A comparative study of classification models for laser-induced breakdown spectroscopy of Astragalus origin.

In this paper, laser-induced breakdown spectroscopy (LIBS) combined with machine learning algorithm was used to study the classification of Astragalus from 10 different origins. First, the spectrum of Astragalus from 10 different origins was collected based on LIBS technology, and support vector machine (SVM) and extreme learning machine (ELM) models were established, respectively. Their accuracy rates were 45% and 30%, respectively. After that, the feature variables were selected as the input vectors of SVM and ELM models using random forest (RF) feature importance ranking, and random forest-support vector machine (RF-SVM) and random forest-extreme learning machine (RF-ELM) models were established, respectively. Their accuracy rates were 83.33% and 58.33%, respectively. To improve classification accuracy, gray wolf optimizer (GWO) algorithm was used to optimize RF-SVM and RF-ELM, respectively, and two classification models of gray wolf optimizer-random forest-support vector machine (GWO-RF-SVM) and gray wolf optimizer-random forest-extreme learning machine (GWO-RF-ELM) were constructed. The accuracy rates of GWO-RF-SVM and GWO-RF-ELM models were 85% and 92%, respectively. The results show that the classification effect of GWO-RF-ELM is the best, and its macro-precision, macro-recall and macro-F1 score were 92%, 100%, 92.04% and 95.86%, respectively. It provides an effective method for the identification of Astragalus. [ABSTRACT FROM AUTHOR]