Study On Determination Of Valid Components In Green Tea By Near Infrared (NIR) Spectroscopy

China is the largest country of the total yield of green tea in the world.Since the existing defects including much subjectivity in sensory results,wasting much time and expense in chemical analysis,detection processes can't meet the demand of rapid determination to the valid compositions of tea in manufacturing process and commercial process.Near infrared(NIR) reflectance spectroscopy is a fast,stable, accurate and nondestructive technique that can be employed as a replacement of time-consuming conventional physical and chemical analysis method in determination of food and agricultural produce.In the dissertation,green tea is the main study objective and the valid components in tea are the main testing indices using NIR.The main novelty and conclusion of the dissertation involve:1.NIR spectroscopy combined with partial least square(PLS) was applied to building the quantitative model to quantitatively predict the valid components content in green tea.In building model,the effects on the spectral preprocessing methods and principal components factors(PCs) to results were discussed.The correlation coefficient(R~2) between the predicted and the reference results for the test set is used as an evaluation parameter for the models:the free amino acids, the total polyphenols and the total antioxidant capacity(TAC) results R~2=0.9016, 0.9042 and 0.9124,respectively.The correlation coefficient(R~2) for the caffeine, the epigallocatechin gallate(EGCG),the epicatechin gallate(ECG),the epigallocatechin(EGC),the epicatechin(EC) and(+)-catechin(C) content models are,respectively,0.9676,0.9604,0.9531,0.9706,0.9212 and 0.9500.It can be concluded that many valid components in tea can be analyzed fast by NIR spectroscopy coupled with the appropriate chemometrics methods,and this real-time,at-site measurement will significantly improve the efficiency of quality control and assurance.2.Selection of the efficient wavelength regions in FT-NIR spectroscopy was used for determination of free amino acid content in green tea.In general,the performance of full-spectrum PLS model might be weaked due to information overlaps in full NIR spectral region.In this experiment,the content of free amino acids in tea was used for experimental target.In order to improve its precision and robustness,interval partial least-squares(iPLS),synergy interval partial least-squares(siPLS),backward interval partial least-squares(biPLS),and genetic algorithm partial least-squares(GA-PLS) were applied to selecting the efficient spectral regions.The four methods were able to produce better prediction models in relation to the full-spectrum model,and the models were simpler and easier. Experimental results showed that the performance of siPLS model was best in the four methods,and the optimal model was achieved with correlation coefficient R~2=0.9105 in prediction set.3.Spectral pretreatment could simplify the prediction models of catechin content in green tea.Redundant near infrared spectral information not linked with components might increase the complexity of model.Net analyte preprocessing (NAP) and orthogonal signal correction(OSC) were respectively used to pretreat the NIR spectra of green tea.In this experiment,the content of EGCG,ECG and EGC in tea were used for experimental target.The NAP and OSC preprocessing method can correct the spectral matrix,and remove the information that is orthogonal to the concentration matrix and can reduce the factors number of calibration.As a result,the number of EGCG,ECG and EGC PLS factors were decreased from 12,14,13 to 2,3,8 for NAP method,and 5,5,7 for OSC method, respectively.This indicates NAP and OSC can efficiently simplify PLS model but does not affect the model's predictive ability.4.Some internal quality index may be the result of concurrent effects of various valid components in green tea,such as TAC.There was complex nonlinear relation between the valid components and NIR spectral.Back propagation neural network(BP-NN),radial basis function neural network(RBF-NN) and least squares-support vector machine(LS-SVM) were implemented for TAC calibration models.The experimental result shows that compared with the results obtained by BP-NN,RBF-NN was an effective predictive method with higher precision,converge and speed.LS-SVM model was developed with a grid search technique and RBF kernel function.SVM was a novel learning technique based on the principles of structure risk minimization(SRM).Comparing to BP-NN and RBF-NN,SVM could not only enhance the generalization ability but also improve the prediction precision,and the correlation coefficient(R~2) reached 0.9691.The overall results indicted that NIR spectroscopy combined with LS-SVM models had the capability to quantitative analysis the valid component with high accuracy.This research offers a new idea to rapidly determine valid components in green tea, and there is also of great significant in improve the level of tea determination, purifying the tea market and maintaining tea brand in our country.