Abstract: The Compressed Sensing (CS) technique is an effective approach for sparse imaging and extraction of scattering structure of targets, which can be applied to target discrimination and recognition. Based on the experimental data from the Forward Looking Virtual Aperture Radar (FLVAR) system, the scattering structures of landmines can be acquired by CS sparse imaging algorithm. Then the sparse scattering structures are parameterized to form the features exploited by classifiers later. In this paper, a novel approach to target discrimination is proposed, which transforms the scattering of landmines to geometrical features, which have strong relationship with its physical characteristics. This new approach not only broadens the methodology for landmine discrimination, but also explores a new way of applying sparse scattering structures to target discrimination.