Deep learning-based detection and evaluation of building changes in rail transit protection zones

Le résumé fourni par la source

Rapid urbanization in China exacerbates risks to rail transit protection zones, necessitating high-precision building change detection. Existing research primarily focuses on algorithm optimization and overlooks the construction of scene-specific datasets. Focusing on protected areas of 16 rail transit lines in Beijing, this study constructs a dedicated training sample set and conducts change detection using Mask R-CNN. The results show that the dedicated training sample set achieves 92.45%, 93.28%, 92.86%, and 86.38% in precision, recall, F1 score, and IOU, respectively, which are significantly improved over common datasets. Compared with the EGRCNN, precision, F1 score, and IOU are improved by 1.04%, 3.62%, and 5.81%, respectively, despite a marginally lower recall. The study demonstrates that a dedicated training sample set enhances the accuracy of building change detection. It emphasizes a shift from pure algorithm optimization toward dataset specialization, offering methodological and practical value for infrastructure safety and sustainable urban development.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.