Statistical and Computational Electromagnetic Approaches for the Feature Extraction of Buried Targets in Ground Penetrating Radar Imaging
Open Access
- Author:
- Idriss, Zacharie
- Graduate Program:
- Electrical Engineering (PHD)
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- October 05, 2021
- Committee Members:
- Soundar Kumara, Outside Unit & Field Member
Ram Narayanan, Chair & Dissertation Advisor
Tim Kane, Major Field Member
Raghu Raj, Special Member
Jesse Barlow, Major Field Member
Kultegin Aydin, Program Head/Chair - Keywords:
- Radar
Radar Imaging
Ground Penetrating Radar
Computational Electromagnetics
Feature Extraction
Waveform Design
Mutual Information
Matched Illumination
Compound Gaussian - Abstract:
- The feature extraction from buried targets is presented from both a statistical waveform design perspective, and a computational electromagnetic target in identification perspective. The statistical perspective provides a stochastic description of the radar scene using the compound Gaussian (CG) distribution. We explore novel waveform design techniques with respect to the Mutual Information (MI) criterion based on CG clutter. The first method presents a waveform that exploits the CG distribution of the scene reflectivity function when projected onto a sparse basis. This is compared to the second method, called the target-specific approach that uses knowledge of the target and clutter frequency response to optimize a matched illumination waveform. In both cases, the Taguchi and particle swarm optimization (PSO) solvers are employed for MI based waveform design optimization. To validate and compare the effectiveness of the optimized waveforms, the resulting scene reflectivity function is estimated using the sparsity-driven regularization radar imaging method. Our experimental results demonstrate that both waveform optimization techniques result in significantly better image reconstruction performance than the traditional LFM waveform; and that the target-specific approach additionally suppresses clutter information in the scene. Secondly, a wide-band direct modeling and feature extraction method for two dimensional geometric objects with arbitrary local coordinate rotations is considered. In particular, we focus on the problem of performing feature extraction for the case of multiple electrically small objects which is pertinent to ground penetrating radar (GPR) applications. We first use low frequency surface current density solutions for a class of simple two dimensional scatterers buried below a rough surface in a lossy half-space. Then for the inverse problem, a novel methodology is developed for pose and location invariant feature extraction derived from basis sets emerging from our proposed modeling framework. A method to deal with possible false targets due to the nonlinear multipath from multiple targets is developed. A noniterative algorithm based on the conjugate of Green's function is developed to solve for the surface current in an unknown domain using multi-frequency, multi-aperture data. Our modeling and feature extraction algorithms are numerically validated for different target shapes buried in lossy soil profiles.