Research And Application Of Intelligent Image Recognition Method For Railway Operation Safety

With the increase of railway operating mileage,the continuous enhancement of our country’s railway transportation strength,in order to ensure the smooth operation of the train,the safety of railway operation is more important.China Energy ShuozhouHuanghua Railway Company is responsible for the operation and management of Shuozhou-Huanghua Railway.In the process of production and operation,the station,section,crossing and other operating places not only have the management requirements for people,machines,objects and other access behaviors,intrusiveness,violations,etc.,but also have the detection,analysis and alarm requirements for the intrusion and interference of operating facilities and the surrounding environment.At present,railway operation and management are mainly monitored and supervised by manually viewing monitoring videos.Monitoring and operation and maintenance methods can no longer meet the requirements of new tasks.Based on the video surveillance network of the project,combined with artificial intelligence,edge computing and other technologies,through the intelligent analysis of the data collected by the existing video surveillance,this thesis realizes the detection of foreign body invasion and illegal behavior in the construction of railway personnel.The system is of great significance to ensure the safety of railway transportation and production.The main research work of this thesis is as follows:(1)In view of the lack of data sets of railway foreign body invasion and violations,the data set is preprocessed and augmented on the basis of its own data set,and the problems such as insufficient data sets are effectively solved.(2)In the aspect of foreign object intrusion detection,a railway foreign object intrusion detection algorithm based on attention and multi-scale targets(YOLOv7-SCT)is proposed.Aiming at the problem of missed detection caused by small targets in monitoring images,the model introduces a small target detection layer,CBAM attention mechanism,and Swin Transformer module to realize the model to identify small targets more accurately in complex environments.Experimental results show that the m AP of the algorithm is 92.5%,which is 3% higher than that of YOLOv7.(3)In the aspect of violation detection,a high-fine-grained behavior recognition algorithm is proposed.Aiming at the problem of human posture distortion and inaccurate key point positioning in complex scenes,a human posture skeleton point extractor based on improved HRNet is designed,an adaptive full-coverage human edge frame target detector based on human body structure is designed,and a human posture key point corrector KPC and a violation discriminator ST-GCN are added to reduce the misjudgment rate of violation detection.Experimental results show that the accuracy of the proposed algorithm reaches 87.0%,which is 1.8%higher than that of the original model.(4)Based on the above algorithm model,an intelligent image recognition system for railway operation safety based on edge devices is designed,which realizes user management,monitoring equipment management,intelligent analysis and early warning,video(image)review,log information management,etc.