面向对象的多种特征极化SAR决策树分类方法

Object-oriented decision tree classification method of multi-feature polarimetric SAR

针对目前极化合成孔径雷达(PolSAR)影像分类单一特征无法获得令人满意的分类结果的问题,该文设计了综合运用纹理和多种极化目标分解特征,结合面向对象分析及CART决策树的分类方法。为验证该方法的有效性,以北京市某区域全极化RADARSAT-2影像为例,按照“影像预处理-目标极化分解-特征参数优化选择-面向对象影像分割-多特征CART决策树分类”的总体思路进行实验,并在特征参数选择时充分考虑各参数之间的相关性、地物的散射特性和分类效果。结果表明:影像特征参数是PolSAR影像分类的关键,恰当的特征参数组合有利于获取准确的分类结果。

Aiming at the problem that a single feature of polarimetric synthetic aperture radar(PolSAR)image classification can't obtain satisfactory classification results,this study designed a classification method which combines texture and polarimetric target decomposition features with object-oriented analysis and CART decision tree?In order to verify the effectiveness of the method,taking a RADARSAT-2 Quad-PolSAR image in a certain area of Beijing as an example,through image preprocessing,target polarization decomposition,optimization of feature parameters,object-oriented image segmentation and multi-feature CART decision tree classification,and the correlation between the parameters,the scattering characteristics of surface features and classification effect were fully considered in the selection of feature parameters.The results showed that the image feature parameters were the key to the classification of PolSAR images,and proper combination of feature parameters was conducive to obtain accurate classification results.