Smart sorting with AI in greenhouse horticulture

Crux Agribotics digitizes and automates greenhouse horticulture. Ferry Mulders, Technology & Safety Officer, enthusiastically talks about "his" robots. "We're a bunch of passionate professionals who make things we're proud of."

Crux Agribotics develops vegetable sorting and packaging lines. The biggest challenge is that they work with unique products. "No two cucumbers are the same," Ferry explains. "It's incredibly complicated to teach a sorting machine to recognize a good cucumber." He immediately raises the distinction between Machine Learning (ML) and AI. "AI is an umbrella term these days. Everything is AI." He rightly points out that sorting machines usually don't use AI, but rather ML.

The big difference? "AI is trained by, for example, a neural network where the products are defined in the training phase and then learns itself what is right and wrong. Whereas ML relies heavily on large data sets, qualitative data, from which the mathematical data can be compared and adjustments made if necessary." The cucumber sorting machine learns from enormous amounts of data what a cucumber is, how to qualify it, and how to package it.

AI vs ML

Crux Agribotics collaborates closely with the greenhouse horticulture industry, giving it unique insight into the practical applications of ML and AI. One example of an AI application is the detection of cucumber stems. The algorithm learns to recognize the stem itself and how the machine can best handle it during packaging.

Ferry uses a tomato processing machine as an example of ML. "Using vision technology, we look at tomatoes and estimate their weight. That depends on the composition of a beefsteak tomato. Afterward, we weigh the entire package and feed that data back into the machine. With that feedback, we've developed an ML loop."

"Every thirty minutes, the machine receives new data. In that half hour, the sun's position has changed, and with it, the weight ratio of the beefsteak tomato." Every half hour, they adjust the algorithms to estimate the weight of the beefsteak tomatoes as accurately as possible. "That has everything to do with the minimum product requirements. A crate must always at least reach the required weight. The producer gives away any extra kilos for nothing." An accurate prediction therefore pays off. "This isn't AI. It's a mathematical model," Ferry emphasizes.

Future

AI is developing rapidly. In the future, Ferry wants to use AI to automatically adjust the machine's recipes and thus increase efficiency. He gives an example: "Machines work with a recipe, a specific input. No two products are the same. If a customer requests that the cucumbers be placed only crosswise, on the short side of the crate, then that has to be entered into the machine – the recipe." Using AI, Ferry wants to be able to advise customers in the future to adjust the recipe. For example, if only long cucumbers are available at that moment, which don't fit crosswise in the crate.

Production Process Automation event

Ferry gives the presentation Smart sorting with AI: accuracy under varying greenhouse conditions during the Production Process Automation event (PPA). View the full program at the website and discover where innovation meets the production floor.