QualityNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.
Deep learning optical sorters spot hard defects in fruit and nuts on the line, so food plants pack cleaner product with fewer hand sorters (Chelan Fruit, Kings River Packing, The Almond Company)
Added 2026-11-18 · Last updated 2026-11-18
Manufacturers in this case: Chelan Fruit · Prima Frutta · Dunstan Hills · Panmure Orchards · Kings River Packing · The Almond Company · all manufacturers using AI
Role
Quality and production in food processing and packing plants (sorting, grading, foreign material removal) · Typical users: food plant quality manager, packhouse or processing plant manager, production supervisor, food safety coordinator, operations director
The problem
Food processors and packers must remove defective product and foreign material at high speed, often during a short season. Hand sorting is hard to staff and inconsistent, and the older camera sorters judge a defect from a few pixels, so subtle defects are missed or good product is thrown away. In cherries, a packer can run 40,000 cherries a minute, as the president of Prima Frutta Packing (Linden, California) explains, leaving no time to sample, and stems, doubles and tangled fruit hide defects from traditional systems; the cherry season runs only from mid-May to the end of June in California. At Chelan Fruit (Pacific Northwest, USA), some cherry lots contained up to 30% culls. At Kings River Packing, a family-owned citrus group in California with three packhouses, defects such as rot, sunburn, clipper cut and long stem are hard for traditional mapping to classify because it does not see the context of the whole fruit. The Almond Company (Madera, California), the almond handling and processing operation of Harris Family Enterprises, processes more than 100 million pounds of almonds a year and faces labor, food safety and quality challenges.
How AI helps
Cameras, lasers and other sensors capture every piece of product on the belt or in the lanes, and deep learning models trained on large sets of real images classify each defect by looking at the whole piece, then eject it or send the piece to the right grade. Operators stay in control of the grade settings, and the software shows the effect of a change before it is applied. At Chelan Fruit, Prima Frutta, Dunstan Hills and Panmure Orchards (New Zealand), TOMRA's LUCAi deep learning runs on the InVision 2 cherry grading platform with models pre-trained on cherry images from several regions, and TOMRA's service team trains new defect models from the customer's data and feedback (for example powdery mildew, insect bites and rain cracks at Prima Frutta). At Kings River Packing, LUCAi was added to the existing TOMRA Spectrim vision systems after a full-season trial on 20 lanes at its Cobblestone Fruit packhouse. At The Almond Company, a TOMRA 5C optical sorter, which combines lasers and cameras with machine learning analysis, sorts confectionery grade almonds and removes foreign material, and its data goes to the TOMRA Insight cloud platform.
Step by step
- Measure today's baseline per product: throughput, hand sorters per shift, defects found in the final quality check, customer claims and good product found in the rejects.
- List the defects your current sorter misses or over-rejects most, with photos of each type, and rank them by claims and lost product.
- Ask the sorter makers for a trial on your own product and worst lots, including the AI upgrade of the sorter you already have.
- Run the trial for a full season or several weeks with the same quality check before and after, and count both missed defects and good product rejected.
- Agree with quality and food safety who changes the grade settings and how foreign material removal is verified, alongside metal detectors and X-ray.
- Track throughput, people on hand sorting, quality check results and claims per lot, and give the vendor feedback on new defect types to train.
