Machine learning in architectural heritage: a review of research trends and applications

Architectural heritage is a tangible record of human civilization and an invaluable cultural asset. Yet traditional conservation and research methods remain costly and inefficient. Although machine learning (ML) is increasingly reshaping architectural heritage conservation, the field still lacks a systematic overview of its rapid development. This study combines bibliometric analysis with a qualitative synthesis of Web of Science literature to map the knowledge structure and evolution of this interdisciplinary domain. The results show rapid growth since 2017, with research activity concentrated in China and Italy and collaboration networks centered on Europe and East Asia. Three main themes dominate the field: (1) defect detection and identification; (2) point-cloud segmentation and digital reconstruction; and (3) analysis, intervention, and inheritance for architectural heritage. The field has evolved from traditional algorithms to deep learning, from 2D image analysis to 3D spatial data, from static models to digital twins, and from basic recognition tasks to broader conservation applications. Future priorities include higher-quality datasets, explainable AI, stronger integration of heritage-conservation theory, attention to institutional and cultural contexts, scale-sensitive research design, and richer multimodal data fusion.