KSCE Journal of Civil and Environmental Engineering Research (대한토목학회논문집)

Korean Society of Civil Engeneers (대한토목학회)
권호 표지
DOI QR코드

DOI QR Code

A Study on the Optimal Convolution Neural Network Backbone for Sinkhole Feature Extraction of GPR B-scan Grayscale Images

GPR B-scan 회색조 이미지의 싱크홀 특성추출 최적 컨볼루션 신경망 백본 연구

  • Park, Younghoon (Bucheon University)
  • 박영훈 (부천대학교 토목공학과)
  • Received : 2024.01.02
  • Accepted : 2024.02.25
  • Published : 2024.06.01
Download PDF
⟨ Previous Next ⟩

Abstract

To enhance the accuracy of sinkhole detection using GPR, this study derived a convolutional neural network that can optimally extract sinkhole characteristics from GPR B-scan grayscale images. The pre-trained convolutional neural network is evaluated to be more than twice as effective as the vanilla convolutional neural network. In pre-trained convolutional neural networks, fast feature extraction is found to cause less overfitting than feature extraction. It is analyzed that the top-1 verification accuracy and computation time are different depending on the type of architecture and simulation conditions. Among the pre-trained convolutional neural networks, InceptionV3 are evaluated as most robust for sinkhole detection in GPR B-scan grayscale images. When considering both top-1 verification accuracy and architecture efficiency index, VGG19 and VGG16 are analyzed to have high efficiency as the backbone for extracting sinkhole feature from GPR B-scan grayscale images. MobileNetV3-Large backbone is found to be suitable when mounted on GPR equipment to extract sinkhole feature in real time.

GPR을 활용한 싱크홀 감지 정확도 강화를 위하여 본 연구에서는 GPR B-scan 회색조 이미지의 싱크홀 특성을 최적으로 추출할 수 있는 컨볼루션 신경망을 도출하였다. 사전 훈련된 컨볼루션 신경망이 바닐라 컨볼루션 신경망보다 2배 이상의 효용성을 가지는 것으로 평가되었다. 사전 훈련된 컨볼루션 신경망에 있어서 빠른 특성 추출이 특성 추출보다 낮은 과대적합을 발생시키는 것으로 나타났다. 아키텍처 종류와 시뮬레이션 조건에 따라 top-1 검증 정확도 크기와 발생 조건 및 연산 시간이 상이한 것으로 분석되어, 사전 훈련된 컨볼루션 신경망 중 InceptionV3가 GPR B-scan 회색조 이미지의 싱크홀 감지에 가장 강건한 것으로 평가되었다. Top-1 검증 정확도와 아키텍처 효율 지수를 동시에 고려할 경우 VGG19와 VGG16가 GPR B-scan 회색조 이미지의 싱크홀 특성 추출 백본으로 높은 효율성을 가지는 것으로 분석되었으며, GPR 장비에 탑재하여 실시간으로 싱크홀 특성 추출을 할 경우에는 MobileNetV3-Large 백본이 적합한 것으로 나타났다.

Keywords

References

  1. Al-Nuaimy, W., Huang, Y., Nakhkash, M., Fang, M. T. C., Nguyen, V. T. and Eriksen, A. (2000). "Automatic detection of buried utilities and solid objects with GPR using neural networks and pattern recognition." Journal of Applied Geophysics, Vol. 43, pp. 157-165, https://doi.org/10.1016/S0926-9851(99)00055-5.
  2. Annan, A. (1999). "Practical processing of GPR data." Mississauga, ON, Canada, Sensors & Software.
  3. Bailia, J., Lahouara, S., Herglia, M., Al-Qadi, I. L. and Besbes, K. (2009). "GPR signal denoising by discrete wavelet transform." NDT & E International, Vol. 42, No. 8, pp. 696-703.
  4. Benedetto, A., Tosti, F., Ciampoli, L. and D'Amico, F. (2017). "An overview of ground-penetrating radar signal processing techniques for road inspections." Signal Process, pp. 201-209.
  5. Gamba, P. (2000). "Neural detection of pipe signatures in ground penetrating radar images." IEEE Trans Geosci Remote Sens, pp. 790-797.
  6. Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang W., Weyand, T., Andreetto, M. and Adam, H. (2017). "MobileNets : Efficient convolutional neural networks for mobile vision applications." arXiv preprint, https://arXiv.org/abs/1704.04861v1.
  7. Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., Le, Q. V. and Adam, H. (2019). "Searching for MobileNetV3." arXiv preprint, https://arXiv.org/abs/1905.02244v5.
  8. Kang, M. S., Kim, N., Lee, J. J. and An, Y. K. (2019). "Deep learning-based automated underground cavity detection using three-dimensional ground penetrating radar." Structural Health Monitoring, Vol. 19, No. 1, pp. 173-185, https://doi.org/10.1177/1475921719838081.
  9. Kingma, D. P. and Ba, J. (2014). "ADAM: A method for stochastic optimization." Proceedings of 3rd International Conference for Learning Representations, San Diego, USA, 2015, arXiv preprint, https://arxiv.org/abs/1412.6980.
  10. Kurbiel, T. and Khaleghian, S. (2017)."Training of deep neural networks based on distance measures using RMSProp." arXiv preprint, https://arxiv.org/abs/1708.01911.
  11. Qin, Y. and Huang, C. (2016). "Identifying underground voids using a GPR circular-end bow-tie antenna system based on a support vector machine." International Journal of Remote Sensing, pp. 876-888, https://doi.org/10.1080/01431161.2015.1137990.
  12. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. and Salakhutdinov, R. (2014). "Dropout : A simple way to prevent neural networks from overfitting." Journal of Machine Learning Research, Vol. 15, pp. 1929-1958.
  13. Tzanis, A. (2013). "Detection and extraction of orientation-andscale-dependent information from two-dimensional GPR data with tuneable directional wavelet filters." Journal Applied Geophysics, Vol. 89, pp. 48-67, https://doi.org/10.1016/j.jappgeo.2012.11.007.
  14. Zhang, Y., Huston, D. and Xia, T. (2016). "Underground object characterization based on neural networks for ground penetrating radar data." Proceedings of the International Society for Optics and Photonics, pp. 1-9.