An unsupervised classification method based on find of density peaks(FDP) is proposed for the polarimetric synthetic aperture radar (POLSAR) image. For the great impact of the boundary and strong points in the POLSAR image, the following density becomes unstable. The saliancy image which is based on the information entropy is proposed to remove these points before classification. The feature in H//A/SPAN space of the remaining pixels is weighted with the saliancy value. Then the unsupervised classification is achieved based on the FDP. In the experiment with the ESAR data, results validate the effectiveness of the new method.