Application of the Curvelet Transform for Clutter and Noise Removal in GPR Data

Abstract

This paper presents a new method based on the curvelet transform to improve the readability of ground penetrating radar (GPR) data during localization works of buried pipes. The localization is achieved by the detection of hyperbolas in the GPR cross section, also called B-scan. However, GPR antenna collects abundant information from the ground, which can partially hide useful information, especially echoes from the pipes (i.e., hyperbolas). In order to highlight these echoes, it is important to reduce noise and to remove undesirable items such as the clutter and column artifacts. For this purpose, we propose to use the curvelet transform in a three processing step method: clutter, noise, and column artifact removal. Thanks to prior information on the distribution of the coefficients and on the orientation of the clutter and artifacts, we are able to reduce them in the B-scan. These steps can be used independently giving more flexibility to the operator. Moreover, the processing is fast to compute and requires only one easily tunable parameter. This is appropriate for a computation in real-time by nonexpert operators. Our method has been applied to real data acquired on a test area and under real conditions. We show the proposed method presents satisfying qualitative and quantitative results compared to other methods.