From lab to factory: A calibration transfer strategy from HSI to online NIR optimized for quality control of green tea fixation

Abstract

Fixation is the key step of green tea processing, usually with moisture content as the quality evaluation indicator. Near-infrared (NIR) spectroscopy has been widely used in tea moisture detection but is mostly realized in the laboratory, and online detection in tea processing is limited. This study explores the possibility of calibration transfer from lab-based hyperspectral imaging (HSI) to online NIR to enable the rapid detection of moisture during the fixation step. Raw spectral data were preprocessed, and direct standardization (DS) and partial least-squares regression (PLS) were used for data calibration and modeling, respectively. The results showed that there were observable differences in the spectral data of the same sample obtained by different instruments. The prediction model based on HSI failed to be transferred directly to the online NIR and has a large root mean square error of prediction (RMSEP) value. Spectral preprocessing with a standard normal variate (SNV) transformation reduced the differences between the instruments. When 40 standard samples were used, the established transferable SNV-DS-PLS model based on HSI achieved the optimal prediction performance with a correlation coefficient of prediction (Rp), RMSEP, and ratio of prediction to deviation (RPD) of 0.94, 2.24%, and 2.76, respectively, for the 50 validated tea samples obtained from online NIR. In conclusion, the model transfer across instruments was successfully realized, and tea moisture detection was brought from the laboratory to the tea factory, which provided a reference model for online moisture detection for other agricultural products.