When a sorter underperforms, the instinct is to blame the camera. In most field cases the camera is innocent — it is being asked to photograph a stream of tumbling, overlapping, unevenly lit product and is doing the best it can with an impossible subject.
Feeding and presentation is the stage that determines whether the optical system ever gets a clean look at each individual particle. It sets a hard ceiling on achievable accuracy, and it is the reason two identical machines can report very different results in different plants.
This article covers what good presentation actually requires, the physical reason throughput and accuracy trade against each other, and the preparation steps upstream that cost almost nothing and recover most of the performance people assume is lost.
What "good presentation" means mechanically
Good presentation reduces to three conditions, and all three must hold simultaneously.
Single layer. Every particle must pass the inspection line without touching or overlapping another. This is the hard requirement, and it is the one most often violated.
Known velocity. Every particle must leave the presentation stage at the same speed from the same point, so the system can compute where it will be at any later instant and aim an air pulse at it.
Known position. The lateral position across the working width must be known, because that is what selects which nozzle fires.
The standard architecture that achieves all three is a vibratory feeder feeding an accelerating belt. The vibratory feeder meters bulk product forward and fans it across the full width. The accelerating belt then runs faster than the feeder, which pulls the pieces apart into a single separated layer, and launches each one off its end on a fixed parabolic path. Industrial optical food sorters are specified with product launch speeds of 2.5–3.5 m/s for exactly this reason.
Because every piece leaves at the same speed from the same point, the downstream system can predict each piece's trajectory — and prediction is what makes a 5–30 ms timed air pulse possible.
Why overlap defeats the algorithm
There is a specific, mechanical reason overlapping product destroys sorting quality, and it is worth understanding because it explains the throughput trade-off.
When two or more defects appear in the same camera frame at the same instant, the ejection system can only remove one of them. The processor has resolved them as a single object; one valve fires; one particle is deflected; the other walks straight into your finished product.
This is not a software limitation or a processing-speed issue. It is a consequence of material arriving faster than the machine can physically separate it. And it means that as feed rate rises, accuracy does not fall gradually — it falls off a cliff once the layer thickness crosses the point where particles begin to overlap.
The same logic applies to occlusion. A defect hidden behind a good kernel is not a detection failure; it was never presented.
Bed depth is where throughput and accuracy trade
Every operator faces the same temptation: increase the feed rate, get more tonnes through the same machine. The physics says otherwise, and the mechanism is the overlap effect above.
Thinner bed depth gives the machine more time and more space to separate each particle, which raises the proportion of defects actually ejected. As the layer thickens, overlap rises, and the proportion of defects that can be resolved and removed falls.
This is why the professional way to size a sorter is to allow headroom rather than to run at the limit. A commonly applied rule in materials-processing procurement is to reserve roughly 20% capacity margin — a plant needing 5 t/day in practice should specify a machine rated higher, because a sorter running permanently at 100% of rated capacity is a sorter running permanently at its worst accuracy.
The corollary is that buying slightly larger than your current volume is usually the cheaper mistake. Undersizing caps your output permanently and degrades the quality you can sell; modest oversizing costs capital once and leaves you the option of better settings.
Preparing the material before it reaches the sorter
The largest and cheapest performance gains are usually upstream of the machine, in preparation. Three variables matter most.
Moisture. High moisture causes product to clump and stick, and clumps cannot be separated into a monolayer by any feeder. For crushed and granular materials, industry guidance keeps moisture content at or below about 5%. If your product arrives wetter, drying or at least conditioning ahead of the sorter will recover more accuracy than any camera upgrade.
Size uniformity. Wide size distribution makes a single presentation and a single nozzle pitch work badly for both ends of the range. For crushed materials, uniform feed particle size in roughly the 2–20 mm band is the usual target. Where the distribution is wide, pre-classification by screening into size fractions before sorting is often the single highest-value change available.
Removal of bulk foreign material ahead of the sorter. A sorter is a precision finishing device, not a primary cleaner. Stones, metal, stalks and heavy trash should be removed by magnets, de-stoners, aspirators and screens before the product ever reaches the optical stage. Every piece of obvious trash you remove mechanically is capacity you have returned to the sorter for the defects that genuinely need optics.
