Determination of dry matter content of tea by near and middle infrared spectroscopy coupled with wavelet-based data mining algorithms

To explore the potential of near and middle infrared spectroscopy application in fast determination of dry matter content (DMC) of tea through the whole process from fresh tea leaf, semi-manufactured tea and to finished tea, samples from seven stages of the tea process were collect and the research was conducted based on data mining algorithms. Kubelka-Munk transform and spectral pre-treatment were adopted for elimination of disturbances caused by irregular appearance of intact tea in diffuse reflectance mode. A wavelet-based data mining algorithm composed of wavelet packet transform and statistical analysis (WPT-SA) was proposed to extract and optimize spectral feature from full-spectrum data. Another data mining algorithm of kernel principal component analysis (KPCA) was also employed for a performance comparison. Regression models were respectively established based on the full-spectrum data, wavelet spectral feature and kernel principal component. Statistical analysis revealed that the wavelet parameters (basis function and scale) were significant for these R^2 and RMSE of determination model and the optimization of wavelet parameters were vital for application of WPT. Modeling results showed that the regression model based on the wavelet spectral feature outperformed the other models, and the optimal regression model obtained a high R^2 of 0.9556, and a low root mean square error of 0.0501. These results indicate that it is feasible to measure DMC of tea in different processing procedures using near and middle infrared spectroscopy, and the proposed feature optimization algorithm (WPT-SA) is an effective data mining approach for enhancing the capability of spectral measurement.