Single Object Detection-Drone (SOD-Drone) Dataset and Single-Channel Small Object Dataset

SOD-Drone is a hybrid dataset for single-object drone detection, comprising a total of 25,856 images. The dataset originates from two primary sources: first, 10,524 self-collected images selected from 200,000 drone frames captured during experiments, after undergoing deblurring and deduplication; second, 10,524 images extracted from three public datasets\u2014Real World, Det-Fly, and MIDGARD\u2014which contain 10,000, 3,178, and 2,154 images, respectively. The dataset integrates diverse perspectives, backgrounds, target poses, and size distributions (small targets: 42%, medium targets: 33%, large targets: 25%) to enhance sample diversity and improve model robustness and adaptability. The data were randomly split into training, validation, and test sets in a 6:2:2 ratio, making it suitable for training and evaluating small target detection algorithms for drones in complex scenarios.