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

[2P5-GS-10-04] Method for improving the identification accuracy of ground penetrating radar images using deep learning

Method for absorbing differences in ground penetrating radar models and underground conditions

〇Tomoyuki Kimoto¹, Ryo Shinno¹, Jun Sonoda² (1. National Institute of Technology, Oita College, 2. National Institute of Technology, Sendai College)

[[Online]]

Keywords:Ground penetrating radar, Social infrastructure inspection, FDTD method

In the inspection of social infrastructure, a ground penetrating radar is used to take an image of radio wave reflection in the ground, and a skilled technician makes a visual judgment, but it is difficult to perform risk factors such as cavities with high accuracy. The purpose of this research is to develop an high accuracy AI system for identifying whether or not it is a risk factor for underground objects from ground penetrating radar images. So far, we have developed a method to train AI systems by generating a large amount of radar images by the FDTD method. However, it is difficult to generate the correct image because the correct physical constants for performing the FDTD method are unknown. In this research, we propose a method to generate a large amount of high-precision radar images ,by a large amount of inaccurate radar images by the FDTD method and real radar images by a ground penetrating radar.