Quantitative Analysis of Ferromanganese Crusts (Fe-Mn Crusts) Using Laser-Induced Breakdown Spectroscopy Combined with Machine Learning

Abstract

Rapid and on-board elemental analysis on the mineral deposits taken from the deep seabed are of great importance for the deep-sea mineral resource survey. In this work, we evaluated the potential of laser-induced breakdown spectroscopy (LIBS) combined with machine learning as a rapid geochemical tool for the quantitative analysis of ferromanganese crusts (Fe-Mn crusts), which is an important type of deep-sea mineral deposits. Both PLS and CNN models were built for the quantification of Fe, Mn, Ti, and Fe-Mn ratio based on the LIBS spectra of Fe-Mn crusts obtained from 39 samples. With the full spectrum as input variables, the linear PLS model shows an overfitting behavior, while the CNN model exhibits superior generalization ability and robustness by automatically learning and extracting important features in the spectra. Thereafter, a feature selection method of SD-SKB was applied on the broadband spectra and the effectiveness of SD-SKB was verified by comparison with the Grad-CAM feature visualization within the CNN model. The performances of the feature-based models are superior or comparable with the full-spectrum models, while the model complexity and computational costs are significantly reduced and the interpretabilities of the models are improved. With the selected variables, the predictive performance of CNN is clearly better than PLS, with the RMSEp values of 0.422 wt.%, 0.532 wt.%, 0.045 wt.%, and 0.031 for Fe, Mn, Ti, and Fe-Mn ratio, and the RSDP values of 1.688%, 2.466%, 3.838%, and 2.605%, respectively. This work demonstrated the capability of LIBS combined with machine learning that could be potentially used for the on-board mineral analysis during the deep-sea mineral resource survey.

Keywords: laser-induced breakdown spectroscopy, ferromanganese crusts (Fe-Mn crusts), Machine learning, convolutional neural network (CNN), feature selection