Monitoring of Composite Insulators in Transmission Lines: A Hydrophobicity Diagnostic Method Using Aerial Images and Residual Neural Networks

Abstract

Hydrophobicity detection is of great significance to the condition monitoring of composite insulators, is an important guarantee for the safe operation of transmission lines. Therefore, this paper proposes a hydrophobicity diagnosis method based on an improved residual neural network. The method consists of an improved residual neural network, an image quality booster, and an image enhancement module. Among them, the improved residual network reduces the convergence time of the model by adding the Focus algorithm. The image quality booster uses a deconvolution algorithm to effectively solve the image distortion problem. At the same time, the image enhancement module improves the generalization ability of the model through the GEM algorithm. On the other hand, the method locates the diagnostic area of composite insulators in a weakly supervised manner and is able to evaluate the hydrophobicity level of insulators at different locations simultaneously. Field tests show that this method can well meet the actual maintenance needs and eliminate the judgment difference of manual detection. Compared with current hydrophobic detection models, the method has greater advantages in terms of detection accuracy and operating speed.