Automatic Detection Method of Dairy Cow Feeding Behaviour Based on YOLO Improved Model and Edge Computing

Area of Science:

• Animal Science
• Computer Science
• Agricultural Engineering

Background:

• Cow feeding behavior is a key indicator of dairy animal health.
• Accurate and rapid assessment of feeding behavior is crucial for dairy farm management.
• Existing methods struggle with accuracy and environmental sensitivity in open farm settings.

Purpose of the Study:

• To develop an advanced method for monitoring dairy cow feeding behavior.
• To address limitations of current detection algorithms in complex farm environments.
• To enhance precision breeding and intelligent animal husbandry through improved behavior analysis.

Main Methods:

• Utilized edge computing for real-time image processing of cow feeding behavior.
• Developed a novel deep learning model, DenseResNet-You Only Look Once (DRN-YOLO).
• Enhanced the YOLOv4 algorithm by integrating a self-designed DRNet backbone and Spatial Pyramid Pooling (SPP) for improved feature extraction.

Main Results:

• The DRN-YOLO model demonstrated improved accuracy (1.70%), recall (1.82%), and mAP (0.97%) compared to YOLOv4.
• Successfully recognized cow feeding behavior in a complex farm environment.
• Overcame challenges of low recognition accuracy and insufficient feature extraction.

Conclusions:

• The DRN-YOLO method effectively analyzes dairy cow feeding behavior in complex environments.
• This approach offers a valuable reference for intelligent animal husbandry and precision breeding.
• The study highlights the potential of edge computing and deep learning in livestock monitoring.