Detection Method of Catenary Hanging String Based on YOLOv7x
[Objective]Aiming at the potential risk caused by overhead contact wire defects during railway opera-tion,an improved YOLOv7x method for overhead contact wire defect identification is proposed.[Method]Firstly,Swin Transformer network is introduced at the end of the backbone feature extraction layer to replace the origi-nal extended and efficient layer aggregation network module,so as to improve the ability of the network to grasp global information.Then the SIoU(SCYLLA-IoU)loss function is used to replace the original network loss func-tion,and the direction penalty mechanism is added to the convergence process of the prediction frame.Finally,CA is integrated with the extended and efficient layer aggregation network module to enhance the global recep-tive field of the neck network module.[Result]Experimental simulation results show that the accuracy of the model trained with the improved algorithm reaches 95.9%,which is 4.7%higher than that of the original YO-LOv7x algorithm,and the detection speed reaches 52 frames per second.[Conclusion]The improved algorithm solves the problem of low detection efficiency in hanging strings defect identification,which may improve the ef-ficiency of detection of hanging strings defect in practice.
detection of catenary hanging string defectsYOLOv7xSwin TransformerSIoU loss functionCA
国家科技期刊平台
NSTL
万方数据
