MMFW-UAV dataset: multi-sensor and multi-view fixed-wing UAV dataset for air-to-air vision tasks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Fixed-wing unmanned aerial vehicles (UAVs) are increasingly utilized in various applications, necessitating robust detection and monitoring systems.
• Existing datasets often lack the multi-modal and multi-view diversity required for comprehensive UAV analysis.
• Accurate identification and tracking of fixed-wing UAVs are critical for airspace management and security.

Purpose of the Study:

• To introduce the MMFW-UAV dataset, the first one-to-one multi-modal image dataset specifically for fixed-wing UAVs.
• To provide a comprehensive resource covering diverse UAV appearances, flight conditions, and aerial perspectives.
• To facilitate research and development in deep learning-based object detection for fixed-wing UAVs.

Main Methods:

• The MMFW-UAV dataset comprises 147,417 images of fixed-wing UAVs captured using zoom, wide-angle, and thermal sensors.
• Images were acquired from multiple aerial perspectives, including top-down, horizontal, and bottom-up views.
• Rigorous quality control, including semi-automatic annotation, manual checks, and refinement, ensured high-quality annotations.

Main Results:

• The MMFW-UAV dataset was evaluated using mainstream deep learning object detection architectures.
• Experimental results demonstrate the dataset's efficacy for fixed-wing UAV identification, detection, and monitoring.
• The multi-modal and multi-view nature of the dataset proved effective in diverse detection scenarios.

Conclusions:

• The MMFW-UAV dataset represents a significant contribution to the field of UAV remote sensing and computer vision.
• It provides a valuable benchmark for developing and testing advanced UAV detection algorithms.
• The dataset is expected to accelerate research and applications leveraging fixed-wing UAV technology.