Transmission line bolts and their defects detection method based on position relationship

Frontiers in Energy Research

Energy Research

Siyu Miao¹  Yu Han¹  Jing Xiong²  Zhenbing Zhao³ 

[1] School of Electrical and Electronic Engineering, North China Electric Power University, Baoding, China;School of Electrical and Electronic Engineering, North China Electric Power University, Baoding, China;Department of Information Engineering, Sichuan Vocational and Technical College of Communications, Chengdu, China;School of Electrical and Electronic Engineering, North China Electric Power University, Baoding, China;Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, North China Electric Power University, Baoding, China;Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding, China;

关键词: transmission line bolts;    bolts defects;    target detection;    attention mechanism;    positional relationship;   

DOI  :  10.3389/fenrg.2023.1269087

 received in 2023-07-29, accepted in 2023-09-05,  发布年份 2023

来源: Frontiers

【 摘 要 】

Introduction: To solve the problems of small proportion of bolts in aerial images of power transmission lines, small differences between classes, and difficulty in extracting refined features, this paper proposes a method for detecting power transmission line bolts and their defects based on positional relationships.Methods: Firstly, a spatial attention module is added to Faster R-CNN, using two parallel cross attention to obtain cross path features and global features respectively, and spatial feature enhancement is performed on the features output from the convolution layer. Then, starting from the spatial position relationship of bolts and their defects, using the relative geometric features of candidate regions as input, the spatial position relationship of bolts and their defects on the image is modeled. Finally, the position features and regional features are connected to obtain enhanced features. The bolt position knowledge on the connecting plate is added to the detection model to improve the detection accuracy of the model.Results and discussion: The experimental results show that the mAP value of the algorithm in this paper is increased by 6.61% compared to the Faster R-CNN detection model in aerial photography of transmission line bolts and their defect datasets, with the AP value of normal bolts increased by 1.73%, the AP value of pin losing increased by 4.45%, and the AP value of nut losing increased by 13.63%.

【 授权许可】

Unknown   
Copyright © 2023 Zhao, Xiong, Han and Miao.