| Presentation | 2020-11-25 Generation of learning images corresponding to differences in ground penetrating radar and underground media for highly accurate identification of ground penetrating radar images with AI Daiki Taga, Tomoyuki Kimoto, Jun Sonoda, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Ground penetrating radar is a technology that detects underground objects by utilizing the reflection of radio waves incident on the ground in places where there is a difference in permittivity, and It is effective for non-destructive sensing of deterioration status such as cavities in social infrastructure. In recent years, by training a large number of radar images in which various types of buried objects generated by physical simulation with FDTD method on the convolutional neural network (CNN), it has become clear that not only the detection of buried objects but identification of the permittivity and size are possible. However, since the frequency and waveform of radio waves emitted by each product of ground penetrating radar are different, it is necessary to regenerate a large number of radar images with the FDTD method for each product. Since the FDTD method requires extremely large computing power, so it is not realistic to regenerate a large number of radar images. In this study, we report that if a large amount of FDTD radar images for one product is prepared in advance, a large amount of pseudo FDTD radar images for the new product can be easily generated using machine learning by preparing only a small amount of FDTD radar images for new product. We also report that we have confirmed that this pseudo FDTD radar image is a valid generated image by a discrimination rate survey by CNN. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Ground penetrating radar / Machine learning / Deep Learning / Image generating / FDTD method |
| Paper # | SANE2020-28 |
| Date of Issue | 2020-11-18 (SANE) |
| Conference Information | |
| Committee | SANE |
|---|---|
| Conference Date | 2020/11/25(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Radar, Remote Sensing and general issues |
| Chair | Akitsugu Nadai(NICT) |
| Vice Chair | Hiroyoshi Yamada(Niigata Univ.) / Makoto Tanaka(Tokai Univ.) |
| Secretary | Hiroyoshi Yamada(Mitsubishi Electric) / Makoto Tanaka(Univ. of Tokyo) |
| Assistant | Shunichi Futatsumori(ENRI) |
| Paper Information | |
| Registration To | Technical Committee on Space, Aeronautical and Navigational Electronics |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Generation of learning images corresponding to differences in ground penetrating radar and underground media for highly accurate identification of ground penetrating radar images with AI |
| Sub Title (in English) | |
| Keyword(1) | Ground penetrating radar |
| Keyword(2) | Machine learning |
| Keyword(3) | Deep Learning |
| Keyword(4) | Image generating |
| Keyword(5) | FDTD method |
| 1st Author's Name | Daiki Taga |
| 1st Author's Affiliation | National Institute of Technology, Oita College(NIT, Oita) |
| 2nd Author's Name | Tomoyuki Kimoto |
| 2nd Author's Affiliation | National Institute of Technology, Oita College(NIT, Oita) |
| 3rd Author's Name | Jun Sonoda |
| 3rd Author's Affiliation | National Institute of Technology, Sendai College(NIT, Sendai) |
| Date | 2020-11-25 |
| Paper # | SANE2020-28 |
| Volume (vol) | vol.120 |
| Number (no) | SANE-250 |
| Page | pp.pp.7-12(SANE), |
| #Pages | 6 |
| Date of Issue | 2020-11-18 (SANE) |