An adaptive change allocation network for coarse-to-fine remote sensing change detection

Change detection in remote sensing images is a crucial task with broad applications in urban planning, land use monitoring, and environmental protection. As a pixel-level task, it requires precise spatial information preservation. However, many existing methods emphasize deep semantic feature extraction while neglecting effective bi-temporal integration and regional change modeling, leading to feature confusion, false alarms, and boundary discontinuities. To tackle these issues, we propose Adaptive Change Allocation Network (ACANet), which uses a coarse change map to guide a dynamic-weight bi-temporal fusion strategy. This enables a coarse-to-fine detection process that enhances change discrimination and improves spatial consistency. Experiments on two benchmark datasets show that ACANet achieves superior performance compared to state-of-the-art methods.