Research On Intelligent Recognition Method Of Typical Defects Of Railway Catenary Based On UAV Image

Catenary is an important part of power supply system of electrified railway,which is a special railway infrastructure to provide continuous traction power for high-speed trains.During the long-term high-speed and frequent operation of the train,the continuous impact of the pantograph and the catenary may lead to damage,loosening,fracture and other diseases of the key components of the catenary support device,affect the technical status and power supply quality of the catenary,and even lead to train safety accidents.Therefore,it is of great significance to carry out high-efficiency and accurate detection and state monitoring for the catenary support equipment for railway maintenance and operation.However,the existing methods of high-speed railway catenary inspection have such problems as low frequency of inspection,poor inspection conditions at night due to maintenance skylight period,limited detection area and so on.Moreover,the existing defect detection systems are often low in automation and detection efficiency,and rely too much on artificial assistance for image defect identification.Therefore,in order to solve the above problems,this paper proposes UAV inspection method for catenary support equipment to improve the railway inspection means and enhance the efficiency and frequency of inspection work.Meanwhile,this paper proposes a method of extracting fasteners and defect discrimination of catenary support device based on computer vision deep learning technology to improve the intelligence of the existing detection system,and the main work is as follows:(1)The UAV inspection method is proposed for the detection of catenary equipment,including the inspection content,the design of inspection scheme,the design of inspection system and the security of patrol inspection,etc.It provides supplementary means for the existing methods of railway catenary inspection and makes up for the shortcomings of the existing methods.For UAV inspection images,an automatic defect location and recognition algorithm is proposed.Through three cascaded networks,the defect detection of catenary support device fastener is gradually realized from coarse inspection to fine inspection.(2)Aiming at the problem of complex background of UAV image detection target,a method of UAV image data enhancement is proposed.Through online data enhancement and mixed sample data enhancement,the adaptability of the algorithm for UAV complex shooting scene is improved.(3)Aiming at the problem of small proportion of pixel area in UAV image,an improved two-stage object detection algorithm is proposed.By adding the fusion path from the network bottom feature to the high-level feature,more detailed features of the target can be retained in the network feature extraction,so as to enhance the recognition ability of the algorithm for small targets;at the same time,the lightweight head network is used to improve the detection speed of the algorithm under the condition of ensuring the accuracy.(4)Because the fasteners of catenary support device are densely and obliquely arranged in the UAV image,the detection algorithm will be interfered,resulting in missing detection.This paper proposes a rotating target detection algorithm RPNet based on UAV image.By adding angle parameter into the regression loss function,the network loss function is improved to make the network fit the deflection direction of the target,and the fastener target is extracted by rotating rectangular regression,so as to improve the object detection accuracy in dense scenes.(5)In view of the fact that there is little defect data in fastener samples,this paper proposes to use siamese neural network to replace the traditional classification model.By comparing the similarity of fasteners in test samples,it can better solve the problem of small defect samples,weaken the category label,and improve the classification ability of small sample defects.