한국농공학회논문집 (Journal of The Korean Society of Agricultural Engineers)

  • 제61권3호
  • /
  • Pages.55-65
  • /
  • 2019
  • /
  • 1738-3692(pISSN)
  • /
  • 2093-7709(eISSN)
한국농공학회 (The Korean Society of Agricultural Engineers)
권호 표지
DOI QR코드

DOI QR Code

딥러닝과 전이학습을 이용한 콘크리트 균열 인식 및 시각화

Recognition and Visualization of Crack on Concrete Wall using Deep Learning and Transfer Learning

  • Lee, Sang-Ik (Department of Rural Systems Engineering, Seoul National University) ;
  • Yang, Gyeong-Mo (Department of Rural Systems Engineering, Seoul National University) ;
  • Lee, Jemyung (Division of Environmental Science and Technology, Kyoto University) ;
  • Lee, Jong-Hyuk (Department of Rural Systems Engineering, Seoul National University) ;
  • Jeong, Yeong-Joon (Department of Rural Systems Engineering, Seoul National University) ;
  • Lee, Jun-Gu (Rural Research Institute, Korea Rural Community Corporation) ;
  • Choi, Won (Department of Rural Systems Engineering, Research Institute of Agriculture and Life Sciences, Seoul National University)
  • 투고 : 2019.02.28
  • 심사 : 2019.05.02
  • 발행 : 2019.05.31
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

Although crack on concrete exists from its early formation, crack requires attention as it affects stiffness of structure and can lead demolition of structure as it grows. Detecting cracks on concrete is needed to take action prior to performance degradation of structure, and deep learning can be utilized for it. In this study, transfer learning, one of the deep learning techniques, was used to detect the crack, as the amount of crack's image data was limited. Pre-trained Inception-v3 was applied as a base model for the transfer learning. Web scrapping was utilized to fetch images of concrete wall with or without crack from web. In the recognition of crack, image post-process including changing size or removing color were applied. In the visualization of crack, source images divided into 30px, 50px or 100px size were used as input data, and different numbers of input data per category were applied for each case. With the results of visualized crack image, false positive and false negative errors were examined. Highest accuracy for the recognizing crack was achieved when the source images were adjusted into 224px size under gray-scale. In visualization, the result using 50 data per category under 100px interval size showed the smallest error. With regard to the false positive error, the best result was obtained using 400 data per category, and regarding to the false negative error, the case using 50 data per category showed the best result.

키워드

NGHHCI_2019_v61n3_55_f0001.png 이미지

Fig. 1 Example photo of cracked input data and non-cracked input data

NGHHCI_2019_v61n3_55_f0002.png 이미지

Fig. 2 Structure of Inception Model (Szegedy, 2015)

NGHHCI_2019_v61n3_55_f0003.png 이미지

Fig. 3 Examination data for crack visualization

Table 1 Examination data for crack recognition

NGHHCI_2019_v61n3_55_t0001.png 이미지

Table 2 Color by region of estimated probability from crack visualization

NGHHCI_2019_v61n3_55_t0002.png 이미지

Table 3 Accuracy of each model for each examination data

NGHHCI_2019_v61n3_55_t0003.png 이미지

Table 4 Result of crack visualization of 100px interval

NGHHCI_2019_v61n3_55_t0004.png 이미지

Table 5 Result of crack visualization of 50px interval

NGHHCI_2019_v61n3_55_t0005.png 이미지

Table 6 Result of crack visualization of 30px interval

NGHHCI_2019_v61n3_55_t0006.png 이미지

Table 7 Ratio of occurrence of false positive (FP) per estimation for interval, examination data and input data per category

NGHHCI_2019_v61n3_55_t0007.png 이미지

Table 8 Ratio of occurrence of false negative (FN) per estimation for interval, examination data and input data per category

