Efficient Cow Detection With CTri ‐ YOLOv11 : From Model Enhancement to Edge Deployment

ABSTRACT Livestock detection is crucial for precision farming but is challenged by the diversity in animal appearance and posture. This study introduces a novel cow detection algorithm that combines a feature reorganization upsampling module with three attention mechanisms. The proposed method enhances fine‐grained spatial feature extraction and employs a multi‐attention fusion strategy to boost the representational power of key features, improving the performance of the YOLOv11 architecture. Evaluated on a self‐built dairy farm dataset, the model achieves a mean average precision (mAP) of 98.4% while sustaining a real‐time speed of 310.3 FPS, significantly outperforming baseline YOLOv11 and other mainstream detectors. Real‐world deployment further validates its effectiveness in supporting automated herd monitoring systems, offering a reliable and efficient solution for enhancing animal welfare and farm management.