Azimuth Ambiguity Suppression in SAR Images Based on VS-KSVD Dictionary Learning and Compressive Sensing

The azimuth ambiguities appear widely in Synthetic Aperture Radar (SAR) images, which causes a large number of false targets and seriously affect the quality of image interpretation. Due to under-sampling in Doppler domain, ambiguous energy is mixed with energy from the main zone in the time and frequency domains. In order to effectively suppress the ambiguous energy in SAR images without loss of resolution, this paper presents a novel method of KSVD dictionary learning based on variance statistics (VS-KSVD) and compressed sensing (CS) reconstruction. According to the statistical characteristics of distributed targets, the dictionary that represents the ambiguities is selected and suppressed by coefficient weighting, in which local window filtering is carried out to remove the block effect and optimize the edge information. Finally, the high resolution images with low-ambiguity can be reconstructed by CS. With the proposed approach, the feasibility and effectiveness of the proposed approach is validated by using satellite data and simulation in suppressing azimuth ambiguity.