Explainable multimodal AI framework adaptively calibrates archaeological sensing workflows

Archaeological digitisation increasingly relies on multiple sensing modalities—photogrammetric 3D reconstruction, hyperspectral imaging, X-ray fluorescence (XRF), and Raman spectroscopy—but each is prone to acquisition-specific errors that degrade downstream analysis. This new framework from an international team led by Nevio Dubbini adds an algorithmic supervision layer that assesses whether acquisitions are statistically consistent, physically plausible, and suitable for integration. Rather than replacing instrument-level calibration, it represents each modality as structured feature spaces encoding geometric, spectral, spatial, and statistical properties, then applies supervised and unsupervised learning combined with explainable AI (XAI) to flag problems.

Tests on real archaeological datasets show the framework reliably catches meaningful acquisition variability, enabling robust quality assessment across heterogeneous sensors. The work is published on arXiv (2608.00074), with links to code, data, and experiment replication tools—making it easy for other teams to adopt.