Thursday, April 25, 2024

Tech provides picture of milk potential

Avatar photo
Herd testing cows for milk volume sampling can be an arduous affair. But an AgResearch proof of concept trial has shown photos and algorithms may prove almost as accurate as an alternative means to measure how individual cows stack up against their herd mates. Richard Rennie reports.
Reading Time: 3 minutes

Herd testing cows for milk volume sampling can be an arduous affair. But an AgResearch proof of concept trial has shown photos and algorithms may prove almost as accurate as an alternative means to measure how individual cows stack up against their herd mates. Richard Rennie reports.

Using 3D photographic technology to estimate a cow’s milk generating capacity was one of the “left field” ideas AgResearch scientist Dr Paul Shorten tossed in the air when looking at how to make cow milk sampling simpler.

He and his team wanted to develop technology that could provide the benefits of individual meters, but only require one or two sensors within the entire farm dairy to achieve that.

“We considered a few options, including just using a single meter and milking times to determine individual cow milk volumes. But the challenge there was to estimate single cow milk volume accurately,” Shorten said.

Coupling 3D photographic imagery with a computer algorithm, the researchers determined that by photographing a cow’s udder prior to milking when it was full, and then afterwards, volume could be calculated.

The technology is similar to that increasingly used in more intensive livestock systems like piggeries and chicken rearing facilities.

A camera unit installed above the livestock’s pens will photograph regularly and calculate average growth rates for individual animals.

“There is also increasing interest globally in using the technology for determination of teat placement in robotic milking processes,” he said.

Working with a small group of 23 cows at Tokanui over autumn, the researchers found their system could estimate milk yield, and an acceptable level of error, when compared to the in-line milk meters they used to calibrate against.

Their method had an error of .8 of a litre compared to .5 of a litre for the milk meter.

“And this is based off just a camera on a smartphone in its proof-of-concept state,” he said.

Refining the concept will include analysing more cows at different times of the year.

“And it may be that when cows are producing more, such as in springtime, the accuracy will lift further,” he said.

While the images were taken via smartphone then separately processed via laptop algorithm, a refined system would feature a camera system in-situ, with the software integrated into it.

Image assessment is becoming a new frontier for managing aspects of large herd behaviour and health.

Dunedin company Iris Data Science has farmers signing up for an early lameness detection system that uses an on-farm camera to collect data points from every cow exiting the dairy.

Using AI, it identifies the cow’s gait and detects any anomalies that may signal impending lameness issues.

Shorten’s work has also demonstrated the potential of using the imagery to identify particular udder traits that affect a cow’s performance.

Their initial work has included using the computer imagery to determine front/rear teat placement, teat length and teat orientation upon the udder, with 94% accuracy.

The value of this work could feed into robotic milking software, but also highlighted the potential for using computer vision for on-farm prediction of udder traits in dairy cows when examining TOPs (traits other than production).

“And it is quite possible this could be applied to other TOPs, including rump angle, udder support and stature,” he said.

In the meantime, he suspects farmers would find the technology useful for management decisions and within herd rankings of cow performance and for making seasonal decisions, such as what cows to dry off based on milk yield.

Shorten’s research incorporating algorithms and imagery is the latest in a lengthy career that has spanned areas as diverse as coronary heart disease modelling, improving meat tenderness and shelf life prediction of food products.

Other work he is involved in at present includes bioacoustics for determining cow behaviour and the use of drones for pasture quality and quantity assessment.

Total
0
Shares
People are also reading