Enhancing Parameters Estimation in Subsurface Imaging with Ground Penetrating Radars

Jyoti Khandelwal, Deepak Pareek

2024

Abstract

Ground Penetrating Radar (GPR) images cannot be directly inverted to obtain subsurface source images, making it a challenging computational problem. This work investigates two fundamentally different approaches to solving the GPR inverse problem: a Hamiltonian Monte Carlo (HMC) method and a U-net neural network. HMC, a Markov Chain Monte Carlo technique, evolves the system state to minimize the objective function, while U-net leverages convolutional neural networks for regression to predict pixel-wise reflectivity values. Extensive experiments on simulated GPR data with varying numbers of sources reveal that HMC’s performance, measured by the Structural Similarity Index (SSI), deteriorates as more sources are introduced. In contrast, the U-net model demonstrates remarkable robustness, maintaining high SSI scores even with numerous sources present. The results indicate that in scenarios with abundant GPR training data and complex source distributions, a U-net model is preferable over HMC due to its superior generalization capability and efficient parallel processing. However, HMC offers interpretability advantages by allowing statistical analysis of individual pixels. This work highlights the trade-offs between the two approaches and provides insights for selecting the appropriate method based on the problem’s complexity and data availability.

Paper Citation

in Harvard Style

Khandelwal J. and Pareek D. (2024). Enhancing Parameters Estimation in Subsurface Imaging with Ground Penetrating Radars. In Proceedings of the 1st International Conference on Cognitive & Cloud Computing - Volume 1: IC3Com; ISBN 978-989-758-739-9, SciTePress, pages 154-160. DOI: 10.5220/0013263400004646

in Bibtex Style

@conference{ic3com24,
author={Jyoti Khandelwal and Deepak Pareek},
title={Enhancing Parameters Estimation in Subsurface Imaging with Ground Penetrating Radars},
booktitle={Proceedings of the 1st International Conference on Cognitive & Cloud Computing - Volume 1: IC3Com},
year={2024},
pages={154-160},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013263400004646},
isbn={978-989-758-739-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Cognitive & Cloud Computing - Volume 1: IC3Com
TI - Enhancing Parameters Estimation in Subsurface Imaging with Ground Penetrating Radars
SN - 978-989-758-739-9
AU - Khandelwal J.
AU - Pareek D.
PY - 2024
SP - 154
EP - 160
DO - 10.5220/0013263400004646
PB - SciTePress