Studies on ANN Models of Determination of Tea Polyphenol and Amylose in Tea by Near-Infrared Spectroscopy

The objectives of the present paper were to build the models for the determination of tea polyphenol (TP) and tea amylose (TA) in tea by near-infrared spectroscopy (NIR). According to the range of 7 432.3-6 155.7 cm~(-1) and 5 484.6-4 192.5 cm~(-1) of NIR spectra, the models are built for determining the contents of TP and TA in tea with the input layer, hidden layer and node ((8, 4, 1) and (7, 5, 1) respectively) in network structure by the artificial neural network. The correlation coefficient (r), the root mean square error of cross validation (RMSECV) and the root mean square error of prediction (RMSEP) were selected as the indexes for evaluating the performance of calibration models. The results show that r, RMSECV and RSECV by the model samples for TP and TA are 0.984 7, 0.460 and 0.123, and 0.947 0, 0.136 and 0.224 respectively, and r, RMSEP and RSEP by the prediction (samples) for TP and TA are 0.980 4, 0.529 and 0.017, and 0.968 2, 0.111 and 0.029 8 respectively. These indicated that the NIR-ANN models can be used to determine the contents of TP and TA in tea.