Discriminant Analysis Of Production Regions And Tea Plant Cultivars,and Prediction Of Chemistry Components Based On Chemical Fingerprint

In China, tea, having a long history, is an important economic crop. But there are some imitations selling as real famous tea in market. To make a judgement of the imitations and real tea, this study carried out for discriminating the production areas and cultivars of tea. And then studying on the NIR with tea to get a quick-time determines of tea biochemical components.In classification analysis of green tea and wuyiyancha tea based on HPLC chemical fingerprint,25common peaks were sorted out for building fingerprint, and with a PCA, the classification accurate rate of the two categories was100%. In classification analysis of Chuan-Yu green tea and Zhejiang green tea, there were24common peaks sorted out. A clusting with PCA results was drawn, and there were three classes in the clusting, the first was chuan-yu green tea, the third was Zhejiang green tea and the second was intercrossed with both the two tea categaries.In discriminant analysis of flatten shaped green tea production regions based on HPLC fingerprint,40common peaks were sorted out when increasing a new HPLC fingerprint chromatogram condition. One discriminant function was obtained for Longjing tea and Flat, it had26variables, and the discriminant accurate rate was92.4%. Similarly, one discriminant function was obtained for Xhlj and Flat, but it had21variabkes, and the accurate rate was95.8%. Two discriminant functions was obtained for Xhlj, Qtlj and Yzlj, even they had36variables, the accurate rate was82.9%. The accurate rate got98.3%for the first-grade and second-grade protection zones of Xhlj with one discriminant function and11variables.In discriminant analysis of flatten shaped green tea production regions based on NIR fingerprint, we developed an efficient procedure for validating the authenticity and origin of tea samples where Partial Least Squares and Euclidean Distance methods, based on near-infrared spectroscopy data, were combined to classify tea samples from different tea producing areas. Four models for identification of authenticity of tea samples were constructed and utilized in our two-step procedure. High accuracy rates of98.6%,97.9%,97.6%, and99.8%for the calibration set, and97.2%,97.5%,97.8%,100%for test set, were achieved. After the first identification step, employing the four origin authenticity models, followed by the second step using the Euclidean Distance method, accuracy rates for specific origin identification were98.4%in the calibration set and96.8%in the test set. This method, employing two-step analysis of multi-origin model, accurately identified the origin of tea samples collected in four different areas.In discriminant analysis of flatten shaped green tea cultivars based on HPLC fingerprint,40common peaks were sorted out. The results for four production region were one function,20variables and81.7%accurate rate for Xhlj;3,23and94.1%for Qtlj;3,20and83.1%for Yzlj;3,19and93.6%for Flat. For all tea samples,3discriminant functions were obtained, they had35variables, and the discriminant accurate rate was86.4%.In discriminant analysis of flatten shaped green tea cultivars based on NIR fingerprint, with the samples manufactured of four cultivars (LJ43, LG, YS and WNZ),4models were established to identify cultivars by PLS. Their identification accuracy rate for calibration set were89.8%,90.9%,96.1%and99.5%, while87.1%,84.2%,96.1%and97.5%for test set, but the general identification accuracy rate for test set was65.6%. After the "first identification" through the combined analysis of the four models for identifying four cultivars, the rate for test set got74.2%, and the "second identification" with Euclidean distance method, the accuracy rate for test set got83.5%.In prediction analysis of flatten shaped green tea biochemical components based on NIR fingerprint and HPLC, linear regression models for7biochemical components by PCR and PLSR were built up separately with a best preprocessing method. And linear fitting equations of predictive values and true values for testing set were calculated. The results showed PLSR getting a better predicton. The best preprocessing methods for all components were Smoothing for GA, EGC, EC and ECG, MSC for C, CAF and EGCG. Linear fitting equation for GA was y=1.0405x-0.0064, and followed by y=0.9227x+0.0922for EGC, y=1.0044x-0.0029for EC, y=0.9572x+0.1425for ECG, y=0.9762x+0.0123for C, y=0.9476x+0.201for CAF and y=0.9823x+0.1864for EGCG Correlation coefficients were0.84,0.88,0.91,0.84,0.86,0.92and0.92in turn.The results in this article could be used for tracing back the origin of famous tea production regions and cultivas. And the prediction analysis results could be used for a quickly detection for biochemical components content with NIR. A comparative study for HPLC fingerprint and NIR fingerprint was carried out in this article firstly.