Clutter Removal in Ground Penetrating Radar with Learned Dictionaries

Since the clutter deteriorates the performance of detection algorithms, removal of the clutter before any detection process is crucial in Ground Penetrating Radar (GPR) systems. In this paper, we propose to separate the GPR data into its clutter and target components by using learned dictionaries. Each patch extracted from the GPR data is decomposed using Orthogonal Matching Pursuit (OMP), then the obtained target patches are merged to form the target data. Detection results provided by the proposed method and the comparison methods Singular Value Decomposition (SVD), Principal Component Analysis (PCA), Robust Principal Component Analysis (RPCA), Nonnegative Matrix Factorization (NMF), Robust Nonnegative Matrix Factorization (RNMF) and traditional Morphological Component Analysis (MCA) for a new dataset containing challenging scenarios demonstrate the superiority of the use of learned dictionaries for clutter removal. Besides, proposed method is faster than traditional MCA method.