UB-SOD dataset for small UAV and bird detection in airport low-altitude surveillance

<p dir="ltr">UB-SOD is a real-world image dataset developed for small UAV and bird detection in low-altitude airport surveillance environments.</p><p><br></p><p dir="ltr">The dataset contains 7,118 JPG images and 18,042 manually annotated object instances, including 10,047 UAV instances and 7,995 bird instances.</p><p><br></p><p dir="ltr">The images were collected from approximately 32 airport environments, covering diverse backgrounds and environmental conditions.

Bounding-box annotations are provided in YOLO TXT, Pascal VOC XML, and COCO JSON formats. Fixed training, validation, and test splits are included.</p><p><br></p><p dir="ltr">According to the COCO object-area definition, 13,288 annotated instances (73.65%) belong to the small-object category, 3,810 (21.12%) to the medium-object category, and 944 (5.23%) to the large-object category.</p><p><br></p><p dir="ltr">The dataset is intended to support research on small-object detection, UAV and bird discrimination, airport low-altitude surveillance, and related computer vision tasks.</p>