An attention-based multiscale transformer network for remote sensing image change detection

The bi-temporal change detection (CD) is still challenging for high-resolution optical remote sensing data analysis due to various factors such as complex textures, seasonal variations, climate changes, and new requirements. We propose an attention-based multiscale transformer network (AMTNet) that utilizes a CNN transformer structure to address this issue. Our Siamese network based on the CNN-transformer architecture uses ConvNets as the backbone to extract multiscale features from the raw input image pair. We then employ attention and transformer modules to model contextual information in bi-temporal images effectively. Additionally, we use feature exchange to bridge the domain gap between different temporal image domains by partially exchanging features between the two Siamese branches of our AMTNet. Experimental results on four commonly used CD datasets - CLCD, HRSCD, WHU-CD, and LEVIR-CD - demonstrate the effectiveness and efficiency of our proposed AMTNet approach. The code for this work will be available on GitHub.1