Research On Foreign Body Detection Technology In Pantograph Area Of Urban Rail Train Based On Deep Learning

In recent years,urban rail transit has entered the stage of rapid development,and the safety impact of road network operation has attracted much attention.Urban rail train pantograph area foreign body detection technique is becoming a hot spot of rail traffic safety research content,the traditional non-contact foreign body detection method in urban rail train detection under the electrical area was slower,insufficient recognition accuracy and poor robustness,deep learning foreign body detection methods have achieved good effect but there are still shortcomings.In this paper,the deficiencies of feature extraction and feature fusion of YOLOv4 network are analyzed,and the feature extraction network and feature fusion network of YOLOv4 are improved respectively,so as to construct a foreign body detection network more suitable for pantograph area of urban rail train.The research work of this paper is as follows:(1)Aiming at the low detection accuracy of YOLOv4 in the process of foreign body detection in the bow area of urban rail trains,a YOLOv4 foreign body detection network based on attention mechanism is proposed.The ECANet attention module was introduced into the feature extraction network of YOLOv4 to enhance the feature extraction capability of the network and make the receptive field pay more attention to the target features while ignoring irrelevant features.Experimental results show that the improved network can effectively improve the accuracy of foreign body recognition.(2)Aiming at the problem of small size foreign body missing in YOLOv4 foreign body detection,an adaptive spatial feature fusion YOLOv4 foreign body detection network was proposed on the basis of previous studies.The feature fusion network in YOLOv4 was replaced by FPN+ASFF structure for adaptive feature fusion.By forming different weights of feature image fusion at different scales,the effective features were highlighted and invalid features were suppressed,and the feature expression ability of the network for small target foreign bodies was enhanced.The loss strategy of α-CIoU boundary box is adopted to optimize targets at all levels and create more space to learn features of high Io U targets,thus improving the training effect of the network.Experimental results show that the improved network can effectively reduce the missing rate of small target foreign body detection.(3)In this paper,the improved YOLOv4 algorithm is applied to the actual pantograph area foreign body detection scene of urban rail train,and a pantograph area foreign body detection system of urban rail train is established.The acquisition scheme of pantograph area image of urban rail train was formulated and implemented,the preprocessing method of foreign body image in pantograph area of urban rail train was selected,and the host computer software for foreign body detection in pantograph area was developed.The improved YOLOv4 algorithm was verified by the urban rail train field passing experiment.The experimental results show that the accuracy rate of the algorithm developed in this paper reaches 82.6%,the recall rate reaches 95.85%,and the foreign body in pantograph area of urban rail train can be accurately detected.