Buried mine classification from three-dimensional radar data

Ground penetrating radar has become an important tool in anti-personnel mine detection. In general, a buried object can be detected by scanning the ground surface with the radar probe. An object whose size is comparable with a typical wavelength (0.1 m @ 1 GHz) gives rise to a complex reflection pattern consisting of a superposition of inverted hyperbolic arcs in three dimensions (3-D). Previous papers have described the codes which use the gradient information in the scattering data to replace each overlapping arc by an 'apex point' describing the 3-D position, amplitude and sign of each arc. This paper introduces the concept of a 'pendant': a series of reflections whose apexes lie close to a vertical column which may be identified as arising from a single object. The amplitudes of these reflections may be used as features for object classification. The method is demonstrated using radar images from a buried isolated mine and also in proximity with other objects. The aim of this classification scheme is the identification of one object (mine) type against more 'point-like' signals from 'clutter' objects which are small compared to the wavelength.