WATCH: AI Framework for Archaeological Site Change Detection

Researchers have developed a new system called WATCH to track changes over time in PlanetScope satellite images from 2017 to 2024, with a resolution of 4.7 meters per pixel, specifically for monitoring archaeological sites. The framework employs three different scoring techniques: Temporal Embedding Distance (TED), which operates without needing training; Self-Supervised Change Detection (SSCD), which merges reconstruction, forecasting, and latent-novelty signals; and a Weakly Supervised (WS) model that uses limited event-month labels. WATCH was evaluated on 1,943 archaeological sites in Afghanistan, utilizing embeddings from six key models and additional handcrafted features. This study, available on arXiv (arXiv:2605.08160), represents a significant step forward in safeguarding cultural heritage through AI.