←→

LIMAP was conceived to address the bottleneck created by the classification and mapping of stone pathologies being carried out completely manually, by testing whether deep learning object-identification techniques could automate the identification, classification, and mapping of limestone deterioration, despite limitations such as the lack of material-specific datasets and incompatibility between Artificial Intelligence (AI) raster outputs and CAD workflows.

To bridge this gap, LIMAP developed a tool integrating AI with photogrammetry into CAD workflows. The team at AP led dataset preparation, curation and validation, while Neural AI managed model training and the vectorisation pipeline. A major achievement was the creation of a proprietary dataset of Maltese stone decay using high-resolution photos, photogrammetric surveys, and point cloud data of historic façades, which was crucial in successfully completing this initial phase of ‘proof of concept’.

Overall, LIMAP leverages AI for documentation, object-identification modelling, restoration, and continuous monitoring, contributing to advancing heritage-led innovation and enabling more efficient heritage conservation workflows to safeguard our shared heritage.