基于卷积神经网络的路表病害识别与测量

PDF全文下载(2193 KB)

中国公路学报 ›› 2018, Vol. 31 ›› Issue (1) : 1-10.

道路工程

基于卷积神经网络的路表病害识别与测量

沙爱民, 童峥, 高杰

作者信息 +

Recognition and Measurement of Pavement Disasters Based on Convolutional Neural Networks

SHA Ai-min, TONG Zheng, GAO Jie

Author information +

文章历史 +

摘要

为进一步提高利用二维图像统计路面病害的精度与效率，将卷积神经网络（Convolutional Neural Network，CNN）技术引入了基于图像分析的路面病害识别与测量。首先，将原始图像进行等尺寸分割作为CNN的训练样本。其次，经结构设计、前反馈算法训练及样本测试3个步骤后，建立病害识别模型（CNN1）。用训练完成的CNN1对所有图像进行病害类型识别并将输出结果作为裂缝特征提取模型（CNN2）和坑槽特征提取模型（CNN3）的训练样本。采用相同步骤建立裂缝特征提取和坑槽特征提取模型，完成训练后，运行CNN2，CNN3对路面裂缝与坑槽图像进行特征提取。最后，分析图像分辨率对3个CNN识别和特征提取精度以及效率的影响。结果表明：CNN1可以准确识别多种病害，CNN2的裂缝长度提取的平均误差为4.27%，宽度提取的平均误差为9.37%，裂缝病害严重等级判断准确率为98.99%；CNN3的单张图像中的坑槽个数测量无误差，单个坑槽面积的平均误差为13.43%，坑槽病害等级判定准确率为95.32%，可见CNN具有较高的测量精度；CNN1在使用CPU的情况下测试完成原始图像平均用时为704 ms·幅^-1，CNN2用时为5 376 ms·幅^-1，采用图形处理器加速后CNN1用时为192 ms·幅^-1，CNN2测试平均用时为1 024 ms·幅^-1，可见CNN在图形处理器加速下效率具有显著优势，相比其他方法，在图像分辨率高于70像素时，CNN对路面裂缝与坑槽的识别与测量具有运算高效、结果精准等优势。

Abstract

In order to further improve the precision and efficiency of pavement disasters using two-dimensional images, convolutional neural network (CNN) was introduced to recognize and measure pavement disasters. First, pavement images were divided into equal-sized maps, which were used as training samples of disaster recognition CNN. Then disaster recognition CNN1 was established by the structural design, before-feedback algorithm and sample tests. Subsequently, all equal-sized maps were input into well-trained disaster recognition CNN1, and the results were used as training samples of crack feature extraction model CNN2 and pit slot feature extraction model CNN3. Meanwhile, the crack feature extraction model CNN2 and the pit slot feature extraction model CNN3 were established with the same procedures. Afterwards, the pavement crack and images of pit slot were extracted by well-trained CNN2 and CNN3. Last but not least, the influences of image resolution ratio on the CNN recognition, precision of feature extraction, and efficiency were analyzed. The results show that various pavement disasters can be accurately recognized by CNN1, and the average errors of crack length and crack width of CNN2 are 4.27% and 9.37% respectively. The accuracy rate of crack disease severity level is 98.99%. There is no error of the number of pit slot CNN3, and the average error of each pit slot area is 13.43%. The accuracy rate of pit slot disease severity level is 95.32%. Thus CNN systems show high accuracy. The disaster recognition CNN1 and feature extraction CNN2 for testing original images in CPU conditions take up to 704 ms·sheet^-1 and 5 376 ms·sheet^-1 on average, whilst the disaster recognition CNN1 and feature extraction CNN2 for testing images after the acceleration by graphics processor in GPU conditions take up to 192 ms·sheet^-1 and 1 024 ms·sheet^-1 on average. Thus CNN systems present high efficiency in GPU condition. Compared with other methods, CNN presents high accuracy and efficiency in recognition and measurement of cracks and pit slot when image solutions are higher than 70 dpi.

导出引用

沙爱民, 童峥, 高杰. 基于卷积神经网络的路表病害识别与测量[J]. 中国公路学报, 2018, 31(1): 1-10

SHA Ai-min, TONG Zheng, GAO Jie. Recognition and Measurement of Pavement Disasters Based on Convolutional Neural Networks[J]. China Journal of Highway and Transport, 2018, 31(1): 1-10

