- English
Tom Meewisse: ‘Workers no longer need to judge manually’
Meewisse Roses has revamped its sorting process. The existing machine with the IRISS Roses vision system from 4More Technology, has been enhanced with additional cameras and AI software for defect detection. While IRISS Defects is still in its learning phase, the benefits are already evident, according to Tom Meewisse and coordinator Katharina Verheul. Sorting is now significantly faster and more accurate.
Advancing Sorting Technology
Until 2016, roses could only be automatically sorted by length, stem thickness, and bud size. The introduction of the Intelligent Rose Inspection & Sorting System (IRISS) by 4More Technology (4MT.NL) took the process to a higher level of precision, uniformity, and efficiency. Early last year, the high-tech company from The Netherlands took another step forward with the development of IRISS Defects—an AI-driven expansion module that rapidly detects defects such as bent stems and necks, mildew symptoms on the underside of leaves and ball centers, a bud deformation that prevents proper blooming.
Extra Actions, Extra Time
“Until recently, our employees had to manually inspect all roses while hanging them in the machine,” says grower Tom Meewisse. “If they noticed a defect, they had to adjust a coding switch above the fork to assign the rose the correct downgrade code, ensuring it was directed to the right bunching station. Experienced workers are quite skilled at this, but the process always takes extra time—especially if the flower needs to be flipped to check the underside of the leaves.”
Automatic Defect Detection
The Bleiswijk-based grower of ‘Red Naomi’ roses did not hesitate when 4MT asked if he was interested in testing a new development: an upgrade of IRISS Roses with a Deep learning module for automatic defect detection. “We already knew from our customers that defect assessment is a limiting factor in the sorting process,” says Tim van der Elst, owner and operations manager of 4MT. “Tom is open to innovation and understands that AI implementation requires time and support.”
Training the Software
Co-owner and head of R&D Wouter Vreugdenhil adds, “You have to train the software to recognize anomalies detected by the cameras. Some are variety-specific. For example, Red Naomi is relatively prone to bud deformation, and mildew spots on the underside of leaves can be harder to identify. Initially, a lot of manual work is required to guide the software in making the right decisions. For most parameters, we’ve gathered and input sufficient data, though some are still in the learning phase. It’s still a ‘work in progress,’ but IRISS Defects is ready for broader implementation in the rose industry. IRISS Roses and the defect module can be installed on all modern sorting machines, regardless of brand. Or we can supply a complete sorting machine that includes the IRISS roses vision system”
Significantly Faster and More Accurate
“The sorting process is still improving,” confirms Meewisse. “It’s not flawless yet, but the speed has increased significantly, and fewer mistakes are made compared to manual assessment. Workers now
only need to hang the roses, while the vision system handles the inspection. We could theoretically reduce our sorting team from five to four workers and still finish sorting 1.5 to 2 hours earlier than before.”
A Win-Win-Win Situation
Work coordinator Katharina agrees. “The team size has already been reduced at times, though not consistently,” she says. “At the very least, the team leader now has more time to oversee operations. Another advantage is that training new workers takes much less time, allowing them to reach full efficiency more quickly. This saves time for us as supervisors as well. Additionally, because people can be inconsistent, but software makes objective decisions based on historical data, the sorting uniformity has also improved. It’s a win-win-win. I’m really happy with it.”
Future Prospects
The users firmly conclude that IRISS Defects is performing well and is a valuable addition to 4MT’s vision system, which has been in use for some time. Meewisse hopes to eventually link sorting process data to the specific greenhouse location where each rose was harvested. “We already collect a lot of data in the greenhouse. If we can correlate stem defects with specific aisle numbers and identify patterns, new optimization opportunities emerge—such as crop protection, scouting, and providing feedback to workers who consistently deliver roses with leaf or stem damage.” “We’re not there yet, and other parties will need to be involved, but it’s only a matter of time,” concludes Van der Elst. “Technology and AI continue
