Enough feed for the cows without waste, thanks to computer vision
Floating Farm
AI in agrifood
We build models using data already captured in your barn, greenhouse or factory, taking you from camera to working model, as we did with computer vision for Floating Farm.
For Floating Farm, we used computer vision to monitor the feeding process and reduce feed waste
For the world’s first floating farm, in Rotterdam, we delivered an end-to-end data solution: from mounting and installing the cameras to implementing a computer vision model that provides real-time insight into the feeding process.
Why do we only find out afterwards how much feed, harvested produce or raw material was lost?
The milking robot, feed wagon, climate control computer and production line controls record what happens every day, but the data sits with four different suppliers. To see how much is wasted, you currently have to add it up afterwards. A model drawing on these sources shows the losses while there’s still time to make adjustments.
At the feed fence, in the greenhouse and on the production line, a model keeps watch
A clear view of the feeding process
For Floating Farm, we installed cameras and implemented an image classification model that provides real-time insights into the feeding process to help reduce feed waste.
Cultivation data on one platform
For tomato producer Agro Care, we are building a data platform with a data warehouse on Databricks, so all its data is available centrally for BI and data science.
Demand for fresh products
A model uses sales, promotions and weather data to forecast demand for products with a short shelf life, so you throw away less and run out of stock less often.
Data from your own machines
The Data Act gives users the right to data from connected machines, including agricultural machinery. We find out what data you can retrieve and connect it.
We start with one source of loss in the barn, greenhouse or factory
-
Choose one source of loss
Leftover feed, crop losses or rejects on the production line: we choose one and identify which machines measure it.
-
Collect and test the data
We collect the data from suppliers and test a model against previous seasons or production runs.
-
Where farmers already look
The result appears where livestock farmers, growers or operators already work, and we monitor whether the model remains accurate.
Does our farm or factory record enough data for AI?
A good fit if
A milking robot, feeding system, climate computer, sorting machine or production line controller has been recording data for years, and a recurring decision about feed, cultivation, planning or quality is currently made on gut feeling.
Not a good fit if
Little has been recorded, or the information exists only on paper and in separate spreadsheets. In that case, you start with recording data and a dashboard, not a model, and we’ll tell you that.
What determines the scope
The number of machines and suppliers the data comes from, how many seasons or production runs of historical data are available, whether a platform is already in place and who will manage the model afterwards.
Who owns the data from our machines?
It is often stored in the supplier’s cloud. The Data Act gives users of connected products a right to that data, and farmers and growers manage their permissions through JoinData.
Barn, greenhouse or factory: where can data deliver value first?
Livestock farming
Choose this if milking robots, feeding systems and sensors already measure a lot, such as feed intake and milk yield. A camera captures what sensors miss; see computer vision.
Crop cultivation and horticulture
Choose this if climate, crop and harvest data are spread across separate systems. Our advice: start with one data platform. That makes forecasting feasible later.
Food industry
Choose this if production line controls, the lab and ERP already record data for each batch. Product rejection, shelf life and customer demand are then good first questions for a model.
From cameras in the barn to a data platform for the greenhouse: our work in agrifood
Data Science Lab has worked with data and AI from Amsterdam since 2016 as a Databricks and Microsoft partner. We helped Floating Farm through our Data For Good initiative, and we’re building on Databricks for Agro Care.
Our first conversation focuses on one machine and the data it already records
Stef Stoelwinder
Business DevelopmentStef is your first point of contact. In our first conversation, we’ll choose one source of loss in the barn, greenhouse or factory and the data associated with it.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which source of waste in your barn, greenhouse or factory do you want to spot first?
Describe the source of loss and the machines or systems that already collect data on it. We’ll reply by email or arrange an initial conversation.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
AI in agrifood means a model learns from data that farmers, growers and food companies already collect: feed and milk data, greenhouse climate and harvest data, and factory batch data and lab results. This helps you spot waste earlier, forecast harvests or demand, or use a camera to help inspect products. These are among the AI applications we build.
Growers and processors in north-east Brabant can read more on our Den Bosch page.
More than they use. In a dairy barn, milking robots, feeding systems and sensors record how much a cow eats and how much milk it produces. In a greenhouse, climate computers, crop records and sorting machines capture temperature, water use and harvest data.
Dairy farmers also complete the KringloopWijzer each year, a calculation tool developed by Wageningen University & Research. In 2016, the dairy sector made it mandatory for farmers supplying a member of the Nederlandse Zuivel Organisatie. Among other things, it shows the nitrogen surplus, ammonia emissions and greenhouse gas emissions per kilogram of milk.
A camera sees things a sensor does not measure. For Floating Farm, we installed cameras and implemented a real-time image classification model for the feeding process. Much of the cows’ feed there comes from urban by-products, such as brewer’s grains, bran and potato peelings, so it changes every day.
The challenge is lighting and variation: changing daylight, dust and new types of feed all need to be included in the training set. Read more about computer vision.
The European Data Act has applied since 12 September 2025. Users of connected products, including agricultural machinery, have access to the raw and pre-processed data generated through their use and may share it with a third party. Derived or substantially enriched data is excluded.
JoinData is the independent data platform where farmers and growers manage their permissions: who can see which data. Your lawyer can assess what this means for your contracts; we work out what you can access technically.
In our blog about data in the F&B industry, we discuss predicting consumer preferences, managing stock and waste for products with a limited shelf life, and sharing information about product origins with customers.
In a factory, it usually starts with batch data from line control systems, the lab and the ERP system. To forecast customer demand, a model uses orders, promotions and the weather; read more about forecasting.
By monitoring it. A different variety, a new by-product in the feed or a changed recipe alters the input, which can make the results less accurate. This is called drift.
Decide in advance who receives an alert and who approves a new version. Read how this works technically in our section on MLOps. Management can sit with your team or with us, through management and continuous improvement.
When a scale, a fixed rule in the feeding computer or a colour sorter already measures what you need to know. Or when you mainly want to know what happened yesterday: a Power BI dashboard using existing data is enough.
If feed, harvests or batches are still recorded on paper or in separate spreadsheets, start with the data foundation. If a simpler solution fits, we’ll tell you.
Applications, the food industry or Brabant: what will you read next?
- AI applicationsAn overview of our AI applications, from data science and computer vision to business intelligence.
- Data in the food and beverage industryWhat data can offer food and supplement brands, from shelf life to provenance.
- AI consultant in Den BoschHow we help growers, processors and suppliers bring together their cultivation, production line and quality data.