中国电力 ›› 2021, Vol. 54 ›› Issue (2): 156-163,196.doi: 10.11930/j.issn.1004-9649.202006315

• 电网 • 上一篇    

基于EfficientDet和双目摄像头的绝缘子缺陷检测

刘逸凡1, 王淑青1, 庆毅辉1, 王晨曦1, 兰天泽1, 要若天2   

  1. 1. 湖北工业大学 电气与电子工程学院,湖北 武汉 430068;
    2. 武汉大学 电气与自动化学院,湖北 武汉 430072
  • 收稿日期:2020-07-07 修回日期:2020-08-12 发布日期:2021-02-06
  • 作者简介:刘逸凡(1996-),男,通信作者,硕士研究生,从事电气工程、绝缘子检测研究,E-mail:embalming@foxmail.com;王淑青(1969-),女,博士,教授,从事计算机控制、人工智能研究,E-mail:254831618@qq.com
  • 基金资助:
    国家自然科学基金资助项目(基于多自主体的复杂智能电网随机博弈与优化研究,51407063)

Insulator Defect Detection Based on EfficientDet and Binocular Camera

LIU Yifan1, WANG Shuqing1, QING Yihui1, WANG Chenxi1, LAN Tianze1, YAO Ruotian2   

  1. 1. School of Electrical and Electronic Engineering, Hubei University of Technology, Wuhan 430068, China;
    2. School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China
  • Received:2020-07-07 Revised:2020-08-12 Published:2021-02-06
  • Supported by:
    This work is supported by National Natural Science Foundation of China (Research on Stochastic Game and Optimization of Complex Smart Grid Based on Multi Agent, No.51407063)

摘要: 绝缘子是输电线路重要的组成部件,带有缺陷的绝缘子会对线路造成隐患,通过图像检测技术可以提高绝缘子缺陷检测的效率,大大减小维护成本。但现有的绝缘子缺陷检测技术有精度不高和检测时间过长等缺点,针对这一问题,提出了基于EfficientDet和双目摄像头的绝缘子缺陷检测方法,首先通过双目摄像头设计了一种数据集采集的方法,解决了目前开源数据集不充分的问题;其次通过前置一个带标签的分类先行算法解决了EfficientDet占用资源过多的问题;最后将提出的算法与3种传统绝缘子缺陷检测算法进行比较,发现提出算法的mAP(平均精度均值)为50.04,优于其他3种算法,绝缘子识别定位准确率达95%以上,缺陷识别定位准确率达90%以上,具有较好的效率与实用性。

关键词: 神经网络, 双目摄像头, 绝缘子, 缺陷检测, 机器视觉

Abstract: Insulator is an important component of transmission lines. The defective insulator will cause hidden dangers to the lines. Image detection technology can improve the efficiency of insulator defect detection and greatly reduce the maintenance cost. However, the existing insulator defect detection technology has the disadvantages of low accuracy and long detection time. Aiming at this problem, an insulator defect detection method is proposed based on EfficientDet and binocular camera. Firstly, a data collection method is designed for binocular camera to solve the problem of insufficient open source data set; Secondly, the problem of excessive resources occupied by EfficientDet is solved by a labeled classification first algorithm; Finally, the proposed algorithm is compared with three conventional algorithms. It is found that the mAP of the proposed algorithm is 50.04, which is superior to other three algorithms. The accuracy rate of insulator and defect identification and positioning is more than 95% and 90%, respectively, which shows the proposed algorithm’s good efficiency and practicability.

Key words: neural network, binocular camera, insulator, defect detection, machine vision