[4K2-J-13] AI application: land and infrastructure

[4K2-J-13-03] Classification of dielectric constant and size of underground radar image using 3D-CNN

〇Tomoyuki Kimoto¹, Ryo Tsuno¹, Jun Sonoda² (1. National Institute of Technology, Oita College, 2. National Institute of Technology, Sendai College)

Keywords:ground penetrating radar, 3D-CNN, classification of underground objects

In this study, to classify the underground objects from the ground penetrating radar (GPR) images by the three dimensional convolutional neural network (3D-CNN). Conservation management of roads is important for maintaining infrastructure, and in order to explore the whole road width at once, it is frequently performed to arrange a plurality of GPR in parallel and sweep. In this study, 3D-CNN is used to simultaneously process this plurality of GPR images and improve the classification performance. As a result, it reports that the classification performance improves for the dielectric constant and the size of buried objects.