Separating Target Response from Surface Feature Ringing with Eigenimage Filtering and Sparse Blind Deconvolution

Abstract:

Sharpening ground-penetrating radar (GPR) images remains a persistent challenge. Two approaches in particular—singular value decomposition (SVD) and sparse blind deconvolution (SBD)—have been respectively shown to denoise GPR images and resolve a reflectivity model effectively. This presentation highlights the treatment of a sample GPR profile acquired over two marked graves in sandy soil. Ringing from partially-buried surface features overprints a target of interest. We use SVD to remove the most dominant features from the profile-- the direct wave, horizontal banding, and the described ringing. For this dataset, the first five eigenimages from SVD capture these structures. Higher-order eigenimages, which are sometimes considered noise, cannot be discarded as they contain signal details critical for SBD. We show that the SBD algorithms benefit from SVD-filtering by returning a physically meaningful reflectivity model and point-like reflectors, respectively, not possible from the original data. SBD on SVD-filtered GPR data results in a reflectivity model that can better capture diffraction patterns produced by the burials as well as very shallow diffractions overprinted by the direct wave and ringing in the original data. The migration of SVD-filtered data provides more distinct and spatially constrained point reflectors regardless of the migration algorithms used. The SVD filter is applied to a series of profiles from a grid survey to reveal the lateral shape and extent of a target that would otherwise be obscured by ringing in several profiles.