Selecting Near-infrared Hyperspectral Wavelengths Based on One-way ANOVA to Identify the Origin of Lycium Barbarum

The effective identification method for Lycium barbarum origins is important to control the source of genuine medicinal materials and enhance market supervision. Near-infrared (NIR) hyperspectral imaging technology has already been preliminarily applied for the identifying Lycium barbarum origins. However, the dimension of hyperspectral data is high, the correlation of adjacent features is strong, and the redundant information is heavy. Therefore, the appropriate wavelength selection method for hyperspectral data can reduce redundant information, and make subsequent classifiers easy to compute. In this study, a near-infrared hyperspectral wavelength selection method based on one-way analysis of variance (ANOVA) is proposed. Since Lycium barbarum of different origins have a different average reflectance at the same wavelength, the wavelengths with significant differences in average reflectance can be selected as feature wavelengths. The results show that the feature wavelengths have good performance for recognition. Compared with the common wavelength selection methods, like the principal component analysis (PCA) and genetic algorithm (GA), one-way ANOVA also shows a better performance. The 37 feature wavelengths are selected by one-way ANOVA, and the average accuracy on the test set after 10-fold cross-validation is 95.25%. These results demonstrated that wavelength selection based on one-way ANOVA can effectively identify the origins of Lycium barbarum.