A deep learning-based approach for feeding behavior recognition of weanling pigs

Area of Science:

• Animal Science
• Computer Science
• Agricultural Engineering

Background:

• Monitoring pig feeding behavior is vital for health and welfare.
• Early detection of feed refusal aids disease control and timely feed replenishment.
• Conventional methods for pig behavior analysis can be slow and less accurate.

Purpose of the Study:

• To develop a real-time deep learning technique for detecting and recognizing pigs in a feeding position.
• To combine pig detection and behavior classification into a single, faster process.
• To improve the accuracy of identifying feeding behaviors in weanling pigs.

Main Methods:

• A You-Only-Look-Once (YOLO) model was adapted with adaptive adjustments for different pig sizes.
• An angle optimization strategy was implemented to improve head detection in feeders.
• The model was trained and tested for detecting feeding and drinking behaviors using Average Precision (AP) at a 0.5 Intersection over Union (IoU) threshold.

Main Results:

• The proposed method achieved high AP for detecting feeding behavior (up to 96.56%) and drinking behavior (up to 89.16%).
• The combined detection and classification approach increased detection speed.
• The model demonstrated superior precision and recall compared to traditional methods.

Conclusions:

• The developed deep learning technique effectively detects pig feeding behavior in real-time.
• This method offers a significant improvement over conventional approaches for monitoring swine welfare and management.
• The study highlights the potential of AI in precision livestock farming.