Delphy: Applying Precision Farming to Greenhouses

270199d3-fb6e-4ded-8c8c-f7185ac910a3_12-ACH Autonome kas Illustratie Pascal verkleind_06494a18_570x377
Image courtesy Resource / Wageningen University & Research.

Last month, I started the course Precision Farming at Wageningen University. Precision farming is about applying the right dosage, at the right place, at the right time, which is easier said than done.

Over the past few weeks, we have had a few guest lectures from companies applying the principles of precision farming. One of these guest lectures was given by Klaas van Egmond from Delphy, a large agricultural company that does a lot in greenhouse horticulture. Klaas is part of the Delphy Digital team. In his presentation, Klaas told us about the latest developments in automation at Delphy.

The broad vision for the coming decade is that greenhouses will involve more robotics and sensors. But you probably already saw that coming. There are a few articles on this kind of thing on AlexGrowsUp. What exactly does this involve in Delphy’s case?

Digital Assistant

Klaas highlighted Delphy’s digital assistant, developed with AgroEnergy, which makes short-term management decisions for the grower based on sensor information, artificial intelligence, and market data. This frees up time and means that Delphy’s consulting services can focus on broader, strategic decisions. The digital assistant works on three time scales for decision making:

  • Every 5 minutes. This is for short-term climate management decisions. The models used here are quite ‘black box’, meaning we don’t really know what’s going on inside, as is the case with neural networks.
  • Every 15 minutes
  • Weekly. This uses transparent ‘white box’ models, so the logic behind these decisions can be understood by us mere humans.

The digital assistant also tries to do this whilst taking spatial variation into account. This is one of the big areas where current agricultural practices can improve yields and efficiency at the same time.

Delphy isn’t the only organisation working on automating greenhouse climate management. Last year, Wageningen University & Research organised the Autonomous Greenhouse Challenge, in which five teams tried to maximise yield in small-scale cucumber greenhouses. The catch was this, though: the teams were not allowed inside the greenhouse. Decisions could only be made through sensor data and models. Delphy was one of the participants in this challenge. There was also a team of professional growers as a control.

proxy.duckduckgo.com.jpeg
Image courtesy HortiNext.

The teams were evaluated as follows:

  • 50% on net profit
  • 30% on the level of the AI (determined by the jury)
  • 20% on resource use efficiency

Perhaps the most interesting outcome of the challenge was that the crops actually survived! Nobody was sure whether this was going to happen. But it did.

What’s more, some teams even got a yield higher than that achieved by the professional growers. Microsoft’s Team Sonoma came first, followed by Tencent and then Delphy. Interestingly, Sonoma won without using the complex sensors used by Delphy. As a result, Delphy are now looking into the real added value of each new sensor type, to simplify things. That said, Klaas claimed that although Delphy’s yields were not the highest, their system was more ready to be applied in practice. It was more robust due to a higher number of sensors.

Disease

One thing that remains hard to detect is disease. By the time the effects of pests can be seen, it is often too late.

Klaas presented a system that goes round the greenhouse, at a rate of around 100 rows per day, taking multispectral images and using laser probes – using a technology called IRIS (infra red interferometer spectrometer and radiometer). With all of this data, they are able to detect plant stress before symptoms are even visible. They can also count the number of tomatoes and their development stage, to give the grower an overview.

If a disease is detected, the system alerts the digital assistant and certain settings are changed accordingly.

Robotics

There wasn’t too much new here, but some interesting examples.

Harvesting tomatoes through robots remains a challenge. One of the challenges is that the clippers used to cut the tomato trusses can accidentally cut the entire tomato stem. Klaas showed us a clever Israeli system that makes cutting the stem physically impossible. As you can see in the video below, clippers have to first go around the truss before cutting the stem. Simple yet clever.

 

Klaas also mentioned a drone that would fly around the greenhouse, hunting for mosquitoes. The drone would then kill the mosquitoes using its rotors. Due to reasons to do with animal welfare, we didn’t get to see a video of this.

Leave a Reply

Your e-mail address will not be published. Required fields are marked *

*