A method for detecting and quantifying damage in masonry architectural heritage based on deep learning of point cloud data

Point-cloud data provide three-dimensional geometry for automated inspection of masonry architectural heritage, addressing limitations of image-based methods in illumination and depth measurement. This study proposes an integrated framework combining preprocessing, binary segmentation using an improved PointNet++ incorporating a Novel Set Abstraction (NSA) module, and morphology-based quantification of damaged area and maximum depth. The model was evaluated on 317 annotated patches collected from four wall segments at different locations and orientations along the Nanjing City Wall. The improved model achieved a mean intersection over union of 73.77%. Compared with the original PointNet++, it increased the damage-class intersection over union from 28.80% to 55.98%. Validation on 30 independently measured damaged regions yielded a mean relative area error of 17.24% and a mean absolute maximum-depth error of 6.77 mm. Under the tested Nanjing City Wall conditions, the framework supports preliminary documentation and monitoring of surface material deterioration; external transferability remains to be validated.