Tea grade evaluation is an important technical method to test the quality of tea, and scientifically building the tea grade evaluation model has an important significance. This paper took 102 Oolong teas as research object and built the Oolong tea grade evaluation model based on interior quality parameters with various characteristic value screening methods in combination with support vector machine algorithm. Meanwhile, by combining hyperspectral technology with chemometrics, this paper built the quantitative forecast model of particle swarm optimization back propagation neural network (PSO-BP) based on characteristic wavelength and sparrow search algorithm optimization least squares support vector machine (SSA-LSSVM) for characteristic quality parameters, and finally verified the chemical value model of quantitative forecast. The results showed that in case of parameter combination of ester catechin, simple catechin, tea polyphenol, aqueous extract, caffeine and epigallocatechin gallate (EGCG), the Oolong tea judgement model had the highest accuracy, the accuracy of training set was 97.22%, and the accuracy of forecast set was 93.33%. The sparrow search algorithm optimization least squares support vector machine (SSA-LSSVM) quantitative forecast model had higher forecast accuracy and lower root mean square error, and the determination coefficient of forecast set R2 ranged from 0.93 to 0.99. By randomly selecting the optimal six forecasted chemical values of 30 Oolong tea samples, the judgement accuracy reached up to 90%. In conclusion, it was feasible to accurately judge different grades of Oolong tea based on interior quality parameter combination, the forecast model based on hyperspectral technology could rapidly and accurately obtain the chemical value, and the forecasted chemical value could accurately judge different Oolong tea grades, which would provide a new analysis method and application example for scientifically judging tea quality and grade.
Full Text Finder