A novel 3D clutter removal for GPR pipe detection via tensor RPCA

Abstract

Ground penetrating radar (GPR) is widely used in underground pipe detection due to its non-destructive and high efficiency characteristics. To improve the accuracy of data interpretation, 3D GPR survey is gradually applied to collect more information to classify pipes and other underground objects. However, its efficiency is often highly deteriorated by strong clutters, especially for nonmetal pipe detection. Recently, low rank and sparse decomposition (LRSD) based methods have been demonstrated their superiority to conventional methods in GPR clutter suppression. However, these methods are suitable for B-scan images, but not effective for 3D GPR data. In this paper, a novel clutter removal method based on Tensor RPCA is proposed for 3D GPR data. Similar to the RPCA, it decomposes the data matrix into a low-rank clutter matrix and a sparse target matrix, but a different cost function is utilized. It minimizes the third-order tensor nuclear norm to achieve global optimization of the clutter and limits the sparsity of the targets by the <i>l</i>-1 norm. Moreover, the randomized singular value decomposition is employed to reduce the computational complexity of the proposed algorithm. Simulation and experimental results show its outstanding performance in clutter removal for 3D data.