Defect Recognition Of Catenary Dropper Based On Image Processing

Catenary is a special power supply system for electrified railway,whose stability of power supplies directly affects the running state of the train.The dropper is an important connecting part between the catenary cable and the contact line in the catenary suspension catenary,which plays the role of adjusting the height and sag of the contact line and bearing a certain capacity.Its state directly determines the stability of the catenary suspension system.Once the dropper fails,it may cause the height of contact line to decline,affect the quality of train current collection,and even cause pantograph catenary accident,which endangers the safety of train operation.But at present,the research on the identification and detection of catenary parts defects mainly focuses on the area of supporting positioning device,and the research on the detection of dropper defects is less,and there are still many problems of practical application.The detection of dropper defects in railway departments is still based on manual inspection and manual image frame by frame discrimination,which has low detection efficiency and high cost of human resources.In view of these problems,this thesis studies the automatic detection algorithm of catenary dropper defects.In this thesis,the catenary parts images collected by the catenary suspension state detection and monitoring device(4C device)in the actual railway line are taken as the research object,and the research status of catenary parts defect state detection at home and abroad is analyzed,as well as the development history of deep learning technology.According to the difference between the image features of different OCS parts and dropper images collected by 4C device,a dropper recognition and defect state detection method based on the combination of image processing and deep learning technology is proposed.According to the different states of the catenary dropper,corresponding algorithms is designed to identify and detect the dropper.And through the actual acquisition of catenary dropper image data testing,the feasibility of this method is verified.In the initial recognition and initial state detection of dropper,aiming at the problem that there are few broken and bent samples of dropper,this thesis proposes a more advanced dense convolutional network(densenet),which can extract deeper image features and realize feature reuses,to replace the feature extraction network in the original faster r-cnn model.The effectiveness of the proposed method is verified by comparing the detection results of the original faster r-cnn model and the faster r-cnn model based on resnet101 feature extraction network.The proposed method can quickly locate the catenary dropper area and detect the dropper image with obvious defects.In order to further distinguishes the dropper image and the suspected dropper device image.According to the analysis of the characteristics of the fixed parts of both end of the dropper and the suspected fixed parts at both ends of the dropper device,this thesis proposes a method to distinguish the dropper clamp and nylon sheath by using the faster r-cnn model again on the basis of the preliminary identification and positioning.In order to further improve the accuracy of detection of dropper clamp and nylon sheath,the size of anchor frame in two-stage faster r-cnn model is adjusted according to the statistics of the size of suspension clamp and nylon sheath marked area in the marked image.Through the detection of the actual image,the proposed method can effectively distinguish the dropper from the suspected dropper,and the model after adjusting the anchor frame can more accurately detect the dropper clamp and nylon sheath.In the detection of broken strands and micro deformation defects of dropper wire,in order to accurately detect the micro defects of dropper wire.Firstly,the method of line detection,image rotation and projection is used to locate the dropper wire accurately;then,the method of calculating the standard deviation of the coordinates of the center line of the dropper wires and the sum of the gray values of the local column of the dropper wire area is used to detect the micro deformation and broken strand defects of the dropper wire.Finally,the feasibility and reliability of the proposed method is verified by the actual image test.