相干斑抑制是极化合成孔径雷达（PolSAR）图像分析的重要预处理步骤。为了更好地抑制极化SAR图像中的相干斑，本文综合目标的异质性和结构信息，提出基于目标异质性的非局部均值滤波方法。首先利用K分布距离度量目标的异质性，并以异质性为基础，保留图像中的点、线等高异质性目标；然后计算图像块之间的异质性差异，最后将其作为度量非局部均值加权滤波像元相似性的权重系数，实现对PolSAR图像的相干斑抑制。实验对比结果表明：本文方法能够有效地抑制相干斑，同时对细节信息和极化信息也具有良好的保持性，能够为后续的图像应用提供支持。

Polarimetric Synthetic Aperture Radar (PolSAR) occupies an important place in remote sensing because it provides richer information about the targets and earth surface compared with single-channel SAR systems. However, PolSAR data is contaminated by speckle noise due to the coherent imaging mechanism, which considerably affects the accuracy of target classification and recognition. Therefore, speckle-noise filtering of PolSAR images is a crucial pretreatment. Nonlocal(NL) means compute the weights between two pixels with similar surrounding neighborhoods (known as patches) instead of two individual pixels. Considering that patches contain structural information, the NL mean filter preserves repetitive structures and performs better than other filters. The key point of the NL algorithm is the similarity criterion setting or the patch weights. This paper proposes a technique to reduce speckle noise using NL means by combining structure and homogeneity similarity. First, image heterogeneity is measured based on the distance of K distribution and is further utilized to distinguish homogeneous and heterogeneous regions. In PolSAR imagery, backscattering from point targets is significantly different from that of distributed media. Strong backscattering from point targets is caused by strong elementary scatterers within a resolution cell. They lack the typical characteristics of speckle and are not random in nature. The preservation of signatures from strong point targets and man-made structures is desired for image interpretation and other applications. In this paper, various samples are collected based on scene heterogeneity. A threshold is utilized to preserve the point and line targets. Then, a new strategy is presented to adapt to the changes in the heterogeneity of the image, which sets the weights of the NL means that were implemented between patches based on the heterogeneity coefficient. Finally, the filtered image is computed. The obtained filter is compared with the refined Lee, mean shift, NLLee, and WisNLTV filters. The qualitative and quantitative aspects of the filters were compared. To compare the ability of the filters to maintain details, corresponding areas in the enlarged span images are shown after filtering with various methods. The proposed method is significantly better on the global and local scales than the existing methods. Moreover, results of H/A/α decomposition show that the proposed method effectively converges the same scattering mechanism and retains complicated scattering mechanisms. The quantitative assessment verifies the equivalent number of looks (a measure of noise reduction), the edge-preserving index, and polarization information preservation on real images. The proposed method has improved filtering performance. The concept of accounting for the heterogeneity coefficient within the NL means algorithm is implemented. The proposed method filters adaptively based on heterogeneity. In addition, comparative results confirm the advantages of the proposed algorithm on both speckle reduction and detail preservation.