基于核的Fisher分类在PolSAR图像的应用

Implementation of Fisher Classification with Kernels in PolSAR Images

提出了利用核函数挖掘物理散射信息,提取极化特征的Fisher分类方法。该方法在Freeman分解的基础上,采用核函数将极化协方差等信息映射到某特征空间,然后在特征空间中进行线性分类,这样能够从PolSAR数据特有的极化散射信息出发,较好利用核方法的优势,改善不同地物类别的可分性。实验表明,该算法能获取有效的分类结果,实现同类相聚,异类分离,具有良好的紧致性。

In this paper, a classification scheme using the kernel function to deal with fully polafimetric SAR images is proposed. Based on the Freeman decomposition, this algorithm introduces one kind of kernel functions and maps the polarimetric coherency matrix elements into feature space, where linear classification could be implemented. By this way, the classification can take advantage of the kernel method and improve the separability of different classes based on physical characteristics of PoLSAR images. Experimental results show the algorithm allows the full exploitation of the information present in the polarimetric image while providing fine performance and good compactness.