Real-Time On-Board Deep Learning Fault Detection for Autonomous UAV Inspections

Inspection of high-voltage power lines using unmanned aerial vehicles is an emerging technological alternative to traditional methods. In the Drones4Energy project, we work toward building autonomous vision-based beyond-visual-line-of-sight (BVLOS) line inspection system. this paper, present a deep learning-based vision system detect faults in components. We trained YOLOv4-tiny architecture-based neural network, as it showed prominent results for detecting components with high accuracy. For running such learning models real-time environment, different single-board devices Raspberry Pi 4, Nvidia Jetson Nano, TX2, and AGX Xavier were used experimental evaluation. Our demonstrated that proposed approach can be effective efficient fully automatic on-board visual inspection.