基于深度卷积神经网络的隧道衬砌裂缝识别算法
|
摘要
衬砌裂缝严重影响了铁路隧道的安全运营,采用机器视觉技术快速获取衬砌图片并进行裂缝识别是国内外的研究热点。衬砌裂缝图像信号具有复杂的特性,存在水渍、污染及其他结构缝等引起的噪声,加之光照不均匀、分布不规律等原因,使得传统的图像处理方法难以快速、准确地检测衬砌裂缝。本文提出一种基于深度神经网络的隧道衬砌裂缝识别算法,有效解决了裂缝识别速度慢、精度低等问题。分类结果精度达到94%,识别速度在GPU(Pascal Titan X)上每张图片仅需0.05 s;分割网络性能均交并比可达到65%,能够准确分割出裂缝形状。该算法具有很好的工程应用价值。
It is a research hotspot for railway to achieve rapid and accurate detection and identification of tunnel lining cracks, since tunnel lining cracks seriously affect the safety of railway tunnels.However,tunnel lining crack image signals have complex features.There are noises induced by waterlogging,pollution and other structural joints,in addition,uneven illumination and irregular distribution.Thus traditional image processing methods are difficult to achieve rapid and accurate detection and identification of tunnel lining cracks.An algorithm for tunnel lining crack identification based on deep convolution neural networks was proposed in this paper,which can effectively solve those problems above. The identification accuracy achieves 94%,and the identification speed is up to 0. 05 s per picture in the hardware environment of GPU(Pascal Titan X).The segmentation performance achieves 65% MIOU(Mean Intersection Over Union). The algorithm can accurately segment crack shape and has good engineering application value.
It is a research hotspot for railway to achieve rapid and accurate detection and identification of tunnel lining cracks, since tunnel lining cracks seriously affect the safety of railway tunnels.However,tunnel lining crack image signals have complex features.There are noises induced by waterlogging,pollution and other structural joints,in addition,uneven illumination and irregular distribution.Thus traditional image processing methods are difficult to achieve rapid and accurate detection and identification of tunnel lining cracks.An algorithm for tunnel lining crack identification based on deep convolution neural networks was proposed in this paper,which can effectively solve those problems above. The identification accuracy achieves 94%,and the identification speed is up to 0. 05 s per picture in the hardware environment of GPU(Pascal Titan X).The segmentation performance achieves 65% MIOU(Mean Intersection Over Union). The algorithm can accurately segment crack shape and has good engineering application value.
引文
[1]柴雪松,李健超.基于图像识别技术的隧道衬砌裂缝检测系统研究[J].铁道建筑,2018,58(1):20-24.
[2]FUJITA Y,MITANI Y,HAMAMOTO Y.A Method for Crack Detection On A Concrete Structure[C]//Pattern Recognition,the 18th Internation Conference.IEEE,2006.
[3]LANDSTROM A,THURLEY M J.Morphology-based Crack Detection For Steel Slabs[J].Selected Topics in Signal Processing,IEEE Journal,2012,6(7):866-875.
[4]李刚,贺昱曜.不均匀光照的路面裂缝检测和分类新方法[J].光子学报,2010,39(8):1405-1408.
[5]褚燕利.基于灰度图像及其纹理特性的裂缝特征提取[J].公路,2010,7(7):131-136.
[6]卢晓霞.基于图像处理的混凝土裂缝宽度检测技术的研究[D].电子科技大学,2010.
[7]HINTON G E,SALAKHUTDINOV R R.Reducing the Dimensionality of Data with Neural Networks[J].Science,2006,313(7):504-507.
[8]BENGIO Y,COURVILLE A,VINCENT P.Representation Learning:A Review and New Perspectives[J].Pattern Analysis and Machine Intelligence,IEEE Transactions,2013,35(8):1798-1828.
[9]SERMANET P,CHINTALA S,LECUN Y.Convolutional Neural Networks Applied to House Numbers Digit Classification[C]Pattern Recognition,the 21st International Conference.IEEE,2012.
[10]BENGIO Y.Deep Learning[M].Massachusetts:MIT Press,2015.
[11]AGATONOVIC-KUSTRIN S,BERESFORD R.Basic Concepts of Artificial Neural Network(ANN)Modeling and Its Application in Pharmaceutical Research[].Journal of Pharmaceutical and Biomedical Analysis,2000,22(5):717-727.
[12]BENGIO Y.Learning Deep Architectures for AI[J].Foundations and Trends in Machine Learning,2009,2(1):1-127.
[13]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Image Net Classification with Deep Convolutional Neural Networks[J].Neural Information Processing Systems,2012,1-9.
[14]SZEGEDY C,LIU W,JIA Y,et al.Going Deeper with Convolutions[J].Computer Vision and Pattern Recognition,2015,1-9.
[15]SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-scale Image Recognition[C]//the International Conference on Learning Representations,2015.
[16]HE K,ZHANG X Y,REN S Q,et al.Deep Residual Learning for Image Recognition[J].Computer Vision and Pattern Recognition,2015(12):1-9.
[2]FUJITA Y,MITANI Y,HAMAMOTO Y.A Method for Crack Detection On A Concrete Structure[C]//Pattern Recognition,the 18th Internation Conference.IEEE,2006.
[3]LANDSTROM A,THURLEY M J.Morphology-based Crack Detection For Steel Slabs[J].Selected Topics in Signal Processing,IEEE Journal,2012,6(7):866-875.
[4]李刚,贺昱曜.不均匀光照的路面裂缝检测和分类新方法[J].光子学报,2010,39(8):1405-1408.
[5]褚燕利.基于灰度图像及其纹理特性的裂缝特征提取[J].公路,2010,7(7):131-136.
[6]卢晓霞.基于图像处理的混凝土裂缝宽度检测技术的研究[D].电子科技大学,2010.
[7]HINTON G E,SALAKHUTDINOV R R.Reducing the Dimensionality of Data with Neural Networks[J].Science,2006,313(7):504-507.
[8]BENGIO Y,COURVILLE A,VINCENT P.Representation Learning:A Review and New Perspectives[J].Pattern Analysis and Machine Intelligence,IEEE Transactions,2013,35(8):1798-1828.
[9]SERMANET P,CHINTALA S,LECUN Y.Convolutional Neural Networks Applied to House Numbers Digit Classification[C]Pattern Recognition,the 21st International Conference.IEEE,2012.
[10]BENGIO Y.Deep Learning[M].Massachusetts:MIT Press,2015.
[11]AGATONOVIC-KUSTRIN S,BERESFORD R.Basic Concepts of Artificial Neural Network(ANN)Modeling and Its Application in Pharmaceutical Research[].Journal of Pharmaceutical and Biomedical Analysis,2000,22(5):717-727.
[12]BENGIO Y.Learning Deep Architectures for AI[J].Foundations and Trends in Machine Learning,2009,2(1):1-127.
[13]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Image Net Classification with Deep Convolutional Neural Networks[J].Neural Information Processing Systems,2012,1-9.
[14]SZEGEDY C,LIU W,JIA Y,et al.Going Deeper with Convolutions[J].Computer Vision and Pattern Recognition,2015,1-9.
[15]SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-scale Image Recognition[C]//the International Conference on Learning Representations,2015.
[16]HE K,ZHANG X Y,REN S Q,et al.Deep Residual Learning for Image Recognition[J].Computer Vision and Pattern Recognition,2015(12):1-9.