Identification And Discriminant Analysis Of Tea Based On Near-infrared Spectroscopy And Hyperspectral Technology

Tea is a kind of healthy beverage,the ancient story'Shen Nong tasted hundreds of herbs,gotten poisoned by seventy-two drags,and was detoxicated by tea' and the present scientific evidences which says that tea have various benefits such as anti-cancer,anti-oxidation,enhancing immunity,lowing blood pressure and blood sugar can prove this.With this concept has been widely accepted,global tea consumption has gradually increased,and the standardization of the tea attracts more and more attention.The accurate identification of tea categories,geographical origins,varieties,classes and the rapid non-destructive testing of biochemical compositions has been an important issue for tea producers and consumers,but it can be sovled by the green,accurate,simple,real-time and non-destructive spectrum technology.In this thesis,the near-infrared spectroscopy(NIR)and hyperspectral technology combined with different chemical metrological methods were applied to establish the relevant mathematical model for the qualitative and quantitative analysis of tea quality in order to provide a method which can be referred to in the future research,and promote the tea standardization process of our country.The contents and results which were studied are as follows:(l)The NIR spectral data of 370 tea samples which included six kinds of tea were collected by six different pre-treatment methods,and were analyzed combined with support vector machine(SVM)algorithm to establish and verify the identification model of tea categories.The results showed that the best SVM model which the identification accuracy,correlation coefficients(R2)and the root mean square error of prediction(RMSEP)were 94.78%,0.92,0.38 can be achieved when the number of tea samples in the calibration set and prediction set were 255 and 115,and the smoothing pre-treatment was employed with the penalty parameter(C)was 105 and the kernel parameter(g)was 0.0015 in the support vector machine model with the radial basis function(RBF)as kernel function.It demonstrated that NIR spectroscopy can be successfully applied to identify tea categories.(2)Taking near-infrared spectral datas of oolong tea from each origins(northern Fujian,southern Fujian,Guangdong and Taiwan)and southern Fujian oolong tea of different varieties(Tieguanyin and non-Tieguanyin)as inputs,combined with support vector machine(SVM)to establish the oolong tea geographical origin discrimination model and southern Fujian oolong tea varieties identification model.The results showed that the best models ean be both obtained based on 1st derivative and normalize scale pretreatmented spectral datas,the identifieation accuracy of validation set,correlation coefficients(R2)and the root mean square error of prediction(RMSEP)were 94.87%,0.87,0.35 and 100%,1,0,respectively.The identification results are significant,showing that the near-infrared spectroscopy has a good application potential on the identification of tea geographical origins and varieties.(3)Hyperspectral technology combined with support vector machine(SVM)classification theory was applied to identify the classes of Tieguanyin tea.Twenty spectral feature parameters was extracted based on the hyperspectral data of tea samples,including red edge amplitude,blue edge position,yellow edge area,red valley reflectivity and normalized difference vegetation indexes,etc.The penalty parameter(C)and the kernel parameter(g)were discussed based on the support vector machine classification model with the radial basis function(RBF)as kermel function by taking these spectral feature parameters as inputs.The identification of Tieguanyin tea classes model was constructed and verified.The best experimental results were obtained using the radial basis function(RBF)SVM classifier with C=106,g=0.0075.The classification accuracy rate of unknown class Tieguanyin tea was 92.86%,showed that hyperspectral technology can be efficiently utilized for rapid,nondestructive and accurate identification of Tieguanyin tea classes.(4)Establish the tea polyphenols quantitative models of Tieguanyin tea using partial least squares(PLS)multivariate statistical methods based on its near-infrared spectral datas and the actual values of polyphenol tested by the methods in GB/T 8313-2008.The results showed that the regression equation obtained by the second derivative combined with normalized scale pretreatmented spectral data is relatively good,but the accuracy is still low.The correlation coefficient(R)and the root mean squared error of cross validation(RESEC)were 0.73 and 0.91 with a larger error.The probabal reasons are as follows:tea polyphenol is a mixture of polyphenols,and the absorption band sensitivity of related groups is low and susceptible to interference,Oolong tea is semi-fermented tea,the composition structure of polyphenols is relatively complex which makes modeling more difficult,slightly less experimental sample volume affects the accuracy of model.