Vision-based Fault Detection Of Key Components Of High-speed EMU Contact Network

As the only source of locomotive power supply,the safe and stable operation of catenary plays a decisive role in the stable running of trains.As the key part of catenary,the geometric parameter of the locator has an important influence on the pulling value and height of the wire.Based on this,the fault detection technology of the key components of the high speed EMU catenary based on the vision is proposed to further improve the convenience and accuracy of the localizer slope detection.The traditional Hough transform and Sobel operator have been widely applied in this paper.As the pillar number is the location information of the slope fault of the catenary locator,the extraction of the pillar information is the basis of the number recognition,and the vertical information of the pillar is rich,so this paper creatively extracts the vertical feature of the image with the Sobel operator before the extraction of the Hough line information.At last,the LeNet-5 model is used to identify the number.In the recognition of locator,a locator recognition technology based on Faster R-CNN model is proposed.The addition of RPN network improves the recognition rate obviously.The RPN network solves the problem that the quality of the previous selected regions directly affects the accuracy of the target detection.Finally,the Hough transform is used to extract the profile of the positioner bracket and locate the locator roughly.Through the determination of the location of the locator and the selection of the fitting line,the slope of the locator is obtained by using the most left and the right end coordinates.