Using various JMP functions, we investigated a method for linking image analysis data of crude tea leaves at the Kagoshima tea market with image data of raw tea leaves used on-site and steamed tea leaves during the manufacturing process.
Using image data of raw and steamed leaves, we were able to predict the appearance color of crude tea with high accuracy using PLS regression and neural models. By fitting the model, we were able to predict the unit price of crude tea with high accuracy based on the variety, bidding date, whether the tea was covered, and the saturation and color of the steamed leaves. In addition, the same tendency was observed when using data obtained from an image analysis app on a smartphone.
On the other hand, although there were missing values in the tea market image data, due to the bivariate relationship, there was no difference in the average unit price depending on whether or not there were missing values, and the tests for homogeneity of variance and Kolmogorov-Smirnov tests showed no difference in variance or distribution, so the missing values were considered to be completely random. Using the missing value imputation function of JMP, the accuracy of the neural model using the imputed data set was lower than that of the model excluding the missing values.
These results suggest that it is possible to predict the quality of the raw tea after shipment from images of the raw materials on-site. In the future, we will increase the number of samples to improve the accuracy of prediction and use the results to provide guidance on picking and manufacturing on-site.
