Multi-Target Behavior Dection Dataset for Simmental Beef Cattle

The construction and release of the Simmental beef cattle multi-objective behavior detection image dataset aims to solve the problem of insufficient high-quality and diversified data in current beef cattle behavior recognition research, and promote the development of intelligent breeding technology based on computer vision. The data was collected from real farm environments, covering different seasons, lighting, and indoor and outdoor scenes, with a total of 3218 images annotated with 10 types of behaviors including lying down, standing, walking, running, eating, foraging, drinking, excreting, fighting, and riding, totaling 11318 annotated instances. We have adopted a semi-automatic process of "automatic pre labeling+manual verification" to ensure the quality and efficiency of labeling, and provide labeling files in PASCAL VOC and YOLO formats. To validate the effectiveness of the dataset, we trained and evaluated behavior detection using the YOLOv8 model. The experimental results indicate that the model has an average accuracy on the validation set（ mAP@0.5 ）Reaching 96.2%, it demonstrates high recognition accuracy for most behaviors, verifying the practicality and reliability of the dataset. This dataset can provide basic data support for research related to beef cattle behavior recognition, health monitoring, and precision breeding management