基于光谱与纹理信息的Worldview-2影像地物分类  

Land Use /cover Classification of Worldview-2 Images Based on Spectral and Texture Information

作　　者：章文龙[1] 林贤彪[1] 仝川[1] 曾从盛[1] 

机构地区：[1]湿润亚热带生态-地理过程省部共建教育部重点实验室,福建师范大学亚热带湿地研究中心,福建福州350007

出　　处：《福建师范大学学报（自然科学版）》2013年第4期46-52,共7页Journal of Fujian Normal University：Natural Science Edition

基　　金：国家基础科学人才培养基金资助项目(J0830521);福建省科技计划重点项目(2009R10039-1)

摘　　要：高分辨率遥感影像分类是遥感影像处理领域中的一个重要的研究方向.选取Worldview-2影像,分别以光谱信息和光谱结合纹理信息为分类数据,采用最大似然法(MLC)和支持向量机法(SVM)进行监督分类,用混淆矩阵对分类结果进行评价.结果表明,9×9为最佳纹理窗口;SVM法分类精度明显优于MLC法;基于光谱结合纹理信息的分类精度明显优于单纯基于光谱信息的分类结果.辅以影像纹理特征,采用SVM法可以较为有效提取Worldview-2地物信息.It is an important research direction to classification the higher spatial resolution remote sensing images in the field of remote sensing image processing. Two classification methods （maximum likelihood method and support vector machine method） and two data sources （spectral data and spectra combined with texture data） of Worldview-2 images were used to the land use/cov- er classification, and the classification results were assessed by confusion matrix. The results showed that, the best texture window was 9 x 9, and the classification accuracy of the support vec- tor machine method （SVM） was obviously better than that of the maximum likelihood method （MLC）. On the other hand, there also have a higher classification accuracy when classified based on spectra combined with texture data than that of the spectral data. In a word, it is an effective way to classify the land use/cover of Worldview-2 images by taking SVM method, and supplemented by the texture data.

关 键 词：Worldview-2影像 纹理特征 SVM MLC 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]