麻豆传媒

Pigs could be issued with biometric passports based on facial recognition technology, giving farmers a more practical and welfare-friendly way of identifying individuals than ear notches or tags, the current industry standards.

Identifying pigs based on their unique facial features could enable them to receive individualised food and veterinary care, and be traced as they go through meat processing. With advanced algorithms and machine learning, it is possible to distinguish between the faces of even the most similar-looking breeds, says at the University of the West of England (UWE) in Bristol, UK.

Zhang鈥檚 team has also been involved in the development of a new ticketless underground travel system based on human facial recognition. 鈥淚t made us think, if we can do it with humans, why can鈥檛 we do it with animals?鈥� says Zhang鈥檚 colleague , who is also at UWE.

To test the idea, Zhang and his team took thousands of photographs of sows from the Large White, Landrace and Hampshire breeds using a single webcam positioned above a watering spout. The researchers placed bars on each side of the spout to help position each pig鈥檚 head in front of the movement-triggered camera and avoid having other pigs in the image. This allowed them to get still shots of dozens of pigs from essentially the same angle every time.

They fed the images into a type of machine learning program called a convolutional neural network, which detected the various subtle differences in pigs鈥� faces, mainly around the animals鈥� snouts, foreheads and eye regions. The software then ran tens of millions of tests that enabled it to learn to identify each pig.

After this training, the system accurately identified the pigs nearly 97 per cent of the time, says Zhang. He presented the research at the in Hartpury, UK, last month, as an example of how facial recognition may be applied to many domestic animals, including horses. The facial recognition program is still under development in collaboration with Scotland鈥檚 Rural College.

Currently, piglets have either numerical stamps on their bodies, microchips under their skin, notches clipped into their ears or tags pierced onto their ears that farmers and veterinary personnel can scan to identify them. This is important for recognising the animals that need specific medications or food based on veterinary evaluations.

It also plays an important role in traceability as animals go through meat processing plants, which is critical for both animal and human health because it provides a record of each animal鈥檚 medical history and origin.

But piercing, notching and microchipping can be painful for pigs and time-consuming for farmers, especially on high-intensity farms, which can hold hundreds or even thousands of pigs, says Smith. When radiofrequency tags 鈥� which work like contactless payment cards 鈥� are used, tag reading only has an accuracy rate of 89 per cent even at a close distance.

The team鈥檚 facial recognition system is more accurate and works with very low-resolution images of 64-by-64 pixels, taken automatically without interfering with the pigs at all, says Zhang. 鈥淲e don鈥檛 actually need expensive systems for this. If the algorithm is good enough, it can do the job,鈥� he says.

at Eastern Institute of Technology in Hawke鈥檚 Bay, New Zealand, finds the system particularly interesting for animal welfare. 鈥淎s technology advances, there are exciting possibilities for non-invasive, unobtrusive identification of individual animals,鈥� she says.

Not only does facial recognition prevent having to physically mark the animals, but it also offers the opportunity to 鈥渃ater to an individual鈥檚 specific [nutritional or veterinary] needs where animals are kept in large groups in intensive farming conditions鈥�, says Waran.