利用极化SAR数据进行土地覆盖分类研究

基于相关矩阵特征向量的目标分解将地物回波复杂的散射过程分解成相互独立的三种单一散射分量:单向散射、双向散射和交叉散射,分别对应各自的目标相关矩阵。目标分解技术降低了散射回波之间的相关性,有利于分析地物散射机理,有助于提高分类精度。对荷兰F levoland地区全极化数据进行分解,经过试验和相关性分析,选用7种数据形成多参数数据组合,对其进行最大似然监督分类,同时进行常规三种极化加相位差的分类和基于复W ishart分布的最大似然分类,逐像元计算混淆矩阵,分析对比三种分类结果的精度,试验表明:相对于常规数据组合分类,基于复W ishart分布的监督分类可以小幅度提高分类精度,而利用目标分解得到多参数组合数据进行分类则有大幅度的提高。

The Research of Land Cover Classification Using Polarimetric SAR Data

Cloude and Pottier have proposed the target decomposition theory for polarimetric SAR data which is based on the eigenvalue analysis of coherency matrix.We can use this method to decompose the data into three scattering components: the single reflection,the double reflection,and the multiple scattering according their scattering mechanism.They are non-correlated and each one has their own coherency matrix.So it is useful to analyze the target scattering mechanism and helpful to improve the classification accuracy.Based on quad-polarization data of Flevoland of the Netherlands,a dataset of feature combinations generated from Cloude and(Pottier's) decomposition after tests and correlation analysis is acquired,including the three decomposition components,the entropy and the scattering angle alpha,together with total power and the phase difference.Image.Another dataset which includes the three basic polarimetric data and the phase difference is also acquired.Then supervised ML classification has been made on the two datasets.The Wishart supervised classification based on coherency matrix has also been made.All the three classification results are presented and the confusion matrices are computed on a per-pixel basis.At last,a conclusion is presented.It proves that the classification accuracy of feature combinations generated from the target decomposition is much better than that of the other two.