摘要
针对极化SAR图像的分类方法多集中在像素级,这些方法不仅运算量大,而且分类效果较差,提出一种利用支撑向量机和超像素分割相结合的方法对极化合成孔径雷达(PolSAR)系统图像分类。首先,利用SLIC算法对Pauli分解后的极化SAR图像进行超像素分割。然后,利用预处理后的数据得到高维的极化特征空间,并利用监督局部线性嵌入(SLLE)算法对高维极化特征进行降维,减少特征空间的冗余信息,提取主要信息。最后,以超像素为处理单元,获得每个超像素内的特征,利用支撑向量机(SVM)对超像素块进行分类,获得初始类别分类结果,之后,使用Wishart分类器再次分类。实验结果表明所提的方法较基于像素点分类的方法能够得到更好的分类效果。
The classification methods for PolSAR images are mostly concentrated at the pixel level. These methods not only have a large amount of computation,but also have a poor classification effect. A method based on SVM and superpixel segmentation for polarimetric SAR image classification is proposed. Firstly,the SLIC algorithm is used to segment the polarimetric SAR image after Pauli decomposition. Then,the preprocessed coherence matrix is used to obtain the high dimension polarimetric feature space,and the Supervised Local Linear Embedding（ SLLE） algorithm is used to reduce the dimension. Finally,superpixel blocks are classified by SVM,and the classification result after being classified by SVM algorithm is classified by the Wishart classifier again.
作者
韩景红
王海江
冉元波
杨建华
HAN Jing-hong;WANG Hai-jiang;RAN Yuan-bo;YANG Jian-hua(College of Electronic Engineering,Chengdu University of Information Technology,Chengdn 610225,China)
出处
《成都信息工程大学学报》
2018年第4期370-374,共5页
Journal of Chengdu University of Information Technology
基金
四川省科技厅应用基础资助项目(2016JY0106)
四川省教育厅重点资助项目(16ZA0209)
关键词
极化SAR
降维
支撑向量机
超像素
分类
polarimetric SAR
dimension reduction
SVM
superpixel
classification
