Weed control with AI instead of chemistry

Autonomous systems help replace pesticides

The agricultural industry is faced with a big challenge when growing crops, as regulations are increasingly restricting the use of pesticides. Spraying such chemicals has often been unavoidable as a way to prevent weeds growing up next to the crop, competing with it for nutrients and leading to lower yields.

A crop plant that suffers major yield losses without active weed control is the sugar beet. The crop is the most widely cultivated sugar plant in German latitudes. Almost 30 million metric tons of sugar beet is harvested in Germany every year. Its importance as a raw material is often not appreciated: in addition to sugar production, various by-products of the sugar beet are used to make feed products, and it is also an excellent starting product for the manufacture of biogas and bioethanol.

Researchers at the Development Center X-ray Technology EZRT at Fraunhofer IIS are therefore working with project partners from the industrial sector on an environmentally friendly and sustainable alternative to herbicides. As part of the BlueBob project, initiated by the seed producer Strube D&S GmbH, the project partners are developing an autonomously navigating field robot that will use state-of-the-art sensor technology, intelligent algorithms and active weeding tools to remove weeds from within crop rows. In combination with the use of conventional weeding tools between crop rows, this will make it possible to attain comprehensive mechanical weed control – which can reduce the use of herbicides in sugar beet cultivation and even eliminate it in the long term.

The biggest challenge for the researchers was to teach BlueBob to precisely differentiate between sugar beets and weeds. To solve this conundrum, the Fraunhofer researchers are employing a combination of special cameras for optically capturing the plant parts in conjunction with an AI algorithm developed especially for this application. Using machine learning techniques, the robot thus decides within fractions of a second where within the row the weeding tool should do its work. The tool is then deployed with centimeter precision as it removes the weed and spares the crop.

The project is a collaboration between Fraunhofer IIS, Strube D&S GmbH and the French robot manufacturer Naïo Technologies, which developed the robot platform with its mechanical components.

“Fundamentally, we can employ deep learning methods to train the algorithm we’re using to work for any crop. This will make the technology useful in all weed scenarios with which farmers are faced,” explains deputy head of the Contactless Test and Measuring Systems department, Oliver Scholz.