Vehicle- or air-borne Ultra-Wide Band Synthetic Aperture Radar (UWB SAR) can perform landmine detection over large areas, where too many false alarms is the major problem for UWB SAR application in practice. In this paper, landmine aspect-invariant feature extraction using the Space-Wavenumber Distribution (SWD) and its associated Support Vector Machine (SVM) with a hypersphere classification boundary are proposed in landmine detection procedure. The proposed methods have been tested by the real data collected with the Rail-GPSAR system..