Fault Detection Of Railway Catenary Design And Implementation Of Intelligent Image Processing System

With the rapid development of Chinas infrastructure,the scale of high-speed railway,which is now an important means of public transport for national travel,has gradually expanded,which also puts forward higher requirements for the operation and maintenance of high-speed railway.Among them,the maintenance of high-speed railway catenary has adopted the catenary suspension monitoring device to collect the first-line images.However,due to the manual inspection of the collected images,the whole inspection process has the problems of low efficiency,tedious and easy to find errors.In the past period of research,people have begun to use the target detection technology based on deep learning to intelligently screen the catenary components,but the implementation of the algorithm is difficult.The main pain points are: 1)The detection model based on deep learning has high requirements for the number of data samples,the number of catenary images is huge,the actual marking work is time-consuming and laborious,and the model production cycle is long;2)At present,there are many kinds of detection algorithms for various parts of the catenary,which are difficult to integrate uniformly,and have little practical significance.In view of the shortcomings of existing methods for detecting catenary defects,this thesis proposes a catenary defect detection and enhancement process based on active learning and deep learning.The main work is as follows: 1)Take the dropper and wire clamp structure in the catenary as the detection target,and use the end-to-end deep target detection algorithm instead of manual screening,which greatly improves the overall detection efficiency.In addition,This thesis studies the practicability of the existing major target detection algorithm structures for this scene,and tries to ensure that the model maintains certain detection accuracy and scalability at the same time.2)Considering the extremely large number of catenary images taken,the difficulty of obtaining defect features efficiently and effectively with random one-time sampling and labeling,and the difficulty of integrating detection algorithms,this thesis based on the idea of active learning,implements an algorithm enhancement process with the end-to-end detection model as the core,and realizes the phased sampling and labeling of catenary images,This enables the model to better learn the characteristics of catenary defects,thus achieving higher performance,reducing the workload of manual labeling,giving space for model expansion,and facilitating the final integration and landing.3)Based on the above contents,this thesis designs and implements an intelligent catenary detection system based on the SSM framework,which implements the basic user management function,and can automatically filter the defects of the collected catenary images,classify and manage the collected data,and the staff can complete further work on the classified historical data.In addition,The labeling module provided by the system is convenient for the staff to label the active learning samples.The test results show that the algorithm of the intelligent catenary detection system designed and implemented in this thesis can reliably filter the collected catenary images,and its active learning strategy can periodically improve the model detection performance.The system itself meets the design requirements,and has certain application value.