What if ranchers could check on the health of their cows just by looking at their faces? That futuristic idea is becoming a reality thanks to CattleFever, a new system developed at the University of Arkansas.
Created by the Institute for Artificial Intelligence and Computer Vision (AICV), the tool uses AI and a thermal camera to check a cow's body temperature. This is an early step towards an automated system that could change the way farmers care for their animals.
The project was led by Trong Thang Pham, a doctoral student at the University of Arkansas, under the direction of Ngan Le, associate professor of electrical engineering and computer science. Le's lab specializes in medical imaging, computer vision, and robotics.
Together, they set out to solve a long-standing problem. Today, a cow's body temperature is measured rectally, a process that is stressful for the animal and labor-intensive for the rancher. CattleFever provides a non-invasive alternative that improves animal welfare and helps detect disease before symptoms spread.
Listen to cow sounds with Edge AI
To train the system, the researchers needed data. Existing datasets for animals such as dogs, cats, horses, and sheep weren't enough, so CattleEyeView, the only dataset for cattle, included overhead RGB photos for tracking herds.
So the team built their own dataset at the Arkansas Agricultural Experiment Station's Savoy Research Facility. The researchers filmed thousands of calves with short videos and thermal cameras, while also checking their temperatures with rectal thermometers to get an accurate baseline.
The researchers then annotated 13 facial landmarks across hundreds of frames: eyes, ears, muzzle, and mouth. This research created CattleFace‑RGBT, a dataset that links visible features and thermal data. Landmark detection tools can now automatically identify calf faces and key features in both RGB and thermal images.
Can AI really estimate body temperature from a cow's face?
Through ablation studies, the team found that readings from the eyes and nostrils most closely matched readings from thermometers. Using these landmarks, the system focused thermal data from those regions.
Heat stress in cattle is a global risk due to climate change
The most accurate predictions came from random forest regression, a machine learning technique that averages results from many decision trees. Results: CattleFever was able to estimate the cow's body temperature to within 1 degree of the rectal thermometer reading.
So far, this system works best when the cow is directly facing the camera. The next challenge is to teach the AI to recognize when a cow is grazing, moving, or turning its head in its natural environment.
“We'll need to take more photos in real-world environments, such as running around, to capture movement in the field.” Pham explained.
The team publicly shared the CattleFace‑RGBT dataset and invited other researchers to build on their work.
“When you discover something new, share it with the world. That's the spirit.” Pham said.
CattleFever represents a leap forward into precision livestock farming, where AI and sensors enable ranchers to care for animals more efficiently and humanely. By reading subtle heat signs on cows' faces, ranchers can detect fevers early, prevent costly outbreaks, and improve the health of their herds.
One day, simply looking into a cow's face could tell more about a cow's health than just eye contact, allowing farmers to build stronger herds and farm smarter.
Reference magazines:
- Chung Tan Pham, Ethan Coffman, Beth Kegley, Jeremy G. Powell, Jiangchao Zhao, and Gan Li. CattleFever: A system that automatically estimates cow fever. smart farming technology. DOI: 10.1016/j.atech.2025.101434
