BRENNEN

Industries / Food, farming and forestry / Agriculture and farming / Produce packing

Sorting and grading

Sorting and grading divide each harvest by size, colour, defects and, more recently, hidden inside quality, into the classes that buyers pay for.

5

Milestones achieved, from 1962 to 2022.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1962

    Grades open European borders

    EEC Regulation No 23 of 4 April 1962 set common standards of quality, sizing and packaging for fruit and vegetables traded between the six member states, from 1 July 1962, for products including cauliflowers, lettuces, onions, tomatoes, apples, pears, apricots, peaches and plums. Produce covered by a standard could cross a member state's border only if it met it, and graded produce was freed from import quotas in steps: "Extra" class by 30 June 1962, Class I by 31 December 1963 and Class II by 31 December 1965.

    eur-lex.europa.eu

    Credit

    Council of the European Economic Community

  2. 1985

    A camera grades fruit

    By the account of Sumio Kawano of Japan's National Food Research Institute, grading took more than half the labour in Japan's packing houses. A grading machine designed in 1985 viewed almost the whole surface of each fruit with one colour camera and paired mirrors, rated its colour on 64 steps from green to red, found surface defects as small as about 0.3 mm across, and sorted 5 fruit a second on each lane. Kawano reported in 1994 that such machines were installed in many packing houses for satsuma oranges, tomatoes and persimmons.

    jircas.go.jp

    Credit

    H. Maeda, whose design was described in 1987; reported by Sumio Kawano

  3. 1988

    A cucumber must be straight

    Commission Regulation (EEC) No 1677/88 required "Extra" and Class I cucumbers to be "practically straight", with an arc no higher than 10 mm for each 10 cm of length, and set minimum weights of 180 g for cucumbers grown outdoors and 250 g for those grown under cover. On 12 November 2008 member states voted to repeal the specific standards for 26 products, cucumbers among them, from 1 July 2009. Ten products, including apples, citrus fruit and tomatoes, kept theirs, but member states could now let shops sell them out of class if labelled.

    LessonShape rules written for trade kept edible food off shelves, and the Commission dropped most of them because, in Fischer Boel's words, "it makes no sense to throw perfectly good products away, just because they are the 'wrong' shape.".

    eur-lex.europa.euec.europa.eu

    Credit

    European Commission; the repeal was proposed under agriculture commissioner Mariann Fischer Boel

  4. 1989

    Peaches sorted by sweetness

    Near-infrared (NIR) light lies just beyond the red end of what the eye can see; it passes a little way into fruit, and sugars absorb some of its wavelengths, so a sensor can estimate sugar content without cutting the fruit. In 1989 Mitsui Mining and Smelting introduced what Kawano calls the first operational machine in the world to sort peaches by sweetness without damaging them, and peaches it sorted were sold as "sweetness-guaranteed". Its later sensor worked out a peach's sugar content in 0.13 seconds, sorted 3 fruit a second per lane, and was extended to apples and Japanese pears.

    jircas.go.jp

    Credit

    Mitsui Mining and Smelting Co.; reported by Sumio Kawano, National Food Research Institute

  5. 2022

    Deep learning grades apples

    Deep learning is a kind of AI that learns to recognise a pattern, such as a defect, from many labelled examples instead of hand-written rules. The grader maker Ellips put its deep learning model, TrueAI, into use in 2022 on dates, pomegranates, apples and onions. In mid-December 2022 Apple King, a packer in Yakima, Washington, switched its Ellips apple grader to TrueAI, and its operations manager said the grader now caught hard defects such as stem-bowl cracks "at a very high level of accuracy with minimal false positives".

    thepacker.com

    Credit

    Ellips; Apple King

Watch for

  1. MRI scans inside the fruit

    Orbem, a German company whose MRI scanners already check eggs for the poultry industry, says an industrial version of hospital magnetic resonance imaging (MRI), read by AI, can find internal browning, hollow heart and rot in avocados and watermelons on a packing line without cutting the fruit. Its chief executive, Pedro Gomez, says the first commercial watermelon scanner went live in Spain with the producer JimboFresh, that packers will pay a fee per pound scanned, and that the first US installations should follow in early 2027.

    Achieved whena packer or other outside party publishes the scanner's detection rate against cut fruit, or a packer reports a full season of routine use.

    thepacker.com

    Claimed by

    Orbem, July 2026

  2. Deep learning for blueberry defects

    TOMRA Food says its LUCAi models, trained on images of real fruit defects collected over several seasons and regions, now grade apples, blueberries, cherries, citrus, kiwifruit and stone fruit, and that its new 5S Blueberry grader detects dehydrated berries. It gives its results in words, with no accuracy figure.

