Research On Parameter Intelligent Monitoring Method And System Based On C2 Continuous Images Of Catenary

These years,China has a great development in the field of high-speed railways.How to ensure the safety and stable operation of high-speed railway trains has become an increasingly important research topic.With the rapid development of deep learning image algorithm research in computer vision,there are many excellent results have been obtained in many image processing tasks.Many research areas have introduced methods of deep learning.In the actual industrial application scenario of high-speed railway catenary detection,the commonly used detection methods currently rely on manual detection or traditional image algorithms.These methods have low efficiency,accuracy and poor robustness in practical application scenarios,and it is difficult to guarantee the safety and stable operation of the high-speed railway catenary system.Therefore,based on the deep learning image recognition algorithm,this paper has carried out related research on catenary non-contact security detection.First,in catenary C2 detection,there is a problem of mileage positioning for continuously acquiring images.This paper uses the Faster R-CNN framework as the basic algorithm to construct a two-layer recognition network model structure.It realizes the fast and continuous license plate recognition of the C2 image of the catenary.According to the specific characteristics of the recognition task,the corresponding optimal frame selection algorithm is designed.At the same time,an image processing algorithms and a continuous number correction algorithms are also designed to make the overall recognition algorithm meet the task requirements.Secondly,the geometric parameters of the wrist-arm structure are also an important detection norm in the safety detection of catenary.In order to achieve the task of detecting geometric parameters of the wrist-arm system based on monocular vision images,this paper proposes a deep learning-based catenary image segmentation and pose estimation algorithm for the wrist-arm system.According to the characteristics of catenary images,the Mask RCNN image instance segmentation network is improved by embedding deformable convolution kernels.And a new greedy algorithm is used in candidate box filtering.So that it can achieve better results in segmentation of catenary image components.Then,based on the mask results after image segmentation,a pose estimation algorithm is designed with the contact arm wrist skeleton.And then,the comparative experiments of a large number of algorithms have proved the superiority and applicability of the algorithm.Finally,in view of the fact that deep learning image algorithms have been widely used in the field of catenary 6C system detection,this paper proposes a software architecture design for a highly concurrent catenary integrated detection and integration service application platform.The design of the platform architecture considers a variety of practical scenarios in contact network detection.It adopts a design that separates front-end and back-end servers from computing servers.It uses Redis database to provide tasks and data buffering services,TF serving to provide underlying algorithm model services,and g RPC as a transmission protocol.It has realized multiple features such as multi-user remote high concurrent requests,multiple hardware platform deployments,distributed server deployment,and reasonable allocation of computing resources.The actual test runs verify the effectiveness and high concurrency of the platform.