Baking Degree Discrimination and Quality Prediction of Tieguanyin Oolong Tea Based on Near-infrared Spectroscopy

Abstract:

[Objective] This study aims to assess the quality of Tieguanyin oolong tea during the baking process. [Method] Near-infrared spectroscopy was used for non-destructive detection of Tieguanyin oolong tea during the baking process and chemometrics was employed to analyze the changes in the intrinsic quality of the Tieguanyin oolong tea. Spectral preprocessing and dimensionality reduction were conducted for the near-infrared spectral data. Support vector machine (SVM) and back propagation neural network (BPNN) were adopted to construct discrimination models for the baking degree of Tieguanyin oolong tea. Combining the near-infrared spectral data with intrinsic quality data, partial least squares regression (PLSR) was employed to build quality prediction models. [Result] According to the changes in intrinsic quality and sensory evaluation results, the baking degree of Tieguanyin oolong tea can be graded into under baking, moderate baking, and over baking. Under the full-spectrum spectral data model, with preprocessing as multiple scatter correction (MSC) and the discriminant model as BPNN, the discriminant effect was the best, and the accuracy rate of the test set was 100.00%. When predicting the four intrinsic quality properties, the prediction model combining the successive projections algorithm (SPA) with PLSR had the best accuracy. The determination coefficients of prediction (RP2) of the best prediction model for free amino acids, tea polyphenols, catechins, and caffeine were 0.949 6, 0.944 3, 0.950 8, and 0.740 0, respectively. [Conclusion] This study realized the accurate discrimination of the baking degree and the rapid prediction of quality of Tieguanyin oolong tea, providing a theoretical foundation for the accurate discrimination and control of oolong tea baking.