【作者】 游丽平；

【导师】 陈友飞； 林广发；

【作者基本信息】 福建师范大学 ， 地图学与地理信息系统， 2007， 硕士

【摘要】 随着高分辨率影像的应用越来越普及，迫切要求人们对高分辨率遥感信息提取进行研究，以满足高分辨率影像信息不断增长的应用和研究需要。然而，传统面向像元的分类方法在对高分辨率遥感影像进行分类时，存在着不能充分利用影像信息、分类精度降低、速度慢等局限性，根据高分辨率遥感影像的特点，本文提出面向对象的高分辨率遥感影像分类方法研究。本文以厦门岛SPOT5遥感影像为例，选取具有典型特征的城区建筑物和城区地类丰富的区域为实验区，以Ecognition软件为平台，对实验区进行面向对象的分类实验。首先，根据不同地物类型的特点，选取地物提取的最优分割尺度对实验区进行分割，构建地物类型提取的分类体系，提取地物类型的特征或特征组合，采用模糊分类法对地物类型进行分类，获得实验区的地物分类结果。最后将实验区面向对象遥感影像分类方法的分类结果与传统面向像元分类方法(如最大似然分类法，最小距离法，马氏距离法以及SODATA聚类法)的分类结果进行对比评价，结果表明，采用面向对象遥感影像分类方法对高分辨率遥感影像进行信息提取时，提取的地物与真实地物具有较高的形状和属性一致性，分类的精度更高，有效得避免了“椒盐现象”，分类结果也更易于理解和解释。更多还原

【Abstract】 With the application of the high-resolution image more and more popular, it is urgently require people to carry on research to classification of the high-resolution remote sensing in order to meet the increasing application and study requirement of the information of high-resolution images. However, when used the traditional pixel-oriented method to classify the high-resolution remote sensing image, it can’t fully utilize image information, should reduce the precision of classification and has slow speed. According to the characteristic of the high-resolution remote sensing image, the paper proposes to use the object-oriented method to classify high- resolution remotely sensed data.Taken SPOT5 image of Xiamen Island as an example, choosing the typical urban building area and landuse abundant area as study areas, and regarding Ecognition software as the platform, the paper carry on the classification experiment to the study areas. The paper dose the research by the follow steps: 1) according to the characteristic of different surface features types, choosing the optimum scale to segment the area to extract the objects; 2) constructing the classification system; 3) extracting the characteristics or characteristic associations of the surface feature types; 4) adopting fuzzy classification to classify to surface feature types, then getting the classification result of study areas. At the end, the paper compares and appraises the classification result between the method of object-oriented and pixel-oriented (such as the maximum likelihood classification, the minimum distance classification, the mahalanobis distance classification, and the Isodata cluster classification) The result indicates that: 1) The extracted surface features have higher shape and attribute consistency with true surface features when used object-oriented method; 2) It has higher precision when used object-oriented method to classify the high-resolution image; 3) The object-oriented method is so effective to reduce the "Pepper and Salt Phenomenon"; 4) The classification result of object-oriented analysis is more easy to understand and explain.更多还原