Automatically Detected Pecking Activity in Group-Housed Turkeys.

Area of Science:

• Animal behavior
• Agricultural technology
• Machine learning in animal science

Background:

• Cannibalism in poultry is a significant welfare and economic concern.
• Early detection systems are needed to mitigate cannibalistic behaviors.
• Pecking activity is a key indicator of poultry welfare and social interactions.

Purpose of the Study:

• To validate a novel automatic pecking activity detection system for turkeys.
• To investigate pecking activity patterns during the rearing phase.
• To assess the system's potential as an early warning for cannibalistic outbreaks.

Main Methods:

• A pecking detection system using a microphone-equipped metallic object and video monitoring was developed.
• Data were collected from female (intact beaks) and male (trimmed beaks) turkeys on two farms.
• A Convolutional Neural Network (CNN) model was employed for automatic peck detection and validated.

Main Results:

• The CNN model achieved high detection performance (sensitivity, specificity, accuracy, precision, F1-score ~90%).
• Pecking activity varied between farms and was highest in the morning hours.
• Daily pecking frequency showed a slight downward trend, with an increase in week 5, but no clear association with cannibalistic outbreaks was found.

Conclusions:

• The automatic pecking detection system is a viable tool for research in poultry behavior.
• Further automation and testing under diverse farm conditions are recommended.
• The system's utility as a direct early warning for cannibalism requires further investigation.