Building Change Detection Method Based on Dual-Task Constrained Deep Siamese Networkand ViT Attention Mechanism

Building change detection has become increasingly important in fields such as land-use planning, urban management, and emergency response. Traditional manual methods are time-consuming and prone to errors, creating a need for more efficient automated techniques. In this paper, we propose a Dual-Task Constrained Deep Siamese Convolutional Network (DCEDSCN) model, which addresses issues such as insufficient feature discrimination, incomplete regions, and irregular boundaries in building change detection. The proposed model integrates the Vision Transformer (ViT) attention mechanism into the encoder, replacing the first layer of the encoder with the Enhanced Vision Module (EVM) and replacing the final blocks with a Multi-Dimensional Feature Fusion (MDFM) module for better feature aggregation. Experimental results on the LEVIR-CD Building Change Detection dataset show that the proposed method outperforms existing approaches in terms of precision, recall, F1-score, and Intersection over Union (IoU), achieving state-of-the-art performance.