Deep learning assisted XRF spectra classification

Area of Science:

• Materials Science
• Analytical Chemistry
• Computer Science

Background:

• Energy-dispersive X-ray fluorescence (EDXRF) spectrometry is crucial for analyzing cultural heritage materials.
• Traditional multivariate analysis of EDXRF data for archaeometry is often complex and time-consuming.
• Artificial intelligence (AI) offers advanced solutions for improving the speed and accuracy of data analysis.

Purpose of the Study:

• To develop and evaluate an autoencoder neural network as a dimension reduction tool for EDXRF spectral data.
• To enhance the extraction of informative features from raw EDXRF spectra for elemental composition analysis.
• To improve the efficiency and sustainability of postprocessing activities in archaeometric studies.

Main Methods:

• Design and implementation of an autoencoder neural network architecture.
• Application of the autoencoder to reduce the dimensionality of raw EDXRF spectral data from canvas paintings.
• Evaluation of the autoencoder's performance in spectrum reconstruction and feature extraction.
• Utilizing the reduced data for efficient classification algorithms.

Main Results:

• The autoencoder effectively reconstructed original EDXRF spectra, enabling informative feature extraction.
• Dimension reduction was achieved, leading to more efficient classification algorithm performance.
• The AI approach demonstrated significant improvements in processing time and reduced manual expert intervention.
• The method proved to be more sustainable for postprocessing analytical data.

Conclusions:

• Autoencoder neural networks offer a powerful and sustainable approach for dimension reduction in EDXRF spectral data analysis.
• This AI technique enhances the efficiency and accuracy of archaeometric studies on cultural heritage objects.
• The proposed method optimizes resource utilization, saving time and expert effort in data analysis.