Simulated Data Based Deep Neural Network Training for Underground Cavity Detection 


Vol. 46,  No. 5, pp. 923-927, May  2021
10.7840/kics.2021.46.5.923


PDF
  Abstract

Recently, underground cavities on urban roads have prompted safety concerns. To provide countermeasures, early detection and identification of cavities are essential. Ground-penetrating radar(GPR) is often used to detect cavity by transmitting electromagnetic pulses and receiving the backscattered radiation from subsurface discontinuities. Collecting data in real-world environment and determining whether cavity exists requires much manpower and time, and there is limitation due to the lack of numbers of cavity data. In this study, to minimize the manpower and save time, we simulate cavity data in place of real-world data. We propose a method of determining the existence of underground cavities by training deep neural network(DNN) model, VGGNet-16, with GPR images produced by simulation. Through simulation, GPR images with and without a cavity were created and used for training. Experimental result showed 92.3% test accuracy in classifying soils with and without cavity models. To ensure the trained model learned cavity features effectively, we use Score-CAM to visualize the model’s representation learning mechanism. Through visualization, we validated that the model learned features that indicate the existence of cavity. In future work, we will verify the similarity between simulated and real-world data and performance of the trained model using real-world data.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

Y. Yoo, D. Kim, M. Lee, J. Lee, "Simulated Data Based Deep Neural Network Training for Underground Cavity Detection," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 5, pp. 923-927, 2021. DOI: 10.7840/kics.2021.46.5.923.

[ACM Style]

Youngjun Yoo, Daehee Kim, Myunghak Lee, and Jaekoo Lee. 2021. Simulated Data Based Deep Neural Network Training for Underground Cavity Detection. The Journal of Korean Institute of Communications and Information Sciences, 46, 5, (2021), 923-927. DOI: 10.7840/kics.2021.46.5.923.

[KICS Style]

Youngjun Yoo, Daehee Kim, Myunghak Lee, Jaekoo Lee, "Simulated Data Based Deep Neural Network Training for Underground Cavity Detection," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 5, pp. 923-927, 5. 2021. (https://doi.org/10.7840/kics.2021.46.5.923)

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)






pISSN : 1226 - 4717
eISSN : 2287 - 3880