Field reconnaissance for checking the state of soybeans in the midst of summer - exhausting, but necessary work when removing new varieties.
Breeders have to wander daily under the scorching sun in critical periods of the growing season to find plants showing desirable features such as early ripening of pods. But, without having the opportunity to automate the detection of these signs, scientists cannot test as many sites as they would like to increase the time to eliminate new varieties into the market.
In the new study of the University of Illinois, scientists predict the time of the maturation of soybeans within two days using images from drones and artificial intelligence, which greatly facilitates work.
"The assessment of the pod maturity requires a lot of time and here it is often possible to make a mistake, since this evaluation system is based on the color of the pod, and there is a risk of incorrectly determining it," says Nicholas Martin, Associate Professor of the Department of Creeding in Illinois and the collaborator of the study. "Many tried to use snapshots from drones to assess maturity, but we are the first to find an accurate way to do it."
Rodrigo Trevizan, a doctoral student working with Martin, taught computers to detect color changes on images from drones collected in five trials, three growing season and two countries. It is important to note that computers were able to consider and interpret even the "bad" images.
"Let's say we want to collect images every three days, but once the clouds appear or it rains, which affects the quality of the pictures. In the end, when you receive data for different years or from different places, they will all look different from the point of view of the number of images, intervals and so on. The main innovation we have developed is how we can take into account all the information received. Our model works well no matter how often the data was going, "says Trevizan.
Trevisan used the type of artificial intelligence, called deep convolutional neural networks (CNN). He says that CNN is like a way to whom the human brain learns to interpret the components of images - color, shape, texture - that is, the information obtained from our eyes.
"CNN detect small changes in color, besides forms, borders and textures. For us, the most important was color. But the advantage of models of artificial intelligence, which we used, is that it would be quite simple to use the same model to predict another characteristic, such as yield or span. So, now that we have these models, people should be much easier to use the same strategy to fulfill many other tasks, "explained Trevizan.
Scientists say that technology will be useful primarily in breeding commercial companies.
"We had sectoral partners who participated in the study that would definitely want to use it in the coming years. And they made a very good, important contribution. They wanted to make sure that the answers are relevant for field breeders that make decisions choosing plants and for farmers, "said Nicholas Martin.
(Source: FarmTario.com. Photo: Getty Images).