摘要
高光谱遥感以其携带的数据量显著增加为代价换取了纳米级的光谱分辨率,使得基于特征光谱信息的目标地物识别成为可能。但如何从大量带有冗余的数据中提取有用信息,是高光谱研究的一个极富挑战性的课题和其实用化的基础之所在。本文以江苏宜兴地区OMIS I数据为例在全面计算影像的统计特征(相关系数、均方差、最佳波段指数、信噪比等)的基础上,结合目视效果对波段集合进行初步缩减和分组;结合地面实测地物光谱详细分析不同地物光谱特性,进行面向土地覆盖易混类别的波段选择;最后总结了OMIS I数据特征选择与提取流程,相关实验证明应用该流程进行特征选择与提取,其后续分类精度较高。
Marked increase of hyperspectral remote sensing data size is a price to pay for waveband width of nm level.It is possible to make the target objects be identified based on diagnostic spectrum information hyperspectral resolution provided.Therefore,extracting useful information from a large quantity data is a challenging task and basis of practical applications.This paper takes OMIS I data of Yixing area,Jiangsu province,China as an example,on the basis of comprehensive computing statistical character of imagery,the bands aggregate is curtailed and carved up combining with visual effect.According to the analysis of different spectrum character,bands selection oriented to land cover sorts that easy to be promiscuous is carried through.In the end,the hyperspectral optimal character selection and abstraction workflow is summarized.Relating experiment proves that the classification precision is preferable after character selection and abstraction by this workflow.
出处
《测绘科学》
CSCD
北大核心
2006年第2期83-85,共3页
Science of Surveying and Mapping
基金
成像光谱技术在土地动态监测中的应用(项目编号:2001AA136020-2)
关键词
特征选择与提取
OMIS
I
土地覆盖
分类
character selection and abstraction
OMIS I
land cover
classification
