Reserch On Track Foreign Body Intrusiondetection Algorithm Based On Object Detection

At present,in the early warning system of foreign matter intrusion in railway security,there are some problems such as low accuracy,unable to locate the position,size and shape of foreign matter through the traditional contact detection.Through the regular inspection by the security personnel on the track,there will also be some problems such as tools left on the track,missing inspection,foreign matter that is not easy to find through the inspection due to the negligence of the inspection personnel The system is objective.In recent years,railway practitioners combine artificial intelligence technology with image and video analysis,which can detect and alarm the invasion of foreign objects in the track,investigate the hidden dangers of train operation safety,and monitor the operation environment in real time.The image analysis technology based on target detection can solve the above problems,but the detection accuracy,detection speed and model of target detection technology are too large,which leads to high deployment cost and can not play the maximum value in the track foreign body detection.This study is based on the research of track foreign body intrusion of target detection to improve the accuracy of track foreign body detection,reduce the frequency of missed detection and false detection,and reduce the cost The scale of the model is improved,and it is easier to deploy.The main contents of this paper are as follows:(1)A private data set of foreign body intrusion detection in the background of railway is constructed.A large amount of foreign body intrusion data is needed in the railway safety detection system.At present,there is no public data set on the network for research.This paper analyzes the installation position,height and angle of the monitoring camera in the railway,and simulates the collection of foreign body intrusion detection data set under the background of railway.Based on the data set of track foreign body intrusion detection constructed in this paper,the algorithm research is carried out.(2)Based on yolov3,the algorithm of track foreign body intrusion detection is studied.Firstly,to solve the problem of insufficient data sets,shuffle is used for data enhancement;Then,in order to improve the accuracy of foreign body detection and reduce the probability of misjudgment,DIo U loss is used to replace the coordinate position loss in yoov3;Then,in order to reduce the storage size of the model,reduce the memory consumption and speed up the reasoning speed,the lightweight network D-Mobilenetv3 optimized by D-SE attention mechanism is used as the backbone network of the algorithm,and the standard convolution in the yolov3 detector head is replaced by the deep separable convolution.The experimental results show that the improved algorithm has higher detection accuracy,smaller model and faster reasoning speed than the original yolov3 algorithm.