Tools
- TOMRA LUCAi deep learning on the InVision 2 cherry grading platform: used at Chelan Fruit, Prima Frutta, Dunstan Hills and Panmure Orchards
- TOMRA LUCAi on TOMRA Spectrim: used at Kings River Packing (Cobblestone Fruit)
- TOMRA 5C optical sorter with TOMRA Insight: used at The Almond Company
- Other tools to evaluate: AI sorting and grading from other optical sorter makers, AI upgrades for the sorters you already own, inline vision inspection systems for packed product
- To start without new software: your grading, reject and labor records and any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below
In the AI tools index: TOMRA LUCAi · TOMRA 5C optical sorter · all AI tools used by manufacturers
Result
Chelan Fruit (TOMRA story in Produce Processing, 2024): even with lots containing up to 30% culls, 40 people were redeployed away from manual sorting while achieving a clean packout. Dunstan Hills: better grading and around 10% more throughput. Prima Frutta: after the 2023 trial season, LUCAi was installed on all its machines, 132 lanes, and and its president says it runs more tons per hour with the same labor force, with good-in-bad errors dramatically reduced (TOMRA release on Postharvest). Kings River Packing (FreshPlaza, 2025): after the 20-lane trial, rollout to its other packhouses across almost 100 lanes; TOMRA states a greater than 99% accuracy rate for difficult defects such as rot, sunburn, clipper cut and long stem. The Almond Company (TOMRA Food release, 2021): production of that product up 20 to 25% after six months, in less time and with less labor; the company expects the sorter to pay for itself within three to five years.
As reported by the source; most figures are vendor or self-reported and not independently verified.
Human review: cautions
- All results come from TOMRA (the vendor) or from articles based on its customer stories; the >99% accuracy is TOMRA's claim for specific citrus defects, not a measured plant result.
- Most examples are fruit packhouses; a processing plant with other products (vegetables, nuts, grains) must test on its own product, lots and defects.
- An optical sorter does not replace food safety controls: keep metal detection, X-ray where needed and your HACCP verification.
- Watch both errors: defects left in the good product and good product sent to rejects; a vendor figure on one side only is not enough.
- Redeployed sorters need other jobs and training; plan the change with the team, especially for seasonal staff.
Try it yourself
You are a quality and operations specialist in a food processing plant. I want to decide whether an AI optical sorter (or an AI upgrade of our current sorter) is worth a trial. I will paste: (1) our sorting and quality records for the last season [date, product, lot, tonnes processed, throughput per hour, hand sorters per shift, rejects in kg, final quality check results by defect type, customer claims]; (2) a list of the defects our sorter misses or over-rejects, with how often we see them. Do this: 1. Calculate per product the hand-sorting hours per tonne, the reject rate and the defect rate at final quality check, and show the worst lots. 2. Rank the defects by impact (claims, lost good product, labor) and say which ones a camera-based AI sorter could reasonably detect and which need another method (metal detector, X-ray, lab test). 3. Write a trial plan with a sorter vendor: products and lots to test, how to measure missed defects and good product rejected, duration and the decision criteria. 4. List the questions to ask vendors (training on our defects, who changes settings, data ownership, cleaning and sanitation, service). 5. Draft a one-page summary for the plant manager with the savings at stake in labor and product. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.
Want more like this? Join the newsletter
Related cases
- AI reads the date code and checks the label on every pack at line speed, so the wrong label or code never ships (Ilchester Cheese, Del Monte Foods)
- Vision defect inspection on the edge (Seagate, Procter & Gamble, Coppenrath & Wiese)
- AI vision grades every board and picks the cut at the edger and trimmer, so more value comes out of each log (Waipapa Pine, Tenon, Hunt Forest Products)
- AI double-checks what the AOI machine flags on circuit boards and clears the false calls, so operators only review real defects (Siemens Rastatt, USI, FermionX)
Sources
- https://produceprocessing.net/article/tomra-technology-help-cherry-packer-refine-operations/
- https://www.freshplaza.com/north-america/article/9740607/u-s-citrus-packer-rolls-out-tomra-s-lucai-deep-learning-tech/
- https://www.potatopro.com/news/2021/tomra-5c-sorter-adds-value-almond-company
- https://www.postharvest.biz/tomra-food/cherry-packhouses-experience-the-efficiency-and-profitability-benefits-of-tomra-lucai-deep-learning-technology
Weekly: 1–2 practical AI use cases for your plant
Free, one email a week. By subscribing you agree to receive the weekly newsletter (and its sponsor message) from AI Agents for Manufacturing. Every email includes an unsubscribe link, and you can unsubscribe anytime. See our Privacy Policy.