摘要
为解决人工检测斜拉桥拉索表面缺陷速度慢、安全性低,传统图像处理方法鲁棒性差、易受环境影响等问题,利用全景成像实现斜拉桥拉索表面缺陷检测。采用PVC管模拟拉索表面,构造孔洞、缝隙、损伤三种常见表面缺陷;四个摄像头获取拉索表面全景图像,无线图像传输设备将数据传送至计算机,通过图像掩膜去除背景并搭建缺陷数据集;利用YOLOX网络训练数据集并在测试集上实验。结果表明:YOLOX网络的均精确度(mAP)为92.77%,帧率(FPS)为18,检测结果均优于其它主流目标检测方法,可实现拉索表面缺陷实时精确检测。
In order to solve the problems like slow speed,low safety,poor robustness of traditional image processing methods,and susceptibility to environmental influences and so on in the manual detection for the surface defects of cables of the cable-stayed bridge,the panoramic imaging is used to achieve the surface defect detection for cables of the cable-stayed bridge.Three kinds of common surface defects like holes,gaps,and damages are constructed by using PVC pipes to simulate the cable surface;the panoramic image of the cable surface is obtained by four cameras,the data is transmitted to the computer through the wireless image transmission device,and the background is removed through the image mask and the defect dataset is constructed;the dataset is trained by using YOLOX network and the experiment is made on the test set.The results show that:the average accuracy(mAP)of YOLOX network is 92.77%,and the frame rate(FPS)is 18.The detection results are superior to other mainstream object detection methods,which can achieve the real-time and accurate detection of cable surface defects.
作者
杨增祥
陈琳
陈悦
Yang Zengxiang;Chen Lin;Chen Yue(Taizhou Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Taizhou 318000,China)
出处
《北方交通》
2023年第12期1-4,共4页
Northern Communications
基金
国网浙江省电力有限公司群创项目(5211T8210002)。
关键词
斜拉桥
拉索表面
全景成像
缺陷检测
YOLOX网络
Cable-stayed bridge
Cable surface
Panoramic imaging
Defect detection
YOLOX network
