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

Area of Science:

• Computer Vision
• Aerospace Engineering
• Artificial Intelligence

Background:

• Unmanned aerial vehicles (UAVs) present significant airspace security challenges due to their increasing use in various applications.
• Existing UAV datasets are insufficient for developing robust counter-UAV systems, often lacking diversity in scenarios, drone types, and environmental conditions.
• Rotary-wing UAVs are prevalent in low-altitude airspace and are critical targets for surveillance and counter-UAV operations.

Purpose of the Study:

• To introduce VisioDECT, a comprehensive, vision-based dataset for multi-drone detection, identification, and neutralization.
• To provide a standardized, scalable, and reproducible resource for training and evaluating counter-UAV systems.
• To facilitate advancements in airspace surveillance, UAV traffic management, and national security.

Main Methods:

• Collected 20,924 annotated images of six different rotary-wing UAV models across sunny, cloudy, and evening scenarios.
• Captured data over 20 months from 12+ locations in South Korea, varying altitudes (30-100m) and distances.
• Provided annotations in .txt, .xml, and .csv formats with detailed metadata and quality verification.

Main Results:

• The VisioDECT dataset offers rich diversity in illumination, weather, and background complexity.
• Benchmark evaluations using state-of-the-art models demonstrate the dataset's suitability for real-time drone defense research.
• The dataset supports detection and classification tasks for multiple UAV models.

Conclusions:

• VisioDECT addresses the limitations of existing datasets, enabling more robust counter-UAV system development.
• The dataset serves as a valuable resource for benchmarking, model training, and evaluation in critical security applications.
• VisioDECT contributes to improving airspace security, UAV traffic management, and national defense capabilities.