A map-based system for patch spraying weeds - weed mapping

This paper reports research on the detection of weeds, the first step in the development of spatially selective treatments, using map-based information to control the application of herbicides. The potential of both visual (human) and automatic detection of weeds has been studied. The use of both spectral reflectance and image analysis techniques have not been found to be sufficiently robust to form a basis for weed patch detection in growing cereal crops. It is concluded that for the immediate future spatially selective weed control will need to be based on visual weed mapping. The project has identified at what times of the year it is possible to map important weed species visually and how it can be done from a range of vehicles (tractor, ATV, combine harvester). Alternative methodologies to input the information, using the AGCO Fieldstar touch sensitive screen and voice recognition software, have been explored. Accurate maps of weed distributions have been created which formed the basis of the subsequent spatially selective herbicide application resulting in herbicide reductions of 9-42%, depending on weed distribution.