滨海光谱混淆区面向对象的土地利用遥感分类

摘要

Land use and land cover information acquisition in coastal spectral confusion region is one of the difficulties for remote sensing information extraction. In this paper, image segmentation and support vector machine classification method were used in order to extract the information of land use/cover with object-oriented technology, based on TM image of March 11, 2007 in Kenli County. The results were compared with that of traditional pixel-based classification. Our results showed that the precision of classification reached 84.83% basing on object-oriented method, which increased by 5.94% and 19.53% respectively in comparison with maximum likelihood method and spectral angle mapper method. It also avoided the "salt and pepper" problem effectively. This study indicated that classification accuracy and efficiency of remote sensing image were improved with object-oriented method, which also provided an effective technological means for fast and accurate information extraction of Land use/cover in coastal spectral confusion region.%滨海光谱混淆区土地利用/覆盖信息获取是遥感信息提取的难点之一,该研究选择黄河三角洲垦利县为研究区,采用2007年3月11日陆地卫星TM遥感影像数据,利用面向对象的土地利用遥感分类技术,通过影像分割和采用支持向量机分类方法对研究区土地利用/覆盖信息进行提取,并将分类结果与传统的基于像元的分类方法进行对比分析.结果表明:面向对象支持向量机的分类精度达到84.83％,比基于像元的最大似然法和波谱角法分别提高了5.94％和19.53％,且有效避免了椒盐现象.说明面向对象的图像分类方法明显提高了遥感影像的分类精度和分类效率,为滨海光谱混淆区土地利用信息的快速、准确提取提供了有效技术手段.