Identification of tea based on CARS-SWR variable optimization of visible/near-infrared spectrum

The identification of tea varieties is essential to obtain high-quality tea that can command a high price. To identify tea varieties quickly and non-destructively, and to fight against counterfeit and inferior products in the tea market, a new method of visible / near-infrared spectrum processing based on competitive adaptive reweighting algorithms-stepwise regression analysis (CARS-SWR) variable optimization is proposed in this paper.

The spectral data of five different tea varieties were obtained by visible / near-infrared spectrometry. The spectral data were preprocessed by the multivariate scattering correction (MSC) algorithm. First, 20 wavelength variables were selected by CARS, and then six optimal wavelength variables were selected using the SWR method, based on the CARS optimal variables. The generalized regression neural network (GRNN) classification model and probabilistic neural network (PNN) classification model were established, based on spectral information from the full wavelength, the CARS preferred wavelength variable, the SWR preferred wavelength variable, and the CARS-SWR preferred wavelength variable.

It was found that the CARS-SWR-PNN model had the best classification effect by comparing different modeling results. The classification accuracy of its training set and test set reached 100%. This shows that the CARS-SWR preferred variable method combined with the visible / near-infrared spectrum is feasible for the rapid and non-destructive identification of tea varieties. © 2019 Society of Chemical Industry.