EMINet-CD: An Edge-Enhanced Mixed Information Interaction Network for Remote Sensing Image Change Detection

Change detection (CD) in remote sensing images (RSIs) is vital for disaster assessment, urban planning, and cropland monitoring. Deep learning methods for automatic discriminative feature extraction have achieved remarkable progress. However, current accuracy is limited by small change targets, blurry boundaries, and illumination/shadow interference. To address these challenges, we propose an edge-enhanced mixed information interaction network. We use a VGG-SwiftFormer backbone to extract hierarchical local-global features and a gated local-global fusion module (GLGFM) to balance them via a gating mechanism. The dual-branch edge difference module for edge information extraction and utilization enhances edge information and suppresses nonedge interference. The GLGFM and the multistage channel-space interaction module are used for two-stage multiscale feature interaction, respectively, performing upsampling and downsampling decoding, to fully align features at different scales. Extensive experiments on four public datasets demonstrate that our method surpasses 12 established CD approaches in both quantitative metrics and visual quality.