Change Feature Refinement and Cross-Scale Interaction Network for High-Resolution Remote Sensing Image Change Detection

Change detection represents a fundamental research area within remote sensing technology. Nevertheless, in practical applications, the spatial arrangement of terrestrial features is frequently highly complex. The changing objects vary significantly in size and exhibit irregular, complex shapes. Many mainstream methods are hindered by the intricate spatial scales of these changed objects, culminating in suboptimal detection performance. To overcome these obstacles, we propose a change feature refinement and cross-scale interaction network (FRCI-Net) for effective change detection. To capture more robust change features, FRCI-Net first employs a feature interaction module (FIM) and a differential feature refinement module (DFRM) to accurately capture change regions while enhancing the boundary details of the change features. To further enhance the recognition ability for multi-scale change targets, FRCI-Net leverages a cross-scale interaction module (CIM) to enable attention-driven integration and consolidation of change features across divers scales, thereby improving the network’s scale awareness of various change targets. Experiments conducted on two benchmark datasets validate that the proposed method surpasses six advanced methods.