Signal enhancement and geometric information retrieval from 2D GPR data with multi-scale, orientation-sensitive filtering methods

摘要

Ground Probing Radar (GPR) is an almost indispen- sable means of imaging near-surface structures and enjoys a very diverse and broad range of applications. Two-dimensional GPR images of the subsurface fre- quently contain geometric information from small scatterers (diffraction hyperbolae) as well as dip-dependent informa- tion from dipping reflectors such as geological bedding, structural interfaces, cracks, fractures, joints, empty or filled cavities associated with jointing or faulting and other conceiv- able structural configuration. The second group of targets, especially fractures, are usually not good reflectors and are spatially localized; in geological, geotechnical and engineering applications their detection is frequently a primary objective. At the same time, GPR is notoriously susceptible to noise. An innumerable variety of natural and artificial buried objects can cause unwanted reflections and scattering. Anthropogenic noise is worse and includes reflections from nearby structures, interference from power lines and telecommunication devices etc. This type of noise is only partially countered with shielded antennae while the extraneous or reflected airwaves, critically refracted airwaves and groundwaves are not easily suppressed during acquisition. Finally, there's systemic noise, frequently manifested as ringing (antenna self-clutter). In many cases, the noise has definite geometrical characteristics (e.g., high-angle crossing clutter). Because the GPR source wavelet is tuned at a single frequency, the information returned by the subsurface structure is usually limited to a relatively narrow band around it and the rest of the spectrum is swamped in noise. Raw GPR data usually require post-acquisition processing, as they pro- vide only approximate target shapes and depths.