摘要
随着公路隧道的不断建设,公路隧道的病害识别问题、养护问题日益增多,传统的隧道衬砌裂纹识别工作需要大量人工参与内业工作,效率不高。文章采用隧道快速检测车所采集的图像数据,利用卷积神经网络,研究得出了一套用于定向识别隧道检测车所拍摄的公路隧道衬砌开裂的模型。根据训练过程中发现的问题,针对裂纹数据集对神经网络识别效果的影响进行了对比研究,证明了按照不同裂纹走势分类图像所训练的神经网络识别能力较好,具有较好的适用性。
With the continuous construction of highway tunnels,the problems of disease identification and mainte⁃nance of highway tunnels are increasing.The traditional tunnel lining crack identification requires a lot of personnel to participate in the indoor work,resulting in low efficiency.Based on the image data collected by the rapid testing vehicle for highway tunnel condition,this paper uses the convolutional neural networks to train a set of models for recognizing the cracks in the highway tunnel linings.According to the problems found in the training process,a com⁃parative study on the effect of different crack data sets on the recognition effect of the neural network models is car⁃ried out,proving that the CNN model trained by images sorted by crack development trends has better recognition ability and better applicability.
作者
江桁
刘学增
朱合华
JIANG Heng;LIU Xuezeng;ZHU Hehua(Department of Geotechnical Engineering,Tongji University,Shanghai 200092;Civil Engineering Information Technology Research Center of Ministry of Education,Tongji University,Shanghai 200092)
出处
《现代隧道技术》
EI
CSCD
北大核心
2020年第5期61-65,共5页
Modern Tunnelling Technology
基金
国家重点研发计划(2018YFB2101004).
关键词
公路隧道
隧道检测车
卷积神经网络
裂纹识别
边缘检测
Highway
Testing vehicle for highway tunnel condition
Convolutional neural network
Crack detec⁃tion
Edge detection