Evaluation For The Quality And Grade Of Black Tea Base On Near Infrared Spectroscopy Analysis

Near infrared reflectance spectra（NIRS）has been developed as a new analysis and research tool due to its characteristics of accurate,fast,efficient,non-destructive and green environmental protection.Black tea plays an important role as the exported production in China.So its quality and grade of rapid and accurate analysis needed to be addressed in the tea industry.Nowadays,more and more NIRS technique has been used for determining major contents in black tea,while still less for tea rating evaluation.Even though black tea ingredients prediction analysis model has been established,but not stability because of the spectrum are susceptible to sample preparation conditions,instrument acquisition parameters,spectral pretreatment,variable interval selection and so on.Therefore,in order to further improve the stability of the near-infrared analysis model and quick to judge tea level,200 parts of tea samples were employed for the nondestructive test by means of NIRS analysis techniques.Different chemical metrological methods（partial least-square method,spectra processing method,selecting efficient spectra regions method）and optimum sample condition method were taking into consideration when modeling.Optimized quantitative analysis models of total nitrogen,water extract,polyphones,caffeine,free amino acids,crude fiber and tea grade were established.At the same time path analysis method was first proposed to determine the key effective factor to level and a regression equation was achieved.The main results of the dissertation involve:1.Sample preparation and loading conditions were determined.Experiments indicated that the repeatability of spectral was best with the powder’s particle size of40-60 mesh,40 MPa pressure sample strength and 4-millimetre thickness.Based on near infrared spectroscopy technique and partial least squares（PLS）,models of the total nitrogen and caffeine content were established.Comparing with conventional sample preparation method,the prediction correlation coefficient R² from 0.9340,0.9765 up to 0.9666 、 0.9767 respectively;the root mean square prediction error RMSEP down from 0.0369,0.0325 to 0.0266,0.0325 respectively.The researchresults showed that the optimal sample preparation conditions was greatly increased the model prediction accuracy.2.Comparison of four kinds of spectrum pretreatment method on modeling.There were standard normal variate（SNV）,min/ max,first derivative（FD）,smoothing.Results showed that: total nitrogen,polyphenols,free amino acids can achieve the best prediction model with SNV method that correlation coefficient（R²）and the root mean square error of prediction（RMSEV）were,respectively,0.8762,0.8033,0.9501 and 0.07,0.51,0.12.While water extract,caffeine and crude fiber with first derivative method and smoothing method respectively were the best.The correlation coefficient（R²）and RMSECV were,respectively,0.646,0.984,0.9868 and 3.19,0.0365,0.113.Studies have shown that spectral preprocessing method could effectively eliminate the influence of non-target factors on the spectrum.3.A predictive model was established by selecting efficient spectral regions combining with the Partial least-squares（PLS）.The synergy interval Partial least-squares（Si-PLS）and genetic algorithm Partial least-squares（GA-PLS）were applied to select efficient spectral regions.Experimental results show that two methods were able to produce better prediction models in relation to the full-spectrum method.With GA-PLS method,the total nitrogen,polyphones,caffeine,free amino acids and crude fiber optimal model offers its R² in 0.9804,0.8982,0.9700,0.9938,0.9920 respectively by calibration set.The water extract get the best R² in 0.7304 and RMSECV in 2.82 by Si-PLS method.Near infrared spectroscopy combined with the characteristic spectrum area screening method can simplify the model and improve the prediction accuracy.The reseach provided a theoretical basis for the latter part of the near-infrared spectrometer at the wavelength range of options for purchase.4.The multiple regression equation relating the tea grade was achieved as:Y=30.189-314.191X₁+5.347X₂-713.852X₃+104.332 X₄+40.569 X₅,where Y is tea grade,X₁,X₂,X₃,X₄,X₅ is total nitrogen,water extract,caffeine,free amino acids,and crude fiber,respectively.The predicting correlation coefficient（R²）is 0.9055.Using path analysis showed that the five composition contents were closely related to the teagrade except for polyphenols,which laid a foundation for the building of tea quality grade model by NIR-PLS.5.The quantitative and qualitative analysis model of black tea level were established.Experimental results showed that correlation coefficient（R²）and the root mean square error of prediction（RMSEV）was,respectively,0.9741;0.323.Identification accuracy rate of the qualitative analysis model was up to 78.5% for unknown samples from the prediction set.A new idea by quick and precise identification of tea grade was offered in this dissertation.