“Nearly Perfect” Accuracy in Cattle Identification Test Using Facial Recognition Technology


A UNIVERSITY researcher developed an AI-based facial recognition system for cattle that demonstrated near-perfect accuracy in early tests.

Human facial recognition technology, used to unlock smartphones and speed up immigration processes at airports, has quickly become normal for people. Today, a University of New England researcher hopes to make it the norm in the livestock industries.

Click on the image to enlarge

For his doctoral thesis, Ali Shojaeipour has developed an artificial intelligence capable of identifying individual cattle via their muzzle patterns with an accuracy of better than 99 pc.

Dr Shojaeipour believes that if it is extended, its technology could replace stock tags and ear tags as a non-invasive and non-tampering method of identifying livestock. Such an innovation would bring significant benefits to both the Australian and global livestock industries.

“We were able to show that biometric livestock monitoring could potentially be used in future agricultural decision support systems for resource management at multiple scales of production,” said Dr Shojaeipour.

Taking the technology from its current proof of concept to commercial application will be a big undertaking, the Iranian-born computer scientist acknowledges, but success could revolutionize the handling of livestock.

Dr Shojaeipour focused on cattle muzzles to achieve “contactless” animal identification. Small variations in shape and pattern make the muzzles as distinct as the human fingerprint.

The researchers previously manually extracted the dimensions of the muzzle from images. For his AI-driven approach, Dr Shojaeipour had to take a two-step approach.

The software first detects the muzzles of cattle in an image and then uniquely identifies the animal using its muzzle pattern. He then built a process based on machine learning that allows the biometric identification of different cattle, able to adapt continuously and as new cattle entered the database.

The results were gratifying: in his test herd of 300 cattle, his system achieved an accuracy of 99.11%.

Dr Shojaeipour is in the process of creating a smartphone app that can identify individual cattle through the phone’s camera.

However, Dr Shojaeipour says his technology requires some development before it is ready for the reality of the mud and grime of farms, feedlots and sales yards.

He built his AI models during the coronavirus lockdown, when his ability to work with real cattle was limited. As a result, “while other people have pictures of their families on their walls, I have pictures of head of cattle.”

Getting the technology to work with moving cattle, with dirt and other distractions, will require considerable work – but Dr Shojaeipour believes it is possible.

He is currently undertaking this work himself, but is seeking support to build a team capable of moving the project forward quickly and translating academic findings into industry impacts. If it is successful, it may not be long before livestock identification takes on a new AI-powered form.

Dr Shojaeipour’s latest publication on his work, Automated Muzzle Detection and Biometric Identification via Few-Shot Deep Transfer Learning of Mixed Breed Cattle, appeared in Agronomy. (Click here to display).

Source: A


Comments are closed.