Practical 3-D Migration and Visualization for Accurate Imaging of Complex Geometries with GPR

Abstract

Obtaining high-quality three-dimensional images that accurately represent target geometries is crucial for reliable interpretation of ground penetrating radar data. Dense data acquisition, three-dimensional migration processing, and visualization are needed to recognize complex target geometries. A novel implementation of the diffraction summation migration algorithm efficiently images three-dimensional geometries by utilizing the efficient properties of fast Fourier transforms to compute diffraction summations. The algorithm decomposes the diffraction summation calculations into spatial and temporal summations. The spatial summations are computed by convolving the data with two-dimensional convolution kernels computed from the three-dimensional point spread functions. Computation times are reduced by performing convolutions in the wavenumber domain using two-dimensional fast Fourier transforms. The migration aperture is easily adjusted in the algorithm to increase computational speeds and reduce processing artifacts associated with large amplitude spikes. Microsoft Visual Studio.NET and C# are used to create efficient and low-cost applications optimally designed for the processing and display of ground penetrating radar data. Migration algorithms are combined with volume rendering and visualization algorithms that utilize DirectX to visually recognize complex geometries in three-dimensions. A gravel test bed containing buried spheres, letters cut from metal sheets, pipes, and drums is imaged with dense and orthogonal 400 and [Formula: see text] ground penetrating radar surveys. A linear array composed of fourteen variably spaced spheres is used to empirically determine the resolution capabilities of commercial ground penetrating radar systems in favorable soils. In addition, letters cut from aluminum sheets are used to test imaging capabilities similar to those used to test human vision. Observed quarter-wavelength spatial resolutions are consistent with theoretical predictions. While horizontal slices work well for displaying horizontal objects, volume rendering is more effective for visualizing dipping objects. Increasing antenna frequency improves spatial resolution and produces less pronounced polarization differences between data acquired with orthogonal surveys.