Artificial neural network for high-throughput spectral data processing in LIBS imaging: application to archaeological mortar. Issue 3 (13th February 2023)

• Record Type: Journal Article
• Authors: Herreyre, N.; Cormier, A.; Hermelin, S.; Oberlin, C.; Schmitt, A.; Thirion-Merle, V.; Borlenghi, A.; Prigent, D.; Coquidé, C.; Valois, A.; Dujardin, C.; Dugourd, P.; Duponchel, L.; Comby-Zerbino, C.; Motto-Ros, V.
• Is Part Of: Journal of analytical atomic spectrometry. Volume 38:Issue 3(2023)
• Page Start: 730
• Page End: 741
• Publication Date: 2023-02-13
• Subjects: Atomic spectra -- Periodicals; Atomic absorption spectroscopy -- Periodicals; 543.0858
• DOI: 10.1039/d2ja00389a
• Languages: English
• ISSNs: 0267-9477

Abstract

Abstract : With the development of micro-LIBS imaging, the ever-increasing size of datasets (sometimes >1 million spectra) makes the processing of spectral data difficult and time consuming. Abstract : With the development of micro-LIBS imaging, the ever-increasing size of datasets (sometimes >1 million spectra) makes the processing of spectral data difficult and time consuming. Advanced statistical methods have become necessary to process these data, but most of them still require strong expertise and are not adapted to fast data treatment or a high throughput analysis. To address these issues, we evaluate, in the present work, the use of an artificial neural network (ANN) for LIBS imaging spectral data processing for the identification of different mineral phases in archaeological lime mortar. Common in ancient architecture, this building material is a complex mixture of lime with one or more aggregates, some components of which are of the same chemical nature ( e.g. calcium carbonates). In this study, we trained an artificial neural network (ANN) for automatic detection of different phases in these complex samples. The training of such a predictive model was made possible by building a LIBS dataset of more than 1300 reference spectra, obtained from various selected materials that may be present in mortars. The ANN parameters (pre-treatment of data, number of neurons and of iterations) were optimized to ensure the best recognition of mortar components, while avoiding overtraining. The results demonstrate a fast and accurate identification of each component. The use of an ANN appears to be a strong means to provide an efficient, fast and automated LIBS characterization of archaeological mortar, a concept that could later be generalized to other samples and other scientific fields and methods.