极化SAR影像分类是极化SAR应用中一项非常重要的研究问题。全极化数据多视极化相干矩阵或协方差矩阵所服从的复Wishart分布是目前极化SAR影像分类领域应用最广泛、最著名的统计模型。但应用该模型需要假定目标的散射分量服从复高斯分布，这使得复Wishart模型与异质性区域数据的匹配效果较差。为解决这一问题，本文研究基于K-Wishart分布模型的极化SAR影像分类。K-Wishart模型基于乘积模型，由Wishart相干斑模型和Gamma纹理分布模型推导而得，该模型将极化信息与纹理信息进行融合，适应了不同条件下的场景描述。 本研究以山东泰安地区徂徕山一带作为实验区，获取了覆盖实验区的一景ALOS PALSAR全极化数据，一景TM数据和土地类型覆盖数据。 首先，通过引入纹理参量，本文发展了一种具有非高斯性质的统计模型，即K-Wishart统计模型。将基于有限混合K-Wishart模型和基于有限混合Wishart模型的极化SAR影像进行分类对比，结果发现，分类精度由74.1935%提高到了88.9276%，且在一定程度上抑制了地形起伏的影响。 其次，本文还发展了一种新的参数估计方法，即基于Mellin变换的矩阵对数累积量参数估计法，对极化SAR影像的K-Wishart分布纹理参数进行估计，不仅简化了参数估计过程，而且提高了估计性能。 此外，本文还提出一种基于有限混合Gamma模型的初始化方法。该方法不仅保证了Wishart和K-Wishart两种分类器进行对比所需要的同等初始条件，而且还降低了时间复杂度。

Classification of polarimetric SAR(PolSAR) has become an important topic of PolSAR applications. The polarimetric covariance matrix of PolSAR is found having a complex Wishart distribution. However, the complex Wishart distribution is under the circular Gaussian assumption. It fits well for measurements over homogeneous regions, but often fails over heterogeneous backscattering media by SAR. To solve this problem, we do research with the K-Wishart distribution. The K-Wishart model is based on the well known product model, with a Gamma distributed texture parameter. It utilises the full polarimetric information and incorporates texture, thus enabling to improve the model to fit of clusters even it is ununiform. The test site is located in the vicinity of of culai Mountain in Taian County, Shandong Province, China. One scene of full polarimetric ALOS PALSAR image, one scene of TM image and a land cover map were acquired for the study. First of all, incorporating the texture virable, we derived the K-Wishart distribution for the covariance matrix with the nature of non-Gaussian. Performing experiment on the classification based on the finite K-Wishart mixture model, the results have shown that the overall accuracy has been increased from 74.1935% to 88.9276%, and to some extent has inhibitted the effects of terrain. Sencondly, we have implemented the novel parameter estimation with matrix log-cumulants based on Mellin transform to estimate the texture parameter in K-Wishart probability density function. The design has simplified the parameter estimation process and been further refined. In addition, we also used the finite Gamma mixture model to make the initialization. This method of initialization not only ensures that the two Wishart and K-Wishart classifiers could make an equal comparison, but also reduces the time complexity.