Siamese Networks for Change Detection in Remote Sensing: A Comprehensive Review

This paper presents a comprehensive review of Siamese Networks applied to change detection in remote sensing imagery, examining various architectures, methodologies, and recent innovations. We analyze the evolution from traditional Siamese architectures to advanced variants advanced variants incorporating transformers, attention mechanisms, and multi-modal processing. Key architectures, including FC-Siam-diff, Dual-UNet, and ChangeFormer, demonstrate robust performance on benchmark datasets like LEVIR-CD and OSCD, achieving F1-scores exceeding 90%. Recent advances in pyramid attention mechanisms, differential fusion, and transformer encoders have significantly improved detection accuracy, particularly for subtle changes and challenging conditions. Despite these advancements, computational complexity, data scarcity, and multi-temporal alignment remain critical challenges. This review emphasizes the importance of hybrid approaches combining CNNs with transformers, attention-guided mechanisms, and semi-supervised approaches for enhancing change detection (CD) capabilities.