Application of time series hyperspectral imaging (ts-hsi) for determining water content within tea leaves during drying
Chuanqi Xie,Xiaoli Li,Pengcheng Nie,Yong He
DOI: https://doi.org/10.13031/trans.56.10243
2013-01-01
Transactions of the ASABE
Abstract:This research investigated the feasibility of using time series hyperspectral imaging (TS-HSI) for rapid and nondestructive determination of water content in tea leaves. Hyperspectral images of tea leaves were obtained at different periods of drying across the wavelength region of 380 to 1030 nm. The reflectance value of the region of interest (ROI) was extracted with ENVI 4.7 software. Different preprocessing methods were applied to determine the best method based on the root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP), residual predictive deviation (RPD), and coefficient of determination (R-2) of the partial least squares regression (PLSR) model. The successive projections algorithm (SPA) was then used to identify the most important wavelengths and reduce the high dimensionality of the spectral data. On the basis of the four effective wavelengths (542, 709, 752, and 971 nm), an SPA-PLSR model was established. Among the different PLSR models, the multiplicative scatter correction-PLSR (MSC-PLSR) model performed best with the highest values of R-cal(2), R-val(2), R-pre(2), and RPD (0.979, 0.961, 0.968, and 5.616, respectively) and the 2 p lowest values of RMSEC, RMSEV, and RMSEP (0.033, 0.045, and 0.040, respectively). However, the SPA-PLSR mode, with only four input variables, was considered to be preferable for determining water content in tea leaves. The SPA-PLSR model obtained R-cal(2), R-val(2), R-pre(2), and RPD values of 0.938, 0.935, 0.946, and 4.292, respectively, and RMSEC, RMSEV, and RMSEP values of 0.055, 0.057, and 0.052, respectively. The results showed that the TS-HSI technique has potential to be a rapid and nondestructive method to detect water content in tea leaves at different drying periods.