Abstract:Fermentation is the most critical step in black tea processing. To address the limitations of traditional methods that rely on manual experience, a rapid and non-destructive detection approach for assessing black tea fermentation quality was developed using a data fusion strategy that combines computer vision and electronic nose technologies. The fermentation degree of black tea was classified into different levels based on the mass fraction of tea polyphenols, and a correlation was established with image and odor information. Qualitative discriminant models for black tea fermentation were developed using different data fusion strategies in combination with random forests (RF), K-nearest neighbors (KNN), and support vector machine (SVM) models, and these were compared with the single sensor models. The results showed that data fusion strategies integrated information from different sensor, providing more comprehensive data, and their discrimination result was better than that of a single sensor. The feature-level data fusion strategies extracted the eigenvalues of different sensors information, simplifying the model data and achieving the superior performance compared to data-level fusion strategies. Among them, the SVM model based on feature-level data fusion achieved the best classification performance, with a classification accuracy rate of 100% in the training set and 95.56% in the prediction set, realizing the rapid and accurate identification of different fermentation degrees of black tea.
