Rapid detection of Pu-erh tea vintage by data fusion strategy based on Terahertz and Raman Spectral technology

The value of long-aged Pu-erh teas increases with time, but it is not easy to distinguish between them due to their similar appearance after fermentation. Therefore, this study was conducted to identify the vintage of Pu-erh tea using spectral fusion technology and chemometric methods. The study collected THz and Raman spectral data for Pu-erh teas produced in 2010, 2013, 2019, 2021, 2022, and 2023. The support vector machine (SVM) was used to create a classification model, and the zebra optimization algorithm (ZOA) was selected for parameter optimization to obtain the ZOA-SVM classifier. The data fusion strategy was then adopted to improve the model’s performance. The analysis of classification performance with low-level and mid-level data fusion indicates that the low-level fusion strategy reduces the model’s recognition accuracy and F1-score to 94.79% and 0.9482, respectively. In contrast, the mid-level fusion strategy improves the accuracy and F1-score to 98.95% and 0.9896, respectively. These results indicate that the information between terahertz and Raman spectral data can complement each other. Using a data fusion strategy combined with the chemometrics method can enable the quick and accurate identification of Pu-erh tea’s vintage, providing an effective solution for the safety detection and vintage identification of tea.