Evaluating Congou black tea quality using a lab-made computer vision system coupled with morphological features and chemometrics
In this paper Ren et al. explores the use of computer vision and machine learning to assess the quality of Congou tea. They start off by explaining the problem, then showing their approach to the solution.
Congou is a popular tea among consumers and used to be the base of the English Breakfast blend*. Most consumers cannot tell the quality of their tea, and so many ways of determining quality have been devised as an attempt to improve trust along the supply chain. These attempts focus on the taste, smell or appearance of the leaves. Most of the ones used today are destructive.
Ren et al. took a different approach. They acquired 100 samples from each class of tea, and spread it evenly over a surface using a vibrating apparatus. They then took a picture with a measurement device next to the tea in order to calibrate the sizes. After that they ran some image filters to assist in differentiating the leaves from the background and scaled the photo to the proper size. This allowed them to automate their data collection, and run it through analysis quickly. The software gathered data based on size, shape and colour of the leaves. This first experiment could not determine the quality of the leaves due to "complexities" (the authors implied it was due to the wrinkling of the leaves as they were harvested and dried).
The authors then tried an SVM machine learning model in an attempt to extract useful data from their samples. Using this approach they were able to get the accuracy of their training model to 100%.
In summary, there is a lot of distrust among consumers and tea companies as far as quality goes. Most methods of determining tea quality require the destruction of the leaves. Ren et al. determined that visual methods can work.
Guangxin Ren, Ning Gan, Yan Song, Jingming Ning, Zhengzhu Zhang,
Evaluating Congou black tea quality using a lab-made computer vision system coupled with morphological features and chemometrics,
Microchemical Journal,
Volume 160, Part A,
2021,
105600,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2020.105600.
Comments