Comprehensive GPR Signal Analysis via Descriptive Statistics and Machine Learning

Document Type

Conference Proceeding

Publication Date



Department of Geological and Mining Engineering and Sciences; Michigan Tech Research Institute


This paper presents a comprehensive analysis of how various soil characteristics impact the features of Ground Penetrating Radar (GPR) received signals. These characteristics include dielectric properties, thickness, number of layers, radar configuration, and surface roughness. The paper conducts an exhaustive analysis using gprMax, simulating diverse soil medium scenarios to demonstrate how these parameters influence the GPR-received signals. The proposed methodology extracts critical features from the received signal for soil characterization through descriptive statistical analysis. The paper then deploys Machine Learning (ML) techniques, specifically a Random Forest (RF) model and Gini Mean Decrease Impurity (MDI) as measures, to identify the most influential features in the dataset. This process extracts a concise set of features from the time domain, followed by an expansion using frequency domain features. The proposed approach not only effectively captures the critical information in the high-dimensional GPR data but also reduces its dimensionality, ensuring the preservation of essential information. Training ML and Deep Learning (DL) models using these significant features, rather than complex raw A-scan data, leads to more accurate soil moisture and subsurface analysis.

Publisher's Statement

© Copyright 2024 IEEE - All rights reserved, including rights for text and data mining and training of artificial intelligence and similar technologies.

Publication Title

2023 IEEE International Conference on Wireless Space and Extreme Environments (WiSEE)