引入商空间粒度计算的全极化SAR影像分类

为了充分利用不同极化特征信息，并将其有效地结合，提出一种结合粒度计算的全极化合成孔径雷达（synthetic aperture radar，SAR）影像分类方法。在不同极化目标分解特征组合的基础上引入影像纹理信息，利用光滑支持向量机（smooth support vector machine，SSVM）对不同特征组合进行类别划分获得粗粒度空间，采用商空间对粗粒度进行合并；根据全极化SAR影像分布特性，以相干矩阵作为新的特征矢量，利用Wishart测度代替传统欧氏距离对差异粒度进行推理，通过合并推理结果与合成论域，获得精细分类结果。采用L波段San Francisco地区和荷兰Flevoland地区的全极化SAR影像进行分类试验，结果表明:利用SSVM算法对全极化SAR影像进行粗粒度划分，并采用Wishart距离对差异粒度推理综合，总体分类效果优于结合纹理信息的Cloude及Yamaguchi4分类结果，且优于基于线性特征融合进行监督分类方法。

An Improved Full Polarimetric SAR Image Classification Method Combining with Granularity Computing of Quotient Space Theory

A new Full Polarimetric Synthetic Aperture Radar (SAR) image classification method is proposed that combines quotient space granularity computing and the texture information to carry out comprehensive classification. Firstly, we classify the Cloude and Yamaguchi4 decomposition characteristics with texture features using Smooth Support Vector Machine (SSVM) algorithm, to get two classification results, which are the quotient spaces. According to quotient space theory, the two particle size layer are synthesized and in accordance with SAR data distribution polarization characteristics, we use the Wishart measure instead of the traditional Euclidean distance to infer the granularity difference and calculate its category, and combine the results of this reasoning with the synthetic domain in order to get exact classification results. To validate the proposed method, polarimetric SAR data acquired by AIRSAR for San Francisco and Flevoland were employed in classification experiments. The results indicate that the classification results obtained by the proposed method arenot only better than the combination of texture information of the Cloude and Yamaguchi4 supervised classification results, but superior to all the features as a feature vector based on a simple feature fusion for supervised classification results.