Abstract: Moisture content in fresh tea leaves is an important index influencing tea quality during processing. In order to rapidly detect moisture content in tea during processing, a nondestructive method was introduced in this paper. Firstly, hyperspectral image data were captured from the fresh tea leaves. Secondly, four kinds of algorithms were used to preprocess the original data. Thirdly, the characteristic wavelength was extracted using stepwise regression analysis. Finally, a quantitative analysis model of the characteristic wavelength and moisture content in the fresh tea leaves was developed by the multiple linear regression and partial least squares regression. Experimental results showed that the best predicted effect of the partial least squares regression was obtained by the pretreatment of orthogonal signal correction after convolution smoothing and stepwise regression analysis. The correlation coefficients of the model calibration set, cross-validation set and prediction set were 0.8977, 0.8342 and 0.7749, respectively. The minimum root mean square errors were 0.0091, 0.0311 and 0.0371, respectively. Thus, hyperspectral technology could effectively detect the moisture content in fresh tea leaves, which would be useful in detecting quality changes in tea processing industry.

Key words: Hyperspectral technique, fresh tea leaf, moisture content, pretreatment