AI tool estimates cattle temperature from a thermal photo: full analysis
A University of Arkansas team says it has built an AI-based system that can estimate a calf’s body temperature from a thermal photo alone, potentially offering a less invasive way to screen cattle for fever. The system, called CattleFever, was reported by the university this winter and published in Smart Agricultural Technology, where researchers described temperature estimates within about 1 degree of rectal thermometer readings. (arkansasresearch.uark.edu)
The project builds on a longstanding challenge in cattle health monitoring: rectal temperature remains a practical benchmark for fever detection, but it takes labor, restraint, and close animal handling. Infrared thermography has been studied for years as a lower-stress alternative, with earlier work showing promise in correlating surface temperatures with rectal temperature, especially around the face and eye region, while also warning that thermography alone may not yet be reliable enough to fully replace standard methods across all settings. (sciencedirect.com)
According to the University of Arkansas, the researchers collected roughly 4,600 paired RGB and thermal frames and manually annotated 600 of them with 13 facial landmarks, including the eyes, ears, muzzle, and mouth. Those annotations were used to train an automated landmarking system for the broader dataset, which the team released as CattleFace-RGBT. In model testing, the most informative regions for predicting body temperature were the eyes and nostrils, and a random forest regression model delivered the best performance. The authors describe the dataset as the first publicly available paired RGB-thermal cattle dataset linked to true temperature measurements. (arkansasresearch.uark.edu)
The university identified doctoral student Huy Pham as first author and Ngan Le, associate professor of electrical engineering and computer science, as senior author, with collaborators from animal science at Arkansas. In the university’s account, the team emphasized that the current results came from animals facing the camera in a pen, and that the next technical hurdle is handling more natural poses, motion, and varied field conditions. That limitation matters, because many promising livestock-monitoring tools perform well in controlled datasets before running into edge cases on commercial operations. (arkansasresearch.uark.edu)
Industry reaction appears to be more contextual than direct at this stage: the broader cattle sector has already shown interest in remote monitoring tools that reduce labor and increase continuous observation, including commercial sensor platforms for reproduction and health management. That makes the Arkansas work directionally aligned with where precision livestock technology is heading, even if this specific tool is still in the research phase. The clearest expert signal from the literature is cautious optimism: facial thermography can help identify physiologic changes, but validation against core temperature under real farm conditions remains the key hurdle. (beefmagazine.com)
Why it matters: For veterinarians, this research points to a future screening layer rather than a near-term diagnostic replacement. If the approach holds up in commercial environments, it could support earlier fever detection, triage of animals needing hands-on exams, and more frequent welfare checks with less stress for calves and less time spent on manual temperature collection. In large herds, even modest improvements in low-touch monitoring can change how quickly sick animals are flagged. But clinicians should read the findings as proof of concept: accuracy within about 1 degree is encouraging, yet decision-making around treatment, isolation, or outbreak response will still depend on robust validation, repeatability, and performance across breeds, ages, ambient conditions, and camera setups. (arkansasresearch.uark.edu)
What to watch: Watch for prospective field studies, broader external validation, and any move from an academic dataset toward a deployable on-farm workflow. The open release of CattleFace-RGBT could accelerate that process by letting other groups test models on the same benchmark, but the real milestone will be whether the system can maintain accuracy when cattle are moving, off-angle, and managed in everyday production environments. (arkansasresearch.uark.edu)