A practical method for scoping optical sorting when size, colour, surface defects, moisture and seasonality all change the image.
1. Define defect classes
The first question is not “which camera?” but “which product must be accepted, rejected or reviewed?”. Defect classes should be linked to a quality or economic consequence.
2. Build representative samples
Natural products vary by variety, season, ripeness, size and surface condition. A useful sample set must therefore include edge cases, not only clean reference products selected for a demonstration.
3. Control product presentation
Overlap, rotation, conveyor vibration and uncontrolled lighting can dominate vision performance. Product presentation and spacing are often as important as the inspection algorithm itself.
| Population | Variation to capture | Evidence |
|---|---|---|
| Good product | Natural colour / shape variation | Representative sample set |
| Critical defects | Smallest / lowest-contrast cases | Defect library |
| Borderline cases | Accept/reject threshold | Review matrix |
| Dirty / wet product | Real operating condition | Dynamic trial |
| Seasonal batches | Recipe stability | Validation by population |
4. Test false reject / false accept trade-offs
Any sorting system trades false accepts against false rejects. The correct threshold depends on product value, downstream risk and whether rejected material can be reworked or downgraded.
5. Validate cleaning and seasonal variation
Moisture, dust, wash-down, temperature and recipe changes must be part of the validation plan. Optical performance that is stable for one hour in a clean test area is not yet production performance.
General technical scoping note. Performance, interfaces and responsibilities must be confirmed project by project.