Answers are grounded in this paper and cite their sources.
In remote sensing (RS) image change detection (CD), edge information provides critical cues for precise detection results and enhances CD performance. However, due to the susceptibility of ground object imaging to seasonal variations, weather conditions and illumination changes, change edges are prone to blending with the background, leading to insufficient utilization of information and false changes. To address this issue, we propose the Edge Integrity Learning Network (EILNet), a novel framework specifically designed for remote sensing change detection (RSCD). First, we construct a change-siamese extraction framework with edge-region decoupled representation to separately learn change edges and regions, thereby improving the accuracy of edge characterization. Then, an explicit edge focus constraint is designed to optimize edge prior learning, where an attention focusing mechanism efficiently aggregates multiscale edge features to enhance edge discrimination accuracy. Furthermore, an implicit hierarchical attention constraint is introduced to facilitate structural information exchange between change regions and edges, refining ambiguous edges. Specifically, hierarchical attention is employed to decouple change foregrounds and backgrounds by guiding multiscale change feature learning with predicted edges, thereby improving change representation performance. In addition, prior edge spatial constraints are leveraged to implicitly learn challenging samples in edge-blurred regions, further suppressing false changes. Experimental results demonstrate that our proposed method significantly improves edge discrimination accuracy in CD, outperforming state-of-the-art approaches across three public benchmarks while effectively extracting and utilizing change-related edge information.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: