Digital eyes help Queensland beef producers level up productivity

Around half of Australia’s national cattle herd is raised in northern Australia – an area typified by properties so enormous that the vast distances between cattle and infrastructure make regular handling and monitoring of a herd logistically challenging, time-consuming, and expensive.  

But you can’t manage what you don’t measure. And in the quest to improve herd productivity, the focus is on the inability to regularly muster and inspect cattle which results in missed opportunities for early intervention in reproductive management – a key productivity driver. This includes identifying non-productive cull cows, and identifying if and when a cow calves. 

RELATED: More than just office yoga: Farmwall’s fresh approach to scope 3 emissions 

Now, a collaboration between the Department of Agriculture and Fisheries, Queensland (DAF) and InFarm is investigating how the ‘other AI’ – artificial intelligence – could help Queensland producers combat low fertility in extensive beef herds. 

Prototype automated camera system locates, identifies, and monitors cattle

At the Queensland Government’s Brian Pastures Research Facility near Gayndah, the collaborative research project has created a prototype camera system to detect cattle and monitor their condition using machine vision and AI.  

Funded through the Queensland Government’s Digital Transformation in Agribusiness Initiative, the project has captured more than 750,000 images and 86,000 videos to train the AI algorithm.  

Founder and Managing Director of InFarm, Jerome Leray, explained the process as ‘Tinder for cows.’ 

“Training an AI comes down to three core things: collecting the data; labelling it to teach the AI what features are important; and then validating the results. We look at an image and ask, ‘Is there a cow here?’ And we swipe right if there is and left if there’s not (or whatever way Tinder works!)” he says.  

It’s simultaneously complex and simple technology. Through repeated exposure to images containing cattle, the AI ‘learns’ to identify features of a cow as distinct from a calf (or pasture, fences, wildlife, and any other object that is also captured in the image.) 

RELATED: Bowen Gumlu Growers Association driving agritech adoption in North Queensland 

“We made sure the AI saw cows from different angles: face on; from the side; from the top as they move through the farm,” added Jerome.  

“We also used images in lush green pastures versus when cows had been there and grazed it to the ground. 

“All those different variables are things the AI needs to see to generalise and learn that what we’re really interested in is the cow.”  

Jerome Leray, Founder and Managing Director of InFarm, explains the technology at Beef2024 as part of the Tech and Innovation Talks program.

Overcoming connectivity and infrastructure challenges

The AI system offers the eyes (and intelligent insights) that can help Queensland producers to operate their business more intensively.  

“We basically digitise the knowledge that we have in our head, and then scale it so it can be applied and leveraged by a computer that’s available 24/7,” Jerome says.  

“It means we can have digital eyes on the animals at all times.” 

But for widespread adoption, researchers realise they need to confront two key constraints that would limit uptake in extensive grazing systems: the connectivity to transmit data, and the infrastructure to situate and power the system.  

Paul Stewart, Principal AgTech Scientist at Department of Agriculture and Fisheries (DAF), explains, “Other industries using [visual] AI, like the dairy industry, have sheds that make it easy to mount a camera system, and power is already available.” 

“But when you’re talking about sparsely populated paddocks, and cattle moving around properties that can cover hundreds to thousands of hectares, the question is: How can we make these systems work?” 

Cattle are monitored with AI technology.

RELATED: ‘I need to be a part of this’: Giaan Rooney on farming and how she’s found her place 

The Brian Pastures project tackles connectivity through a hub and spoke model with a central connection at the office synced to cameras located at strategic points across the property – powered by solar panels and batteries. 

Paul explained, “Brian Pastures is a very hilly property with lots of peaks and valleys, which makes connectivity challenging. Our hub and spoke model, using repeaters where necessary, delivers point to point connectivity of about 19 kilometres. But our system is overkill for the average producer. 

 “The long-term goal is that the system, once trained, performs its tasks remotely, only sending alerts when it identifies an issue a producer needs to address – greatly reducing connectivity requirements.” 

On the infrastructure challenge, DAF is using a temporary structure made of portable panels to mount the cameras, but Paul says they’ll move to a single pole to replicate real world use. 

RELATED: Five fast facts to get your boss to sign off on your evoke^AG. 2025 leave 

“We’ve been testing around watering points such as troughs and dams. But during the wet season that can be problematic, because cattle are not coming in as often. So, we’re also going to look at places like lick stations, gateways, and lanes that cattle walk through regularly.”    

Paul Stewart, Principal AgTech Scientist at Department of Agriculture and Fisheries (DAF)

Diverse use cases deepen the value proposition

Jerome sees the value proposition from two sides: the producer and the industry.  

“For the producer, it’s production and profit. Did the cow produce a calf? Did she lose the calf? When? Did she abandon her calf at the watering point? Once we know this stuff, we can make management decisions around who our cull cows might be. That’s just one easy win for northern Australia.

But the system is also being developed to help with other challenges impacting production, such as buffalo fly which can affect feeding and weight gain, and bottle teat as a contributor to calf mortality. Weight estimation, growth rates, and body condition scores (an estimate of fat reserves) are also potential use cases.  

RELATED: Soaring from startup to global company: How Bird Control Group’s lasering in on success and pests 

“We had a vet train the AI to provide a body condition score using 3,000 images in just half a day. It needs more training to validate, but even in that short space of time, the AI is within 0.6 of a body condition score determined by a human,” said Jerome. 

From the industry’s point of view, the value proposition is just as strong. The system will be developed into remote surveillance nodes that can detect animal injuries and lameness, and symptoms of emergency animal diseases such as Lumpy Skin Disease and Foot and Mouth Disease – offering an early warning system to secure the industry in the event of an outbreak. 

“The potential benefits of this technology are very exciting,” concluded Paul.  

“Not to mention, far reaching. Once upon a time, it was science fiction that we’d have eyes and ears across our properties to make management easier and more effective.” 

“Now, we’re inching closer to that being reality.”