Presentation 2020-01-30
Embedded object identification from ground penetrating radar image by semi-supervised learning using variational auto-encoder
Tomoyuki Kimoto, Jun Sonoda,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Recently, deterioration of social infrastructures such as tunnels and bridges becomes serious social problem. It is required to rapidly and accurately detect for abnormal parts of the social infrastructures. The ground penetrating radar (GPR) is e?cient for the social infrastructure inspection, but, it is di?cult to identify the material and size of the underground object from the radar image obtained the GPR. In our previous studies, we have massively generated the GPR images by a fast ?nite-di?erence time-domain (FDTD) simulation, we make learned the generated GPR images to the convolution neural network (CNN). As the results, it has been clarified that the relative permittivity and size of the object can be identified from the underground radar images. However, when using real images of construction sites, correct labels such as the relative permittivity of buried objects can only be examined by digging the ground, and supervised learning is not practical. In this study, we use the unsupervised learning of radar images using variational auto encoders (VAE), and mapping high-dimensional information of radar images to latent space. And we report that the identification rate is improved, by semi-supervised learning to give correct labels to some of these latent variables.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Grand Penetrating Rader / Deep Learning / Variational Auto Encoder / Semi-Supervised Leraning
Paper # EST2019-80
Date of Issue 2020-01-23 (EST)

Conference Information
Committee EST
Conference Date 2020/1/30(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Beppu International Convention Center
Topics (in Japanese) (See Japanese page)
Topics (in English) Simulation Technique, etc.
Chair Akimasa Hirata(Nagoya Inst. of Tech.)
Vice Chair Shinichiro Ohnuki(Nihon Univ.) / Masayuki Kimishima(Advantest) / Jun Shibayama(Hosei Univ.)
Secretary Shinichiro Ohnuki(National Inst. of Tech.,Sendai College) / Masayuki Kimishima(Aoyama Gakuin Univ.) / Jun Shibayama
Assistant Takahiro Ito(Nagoya Inst. of Tech.) / Kazuhiro Fujita(Fujitsu)

Paper Information
Registration To Technical Committee on Electronics Simulation Technology
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Embedded object identification from ground penetrating radar image by semi-supervised learning using variational auto-encoder
Sub Title (in English)
Keyword(1) Grand Penetrating Rader
Keyword(2) Deep Learning
Keyword(3) Variational Auto Encoder
Keyword(4) Semi-Supervised Leraning
1st Author's Name Tomoyuki Kimoto
1st Author's Affiliation National Institute of Technology, Oita College(NIT, Oita)
2nd Author's Name Jun Sonoda
2nd Author's Affiliation National Institute of Technology, Sendai College(NIT, Sendai)
Date 2020-01-30
Paper # EST2019-80
Volume (vol) vol.119
Number (no) EST-407
Page pp.pp.7-12(EST),
#Pages 6
Date of Issue 2020-01-23 (EST)