    Achieved whenan outside party publishes detection rates against hand-graded samples of the same fruit.

    freshplaza.com

    Claimed by

    TOMRA Food, September 2026

Open

  1. 01

    Defect areas, measured not guessed

    An open test set of about 2,000 apples of one or two varieties in which every defect has a measured size. Photograph each apple from six sides in a light box beside a printed scale, and have two people trace each bruise, scab spot and skin fault on the photos so its area is known in square centimetres, which fixes the class the fruit should get under the EU standard. Then have three graders trained on the standard grade the same fruit by eye as Extra, Class I, Class II or out of grade, and score open vision models on the photos.

    The brief
    Credit · unclaimed
  2. 02

    Class II, tasted blind

    A same-lot test of whether an apple's cosmetic class says anything about how it eats or keeps. Take Class I and Class II apples graded out of the same bins. Measure sugar in degrees Brix (the percentage of dissolved solids, read with a refractometer), firmness with a penetrometer, weight loss and the first signs of rot over four weeks in a refrigerator and at room temperature. Run a blind tasting of peeled slices.

    The brief
    Credit · unclaimed

Open 01 · the brief

Defect areas, measured not guessed

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open test set of about 2,000 apples of one or two varieties in which every defect has a measured size. Photograph each apple from six sides in a light box beside a printed scale, and have two people trace each bruise, scab spot and skin fault on the photos so its area is known in square centimetres, which fixes the class the fruit should get under the EU standard. Then have three graders trained on the standard grade the same fruit by eye as Extra, Class I, Class II or out of grade, and score open vision models on the photos.

Why it is still open

The EU apple standard sets its limits as areas: Class I allows bruising up to 1 cm² and scab up to 0.25 cm², Class II bruising up to 1.5 cm² and scab up to 1 cm² (). Graders judge these areas by eye, and makers of AI graders publish no accuracy against the classes: TOMRA describes LUCAi's results in words only (). The nearest open set found, RaspGrade from Italy's Fondazione Bruno Kessler (2025), gives one expert's ripeness grade per raspberry, under a non-commercial licence (). A Hugging Face search for apple grading datasets on 20 November 2026 found one empty repository.

eur-lex.europa.eufreshplaza.comtev.fbk.eu

Built with what exists

A turntable and light box; a phone or USB camera; a colour and scale card; OpenCV to measure defect areas; Label Studio for the graders' entries; open vision models such as DINOv2 or OpenCLIP, fine-tuned in PyTorch.

  1. First step

    Buy one carton each of Class I and Class II apples of the same variety at a wholesale market and photograph 100 of them to settle the light box and scale.

  2. Done looks like

    2,000 apples with traced defect areas and the class those areas call for, released under an open licence; for each of the three graders and at least two models, the share of fruit put in that class; agreement between the graders reported as Fleiss' kappa (a standard measure of how often raters agree beyond chance); and the graders' area estimates beside the traced areas for at least 300 defects.

Open 02 · the brief

Class II, tasted blind

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A same-lot test of whether an apple's cosmetic class says anything about how it eats or keeps. Take Class I and Class II apples graded out of the same bins. Measure sugar in degrees Brix (the percentage of dissolved solids, read with a refractometer), firmness with a penetrometer, weight loss and the first signs of rot over four weeks in a refrigerator and at room temperature. Run a blind tasting of peeled slices.

Why it is still open

When the EU dropped shape rules for 26 products in 2009 it kept classes for apples and nine other products (), and the apple classes are set by visible faults such as russeting (rough brown skin), bruises and skin defects (). A web search on 20 November 2026 found studies of shoppers' dislike of blemished fruit but no published comparison of eating and keeping quality between classes taken from the same lot.

ec.europa.eueur-lex.europa.eu

Built with what exists

A digital refractometer; a fruit penetrometer; a kitchen scale; a refrigerator at about 4 °C and a room-temperature shelf, each with a cheap temperature logger; randomized codes for the tasting; R or Python for the statistics.

  1. First step

    Ask one packhouse for 50 Class I and 50 Class II apples graded from the same bin and start the four-week keeping test.

  2. Done looks like

    At least 200 apples per class from at least three lots, every measurement published per fruit, and at least 60 blind tasters, with the differences between classes and their confidence intervals for Brix, firmness, days to first rot and preference.