Integrated Convolution and Attention Enhancement-You Only Look Once: A Lightweight Model for False Estrus and Estrus...

Area of Science:

• Artificial Intelligence
• Animal Science
• Computer Vision

Background:

• Accurate estrus detection is vital for sow productivity and farm profitability in intensive pig farming.
• Short estrus duration and false estrus complicate timely insemination.
• Existing methods lack the precision needed for real-time monitoring in diverse conditions.

Purpose of the Study:

• To develop and validate an enhanced YOLOv8 model (ICAE-YOLO) for accurate vulvar detection to identify sow estrus stages.
• To improve the precision of estrus detection by classifying five states, including pseudo-estrus.
• To provide a robust solution for real-time, automated estrus monitoring in challenging farming environments.

Main Methods:

• An enhanced YOLOv8 model, ICAE-YOLO, was developed, integrating Convolution and Attention Fusion Module (CAFM), Dual Dynamic Token Mixing (DDTM), Dilation-wise Residual (DWR), and Focaler-Intersection over Union (Focaler-IoU).
• The model was trained and tested on a dataset of 6402 sow estrus images.
• Performance was benchmarked against YOLOv8n, YOLOv5n, YOLOv7tiny, YOLOv9t, YOLOv10n, YOLOv11n, and Faster R-CNN.

Main Results:

• ICAE-YOLO achieved a mean Average Precision (mAP) of 93.4% and an F1-Score of 92.0%.
• The model demonstrated superior recognition accuracy compared to all benchmarked models.
• ICAE-YOLO maintained a favorable balance between model size (4.97 M) and computational efficiency (8.0 GFLOPs).

Conclusions:

• The proposed ICAE-YOLO model offers highly accurate and real-time estrus detection in sows.
• This technology can significantly enhance automated estrus monitoring in intensive pig farming.
• The model's performance provides a strong foundation for improving sow reproductive efficiency and farm profitability.