Classification accuracy improvement of portable laser-induced breakdown spectroscopy based on spectral feature augmentation

Spectral features have a crucial influence on the qualitative analysis performance of portable laser-induced breakdown spectroscopy (LIBS) technology. To solve the problem of poor classification performance caused by the weak characterization ability of conventional single feature forms such as spectral intensity feature, a classification method based on spectral feature augmentation (SFA) was proposed. It was realized through the mining, extraction and fusion for 6 different types of features including spectral intensity (SI), spectral peak area (SPA), full width at half maximum (FWHM), standard deviation (SD), signal-to-background ratio (SBR), and signal-to-noise ratio (SNR). Using this method, the classification accuracy of 75 sedimentary rock samples was increased from 87.2% of the conventional spectral intensity method to 95.5% of the SFA method. In addition, the classification experiment of 24 metallic mineral samples was used to test its generalization ability, and their classification accuracy was increased from 77.5% to 91.7%. The results indicate that the SFA method can improve the characterization ability of feature vectors and the classification model's accuracy by mining more effective information from a limited number of spectra and then successfully integrating and utilizing it. This method provides a new approach for improving the qualitative analysis performance of portable LIBS