Features for detecting obscured objects in ultra-wideband (UWB) SAR imagery using a phenomenological approach

Abstract

We describe a technique that is based on phenomenological principles for detecting man-made objects in ultra-wideband (UWB) synthetic aperture radar (SAR) imagery. The UWB sensor (50 MHz to a GHz) provides penetration and high resolution necessary for detecting man-made objects that are obscured by some random media. Traditional 2-D pattern matching techniques are not effective since the SAR image of an object is distorted by the obscuring media. An electromagnetic model is used to predict the backscatter received from the scene objects. We show that the backscatter is highly dependent on the aspect angle of the incident wave on the object. Man-made objects exhibit specular reflections, at certain aspect angles while natural objects generally do not. Aspect-angle signatures of objects are established using a multi-aperture approach which essentially reconstructs the SAR image of an object over smaller subapertures of the full synthetic aperture. These signatures are matched to the object's polarimetric counterparts to establish feature vectors, which are used for detection. This procedure is applied to detection of man-made vehicles that are embedded in a deciduous forest and obscured by foliage. Results are presented using real data.