遥感影像分类方法的研究

该文基于ERDAS的KnowledgeEngineer分类方法原理，提出一种多信息源、智能化、程序化的阈值分类技术，利用空间模型语言SML（SpatialModelerLanguage）编程实现遥感影像的分类，进而克服了传统分类方法只能针对单一信息源的局限。研究工作以1999年ETM＋遥感影像临港新城为例，将该方法与传统的监督分类方法进行比较和精度评价。结果表明，阈值分类法比监督分类法分类精度高，指标Kappa系数由0.6109提高到0.8204。该方法可通过模块实现多信息源的调用，从已分类图像中提取确认的分类信息，达到一定的智能化，减少人为的重复性操作。

Research on methods of remote sensing image classification

Based on the principle of Knowledge Engineer in ERDAS, a multi-source of information, intelligence,and programmed threshold classification is proposed in this thesis. The Spatial Language program is applied to conduct the classification of remote sensing image,and then,to overcome the single source limitation of the traditional classification methods.Using the Lingang New City ETM＋ RS image acquired in 1999 as the training example,comparing this method with the supervised classification, we evaluate the accuracy of classification by the Kappa index.The final result shows that the Kappa value is improved from 0.6109 to 0.8204. Applying this program,we can collect and gather information from multiple sources, and then extract identified patches,realize intelligent classifying procedure as well as reduce laboriously repetitive operations thoroughly.