Artificial intelligence (AI) has made its way into agriculture in various ways, providing new technologies to enhance production agriculture. At the University of Arkansas, researchers developed a tool, the CattleFever system, that uses AI and thermal and RGB color cameras to detect cattle body temperature.
Traditionally, cattle temperatures are taken rectally. With the CattleFever system, this can reduce labor required to track herd health. Temperature is a key symptom for many diseases, so this system allows for faster detection and treatment.
Research Background
The University of Arkansas is equipped with an Artificial Intelligence and Computer Vision Lab, directed by Ngan Le, associate professor in the department of electrical engineering and computer science. She explains one of her key research directions is precision agriculture with artificial intelligence and computer vision.
Previous projects have focused on poultry, but broader agriculture-related projects, including cattle welfare, are on the horizon.
Le says, “This motivation led me to initiate collaborations with colleagues in the department of animal science, including Dr. Kegley, Dr. Powell and Dr. Zhao to combine their expertise in cattle with our strengths in AI and computer vision.”
This project initiative was closely supported and funded by the University of Arkansas division of agriculture.
Platform Construction
To build CattleFever, researchers needed data. However, the existing data for cattle only provided overhead rather than thermal images. So, the group built their own dataset using thermal images of calves. Collaborating with the Savoy Research Complex at the university, calves were recorded with synchronized RGB cameras, technology that captures images with red, green and blue light, and thermal cameras.
Rectal temperatures were also recorded for a base in the dataset. Technical team members, Trong Thang Pham and Ethan Coffman, along with several undergraduate students developed a semi-automated annotation and data processing system. More than 600 recorded frames were used to train the system in what to look for. This data all served as a benchmark for the CattleFever system.
All images gathered were linked to thermal and RGB images. Landmarks in 13 different places, such as eyes, ears, muzzle and mouth, on the animal were established.
“These landmarks allow the system to localize individual facial regions, and the thermal camera then measures the temperatures in those regions,” Le says.
The eyes and nostrils read closest to the rectal temperatures, so these landmarks were established as focus areas for thermal image readings. A machine-learning approach was used to predict data results.
These technology trainings resulted in CattleFever being able to automatically detect animal temperature within 1 degree of the rectal reading. Le explains that as more data is collected in real-life environments, the more accurate the system will become.
Project Outlook
In these studies, all cattle were directly facing the thermal cameras.
“‘We probably need to take more photos of them in the real-world settings, such as running around, to capture their motion in the field,” Pham explains.
Teaching the cameras how to recognize and interpret a cow’s face in real-world environments is the next step. Le explains further features like environmental and audio sensors will be added to increase animal welfare monitoring accuracy and lead to more developments of indicators like common symptoms or early signs of illness. At this point, additional funding is being sought to continue more research on this project.
Eventually, the goal is for producers to have access to technology like this. This could look like a monitoring system of cameras set up that are synched to a mobile interface or app.
Le says, “While the current work represents an important first step, we are excited about continuing to develop technologies and expanding its capabilities to support the real-world agricultural applications.”
