一种基于光谱与纹理特征的多光谱遥感图像地物分类方法  

A method of multispectral remote sensing imagery classification based on spectral and texture feature

作　　者：胡蕾[1,2] 侯鹏洋 

机构地区：[1]北京航空航天大学计算机学院,北京100191 [2]江西师范大学计算机信息工程学院,南昌330022

出　　处：《中国科技论文》2015年第2期197-200,共4页China Sciencepaper

基　　金：国家自然科学基金资助项目(61262036)

摘　　要：基于光谱特征对高分辨率多光谱遥感图像进行地物分类易形成噪声,提出一种光谱特征与纹理特征相结合的地物分类方法。首先基于光谱特征与纹理特征利用四叉树技术,对图像进行分块处理,以图像块的方式提取地物的光谱特征和纹理特征,然后采用支持向量机(support vector machine,SVM)对图像块进行地物分类,并通过区域增长方法对边缘区域进行处理,使得分类区域边界清晰。对Quickbird多光谱遥感图像进行实验,实验结果表明,该方法地物分类结果精度较高、区域一致性强、噪声少。In high resolution multispectral remote sensing imagery,it is common for the ground object classification based on spectral feature to generate much noise.A classification method is proposed,in which texture feature is introduced and fused with the spectral feature.Firstly,an image is blocked based on spectral and texture features with quad-tree method.With the blocks,spectral and texture features of ground object can be extracted better.Then,(support vector machine)SVM is applied for ground objects classification for the blocks.And the edge blocks processed to make the edge more accuracy through region growing method.With the Quickbird multispectral remote sensing imagery,experimental result shows that the proposed method presents higher classification accuracy and better the compact of classification region with lower noise.

关 键 词：多光谱遥感图像 地物分类 图像分块 光谱特征 纹理特征 支持向量机 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]