Deep learning based estimation of wall parameters for through-the-wall imaging

Abstract

This project explores the use of deep learning algorithms, particularly Convolutional Neural Networks (CNNs), to estimate wall parameters in Through-the-Wall Imaging (TWI). TWI utilizes Ground Penetrating Radar (GPR) to detect objects hidden behind walls, but distinguishing between the target and clutter from the wall can be quite challenging. Traditional signal processing techniques often struggle with complex wall structures, which results to detection inaccuracies. This project uses GPRMax software, an open-source radar simulation software, to create a dataset of synthetic B-scan images. CNN models are then developed using TensorFlow, which are trained on different datasets to learn the relationship between the B-scan data and simulation parameters. This project begins by estimating wall parameters and then expands to include the prediction of object parameters such as position and permittivity.