Casella, an assistant professor of data science for animal systems in the College of Agricultural Sciences and member of the Penn State Institute of Computational and Data Sciences, noted that this project marks the first documented attempt to assess drone-based behavioral tracking in commercial-style poultry settings.
“This work provides proof of concept that drones plus AI can potentially become an effective, low-labor method for monitoring turkey welfare in commercial production,” Casella said. “It lays the groundwork for more advanced, scalable systems in the future.”
During the project, the drone captured video four times per day while flying above a flock of 160 turkeys between five and 32 days old at the Penn State Poultry Education and Research Center. From this footage, researchers pulled individual frames and built a dataset of more than 19,000 images showing various actions, including feeding, drinking, standing, huddling, sitting, wing flapping, and perching.
The images were used to train a YOLO computer vision model. The strongest model identified 87% of behaviors present and classified actions with 98% accuracy.
As Casella explained, “The study shows that a drone equipped AI system can accurately detect turkey behaviors.” He added that this approach could reduce labor needs and provide continuous, noninvasive welfare monitoring in commercial operations.