Combining clutter learning with LS for improved buried target detection in GPR

The classical least squares (LS) filtering method is combined with the subspace-based methods for buried target detection in ground penetrating radar (GPR) images. The LS method is used to estimate the next A-scans from previously observed A-scans which are assumed to belong to clutter samples. Generally, A-scans used in the initial step are accepted as clutter for the LS to work correctly. However, this is not guaranteed and if the first observed A-scan samples contain any target information, LS method will fail. To avoid target component presence in previously observed A-scans, the pre-processing step is integrated to keep only the clutter information. This step is based on obtaining clutter information from GPR image by using subspace-based methods. Various subspace-based methods are used to validate the efficiency of the proposed pre-processing step compared to the classical LS method. This additional preprocessing step does not bring any computational burden and is appropriate for real-time target detection.