Ore Classification and Recognition Based on Confocal LIBS Combined With Machine Learning
Su Yun-peng,He Chun-jing,Li Ang-ze,Xu Ke-mi,Qiu Li-rong,Cui Han
DOI: https://doi.org/10.3964/j.issn.1000-0593(2023)03-0692-06
2023-01-01
Spectroscopy and spectral analysis
Abstract:Mineral classification and identification is an important area in the field of geological research, which is of great significance to geological exploration and environmental evolution. However, the traditional ore classification and identification methods rely on professionals to conduct manual identification through the shape and physical properties of the ore, which has strong subjectivity and low accuracy. Laser-induced breakdown spectroscopy (LIBS) is suitable for geological research due to itselement "fingerprint" characteristics, high sensitivity and fast on-line detection. In this paper, we use confocal laser-induced breakdown spectroscopy combined with machine learning to improve the accuracy of ore classification and recognition. The confocal LIBS system is used to obtain the spectral data of 8 natural ore samples (Gold, Copper, Silver, Hematite, Aluminum, Galena, Apatite and Sphalerite). Principal component analysis (PCA) is used to reduce the dimension of the data, Linear discriminant analysis (LDA), nearest neighbor rule (KNN) and support vector machine (SVM) are used for high-precision classification and recognition of feature spectral lines. Firstly, a standard copper is employed as the sample to conduct the comparison experiments between non confocal LIBS system and the confocal LIBS system for the stability and its influence on the cumulative contribution rate of PCA principal components. The results show that compared with non-confocal LIBS system, the stability of the confocal LIBS system is improved by 63. 75%, and the cumulative contribution rate of principal components is increased by 17. 81%. Then, the confocal LIBS system is used to obtain the spectral information of the above eight ore samples with data preprocessing, such as denoising. PCA is used to extract the ore feature data, and the first 10 -dimensional feature space with a cumulative contribution rate of 99. 4% is retained. Finally, the feature data are combined with LDA, KNN and SVM to build a classification model for classification and recognition. The experimental results show that the classification accuracy of PCA combined with LDA and KNN is 95. 78% and 92. 58% respectively, while that of SVM can reach 97. 89%. Therefore, combining confocal laser-induced breakdown spectroscopy with PCA and SVM can provide a fast and accurate classification and recognition method for geological exploration and mineral recognition and has wide application prospects.