Interview with a Scientist who Promotes an Agritech Project Joined by 2,245 Teams Worldwide to Achieve a Safe and Reliable Food Supply

“In wheat cultivation and research, we count the number of wheat heads per unit of area to determine the yield. This is mostly done visually, so it requires a lot of time and effort. There has been research into models for automatic measurement of wheat heads, but they are based on limited experimental data, and they lack versatility. Creating a large-scale, highly versatile model for automatic measurement requires instant improvement in the diversity of our data. That is why we launched this project. First, we compiled more than 4,500 images of wheat that we labeled^*2 to build a Global Wheat Head Detection dataset^*3, and we made this available.

1. *2.The process of attaching correct answers to data to be used by AI for learning (annotation). For example, images of wheat are labeled with information such as the variety, soil condition, and climate conditions.
2. *3.A collection of data to be processed by a program.

• 

  Assistant Professor Wei Guo of the Institute for Sustainable Agro-ecosystem Services at the University of Tokyo’s Graduate School of Agricultural and Life Sciences. He came to Japan in 2006 and completed his doctoral course (Doctor of Agriculture) at the Graduate School of Agricultural and Life Sciences, University of Tokyo in 2014. He has been in his current position since 2019. He specializes in agricultural informatics and plant phenotyping.

“The sizes and photographing methods of the images of wheat we collected from various countries were different, so we first had to make the images consistent based on standards we set within the project. We also had to establish criteria for producing the learning data^*4, such as whether to include nodules (thornlike projections from the wheat tips) when labeling, or whether to label wheat heads when only part of them is visible. Without production criteria for learning data, we could run into problems such as one researcher labelling things that others did not label. To avoid this, researchers in each country sent labeled data to one another for corrections. They went back and forth many times with the project data to determine the production criteria.”

1. *4.Data used by AI for learning.

• 

  Image dataset of wheat heads that have been labeled (photos courtesy of Global Wheat Dataset)

“Although we promoted the competition at conferences and to the scientific community, I think it helped that we collected the models with Kaggle^*5, which is frequented by data scientists from around the world. It was a novel idea to have a competition themed around wheat, since most of them involve cars or people.”

1. *5.A platform for data science competitions where companies and researchers submit data and data scientists compete to create the best models for them.

“It was very informative to see the excellent models that the engineers created using their way of thinking. To improve the accuracy of image analysis, it’s important for us to think as engineers would think. If agricultural scientists like us can adopt the engineer way of thinking that data scientists have, I think we will be able to further accelerate agriculture that uses data to predict growth.”

“Building the image dataset and holding the competition were steps toward realizing smarter agriculture using AI on farms. To add more diversity to our research, I want to develop future activities by adding members from regions that are not currently participating, such as Africa, South America, and India. In the future, we hope to create a venue for providing big data on agriculture and encourage people in countries and territories throughout the world to take greater interest in farming.”

• 

  Dr. Guo’s laboratory is building a program that can recognize wheat heads in real time. It learns multiple images of wheat displayed slideshow style on an AI computing platform connected to a camera.

“I want to solve food-related problems. I talk about starting a revolution, but we need diversity that can be realized in many different countries and territories. What is a diverse revolution? What comes next after tractors? That’s what I have been trying to determine. The technology that would start a revolution has not yet been introduced; and when it is, cooperation of various companies will be essential so that it can be used in various farming circumstances. Kubota is one of those companies. I hope we can combine our cutting-edge knowledge and technologies to achieve the vision of ‘easier agriculture’ that enables stable, high-yield production of safe and secure food in any country or territory.”

• 

  “I envision farming that people can do even when they are asleep,” Dr. Guo says with a smile.

“The research being conducted with the Global Wheat Dataset project is part of what is necessary to make agricultural machinery more intelligent in terms of assessing the growth of crops and performing the work needed at each step. This is in line with Kubota’s vision for the future of farming – providing a stable supply of safe and reliable food. The image dataset created for this project along with the models that won the competition will be made open source so they can serve as a common platform that people involved in farming around the world can use to improve agriculture. And the efforts to build this common platform are being supported by Kubota in the form of sponsorship.”

“In addition to using the open source models and image datasets, we can apply these outstanding modeling methods and image dataset labeling techniques to our own development. And by using this dataset and its massive amount of data, we could dramatically accelerate data processing speed. For example, crop growth data collected over many years based on location information can be analyzed from multiple angles, including soil and climate conditions and watering methods, to predict crop growth with greater accuracy. This can help AI suggest what farming work needs to be done next. This is the vision for data-driven precision agriculture that we aim for at the Advanced Systems R&D Center. Another benefit is the ability to work closely with smart agriculture researchers around the world and experience the latest global technologies they are studying.”