摘要
为了提高钢桁架桥螺栓病害检测和识别效率、完善分析方法,本文在无人机航拍视频的基础上,提出了一种基于深度学习的螺栓病害智能识别方法。通过混合高斯算法、Canny边缘检测、最小包围圆算法等传统图像处理手段对航拍视频进行预处理,实现钢桁架桥螺栓图像的批量化提取,并通过对螺栓图像采取缩放、旋转、变形等措施拓展螺栓图像的样本数;采用迁移学习引入深度学习模型INCEPTION-V3,经过训练,当螺栓数据测试集上的准确率大于95%时,可满足工程精度需求;并将该方法应用于实际工程,当把0.8设置为计算螺栓病害概率的分割点时,该方法对螺栓病害具有较好的识别效果,同时能够实现自动化、智能化,避免人为主观判断带来的影响。
To improve the efficiency of detection and identification of bolt damage of steel truss bridges and optimize the analysis methods,a method for intelligent identification of bolt disease based on deep learning was proposed,using unmanned aerial vehicle(UAV)video.The UAV video was pre-processed by traditional image processing methods such as Gaussian algorithm,Canny edge detection and minimum bounding circle algorithm.Batch extraction of steel truss bolt images was realized,and measures such as scaling,rotation and deformation of bolt images were taken to expand the number of samples of the bolt image.The deep learning model of INCEPTION-V3 was introduced to transfer learning.After training,the accuracy rate on the bolt data test set was greater than 95%,which could meet the needs of engineering accuracy.Results showed that when 0.8 was set as the dividing point for calculating the probability of disease,the method had a better recognition effect on bolt disease,and could realize automation and intelligence,and avoid the impact of human subjective judgment.
作者
沈浩
江臣
陈宇文
王国香
李枝军
SHEN Hao;JIANG Chen;CHEN Yuwen;WANG Guoxiang;LI Zhijun(Jiangsu Transportation Research Institute Co. Ltd., Nanjing 210017, China;Jiangsu Provincial Transportation Engineering Construction Bureau, Nanjing 210004, China;College of Civil Engineering,Nanjing Tech University, Nanjing 211800, China)
出处
《南京工业大学学报（自然科学版）》
CAS
北大核心
2020年第5期608-615,共8页
Journal of Nanjing Tech University(Natural Science Edition)
基金
江苏省交通运输科技项目(2017-2-10,2019Y18)。
关键词
钢桁架桥
螺栓病害
无人机航拍
图像处理
迁移学习
神经网络
steel truss bridges
bolt diseases
unmanned aerial vehicle(UAV)video
image processing
transfer learning
neural networks
