The sourced answer
How do optical sorters detect and remove defective potatoes, and what should a processor look for when choosing one?
Optical sorters image each potato with cameras and sensors, use software to classify defects, and then eject rejects with pneumatic air jets or mechanical fingers (Springer). The most advanced systems also use infrared, NIR and X-ray to assess internal quality.
How detection works
- Camera-based vision: These systems detect external defects such as green skin, cuts, bruises, scab and cracks at 5-50 tubers per second (Noordam et al., 2000, Machine Vision and Applications).
- 360-degree coverage: Key Technology's Oculus sorter combines infrared cameras, optimized for greening, with digital color cameras. It identifies bumps, notches, skin discoloration, mechanical damage, common scab, cracks and black dot. Newtec's Celox-P-UHD uses six cameras and a patented vibrating-roller system that scans each potato up to 45 times.
- Internal defects: Near-infrared (NIR) sorting can detect hollow heart and internal browning non-destructively (Walsh et al., 2020, Postharvest Biology and Technology). Hyperspectral imaging is an emerging approach to comprehensive quality assessment (Gowen et al., 2007). The most sophisticated multi-sensor systems combine hyperspectral, NIR, InGaAs infrared and X-ray to assess dry matter and sugar content.
- Foreign material and fine defects: TOMRA cites 98%+ foreign material detection accuracy on its 5A system and says some belt-based configurations can detect defects smaller than 1mm. These are vendor claims.
- Disease discrimination: Newtec's AI can tell Rhizoctonia fungal damage from wormhole damage. The two look alike but call for different agronomic responses.
Removal
Once a tuber is classified as defective, pneumatic air jets or mechanical fingers divert it (Springer). Newtec's machine can sort into as many as 16 product categories at once. Typical examples are Class A under 50mm, Class A over 50mm, Class B and rejects.
What a processor should look for when choosing one
- Throughput: Newtec's Celox-P-UHD handles up to 30 tonnes/hour, depending on tuber size. Seed-potato inspection lines run roughly 15-18 tons/hour because each tuber is scrutinized more closely. Large packing lines that combine optical grading with mechanical sizing reach up to 90 tons/hour (Tolsma-Grisnich and Wyma materials). Match capacity to your peak intake.
- Defects you actually face: Check that the sorter handles your main problems, such as greening, scab, rhizoctonia, wireworm, rot, or foreign material. Ask for demonstrations on your own crop.
- Sensor type: Cameras only see the surface. If hollow heart, internal browning, dry matter or sugars matter, as in French fry or chip processing, you need NIR, X-ray or hyperspectral capability.
- Accuracy for your task: Peer-reviewed accuracy is strong for well-defined, coarse tasks, such as 98%+ for sprout detection and about 90% for cultivar or processing-suitability classification. It drops for finer tasks like pixel-level internal classification and multi-category shelf-life prediction. Be cautious about vendor claims on those.
- Sorting flexibility: Look at the number of output categories and the lane configuration (4, 8 or 12 lanes on the Celox-P-UHD). Mechanical roller or screen graders separate by size, typically seed under 55mm, ware 55-85mm and oversize over 85mm (USDA). Check whether the optical sorter will work alongside them.
- Labor savings: A seed-potato farm processing about 5,000 tons a year reported that it no longer needed outside manual sorters after installing an AI sorter. Work that took several people at the belt was reduced to one supervisor (Tolsma-Grisnich case material).
Key takeaway: Choose a sorter based on your defect profile, whether you need internal quality detection, and your peak throughput. Validate vendor accuracy claims on your own potatoes, and remember that fine-grained and internal classification are still less reliable than surface defect detection.