基于改进YOLOv3的铁路关键作业流程自动鉴别系统

Automatic identification system of railway critical operation process based on improved YOLOv3

针对铁路列车关键安全点的检测问题,提出一种改进YOLOv3网络的铁路关键作业流程自动鉴别系统,提高了对小物体特征的检测精度。使用残差网络对darknet53结构进行改进,采用FPN网络加强特征提取以及融合(104,104,24)的检测分支,通过YOLO_Head获得预测结果,从而提高小目标检测精度。在铁路数据集上的实验结果表明,改进的YOLOv3网络在较小的关键点的检测上有了较大突破,列车中电箱锁的置信度提升了0.33,精准度提升了21.39%,召回率提升了10.34%,F1值提升了0.16;3种关键特征的mAP提升了12.05%。

Aiming at the detection problem of railway train critical safety points,an automatic identification system of railway critical operation process based on improved YOLOv3 network is proposed,which improves the detection accuracy of small object features.The residual network is used to improve the darknet53 structure,FPN network is used to strengthen feature extraction and fusion of detection branches(104,104,24),and YOLO_Head is used to obtain prediction results,so as to improve the accuracy of small target detection.The experimental results on the railway data set show that the improved YOLOv3 network has a great breakthrough in the detection of small key points.The confidence,accuracy,recall rate and F1 value of the train electric box lock are improved by 0.33,21.39%,10.34%and 0.16 respectively.MAP for 3 key features is improved by 12.05%.