Transmission Line Component Inspection Method Based on Deep Learning under Visual Navigation

Abstract

The efficiency of the operation and maintenance for transmission lines can be greatly enhanced with the help of UAV. However, the images of the transmission line component collected by UAV could be spoiled due to the incorrect operations of the controllers, which causes the defects of the transmission line component couldn’t be diagnosed accurately. A visual inspection algorithm for the UAV based on the sensing of the tower components is proposed which uses a parallel network structure and can be applied on embedded devices, which based on Faster R-CNN target recognition framework and the ZFNet. Thus, the classification and posture of the component can be acquired. The UAV’s pose and position can be adjusted by the distance sensing model and components’ posture. Results show that the maximum mAP (mean Average Precision) of component recognition and posture perception are 0.95 and 0.91, respectively. Additionally, the posture of the UAV can be adjusted effectively and optimum images of the component can be obtained.