Research On Key Technologies Of Detection For Catenary Locator Of High-speed Train

The locator of the overhead catenary system(OCS)for high-speed railway holds the contact wire,which plays a key role in ensuring the pantograph has good current collecting quality.Locator is vulnerable to abnormal slope caused by excitation from pantograph-catenary coupling system,which threatens the safe running of train.Traditional locator detection methods include manual inspection and traditional image processing technology.The manual inspection method relies on the experience of the inspectors,and the labor intensity is too high.Using traditional image processing technology to detect locators,the accuracy and real-time performance of the detection are poor.In recent years,computer vision technology based on convolution neural network has been gradually applied in the field of locator detection,hoping to break through the bottleneck of traditional image processing technology and improve the accuracy and speed of locator detection.On the basis of previous locator detection algorithms,this paper designs a locator visual detection model based on cascade of initial location and precise location,and proposes a object detection model combining Feature Pyramid Networks(FPN)and Faster Regions with Convolutional Neural Network(Faster R-CNN).Firstly,the FPN combined Faster R-CNN detection model is used to locate the key areas of the locator.Secondly,the image processing technology is used to accurately detect the locator in the key areas.The overall locator detection system adopts the strategy from rough detection to precise detection.The specific work is as follows:In the process of initial positioning of key areas of OCS locators,the differences between the main object detection algorithms are compared in detail.According to the actual situation of locator detection,a detection model based on Faster R-CNN is proposed,which is optimized and improved by FPN.So that the locator detection model based on FPN combines Faster R-CNN can integrate more abstract and more semantics deep convolution features with higher resolution and more detailed information shallow convolution features.A multi-scale convolution feature with more abundant information is formed.Through the experimental analysis of the catenary images collected by the high-speed train Catenary Inspection vehicle,the locator detection model presented in this paper shows excellent performance in terms of detection accuracy and speed.After obtaining the key area of the locator based on FPN combined Faster R-CNN locator initial positioning detection model,image processing technology is used to complete the process of locator precise positioning.Firstly,image preprocessing operations such as gray level transformation and smoothing are performed on catenary images.Secondly,LSD(Line Segment Detector)is used to detect the key areas of the locator.Finally,according to the characteristics of the locator's structure,a filter mechanism is designed to screen and calculate the best fitting locator's line,so as to realize the accurate detection of the locator.