雷达植被指数与香农熵在GF-3影像分类中的应用

Application of radar vegetation index and Shannon entropy in GF-3 image classification

针对当前雷达影像分类过程中极化特征组合可提高分类精度这一问题,该文探讨了极化特征组合在高分三号(GF-3)数据地物分类中的应用,以北京市昌平地区GF-3全极化合成孔径雷达(SAR)数据为例,首先对数据进行最优极化分解选择,选取Cloude分解数据为最优数据;然后基于相干矩阵的特征值,提取特征参数雷达植被指数(RVI)与香农熵(SE);最后组合这些极化特征对影像进行决策树分类,并与传统支持向量机(SVM)分类方法的精度进行比较。结果表明:本文采用的RVI与SE极化特征组合的决策树分类方法,较传统SVM分类方法的精度,有一定的提高,尤其是对旱地与有林地、城镇建筑用地与工业用地的区分,效果更佳。

In view of the current radar image classification process,the polarization characteristic combination can improve the classification accuracy,in this paper,the application of polarization characteristic combination in the classification of GF-3 data characteristic was discussed. Taking the GF-3 full-polarization synthetic aperture radar(SAR) data of Changping area in Beijing as an example,carrying out the optimal polarization decomposition of data firstly,selecting Cloude decomposition data as the optimal data,and then extracting the characteristic parameter radar vegetation index(RVI) and Shannon entropy(SE) based on the eigenvalues of the coherent matrix,and finally combining these polarization characteristic to classify the image into decision trees,and the accuracy of traditional support vector machine(SVM) classification methods was compared. The results showed that the decision tree classification method based on the combination of RVI and SE polarization characteristic was better than the accuracy of the traditional SVM classification method,especially the distinction between dry land and forest land,urban residential land and industrial land,the experimental results were better.