Change probability-aware graph for multimodal change detection

Multimodal change detection (MCD) has emerged as a pivotal technology for monitoring surface changes based on remote sensing images. Recently, many graph structure-based methods have been proposed, leveraging the construction and comparison of consistent structural features across multimodal images to extract changes. However, these graph structures are vulnerable to disturbances from changed regions, potentially compromising the precision of change detection (CD). To address this, this study introduces a novel approach for MCD, termed change probability-aware graph (CPaG). The proposed CPaG utilizes image superpixels as graph vertices, thereby representing the image’s structural features. In constructing connections between vertices, the method meticulously assesses the similarity among them and the change probabilities of neighboring vertices. This approach enhances the precision with which structural features are established and enables a more accurate assessment of the structural disparities between multimodal images. Given that the change probability of neighboring vertices is derived from the structural differences in multimodal images, the study has devised an iterative framework for calculating these probabilities and adjusting the connection weights between vertices. Upon the conclusion of the iterative process, a change intensity map (CIM) is obtained, which delineates the change intensity (CI) for each superpixel within the multimodal images. By applying binary segmentation to CIM, a binary change map (CM) is generated. The efficacy of the proposed CPaG is substantiated through experiments conducted on six multimodal and four unimodal image datasets, as well as comparisons with state-of-the-art methods.