Rapid authentication of geographical origins of Baishao (Radix Paeoniae Alba) slices with laser-induced breakdown spectroscopy based on conventional machine learning and deep learning

Abstract

The geographical origin of Baishao (Radix Paeoniae Alba) affects the components and content, which in turn affects its pharmacological action. Laser-induced breakdown spectroscopy (LIBS) was combined with conventional machine learning and deep learning methods to rapidly discriminate the geographical origins of Baishao slices without sample preparation. The influence of spatial variation of Baishao slices on the LIBS signal was investigated. The spectra that were averaged using 16-point spectra showed the best origin identification performance, with an accuracy of 96.7% as determined by partial least squares-discriminant analysis (PLS-DA). Meanwhile, the spectra obtained from a single point after voting showed the best origin identification performance using ResNet, with an accuracy of 95.0%. The preliminary results indicate the feasibility of using LIBS and machine learning for rapid, accurate, in situ origin identification of Baishao slices, which provides an approach for quality and adulteration supervision.