What is your goal?

In my research, computer vision, namely image processing, technologies are used to analyze natural and agricultural environments. The subjects of my research are wide-ranging, including meadows, forests, crops, soils, and post-harvest vegetables and fruits. Here, I will discuss discrimination between weeds and crops.

 

What made you start your current research?

When I was a student, I had an opportunity to help a farmer engaged in organic, agrochemical-free cultivation with his farm work. Agrochemical-free cultivation meant that we had to manually remove weeds as we were not able to use agrochemicals (herbicides). It was extremely tough to manually remove such a huge amount of weeds every day in the midsummer heat across vast farm lands (the farm lands in Hokkaido are particularly vast!). It was not only physically but also mentally draining. That experience led me to think that a machine or robot that can automatically remove weeds would make the work of farmers easier.

 

How is this related to our lives?

As it requires manpower, agrochemical-free cultivation is time-consuming and costly. Accordingly, many farmers currently have to resort to herbicides. By developing a machine or robot that automatically removes weeds, agrochemical-free cultivation will become more popular. This in turn will allow consumers to buy safer foods at reasonable prices, preventing sprayed herbicides from adversely affecting natural environments.

 

Specifically, what are the contents of your research?

Although we talk about weeds being automatically removed using machines and robots, what we are attempting to achieve is actually the same as manual labor by humans. Humans first see a plant on farm land using their eyes, then use their brains to determine if it is a crop or weed, and finally remove only weeds by hand. On the other hand, automatic weeding machines and robots take a picture of a plant using a camera instead of an eye, determine if it is a crop or weed using a computer instead of a brain, and then remove only weeds using a bladed machine or robot hand.

Weeds and crops come in diverse shapes, colors, and sizes of leaves depending on their variety (Fig. 1). Furthermore, even with the same kind of plants, the shapes, colors, and sizes of leaves are individually different. In addition, since the conditions of sunlight vary depending on whether the weather is nice or cloudy and whether the time is morning, daytime, or evening, images shot by the camera appear different. Plants are not artificial matter but natural phenomena. As outdoor natural environments, farm lands are not under uniform lighting conditions unlike indoor settings. To clearly discriminate between images of weeds and those of crops within such ambiguous and complex natural environments, advanced information processing technologies are required. Fig. 2 shows how crops and weeds were discriminated using images shot from a tractor as it traveled over rows (two rows) of crops.

Besides weeds, I am also studying image processing with a variety of subjects. Fig. 3 shows goya (bitter gourd), a specialty vegetable indigenous to Okinawa Prefecture, with which I have tried to estimate the ripeness and components thereof using image processing without cutting the vegetable (nondestructive inspection). Fig. 4 shows orange trees in Florida, United States, with which I am studying how to count the number of fruits (yield estimation) by discriminating them from leaves even while the fruits are still unripe (green). As described above, although my laboratory is in Hokkaido, my research subjects are scattered across Japan, including Okinawa, with some even located overseas, such as in the United States.

Fig. 1 A crop (sugar beets) and weeds (four types)

 

Fig. 2 Discrimination between crops and weeds (in rows of crops, green: crops; red: weeds)

 

Fig. 3 Nondestructive inspection of goya Fig. 4 Detection of green orange fruits