基于支持向量机的极化SAR图像分类

Classification of Polarimetric SAR Image Based on SVM

极化合成孔径雷达可以同时得到地面场景在不同极化组合下的雷达图像,极大地丰富了获取的地物目标信息量。针对极化SAR图像特征提取和目标分类的困难,由4种基本极化组成样本向量,运用基于统计学习理论的支持向量机以及不同的核函数设计分类器,提出了一种新的极化SAR图像分类算法,并对实测极化SAR数据进行分类实验。结果表明,将支持向量机分类器应用于极化SAR图像分类中是可行和有效的,并且通过选择适当的惩罚系数,可以进一步提高分类效果。

Polarimetric synthetic aperture radar can obtain images of ground scenes in the different polarization states, thus it can enrich target information. Because the feature extraction and target classification of polarimetric SAR image are very difficult, a new algorithm of target classification is proposed. Samples are composed of four kinds of polarization. Classifier can be designed using support vector machines based on statistical learning theory and different kernel functions, and classification experiment can be implemented with polarimetrie SAR data. The results indicate it is feasible and efficient to apply SVM classifier to polarimetric SAR image, and classification efficiency can be improved by selecting penalty coefficient.