Industries / Food, farming and forestry / Agriculture and farming / Seed production
Cleaning and sorting
Cleaning and sorting, which the trade calls conditioning, takes seed as it comes from the field and removes chaff, stones, weed seed and broken, light or off-color seed, so that what is bagged is sound seed of one crop.
5
Milestones achieved, from 1919 to 2022.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1919
Seed sorted by weight
A specific-gravity separator, or gravity table, blows air up through a tilted, rocking deck so that seed of one size forms layers by density: light seed floats and runs to the low side, and heavy seed is walked to the high side. Edwin Steele and Henry Sutton of Sutton, Steele & Steele in Dallas, Texas, first built it to concentrate metal ores without water, and by 1919 it was used on field seeds, peas, beans and corn, by the account of the firm's successor, Triple/S Dynamics. A 1968 USDA handbook describes how the seed industry adapted the mining machine to separate crop seed from contaminants.
Credit
Edwin Steele and Henry Sutton (Sutton, Steele & Steele)
1998
Sorting by chlorophyll glow
Seed that did not finish ripening keeps chlorophyll in its coat, and chlorophyll fluoresces (gives off light of its own when lit at certain colors). Henk Jalink and colleagues in Wageningen sorted cabbage seeds one by one by this signal and raised a lot from 90% to 97% normal seedlings by removing the 13% of seeds with the strongest signal. In 2020 a University of São Paulo team confirmed on commercial tomato and carrot seed lots that a stronger signal meant lower vigor.
Credit
H. Jalink, A. Frandas, R. van der Schoor and J.B. Bino (Centre for Plant Breeding and Reproduction Research, Wageningen)
2010
A sorter for small lots
Thomas Pearson, an engineer with the USDA's Agricultural Research Service (ARS) in Manhattan, Kansas, built a compact sorter that photographs each falling seed in color and drops it into one of two bins; National Manufacturing of Lincoln, Nebraska, has sold it since 2010. In ARS tests it told hard red from hard white wheat kernels with 98.6% accuracy. By ARS's 2013 account, plant breeders use it to separate seed, a major pea and bean breeder uses it to remove damaged seed, and native grass seed for revegetating western public lands is sorted with it to remove unwanted species.
Credit
Thomas C. Pearson (USDA ARS); National Manufacturing
2016
Weeds in weed-free mixes
Seed mixes of native plants sold for US pollinator habitat carried seed of Palmer amaranth, a prolific weed with populations resistant to several groups of herbicides, onto conservation land in Iowa, Minnesota and other Midwest states. University of Illinois researchers traced it to at least one company's mix whose tag claimed it was 100% weed-free, identifying the weeds by germinating the seed and growing the plants.
LessonA tag that says weed-free is only as good as the checks behind it; the agencies running the program reviewed the tags but did not inspect the mixes.
Credit
Aaron Hager and colleagues, University of Illinois
2022
Sorting by predicted germination
Seed-X, an Israeli company, makes the GeNee Sorter, which photographs each seed and uses computer vision and machine learning to predict qualities such as germination and purity, then rejects seed predicted to fail. Advanta Seeds, part of the UPL group, put it into its processing line in May 2022 after a trial of several months, and said it let the company recover seed lots that had failed germination standards.
Credit
Seed-X; Advanta Seeds
Watch for
Sound waves push seeds aside
Most optical sorters knock rejected seed out of the stream with jets of air. A University College London team paired a camera with ultrasonic phased arrays (grids of small emitters that focus beams of sound) to push chosen seeds off a moving conveyor belt by sound pressure, sorting by size and color and removing debris. The team says the push can be set for each seed, so seed can be split into more than two grades.
Achieved whena sorter maker or seed company runs it at commercial speed and publishes its accuracy against an air-jet sorter on the same seed lots.
Claimed by
Hardwick, Morgan and Hirayama, University College London, in Nature Communications, July 2025
Open
01
A seed-sized impact logger
A sensor capsule no larger than a big bean seed, about 15 mm, that records every knock it takes as it rides with real seed through a conditioning line's elevators, cleaners, gravity table and bagger, so the plant can see where seed is being cracked.
The briefCredit · unclaimed02
Open sorter judged by germination
A low-cost single-seed sorter, with a camera, an open image model and an air jet, whose design, training images and code are published, and which is judged the way seed is sold: by how well the seed it keeps germinates.
The briefCredit · unclaimed
Open 01 · the brief
A seed-sized impact logger
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A sensor capsule no larger than a big bean seed, about 15 mm, that records every knock it takes as it rides with real seed through a conditioning line's elevators, cleaners, gravity table and bagger, so the plant can see where seed is being cracked.
Why it is still open
Impact loggers already ride with produce, but at fruit and tuber size. A University of Georgia thesis (2010) notes that the commercial Impact Recording Device is at least 57 mm across, and built a 25.4 mm sphere for blueberries because no smaller sensor existed. Searches in November 2026 found loggers for potatoes, apples and blueberries, and seed sensors that count seed in planters, but none that rides with seed through handling.
Built with what exists
a miniature three-axis MEMS accelerometer, a low-power microcontroller and a coin cell on a small flexible circuit board, cast in resin to the shape and weight of a large bean or peanut seed; a drop-test rig for calibration; the standard sodium hypochlorite soak test for cracked seed coats as the damage reference.
First step
Build a 25 mm prototype from off-the-shelf parts and calibrate it by dropping it onto steel from known heights.
Done looks like
20 runs of the logger through one working seed line, run by the plant's staff, each giving an impact map by machine, and seed sampled after each machine tested for cracks, showing whether the machines with the hardest knocks crack the most seed.
Open 02 · the brief
Open sorter judged by germination
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost single-seed sorter, with a camera, an open image model and an air jet, whose design, training images and code are published, and which is judged the way seed is sold: by how well the seed it keeps germinates.
Why it is still open
The sorters that predict germination seed by seed, such as Seed-X's GeNee Sorter, are closed products. The low-cost designs found sort food grains by appearance: a 2025 lentil sorter by Davy Rojas Yana and Edwin Salcedo separated grains with 87.2% accuracy at 8 g a minute, and an open-hardware coffee sorter project aimed at smallholders gives only design estimates (under US$300, 250 to 300 beans a second on a Raspberry Pi 3); neither reports germination. Open image sets exist, such as 3,000 pea seeds labeled live or dead (Men and colleagues, 2026), but searches in November 2026 found no sorter built and tested on them.
Built with what exists
a Raspberry Pi 5 with its camera module; a small solenoid air valve; an image classifier trained with PyTorch, starting from the AgML pea seed set; a standard germination test run by a seed laboratory as the referee.
First step
Photograph 1,000 seeds of one lot one at a time, germinate each in its own cell of a tray, and see whether a simple model can tell from the photo which seeds will grow.
Done looks like
For three seed lots of one vegetable crop that start below 85% germination, the seed the sorter keeps germinates at least 5 percentage points higher than the unsorted lot in a standard laboratory test, with the share of good seed thrown out reported, and the design and data published.