文献类型：期刊文献

中文题名：基于对象的CHRIS遥感图像森林类型分类方法研究

英文题名：THE STUDY ON THE FOREST TYPES CLASSIFICATION METHOD OF CHRIS REMOTE SENSING IMAGE BASED ON OBJECT

作者：李小梅[1] 张秋良[1] 李增元[2] 谭炳香[2]

第一作者：李小梅

机构：[1]内蒙古农业大学林学院;[2]中国林科院资信所

年份：2010

期号：2

起止页码：31-36

中文期刊名：内蒙古农业大学学报：自然科学版

收录：北大核心:【北大核心2008】；

基金：林业公益性行业科研专项(200804027-03);林业科技支撑计划(2008BAD-BOBO302);国家自然科学基金(40601070);国家高技术研究发展计划(2006AA120105)

语种：中文

中文关键词：面向对象;分类;影像分割;最近邻法;CHRIS影像

外文关键词：Object-oriented; classification; image segmentation; nearest neighbor method; CHRIS image

分类号：S771.8

摘要：高光谱遥感森林类型分类中采用传统基于像素分类方法精度较低,本文通过高光谱遥感影像的特征,采用面向对象的最近邻监督分类方法对高光谱CHRIS影像进行分类实验,首先对影像进行多尺度分割,然后将分割对象信息、形状特征及上下文联系等特征构成特征空间进行最近邻监督分类,并与传统的基于像素的最大似然分类方法进行比较分析,结果表明,面向对象的最近邻法能够较好的识别森林类型,总精度为89.06%,kappa系数为0.82,而最大似然法分类精度为85.75%,kappa系数为0.79,其分类精度明显高于最大似然法,这表明该方法适合高光谱遥感影像分类,为今后的高光谱遥感森林类型分类能够起到技术参考和理论依据。
The accuracy of hyperspectral data in forest type classification using traditional pixel-based method is very low.In this paper,we analyze the characteristics of hyperspectral data and then choose the object-oriented nearest neighbor supervised method to classify the CHRIS image.Firstly,we segment the image basing on the multiresolution sense.Secondly,adopt nearest neighbor supervised method to class the composite dataset which is combining with the information of segmented objects,shape characteristics and context features.At last,we compare the results of the two ways-the traditional maximum likelihood method based on pixel and the nearest neighbor supervised method based on object.The total accuracy is 89.75% and 96.23%,which shows that the latter approach better for forestry type classification.Through this experiment,we recognize that the object-oriented nearest neighbor supervised method has a definite advantage in dealing with hyperspectral data classification and it can play a technical reference and theoretical basis for the future forest type classification of hyperspectral image.