Multimodal knowledge graph for Chinese architectural heritage based on semantic-geometric alignment

Traditional Chinese architectural heritage is documented through historical texts, field photographs and HBIM geometry, yet these sources are usually disconnected. This study proposes a multimodal knowledge graph framework that aligns textual semantics, visual evidence and 3D components for Chinese architectural heritage. We construct YingzaoOnto by combining Yingzao Fashi, Gongcheng Zuofa Zeli, field surveys and interoperable heritage and spatial ontologies. The proposed Semantic-Geometric Alignment module maps text, image and voxelized HBIM features into a shared space through contrastive learning, positive-negative sampling and ontology-guided geometric constraints. Experiments on a curated text-image-HBIM dataset show improved cross-modal retrieval, HBIM semantic enrichment and period-matched modular rule validation. The resulting ChAHMKG provides a traceable knowledge infrastructure for conservation documentation, geometric checking and evidence-based restoration.