The drone that sees like a bird: AI system sets a new standard for turkey monitoring

Researchers extracted single frames from the videos and manually tagged each turkey’s behavior, including feeding, drinking, sitting, standing, perching, huddling, and flapping its wings. (Image: Penn State)

As Americans prepare for holiday meals, researchers at Penn State turned their attention to turkeys in a different way. Their new work raises one clear question for producers: can drones help monitor flocks more quickly and safely on large farms?

They used a small drone fitted with a colour camera. They recorded videos of young turkeys four times each day. The study ran from five to thirty-two days of age. The flights covered the full enclosure area each time. The footage was taken at the Penn State Poultry Education and Research Centre. The team then extracted images from the videos for analysis. The study appears online ahead of the December Poultry Science issue.

Researchers labelled more than nineteen thousand behaviour examples. These included feeding, drinking, sitting and standing actions. They also marked perching, huddling and wing flapping behaviours. The labelled data trained a computer vision model. The team used YOLO, a widely used AI system. The model detects objects and actions within images quickly.

(Image: Penn State)

The strongest version detected 87% of behaviours. It identified specific actions with 98% accuracy. Researchers said the results were strong for farm use. They noted farms often present visual clutter naturally. They added the system worked effectively despite these conditions.

Monitoring turkey welfare takes large labour resources today. Farm workers must observe flocks closely and regularly. The drone system offers a low-labour alternative now. It allows continuous monitoring without direct handling. It could also reduce staffing demands on large farms. Researchers said it supports better welfare decisions overall.

Senior author Enrico Casella guided the project. He works in data science for animal systems. He is based in Penn State’s College of Agricultural Sciences. He also works with the Institute of Computational and Data Sciences. The study’s first author was Giulio Calderone in Palermo. Several Penn State researchers also contributed.

The U.S. Department of Agriculture provided main funding. Support also came from Penn State computational science teams. The results now set the stage for follow-up projects. Researchers said future systems may become more scalable soon.