Online quantitative determination of Wuyi Rock Tea quality compounds by near infrared spectroscopy

Abstract: Wuyi Rock tea (Camellia sinensis) is a kind of Oolong tea. It typically is of high sensory quality in China. Its cultivation and processing have become pillar industries in northern Fujian Province, China. Near infrared reflectance spectroscopy (NIRS), a kind of online inspection technology, has been widely used in many areas for rapid qualitative and quantitative analysis. The purpose of this research is to apply NIRS technology to the routine online inspection processing of Wuyi Rock tea. Totally 70 Wuyi Rock tea samples were collected from 10 different local companies. Four chemical components including moisture, caffeine, tea polyphenols and crude fiber were measured in laboratory based on Chinese National Standard methods. The statistic analysis results showed that the distribution range of moisture, tea polyphenols, caffeine and crude fiber were 6.02%-8.09%, 9.07%-15.65%, 18.41-30.22 g/kg and 9.19%-16.62%, respectively. Their standard distribution values were 0.424%, 1.353%, 2.613 g/kg and 1.497%. TQ Analyst 8.5 spectrum metrology analysis software was applied to develop the model. Several mathematical pretreatment methods and model results were analyzed and compared. The results showed that moisture can be predicted by PCR model with First derivative and Savitzky-Golay smoothing pretreatment in spectrum range of 5110-4933 cm-1. The correlation coefficient and mean square error of determination for the prediction model of moisture were 0.9815 and 0.0812%, respectively. The tea polyphenols can be predicted by PCR model in spectrum range of both 4736-4644 and 5936-5712 cm-1 without derivative and smoothing pretreatment. The correlation coefficient and mean square error of determination for the prediction model of tea polyphenols was 0.7427 and 0.748%, respectively. The caffeine can be predicted by PCR model in spectrum range of 4678-4289 cm-1 without derivative but with Savitzky-Golay smoothing pretreatment. The correlation coefficient and mean square error of determination for the prediction model of caffeine was 0.8601 and 1.39 g/kg, respectively. The crude fiber can be predicted by PCR model in spectrum range of 4478～3857 cm-1without derivative but with Savitzky-Golay smoothing pretreatment. The correlation coefficient and mean square error of determination for the prediction model of tea polyphenols was 0.8840 and 0.580%, respectively. Compared with the requirement of deviation in Chinese National Standard methods, the moisture model was better than its standard's error. The caffeine model, tea polyphenols model and crude fiber model were not as good as their standard's errors. But they were good enough for the routine online inspection.