Single-file feed
Product is separated so it passes the camera one piece at a time. The inspection and ejection zones are kept apart.
Optical sorting
ARYAS VisionSort
A vision-based sorting system. We design the belt, the enclosed light cabin, the camera and the classification as a single chain, so mouldy, rotten, discoloured and foreign-material product leaves the line.
What it does
VisionSort inspects dried-food product piece by piece and physically removes what fails. The same machine is reconfigured for different product families — nuts, dried fruit and mushrooms among them.
What distinguishes the system is not the camera but the conditions the camera works in: controlled illumination, a fixed viewing geometry, and decision rules tuned to the product family.
The processing chain
Six stages from the feed point to the ejection point
Enclosed light cabin
External light is blocked, illumination is held constant, and the camera is isolated from vibration.
Imaging
Synchronous capture for a moving line. Product with irregular surfaces is imaged from more than one angle.
Classification
Each image passes through the computer-vision model separately and produces its own findings.
Decision
Per-image findings are merged into a single verdict: accept, reject, or hygiene risk.
Physical ejection
The product is removed by computing the delay between decision and actuator together with where the product is at that moment.
Why an enclosed cabin
Image quality is usually won or lost in the cabin, not in the model.
The enclosure blocks external light, holds the illumination steady and protects the camera from vibration. The right light makes a defect visible; the wrong light blinds even the best model. That is why the light geometry — bright field, dark field, backlight, dome, coaxial, low-angle — is chosen before the model, according to the defect type.
A fixed camera under controlled light removes most of the variation a handheld or moving camera introduces. The model then solves one problem instead of two.
External light isolation and constant illumination
Light geometry selected for the defect type
Vibration isolation and a fixed working distance
The decision model
Every piece resolves to one of three outcomes — and the two reasons for rejection are not conflated
No finding
The product stays on the line and continues through the normal flow.
Cosmetic or separable finding
Discolouration, mechanical damage or foreign material. Removed from the line, and can be routed to a re-sorting pass to reduce the loss of sound product.
Food safety finding
Mould or rot. Removed unconditionally and never re-sorted.
A finding may be visible from one angle and hidden from the others — the product's own body can occlude it. Findings are therefore merged with a "real if seen from any angle" rule rather than "all angles must agree". In a quality and hygiene context, the cost of missing a finding is far higher than the cost of one extra inspection.
Defects detected
The colour coding shows which verdict each defect leads to
Mould
Mould growth on the surface.
Rot and spoilage
Moisture-driven spoilage, softening, staining.
Insect and worm damage
Infestation and insect-bored holes.
Foreign material
Soil, twigs, leaves, stones.
Bruising and breakage
Mechanical damage, fragmentation, deformation.
Colour and darkening
Colour deviation and oxidation-driven darkening.
Product profiles
The same machine, configured to the product family
Nuts
- What varies
- Shell texture, specular glare from oily surfaces, size distribution.
- What gets reconfigured
- Light geometry, glare suppression, classifier profile.
Dried fruit
- What varies
- Colour range, surface wrinkling, stickiness.
- What gets reconfigured
- Colour reference, feed separation, decision thresholds.
Mushrooms
- What varies
- Irregular body shape, pale surface, occluded defects.
- What gets reconfigured
- Number of imaging angles, light angle, decision rules.
Line integration
The machine is built to run as part of your existing line, not beside it.
PLC triggering and status exchange
Communication over industrial protocols such as OPC-UA and Modbus TCP/RTU
Defect counts reported to SCADA or MES
Output suitable for feeding HACCP and ISO 22000 traceability records
How we start
Visibility comes first, not the model.
For each product family the imaging chain — wavelength, sensor, lens, light, filter — is designed for that product. Work begins with a visibility trial on your own samples: we first demonstrate physically which light and which wavelength show the defect at the highest contrast. The model comes after that.
A visibility trial with your samples
Light and wavelength chosen for the defect type
A classifier and decision rules specific to the product family
Mechanical and cabin design matched to your line conditions
Frequently asked questions
Which products can it be used for?
Dried food products, principally nuts, dried fruit and mushrooms. The light geometry and classifier profile are reconfigured for each product family; the machine is not locked to a single product.
Why are mould and rot a separate category?
Because the consequence differs. Discolouration or bruising is a cosmetic finding and can go to a re-sorting pass. Mould and rot are hygiene findings; they are removed unconditionally and never reassessed.
Will it integrate with our existing line?
That is the intended way to run it. PLC triggering, communication over industrial protocols and reporting to SCADA or MES are within the system's design scope. The mechanical layout is determined by the conditions of your line.
Why image from more than one camera?
The product's own body can hide a defect from a single angle. With multiple angles, findings are merged under a "real if seen from any one of them" rule, which lowers the number of defects that slip through.
How does the process begin?
With a visibility trial on your samples. We physically demonstrate which light and wavelength show the defect best; the classification work is designed after that.
Let's Assess Your Product Together
Talk to our team about a visibility trial with your own samples