Level: Beginner
【Presenter】
Kagoshima Prefecture Hokusatsu Regional Promotion Bureau Tsuyoshi Tomihama
[Presentation summary]
At the Kagoshima Prefecture tea market, in order to improve the quality of crude tea, the appearance and light color of the crude tea being bid for are photographed with a digital camera, and the numerical values obtained from the texture and chromaticity analysis of the images are captured on smartphones along with the unit price and images. Feedback is provided to each farmer. This time, we used JMP's various functions to examine a method for explaining and predicting the values of crude tea components, which greatly affect the taste and aroma of tea, from cultivation information and image analysis data.
We used a data set of 1,292 samples of Ichibancha tea, which were delivered to the tea market from various production areas within the prefecture, analyzed by image, and analyzed for composition using the near-infrared method after bidding. Principal component analysis using Ichibancha crude tea component values (total nitrogen, free amino acids, theanine, fiber, tannin, caffeine, vitamin C) revealed that the characteristics of the crude tea being bid for are mainly based on indicators related to total nitrogen and fiber. 74.4% could be explained by component 1 and principal component 2, which is an index related to tannin and caffeine content, and unit price had a positive correlation with total nitrogen and a negative correlation with fiber.
From the control chart, there are many cases where total nitrogen and fiber deviate from the control limits in the latter half of operation (mid- to late-ripening varieties), and the lower specification limit for total nitrogen is 5% content, and the upper specification limit for fiber is 22% content. The nonconformity rates were 4.9% and 6.4%, respectively.
In order to predict total nitrogen and fiber from crude tea image analysis data, we applied PLS regression and response surface models using cultivation information and 10 image analysis items as parameters. As a result, the values of both components were found to be significantly different depending on bidding date and variety. In addition, we were able to explain this with a model that incorporates the image analysis item ``white stems'' (an indicator of the ripeness of new shoots to be picked). Furthermore, with the design space function of the profile, if ``Main varieties of medium to late maturing are harvested and manufactured so that the white stems are level 3 or less by May 7th'', the proportion of total nitrogen and fiber within the specification will be 98.9%. It was predicted that it would be possible.
As described above, crude tea component values can be predicted from cultivation information and image analysis data, and a cultivation index that maintains within specifications has been obtained. We will introduce an example of how this information is utilized for guidance in on-site harvesting and manufacturing.
[Presenter profile]
Graduated from Kyushu University Graduate School of Agriculture in 1996
Joined Kagoshima Prefecture in 1998
[Co-presenter]
Kawasaki Kiko Co., Ltd. Kyushu Branch Shoichi Sato
