Buried object characterization by data-driven surrogates and regression-enabled hyperbolic signature extraction.

Area of Science:

• Geophysics
• Artificial Intelligence
• Electromagnetics

Background:

• Ground Penetrating Radar (GPR) is crucial for subsurface exploration.
• Characterizing buried objects using GPR data, especially their size and position, is computationally intensive.
• Current methods struggle with simultaneous estimation of multiple geophysical parameters.

Purpose of the Study:

• To develop an efficient artificial intelligence (AI)-based methodology for characterizing buried cylindrical objects.
• To estimate geophysical parameters including depth, lateral position, and radius simultaneously.
• To reduce the computational cost associated with GPR data processing.

Main Methods:

• Utilized Finite-Difference Time-Domain (FDTD) simulations with gprMax for generating GPR B-scan data.
• Developed a data-driven surrogate model using a modified multilayer perceptron (M2LP) deep learning framework.
• Extracted 1D hyperbolic signatures from 2D B-scan data via linear regression to reduce dimensionality.

Main Results:

• The M2LP framework achieved an average mean absolute error of 10 mm and an 8% average relative error in object characterization.
• The methodology demonstrated superior performance compared to Multilayer Perceptron (MLP), Support Vector Regression Machine (SVRM), and Convolutional Neural Network (CNN).
• The approach was validated using noisy data and real-world measurement data, confirming its robustness and applicability.

Conclusions:

• The proposed AI-based surrogate modeling approach offers a computationally efficient and accurate method for buried object characterization using GPR.
• The extraction of hyperbolic signatures provides a robust link between GPR data and geophysical parameters.
• The M2LP framework shows significant potential for real-time GPR data analysis and interpretation.