Adaptive Edge-Guided Fusion Network for Remote Sensing Image Change Detection

Change detection in remote sensing images has become increasingly important in land resource management and urban ecological development. Though existing change detection networks deepen network depth to enhance model representation capabilities, they result in insufficient utilization of shallow-layer features and limited extraction of edge information. To address this, we propose an Adaptive Edge-Guided Fusion Change Detection Network (AEGFNet). To fully leverage shallow-layer features, the designed Edge-Guided Branch (EGB) first filters edge features via wavelet transformation, then further processes feature interactions to obtain rich edge information. Finally, this edge information is adaptively fused into the main features to improve the network's detection performance. Beyond this edge-guided optimization, a cross-scale interaction module (CSIB) based on Transformers is designed to fuse multi-scale features with different semantics, which further enhances the remote sensing change detection network's ability to perceive differences in dual-temporal images. A hierarchical fusion module (HFM) is additionally proposed in the decoding stage, which progressively fuses decoding features from the previous layer to retain more critical deep-layer information. Experiments conducted on representative remote sensing datasets such as LEVIR, WHU-CD, and SYSU-CD demonstrate that the proposed method outperforms existing state-of-the-art change detection methods in terms of detection performance. The code is available athttps://github.com/HyperSystemAndImageProc/AEGFNet.