Deep Architecture for High-Speed Railway Insulator Surface Defect Detection: Denoising Autoencoder With Multitask Learning

Dongkai Zhang,Long Yu,Shibin Gao,Gaoqiang Kang
DOI: https://doi.org/10.1109/TIM.2018.2868490
IF: 5.6
2019-08-01
IEEE Transactions on Instrumentation and Measurement
Abstract:The insulator is an important catenary component that maintains the insulation between the catenary and earth. Due to the long-term impact of railway vehicles and the environment, defects in the insulator are inevitable. Recently, automatic catenary inspection using computer vision and pattern recognition has been introduced to improve the safety of railway operation. However, achieving full automation of insulator defect detection is still very challenging due to the visual complexity of defects and the small number of defective insulators. To overcome these problems, this paper proposes a novel insulator surface defect detection system using a deep convolutional neural network (CNN). The proposed system consists of two stages. First, a Faster R-CNN network is adopted to localize the key catenary components, and the image areas that contain the insulators are obtained. Then, the classification score and anomaly score are determined from a deep multitask neural network that is composed of a deep material classifier and a deep denoising autoencoder. The defect state is determined by analyzing the classification score and anomaly score. Experiments of the catenary insulator defect detection along the Hefei–Fuzhou high-speed railway line indicate that the system can achieve high detection accuracy.
Computer Science,Environmental Science,Engineering
What problem does this paper attempt to address?