Identifying underground voids using a GPR circular-end bow-tie antenna system based on a support vector machine

Abstract

ABSTRACTThis article presents a pattern recognition approach based on discrete cosine transform (DCT) and polarization attributes to identify voids underground in ground penetrating radar (GPR) images. Features vectors composed of DCT coefficients supplying to the support vector machine classifier are studied, together with their linearity. To evaluate the proposed method, we hold an experimental study of GPR data obtained by numerical modelling of a GPR circular-end bow-tie antenna system using the finite-integration technique based simulator. In order to form the database of features, different acquisition scenarios of GPR models are achieved by varying buried object’s material type, position, shape, size parameters, and electric characteristics of subsurface medium. The robustness of the proposed system towards degradation of the generated GPR data is tested by introducing Gaussian and speckle noise. The experimental results show the proposed approach exhibits encouraging performance in terms of void identification with errors of object apex position in GPR images.