AI in Plant Breeding.
Ever since I joined KWS, I have been concerned with the question, how we can use cutting-edge technology to improve plant breeding. A good example for this is an ongoing project for the Vegetables business unit, which aims to automate the phenotyping of pepper fruits with the help of artificial intelligence (AI).
Digitalizing the phenotyping of peppers proves to be somehow challenging in practice. After all, the pepper types which KWS breeds are very diverse. In the future, AI and computer vision will support the colleagues at our breeding stations by automatically measuring and analyzing the fruits by using images. Let’s have a closer look at how that works exactly.
During the development of a new variety, breeders and their colleagues from Product Management and Sales align and update the market requirements. This helps to fine tune breeding program goals and ensure new products match the value chain expectations. Traditionally, the process of phenotyping that breeders conduct – where traits like weight, shape, and flesh thickness are being measured – is manual and time-consuming. In our project, we are developing an AI-based application that automatically measures and analyzes the fruits of different pepper types using images. AI enables us to reliably identify a potentially unlimited number of fruits in an image. These can then be measured precisely using conventional image analysis. As a result, the tool reduces the manual workload, but above all, increases the quantity and quality of the data our breeders obtain.
“With AI, we can reduce the manual workload, but above all, it increases the quantity and quality of the data our breeders obtain.”
Learning to recognize peppers
Unlike other applications, AI does not simply apply predefined procedures to a problem, but independently develops rules that can be used to solve certain types of tasks. To do this, however, it must be trained with sample data. That is why we use images for training purposes in which the individual fruits and peduncles are color-coded. AI analyzes these sample solutions and gradually learns to correctly identify pepper fruits.
“In many ways, artificial intelligence learns like us humans: it must first have seen a variation many times before it can reliably recognize it.”
The results so far
The progress is pretty exciting. The AI tool we are working on in close exchange with our breeder colleague Pablo Miranda Fernandez, who works at the KWS breeding station in Almeria, shows the potential to reduce the manual workload of our breeding teams and to increase the amount of data we can collect. It not only speeds up the process but also enhances the accuracy and quality of the data, providing breeders with a more comprehensive basis for making decisions. In the coming period, we will focus on optimizing the tool and testing it extensively at breeding stations to ensure reliability for all pepper types.
Exploring other paths
Looking ahead, this tool is just the beginning. At KWS, we are already exploring AI-based phenotyping applications for other vegetables, such as cucumbers, tomatoes, melons, watermelons, and beans. AI won’t replace human expertise, but it will certainly enhance our ability to breed innovative, customer-focused varieties more efficiently. This journey has only just begun, and I’m excited to see how these new technologies will shape the future of plant breeding.
