| With the rapid development of the railway and the expansion of its network scale,the requirements for the safe operation of the current railway traction power supply system are higher than before,and the maintenance and protection measures for the traction substation equipment are gradually increasing,making the safety management of the field operation of each traction substation become an important task.This paper introduces the detection of helmets and human body in the operation site of traction substation.Because helmets and human body are not small targets,the balance between detection accuracy and detection speed can be considered when selecting or improving some target detection algorithms.Focus on the use of yolo-v3 algorithm for helmet detection,yolo-v3 algorithm modeling,algorithm network improvement.On the basis of yolo-v3,the feature of safety helmet carrying is extracted by constructing the depth residual network,which is df-yolo-v3.In this paper,the training set is established based on the screenshots of the monitoring video of traction substation.At the same time,the test analysis is carried out on fast r-cnn,fast r-cnn,ssd300,ssd512 and other algorithms,and AUC images of ten algorithms are drawn for analysis.The experimental results show that yo-v3 has obvious advantages in real-time detection,and the transmission rate meets the requirements of real-time detection.Compared with yolo-v3,the detection effect of df-yolo-v3 optimized by deep residual network is obviously improved,which shows that the multi-scale network structure improved in this paper is effective.Although the transmission rate of df-yolo-v3 is not the highest,its transmission rate has met the needs of real-time detection.The algorithm meets the requirements of real-time safety detection in the operation site of traction substation,and establishes an automatic detection platform for helmet wearing,needs analysis,platform design,platform implementation and platform test,etc.,and designs and realizes video / picture recognition,personnel information collection,helmet identification,alarm function,all-weather intelligent analysis,monitoring equipment management and equipment management.,monitoring record query and other functions.At present,the research results have been put into use in the operation site of traction substation,and the effect of trial operation is good.Through spot check,it is found that after the automatic detection platform for wearing safety helmet is put into use,the wearing rate of safety helmet rises from 90% to 98%.The development and implementation of automatic detection platform for wearing safety helmet has practical significance and value. |