Validation of a method for estimating the extent of segregation on bridge slab using ground penetrating radar with deep learning

Abstract

In this paper, we proposed the application of the 3D Convolution Neural Network (3D-CNN) on the Ground Penetrating Radar (GPR) data to detect segregation on deck slab on two real bridges. We set this up as a classification problem whether the damage exists in a certain area. First, high accuracy of Area Under the Curve (AUC) was obtained when we trained and evaluated the models on each single bridge. Second, it decreased on the models which we trained with one bridge and evaluated with another. It shows that the features of radar data required to estimate are different on the bridges. Finally it was high on the models which we trained and evaluated on the both, and it shows a possibility that the generalization performance can be acquired by collecting radar data on various bridges.