Research On Digital Image Processing And Recognition Of Pantograph And Catenary Operation State Of High-Speed EMU

Railway transportation is one of the important modes of transportation in our country.The national economy of our country puts forward higher requirements for railway transportation.China’s high-speed EMU obtains the electric energy in the power grid through the catenary pantograph,and then drives the train and ensures the operation of control or auxiliary equipment.Therefore,catenary pantograph is the key to ensure the safe and stable operation of the whole EMU system.The highest speed of high-speed EMUs in China can reach 350 km / h,so pantograph is affected by high-speed friction,strong wind resistance,strong impact and highfrequency vibration.In this case,the pantograph is prone to structural damage or unreasonable position and other faults.At present,China’s high-speed EMU is equipped with on-board catenary operation status monitoring device(3C)for pantograph and catenary operation monitoring.Through the monitoring video recorded by the device,the video and data status under visible light and infrared light at the time of pantograph failure can be understood afterwards.However,the current purpose of the device is to monitor and review the records afterwards,and the identification method is to rely on the visual inspection of maintenance workers,so it is impossible to record the suspicious abnormal state and predict the occurrence of faults.Therefore,the automatic identification of the monitoring video of the device can not only greatly improve the accuracy of fault identification,but also can be used as a reference for the maintenance of high-speed EMU.It can save labor cost,ensure the safe and stable operation of EMU and improve the efficiency of maintenance.This paper studies the application of image processing and recognition technology in the image acquired by 3C.During the normal operation of EMU train,the computer system can process and recognize the real-time image obtained by the pantograph monitor,and get the realtime status of the pantograph itself and its contact relationship with the contact network.In case of pantograph or pantograph network relationship fault,it can give an alarm in real time so that the driver and maintenance personnel can take measures to deal with it in time to prevent the accident from expanding.The research contents of this paper are as follows:I.Research on the relationship between pantograph and catenary of EMU:(1)Analyze several types of pantograph and catenary faults: automatic pantograph dropping,pantograph bouncing,contact point arcing,structural damage,etc.,and establish expert database or neural network discriminant function for fault discrimination.(2)Understand the characteristics of the network fault,extract the characteristics of the specific monitoring location,image recognition and fault identification.(3)The faults of pantograph and catenary are classified.After classification,relative position and relative angle are used to identify the corresponding faults.II.Research on image processing: Based on MATLAB language,the image can be processed as follows:(1)Image enhancement.It includes reducing the light part and brightening the dark part.(2)Image noise reduction.Gaussian noise and salt and pepper noise in the image can be filtered by means of mean filtering,sequence statistical filtering or adaptive filtering.(3)Image restoration.The degraded image caused by defocusing blur can be obtained by inverse filtering or Wiener filtering;the degraded image caused by motion blur can be obtained by Wiener filtering and its derivative algorithm or blind deconvolution.III.Research on image recognition:(1)Feature extraction.Through radon,we can extract the geometric features in the image,and through the transformation of the geometric features,we can identify the important parts of the pantograph in reality.(2)Feature analysis.Through the analysis of the data characteristics of the monitored parts obtained in the previous step,it can be found that the data characteristics of the monitored parts conform to the normal distribution no matter the relative position or the structure size.Because it is known that the normal distribution itself conforms to the "3σ criterion",the abnormal characteristics outside the distribution range can be considered as abnormal characteristics,and alarm records are required.(3)Artificial neural network technology.In addition to analyzing the relationship between features for image recognition,we can also use CNN convolution neural network for image recognition.Convolutional neural network is used for direct deep learning of the digitized image.By giving normal and abnormal images,CNN can continuously adjust the weight of each pixel in the learning process,and finally get a complete training function;through the training function,new images to be tested can be detected.IV.Research on program design: summarize and summarize the above methods,and use the GUI page design of MATLAB itself to get a complete program that conforms to the technical route of this paper.The final functions of the program include image processing,image recognition,image data transformation,output of training function and detection results.Through the research on the processing and recognition of the pantograph catenary monitoring image obtained by 3C device,this paper basically realizes the recognition function of the collected image,which can carry out real-time image-based state monitoring,and can realize real-time monitoring of pantograph catenary contact state and suspicious fault alarm record during the operation of EMU.