摘要
滨海湿地信息提取对于准确掌握滨海湿地分布现状、保护与管理滨海湿地珍稀资源具有重要意义。通过可分性指数筛选极化分解特征并利用随机森林法对全极化SAR影像进行分类,以提高滨海湿地保护区地物信息提取精度。选取辽宁省辽河口湿地自然保护区作为研究区域,基于国产高分三号全极化雷达影像,采用5种极化目标分解方法提取极化特征,利用可分性指数优化特征选择,最后利用随机森林法进行辽河口自然保护区地物分类及精度评价。实验结果表明,基于优化选择的极化特征地物分类精度可达75.47%;优化选择后的极化特征参数能够有效避免信息冗余,提高滨海湿地保护区地物信息提取精度。
The study on the information extraction methods of coastal wetlands is highly significant for accurately grasping the distribution status of coastal wetlands and for protecting and managing the rare resources in coastal wetlands.To improve the information extraction precision of surface features in coastal wetland conservation areas,this paper screens the polarimetric decomposition features using the separability index and classifies fully polarimetric SAR images using the random forest method.The details are as follows.Based on the domestic GF-3 fully polarimetric radar images of the Liaohe River Estuary National Nature Reserve in Liaoning Province,five polarimetric target decomposition methods were used to extract polarimetric features,the separable index was adopted to optimize feature selection,and finally the random forest method was utilized to conduct the classification and accuracy assessment of surface features in the study area.The experiment results show that the classification accuracy of surface features in wetlands based on optimized polarimetric features was up to 75.47%.Meanwhile,the optimized polarimetric feature parameters can effectively avoid information redundancy and improve the information extraction accuracy of surface features in coastal wetland conservation areas.
作者
何陈临秋
程博
陈金奋
张晓平
HE Chenlinqiu;CHENG Bo;CHEN Jinfen;ZHANG Xiaoping(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《自然资源遥感》
CSCD
北大核心
2021年第4期105-110,共6页
Remote Sensing for Natural Resources
基金
国家自然科学基金重点项目“基于认知计算的遥感卫星下行数据即时服务的理论与方法研究”(编号:61731022)资助。
关键词
滨海湿地
高分三号
极化特征选择
coastal wetland
GF-3
selection of polarimetric features
作者简介
第一作者:何陈临秋(1993-),女,硕士研究生,主要从事雷达遥感湿地信息提取方法研究。Email:heclq@radi.ac.cn;通信作者:程博(1974-),男,博士,高级工程师,主要从事遥感卫星信息处理及应用研究。Email:chengbo@radi.ac.cn。
