基于CNN和RFC的极化SAR图像分类  

A Method of PolSAR Image Classification Based on Convolution Neural Network and Random Forest Classifier

作　　者：陈彦桥 张泽勇 陈金勇[1,2] 高峰[1,2] 柴兴华[1,2] CHEN Yanqiao;ZHANG Zeyong;CHEN Jinyong;GAO Feng;CHAI Xinghua(The 54th Research Institute of CETC,Shijiazhuang 050081,China;CETC Key Laboratory of Aerospace Information Applications,Shijiazhuang 050081,China)

机构地区：[1]中国电子科技集团公司第五十四研究所,河北石家庄050081 [2]中国电子科技集团公司航天信息应用技术重点实验室,河北石家庄050081

出　　处：《无线电工程》2021年第2期87-91,共5页Radio Engineering

基　　金：航天信息综合应用关键技术研究(2019)(SXX19629X060)。

摘　　要：卷积神经网络(Convolution Neural Network,CNN)在极化合成孔径雷达(Synthetic Aperture Rader,SAR)图像分类的应用中在类别非边界区域取得了好的分类结果,在类别边界区域没有取得好的分类结果,随机森林分类器(Random Forest Classifier,RFC)在极化SAR图像分类领域也得到了不错的分类结果,但是在类别非边界区域的分类结果没有CNN好。提出在类别边界区域使用RFC的分类结果,在类别非边界区域使用CNN的分类结果,通过此种集成学习方式得到了更好的分类结果。In the application of Polarimetric Synthetic Aperture Rader(PolSAR)image classification,Convolution Neural Network(CNN)can obtain good classification results in the category non-edge area,but cannot obtain good classification results in the category edge area.Random Forest Classifier(RFC)has obtained good results in the PolSAR image classification,but its classification results in the category non-edge area are not as good as CNN.In proposed methods,the classification results of CNN are used in the category non-edge area,and the classification results of RFC are used in the category edge area.Better classification results can be obtained by using this kind of integration learning method.

关 键 词：卷积神经网络 随机森林分类器 集成学习 极化合成孔径雷达 图像分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]