Speaker
Description
This research focuses on the application of advanced digital technologies to the study and restoration of Nuragic pottery. The reconstruction of fragmented archaeological artefacts is a key challenge in cultural heritage research, enabling both the physical restoration of objects and a deeper understanding of their intact shape, typology, and function. Traditional manual reassembly requires time-consuming manual efforts by restorers and archaeologists, while recent advances in 3D digitisation and AI may provide new tools to face this challenge. A recent collaboration between the INFN-CHNet, the Soprintendenza Abap per la città metropolitana di Cagliari e le province di Oristano e Sud Sardegna and Fondazione Barumini has led to the development of an innovative digital pipeline. The project is structured around three main objectives:
1) Dataset: creation and publication of a large open-access dataset of 3D models of intact pottery vessels and their digitally fragmented version usable to train neural networks in object reconstruction tasks.
2) Classification: design, test and validation of a Deep Learning model capable of automatically identify pottery typologies processing shards in the form of 3D point clouds.
3) Partial reassembly: development of a Deep Learning model capable of evaluating adjacency matches between fragment pairs from the analysis of their geometry and fracture surfaces.
The State-Of-The-Art AI-based technologies for the analysis of 3D data will be employed (e.g. PointNet++, DGCNN, GATv2, PointTransformer etc.) with the goal to support the investigation and physical restoration of Nuragic pottery (1700 B.C.E-1000 B.C.E) found at sites under archaeological investigation by the Soprintendenza. Preliminary results of this work will be presented.