Quality assessment and grading of onions are often subjective and vary across procurement centers, resulting in disputes and inconsistencies.
Ministry of Consumer Affairs, Food & Public Distribution · Fitness & Sports · Software
Almost everything here favours you — cheap props, self-built data, an invisible theme label and a judge-verifiable result — but the description defines nothing, so go and find the actual NAFED grading specification and build to that instead of to thresholds you chose.
What it actually is
When onions are bought at a government procurement centre, a person eyeballs the lot and decides what grade it is, and two centres will grade the same sack differently. Farmers dispute the call and there is no record to settle it. The ask is a phone app that photographs the onions and grades them the same way every time.
What to build
A mobile app where an officer photographs a spread onion lot and gets an instant grading report, built on a detector that segments individual bulbs in a cluttered heap and classifies each into the four defect classes the statement names — damaged, rotten, sprouted and undersized — with size estimated against a reference object placed in frame so undersized is measured in millimetres rather than guessed from pixels, aggregating per-bulb results into the Grade A and URS percentage split the statement asks for, and emitting a timestamped, geo-tagged digital quality report the farmer and the centre both receive.
Smallest thing that wins the room
Have a judge tip out their own bag of onions with a coin in the frame, photograph it, and get a per-bulb overlay with a Grade A and URS percentage they can check against what they can see.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 23% of the 226 · #175 of 226 by expected field
A normal-sized field. Your idea has to be good, not miraculous.
Why: central ministry statements sat below the average.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
4/5A genuine hidden gem — physical props you can carry in, a dataset you can create yourself, a result a judge verifies by looking, and the portal has filed onion grading under Fitness and Sports so nobody browsing Agriculture or FoodTech will ever find it.
Feasibility
4/5This is one of the few statements where you can build your own dataset in an afternoon — onions cost almost nothing, the four defect classes are externally visible, and you can photograph hundreds of labelled bulbs under varied lighting yourself rather than hunting for a corpus that does not exist.
Innovation scope
4/5The description is five bullets with no method, no architecture and no interface prescribed, so how you segment a heap, how you handle occlusion and how you turn per-bulb results into a defensible lot percentage are all yours to design.
Clarity
2/5Under 350 characters with no background or description section at all — it never defines what Grade A or URS actually mean in millimetres or defect tolerance, never states an accuracy target, and never says whether you are grading a sample or a whole lot, which is the difference between a vision problem and a sampling problem.
Effort
MediumA detection model, a size calibration step, an aggregation rule and a report-generating mobile app is a contained build with one model and one screen, which is genuinely modest by the standards of this portal.
Demo-ability
EasyYou can put real onions on the table and let a judge photograph them, and because the defects are visible to the naked eye the judge can immediately check whether your grading is right — verification in the room is worth more than any interface.
In its favour
- Green flag: The Fitness and Sports theme label is about as wrong as it gets, so this statement is effectively invisible to anyone browsing by theme
- Green flag: You can build your own labelled dataset from a sack of onions and a phone, which removes the data dependency that kills most vision statements
- Green flag: Physical props mean the judge participates in the demo rather than watching it, and a correct grade they can verify themselves is far more persuasive than a reported accuracy number
- Green flag: Because the description is nearly empty, you get to define the grading specification yourself — anchor it to the published NAFED or DoCA procurement size and tolerance norms and your scope becomes defensible rather than arbitrary
Against it
- Red flag: Internal rot is invisible from outside and is precisely the defect that causes procurement disputes, so a purely external vision system cannot detect the failure mode that matters most — say this before a judge says it to you
- Red flag: Estimating the percentage composition of a whole lot from a photograph of its surface is a sampling problem, not a vision problem, and a heap's top layer is not a random sample of the sack
- Red flag: Undersized is a physical measurement and pixel counts mean nothing without scale, so a reference object in frame is not optional — without it your size class is fabricated
- Red flag: Grade A and URS are never defined in the statement, so if you invent thresholds a procurement officer will simply say your grades are not the ones they use
What you will be writing
- YOLOv8-seg instance segmentation of clustered bulbs
- reference-object pixel-to-millimetre calibration
- EfficientNet defect classification head
- TensorFlow Lite on-device inference
- Flutter camera capture app
- PDF quality report generation
- Agricultural produce grading
- Computer vision
- Procurement transparency
Prior art to read before you start
onion quality grading · produce defect classification from images · on-device agricultural inspection app
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.