WITH the recent Australian record of 59,487 sheep tags scanned on sheep sold in one day last December at the Hamilton Regional Livestock Exchange, it may be of interest to many local graziers to learn that the National Livestock Identification System (NLIS) concept has a significant potential technological augmentation on the horizon.

A New Zealand-based scientist, has with his project team, been working on adapting facial recognition systems for not just telling sheep apart, but tracking behaviour and assessing problems before they become visible, and his early days were here in Western Victoria.

neXtgen Agri chief executive, Doctor Mark Ferguson grew up in Hopetoun, north of Horsham, and his career took him to Hamilton in 1998 first as a research officer and then as a sheep production research scientist for the Department of Environment and Primary Industries which “involved managing key areas of a large Merino ewe research project funded by Australian Wool Innovation (AWI)”.

“The project known as ‘Lifetime Wool’ aimed to quantify the impact of maternal nutrition on the lifetime performance of progeny,” he said.

“I was responsible for the management of data collection, analysis and interpretation for the Victorian component of this national project.

“’Lifetime Wool’ experienced unprecedented success in changing farmer attitudes towards the nutritional management of breeding ewes and resulted in significant practice change across the sheep industry in Australia.”

After a similar position in Perth for four years and then also lecturing at Murdoch University, he travelled across the Tasman to manage production science for The New Zealand Merino Company including investigating genetic resistance to footrot, developing training material materials and monitoring welfare with on-sheep sensors.

Then roughly five years ago a client made an off-the-cuff comment to adapt human facial recognition technology to sheep; “if it (worked) in humans, it must be able to work in sheep – that’s probably about as throwaway as it was” and he set about investigating if the concept was even plausible.

“I then looked into it a bit … I had no idea about what machine learning even was,” Dr Ferguson said.

With further inspiration coming from a USA conference he attended, where one speaker promoted artificial intelligence as a critical future trend, he came back to New Zealand “to give it a crack” and secured AWI funding to do a proof-of-concept and “turn possible into practical”.

The challenges of combining agriculture experience and technological skills to solve problems have made the project unique.

“I’ve had five people working on it, and they’re all obviously non-Ag people,” Dr Ferguson said.

“The project is quite interesting, so we don’t have a lot of trouble getting people but getting them to think about Ag problems has been … one of the challenges.

“Mainly around animal behaviour and making sure you’ve got cameras in the right spot … building the machine learning is a massive task – that’s taken up a lot of our time.

“(We’ve) always (had) in mind thinking about how a farmer would need to do it.

“The biggest challenge was to get several images of the same sheep over different times … it’s not easy.

“For (the) project, four cameras were added to a manual weigh crate to capture photos of sheep in a standard environment. The cameras were in front, behind, over-head and to the side of the sheep. The team took roughly 100 images from each camera of each sheep.

“More than 4000 sheep were photographed resulting in almost 1.5 million images or 500GB of data.

“Once the model was appropriately trained, the machine was able to determine the identity of a sheep based on seeing a single image of the sheep with up to 99 per cent accuracy.”

Dr Ferguson said this meant whilst an individual sheep “can be identified regardless of whether the tag can be visually or electronically read” and this has application for things like stolen stock – “you can cut an ear tag out, but you can’t replace a face” - the main reason for the project was to enhance management of the animals.

“We’re really focussing on lamb to dam, so working out which lambs belong to which ewes,” he said.

“(It’s) so we can work out which are our most productive ewes in our flocks.

“One of the things I’m interested in is when they give birth so that we can accurately age (lambs) so that when we’re working out what weight they wean we know that one was 60 days and one was 100 days and we can account for that.

“Also welfare monitoring, things like fly strike or worms or dags or whatever, those things are all very possible with a camera … it’s things that happen in the paddock, that’s where we’re interested in, we’re not so interested in the yard-based stuff at the moment.”

He was also pleased to report on the predictive ability of the technology on the liveweight of sheep.

So far the system has shown “considerable promise” in its accuracy and needs some “fine-tuning” but potentially having strategically-placed cameras on a farm giving liveweights of sheep would definitely be an advantage.

He is grateful for the support of the New Zealand Government with innovation funding to help the project along, but believes the generally less hilly terrain in Australia will be where a lot of interest will come from, as the technology will have less challenges to overcome.

“(The) flat paddocks around Western Victoria will be pretty handy … (it will be) easier for the tech to work there than it is here,” Dr Ferguson said.

“I’d imagine this will have better uptake in Australia.”

While he doesn’t see the technology replacing the NLIS, he would like to think this would be “supercharging” identification database tech “rather than replacing it” and could be extended to cattle and other animals in the future.

“I guess my personal bias is sheep, so we’ve done sheep,” Dr Ferguson said.

“We’ve definitely got other species on our mind – maybe I should’ve started with cows, there’s less of them, it might’ve been easier!”