ALCEO project will develop new Artificial Intelligence methods for processing different types of EO data with the aim of semi-automatically identifying, in archaeological areas, 'anomalies' that can be recognised as looting pits.

The project will proceed by:

1. Developing a pipeline for automatic dumping of EO datasets through open platforms.
2. Designing an 'anomaly'-detection algorithm on time series images, to detect looted sites.
3. Designing an experimental set-up for comparing, on a benchmark dataset, the proposed classification algorithm to the state-of-the-art ones.
4. Archaeological ground-truthing in areas notoriously subject to looting activities, to verify automated identifications and test the accuracy of the predictions.
5. Producing a digital map of the identified looted sites to be provided to relevant authorities (Interpol, Carabinieri TPC).

END-USERS

Law Enforcement Agencies represent ALCEO's PRINCIPAL END-USERS.

The project has been endorsed by:

1. INTERPOL - International Criminal Police Organization (Works of Art Unit)
2. MIC - Italian Ministry of Culture - Coordinator of actions & investigations of the Carabinieri Command for the Protection of Cultural Heritage (Carabinieri TPC)

NEEDS

A novel system able to:

1. provide a faster and reliable way to detect both past and ongoing illicit digging activities;
2. monitor efficiently areas secluded or difficult to patrol (i.e., deserts, farmlands, etc.);
3. understand and follow the evolution of looters' behaviour by analysing the patterning of pits;
4. facilitate the matching of recovered and seized archaeological items with their provenance;

Within the ALCEO project is the development of the Active Learning platform. The platform serves to reduce manual labelling effort by continuously improving model performance through human annotation on the most informative samples.

The model's steps include:

1.    Uncertainty Extraction: Gathering per-tile confidence scores from inference runs.
2.    Sample Selection: Identifying top-N high-uncertainty tiles.
3.    Expert Annotation: Reviewing and labelling of the selected tiles.
4.    Data Integration: Versioning new annotations and triggering pipeline updates.
5.    Model Retraining: Retraining the model on new datasets to improve accuracy.