[2P5-GS-10] AI application: detection / identification

[2P5-GS-10-03] Underground object recognition using interpolated GPR images

Kaito Tokushige¹, 〇Yasushi Kanazawa¹, Jun Sonoda², Tomoyuki Kimoto³ (1. Toyohashi University of Technology, 2. National Institute of Technology, Sendai College, 3. National Institute of Technology, Oita College)

[[Online]]

Keywords:GPR image, Image interpolation, Underground object recognition

We propse a method for recognising an underground object using interpolated ground penetrating radar images. A ground penetrating radar can measure the condition of the underground directly but only see below a survey line. So, for dense measurements, it is necessary to set many survey lines in the area. In this paper, we first generate intermediate images between sparse survey lines by GAN-based network, and then recognise the size and position of the underground object using CNN-LSTM-based network. The effectiveness of our method was confirmed by experiments using simulated images.