Automated dimensional pattern analysis of Ming Great Wall hollow defense towers using point cloud segmentation

Accurate dimensional documentation of architectural heritage elements is critical for conservation, restoration, value assessment, and sustainable reuse; however, conventional measurement approaches remain inefficient and difficult to apply at large scales. This study proposes an automated framework for the dimensional extraction and pattern analysis of architectural heritage units based on point-cloud semantic segmentation. High-resolution 3D point clouds of hollow defense towers were generated using low-altitude UAV photogrammetry. Components were extracted using the PTv3 semantic segmentation algorithm. Geometric feature-based algorithms were applied to compute dimensional attributes, while statistical and GIS-based spatial analyses were employed to identify morphological and spatial patterns. Applied to 602 towers, the framework established a comprehensive dimensional and morphological database. Results reveal consistent tower length, dominant near-square plans, variable width, and clear spatial clustering, suggesting differentiated construction strategies under unified building standards. The proposed approach enables scalable, reproducible, and quantitative analysis of architectural heritage.