Abstract: [Purposes]Traditional building change detection methods for remote sensing images have limitations in preserving high-resolution details and global semantic modeling of complex scenes.Aiming at the demands of regular,automatic and regular,automatic and efficient change detection for urban buildings,this paper carries out research on method improvement and practical application.[Methods] A two-stage collaborative detection framework integrating Transformer-based change detection and HRNet-based building extraction is constructed.First,Transformer is adopted to model the global semantic relationships between bi-temporal images and locate change regions.Then,HRNet is utilized to retain high-resolution spatial details,so as to realize refined building extraction and change type identification within candidate change areas.[Findings]Taking Shenzhen city as the research area,application verification is conducted from pilot tests in typical zones to full-city promotion.The effectiveness of model-generated change clues is further analyzed from the perspective of practical business work.Experimental results demonstrate that the proposed method achieves favorable feasibility and performance in high-density urban environments.[Conclusions]The proposed method can provide technical support for refined natural resources management and smart city construction,and possesses great practical engineering application value．

Key words: building change detection, collaborative detection, Transformer, HRNet, high-resolution remote sensing imagery