Classification accuracy improvement by data preprocessing in handheld laser-induced breakdown spectroscopy

Qualitative analysis using handheld laser-induced breakdown spectroscopy (HH-LIBS) usually suffers from spectral fluctuation. To reduce spectral discreteness and improve classification accuracy, 3 different data preprocessing methods namely minimum standard deviation (MSD), minimum distance (MD), and Weibull distribution (WD) were proposed. The classifications of 15 rock samples using the linear discriminant analysis (LDA) algorithm assisted with these preprocessing methods were carried out. The results showed that the relative standard deviations (RSDs) of the spectral intensities were reduced from 44.39% of the original spectra to 25.28, 19.67, and 27.26%, respectively. The classification accuracies of the rock samples were increased from 93.07% to 99.05, 97.04, and 99%, respectively. The results demonstrate that the preprocessing methods provide an effective approach for improving the analytical performance of HH-LIBS.