Rapid Determination Of The Quality Components In Green Tea Based On Near Infrared Spectroscopy (NIRS)

The tea sensory evaluation is of much subjectivity and the chemical analysis istime consuming and of high cost. As a result, detection processes can’t meet thedemand of rapid determination to the compositions of tea in manufacturing andcommercial process. Near infrared spectroscopy technique (NIRS) has the advantagesof non-destruction, rapid, convenience, environmental protection and real time onlineanalysis. It has been widely applied to quality assessment of agricultural products. ByNIRS to quality analysis of tea, is of great significant in improving the level of teadetermination and detecting various quality indexes rapidly at the same time.The samples110green tea were employed as the study object. NIRS combinedwith partial least squares (PLS) was applied to develop quantitative analysis modelsof moisture, total nitrogen, total crude fiber, water extract, tea polyphenols (TPP),caffeine and free amino acids (AA) of green tea by cross validation. The effects ofspectra of different sample shape, number of calibration sets and spectrapreprocessing methods to results were discussed. The optimal calibration models wereevaluated and the prediction performance was validated by independent validationsets.The NIR spectra of integrated samples, crushed samples and the green tea soupwere collected. The effect of the three kinds of shape on the models was discussed. Itshowed that the optimal model of each component was developed by crushed samples.The models of water extract, TPP, caffeine and AA established by green tea soup werenot ideal. By the calibration sets in the proportion of50%,60%,70%,80%,90%, themodels of each component were established respectively. The results showed thatwith the increase of the proportion of calibration sets, the root mean square errors ofcross-validation (RMSECV) reduced and became steady in the end. In the proportionof70%, the veracity of models of seven components was maximal and the RMSECVwas minimum with0.162,0.099,0.426,0.467,0.377,0.104and0.212, respectively.Spectroscopy pretreatment methods included non-preprocessing, multiplicationscatter correction (MSC), standard normalized variate (SNV), first derivative (1stDer),1stDer added MSC and1stDer added SNV. By internal cross validation: the moisturemodel with1stDer was the best, the models of total nitrogen and caffeine with1stDer+SNV were the best, the water extract optimal model was with SNV, theoptimal models of total crude fiber, TPP and AA were all with1stDer added MSC. Thedetermination coefficient (R2c) of internal cross-validation was0.9862,0.9324, 0.8707,0.9367,0.9334,0.9609and0.9238, respectively. The root mean squareerrors of cross-validation (RMSECV) was0.182、0.098、0.394、0.452、0.366、0.078and0.187, respectively; the determination coefficient (R2p) of externalvalidation was0.9847、0.9183、0.8598、0.9388、0.9221、0.9599and0.9318,respectively. The root mean square errors of prediction (RMSEP) was0.234、0.102、0.459、0.473、0.496、0.090and0.230, respectively. The ratios of prediction todeviation (RPD) was8.42%、3.50%、2.68%、4.09%、3.65%、5.34%and3.98%,respectively.The RMSECV and RMSEP of all models were less than0.5. The R2pexceeded0.9and RPD exceeded3%, except for crude fiber model with0.8598,2.68%, respectively. The results showed that NIRS could be used for quantitativedetermination of seven kinds of components in green tea rapidly. The determination ofmoisture content, caffeine in crushed green tea samples can be achieved highprediction accuracy and fast NIRS analysis, but the determination of total crude fibercan be only achieved rudeness estimation.