AI pre-sorters reach Dutch potato stores: €100,000 entry price

Dutch potato and onion growers now pay at least €100,000 for camera-based pre-sorters that remove stones, clods and defective produce during store loading, with all-in costs running to €80,000-100,000 per metre of working width. The Netherlands has become the key testing ground for the technology, and manufacturers have moved beyond basic separation into automated quality grading.

The first pre-sorters capable of handling freshly harvested crops during store loading are now commercial reality. They build on the same principles as camera-based inspection with image recognition and size grading. Removing clods and stones — essentially anything that is not a potato or an onion — already works well and saves manual labour. Cameras can also reject produce on external characteristics.

A little soil is not a problem in storage and can even buffer moisture fluctuations. Too much means clods must be picked out. Mechanical clod separators exist for smaller size grades, and more sophisticated machines, particularly for washed produce, can recognise traits such as green colouration. Growers are interested, and some are already investing.

What can the machines remove?

The latest systems sort for quality at store loading: misshapen or green potatoes, growth cracks, wireworm damage, cuts, mother tubers and rot, plus onions with missing skins or Fusarium infection. Storing rotten potatoes or mother tubers can compromise the whole crop, and unwanted material wastes valuable store space.

The technology can create a third output stream of rejected potatoes, saleable for example as livestock feed. A fourth stream, splitting the clean crop into two size grades, is technically possible but not yet operational. Manufacturers are working on it.

"We develop in response to market demand," said Remco Theunisse, head of marketing at Flikweert Vision. In a few years the company's QualityGrader has established camera-based quality sorting as an alternative to manual inspection on grading lines. The market is now asking for the process to move upstream to store loading. Freshly harvested crops present more variation than dry produce: wet clay and heavily coated potatoes may not even look like potatoes, and optical recognition relying solely on external appearance has its limits.

What about throughput and accuracy?

All three machines seen operating had experienced a blockage at some point. In practice, there is a trade-off between throughput and accuracy. Throughput is expressed in tonnes per hour, but with a camera-based sorter what matters is the number of objects the system can assess and remove per second. On that basis, suppliers currently appear fairly evenly matched.

Throughput depends less on the technology than on the width of the sorting unit, and begins with spreading the crop evenly across the belt. With the exception of Downs, pre-sorters move produce beneath the cameras without rotating it, so cameras cannot inspect every side, making pre-sorting somewhat less precise. Downs rotates the produce, enabling more accurate sorting but requiring greater computing power and more expensive technology; spinning the rollers clean also reduces throughput.

"It saves us 4 people on the belt," one user told Future Farming. Where manual labour is the alternative, payback can be calculated quickly, provided labour is available. Efficient pre-sorting also improves subsequent grading before delivery. Annual charges for licences and updates may apply on top of the purchase price, though some operating machines remain pre-production models and suppliers cannot always give firm details.

The machines

Downs Crop Vision gained a new screen and software this year. Rollers rotate potatoes — at 90-110 rpm, rising to 300-400 rpm for self-cleaning — so cameras photograph tubers from all sides, sorting for misshapen growth, greening, scab, wireworm and slug damage. The standard machine separates three streams, with the two size-grade option planned for next year. Widths reach 280 cm, with throughputs of 70 tonnes per hour achievable on large potatoes.

Flikweert Divider Select runs potatoes on a flat belt, taking one image per object to decide potato or reject, and recognising green potatoes, mother tubers, cuts and rot. It is available in widths up to 3 metres at roughly €80,000 per metre. The Divider Pro takes three images per object with more computing power, sorting more accurately for a wider range of characteristics. A separate RejectSeparator recovers potatoes from the reject stream.

Tolsma-Grisnich Optica S — the first five units are operating this year, including one in the Netherlands. Based on the Optica Q, which identifies scab, Rhizoctonia, misshapen tubers and wireworm damage across three streams, the simpler S has one ejector finger bar and two streams. It recognises Fusarium in onions and runs on a rubber-coated web conveyor, chosen because it resists stone damage. It handles 60-80 tonnes per hour, comes in 2.40 metres only, and starts at €200,000.

TOMRA 3A removes clods, stones, anything that is not a potato, and can separate green potatoes. Several machines operate in the Netherlands, where a prototype successor is running this season. TOMRA says the successor will offer more capabilities but has released no details. Its launch is scheduled for Interpom in Kortrijk, Belgium, from 29 November to 1 December 2026.