Typical Defect Detection Technology of Transmission Line Based on Deep Learning

• Published in: 2019 Chinese Automation Congress (CAC) pp. 1185 - 1189
• Main Authors: Wanguo, Wang, Zhenli, Wang, Bin, Liu, Yuechen, Yang, Xiaobin, Sun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2019
• Subjects: Convolution; deep learning; Feature extraction; Insulators; Machine learning; Mathematical model; object detection; Power transmission lines; Training; transmission line
• ISSN: 2688-0938
• DOI: 10.1109/CAC48633.2019.8996643

Detection of line component defects in UAV transmission line inspection is always a difficult problem. In order to solve the problem of identifying the defects of insulators and anti-vibration hammers in transmission lines, a target detection technique based on deep learning is proposed to diagnose the typical defects of transmission lines. SSD algorithm based on candidate regions is selected to locate and identify the defects. Firstly, the influence of different feature extraction networks and network parameters on the accuracy and speed of target detection is studied. The network is improved by adjusting network parameters and model optimization method, and the network parameters that are most conducive to transmission line defect detection are selected.Then, the influence of different data enhancement methods on the accuracy of defect detection is studied. The training samples are expanded by multi-scale training and horizontal mirror method to further improve the accuracy of target detection.Finally, the recognition and classification experiments are carried out using the actual images collected by UAV. The experimental results show that the target detection method based on deep learning can accurately locate the location of defects from the transmission line image, and can be applied to the fault diagnosis task in the transmission line scene.