摘要
基于eCognition Developer平台,以泰安市QuickBird影像为数据,采用面向对象多尺度分割、最邻近和隶属度分类,充分利用高分辨率QuickBird影像具有的丰富光谱、形状、纹理和结构等地物信息,对实验区进行分类并提取住宅建筑物信息。实验表明,与传统逐像元分类法相比,面向对象分类法有效地避免了分割区域的离散破碎,地类信息的提取更加完整、精确、高效。
Traditional pixel-based classification method only uses the spectra feature of images.However,object-oriented method uses multi-scale segmentation based on a region combination algorithm and integrates spectral and spatial features in the segmentation based on image objects.It uses image object as the smallest classification unit and it is more suitable for high-resolution image analysis and processing.With eCognition developer platform and QuickBird image of Tai'an city,this paper used multi-scale segmentation,nearest neighbor and membership functions classification and took full advantage of the spectrum,shape,texture,structure features and other spectra feature of QuickBird image to extract residential building features in experimental zone.The results show that object-oriented classification method effectively avoid the fracture of segmented regions and is more integrated,more accurate and more efficient in the extraction of images comparing with the traditional pixel-based classification method.
出处
《地理空间信息》
2013年第1期67-69,13,共3页
Geospatial Information
基金
湖南省教育厅优秀青年基金资助项目(11B046)
空间数据挖掘与信息共享教育部重点实验室(福州大学)开放课题(201007)
湖南科技大学研究生创新基金资助项目(S120035)
关键词
面向对象
多尺度分割
高分辨率
住宅建筑物
object-oriented,multi-scale segmentation,high resolution,residential building
