Real-Time Detection of Cattle Behavior Using YOLOv9 with AutoAugment and Mixup Data Augmentation Techniques

As the number of livestock farms in Korea decreases, the scale of individual farms has grown, necessitating advanced methods for efficient livestock management. Recent studies have increasingly applied computer vision and deep learning to monitor livestock behavior, health, and disease, aiming to enhance productivity and animal welfare. This study introduces a real-time cattle behavior detection system utilizing YOLOv9, a state-of-the-art object detection algorithm known for its high accuracy and efficiency. The YOLOv9-s model, optimized for real-time detection with minimal computational overhead, was employed. To further enhance detection accuracy, we incorporated AutoAugment, a data augmentation technique that automatically selects the optimal augmentation policies, and mixup, a method that improves model generalization by creating new training samples through linear interpolation of existing ones. The application of AutoAugment improved the mean Average Precision (mAP) from 0.926 to 0.941, while the addition of mixup raised the mAP to 0.947, representing a 2.1% performance increase. These results confirm the system's capability for accurate and efficient real-time detection of cattle behavior.