A Semantic Classification Approach for the Aachen Cathedral
Keywords: Machine Learning, Aachen Cathedral, Digital Survey, Classification, Segmentation
Abstract. This paper presents a semantic classification workflow applied to the UNESCO World Heritage site of Aachen Cathedral, integrating 3D survey data and Machine Learning (ML) techniques. The research focuses on the Westwerk area and combines Terrestrial Laser Scanning (TLS) and photogrammetric point clouds for material and construction technique identification. A hybrid approach was adopted: 3D geometry-based classification using supervised algorithms (Random Forest) and 2D image segmentation via META’s Segment Anything Model (SAM). The 3D method proved suitable for distinguishing features with strong morphological differentiation, while the 2D approach was more effective for visually subtle or geometrically similar elements. Annotated 2D masks were projected onto the 3D model to improve classification reliability.
The study demonstrates the potential of integrating spatial and radiometric data for scalable, semi-automatic classification of historic masonry. Results contribute to the development of enriched, semantically annotated 3D datasets supporting documentation, conservation planning, and future research on complex heritage sites.