SwinChange-DE: A Detail-Preserving Hierarchical Transformer Framework for High-Resolution Remote Sensing Image Change Detection

High-resolution remote sensing image change detection plays an important role in urban monitoring, disaster assessment, and land-cover analysis. However, complex spatial structures and diverse object scales make accurate change localization challenging, while progressive downsampling can obscure object boundaries and small changes. To address these challenges, this paper proposes SwinChange-DE, a detail-preserving hierarchical Transformer framework for high-resolution remote sensing image change detection. A Swin-inspired encoder extracts spatial–semantic representations at multiple resolutions through efficient contextual modeling. To bridge the semantic gap across feature levels, an adaptive fusion strategy refines shallow features under deep semantic guidance before aggregation. A detail enhancement module further leverages fused change information to guide the integration of high-resolution features from both temporal images during decoding, supporting boundary recovery and the detection of small-scale changes. Together, these components connect contextual representation, cross-scale alignment, and precise change localization within a unified framework. Extensive experiments were conducted on the LEVIR-CD+ and WHU-CD benchmarks. SwinChange-DE achieves F1-scores of 87.66% and 87.43%, respectively, outperforming the best-performing competing methods by 0.25% and 0.93% in F1-score. The results demonstrate the effectiveness of the proposed framework in improving boundary delineation and recovering subtle changes in high-resolution remote sensing imagery.