Remote sensing techniques and Machine Learning analyses in archaeology: a methodological approach to territory investigations

This research, conducted within the framework of the In.Res.Agri project, explores the integration of advanced remote sensing and artificial intelligence techniques to investigate archaeological landscapes in the Campania region (Italy), with a focus on the regions of Irpinia and the Campanian Plain. The study utilises high- and very-high-resolution multispectral satellite imagery, processed through software such as eCognition, with the objective of identifying and analysing land divisions, including Roman centuriation, and other archaeological features. The methodology combines traditional topographical analysis with both semi-automated and fully automated deep learning approaches, notably the training of convolutional neural networks (CNNs) for the automated detection of linear features. Preliminary results demonstrate the software’s capacity to validate known archaeological layouts while revealing previously unidentified structures, such as ancient road networks, in challenging terrain. The continuous enhancement of CNN models is intended to achieve fully automated detection, with the objective of accelerating landscape analysis and enabling field validation. The project under discussion demonstrates the considerable potential for the integration of geospatial analysis and AI-driven technologies to enhance our comprehension of archaeological landscapes. It is important to acknowledge, however, that such endeavours are not without their limitations, and that there is a necessity for ongoing refinement and on-site verification.