Deep Learning-Enabled Multi-Drone System for Automated Aircraft Visual Inspection

Reliable aircraft inspection is essential for maintaining structural integrity and operational safety, yet conventional visual inspection can be slow, labour-intensive, and sensitive to unfavourable lighting conditions. This work presents a cooperative two-UAV framework for automated aircraft surface inspection in which image acquisition and illumination are assigned to separate aerial platforms. The imaging UAV observes the inspection surface from an approximately normal direction, while the second UAV provides directional lighting from an adjustable oblique angle to improve crack visibility. A YOLOv8-based deep-learning detector is combined with conventional computer vision methods for surface-crack analysis. The proposed framework is evaluated through high-fidelity Gazebo simulations and laboratory experiments using a Parrot Anafi UAV. Directed illumination increased the number of extracted local features from 993 to 2,033, corresponding to an improvement of more than 100%. In addition, the crack-detection output improved by approximately 20–30% under suitable directional illumination. The results indicate that independently controlling illumination can enhance the visual representation of small surface defects and support more reliable automated aircraft inspection.