摘要
针对现有技术方案在铁轨弹条检测中存在检测率低或无法满足实时性要求等问题,提出了一种基于Yolov3算法的铁轨弹条检测方案。构建以卷积神经网络为主的网络结构,在每一层卷积层后加入残差块结构,加强了网络的特征学习能力;引入特征金字塔结构,加强网络层特征复用,实现对小目标检测精度的提升。相较于其他深度残差网络,所构建的网络结构检测率为95.10%,平均检测速度大于70 FPS,在检测精度未明显下降的情况下,实现了检测速度的大幅提升。在各种试验环境条件下,算法的鲁棒性良好,因此所提方案可满足实际的工程运用需求,实现对铁轨弹条的全天候检测。
Aiming at the problems of low detection rate or inability to meet real-time requirements in the detection of spring bar fasteners in the prior art solutions,a spring bar fastener detection scheme based on Yolov3 algorithm was proposed.In this scheme,the network structure based on convolutional neural network was constructed,and the residual block structure was added after each convolution layer to strengthen the feature learning ability of the network;by introducing the feature pyramid structure and strengthening the feature reuse in the network layer,the accuracy of small target detection can be improved.Compared with other depth residual networks,the detection rate of the constructed network structure is 95.10%,and the average detection speed is greater than 70 FPS.The detection speed is greatly improved without significant decline in detection accuracy.The algorithm has good robustness under various experimental environment conditions,so this scheme can meet the actual engineering application requirements and realize the all-weather detection of spring bar fasteners.
作者
张楠
厉小润
王森荣
林超
Zhang Nan;Li Xiaorun;Wang Senrong;Lin Chao(College of Electrical Engineering,Zhejiang University,Hangzhou Zhejiang 310063,China;China Railway Siyuan Survey and Design Group Co.,Ltd.,Wuhan Hubei 430027,China)
出处
《电气自动化》
2023年第2期22-24,共3页
Electrical Automation
关键词
铁轨弹条
Yolov3算法
特征金字塔
目标检测
残差结构
pring bar fastener
Yolov3 algorithm
feature pyramid network
target detection
residual structure
