Tracing a cow’s ‘fingerprint’ through her unique muzzle

In Australia, University of New England computer scientist Dr Ali Shojaeipour developed artificial intelligence tools to identify individual cattle with 99 per cent accuracy using their muzzle patterns.

Facial recognition software identified small variations in the shape and patterns of the cattle’s muzzles.

Dr Shojaeipour was able to build a self-learning biometric identification model that was capable of continually adapting as new data was entered via new photos of additional cows.

His initial modelling using 300 cows indicated 99.11 per cent accuracy of muzzle identification.

However, in a UNE article in 2021, he said his technology needed further development to ensure accuracy in the manure and dirt reality of farms, feedlots and saleyards.

Now Dr Phillip Zada, CEO of Australian software engineering company Stoktake, has developed the technology that ensures accuracy even when a cow’s muzzle is covered in manure.

He is taking Stoktake onto the farm, with a research project at Ellinbank Smartfarm, the dairy industry’s leading research farm in West Gippsland.

Stoktake, like other livestock facial recognition programs developed in recent years, is designed for farmers to take photos or videos of their cattle using their smartphone or tablet.

Stoktake also integrates across other software programs such as those providing animal management, feed, fertility and productivity data for each cow.

Dr Zada is also the Z Ware CEO, a company which designed the software engineering program as part of a brainstorming session, or hackathon, between employees.

“I invited Ali to work with us, and in the early days of the project, we were able to use some of his research,” Dr Zada said.

“Not only were we able to develop Stoktake and get it to work, we were able to get it to work differently to anyone else.

“It will work even when there is dirt or manure on the cow’s muzzle.”

Stoktake’s initial database of 500 cows was built through field tests and pilot programs, then expanded to 1500 cows, with a 99.65 per cent match rate on the muzzle.

Dr Zada said a mix of corporate and government funding had enabled global trials in countries including South Africa, Botswana, Brazil, the United States and others — leading to a database of 25,000 dairy and beef cows.

Now Dairy Australia is funding a research project at Ellinbank Smartfarm using Stoktake.

“The interest has come from small and large producers — the small producer with 30 to 40 cows is just as interested as the larger producers in Brazil with half a million cows,” Dr Zada said.

“Every country has their own issues with stolen cattle. Some countries like Australia have a national registration system, some don’t.

“One of the most common technologies out there are eartags. But when cows get stolen, it’s not that hard to take the eartag out.”

In Australia, research by PwC has identified an estimated 28,000 cattle are stolen each year, at an estimated $105 million annual cost to the industry. This data is across the Australian dairy and beef industries.

“After trialling Stoktake against crime in several countries across 12 months, we know it works effectively,” Dr Zada said.

He spent time last year liaising with police, academics, farmers and others to develop Stoktake as a solution against crime for Australian farmers.

One of the first challenges working with dairy cattle at Ellinbank was to sort out the assumptions within Stoktake.

“We’ve been experimenting with camera placements and in different lighting and weather conditions,” Dr Zada said.

“One of the things we uncovered at the beginning was the cows at Ellinbank came through the yards to the dairy about two hours later than we expected.

“This meant the lighting in the yard wasn’t ideal for identifying cows by muzzles.

“We realise we need to improve lighting in the yard, given when cows generally come into the dairy for milking.

“But in the bail, there was sufficient lighting to identify the muzzle.

“So the first stage of this project at Ellinbank has been about identifying the best placement for the cameras.”

The cameras, firstly, take the initial photo of the cow, and secondly, use the engineered software algorithm to facial recognise the cow.

Stoktake can be integrated with other livestock management systems, meaning it can be part of the program on a farm that is already correlating herd management, fertility and reproduction data, feed management, vaccination records and other health data, weight and fat scoring, location on the property and sales data.

“It was built for integration with other software programs,” Dr Zada said.

“It also has the potential to identify cows that are coming through the yards that shouldn’t be, because they’re sick and are supposed to be milked separately.”

He said Stoktake can be used offline.

“In Botswana, farmers recorded videos and photos of their cattle on their phones, then uploaded that information to the database when they have a mobile connection.”

As well as the research project using dairy cattle at Ellinbank Smartfarm, Stoktake is scaling up this year through two additional partnerships — one in Brazil involving 170,000 head of a mix of dairy and beef cattle, and one in South Africa with 50,000 head of a mix of dairy and beef cattle.

There is an Australian patent pending confirmation for Stoktake in its final stages.