Positioning And Abnormal Status Detection Of Catenary Dropper And Pipe Cap In High-speed Railway

During the rapid development and operation of high-speed railways,the safety inspection of high-speed railway infrastructures has always been the top priority of the traffic safety for electrified railways.In high-speed railway catenary support suspension devices,the droppers and pipe caps are important key components.Due to construction defects or vibration and impact during the long-term operation of the train,it will inevitably cause the catenary support suspension device to malfunction,which will have a huge impact on the stable operation of the train.Therefore,it is particularly important to find the faulty parts in time.However,the current fault detection mainly relies on manual and traditional image processing technology,the efficiency is not high and the detection accuracy is low,and the automatic detection of abnormal conditions of droppers and pipe caps still has difficulties.Therefore,intelligent identification algorithms for the positioning and fault detection of catenary droppers and pipe caps need to be researched to guide the disappearance of catenary safety hazards.The main research work are as follows:First,this article takes the droppers and pipe caps in the high-speed railway catenary support suspension as the research object,analyzes the characteristics of the catenary images collected by the C4 subsystem,and briefly preprocesses the catenary images: using image data cleaning to eliminate overexposed and pure black background images that are not related to catenary droppers and pipe caps detection;in order to remove the interference of environmental noise in the image,the RR-DCT is used to denoise,the random subsampling of Poisson-disk sampling realizes the trade-off between image noise reduction performance and computational efficiency;for catenary image quality problems under low light conditions,AINDANE algorithm enhances the brightness and contrast of low-light images through global brightness adjustment combined with local contrast adjustment.The pre-processed catenary images are improved in quality and can better meet the requirements of subsequent detection algorithm.Then,this paper analyzes deep learning target detection algorithms based on candidate regions.Aiming at the shortcomings of Faster R-CNN,an improved Faster R-CNN network localization model is proposed: extracting the features of the network by skip-layers,combining feature maps at all levels to obtain high-level semantics and low-level high-resolution location information at the same time,so that the efficiency of feature extraction is improved;at the same time,to improve recognition accuracy,a lightweight RPN is used to generate fewer region proposals;accelerate the generation process of candidate regions by adjusting the position of the Ro I pooling layer to improve the overall speed of the network.An image data set of catenary droppers and pipe caps was established,and the improved Faster R-CNN network was trained.The superiority of the generated catenary dropper and pipe cap positioning model was verified through actual railway lines tests.The foundation of dropper and cap abnormality detection is laid.Finally,this article addresses the problem of SSD networks that use shallow features to detect small targets.Due to the lack of semantic information,small targets are not well detected,and an improved SSD detection network is proposed.The classifier Residual-101 is combined with SSD,and an additional deconvolution layer is added at the same time to fuse the deep features of the image with the shallow information to improve the ability of shallow representation and improve the detector's detection ability for small targets such as droppers and pipe caps.On the fault detection of catenary droppers and pipe caps,this paper proposes a cascade network for abnormal state detection of droppers and pipe caps.The improved Faster R-CNN network and the improved SSD network are cascaded.The fault condition is judged on the positioning result of droppers and pipe caps.The experimental results show that the cascade network model composed of the improved network in this paper can quickly and accurately judge the various states of catenary droppers and pipe caps,and has strong robustness on different railway lines.