GPR antipersonnel mine detection: improved deconvolution and time-frequency feature extraction

This work deals with the processing of GPR (ground penetrating radar) signals for AP (anti-personnel) mine detection. It focuses on two steps in this processing, namely the deconvolution of the system impulse response, and the extraction of target features for classification. The objective of the work is to find discriminant and robust target features by means of time-frequency analysis. Deconvolution is an ill-posed inverse problem, which can be solved with regularization methods. In this paper a deconvolution algorithm, based on the iterative v-method, is proposed. For discriminant feature selection the Wigner distribution (WD) is considered. Singular value decomposition (SVD) along with the concept of the center of mass as the most robust feature are used for feature extraction from the WD. The proposed normalized time-frequency-energetic features have a good discriminant power, which doesn't degrade with increasing object depth.