Dynamic Serpentine Convolution with Attention Mechanism Enhancement for Beef Cattle Behavior Recognition

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Beef cattle behavior recognition is vital for intelligent farming but faces accuracy challenges.
• Traditional methods struggle with complex behavioral identification and low precision.

Purpose of the Study:

• To develop and evaluate the YOLOv8n_BiF_DSC algorithm for non-intrusive beef cattle behavior recognition.
• To enhance feature extraction and context dependency capture for improved accuracy.

Main Methods:

• Collected video data of 45 beef cattle exhibiting nine behaviors using fixed and mobile cameras.
• Augmented data to create a dataset of 34,560 samples.
• Improved YOLOv8n with dynamic snake-like convolution and BiFormer attention mechanisms.

Main Results:

• YOLOv8n_BiF_DSC achieved 93.6% accuracy, outperforming original YOLOv8n by over 5%.
• Achieved 96.5% average precision at IoU 50 and 71.5% at IoU 50:95.
• Demonstrated high accuracy (98.9%) in recognizing specific behaviors like lying down.

Conclusions:

• The YOLOv8n_BiF_DSC algorithm offers superior feature extraction and data fusion capabilities.
• It shows high robustness and adaptability for intelligent beef cattle management.
• Provides a strong foundation for advanced livestock monitoring systems.