Origin Detection of Illicium Verum Hook. f. Based on Sensor Array Optimization

Abstract

In order to improve the identification ability of electronic nosetostar anise from different areas, this study uses sensor array optimization, Particle Swarm Optimization (PSO) and Support Vector Machine (SVM) to effectively improve the discrimination and prediction ability of electronic nose. Firstly, the initial sensor arrays are selected according to the aroma components of star anise. On the basis of extracting sensor eigenvalues to form the initial feature matrix, the final sensor arrays are selected by combining variable correlation analysis and coefficient of variation analysis (RSD). Linear Discriminant Analysis (LDA) is applied to the sensor array before and after optimization to increase the group spacing of star anise from different producing areas. Particle swarm Support Vector Machine (PSO-SVM) and Genetic Support Vector Machine (GA-SVM) are used to distinguish the producing areas of star anise samples. The accuracy of PSO-SVM training set is 99.16%; that of test set is 93.33%; that of GA-SVM training set is 99.16% and the accuracy of test set is 90%. The results show that PSO-SVM model has high precision and convergence accuracy and it is more feasible to distinguish the origin of star anises.