Laser & Optoelectronics Progress, Volume. 58, Issue 12, 1210022(2021)

Insulator Defect Recognition in Aerial Images Based on Gaussian YOLOv3

Quan Wang1 and Benshun Yi2、*
Author Affiliations
  • 1FiberHome Technologies Group, Wuhan, Hubei 430074, China
  • 2Wuhan University, Wuhan, Hubei 430072, China
  • show less

    The safety monitoring of insulators on transmission lines is particularly important because these insulators are exposed to strong electric fields and harsh environments. To quickly and accurately identify insulators in aerial images, an insulator detection algorithm based on Gaussian YOLOv3 (you only look once) is proposed in this work. First, in order to output the prediction box, the output of the network is increased and the loss function of the network is improved. Then, the mean and variance of the corresponding prediction box coordinate are output by combining the strategy of Gaussian distribution. Finally, the overfitting problem of small data sets is resolved via multi-stage transfer learning. Experimental results show that the algorithm can accurately determine the location of an object. Detection accuracy levels of 93.8% and 94.5%, which are better than the Faster regional convolutional neural network and YOLOv3 algorithm under the same conditions, are realized for insulators in the test set and insulator defects, respectively. The detection accuracy is important for power transmission. Moreover, the intelligent detection of line insulators yields a certain reference value.

    Tools

    Get Citation

    Copy Citation Text

    Quan Wang, Benshun Yi. Insulator Defect Recognition in Aerial Images Based on Gaussian YOLOv3[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210022

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Aug. 24, 2020

    Accepted: Nov. 4, 2020

    Published Online: Jun. 21, 2021

    The Author Email: Yi Benshun (yibs@whu.edu.cn)

    DOI:10.3788/LOP202158.1210022

    Topics