Research On Pantograph Recognition And Damage Detection Of Skateboard Contact Surface Based On Computer Vision

Electric locomotives are the main force of long-distance transportation.During driving,the locomotive obtains electric energy through the contact of the pantograph slide plate and the contact net.A good pantograph slide plate can ensure that the locomotive obtains good electric energy during operation.The pantograph skateboard also has a certain service life.The cracks and burns of the skateboard are the most common failures in driving.Once these failures are found,a new pantograph skateboard must be replaced in time to avoid serious driving accidents,reduce economic losses,and ensure Personnel safety.At present,the damage detection for the contact surface of the pantograph slide plate still relies on manpower.It requires workers to board the roof to detect whether the slide plate has cracks,burns,and other defects.There are certain subjective errors and potential safety hazards.Therefore,a new detection method that is fast and safe in combination with the actual site is needed.On the basis of investigating the actual situation,this article takes advantage of the feature that electric locomotives need to be returned to the maintenance plant for maintenance inspection,and image processing technology is used to detect the contact surface of the pantograph slide plate,and the algorithm is used to determine whether the slide plate has cracks and large-area burns.The inspection process does not affect the normal driving plan of the electric locomotive and meets the requirements of onsite inspection.In this paper,we use obliquely downwards to shoot down the bow video,apply machine learning theory and image processing technology to identify the pantograph and detect the skateboard area.The main work is as follows:(1)After the train returned to the preparation plant,the lowering bow video was shot diagonally downward on the preparation platform,and the video was decomposed by ffmpeg to obtain a large number of lowering bow images,including images containing the pantograph and images not containing the pantograph.(2)Aiming at the problem that the decomposed image is too dark or underexposed,the improved light correction MSR algorithm is used to enhance the image quality,which solves the problem that the image is too dark and the subsequent fault recognition of the skateboard is difficult.(3)The filter contains images of pantograph skateboards.An 8-layer convolutional neural network is designed for the pantograph skateboard.A large number of samples are used to train the network model,and the images containing the pantograph skateboard are identified and detected.A large number of image verifications show that the accuracy of pantograph recognition and screening using convolutional neural networks can reach 97.4%,which can accurately complete pantograph detection.(4)The edge detection process is performed on the image containing the pantograph skateboard,and different edge detection operators are used to compare the effects.Finally,the Canny operator is used to obtain the edge contours of the pantograph and the skateboard,paving the way for subsequent recognition and extraction of the skateboard area.(5)The image after edge detection uses Hough line detection to find the straight edge of the skateboard.According to the special long horizontal line of the skateboard in the image,the upper and lower edges of the skateboard are found through parameter settings.The area surrounded by the detected straight line is the skateboard area.The detected skateboard area has a certain tilt,and the perspective transformation can be used to correct the distortion of the detected skateboard,which avoids the error of subsequent damage detection of the contact surface of the skateboard.(6)Perform crack detection and burn detection on the contact surface area of the skateboard obtained after calibration.In view of the failure of the slide plate crack,the second-generation curve wave transform is used to identify the false cracks and joint interference,and the existing crack length can be calculated by morphological processing.According to the burn characteristics of the skateboard,the percentage of the area occupied by the self-color pixels is used to detect the burn,and according to the conclusion,it is determined whether to replace the new skateboard strip.