Skip to main content
SciX 2025 Main banner

Oral Contributed Presentation

CHEM

CHEM03: Young Investigators in Chemometrics

Applying Machine Learning to Micro-Spatially Offset Raman Spectroscopy for Analyzing Heterogeneously Degraded Pigments

Thursday, October 9, 2025
11:30 AM - 11:50 AM EDT
Location: Meeting Room #3
In fifteen words or less, explain the significance of this contribution (Novel Aspect).: We developed an automated machine learning approach to detect degradation products in micro-SORS spectra.

Abstract Text: Micro-Spatially Offset Raman Spectroscopy (micro-SORS) is a non-invasive technique that is used in the Cultural Heritage field to investigate painted layer sequences and heterogeneous degradation diffusion processes in artworks. Information about the distribution of pigments and their degradation products within the layers is essential for the knowledge of the artist’s technique and the conservation history of the objects. Numerous degradation products are present within the layers in varying concentrations, and some signals overlap, which makes identifying the compounds difficult and hard to compare to literature references. In our study we applied cheminformatics to speed up the analysis of the distribution of pigments and their degradation products. We developed a method consisting of a combination of machine learning techniques to identify important shifts in the Raman spectra that differentiate between compounds corresponding to the intact and the degraded pigment. By using this method, large amounts of spectra can be analyzed, and important patterns in the intact and degraded spectra can be highlighted, which would be missed by visual inspection. Similar spectra are grouped together using hierarchical clustering, and important distinguishing features are determined using explainable machine learning. These important features are identified and used to create micro-SORS maps, which are visual representations of the compositions of the pigment layers. The micro-SORS maps allow conservators to determine the depth and breadth of the pigment degradation so that they can take action to prevent further damage. Combining machine learning with Raman spectra analysis allows for an automated method to analyze large amounts of data and to identify important shifts without prior knowledge of the composition of the samples.