The Weedcer image-processing system could prove to be an important tool for Norwegian agriculture and for the thousands of farmers who spray their fields with weed-killer every year. The program can distinguish between weeds and useful plants in real time during spraying, making it possible to spray only the plants that need to be eliminated.
The project involves Bioforsk Plantehelse, which has identified damage thresholds, i.e. the level of infestation required to trigger spraying, the agriculture technology and equipment manufacturers DAT and Adigo, and scientists from SINTEF, who have contributed algorithms and image-processing techniques.
Weeds with rounded leaves
“To be able to distinguish between weeds and other plants in the soil, we have identified characteristics and information in images that can discriminate and recognize different objects. For example, maize leaves are long and narrow, while weeds tend to have more rounded leaves,” says Kristin Kaspersen.
When a video camera films the soil and sends its images to the computer, they can be analysed in real-time, the different types of leaves can be recognized, and the amount of weeds and thus of weed-killer can be determined.
The next steps
The project, which has been underway since 2007, is an example of how pattern recognition can be used to solve problems in which the amount of data is too big to handle or where the data are diffuse and difficult to deal with. Adigo has produced a self-navigating four-wheeled robot that can drive around a field. The idea is to integrate the mechanism into a box on the spray boom mounted behind the tractor. Øyvind Overskeid of Adigo says that the new system will be ready for the market in 2012. The Research Council of Norway is financing development of the product.