Identifying a ewe to be culled or getting an accurate measure of pasture quality could be as easy as taking a picture.
Former AgResearch employees Greg Peyroux and Benoit Auvray from Iris Data Science in Dunedin are looking at agricultural applications for artificial intelligence (AI) using visual imaging.
And they have found a few.
Initial work is on using computers to accurately analyse images to recognise and identify the faces of individual sheep and the pasture composition of quality species and weeds.
There are also plans for an application to optimise fertiliser effectiveness but exactly what and how is still in the design phase.
Auvray says to develop a practical use for AI means initially training computers to recognise individual sheep, pasture species or weeds then transfer that information into a format farmers can use.
Director and data scientist Auvray says facial recognition for sheep could allow a picture of a ewe having birthing problems to be taken in a paddock on a smartphone.
The image can be transferred to a computer and through facial recognition she can be identified and culled the next time she is through a drafting race.
Iris is also developing a computer-generated condition score for pasture quality and another system monitoring weed infestation to let farmers respond before it becomes a widespread problem.
Pasture images are collected from a GoPro camera mounted on a quad bike or a drone, which the computer then analyses to determine pasture quality, pasture species or degree of weed infestation.
Auvray says they are working on a pasture condition score in conjunction with some farmers but at this stage it measures only the amount of clover and other pasture species and not the volume of drymatter, something requiring further research.
Feedback from dairy farmers indicates a desire and a need for pasture quality measurement, evident by how stock avoid grazing parts of a paddock.
“Dairy farmers really want to know more than just how much drymatter is in the paddocks,” he says.
Identifying a looming weed problem could allow the precision application of a herbicide before the issue becomes widespread.
Peyroux says it is early days for the sheep facial recognition research and he hopes to have a prototype trial on a farm by the end of the year.
The technology could replace ear tags, which can be lost, make it easier to single out individuals or, worked in with an automated drafting system, to split off groups according to age or sex.
The technology is similar to that being rolled out for use among humans in Europe, China and the United States, which, in some cases, has caused consternation with civil liberty groups.
Police in England and Wales are using automated facial recognition to scan crowds for suspected criminals while San Francisco city officials have banned police and other agencies from using the technology.
Auvray says images of the faces of individual sheep, pasture plants and weeds showing a variety of angles are fed into a computer.
That trains the computer to automatically identify specific and relevant features but as the computer learns the process it narrows the key features to focus on and in future the volume of initial data required will diminish.
The facial data can be collected in sheep yards by a computer the size of a cigarette packet, called an Edge Device.
There are still some questions to be answered, such as whether the computer can continue to identify an individual sheep as it ages.
Peyroux says as well as replacing electronic identification and plastic ear tags, facial recognition technology can be used to identify stolen stock.
He hopes it will work out cheaper than tagging every sheep.
Iris is also looking to use imagery technology to measure body condition in sheep based on changes in the body condition and weight gain over a 12-month period.
Work is still under way developing that technology, but the process will track changes in weight and refer that to the body score the previous year to come up with a new body score measure.
In the US researchers are looking at facial recognition in cattle and in China similar work is under way in pigs but Peyroux believes his company is the first to try it for sheep.
Peyroux and Auvray were both employed at AgResearch Invermay, Peyroux as a product development manager and Auvray as a statistician.
When AgResearch announced plans to down-size Invermay and shift most of the scientists to Lincoln they opted to stay in Dunedin.
Peyroux established Iris in 2013 and Auvray worked at the University of Otago in the mathematics and statistics department and for Beef + Lamb NZ Genetics before joining the fledgling business.
MORE:
Greg Peyroux is keen to hear from farmers with feedback or ideas on the projects or who might want to be involved in some of the development work: www.irisdata.co.nz
