Elemental Feature Extraction from Historical Pigments Through X-Ray Fluorescence Spectroscopy and Unsupervised Machine Learning

The analysis of historical pigments contributes significantly to understanding the materials and techniques used in artworks and in preserving cultural heritage. This work introduces a novel methodology for classifying historical pigments combining X-ray fluorescence (XRF) spectroscopy with machine learning techniques. We applied this approach to a representative heterogeneous dataset of historical pigments from the open-access spectral library INFRA-ART, as well as commercial oil colors and pigments with different particle sizes. A comparative analysis through principal component analysis (PCA) and hierarchical cluster analysis (HCA) demonstrates the advantages of the full-spectrum method over conventional peak-based strategies, offering improved classification performances and robustness. Employing the entire spectrum, it is possible to access additional key features for pigment discrimination that are discarded during the computation of the traditional methods and it is possible to have an efficient feature extraction even in more complex samples. This approach offers significant advantages by allowing the simultaneous processing of extensive datasets, which is useful for interpreting real-world scenarios in cultural heritage that are characterized by high heterogeneity.