Two Saturdays ago I went to TechnoHouse.
If you think that sounds like a big party, you’re not the only one. In some ways it was (that’s definitely what they were going for). But to put it broadly, TechnoHouse was an event organised at Rijk Zwaan’s headquarters in Westland, for sharing knowledge, networking, and showcasing the latest in breeding vegetable crops and greenhouse horticulture.
The day started with an enthusiastic welcome from Jim Stolze, our host. Jim described himself as quite the nerd, involved in artificial intelligence and that sort of thing. He knew how to liven up the crowd, just like you’d expect from the event’s name.
The Q&A set the stage for the rest of the day. One of the big themes was innovation, in terms of the whole industry. The Dutch are world leaders in greenhouse horticulture, but the competition in Silicon Valley, Israel, and the Far East is rising. Got to stay on top of things.
Rijk Zwaan is constantly innovating. The general tone of the Q&A was: “Thinking of working at a big electronics company like ASML? Come here instead.” There’s plenty going on.
We then went off to our separate presentations, done in small wooden greenhouses with headphones, like a silent disco.
Deep Learning Landscape – Adel Zakirov
The first presentation I registered for was about machine learning at Rijk Zwaan, from a practical perspective. Rijk Zwaan uses a lot of machine learning, for things like image processing and phenotyping their new varieties. Soon they will be using machine learning for genetic information as well.
Adel started off by telling us that if it’s written in PowerPoint, it’s probably artificial intelligence. If it’s written in Python, it’s probably machine learning. ‘Artificial intelligence’ is a vague buzzword. Deep learning is a powerful sub-category of machine learning that uses artificial neural networks.
2-5 years ago (says a lot about the pace of deep learning developments), the approach used for neural networks used to just be ’stack more layers’. Now, however, there is a more methodical approach.
Adel walked us through Rijk Zwaan’s approach to deep learning models, and the software used to do this. To train the model, a lot of images have to be annotated. Rijk Zwaan uses software called Supervisely (owned by Rijk Zwaan) to make this less tedious. Other software used included GitLab and DVC for version control, PyPI architecture, and MLflow to track model performance. Once a model has been developed, it continuously gets refined and updated. This finetuning is what takes the most time.
At the end of this presentation, we went to the next session, to the tune of ‘I Like To Move It’, played by our DJ.
Design Thinking – Vivian Baas
The next presentation was less technical, but at least as valuable. Vivian has a background in industrial design and worked outside of greenhouse horticulture for a few years before joining Rijk Zwaan.
In her presentation, she explained a few universal design principles. One of these was prototyping. Prototyping doesn’t have to be expensive. Her example of this was the layout of every McDonald’s kitchen, which was tested using chalk mockups. Cheap to do, but extremely influential on a large scale.
Rijk Zwaan has clients all over the world, so context always has to be taken into account. Low-tech or high-tech? Traditional or progressive? Experienced or novice? Data-driven or based on gut feeling?
You’d be forgiven for thinking Dutch growers would be the most receptive to new technologies, but in reality the opposite is true. Since the Dutch industry is so advanced, a lot of growers can achieve impressive yields on gut feeling alone. This makes them sceptical of data-driven decision support. They see it as a burden. However, since the hardware is already being used in most greenhouses anyway, not logging all this data would be a shame. Thinking about user-friendliness is crucial for companies like Rijk Zwaan, to get growers to log more data.
Another anecdote Vivian shared was that the light green text used in Rijk Zwaan’s apps was very hard to read in the Indian sun. Just a small change made a huge difference.
Seeds and vegetable varieties are the foundation of everything in horticulture. They affect everyone, from the nursery, to the grower, to the supply chain, to the consumer. Lots of users to take into account. The same applies to the software developed by Rijk Zwaan. Every user has different needs.
In this session, I spoke with a few people about Rijk Zwaan and other companies in Westland.
Contrasting Dutch growers, a new grower in say Kazakhstan has zero experience, making them more open to using data. Even if the data-driven approach is only 80% as good as a Dutch grower’s intuition, that can be very helpful for inexperienced growers.
I also met Leo Tetteroo, who I spent a day with at Priva back in 2016 for a two-day mini-internship. Amongst other things, we discussed machine learning being done at Priva, including the TopCrop project. In this project, plants are being monitored from above, using cameras.
In-House Software Development – Niek de Jong
Niek’s background is in business and IT. He showed us how Rijk Zwaan develops its software.
Although Vivian’s anecdotes were about software for Rijk Zwaan’s clients, a large proportion of the software developed by Rijk Zwaan is for Rijk Zwaan’s own employees.
Niek showed us a software development environment called OutSystems. Outsystems is a low-code platform that makes it easy to develop apps for different platforms. It’s very visual, with lots of dragging and dropping.
One of the challenges Niek highlighted was that when more apps get developed, they tend to get overlaps in functionality. Despite this, they are not always connected to each other. It’s a big mess. Because of this, it is important to make software as generic and universal as possible. It can be tempting to take shortcuts to start with, but keeping it generic prevents a lot of headaches later on.
Our tour started at Rijk Zwaan’s warehouses, where an incredible amount of seeds are stored. To give you an idea, they are stored in ‘light’ polystyrene boxes that weigh over a hundred kilos (that’s the empty box, no seeds). All of Rijk Zwaan’s seeds – wherever in the world they are sold – go through one of Rijk Zwaan’s two warehouses in The Netherlands.
Their brand new state-of-the-art warehouse looks like what you’d see at Amazon or Alibaba: a huge room full of crates being moved around by robots. I wish I could show you a video, but there was nothing I could find online!
The warehouse is climate controlled to improve the seeds’ lifespans. Not needing to take human comfort into account allows for more options when it comes to climate control, though there are still some minimum safety requirements. The warehouse has only one climate zone, since all seeds pretty much have the same requirements.
The next thing we saw was a machine that sorted seeds based on size: a huge time-saver. The main principle behind the machine was a tray with holes. The machine works for all crops’ seeds. All that has to be changed are some settings and the holes in the tray.
We then went to an area where the seeds were packaged into bags and labelled. Although the seeds are counted individually, the bags are also weighed to prevent mistakes. There is a fine balance between being efficient with seeds and underdelivering – and efficiency is important, since a bag of cucumber seeds could cost €1000!
Human Intelligence Session & Closing Remarks
Everyone got back together in front of the main stage, for the final session, called ‘Human Intelligence’. Henk van Lenteren, seed quality manager, was invited up on stage.
Henk has been working at Rijk Zwaan for over 40 years and knows the ins and outs of the company like few other people. As seed quality manager, he is involved in quite a few teams, like breeding and production. His job is to connect different specialists, which will become an important task for companies doing complex things like Rijk Zwaan (dare I use the overused term ‘multidisciplinary’?).
What about artificial intelligence? Will there be anything left for us to do at all? Henk’s conclusion was that artificial intelligence will and should be used as a helper for human intelligence. The art of breeding is still an art – we just have more tools at our fingertips to do it than ever before.
All in all, TechnoHouse was an amazing and encouraging day. Many thanks to Rijk Zwaan for inviting me and others to your headquarters and for organising a fantastic show!