Digital imaging technologies are widely used in industrial inspection, product quality testing, automation control and process optimization. Likewise, in agriculture, the use of machine vision can raise productivity, reduce costs and protect the environment in the process. For that reason, researchers at Wageningen University in the Netherlands have developed a novel system to automatically recognize and combat weeds in the field.

Battling weeds: swatting a fly with a bazooka
Just as weeds are barely managable in the average garden, they are no less unwieldy on farms. One typical example is the wild potato that grows in sugar beet fields. The plant in question results from leftover potato plants that then grow back wild in years when the crop is rotated to sugar beets. The potato weed releases nematodes and can also spread diseases such as phytophtora infestans across the field. In multi-acre fields, it is impossible to manually dig up each individual wild potato, necessitating chemical means as the only real solution: thus, to eradicate a few scattered weeds, the entire field must be sprayed with pesticides that of course also come into contact with the intentionally cultivated crop.

The Netherlands are known for their advanced and highly productive agricultural sector, occupying one of the top spots in global agronomy. Thus, it is no wonder that Dutch researchers have taken on the problem of weed eradication. This five-year-long project of the Farm Technology group of Wageningen University was part of a dissertation research project and was supported by the STW Technology Foundation in Utrecht.

Giant field scanner
The goal of the project was to recognize and to precisely attack potato weed plants without contaminating the entire field surface and, above all, the sugar beet harvest. It quickly became evident that an imaging system would be necessary to differentiate between plants in the field and accordingly to spray only those to be destroyed.

With the support of Dutch imaging specialist Data Vision, Allied Vision Technologies‘ sales partner in the Benelux countries, researchers developed a mobile system that can be hauled by a tractor.

This giant scanner captures the ground surface and recognizes the wild potato plants, which are then subsequently targeted and treated with a herbicide microsprayer.

Recognizing and targeting weeds
The entire system is mounted on a trailer that can be towed over the field.

During the first stage, the weed must be detected. To this end, two Marlin F-201 digital cameras by Allied Vision Technologies are engaged. Both cameras are equipped with 2 megapixel sensors; one of the cameras is a color version, while the other is an infrared sensitive monochrome variety with a 780 nm IR pass filter. The cameras serve two functions: first, they localize the plants against the earth background; then, they identify the plants either as weed or agricultural crop.

In a captured ground area of 150 x 18 cm, resolution of just over 1 pixel/mm is achieved (1.08 Pixel/mm). For optimal, comparable images, the cameras are installed in a bin that is open from below, allowing five Xenon lamps to provide controlled illumination. Furthermore, a distance measurement device is mounted on one of the trailer wheels, documenting the exact position of the image on the field.

To control the image capture and analysis, Dutch developers decided on National Instruments hardware (NI PXI system with Virtex-5 FPGA) and software (NI LabView).

Intelligent image analysis accomodates naturally occurring aberrations
Image analysis must first and foremost recognize the weeds. Supporting this task, the system is adaptive. "On a mechanically planted field, the path of the furrows is a clearly defined constant: anything growing between the furrows can only be weed," explained Dr. Ard Nieuwenhuizen. Still, weeds can also grow in the furrows themselves between the cultivated plants. Therefore, this system concentrates first on plants beyond the furrow and analyzes their color and infrared properties. Having learned the characteristic features of the wild potato, the software can differentiate between plant types on the furrow as well, even when located between sugar beet plants.

"One additional challenge for an imaging system in agriculture is that the ground properties, such as water or nitrogen content, aren't homogenous in nature, which leads to variations in the color properties of plants," according to Nieuwenhuizen. To circumvent this problem, the system recalibrates anew every ten meters so that only adjacent plants are compared to each other.

Once the undesired growth is identified, it needs only to be treated with the microsprayer. This device, new to the field of agriculture, sprays small quantities of herbicide in 5 microliter drop increments directly onto the leaves of the wild potato. Traveling at 3 kph, the sprayer works with a precision of ±15 mm.