GPR data noise attenuation on the curvelet transform

Abstract

Signal extraction is critical in GRP data processing and noise attenuation. When the target depth is shallow, its reflection echo signal will overlap with the background noise, affecting the detection of arrival time and localization of the target. Thus, we propose a noise attenuation method based on the curvelet transform. First, the original signal is transformed into the curvelet domain, and then the curvelet coefficients of the background noise are extracted according to the distribution features that differ from the effective signal. In the curvelet domain, the coarse-scale curvelet atom is isotropic. Hence, a two-dimensional directional filter is designed to estimate the high-energy background noise in the coarsescale domain, and then, attenuate the background noise and highlight the effective signal. In this process, we also use a subscale threshold value of the curvelet domain to filter out random noise. Finally, we compare the proposed method with the average elimination and 2D continuous wavelet transform methods. The results show that the proposed method not only removes the background noise but also eliminates the coherent interference and random noise. The numerical simulation and the real data application suggest and verify the feasibility and effectiveness of the proposed method.