Hybrid AI-Based Annotations of the Urban Walls of Pisa for Stratigraphic Analyses

The paper presents the preliminary results of the application of automatic classification techniques, using Artificial Intelligence, to the interpretation of the survey data of some important sections of the walls of Pisa, in order to identify the distinctive features that have characterized the evolution of lithotypes over time. A supervised Machine-Learning procedure, leveraging the Random Forest, is applied for the semantic segmentation of many tracks of the Pisan walls. The methodology, applied to the under-researched domain of fortified heritage, is tested to support the stratigraphic analysis as well as the description of the lithotype evolution and succession. We further investigate the transfer of semantic annotations on lithotypes from 2 to 3D data, as part of a broader research project on different ways of storing, retrieving and transferring important architectural heritage knowledge across different types and methods of representation. Two relevant tracks of Pisan walls are considered: the walls in Tersanaia and the track between Torre Santa Maria and Porta murata di Santo Stefano.