Sheep Behavior Analysis Dataset

To realize sheep behavior detection and accurately identify sheep behavioral actions (including drinking, resting, standing and eating), this paper constructs the Sheep Behavior Analysis Dataset.This study captures various behavioral scenes of sheep in a natural farm environment, collecting sheep behavior images under diverse conditions including normal lighting, low lighting, strong lighting, occlusion, blur, and nighttime.

To ensure the quality of the dataset, 1,646 high-resolution high-quality images were obtained after manual screening, and all images were manually annotated. To evaluate the quality and authenticity of this dataset as well as the performance of sheep behavior detection models, seven models were tested: Faster R–CNN, YOLOv5s, YOLOv8n, YOLOv11n, YOLOv12n, RT‑DETR, and YOLO‑CFED. Experimental results demonstrate that this dataset achieves good adaptability in sheep behavior recognition tasks, and can provide high-quality data resources for both research and practical applications in sheep behavior detection.