Classification and recognition of black tea with different degrees of rolling based on machine vision technology and machine learning algorithms

Rolling is an important process for the formation of sensory quality of black tea, and it is also the key process to lay the material foundation for the subsequent fermentation. Currently, in the production practice, the evaluation of the degree of rolling still mainly depends on the sensory indicators such as the color change of the rolled leaves. This processing suitability evaluation methods based on subjective experience have the problem of insufficient stability. Based on this, according to the color features of samples under different rolling time series, combined with the classification models, this study aims to establish the objective classification methods for the degree of black tea rolling. Firstly, this study employed a portable colorimeter and a machine vision system to synchronously capture the mean values of both local and global color features from samples at different rolling stages. Then, classification models are constructed based on local and global color information respectively, and the feasibility of using RGB, HSV, and La*b* color models to evaluate the degrees of rolling is investigated. The effects of different color feature variables on the classification accuracy of the models are systematically compared, and the performance differences between linear and nonlinear classification methods are analyzed. The results show that the best classification effect can be obtained by using B-S-a* feature variable combination and nonlinear classification method. Among them, RBF-SVM and RF are identified as the optimal modeling methods. The overall accuracies of the models are 97.5 %, and they show high precision and recall rate in the classification of samples with different rolling degrees. This study provides a quantifiable evaluation method for the rapid and objective discrimination of the rolling degree of black tea.