摘要
针对人工检查接触网吊弦线夹缺陷耗时耗力且效率低下的问题,提出利用深度学习实现吊弦线夹状态分类的方法.首先利用加入特征金字塔和K-means算法改进的Faster R-CNN算法准确地定位到吊弦线夹,然后采用加入SENet注意力机制模块的Inception-ResNet-V2网络对接触网吊弦线夹螺母的缺失、松脱、正常三种状态进行高效准确的自动分类,达到计算机辅助检查的效果.仿真实验结果表明:该方法对吊弦线夹三种状态的分类准确率较高,平均准确率达到了96.61%,并具有高精度、易泛化的特点,为接触网零部件的缺陷检测任务奠定了必要的基础.
Aiming at the problem of time-consuming,labor-consuming and inefficient manual inspection of catenary dropper clamp defects,a method of realizing dropper clamp state classification by deep learning is proposed.Firstly,the improved Faster R-CNN algorithm with feature pyramid and K-means algorithm is used to accurately locate the dropper clamp,and then the Inception-ResNet-V2 network with SENet attention mechanism module is used to automatically classify the missing,loose and normal states of catenary dropper clamp nut with high efficiency and accuracy,so as to achieve the effect of computer-aided inspection.The simulation results show that this method has high classification accuracy for the three states of catenary clamp,with an average accuracy of 96.61%,and has the characteristics of high precision and easy generalization,which lays a necessary foundation for the defect detection task of catenary parts.
作者
梅小云
顾桂梅
陈充
张存俊
MEI Xiao-yun;GU Gui-mei;CHEN Chong;ZHANG Cun-jun(School of Automaton and Electrical Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;China Railway Lanzhou Bureau Group Co.,Ltd.,Lanzhou 730030,China)
出处
《兰州交通大学学报》
CAS
2022年第1期61-67,共7页
Journal of Lanzhou Jiaotong University
基金
甘肃省科技计划资助(20JR10RA216)。
