Rapid identification of green tea varieties based on FT-NIR spectroscopy and LDA/QR

Abstract

There are many substances beneficial to human body in tea. In this study, we put forward innovative strategies to quickly and harmlessly identify Chinese green tea varieties. Near-infrared (NIR) spectrometer was used to collect NIR spectral data of tea samples, and the data were preprocessed by Savitzky-Golay (SG) filter to eliminate noise of spectral data. Three feature extraction algorithms: principal component analysis (PCA) combined with linear discriminant analysis (LDA), LDA/QR, generalize singular value decomposition (GSVD) were performed to decrease the dimension and compress the spectral data. Finally, k-nearest neighbor (kNN) classifier was utilized to classify the samples according to the NIR spectra of the samples. PCA combined with LDA, GSVD and LDA/QR had the classification accuracy rates 94.19%, 91.86% and 98.84%, respectively. So, LDA/QR showed the highest classification accuracy in classification of NIR spectra of tea samples. We believe that the combination of NIR spectroscopy and feature extraction algorithms can quickly identify the types of tea samples. This method may have the potential to identify other varieties of food.

Keywords:
Chinese green tea; near-infrared spectroscopy; Savitzky-Golay filter; discriminant analysis; LDA/QR