Manufacturer and authenticity identification of Chinese Ejiao based on laser-induced breakdown spectroscopy and machine learning algorithms

In this paper, a method combining LIBS technology and machine learning algorithms was proposed to identify Ejiao rapidly and accurately. The experiment was divided into sample preparation, spectra acquisition, data preprocessing, feature selection, model training and model evaluation. Several Ejiao samples from different Ejiao manufacturers were collected, and there were two fake samples made by illegal manufacturers. All the samples were appropriately cut without other processing. The LIBS spectra of these samples were detected in the wavelength from 200 nm to 430 nm. Noise removal and intensity normalization were carried out; then the PCA method was used to select the features. Machine learning algorithms like SVM, RF and ANN were used to build the Ejiao identification model, and the model performance was evaluated to explore the feasibility of using the LIBS spectrum to label Ejiao. The result showed that, for the Ejiao manufacturer identification problem, the best model accuracy of the three algorithms is 96.67%, 96.19% and 97.62% respectively; for the Ejiao authenticity identification problem, good model performance was obtained, for instance, the average Fl-measure for the three methods was 98.33 %. It also showed that the model performance would be further improved as the size of the training data set increased. The ANN algorithm obtained the best model performance, and the manufacturer identification accuracy was 98.57%, and the authenticity identification Fl-measure was 99.01%. The result indicated that our method could establish a practicable model for the manufacturer and authenticity identification of Ejiao.