VisioDECT: A robust dataset for aerial and scenario based multi-drone detection, identification, and neutralization

The rapid proliferation of unmanned aerial vehicles (UAVs) for logistics, surveillance, and civilian applications continues to pose significant challenges to airspace security, particularly through unauthorized or malicious deployments. Existing UAV datasets are limited in scope, often focusing on single-drone scenarios, synthetic imagery, or restricted environmental conditions, thereby constraining the development of robust counter-UAV systems. To bridge these gaps, we present vision-based drone detection dataset named as VisioDECT, a comprehensive and scenario-rich dataset for multi-drone detection, identification, and neutralization. The dataset comprises 20,924 annotated images and labels from six UAV models (Anafi-Extended, DJI FPV, DJI Phantom, EFT-E410S, Mavic Air 2, and Mavic 2 Enterprise), captured across three distinct scenarios (sunny, cloudy, and evening) at varying altitudes (30–100 m) and distances. Importantly, all UAVs included in this dataset are rotary-wing (multirotor) platforms, which dominate low-altitude airspace and are the most commonly encountered in real-world surveillance and counter-UAV scenarios. Data were collected over 20 months from more than 12 locations in South Korea, ensuring diversity in illumination, weather, and background complexity. Each sample is provided in three standard formats (.txt, .xml, .csv), with detailed metadata and quality-verified annotations for detection and classification tasks. Illustrative benchmark evaluations using state-of-the-art detection models (e.g., DRONET, YOLO variants) are included solely to validate the quality and practical usability of the dataset for real-time drone defense research. VisioDECT provides a standardized, reproducible, and scalable resource that enables benchmarking, model training, and evaluation for airspace surveillance, UAV traffic management, and national security applications.