Rapid and Accurate Classification of Coal by Laser-induced Breakdown Spectroscopy Coupled with Plasma Image-Spectrum Fusion Strategy

Abstract

High-accurate evaluation of coal quality is crucial for efficient energy use. Laser-induced breakdown spectroscopy (LIBS), a fast and efficient detection method, has become a key focus in the field of coal quality detection. However, the complexity of coal matrix significantly limits the classification accuracy of LIBS, hardly meeting the current needs. To overcome this challenge, a novel plasma image-spectrum fusion method was proposed in this study. It provided supplementary information to the spectra through the plasma images, thus significantly improving the accuracy of coal classification models. To verify the classification accuracy of this method, a systematic experimental study was carried out. The spectra, coaxial plasma image, and paraxial plasma image of coal samples were obtained simultaneously, and three different data fusion schemes were compared with the traditional spectra scheme. Meanwhile, three machine learning models, k-nearest neighbor (KNN), decision tree (Tree), and support vector machine (SVM) were used to verify the broad applicability of this method. The results demonstrated that the performance of the classification models based on fusion schemes significantly improved compared with those based solely on spectra. Among them, the classification models based on spectra coupled with coaxial plasma images, and paraxial plasma images achieved the best classification results. The accuracies of the prediction set of these three models (KNN, Tree, and SVM) were improved from 76.51%, 91.23%, and 93.69% to 89.05%, 97.90%, and 96.65%, respectively. Therefore, the relevant results fully verify the feasibility and wide applicability of this method.