How do you integrate an egg grader with your systems?

What does each field in the record mean?

Each record has an identification number for the egg, the class, the scores for the three defect classes (clean, dirty and cracked), a list of reasons in plain text and a block of measurements. The class is the final decision. The reasons explain the decision, and the measurements are the numbers the reasons use.

The measurement fields are: the dirt fraction, the crack length in multiples of the reference radius, solidity, ellipse fill, the axis ratio, and whether the egg touches the image border. An egg that touches the border is not graded: it receives the unverified class.

An example from the demonstration image: a cracked egg recorded the reason crack length 1.22 r against the limit of 0.80 r. Anyone reading the record sees the reason and the number that produced it, without consulting another source.

How do you use the records to count classes and filter batches?

Counting by class is the simplest use of the data: add up how many eggs of each class passed through a batch, a shift or a shed. The link to the batch, the shift or the time has to be made in your own system, because the description of the per-egg record does not cover those fields.

It also helps to aggregate by reason. An increase in dirty eggs may come from more dirt on the shells or from a change in a threshold, and only the recorded reason and measurement show the difference. A dashboard that shows only the rate of each class does not say why it changed.

When is it safe to trigger an automatic action from this data?

In most cases, start with the dashboard and not with the automatic action. A diverter or an alarm that reacts to every record turns a threshold error into an immediate loss of product.

A per-egg record sees only the surface, so actions based on it can only cover what the surface shows. Any decision that depends on the air cell, the yolk or hairline cracks still depends on candling.

1. Validate the integration with labelled eggs from your line before turning on any action.
2. Run the dashboard in observation mode for a period and compare it with the team's assessment.
3. Define who receives the alerts and what to do when the rate of a defect changes suddenly.
4. Record every automatic action with the rule it used, so that an audit can rebuild it.

Calibration with labelled eggs is the step that validates the thresholds before integration: see calibrating with labelled eggs.

Why should the same image always produce the same record?

Vigiovo is deterministic: the same image and the same configuration give the same result. This helps integration, because if a number changes without explanation, the first suspect should be the configuration or the image, and not the classifier itself. On this topic, see reproducible inspection.

For that reason, keep the configuration in use and the date of each change alongside the records. Without that, a difference between two weeks cannot be explained.

Frequently asked questions

Can the Vigiovo record feed a quality dashboard?

Yes. Each egg comes out as a structured record with the class, scores, reasons and measurements. What the dashboard shows and how the data is aggregated is up to your team, and the numbers should be read with the limit that the camera sees only the surface of the egg.

Can I use the scores as a confidence level?

No. The scores come from a fixed formula applied to the measurements, not from probabilities estimated from data. A real confidence measure would need a separate validation. Use the measurements and the reason to understand each decision.

Does integration replace candling?

No. The camera sees the surface of the egg, and candling sees the inside, including the air cell, the yolk and fine cracks. Camera data should not be used to claim the internal quality of an egg. Candling remains the step that checks those criteria.

Sources

1. Como o Vigiovo classifica ovos sem IA / How Vigiovo grades eggs without AI, Vigiovo