Unsupervised multimodal change detection based on adaptive optimization of structured graph

Multimodal Change Detection (MCD) is essential for disaster evaluation and environmental monitoring by integrating various remote sensing data to monitor surface changes. However, the significant imaging differences in multimodal images render traditional unimodal change detection (UCD) methods ineffective. This paper proposes a novel method for MCD using an adaptive optimization of the structured graph (AOSG) to mine comparable structural features across multimodal images. The proposed method first constructs an adaptive structured graph that captures the structural features of multimodal images. It then cross-maps these features to other image domains to measure change intensity (CI). Moreover, the method incorporates post-mapping structure changes in structured graph and discrepancies in multimodal structured graphs, enhancing its ability to measure structural differences. Through iterative optimization, the optimized structured graph is constructed by examining the change attributes of neighbors in the structured graph, and forward and backward CIs are then produced. By fusing these CIs, the final CI is obtained and subsequently segmented to derive the change map (CM). Through experimental evaluations on six multimodal datasets and four unimodal datasets, the results consistently demonstrate the effectiveness of the proposed AOSG.