A New Clutter Removal Method Based on Direct Robust Matrix Factorization for Buried Target Detection

Clutter decreases severely the performance of target detection algorithms in ground-penetrating radar (GPR) imaging systems. Low rank and sparse decomposition (LRSD) methods divide the data into its clutter and target components by rank minimization with sparsity constraint. This paper proposes a direct solution for LRSD decomposition of the GPR data unlike robust principal component analysis (RPCA) which uses a nuclear norm relaxation. The non convex optimization problem is solved by successive partial singular value decompositions (SVD)s and soft thresholding operations and does not require any parameter computation. The visual and numerical comparisons for both simulated and real data show the superiority of the direct robust matrix factorization (DRMF) over the relaxation solution RPCA as well as over the traditional low rank methods SVD and PCA.