Discrimination of Storage Time of White Tea Using Hyperspectral Imaging

Abstract

The storage time is a main factor determining the value of white tea. To discriminate the storage time rapidly and nondestructively, a new method was applied in this paper. First, hyperspectral image data were captured from Shoumei of 3, 6 and 10 years storage time. Second, four kinds of algorithms were applied to preprocess the original data, savitzky-golay Filter, standard normal variate, minmaxscaler, and multiplicative scatter correction. Third, support vector machine, partial least squares with linear discriminant analysis and logistic regression were built based on the data of the full spectra. Finally, the best combination of preprocessing algorithm and model could be found by comparing the confusion matrix, precision and recall. The results showed that the best classification performances were obtained with the support vector machine after the pretreatment of standard normal variate. The precision of the calibration sets was 90.83%, and that of the prediction sets was 86.02%. Therefore, it is possible to use hyperspectral imaging to discriminate white tea of different storage time in tea industry.