Nondestructive measurement of moisture content of green tea in primary processing based on diffuse reflectance spectroscopy

Moisture content of tea is an important index affecting processing quality of tea. To realize fast measurement of moisture content of tea in processing, this paper put forward a nondestructive way to measure moisture content of green tea in primary processing based on diffuse reflectance spectroscopy technique. A visible-near infrared (Vis/NIR) spectroradiometer was adopted for scanning diffuse reflectance spectra of 568 samples in the range of 325-1 075 nm wavelengths. These samples were from eight procedures in primary processing, moisture contents of samples were immediately measured after spectral scanning. For obtaining high-dimension spectral data, wavelet transform (WT) was used to reduce of dimensionality and extraction of wavelet coefficients. The capability of low-frequency wavelet coefficients was evaluated for extracting spectral characteristic, and the result indicated that it was effective to mine characteristic information from spectra by wavelet transform. Three regression algorithms including partial least square (PLS), artificial neural network and least square support vector machine (LS-SVM) were used to develop models for determination of moisture content respectively. It could be found that LS-SVM model obtained the optimal result with rc =0.9985 and rv =0.9875. These results indicated that it is feasible to measure moisture content of green tea nondestructively and fast based on diffuse reflectance spectroscopy, WT is an effective method for extraction of characteristic from spectra, and LS-SVM algorithm can be broadly used for regression analysis with high precision and strong generalization.