AI-Based Cow Detection and Tracking to Monitor Real-Life Livestock Farming

The adoption of autonomous technologies in livestock farming has revolutionized traditional management practices, offering unprecedented opportunities for improved efficiency and animal welfare. Monitoring the movement and feeding of animals is essential for measuring the well-being of each cow automatically. This is a key step for self-operating farms. The flexible nature and unpredictable actions of cows in complicated farm settings make it difficult to follow each cow in a group. To address these difficulties, this paper introduces a new method, YOLOv7s-CA+DeepSORT-ViT, for cow tracking and feeding monitoring in autonomous livestock farming systems. The proposed approach integrates coordinate attention (CA) into YOLOv7 to enhance face detection performance and embeds a vision transformer (ViT) into the reidentification (reID) network DeepSORT to improve target recognition accuracy. Comparative evaluations against baseline methods, including SORT, BoTSORT, and ByteTrack demonstrate the superior performance of the proposed approach in multitarget tracking accuracy and reduction of ID switching times. Precision livestock farming is highlighted as a critical domain, emphasizing the significance of advanced tracking methods in addressing the challenges of complex farming environments. The findings of this study underscore the potential of the YOLOv7s-CA+DeepSORT- ViT approach to enhance cow tracking and feeding monitoring in autonomous livestock farming systems.