An artificial intelligence-based semiquantitative method based on visible spectroscopy and imaging to analyse inorganic red pigments in wall paintings

Artificial Intelligence (AI) and Machine Learning (ML) are revolutionizing data analysis by introducing innovative and enhanced methods for data processing. In this article, a chemometric semiquantitative model based on visible spectroscopy and digital image colorimetry was applied to estimate the metal content in inorganic pigments. The model utilized Support Vector Machines (SVM) and Artificial Neural Networks (ANN) regression methods to correlate spectral and colorimetric data with elemental composition. Replicas were prepared and painted with three red inorganic pigments (cinnabar, hematite and minium), and they were analysed using portable X-ray fluorescence, visible reflectance spectroscopy, and digital imaging. Cross-reference between elemental and colorimetric information was performed using Support Vector Regression and Artificial Neural Networks, and the models were validated through Venetian-blinds cross-validation. In the calibration step, Root Mean Square Errors (RMSE) for Fe, Pb, and Hg were 0.03, 3.5, and 3.0 %, respectively, with correlation values (R²) of 0.99, 0.90, and 0.94. For the prediction set, RMSE was 3.0, 2.6, and 2.3 %, for Fe, Pb, and Hg, respectively, with R² of 0.83, 0.92 and 0.81. This article demonstrates that innovative data treatment models, coupled with non-invasive and portable techniques, allow us to estimate the content of elements in inorganic pigments in Cultural Heritage samples.