Researchers from Penn State University explore a groundbreaking approach that could improve turkey welfare while simultaneously reducing costs for producers with the help of drone technology.
The innovative study deployed a small drone, fitted with a camera and powered by advanced computer vision—an AI technology designed for interpreting visual data—to autonomously observe and identify the behaviours of turkeys. This research represents the inaugural investigation into the capability of drones, enhanced with computer vision, to discern distinct behaviours of turkeys captured from an aerial perspective.
Dataset of Behaviours
In the study, researchers collected footage of 160 young turkeys aged 5 to 32 days, recording their behaviours four times daily at the Penn State Poultry Education and Research Centre. The drone’s flight path was meticulously crafted to ensure comprehensive coverage of the area, optimizing the data collection process.
During the research, individual frames from the recorded footage were manually labelled, creating a rich dataset comprising over 19,000 instances of specific turkey behaviours, including:
- Feeding
- Drinking
- Sitting
- Standing
- Perching
- Huddling
- Wing flapping
Utilizing this dataset, the team trained and validated a computer vision model known as YOLO (You Only Look Once), a highly regarded object and action detection framework.
Detecting Specific Behaviours
After evaluating various versions of the YOLO model, researchers determined that the most effective model demonstrated an impressive 87% accuracy in identifying present behaviours and an outstanding 98% accuracy in detecting specific behaviours.
Enrico Casella, the senior author and assistant professor of data science for animal systems at Penn State, expressed optimism regarding the findings. “These metrics are quite promising, particularly for behaviour classification in real-world farm environments, which often present complex visual challenges,” stated Casella.
“This work establishes a proof of concept that integrating drones with AI could become a highly efficient, low-labour strategy for monitoring the welfare of turkeys in commercial settings,” he added. “It lays the foundation for more advanced and scalable monitoring systems in the future.”
The implications of this research are significant, providing a promising avenue for more effective welfare monitoring in commercial turkey production while easing labour requirements and reducing the need for constant human oversight.
For further insights, access their study published ahead of the December issue of Poultry Science.
This format enhances readability and is optimized for seamless integration into a WordPress site. The structured HTML allows for clear presentation of the content while ensuring that important details are highlighted.
