Self-learning sprayer not yet fully practice-ready

Selectively spraying volunteer potatoes in beet fields is a time-consuming process. In order to automate this, Wageningen University & Research (WUR) in the Netherlands developed an autonomous spraying system that is able to recognise and simultaneously treat the volunteer potatoes.

At this moment, the system is not yet as accurate as it should be. During the Innovation Day ‘Dág onkruid!’ (‘Goodbye weeds!‘), where the system was demonstrated, we could see a large percentage of beets also being sprayed by the machine.

Cameras and data analysis

The sprayer uses images provided by cameras and data analysis to separate weeds from crops, after which it should be able to perform spot-wise application. WUR took the machine to a field where volunteer potatoes were deliberately planted.

The fact that during its deployment the sprayer touched too many beets shows that further development is required.

Text continues underneath image

The system uses a narrow spraying boom, with 10 cm of space between the spraying tips, and 4 cameras. Self-learning software should be able to separate beet plants from volunteer potatoes. – Photo: Matthijs Verhagen

Useless dataset

The machine’s task was made a lot more difficult due to the fact that many plants were affected by insects and drought, and looked different than those of last year. This makes last year’s dataset basically useless.

WUR wil continue to develop the system, so it will eventually be an alternative to manual labour. Right now, at the testing facility in Valthermond, € 150 per hectare is spent on manual labour to remove volunteer potatoes. It is expected that 2 or 3 rounds are necessary to remove all the volunteer potatoes from the fields. That adds up to around € 300 tot € 400 per hectare.

Robotti field robot

WUR demonstrated the self-learning sprayer together with the recently purchased Robotti field robot. This robot is able to navigate autonomously and find its way in the field. The sprayer in this case was mounted on the robot. It could also be mounted onto a tractor.

It will take some time before we can see the self-learning sprayer work in the fields. The software needs to be refined, and there should be less distance between the nozzles, in order for the machine to work more precisely.