precision purification of potatoes with computer vision and air pressure

Traditional solutions such as clod separators use density differences or mechanical separation methods. They are a good first step in the treatment process. Although robust, these systems offer limited control over the separation process and are prone to wear and tear or incorrect sorting.

That is why Dema-Construct wants to go one step further. This has now been achieved with an automated, vision-controlled sorting machine with higher accuracy, better control and minimal mechanical friction.

Optical sorters in themselves are not new. But where others work with mechanical valves, Dema-Construct deliberately chose to use air pressure. That means less wear and tear and thus more long-term reliability.

Moreover, they also focused on sorting unwashed potatoes, a much more complex challenge than the washed varieties.

This combination called for a completely different approach: no classical algorithms, but an AI-driven solution that makes decisions in real-time, with extreme precision in timing.

Potatoes come in all shapes, colours and textures. The variation in shape, colour and texture is so great that some stones are more visually close to potatoes than potatoes are to each other.

Bringing vision, edge computing and AI together in a robust industrial design made this complex challenge workable and scalable. A fine example of collaboration between domain knowledge and technology.

A real-time sorting system for agricultural products requires an interplay of vision, tracking and control under strict timing. The main challenges were:

• Visual similarity: potatoes vary widely in shape, colour and texture due to factors such as variety, soil type and weather. This natural variation makes distinguishing from impurities such as stones and clods particularly challenging.
• Defect recognition: in addition to waste, potatoes with all kinds of defects have to be removed.
• Object tracking under variable conditions: density, speed and heterogeneous object shapes influence detection and trajectory determination.
• Precision control: only a small time window between conveyor and collection bin is available to eject impurities. The ejection mechanism must be accurate to a millisecond.
• Robustness: classical systems are subject to wear (finger mechanisms), especially on impact with hard objects.

Together with Dema-Construct, Vintecc developed a fully integrated system consisting of vision, tracking and air pressure sorting, tailored to extreme real-time requirements:

Vision & Detection

• Two cameras permanently monitor the conveyor belt and feed the real-time AI model with images.
• Object recognition is done via trained AI image processing models that recognise a huge variety of 'waste': from clods and stones to lost machine parts or even historical objects such as obbuses.
• The system also distinguishes seven types of defects in potatoes. This allows highly targeted choices to be made in the sorting process, tailored to specific quality levels and tolerances.
• Thorough data curation, dataset quality analysis and advanced augmentation allowed training an AI model that remains robust under these conditions.

Trajectory forecast

• Each object is tracked individually, predicting position, speed and acceleration until the launch phase.
• Both impurities and potatoes are tracked. Not only to detect, but also to mark safe zones around the potatoes so that they are not accidentally blown away.

Precision control

• The moment objects hover between the belt and the collection bin, a targeted shot of air is given via a row of air valves, if necessary.
• Air is emitted in short pulses, allowing precise removal of small and large objects.
• Timing is crucial: a shot too early or too late results in tipping or false ejection. The software determines the optimal shot timing with sub-frame precision.

Operational reliability

• The system contains only air conduction and electromagnetically controlled valves. No wear-sensitive mechanics such as moving fingers.
• Because there is no physical contact with objects, there is also no possibility of damage to both potatoes and the machine itself.

• Precise classification and removal of impurities.
• Tracking of objects and velocities allowing precise shooting.
• High uptime and low maintenance requirements, thanks to the absence of moving mechanical components.
• Scalable design suitable for different production lines or agricultural streams.
• Real-time insight into sorting processes through integrated visualisation and logging.

By complementing mechanical sorting with a combination of computer vision, real-time tracking and air control, Dema-Construct, with support from Vintecc, is realising a new standard in agricultural automation. The system's accuracy, speed and reliability make it ideal for applications where product quality and uptime are crucial.

This solution shows how industrial processes can be controlled with millisecond precision. Fully software-driven, with no mechanical compromises.