The Spatio-Temporal Weighted Adjustment Network for Remote Sensing Image Change Detection

With the rapid advancement of deep learning, substantial progress has been achieved in remote sensing change detection (CD). However, there are still two key challenges. First, the widespread scene context interference hinders the accurate detection of change regions; second, the existing methods are difficult to simultaneously detect change regions across different scales. To address these issues, this article presents a cross-spatio-temporal weight adjustment network (CWA-Net) with three core optimizations. First, we propose a cross-spatio-temporal differential fusion attention mechanism, which utilizes differential features extracted by the backbone network to enhance bitemporal features. Through the coordinated use of multiple attention mechanisms and channel exchange, the mechanism promotes deep interaction and fusion of bitemporal features, effectively reinforcing change region representations while mitigating scene interference. Second, we design a multiscale selection and aggregation module that adaptively selects and aggregates the optimal scale features from multiscale features, enhancing the model’s capability to capture change regions at different scales. In addition, we put forward a two-type change-feature complementarity strategy, which reweights change features extracted via subtraction and concatenation during the aggregation of multiscale feature maps, thereby enhancing feature complementarity and enriching change information. Finally, extensive experiments on four remote sensing CD datasets demonstrate that CWA-Net, based on a simple backbone network ResNet18, outperforms existing state-of-the-art SOTA methods.