The Key Technology Of Intelligent Image Processing For Defects Detection Of High-Speed Railway Catenary

With the development of China's high-speed railway,the scale of the railway network has been continuously expanded,and higher requirements have been put forward for China's railway operation management.To ensure operational safety,it is urgent to realize modern and high-efficiency inspection of high-speed railway lines.The popularity of the high-speed railway contact network suspension state detection and monitoring device(4C device)improves the efficiency of image acquisition.However,for the collection of a large amount of image data,the current detection method is still manually checked,and the efficiency is low,and the failure is not processed in time.The problem.The image intelligent analysis technology based on the traditional image processing algorithm is often poorly universal,and it is difficult to achieve the performance instead of artificial.In the face of a large number of contact network acquisition images,further research is needed.With the development of GPU hardware and the support of big data,deep learning has developed rapidly,occupying an important position in the fields of computer vision and natural language processing.Due to its excellent feature expression,deep convolutional neural networks can achieve more accurate and robust performance than traditional image processing algorithms when data is sufficient.Therefore,this paper applies the deep learning algorithm to the high-speed rail contact network fault detection scenario to solve the problem of poor universality of traditional image processing algorithms.In this paper,the problem of two important components of the contact network,the suspension string and insulator fault detection is studied.In the chord fault detection,the deep learning target detection algorithm Faster R-CNN is used to distinguish the normal chord from the obvious chord.Based on the positioning result of the normal sling,the Hough transform-based line detection algorithm is used to judge the chord.Is there a slight slack failure?In the insulator fault detection,the deep learning target detection algorithm RRPN is used to locate the insulator with angle information.On the basis of the rotating insulator image,a fault identification method for the same insulator is obtained one by one,which avoids the acquisition.The problem of poor image consistency.The deep learning algorithms in the two fault detection problems can achieve high precision,robust and fast detection,improve the universality of the algorithm,and reduce the design difficulty of the subsequent algorithms.The experimental results show that both fault detection algorithms can achieve a higher recall rate.Although there is a certain degree of false alarm,the fault image output by the fault detection algorithm can be manually checked,and the fault image can be hardly missed.Under the premise,the number of images that need to be viewed by the manual is greatly reduced,and the efficiency of the 4C device is improved,which provides an idea for the intelligent analysis technology of the high-speed rail contact network image.