Non-destructive classification of ancient mural pigments by hyperspectral imaging

Abstract

Given the vulnerability and value of ancient murals, there is an urgent need to identify, restore and preserve their pigments. This study develops an image spectral fusion (ISF) method integrating hyperspectral imaging with optimized superpixel segmentation and spectral processing to achieve rapid, non-destructive pigment classification. Applied to the Yungang Grottoes murals, the Support Vector Machine model based on ISF realizes the superpixel-level classification of ancient mural pigments with an accuracy of 87 %. External validation demonstrates its excellent classification performance across diverse mural preservation states. Spectral characterization analyses reveal the potential of the method in pigment identification through spectral matching, and pigment mixtures classification. This non-destructive, contactless detection method can serve as a methodological foundation for pigment identification in murals.