Change Detection on Remote Sensing Images Using Multi-Dimensional Attention Network

Change detection in remote sensing images plays a vital role in applications like urban planning and land resource management. Despite its importance, challenges persist due to complex backgrounds, which often lead to imprecise edge detection and missed small-scale target changes. These limitations highlight the need for methods that can robustly extract fine-grained details without sacrificing computational efficiency. To address these challenges, this study proposes a multi-dimensional attention network (MDANet) for change detection in remote sensing imagery. It incorporates a novel multi-dimensional attention mechanism that effectively captures fine-grained details for small object detection, while maintaining computational efficiency and robustness in handling complex scenes. Firstly, it introduces the Attention Feature Fusion Module (AFM), which extracts critical channel features and locates regions of interest to preserve detail and improve edge precision. Secondly, the Multi-Scale Feature Enhancement Module (MFEM) is employed, integrating multi-scale convolution branches to capture a broader context, thus addressing the issue of missing small-scale changes. The proposed MDANet was tested on three widely used datasets-LEVER-CD, WHU-CD, and SVCD-and showed better performance compared to existing state-of-art methods. Results indicate that MDANet effectively detects small-scale object changes, achieving superior overall accuracy compared to other competing methods.