A polynomial interactive reconstruction method based on spectral morphological features for the classification of gem minerals using portable LIBS

Abstract

Spectral features are one of the most important factors affecting the classification performance of laser-induced breakdown spectroscopy (LIBS). In order to improve the classification accuracy of portable LIBS, the spectral morphological feature that includes kurtosis and skewness features of the spectral profile was discovered, its utilization method – the polynomial interactive reconstruction (PIC) method – was proposed in this study. A classification experiment of 24 kinds of gem minerals was carried out; the classification accuracies of the conventional spectral intensity method and whole spectrum (WS) method for the samples were only 82.9% and 81.6%, respectively, which improved to 93.9% using the proposed PIC method. Furthermore, the accuracy was also significantly higher than that of other single original features. Due to the high price of samples and their insufficient supply, the generality of the experiment may need further experimental verification. But the experimental results can still show that the proposed feature mining and reconstruction method can significantly improve the classification performance in portable LIBS at least for the present application. This study proves the potential of combining spectral analysis and feature engineering technology to enhance the performance of LIBS qualitative analysis.