Answers are grounded in this paper and cite their sources.
This framework uses machine learning to accurately detect decay types in tropical heritage structures, indicating improved diagnostic capabilities.
Diagnosing chromatic decay—a key indicator of material deterioration—in tropical architectural heritage presents a persistent challenge in conservation informatics. Conventional methods, including two-dimensional imaging, manual inspection, and photogrammetry, are often characterised by their labour-intensive nature and lack of expediency when scaled up. Furthermore, these methods are usually insufficient in capturing the subtle chromatic variations that occur across diverse materials. These limitations are particularly problematic in humid tropical environments, where biological patina, staining, and water-induced decay progress rapidly. This study examines the development of a versatile diagnostic framework designed explicitly for tropical heritage sites. The research employs a dual machine learning methodology applied to RGB-enhanced photogrammetric point clouds collected from four heritage sites in Semarang, Indonesia. The initial approach utilises unsupervised hierarchical clustering on HSV-transformed point clouds to detect chromatic variations. The second approach uses a supervised Random Forest classifier, trained on manually annotated UV maps, to detect specific decay types. Both methods were evaluated against ground-truth data to assess their accuracy and scalability in classifying biological patina, stains, and surface growths. The findings suggest that the unsupervised clustering algorithm exhibited superior performance, attaining a precision of over 85% and an F1-score of more than 0.83 across all sites. The system’s flexibility, independence from manual annotation, and robustness to variable lighting and geometric conditions make it particularly effective for scalable diagnostics. The proposed framework provides a practical and transferable solution for enhancing digital heritage conservation workflows in tropical regions.
No takes yet. Share an insight, caveat, or question.
Gbran et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: