A Detail-Preserving Frequency-Guided Cross-Scale Fusion Network for Remote Sensing Change Detection
Abstract
Remote sensing change detection (RSCD) aims to identify changed regions from bitemporal remote sensing images and plays an important role in urban monitoring, disaster assessment, and land-resource management. However, accurate RSCD remains challenging due to scale variations, boundary detail loss, pseudo changes, and efficiency demands in high-resolution scenarios. To address these issues, this paper proposes FG-CFNet, a detail-preserving frequency-guided cross-scale fusion network. Specifically, LSAM preserves local details during hierarchical semantic aggregation; MSWA decomposes bitemporal features into low-frequency semantic and directional high-frequency components to capture structural and boundary changes; and UCFM uses frequency-guided features and a change-prior map to guide cross-scale fusion, enhancing true changes while suppressing pseudo changes. Experiments on LEVIR-CD and WHU-CD demonstrate the effectiveness of the proposed method.