基于语义网的遥感图像分类  

Remote Sensing Image Classification Based on Semantic-net

作　　者：张亚亚[1] 张立民[1] 刘小伟[1] 徐涛[1] 

机构地区：[1]海军航空工程学院电子信息工程系,山东烟台264001

出　　处：《计算机技术与发展》2015年第5期218-223,共6页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61102167)

摘　　要：为充分利用遥感图像的先验知识,提高地物信息提取的效率,针对以往低分辨率遥感图像基于像素知识分类出现的问题,提出了一种基于对象的语义网的遥感图像知识分类框架。通过利用语义网络表示地物种类之间的层次关系,研究地物种类的特征属性,通过多种属性集共同描述对象,研究基于模糊隶属度函数的图像对象及其概念的匹配程度,对图像进行精确分类。在InterIMAGE系统中的仿真实验结果表明,通过与人工解译的真值图进行对比,该方法在不需要特定提供样本的情况下,有效提高了地物信息提取的效率和准确度。For full use of priori knowledge for remote sensing images to improve the efficiency of land cover information extraction,in view of the problem on low-resolution remote sensing images on pixel knowledge-based classification,present a remote sensing image knowledge classification framework of semantic-net based on the object. Using semantic network to represent the land cover categories hierarchical relationship,research the feature of land cover categories,and by multi-attribute sets to describe the objects,study the matc-hing degree of image objects and concepts based on fuzzy membership function to interpret image. Simulation results on InterIMAGE sys-tem show that the method has improved the efficiency and accuracy in the case of without special samples,compared with the manual in-terpretation of the ground truth image .

关 键 词：语义解译 基于知识 隶属度函数 InterIMAGE系统 区域匹配 遥感图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]