AI and Deep Learning for Image-Based Segmentation of Ancient Masonry: A Digital Methodology for Mensiochronology of Roman Brick

In the field of building archaeology, the analysis of wall surfaces represents a fundamental tool for the study of an architecture and its construction phases. In fact, masonry stores valuable information regarding not only used materials and construction techniques but also transformations happen over time for natural events or anthropic interventions. The traditional approach to the analysis of building materials is mainly based on direct observation and manual annotations based on orthophotos obtained through photogrammetric surveys. This process, while providing a high degree of accuracy and understanding, is extremely time- and resource-consuming. In addition, the lack of standardised procedures for the statistical analysis of measurements leads to data that are difficult to compare for different contexts. Time and subjectivity are ultimately the two main limitations that most hinder the diffusion of the mensiochronological approach and for this reason, the most recent artificial intelligence solutions for the segmentation and extraction of measurements of individual masonry components will be addressed. Finally, a workflow will be presented based on image segmentation using machine learning models and the automatic extraction and statistical analysis of measurements using a script designed specifically by the author for the mensiochronological analysis of Roman brick masonry.