The use of crop protection products is under pressure. It is an important part of the EU’s Farm2Fork strategy, but in the Netherlands, too, efforts to reduce use have been underway for a long time. In the meantime, many agents are on the “Candidates for Substitution” list, agents that should no longer be used in the foreseeable future. In AGROS – Weed Control, we are working on solutions to minimize pesticide use through weed recognition combined with spot spraying. In most cases, it is about detecting a weed plant first and then spraying the very small spot of that plant.
The work falls into 2 parts. Developing the recognition algorithm requires a lot of images of the weeds and the cultivated crop, which must have been collected under a multitude of conditions. The next step is to annotate these images. This involves marking for each image what the weed is and what the cultivated plant is. A lot of work, in other words. With these images, the recognition algorithm is trained and loaded on the spot sprayer. The spot spray technique requires a spray boom with many nozzles at small distances that can deliver the pesticide drift-free, exactly in the right spot. This too is proving to be no easy task, but gradually the first products are entering the market.
In the research we use our own machine (see photo) with 3 RGB cameras under the hood and a data processing and control unit on the back for actuating the nozzles.
In the project we started with the recognition of potato storage in sugar beets. We are currently working on recognition of weeds in an onion crop. The intended control system is based on hoeing between the rows, then addressing the escaped weeds in the row by spot-spraying and spot-cropping. This not only prevents a temporary drop in the growing curve due to spraying, but also leads to a significant reduction in pesticide consumption. We hope to have this technique working next year in a pilot environment and to be able to collect the first figures on used pesticides. Expectations are high.