"During a traditional quality inspection, a person checks all the products. Over the span of a week, that person isn't always equally alert and consistent. An employee may be sharper on a Monday afternoon than in the morning. And, come Friday afternoon, you can forget about accuracy. With digital quality control, a camera box consistently inspects the punnets from all sides, 24 hours a day, 7 days a week," begins Thomas Wennekers of Gearbox, explaining an advantage of digitizing quality control.
Some eight years ago, chances to use artificial intelligence in the horticultural sector led to the founding of Gearbox in the Netherlands. "AI is a comprehensive concept. For us, it primarily means image recognition, so we can use technology rather than humans to do inspections. We train algorithms to do quality checks on every conceivable type of fruit and vegetable's external characteristics."
"We're currently working with the largest products like grapes, soft fruit, tomatoes, cucumbers, and bell peppers. Besides fruit and vegetables, the GearStation also inspects various kinds of potted plants and cut flowers such as roses, gerberas, chrysanthemums, lilies, and tulips, both loose and in final packaging," says Thomas.
Digitalization
Though the technology resembles optical sorting, Thomas explains that digital quality inspections can assess many more external characteristics. "Most optical graders sort products by weight and a single external quality, like color, or being good or bad. We took that to the next level. For instance, our digital quality inspection can detect mold, nibbling damage, bruises, soft spots, color differences, soft fruits, and cracks in the fruit flesh."
Thomas believes digitalizing quality inspections offers several benefits. "Automating the whole process saves on labor, which is a must nowadays. Also, there´s an increased focus on the delivered quality, and digital quality control's byproduct is data," he says. Although it is a byproduct, this data can offer many possibilities. "It's still in its infancy, but that data is becoming increasingly important. More and more cultivation aspects are data-driven, but that data can also be the starting point for improvements."
The GearStation checks the punnets from all sides with a camera box
Training and targeted control
Thomas cites using data to train and manage staff. "You can combine the data with labor registrations, thus addressing harvesting quality with employees. We can then know which worker harvested which trolley in which path. And after the quality check, the picker and harvesting quality is known," he explains. The GearStation can also pick up things like damage caused by caterpillars or whiteflies. "The system knows in which greenhouse and which path the problem occurs. Growers can then combat the issue in a far more targeted way, namely only where a deviation was detected."
Gearbox uses deep learning to train the GearStation algorithm. "We use lots of examples to teach it to recognize things," says Wennekers, adding that plenty is involved in that. "Say I want to teach a computer to recognize a dog; I can say that it has four legs, a tail, two ears, and fur. If I then take a photo of a cat, it's recognized as a dog because it has the same characteristics. That's why we feed thousands of photos of dogs to the algorithm, and with the help of deep learning technology, the algorithm will look for characteristics that make a dog a dog. Ultimately, the system will be able to recognize a dog. That's where we excel: using deep learning to better recognize crop characteristics."
Snowball effect
While GearStation currently only does quality control, there are expansion plans. "We eventually want to add automatic robot packaging to the system. These are ongoing developments that I expect we'll soon be able to share. We're getting better and better at it, but the only step you can take with AI is to use it in practice, so the algorithms become increasingly accurate. The more you use the systems, the better the algorithms become, and the more accurate the result. Then, more and more people will start using it, which will have a snowball effect. We're really still on the threshold of AI in horticulture, and it will only keep growing," Thomas explains.
Thomas Wennekers
He considers digitizing greenhouses and fruit the first step toward future automated cultivation. "To make progress in horticulture, you must scale up. To afford innovations, large growers must grow even larger. AI-based tools require many hours of innovation, so come with a hefty price tag, making it very difficult for smaller growers to keep innovating." Thomas, though, recognizes the necessity. "Eventually, due to upscaling, one person won't be able to control an entire greenhouse alone, so it will have to be further automated. Autonomous cultivation is a big part of that," he points out.
Joint innovation
Because of this upscaling, Thomas would like to see digitization and AI placed higher on the agenda of, for example, the Dutch government and growers' associations and, above all, see more collaboration. "It's tough to innovate alone. Not all innovations succeed, so finding solutions together is vital. The investments are so large, and there's a risk that the desired result won't be achieved, so the only way to achieve this is to work together. Given the issues of our time – labor availability, the pressure on crop protection, increasing quality requirements, and high energy costs – I think developments are moving too slowly. If you want to respond to these aspects, some urgency is called for," Thomas concludes.
For more information:
Thomas Wennekers
Gearbox
Tel: +31 (0)174 295714
thomas.wennekers@gearboxinnovations.com
www.gearboxinnovations.com