摘要
针对目前无人机航拍的绝缘子图像数据集中缺陷绝缘子样本数量较少的问题,提出一种融合图像数据集制作方法;同时针对绝缘子缺陷检测准确率低且检测网络参数量和计算量大的问题,提出一种改进的YOLOv5神经网络用于绝缘子的缺陷检测。首先采用无人机拍摄真实背景下绝缘子图像,然后借助机器人搭载双目相机和工业相机制作融合图像数据集,以平衡数据集正负样本比例,最后用神经网络对两种方式采集的数据集训练和验证。实验结果显示,采用GhostNet模块改进的YOLOv5网络缺陷检测mAP达98.43%,相比于未改进的YOLOv5网络mAP提高了6.02%,且通过机器人制作的绝缘子融合图像数据集具有与无人机拍摄的真实背景数据集相同的效果。因此,改进的YOLOv5神经网络能满足绝缘子缺陷检测要求。
Aiming at the problem that the number of defective insulator samples in the insulator image dataset taken by UAV is small,a method of making fusion image dataset is proposed.At the same time,aiming at the problems of low accuracy of the insulator defect detection and large amount of parameters and calculation of detection network,an improved YOLOv5 neural network is proposed for insulator defect detection.First,the insulator image under the real background is taken by UAV,and then the fusion image data set is made by using the robot equipped with binocular camera and industrial camera to balance the proportion of positive and negative samples in the data set.Finally,the neural network is used to train and verify the data set collected in both ways.The experimental results show that the improved YOLOv5 network defect detection mAP using the GhostNet module is 98.43%,which is 6.02%higher than the unimproved YOLOv5 network mAP,and the insulator fusion image data set made by robot has the same effect as the real background data set taken by UAV.Therefore,the improved YOLOv5 neural network can meet the requirements of insulator defect detection.
作者
王年涛
王淑青
黄剑锋
要若天
刘逸凡
WANG Niantao;WANG Shuqing;HUANG Jianfeng;YAO Ruotian;LIU Yifan(Hubei University of Technology,Solar Energy Efficient Utilization and Storage Operation Control Hubei Key Laboratory,Wuhan 430068,China;School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China;Wuhan National Laboratory for Optoelectronics,Huazhong University of Science and Technology,Wuhan 430074,China)
出处
《激光杂志》
CAS
北大核心
2022年第8期60-65,共6页
Laser Journal
基金
国家自然科学基金(No.61873195)。
作者简介
王年涛(1995-),男,硕士研究生,主要研究方向为接触网绝缘子缺陷检测。E-mail:lentowang@hbut.edu.cn;通信作者:王淑青(1969-),女,博士,教授,从事智能检测与控制研究。E-mail:254831618@qq.com。
