Preliminary Construction Of NIR Calibration Model For Green Tea Qualities

Near infrared spectroscopy has tended to maturation in theroy, technology and application after half century. Being capable of making nondestructive, rapid, high efficient, convenient and online analysis, NIR technique is used in many fields. And it is one of analytical technique which has made rapid progress.The calibration model of NIR possesses sample and constituent dependence when it is used for predication. To make use of the advantage of NIRs, the individual calibration models must be developed according to the corresponding sample status and constituents. Thus, the key work of NIR technique is to develop the robust and sccurate calibration models.The goal of this project was to develop calibration models for moisture, free amino acids(AA) and total tea polyphenols(TPP) of green tea by using different spectra pretreatment methods and regression methods, which could be capable of analyzing amount of green tea samples, and to reduce the analysis cost greatly and to increase the analysis efficiency and to accelerate the usage frequency of the instrument. Meanwhile, this project analysed the effects of different sample size on calibration models.Spectroscopy pretreatment methods were multiplication scatter correction (MSC), standard normalized variate (SNV), Svitzky Golay (SG), Normalization, first derivative and second derivative;regression methods were principal components regression (PCR) and partial least square (PLS). The results indicated that the PLS was the optimal method for calibration models of three components. It indicated that by internal cross validation : the moisture model with SG9 added first derivative was the best, the standard error of calibration (SEE), standard error of cross- validation (SEP) and correlation coefficient (r)of calibration set (C-set) and validation set (V-set)wereO.1486, 0.1685> 0.994, 0.9925;the AAmodel with SNV was the best, its SEE, SEP, SEE-r and SEP-r were 0.2866 > 0.9498 > 0.2481 , 0.9514;the TPP model with first derivative added normalization to unit length was the best, its SEE, SEP, SEE-r and SEP-r were 1.086, 0.8946, 1.093, 0.8344. TPP calibration model had less effect than moisture and AA calibration models.Compared four sample sizes 7g, lOg, 12g, 15g, the result is that: different sample sizes had little effect on calibration models when all samples had to no phenomena of transmission. According to effects of three calibretions, samples had some transmission when using 7g size.This project also studied the difference between the model using original spectra and that using average spectra. The result was that the former was better than the latter, and the former was more able to avoid error by man or instrument.About hundred green tea samples were chosed as materials. These samples had good represention in aspects of different sorts, figures and producing areas, and wide content range of three components(moisture, free amino acids and total tea polyphenols). Otherwise, the NIR instrument we used had leave unused by the time of this project started, because of no calibration model for tea. Thus, preliminary construction of NIR calibration model for green tea qualities was not only to make a basis for construction future models for other tea and other study in green tea, but also to accelerate the usage frequency of the instrument and avoid wasting resource.