Machine vision technologies can help improve food production right at its roots by helping farmers identify and kill weeds in their fields. Researchers at Wageningen University have developed a system that automatically does just that; such a system could help raise productivity, lower costs and protect the environment. The researchers, who work in the university’s Farm Technology Group, set out to develop a system that would recognize and attack a weed called the wild potato, which can be a nuisance to sugar beet farmers. When crops are rotated from potatoes to beets, the leftover potato plants that grow back wild during sugar beet years release nematodes and also can spread diseases across the field. The weed-fighting system can distinguish between intentionally cultivated crops, in green, and weeds, in red. Images courtesy of Wageningen University. A home garden hobbyist can dig up wild potatoes by hand, but in a large multiacre field, this is impractical, if not impossible. Large-scale beet farmers often must turn to sprayed herbicides, which naturally will come into contact with the beets as well as the potatoes. The researchers’ aim was to automatically distinguish the potato weeds from the beet plants, so they turned to machine vision. Dutch imaging specialist Data Vision, a sales partner of Allied Vision Technologies (AVT) in the Benelux countries, helped the team design a portable scanner system that can be towed over a field by a tractor. When the scanner spots wild potatoes, a microsprayer releases herbicide over that area. This eliminates the need for spraying chemicals blindly over large patches of ground, cutting down on the farmer’s herbicide expenses and reducing the amount of pesticide sprayed onto the beet crop. The team knew that the image analysis system had to be adaptive. “On a mechanically planted field, the path of the furrows is a clearly defined constant,” said researcher Dr. Ard Nieuwenhuizen. “Anything growing between the furrows can only be weed.” But weeds also can grow right from the furrows themselves, sneaking in between the intended cultivated plants. Shown is a schematic of the weed-fighting system. So the researchers taught the software to tell the difference – in terms of IR properties and colors – between sugar beets and wild potatoes. Two Marlin F-201 industrial digital cameras from AVT – each equipped with 2-megapixel sensors, one in color and the other an IR-sensitive monochrome sensor with a 780-nm IR pass filter – distinguish the plants from the earth and then identify each as either weed or crop. Five Xenon lamps illuminate the ground below the unit, and a distance-measurement device on one of the trailer’s wheels pinpoints the image’s location. National Instruments hardware (NI PXI system with Virtex-5 FPGA) and software (NI LabView) are used to capture and analyze the images. The system homes in on plants that have grown outside the furrow but then also looks for possible weeds within the furrow. And it recalibrates itself every 10 m so that it compares adjacent plants only to each other: Nieuwenhuizen noted that, in nature, variations in ground properties including water and nitrogen content can result in varying color properties even within the same species of plant. When the system confirms the presence of a weed, it deploys the microsprayer, which dispenses herbicide in 5-µl-drop increments directly onto the wild potato’s leaves with a precision of ±15 mm. Although the team has declared the prototype a success, the system needs further development before it can be marketed and distributed commercially. The researchers want to teach it to fight various kinds of weeds. To address the needs and wishes of organic farmers, another research team at Wageningen University and Research Center is working on another weed-recognition system that will mechanically remove invasive plants instead of spraying them.