Presentation differs by architecture
The three architectures present product differently, and each suits different materials:
- Free-fall chute machines accelerate product down inclined channels by gravity. Compact, high-capacity, and the standard choice for free-flowing granular product such as rice, grain, beans and sesame.
- Belt machines convey product flat on a moving belt and launch it from the belt end. Chosen where product must not free-fall — fragile, large or irregular items, including many nuts and kernels where impact damage matters.
- Waterfall / cascade configurations for specific high-flow duties.
We treat the selection reasoning separately in chute-type versus belt-type color sorters; the point here is narrower — whatever the architecture, the feeder must deliver a monolayer at constant velocity, and the architecture determines how that is achieved.
Presentation faults worth checking
When results slip, these are the causes worth eliminating before touching any software setting:
- Bridging or rat-holing in the hopper, which produces surging feed — alternating starved and overloaded, and wildly inconsistent accuracy
- Uneven distribution across the width, leaving some channels starved while others are overloaded. The machine is only as good as its worst-fed channel
- Dust accumulation, which both fouls optics and indicates product being abraded in the feed path
- Worn or mistuned vibrator mounts, giving irregular metering
- Feed rate set above the monolayer limit — the most common cause, and the one operators resist because it is the only one that requires slowing down
A practical commissioning checklist
Before accepting a machine or after any performance complaint, verify presentation in this order:
- Is the layer a monolayer at the inspection line? Watch it, do not assume it.
- Is distribution even across the full width?
- Is the feed rate at or below rated capacity, with margin?
- Is moisture within specification for the product?
- Is size distribution controlled, or has it widened since commissioning?
- Is bulk foreign material being removed upstream?
- Is the viewing window clean and the background calibrated?
Most "camera problems" resolve somewhere in that list.
How JIACUI machines handle presentation
Presentation hardware is specified to the product, not averaged across the range.
The chute range — CS-QB16 (20–100 kg/h, suited to laboratory and small-batch work), CS-HA32, CS-4A, CS-5A, and the CS-7A at 6–12 T/H — uses gravity-accelerated channels matched to free-flowing product. The belt range — CS-LA300, CS-LA600, CS-LA1200 and the double-deck CS-LA1200D — carries material flat for fragile or irregular product, with the double-deck arrangement enabling a second sorting pass that recovers good product carried over in the first.
Our published baseline remains sorting accuracy ≥99%, stated at its test condition: polished japonica-type rice at approximately 2% foreign content. That figure is a machine specification; the achieved figure on your product will depend substantially on the six presentation checks above.
FAQ
My sorter misses defects at high feed rates. Is the camera too slow?
Rarely. At high feed rates the cause is almost always bed depth — particles overlapping in the frame, so the system resolves two defects as one and can only eject one. Reduce feed rate and rejection efficiency will usually recover immediately, which confirms the diagnosis.
Should I buy a bigger machine than I need?
Modestly, yes. Reserving around 20% capacity margin is standard practice, because a machine run at its rated limit is run at its worst accuracy. The reverse — buying far larger than your volume — ties up capital for capacity you will not use.
Can wet product be sorted?
Not well. Moisture causes clumping that defeats monolayer formation, and it also fouls optics and chutes. Bringing moisture to within specification, commonly around 5% or below for granular materials, is usually a prerequisite rather than an optimisation.
Why is one channel performing worse than the others?
Uneven distribution across the working width. The feeder is delivering more to some channels than others, and the overloaded channels are running above their monolayer limit. Rebalance the feed before assuming a hardware fault.
Does pre-cleaning really matter that much?
It is the highest-return change available in most plants. Mechanical removal of stones, metal, stalks and fines ahead of the sorter returns that capacity to the optics for the defects that actually require it, and it reduces dust load on the optics.
How often should I re-check presentation?
Make it a shift-start check, because clumping, bridging and moisture drift are all time-varying. Our maintenance checklist sets out where presentation checks sit in the daily and weekly routine, and ROI and capacity planning covers how to size throughput against it.
