Data Science

Part of this research was supported by Grant-in-Aid for Scientific Research (B), “Construction of a Very Large-scale Analysis Platform for Photosynthesis and Material Production Processes in Rice and Search for Useful Genetic Resources”.

The most direct indicator of the growth rate of a crop is the total dry matter weight, or biomass, present in the above-ground area and the yield. These are the most important traits in the science of crop production. However, measuring them is labor-intensive. This is a major obstacle to the selection of high-yielding varieties and to the accurate determination of production yields at production sites around the world. We therefore considered the possibility of applying machine learning techniques, especially deep learning, which has been making rapid progress in recent years, to biomass and yield estimation. The first step in deep learning is to prepare a large set of images (in this case, photos of rice or soybean plantations) and their corresponding true values (in this case, actually measured biomass and yield). Using this data set, we then build a model to accurately estimate biomass and yield from the images. In order to build a better model, it is important to collect a large amount of data and to include a wide range of variations in the data.

In collaboration with various research institutions, we have collected images of rice plantations from around the world and the corresponding biomass and yield. We have collected thousands of images of rice plantations around the world and the corresponding biomass and yields. The deep learning model we built based on these data is named “Rice Scouter,” and when rice images are input to Rice Scouter, biomass and yield can be estimated instantly, whereas previously it would have taken a long time to measure them. This technology was developed in collaboration with a company, and is now available free of charge as a smartphone application called “HOJO” (iOS and Android versions). It is expected that this technology will promote the development of high-yield varieties and cultivation technology support in developing countries. We plan to further improve the accuracy of this technology and apply it to other crops such as soybeans.