Deep learning for ground penetrating radar image processing

Abstract

Ground Penetrating Radar (GPR) is a useful technique that uses radar pulses to image the subsurface. It is a non-intrusive method of surveying the sub-surface to detect underground utilities such as pipes, cables, etc. The GPR images usually come in three variations, either as an A-scan, B-scan, or C-scan images. Firstly, this paper will discuss how GPR can be used for detecting tree roots underground and discuss how factors like permittivity will affect the overall B-scan image. Secondly, this paper will also talk about an open-source forward-based solver software called gprMax that simulates electromagnetic (EM) wave propagation. It solves Maxwell’s equations in three dimensions (3D) using the Finite-Difference Time-Domain (FDTD) method. It was designed for modelling GPR applications, but it can be also used to model many other electromagnetic wave propagation applications. Thirdly, this paper will also discuss how Deep Learning Techniques can be used to create a surrogate Deep Neural Network (DNN) model for forward modelling of GPR images to solve a problem that National Parks Board (NParks) are currently facing.