Detection of Pig Movement and Aggression Using Deep Learning Approaches.

Area of Science:

• Animal Behavior
• Machine Learning in Agriculture
• Animal Science

Background:

• Manual observation of pig aggression and movement is time-consuming, labor-intensive, and subjective.
• Current methods are impractical for large farms and do not cover the entire pig growth cycle.
• Accurate behavioral data is crucial for understanding pig social hierarchies and for breeding selection.

Purpose of the Study:

• To develop an efficient and accurate deep learning-based method for detecting and recognizing pig motion and aggressive behaviors.
• To provide a reliable system for recording pig movement duration and aggressive acts for breeding selection and behavioral research.
• To overcome the limitations of manual observation in terms of time, labor, and subjectivity.

Main Methods:

• Collected video data from commercial farms with pigs in stable social groups.
• Developed a deep learning model using an improved EMA-YOLOv8 and a target tracking algorithm.
• Annotated specific aggressive behaviors including head-to-head tapping, head-to-body tapping, neck biting, body biting, and ear biting.

Main Results:

• The deep learning model achieved an average precision of 96.4% in detecting pig identity and fighting behaviors.
• Model detection results showed high correlation with manual recordings (R² of 0.9804 and 0.9856).
• The system accurately identified motion duration and aggressive behaviors under natural farm conditions.

Conclusions:

• The proposed method offers an accurate and effective solution for monitoring pig behavior.
• This technology provides reliable data for studying pig social hierarchies and selecting for health and aggression phenotypes.
• The AI-driven approach overcomes the limitations of traditional manual observation methods in pig welfare studies.