Abstract: Based on the sparse characteristics the Compressed Sensing (CS) technique can achieve the sparse image and scattering structure of the target. In this paper the CS technique is applied in the Vehicle-mounted Stepped-frequency Forward-looking Ground-penetrating Virtual Aperture Radar (SFGPVAR) system as well. By introducing the electromagnetic model the landmine can approximate to a symmetrical cylinder with isotropic characteristic. Based on analyzing the results of simulations and experiments the metal landmine has the structure of two isolate point scatters. The distance between the two point scatters is independent of the incident angle of radar waves, but has the only relation with the physics dimension of landmine. This satisfies the sparseness required by the CS in the imaging space, and as a result of that the structure of landmine can be extracted by the CS algorithm. At the end of this paper the feasibility of the CS algorithm on the scattering structures extraction of the landmine and the stability of the landmine feature with two point scatters are proved by the experiment data from the SFGPVAR system. This novel approach not only extends a new way for the feature extraction of landmine, but also detection and classification for the targets with simple scattering structures.