An Australian farm is set to trial sheep facial recognition technology made by New Zealand-based agricultural technology company NeXtgen Agri.
CEO Mark Ferguson said the trial is expected to kick off “within the month” on his parent’s farm in north-western Victoria.
The setup involves a camera, a single-core computer and "five or six different neural networks" that are designed to match ewes with lambs.
Due to the size of the image files, it is envisioned that processing will happen "at the edge" - in the paddock - rather than by trying to transfer the data out to cloud. It may be that only the results are transmitted out, or alternatively that they periodically downloaded out in the field.
Lamb matching, Ferguson said, "has been a massive hindrance to the development of industry because you never know which are your most productive ewes."
"There's thousands of ewes in the paddock, but you don't know which one's producing the best lambs or the most lambs," he said.
"You've got parts of that information, but not the whole thing, and so that's holding back our ability to be more efficient in sheep agriculture, which has got heaps of benefits from climate change through to like methane intensity through to profitability.
"The trial is about seeing whether the camera can do that autonomously, so work out who's hanging around each other and use that proximity to work out who's related to each other."
The Australian trial marks the first time the technology has been tested "end-to-end".
Ferguson said that a number of technical challenges had been overcome, including around the operation of the camera, and the functionality of the AI models.
The pace of change of hardware had also been a challenge in the project, as had tailoring the solution around "sheep behaviours".
"You obviously need animals to rock up near a camera [to perform a match]," Ferguson said.
It is hoped the trial will enable NeXtgen Agri to iron out any bugs before a planned commercialisation of the technology.
“Obviously we're not expecting it to be perfect but we need it perfect enough that people can get value out [of it]," he said.
Ferguson said there could be other applications for the technology, including for tracking animals in transport.
"If you wanted to make sure that a certain set of animals that had been loaded on a truck somewhere were the same set of animals that came off a truck somewhere else, it would be useful technology," he said.
“There’s lots of parts of the value chain that could use the technology.
"Particularly if you extend that into machine vision generally, there's a whole heap of stuff that will happen in livestock, that’ll be based on computers and cameras working together.
“I definitely see this as bit of a tip of the iceberg."