Producers needed to develop cow facial recognition software

Researchers at the University of Kentucky need your help in developing facial recognition software for beef cattle.

Amanda Radke

January 30, 2019

6 Min Read
jackson cow map small3.png

Earlier this week, I shared some information about a Chinese company that is working on software that uses artificial intelligence to recognize and identify individual cows on a dairy farm to monitor animal health and wellness.

Here in the United States, Cargill has also invested in similar technologies that would help producers manage their individual animals based on facial recognition.

READ: Emerging technology could identity cattle through facial recognition

Following my blog post, I was delighted to hear from Josh Jackson, University of Kentucky (UK) assistant Extension professor of livestock systems, who shared with me how UK is working to develop and hone facial recognition software for cattle.

Jackson says they need producers help in developing this software. To assist in this endeavor, it’s as simple as snapping a photo on your cell phone.

Jackson says, “Here at UK, we are trying to develop and hone facial recognition software. This software is not for monitoring people but, rather, for identifying cows as part of our drone-related research. Several engineers, including myself, are trying to figure out how well this can work. We are turning to you, the public, for a crowdsourced solution.

“Crowdsourcing is a relatively new concept which is aided by the internet and social media. Crowdsourcing allows for many people to come together to find a solution to a problem. We are looking for help from anyone and everyone across the state and beyond, so please share this on your favorite social media platform tagging the photos with #ukcowmap. We need photographs of cow facial features or, in essence, the cow’s portrait. Please submit photos to [email protected].”

Related:Drones, smart ear tags & cameras: The case for using technology in ranching

Currently, Jackson and his team are looking at cattle on the UK beef farm and farms in the surrounding area to gather a diverse set of images of cattle within the region. However, more individual photos are needed to further enhance the efficacy of this software.

“We are looking for any and all cattle facial photos,” says Jackson. “We are trying to compile as extensive a database as possible. Therefore, we want pictures of bulls, heifers, calves, cows, steers of all ages. We are also looking for photos from all breeds: Angus, Charolais, Hereford, etc., and even a yak if you have one. It makes no difference to me if the cow is purple. We are expecting the identification of registered Angus/predominately Angus to be the most difficult to distinguish as they will lack the unique patterns that a Hereford could possess. Also, we are expecting a lack of contrast of features with the Angus to create further obstacles.”

He provides tips for taking photos and says, “Having the cows in a chute or head gate is preferred but not essential. Having a consistent background like a chute makes it easier to remove unnecessary aspects and process. Use the best lighting possible. Try to avoid using the flash as most cattle will be apprehensive of sudden bursts of light.

“Do you need a high-resolution camera? No. If you really want to take pictures with a high-resolution camera, by all means, go ahead, but cell phone quality is just fine, as most are 12 MP. Pictures where any of the cow’s facial characteristics are blurry are not ideal. Don’t get close enough for the cow to hit you or the camera/phone but do try to get the entire face in the photo. If you want to take several shots from different angles all around the face of the same animal that is perfectly acceptable, as well.

“Again, we will take all the photos that we can get. We don’t care if the cow is smiling, frowning, bellowing, ruminating, or pondering. Ears can be forward, back, to the side, or whirling all around. We need to characterize and train our program as much as possible.”

Ultimately, this gallery of images will help strengthen the programmers as they account for variables including lighting, hair growth and loss, age, flies, mud and other external factors.

So how will your photographs be used?

Jackson explains, “We are going to be using aspects and advances in machine learning to autonomously identify animals. Eighty percent of the cattle photos will be utilized for training and the remaining 20% will be used for validation. Most human-based facial recognition databases and programs would require anywhere from 10 to 10,000 photographs of subjects to properly train the algorithms.”

Is it worth your time? What’s the benefit for participating producers?

“The software developed could potentially be more applicable to you if it could actually recognize your cattle,” said Jackson. “For instance, I would like to envision a future where I could have my UAV go out and not only count but identify which animal is where.

“Assuming we can successfully identify animals in field, the next goal would be pinkeye detection and notification. That way, if there is an issue or if I needed to go check a specific animal, the drone would have found this animal for me (prior to it getting dark, of course).

“Also, we are trying to think of the future, as this could save many producers time and potentially money if disease events can be detected sooner. I want to make it easier to farm. As the adage goes, ‘We’ll never know until we try.’ So, with your help and #ukcowmap, we’re trying.”

In addition to facial recognition, UK researchers are exploring ways to automate recognition by ear tag numbers. Yet, this route has its challenges, as well.

“Ear tags can be covered by dirt, the distance can be too far, the tag can be stuck up in the ear or behind the fly tag, or the tag can be missing (5-20% of cattle ear tags are lost), to name a few of the factors,” he says. “We are also looking at different RFID systems or actively powered tags to help the UAV identify and locate cattle in the field. Thermal images can tell you a cow is there but not necessarily which cow.”

Again, if you feel so inclined, you can help the developers at UK perfect their facial recognition software by simply taking portraits of your cattle and posting them on social media using the hashtag #ukcowmap or emailing the images to [email protected].

I look forward to seeing how this software could be implemented on the ranch and in the feedyard in the years to come. This is a clear indication that new and emerging technologies will continue to enhance and improve the way we produce beef from pasture to plate. What a great and exciting time to be in the beef business!

The opinions of Amanda Radke are not necessarily those of beefmagazine.com or Farm Progress.

Subscribe to Our Newsletters
BEEF Magazine is the source for beef production, management and market news.

You May Also Like