Aiming to remove the false alarm caused by azimuth ambiguity in SAR imagery during the process of ship target detection, a method based on the improved H/α/Wishart unsupervised classification is proposed. First, the scattering echo peak zone of target is extracted and the H/α classification results are treated as the initial cluster centers. Second, the ship target and azimuth ambiguity are identified by comparing and analyzing the structure of each region. It is showed by experiment that the method can detect ships based on the azimuth ambiguity; thus the false alarm rate in SAR-based ship detection is reduced.