Cow 4172 steps onto the rotary platform of a 160-cow dairy in Flemish Brabant at 06:30. Her RFID transponder came off near the slurry channel last Tuesday, so for five milkings her yield has been logged against the wrong animal and her concentrate ration has gone to a heifer two stalls down. Nobody noticed, because on a rotary parlour at dawn nobody has time to notice. That small, expensive category of error is what GEA, one of Europe's largest dairy equipment makers, says it can now remove: on 30 September the company announced DairyNet Cow Verification, a camera system that recognises each cow by sight and corrects identification mistakes while the platform keeps turning.
What did GEA actually announce?
A camera-based identification system for rotary milking parlours, announced on 30 September 2026 and due to be shown at EuroTier in Hanover from 10 to 13 November. Several cameras film each cow from fixed angles as she enters a stall. Computer vision and machine learning compare her body shape, build and coat pattern against historical images, verify that she is who the RFID transponder claims, and quietly fix the record when the transponder is wrong or missing. It retrofits onto existing GEA rotary parlours running the company's DairyNet herd management system.
The interesting part is what it does not require. No new ear tags, no extra sensors on the animal, no change to the milking routine. The system runs in the background and is self-learning, improving as it accumulates images of the herd. According to GEA's announcement, it can identify an animal even after she has lost her transponder entirely, provided the cameras saw her during the learning phase.
Why ear tags and transponders are not enough
Because electronic identification fails in ordinary, costly ways. Transponders fall off in cubicles and slurry channels. Antennas misread when two cows crowd a race. A busy rotary assigns the read to the wrong stall. Every misread cascades into the data a modern dairy runs on: milk recording, concentrate allocation, heat detection, treatment withholding periods.
A wrong record is worse than no record. If a treated cow's withholding period sits on the wrong animal's file, the risk is antibiotic residue in the tank, and in Belgium or the Netherlands that is a rejected collection and a penalty, not a warning. If the ration computer feeds the wrong cow, you pay twice: concentrate wasted on one animal, production lost on another. GEA's pitch is essentially that herd data is only worth what its weakest identification read is worth, and cameras are a second, independent witness.
The quieter story: cameras that already watch how cows walk
Visual identification is the newest layer on a stack GEA has been assembling for a while. In March 2024 it acquired CattleEye, a Northern Irish company whose software watches cows from a single 2D camera above the parlour exit and scores their gait for lameness, no wearables involved. That technology now sits in the same DairyNet portfolio the new verification system plugs into.
The evidence behind it is unusually solid for agritech. A validation study published in Frontiers in Veterinary Science in 2023 ran the CattleEye system on three commercial Holstein herds of roughly 180 to 750 cows in Northwest England and North Wales, against 6,040 veterinary mobility scores. Agreement with the vets exceeded 80 percent, and the algorithm proved more sensitive than a trained human assessor at flagging cows with painful foot lesions, detecting 52 percent of them against the assessor's 29 percent.
Lameness is where the money is. A 2023 bioeconomic model in the Journal of Dairy Science, built on European price data, put the cost of a lame cow at about €307 per year, and a cow with digital dermatitis at nearly €392 per year once lost milk, fertility and culling are counted. On a 150-cow herd with typical prevalence, detection that shaves a few weeks off each case pays for a camera faster than most farm software ever will. And the direction matters more than the product: a camera does not fall off, needs no battery, and watches every animal every day. Gait scoring, body condition scoring, identity checks and feed-bunk monitoring are all converging on the same cheap hardware.
When this is not the right move
Honesty first: a fair share of dairy farms should ignore this launch. The verification system targets rotary parlours with DairyNet, which places it on larger herds; if you milk 60 cows through a herringbone or robots from another maker, this product is not for you yet. If your herd is small, your parlour is paid off and transponder losses are rare, the arithmetic will not close. If the internet connection in the parlour barn is poor, cloud-dependent analytics will frustrate you before they help you.
Ask where the data lives before you sign. Images of your cows, your yields and your treatments are competitively meaningful. You should know whether you can export your herd history in a usable format, what the supplier may do with it in aggregate, and what happens to both when you next replace the parlour. Any system that only works inside one manufacturer's ecosystem deserves those questions in writing.
Frequently asked questions
Does the EU AI Act restrict cameras that identify cows? No. The Act's biometric identification rules protect people, not livestock, so recognising cows by coat pattern is unregulated as such. Mind the edges, though: if parlour cameras also capture staff, GDPR applies to those images, and your employees should know what is filmed and why.
Do I have to replace my milking equipment to use AI like this? Not necessarily. GEA's verification system is a retrofit for its own rotary parlours, and camera-based gait scoring such as CattleEye works above a parlour exit regardless of milking hardware. The pattern across suppliers is cameras layered onto existing infrastructure rather than new machines.
How reliable is camera-based lameness detection really? Good enough to act on. In the 2023 Frontiers in Veterinary Science validation, agreement with veterinary mobility scores exceeded 80 percent and the system caught more painful lesions than a trained assessor. It is a screening tool: it tells your hoof trimmer where to look first, not what to do.
Is this only for big herds? For identity verification, today, mostly yes, since it targets rotary parlours. Camera gait scoring already fits mid-sized European herds, and vision hardware keeps getting cheaper. The sensible move for a 60-cow farm is to watch the cost curve rather than be first.
Where this fits in your own thinking
The lesson from this launch is not about one German manufacturer. It is that in dairying, as across agrifood, the AI that earns its keep removes a specific, costed error, wrong cow, wrong ration, lame too long, rather than promising transformation. Finding that error in your own operation, on the farm, in the cooperative, on the packing line, is the real work, and worth doing before any purchase. That is the kind of thinking, and building, we do at Cresly for farms, cooperatives and food producers. If you are weighing where AI would actually earn its keep in your business, write to [email protected].