基于有限混合模型的极化SAR影像分类方法研究

发布时间：2018-05-13 14:21

本文选题：极化SAR分类 + Wishart分布　；参考：《西安科技大学》2013年硕士论文

【摘要】：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ALOSPALSAR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。
[Abstract]:The classification of polarized SAR images is a very important research problem in the application of polarized SAR. The complex Wishart distribution followed by multi-polarimetric coherent matrix or covariance matrix of fully polarized data is the most widely used and most famous statistical model in the field of polarimetric SAR image classification. However, it is necessary to assume that the scattering component of the target is distributed from the complex Gao Si, which makes the matching effect between the complex Wishart model and the heterogeneity region data poor. In order to solve this problem, this paper studies the classification of polarized SAR images based on K-Wishart distribution model. K-Wishart model is derived from Wishart speckle model and Gamma texture distribution model based on product model. It adapts to the scene description under different conditions. Taking Culai Mountain area in Tai'an area as the experimental area, a scene ALOSPALSAR full polarization data, a scene TM data and land type coverage data were obtained in this study. Firstly, by introducing texture parameters, we develop a non-Gao Si statistical model, that is, K-Wishart statistical model. The polarimetric SAR images based on finite mixed K-Wishart model and finite mixed Wishart model are compared. The results show that the classification accuracy is increased from 74.1935% to 88.9276, and the effect of terrain fluctuation is restrained to some extent. Secondly, this paper also develops a new parameter estimation method, which is matrix logarithmic cumulant parameter estimation method based on Mellin transform, which not only simplifies the process of parameter estimation, but also estimates the texture parameters of K-Wishart distribution in polarized SAR images. Moreover, the estimation performance is improved. In addition, an initialization method based on finite hybrid Gamma model is proposed. This method not only guarantees the same initial conditions required for the comparison of Wishart and K-Wishart classifiers, but also reduces the time complexity.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TN957.52;P225.1

【参考文献】