NGHHCI_2019_v61n3_55_t0008.png 이미지

참고문헌

  1. Boeing, G., and P. Waddell, 2016. New insights into rental housing markets across the United States: web scraping and analyzing craiglist rental listings. Journal of Planning Education and Research 1-20. doi:10.1177/0739456X16664789.
  2. Byun, G. J., H. W. Song, W. Choi, and S. M. Wu, 2000. Crack evaluation technique and crack repair of concrete structures. Magazine of the Korea Concrete Institute 12(6): 97-108. https://doi.org/10.22636/MKCI.2000.12.6.97
  3. Chen, G., H. Mu, D. Pommerenke, and J. L. Drewniak, 2004. Damage detection of reinforced concrete beams with novel distributed crack/strain sensors. Structural Health Monitoring 3(3): 225-243. doi:10.1177/1475921704045625.
  4. Chilamkurthy, S., 2019. Transfer learning tutorial, https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html. Accessed 14 Apr. 2019.
  5. Devikar, P., 2016. Transfer learning for image classification of various dog breeds. International Journal of Adavanced Research in Computer Engineering & Technology 5(12): 2707-2715.
  6. Hassan, S. I., D. L. Minh, S. H. Im, G. B. Min, J. Y. Nam, and H. J. Moon, 2018. Damage detection and classification system for swer inspection using convolutional neural networks based on deep learning. Journal of the Korea Institute of Information and Communication Engineering 22(3): 451-457. doi:10.6109/JKIICE.2018.22.3.451.
  7. Kang, H. K., and S. G. Hong, 2008. Causes and mitigation of concrete cracking. Magazine of the Korea Concrete Institute 20(5): 61-68. https://doi.org/10.22636/MKCI.2008.20.5.61
  8. Kim, W., 1995. Major causes for deterioration in concrete structures. Magazine of the Korea Conrete Institute 7(6): 14-22.
  9. Krizhevsky, A., I. Sutskever, and G. E. Hinton, 2012. ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems 1097-1105. doi:10.1145/3065386.
  10. LeCun, Y., Y. Bengio, and G. Hinton, 2015. Deep learning. Nature 521(7553): 436-444. doi:10.1038/nature14539.
  11. Lee, B. Y., S. T. Yi, and J. K. Kim, 2007. Surface crack evaluation method in concrete structures. Journal of the Korean Society for Nondestructive Testing 27(2): 173-182.
  12. Lee, Y. I., B. H. Kim, and S. J. Cho, 2018. Image-based spalling detection of concrete structures using deep learning. Journal of the Korea Concrete Institue 30(1): 91-99. doi:10.4334/JKCI.2018.30.1.091.
  13. Pan, S. J., and Q. Yang, 2010. A Survey on transfer learning. IEEE Transactions on knowledge and data engineering 22(10): 1345-1359. doi:10.1109/TKDE.2009.191.
  14. Pour-Ghaz, M., A. M. Asce, T. Barret, T. Ley, N. Materer, A. Apblett, and J. Weiss, 2014. Wireless crack detection in concrete elements using conductive surface sensors and radio frequency identification technology. Journal of Materials in Civil Engineering 26(5): 923-929. doi: 10.1061/(ASCE)MT.1943-5533.0000891.
  15. Simard, P. Y., D. Steinkraus, and J. C. Platt, 2003. Best practices for convolutional neural networks applied to visual document analytsis. In Proceedings of the Seventh International Conference on Document Analysis and Recognition 2: 958-962. doi:10.1109/ICDAR.2003.1227801.
  16. Song, G., H. Gu, Y. L. Mo, T. T. C. Hsu, and H. Dhonde, 2007. Concrete structural health monitoring using embedded piezoceramic transducers. Smart Materials and Structures 16: 959-968. doi:10.1109/ICDAR.2003.1227801.
  17. Suaris W., and V. Fernando, 1987. Detection of crack growth in concrete from ultrasonic intensity measurements. Materials and Structures 20: 214-220. doi:10.1007/BF02472738.
  18. Szegedy, C., V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, 2016. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition 2818-2826. doi:10.1109/CVPR.2016.308.
  19. Szegedy, C., W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, 2015. Going deeper with convolutions. The IEEE Conference on Computer Vision and Pattern Recognition 1-9. doi:10.1109/CVPR.2015.7298594.
  20. Xia, X., C. Xu, and B. Nan, 2017. Inception-v3 for flower classification. Image, Vision and Computing, 2017 2nd International Conference on IEEE 783-787. doi:10.1109/ICIVC.2017.7984661.
  21. Yoon, W. H., 1991. About crack of concrete structure. Magazine of the Korea Concrete Institute 3(3): 66-74. https://doi.org/10.22636/MKCI.1991.3.3.66