Personnel Abnormal Behavior Recognition In Railway Freight Yard Based On Deep Learning

The abnormal behavior of personnel is a key factor affecting the loading and unloading safety of railway freight yards.At present,the abnormal behavior is mainly monitored by manual methods.With the development of deep learning,the use of deep learning technology for video image processing and analysis is more efficient than manual recognition.Aiming at the problems of low efficiency and untimely supervision in the current manual monitoring mode,this paper studies the abnormal behavior recognition of railway freight yard personnel based on the deep learning method.The specific research work includes the following three aspects.(1)Detection of personnel objects and helmet wearing conditions based on improved YOLOv5 s.In order to suppress the interference of irrelevant features and enhance the important features,the effective channel attention ECA module is combined with the large kernel attention LKA module,and the ECA-LKA attention module is designed,and the ECALKA is used to improve the YOLOv5 s backbone network and improve the features of the model.extraction capacity.Secondly,for the original YOLOv5 s to detect small-sized objects,there is a problem of missed detection.In its feature fusion network part,the upsampling layer is further increased to obtain a larger-sized feature map,and it is fused with shallow features to retain the positioning information of small objects.In addition,based on Bi FPN weighted bidirectional feature pyramid network,adaptive weights are added to different feature graphs to balance the importance of different features.Compared with the original model,the Precision,Recall and Mean Average Precision(m AP)of the improved YOLOv5 s model are improved by 1%,2% and 2.1% respectively.(2)Behavior recognition is realized by combining Deep Sort and Slow Fast algorithms.The temporal information of people’s behavior changes in the video is extracted through the fast channel of Slow Fast,the spatial location information is extracted through the slow channel,and the behavior recognition is completed by fusion processing of spatiotemporal feature information.And by combining the Deep Sort object tracking algorithm,the person identity association between the front and rear video frames is realized,so as to complete the longterm person behavior recognition.In addition,using the improved YOLOv5 s object detection algorithm as the detector of Deep Sort.(3)On the basis of object detection algorithm and behavior recognition algorithm,an abnormal behavior recognition system of railway freight yard personnel is constructed.The system mainly includes a data access module,a object detection module,and an abnormal behavior recognition module.Load the trained model for object detection and identification of abnormal behavior of personnel.When abnormal behavior of personnel is identified and the time is greater than the set threshold,an alarm will be issued to prompt the staff to take relevant measures to prevent further abnormal behaviors and safety accidents.