Artificial Intelligence for Detecting Surface Alteration Phenomena in Stone-Built Heritage: The Case of the ‘Unfinished Church’ of Venosa

Abstract

Point clouds and 3D models have become essential not only for the digitisation process but also for the non-invasive assessment of deterioration and potential decay mapping in Cultural Heritage, particularly in the built environment and architectural landmarks. These resources facilitate precise digital inspections and enable a comprehensive analysis of the morphological and material properties of heritage assets, in strict alignment with conservation principles. Recent advancements in Artificial Intelligence have further refined 3D data and image processing, introducing sophisticated techniques for segmentation and classification through both supervised and unsupervised learning paradigms. Building upon these breakthroughs, this study explores the semi-automatic identification of surface alterations in the stone masonry of the south-east façade of the ‘Unfinished Church’, which is part of the Most Holy Trinity Complex in Venosa (Southern Italy). The mapping process started with the photogrammetric point cloud, employing RGB colour-detection techniques, followed by the implementation of two Machine Learning algorithms (Fast Random Forest and K-Nearest Neighbours) to examine the UV texture of the polygonal model. Comparative analyses, both quantitative and qualitative, were conducted to assess the effectiveness of these methods in identifying and classifying alterations, highlighting their potential to support preservation efforts and guide future maintenance strategies.