Rapid and real time detection of black tea rolling quality by using an inexpensive machine vison system

Rolling is a crucial step in black tea manufacturing. Rolling is achieved through a rolling machine, which achieves the processes of crushing, tearing, and curling. At present, the lack of intelligent detection for black tea rolling quality has limited the development of automation in the black tea industry. This study established qualitative and quantitative prediction models based on machine vision technology and chemometric methods for detecting black tea rolling quality. High-performance liquid chromatography (HPLC) was employed to quantify the contents of catechins and theaflavins (TFs). A qualitative discrimination model for rolling time was constructed using the DarkNet-53 convolutional neural network (DarkNet-53 CNN) with a dynamic learning rate, while a quantitative prediction model for TFs content was developed using the improved sparrow search algorithm optimized support vector regression (ISSA-SVR) during the black tea rolling process. The results indicated that, compared to other models, the DarkNet-53 CNN model presented in this study achieved superior discrimination of rolling time, attaining an overall accuracy of 97.82%. The ISSA-SVR model demonstrated considerable advantages in predictive performance, with an Rp value exceeding 0.9 and an RPD value surpassing 7.5. Therefore, this research introduces a low-cost and reliable method for rapid detection of black tea rolling quality for the first time.