Product Recognition, Classification and Counting
Recognise and sort products on the conveyor by shape and colour.
Counting and sorting are prime examples of work people can do, but cannot sustain at the same accuracy over long periods.
The model recognises the product by shape, size and colour together; touching or overlapping products are separated one by one with instance segmentation. Counting runs in sync with conveyor speed, and encoder data prevents double counts and missed items. The classification result can drive a sorting actuator directly, so inspection does not stop at a report — it steers the line itself.
How it works
Touching products are split
Instance segmentation processes overlapping products as distinct objects.
Shape and colour together
Geometry and colour are combined to determine product type; colour deviation is measured too.
Command to the actuator
Class information goes to a pusher, air jet or robot; the product is routed to the right lane.
Where does the counting discrepancy arise?
The gap between warehouse and line counts is an accepted uncertainty at most plants. Camera-based counting makes it measurable; once you see at which station and on which product type it accumulates, inventory correction becomes record-keeping rather than guesswork.
| Single-file flow | 99.6% |
| Dense flow | 99.1% |
| Touching products | 98.2% |
| Transparent product | 96.4% |
| Mixed sizes | 97.8% |
Accuracy on transparent products depends on the lighting method; adding back-lighting brings it close to single-file flow levels.
| W1 | 2.8 |
| W2 | 2.4 |
| W3 | 2.0 |
| W4 | 1.6 |
| W5 | 1.3 |
| W6 | 1.1 |
| W7 | 1.0 |
| W8 | 0.9 |
The gap does not reach zero; once measurable it is pulled into a manageable range.
Use cases
Product count and size
Size control, counting and ripeness classification.
Live counting
Award-winning fish counting: individual counting in dense flow.
Part counting
In-box part counting and sorting by type.
Sorting by colour
Separating colour-coded products on the line.
Technical requirements
| Camera | High frame-rate camera; line-scan option depending on line speed |
| Sync | Encoder connection locks counting to conveyor speed |
| Lighting | Diffuse top lighting; back-lighting for transparent products |
| Sorting | Class-based routing via air jet, pusher or robot |
| Output | Live count, type distribution, shift and batch reports |
Frequently asked questions
How does Product Recognition, Classification and Counting work?
Counting and sorting are prime examples of work people can do, but cannot sustain at the same accuracy over long periods.
Are our existing cameras suitable for this solution?
Required camera specification: High frame-rate camera; line-scan option depending on line speed. Before installation we assess your camera inventory free of charge and report which points are suitable and, if needed, which camera should be added.
What results and indicators does it provide?
Key indicators: Counting accuracy: 99.3%; Maximum flow rate: 1,000 units/min; Separable product classes: 8. Measured counting accuracy: 99.3%. Accuracy on transparent products depends on the lighting method; adding back-lighting brings it close to single-file flow levels. Thresholds are calibrated on your own site during the pilot.
Is it compliant with KVKK, Türkiye's data-protection law?
Footage is processed on an edge server inside your facility; raw video is not sent to the cloud. Facial recognition is not used; analytics outputs are anonymous and event clips are deleted automatically when the defined retention period ends.