Acousto-optic tunable filter near-infrared spectroscopy for in-line monitoring liquid-liquid extraction of Gardenia jasminoides Ellis based on statistical analysis.

This study aimed to monitor liquid-liquid extraction of Gardenia jasminoides Ellis (Zhizi in Chinese) using in-line near-infrared spectroscopy. Shanzhiside (SZS), deacetyl asperulosidic acid methyl ester (DAAME), genipin-beta-D-gentiobioside (GG), geniposide (GS), total acids (TA) and soluble solid content (SSC) were selected as quality control indicators, and measured by reference methods. Both partial least-squares regression (PLSR) and back propagation artificial neural networks (BP-ANN) were applied to create models to predict the content of above indicators. Paired-samples t-test and nonparametric test were used to compare differences in predictive values between two models of each indicator. Relative standard error of prediction (RSEP) and mean absolute percentage error (MAPE) were used to evaluate the predictive accuracy of the established models. The results showed that there was no significant difference in predicting DAAME, GS and TA between two models. However, PLSR model gave better accuracy in predicting GG and SZS than BP-ANN model. The BP-ANN model of SSC was better than PLSR model. This study shows that NIR spectroscopy can be used for rapid and accurate analysis of quality control indicators in the liquid-liquid extraction of Zhizi. Simultaneously, this study can serve as technical support for the application of NIR spectroscopy in the industrial production process.