基于局部相似自组织的遥感图像自动识别算法

An algorithm for automatic recognizing remote sensing image based on self-organization of local similarity

提出了一种根据局部特征进行图像描述和自动学习的识别算法。该算法能对地表遥感图像进行外貌分析 ,利用地貌特征进行图像分割 ,识别出图像中的丘陵、森林、沙漠、冲积扇等地形。通过用一个大小可变的、边界模糊的窗口对图像进行大量取样 ,再利用这些样本来训练支持向量机 ,并使用该支持向量机进行模式分类 ,进行基于某些类型局部模式的相似性的自组织聚集 ,从而获得对遥感图像的整体性描述或理解。最后给出该方法在一些真实的遥感图像中的运用和分类实验的结果。

In this paper, a new pattern or feature abstraction algorithm is developed based on local attributes of remote sensing image, in order to make a physiognomic image analysis of the earth′s surface and acquire description where there are the foothill, forest, desert or alluvial pie slice etc. A size-changeable and edge-fuzzy window are designed to get many samples via sliding around image. All of these samples are served for the learning of a Support Vector Machine to make patterns classification. A self-organized comparability for assembling can be obtained based on the similarity of some types of local patterns and form a holistic understanding to remote sensing image. The results of classification experiment and application to some actual visible light images are presented.