An automated mineral classifier using Raman spectra

We present a robust and autonomous mineral classifier for analyzing igneous rocks. Our study shows that machine learning methods, specifically artificial neural networks, can be trained using spectral data acquired by in situ Raman spectroscopy in order to accurately distinguish among key minerals for characterizing the composition of igneous rocks. These minerals include olivine, quartz, plagioclase, potassium feldspar, mica, and several pyroxenes. On average, our classifier performed with 83 percent accuracy. Quartz and olivine, as well as the pyroxenes, were classified with 100 percent accuracy. In addition to using traditional features such as the location of spectral bands and their shapes, our automated mineral classifier was able to incorporate fluorescence patterns, which are not as easily perceived by humans, into its classification scheme. The latter was able to improve the classification accuracy and is an example of the robustness of our classifier.

Graphical abstractFigure optionsDownload full-size imageDownload as PowerPoint slideHighlights► A spectroscopic mineral classifier was built using an artificial neural network. ► Minerals were selected for compositional characterization of igneous rocks. ► We used two sources of spectral data to ensure the robustness of our classifier. ► The classifier learned differences in spectra that are hard to perceive by humans.