In order to better preserve the structural information and polarization scattering information while suppressing speckle, an adaptive non-local mean filtering algorithm based on coefficient of variance (C.V) is proposed. The algorithm combines the statistical properties of the image sub-blocks with the polarization-scattering-characteristics of the target points to filter the homogeneous pixels, then introduces the C.V adaptive selection smoothing coefficients to calculate the weights used for filtering, and finally performs non-local mean filtering on the homogeneous pixels. The experimental results of polarimetric synthetic aperture radar data collected by different systems show that compared with the refined LEE filter, NL-Pretest filter and the latest filter, the proposed algorithm can effectively remove the speckle noise, and the edge and polarization scattering characteristics of the image are better maintained.