Automated defect classification in sewer closed circuit television inspections using deep convolutional neural networks

被引:202
作者
Kumar, Srinath S. [1 ]
Abraham, Dulcy M. [1 ]
Jahanshahi, Mohammad R. [1 ]
Iseley, Tom [2 ]
Starr, Justin [3 ]
机构
[1] Purdue Univ, Lyles Sch Civil Engn, W Lafayette, IN 47907 USA
[2] Louisiana Tech Univ, Dept Civil Engn, Ruston, LA 71270 USA
[3] RedZone Robot, Pittsburgh, PA USA
关键词
Wastewater pipelines; Underground infrastructure; Automation; Robotics; Closed circuit television (CCTV); Convolutional neural networks; Condition assessment; Deep learning; Artificial intelligence; MORPHOLOGICAL SEGMENTATION; PIPE DEFECTS;
D O I
10.1016/j.autcon.2018.03.028
中图分类号
TU [建筑科学];
学科分类号
0813 ;
摘要
Automated interpretation of sewer CCTV inspection videos could improve the speed, accuracy, and consistency of sewer defect reporting. Previous research has attempted to use computer vision, namely feature extraction methods for automated classification of defects in sewer CCTV images. However, feature extraction methods use pre-engineered features for classifying images, leading to poor generalization capabilities. Due to large variations in sewer images arising from differing pipe diameters, in-situ conditions (e.g., fog and grease), etc., previous automated methods suffer from poor classification performance when applied to sewer CCTV videos. This paper presents a framework that uses deep convoluted neural networks (CNNs) to classify multiple defects in sewer CCTV images. A prototype system was developed to classify root intrusions, deposits, and cracks. The CNNs were trained and tested using 12,000 images collected from over 200 pipelines. The average testing accuracy, precision and recall were 86.2%, 87.7% and 90.6%, respectively, demonstrating the viability of this approach in the automated interpretation of sewer CCTV videos.
引用
收藏
页码:273 / 283
页数:11
相关论文
共 43 条
[1]  
Abadi M., 2015, TENSORFLOW LARGE SCA
[2]  
[Anonymous], 2012, Buried no longer: Confronting America's water infrastructure challenge
[3]  
[Anonymous], 2017, REP CARD AM INFR
[4]   The relevance of sewer deterioration modelling to support asset management strategies [J].
Caradot, Nicolas ;
Sonnenberg, Hauke ;
Kropp, Ingo ;
Ringe, Alexander ;
Denhez, Stephane ;
Hartmann, Andreas ;
Rouault, Pascale .
URBAN WATER JOURNAL, 2017, 14 (10) :1007-1015
[5]   Deep Learning-Based Crack Damage Detection Using Convolutional Neural Networks [J].
Cha, Young-Jin ;
Choi, Wooram ;
Buyukozturk, Oral .
COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING, 2017, 32 (05) :361-378
[6]  
Chae M. J., 2008, DIGITAL PIPELINE IMA, P1, DOI [10.1080/1573062X.2015.1076857, DOI 10.1080/1573062X.2015.1076857]
[7]   The devil is in the details: an evaluation of recent feature encoding methods [J].
Chatfield, Ken ;
Lempitsky, Victor ;
Vedaldi, Andrea ;
Zisserman, Andrew .
PROCEEDINGS OF THE BRITISH MACHINE VISION CONFERENCE 2011, 2011,
[8]   NB-CNN: Deep Learning-Based Crack Detection Using Convolutional Neural Network and Naive Bayes Data Fusion [J].
Chen, Fu-Chen ;
Jahanshahi, Mohammad R. .
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, 2018, 65 (05) :4392-4400
[9]  
Cleys T., 2017, NEXT GENERATION LASE, P226, DOI [10.1061/9780784480885.022, DOI 10.1061/9780784480885.022]
[10]   The consistency of visual sewer inspection data [J].
Dirksen, J. ;
Clemens, F. H. L. R. ;
Korving, H. ;
Cherqui, F. ;
Le Gauffre, P. ;
Ertl, T. ;
Plihal, H. ;
Mueller, K. ;
Snaterse, C. T. M. .
STRUCTURE AND INFRASTRUCTURE ENGINEERING, 2013, 9 (03) :214-228