Vision Transformer-based Change Detection in optical and SAR Remote Sensing Images

Change detection (CD) in remote sensing plays a crucial role in monitoring land cover changes and environmental transformations. Currently, there is a lack of adaptability to different image types. In this work, we propose CD-ViT, an innovative change detection method based on the Vision Transformer (ViT). The proposed framework integrates complementary information from optical and SAR NDVI images through a cross-attention fusion module, followed by a multi-attention UNet decoder to generate highly accurate change maps. Extensive experiments were conducted on several geographical areas, comprising a total of 145,161 patches. Across all studied regions, CD-ViT outperforms state-of-the-art methods, achieving a precision of 94.3% and an F1-score of 94.1%.