Skip to content
PenzaıHUMAN TECH
The Penzai mascot watches peppers on a camera grading line
Phase 7 of 8 · In design

AI grading and quality control: the AgriAI Quality agent

The AgriAI Quality agent measures size, colour and defects of every fruit with computer vision and sorts it into the right grade, logging quality by batch.

Quality agent

Every fruit in the right grade.

What it does

  • Measures size, colour and shape of every fruit with computer vision
  • Spots defects: bruises, sunscald, rot, deformities
  • Assigns grade and size class under marketing standards
  • Sorts fruit automatically on the grading line
  • Links defects to causes in the greenhouse and alerts the other agents
  • Logs quality statistics by batch and customer

How it works, step by step

  1. Fruit reaches the line with its source crate code
  2. Cameras photograph them from several sides and measure size and colour
  3. The model detects defects and assigns the grade
  4. The line sorts the fruit and the agent logs batch quality

Agronomic deep dive

The science behind the agent

Quality isn't created in the packhouse: it is measured. Size, colour and defects tell what happened in the greenhouse in the weeks before, and measuring them precisely helps both to sell better and to grow better.

The rules: marketing standards

In Europe sweet pepper has a specific marketing standard, in Implementing Regulation 543/2011: Extra, I and II classes, minimum requirements, tolerances and sizing. Customers' own specifications come on top.

The agent turns standards and specifications into measurable rules: diameter, weight, colour uniformity, defects allowed per class. That keeps grading the same every day, whoever is on shift.

Machine vision on the line

Machine vision for fruit and vegetables is a mature field: the review by Cubero and colleagues describes how colour and multispectral cameras with controlled lighting measure size, colour and external defects at high speed.

The hard part is small or ambiguous defects, like a light bruise or the start of rot. For those the agent uses several images of the same fruit and, if unsure, sends it for human inspection instead of guessing.

From defect to cause

Many defects have a cause in the greenhouse. Sunscald depends on radiation and shading; blossom-end rot is linked to calcium transport into the fruit, itself influenced by irrigation and humidity; cracking depends on swings in water supply.

Thanks to crate traceability, the Quality agent knows which zone and harvest day a defect comes from, and alerts the Climate and Fertigation agents. This is where the chain stops being a series of hand-offs and becomes a system that learns.

The parameters that matter

ParameterReferenceWhy it matters
Commercial classesExtra, I, II (EU Reg. 543/2011)They set price and destination
SizeDiameter or weight per standard and specUniform, compliant packs
ColourUniform, per commercial typeThe buyer's first criterion
Allowed defectsDifferent tolerances per classA defect out of tolerance downgrades the batch

Reference values from the agronomic literature for soilless greenhouse vegetables: always calibrate them to crop, variety, stage and local climate.

The sensors it uses

What each one measures, where it goes and the detail that makes the difference. “Core” is needed to start, “Extra” adds precision. See the full list for all agents

  • Grading line camerasCoreSize, colour, shape and defects of every fruit, from several sidesIn a booth with controlled lighting over the beltBooth lighting must be kept constant and calibrated with reference colour samples.
  • Line scales and checkweighersCoreWeight of every fruit and every packOn the grading belt and at the packing exitA daily check with test weights avoids weight disputes.
  • Barcode and QR readersCoreThe code of crates, batches and packs at every stepAt line entry, packing and dispatchEach scan is a link in traceability: miss one and the chain breaks.

The most common mistakes

  • Uncontrolled line lighting: measured colour changes with the light
  • Grading without a written spec: each shift decides its own way
  • Rejecting without logging why: the cause information is lost
  • Not calibrating cameras with reference samples
A measured defect is information for the greenhouse; a defect that is just rejected is simply a loss.

Frequently asked questions

Don't grading machines already exist?

Yes, machine vision for fruit and vegetable quality has been studied for decades. What the agent adds is linking each defect to its cause in the greenhouse, so the problem is fixed at the source.

Which rules does it follow for grades?

EU marketing standards for fruit and vegetables and your customers' specifications, which you can load as rules.

Become a pilot farm

Free digital check-up: we look at systems, consumption and problems and tell you whether and where AgriAI can help.

Apply as a pilot farm / I have land to bring back / I'm an agronomist: join the scientific team