Research On Intelligent Image Recognition Method For High-speed Railway Operation Safety

High speed railway is a complex giant system,and subsystem,facilities,equipment failure may endanger the safety of high-speed rail operation.The image and video detection and monitoring systems are widely used in the whole railway for the operation safety of high-speed railway,such as mobile equipment,infrastructure status,operation environment and so on.Although the existing detection and monitoring system and image data application level can meet the requirements of high-speed railway infrastructure equipment status and operation environment safety detection and monitoring,they are distributed in different regions of the road network,and are still in the local application stage of human-computer joint interpretation and verification of safety hazards and faults,which requires a lot of labor and time-consuming.Make full use of the existing information infrastructure resources in the main data center of China Railway Group and the railway data service platform,solve the problem of intelligent image recognition for high-speed rail operation safety,and realize the intelligent upgrading of high-speed railway operation safety image detection and monitoring system,which will greatly improve the work efficiency and reduce the work cost.In recent years,the convergence of the advantages of deep learning,edge computing,cloud computing and other information technologies has promoted the developmet of artificial intelligence from technology research to industry application.The combination of deep learning and other artificial intelligence technologies with high-speed railway operation safety image scene can realize operation state intelligent recognition of highspeed railway facilities and opration environment real-time detection.Finally,it can realize image-based hidden danger investigation,defect detection and fault diagnosis of high-speed rail operation safety.This paper will focus on the intelligent recognition method and application of high-speed railway facilities and equipment operation safety image and mainly got the following innovative achievements.(1)Semi automatic annotation method of high-speed railway operation safety image based on deep active semi-supervised learning.Aiming at the problem of the efficiency of image annotation in the process of intelligent recognition of massive high-speed railway operation safety images by using deep learning method,an automatic annotation method of high-speed railway operation safety images based on deep active semisuperided learning is proposed.Active learning and semi-supervised learning are introduced into the fine-tuning process of convolution neural nerwork in the way of continuous iterative paradigm,and the deep convolution neural network can learn the features of high-speed railway operation safety image through an incremental way,so that the fast annotation of high-speed railway operation safety image data can be realized.Combining the above method with image intelligent recognition of TEDS,aiming at the automatic annotation of EMU operation safety image data,the semantic annotation problem of EMU operation image is transformed into EMU structure subsystem classification and component object detection problem,and the annotation task is divided into two stages: coarse annotation and fine annotation.Then a two-level association method based on deep active semi-supervided learning is proposed for the hierarchical structure of EMU component classification,which quantifies the relationship between data annotation number and target task performance.With less annotating data,higher performance can be achieved in the target task.(2)Defect detection method of EMU operation safety image based on convolution neural network.Aiming at the low accuracy of automatic defect identification by using imag matching method in TEDS system,this paper proposes a defect detection and segmentation model of EMU operation safety image based on convolutioan neural network,which analyzes the characteristics of EMU operation safety image and the defects,optimizes the region-based object detection model.The deformable convolution(DCN)with changeable receptive field is used to adapt the diversity of defect shap and size.The online hard example mining(OHEM)is used to select hard samples that are input into the prediction nerwork again to balance the proportion of positive and negative samples.This method overcomes the proble of the imbalance of positive and negative samples under the complex background and the variety of defect shape and size.(3)Fault identification method of catenary suspension operation state based on lightweight convolution neural network.Aiming at the problem of small fastener object defect detection in high-speed railway catenary suspension operation monitoring image,this paper analyzes the characteristics of catenary suspension image and compares the difference between catenary suspension fastener defect detection and natural image object detection.The problem of the fastener defect detection is transformed into two processes of fastener detection and fine recognition of operation state.A method of fastener defect recognition based on two-level lightweight convolutional neural network is proposed.Firstly,a fastener detection model composed of lightweight feature extraction network,global attention module,mutually enhanced classifier and detector is designed to realize the efficient detection of target fastener instances.Sencondly,a lightweight multi label classification network is built to complete the fine recognition of fastener working state.(4)The general framework of high-speed railway operation safety image intelligent recognition platform.This paper analyzes the application status and characteristics of high-speed railway operation safety image detection and monitoring information system in the railway network,and proposes the overall architecture of high-speed railway operation safety image intelligent recognition platform,which build with the mode of “edge+cloud’’.On the basis of the unified railway big data service platform,an intelligent image recognition platform for high-speed railway operation safety is constructed,and the process of edge computing is designed in combination with the business scope.