Shp2gml: semantic 3D model generation for Ming and Qing historical buildings at multiple LoDs using domain knowledge and multi-source data

Ming and Qing historical buildings are key components of China’s cultural heritage, characterized by refined craftsmanship and unique spatial layouts. However, UAV restrictions and heritage protection policies often limit the availability of comprehensive 3D data. To address this, we propose shp2gml, a semantic 3D modeling method that integrates building footprints, façade imagery and unstructured web text under data- constrained conditions. First, domain knowledge and deep learning are combined to extract building features from multi-source data. Then, by analyzing architectural characteristics and applying domain-specific rules, a modeling algorithm is designed to generate semantic 3D models. Symmetry priors are further introduced to infer missing elements and enhance texture-to-geometry mapping. Experiments show that shp2gml produces LoD0 to textured LoD3 models, achieving an overall F1-score of 78% and >0.80 accuracy for key entities in NER task. Compared with existing models, roof-type accuracy reaches 100% and door/window completeness improves to 83%, demonstrating its effectiveness for cultural heritage modeling and GIS-based applications. This study not only advances semantic 3D modeling for architectural heritage but also offers new perspectives for interdisciplinary research in GIS and heritage digitalization.