An Annotated Aerial Dataset for Training and Benchmarking Computer Vision Models in Poultry Behavior Analysis

We present a dataset of annotated aerial images of turkey behaviors acquired using an Unmanned Aerial Vehicle (UAV) in an experimental poultry facility. Video recordings were collected from 160 Nicholas Select turkey toms between 5 and 32 days of age and converted into image frames. The dataset comprises 2, 388 images with bounding box annotations for eight behavioral categories: feeding, drinking, sitting, standing, perching, huddling, wing flapping, and dead. Each image contains multiple individuals captured from an oblique aerial perspective, reflecting realistic flock conditions. The images are provided with object detection Annotation encoded in YOLO format. This dataset supports the development, training, and benchmarking of computer vision models for automated poultry behavior analysis. It provides a resource for research in Precision Livestock Farming (PLF), particularly for non-invasive monitoring of animal behavior using aerial imagery.