote sensing images

Abstract

Recently, Transformer has become a popular tool for change detection (CD) in remote sensing images due to its ability to model global information. However, the existing Transformer models lack specific designs tailored for change information and do not adequately consider the fusion of features at different levels, thereby impeding their ability to distinguish the changes of objects of interest in complex backgrounds. To tackle these issues, we propose a multi-level DiffFormer feature grouping fusion network (MDGF-CD). Specifically, the proposed MDGF-CD develops a DiffFormer structure to form the backbone for extracting multi-level features from the original input image pairs. DiffFormer better matches the CD task by considering the change and global information in the attention mechanism. Then, we define a novel multi-level feature grouping fusion method to effectively integrate features across different levels, which adopts a grouping pattern to fully fuse low-level spatial details and high-level abstract semantics. In addition, a spatial change-aware module is designed to preserve low-level spatial details for better detecting subtle changes. Experimental results on the LEVIR-CD, WHU-CD and CLCD datasets demonstrate that the proposed MDGF-CD outperforms the existing state-of-the-art methods.