Determination of tea polyphenols content by infrared spectroscopy coupled with iPLS and random frog techniques

The potential of infrared spectroscopy for fast determination of tea polyphenols (TP) of 14 cultivars of tea trees was investigated based on data mining technique. And the TP determination models were respectively developed for large leaf cultivars, middle leaf cultivars and all the cultivars. Interval partial least squares (iPLS) was proposed to extract and optimize feature from full-spectrum data. Regression models were respectively established based on PLS, iPLS and biPLS. Modeling results showed that the model based on the biPLS with the optimal subinterval selection (2452-dimensional wavenumbers) outperformed the other models, and the optimal regression model was obtained with high validation correlation of 0.9059, and low RMSE of 1.0277. On the basis of the optimal subinterval selection from biPLS, a further excavation of characteristic wavenumber was done by random frog. Thus, 18 optimal wavenumbers were selected for the TP determination, and the corresponding linear formula of the TP measurement was established. The results proved the feasibility of infrared spectra for measurement of the TP content of tea.