Mid-level data fusion of Raman spectroscopy and laser-induced breakdown spectroscopy: Improving ores identification accuracy.

The identification of ore samples is of great scientific significance for mineral exploration, and geological evolution research on the planets. Attributed to the changes in the composition and structure of the same ore, the fusion of multiple technologies can effectively meet the comprehensive and accurate analysis of actual samples compared with a single technology. We develop an efficient method of applying the combination of Raman spectroscopy and laser-induced breakdown spectroscopy (LIBS) to ores identification. We construct a convolutional neural network (CNN) model and train it with mid-level Raman-LIBS fusion spectra of ores. Also, we develop a hybrid feature selection method AVPSO based on analysis of variance (ANOVA) with the particle swarm optimization (PSO) to improve the classification performance of the model. Compared with the model features visualized by Grad-CAM method, the similarity selected features verify the effectiveness of the AVPSO method. The identification of mid-level fusion strategy provides the best accuracy of 98%, while the accuracies of Raman and LIBS are slightly lower with values of 87.9% and 91.3%, respectively. The proposed method is of great significance for the rapid and accurate identification of ore samples.