中图分类号： U416.2

参考文献

[1] TOSHIHIKO F,KEIJI T,MAKOTO N,et al.Automatic Pavement-distress-survey System[J].Journal of Transportation Engineering,1990,116(3):280-286.
[2] LUO W,LI L,WANG K.Automated Pavement Horizontal Curve Measurement Methods Based on Inertial Measurement Unit and 3D Profiling Data[J].Journal of Traffic and Transportation Engineering:English Edition,2016,3(2):137-145.
[3] 曹建农,张昆,元晨,等.用Mean Shift实现路面裂缝损伤自动识别与特征测量[J].计算机辅助设计与图形学学报,2014,26(9):1440-1459. CAO Jian-nong,ZHANG Kun,YUAN Chen,et al.Automatic Road Cracks Detection and Characterization Based on Mean Shift[J].Journal of Computer-aided Design & Computer Graphics,2014,26(9):1440-1459.
[4] 李清泉,邹勤,毛庆洲.基于最小代价路径搜索的路面裂缝检测[J].中国公路学报,2010,23(6):28-32. LI Qing-quan,ZOU Qin,MAO Qing-zhou.Pavement Crack Detection Based on Minimum Cost Path Searching[J].China Journal of Highway and Transpor,2010,23(6):28-32.
[5] LI Q,ZOU Q,ZHANG D.FoSA:F^* Seed-growing Approach for Crack-line Detection from Pvement Images[J].Image and Vision Computing,2011,12(29):861-872.
[6] 沈照庆,彭余华,舒宁.一种基于SVM的路面影像损伤跨尺度识别方法[J].武汉大学学报:信息科学版,2013,38(8):993-997. SHEN Zhao-qin,PENG Yu-hua,SHU Ning.A Road Damage Identification Method Based on Scale-span Image and SVM[J].Geomatics and Information Science of Wuhan University,2013,38(8):993-997.
[7] BARAT C,CHRISTOPHE D.String Representations and Distances in Deep Convolutional Neural Networks for Image Classification[J].Pattern Recognition,2016,54:104-115.
[8] SHI B,BAI X,YAO C.Script Identification in the Wild Via Discriminative Convolutional Neural Network[J].Pattern Recognition,2016,52:448-458.
[9] LENG B,GUO S,ZHANG X,et al.3D Object Retrieval with Stacked Local Convolutional Autoencoder[J].Signal Processing,2015,112:119-128.
[10] XU J,LUO X,WANG G.A Deep Convolutional Neural Network for Segmenting and Classifying Epithelial and Stromal Regions in Histopathological Images[J].Neurocomputing,2016,191:214-223.
[11] LECUN Y,BERNHARD B,BOTTOU L,et al.Handwritten Digit Recognition with a Back-propagation Network[J].Neural Information Processing Systems,1990,75:396-404.
[12] LECUN Y,BENGIO Y,HINTON G.Deep Learning[J].Nature,2015,521(7553):436-444.
[13] ZBONTAR J,LECUN Y.Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches[J].Journal of Machine Learning Research,2016,17:451-460.
[14] XU T,ZHU F,EDWARD K,et al.Dual Many-to-one-encoder-based Transfer Learning for Cross-dataset Human Action Recognition[J].Image and Vision Computing,2016,55(2):127-137.
[15] EARNEST P,KRISHNA M.Human Action Recognition Using Genetic Algorithms and Convolutional Neural Networks[J].Pattern Recognition,2016,59:199-212.
[16] JORDI M,ALBERTO A,ROBERTO P.Local Deep Neural Networks for Gender Recognition[J].Pattern Recognition Letters,2016,70:80-86.
[17] JIA F,LEI Y,LIN J,et al.Deep Neural Networks:A Promising Tool for Fault Characteristic Mining and Intelligent Diagnosis of Rotating Machinery with Massive Data[J].Mechanical Systems and Signal Processing,2016,72-73:303-315.
[18] TONG Z,GAO J,HAN Z,et al.Recognition of Asphalt Pavement Crack Length Using Deep Convolutional Neural Networks[J].Road Materials and Pavement Design,https://doi.org/10.1080/14680629.2017.1308265.
[19] CHA Y,CHOI W,BUYUKOZTURK O.Deep Learning-based Crack Damage Detection Using Convolutional Neural Ntworks[J].Computer-aided Civil and Infrastructure Engineering,2017,32:361-378.
[20] TONG Z,GAO J,ZHANG H.Recognition,Location,Measurement,and 3D Reconstruction of Concealed Cracks Using Convolutional Neural Networks[J].Construction and Building Materials,2017,146,775-787.
[21] ZHOU S,SHEN W,ZENG D,et al.Spatial-temporal Convolutional Neural Networks for Anomaly Detection and Localization in Crowded Scenes[J].Signal Processing:Image Communication,2016,47:358-68.
[22] FENG L,PO L,LI Y,et al.Integration of Image Quality and Motion Cues for Face Anti-spoofing:A Neural Network Approach[J].Journal of Visual Communication and Image Representation,2016,38:451-460.
[23] HU Y,CHANG H,NIAN F,et al.Dense Crowd Counting from Still Images with Convolutional Neural Networks[J].Journal of Visual Communication and Image Representation,2016,38:530-539.
[24] FAN H,ZHOU E.Approaching Human Level Facial Landmark Localization by Deep Learning[J].Image and Vision Computing,2016,47:27-35.
[25] NIAN F,LI T,WANG Y,et al.Pornographic Image Detection Utilizing Deep Convolutional Neural Networks[J].Neurocomputing,2016,210:283-293.
[26] ADRIANA T,ABDON D,MONICA B,et al.Prediction of Compressive Strength of Concrete Containing Construction and Demolition Waste Using Artificial Neural Networks[J].Construction and Building Materials,2013,38:717-722.
[27] DONG Y,WU Y.Adaptive Cascade Deep Convolutional Neural Networks for Face Alignment[J].Computer Standards & Interfaces,2015,42:105-112.
[28] FENG L,PO L,LI Y,et al.Integration of Image Quality and Motion Cues for Face Anti-spoofing:A Neural Network Approach[J].Journal of Visual Communication and Image Representation,2016,38:451-460.

基金

“十二五”国家科技支撑计划项目（2014BAG05B04）；交通运输部建设科技项目（2014 318 223 010）

PDF全文下载(2193 KB)

5267

Accesses

Citation

Detail

段落导航

摘要
Abstract
关键词
Key words
引用本文
参考文献
基金

收稿日期	出版日期
2017-01-07	2018-01-20
发布日期
2018-02-05

选择文件类型/文献管理软件名称

选择包含的内容

摘要

Abstract

关键词

Key words

引用本文

{{custom_sec.title}}

{{custom_sec.title}}

参考文献

{{custom_fnGroup.title_cn}}

脚注

基金

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

摘要

Abstract

关键词

Key words

引用本文

{{custom_sec.title}}

{{custom_sec.title}}

参考文献

{{custom_fnGroup.title_cn}}

脚注

基金