【摘要】 研究一种新型遥感图像分类器。它采用了一种称为ES(ExtractingShape)的新算法,即从遥感图像直接提取景物的矢量图形,由此获得图形元素和数学特征,并通过在波谱空间和形状特征空间的两次聚类给出景物的分类结果。通过引入知识库和捕捉用户调整事件获得新知识,使该分类器具有学习能力。分类精度高于国际知名的GIS软件IDRISI40%以上,在普通配置的台式个人电脑上,对彩色遥感图像的分类时间比IDRISI少23%。更多还原

【Abstract】 A remotely sensed image classifier is discussed in this paper and a novel algorithm,named ES (Extracting Shape),is as the core of the classifier. In ES, a vector graphics is abstracted from a remote sensing image,then vector elements and mathematic features can be obtained from the graphics and a classifying result can be generated by twice clustering in spaces of spectrum and shape features respectively. The classifier is provided with learning capability through establishing repository which can obtain new acknowledge from the adjusting events by users. The classification precision is over 40 percent better and operation time is 23 percent less by using the authors’ classifier(hereafter call it as ″ES classifier″) than those by using these classifiers in famous GIS software IDRISI on a personal computer.更多还原