A Lightweight Model for Small-Target Pig Eye Detection in Automated Estrus Recognition.

Area of Science:

• Animal Science
• Computer Vision
• Artificial Intelligence

Background:

• Accurate sow estrus identification is critical for efficient large-scale pig farming and economic benefits.
• Short estrus duration and subjective human assessment hinder precise insemination timing.
• Automated, non-contact estrus detection methods are needed to overcome current limitations.

Purpose of the Study:

• To develop and validate an improved algorithm, Enhanced Context-Attention YOLO (ECA-YOLO), for automated sow estrus detection.
• To leverage ocular features (eye's spirit, color, shape, morphology) for accurate estrus stage classification.
• To enhance the model's performance in detecting and classifying sow estrus under complex farming conditions.

Main Methods:

• Proposed an Enhanced Context-Attention YOLO (ECA-YOLO) algorithm based on YOLOv11, incorporating MSCA, PPA, and GAM modules.
• Utilized ocular appearance features across different estrus stages as key indicators for classification.
• Implemented an Adaptive Threshold Focal Loss (ATFL) function to improve sensitivity to challenging samples.
• Trained and validated the model on a dataset of 4461 sow eye images during estrus.

Main Results:

• ECA-YOLO achieved a mean average precision (mAP) of 93.2% and an F1-score of 88.0%.
• The model demonstrated superior performance compared to benchmarks including YOLOv5n, YOLOv7tiny, YOLOv8n, YOLOv10n, YOLOv11n, and Faster R-CNN.
• Achieved 75.53 frames per second (FPS) with 5.31M parameters, indicating efficient real-time processing capabilities.

Conclusions:

• Ocular features are feasible indicators for automated sow estrus detection.
• ECA-YOLO demonstrates significant potential for real-time, accurate monitoring of sow estrus in intensive pig farming.
• The study provides a foundation for developing advanced automated systems to optimize breeding management in swine production.