面向对象的土地覆盖信息提取方法研究

Land Cover Extraction Based on Object Oriented Analysis

利用遥感方法提取土地覆盖信息时,常用的自动分类方法只能利用遥感数据的光谱信息,而面向对象的分类方法在综合利用遥感数据光谱信息、纹理特征、拓扑关系和加入专题信息进行多尺度分割获取对象后的基础上,通过对对象的目视识别建立模糊判别函数来进行分类。利用面向对象的方法成功提取了三峡库区奉节县中部的土地覆盖信息,得到了较为满意的结果,利用野外采样数据进行验证其精度达89.2%。

Conventional pixel-based image analysis is used frequently to extract land cover feature from remotely sensed data. Therefore only the spectral information is used in the process of classification. However ,object-oriented image analysis can overcome the limitation stated above. It is based on fuzzy logic functions, uses multiscale image segmentation techniques, allows the integration of a broad spectrum of different object features,such as sepctral values,shape and texture. In the paper, the need of land cover classification is satisfied by the result of object oriented analysis in the middle of Fenjie county of the Three Gorges Area, as well as the feature extraction. At the mean time , there is a validation by using in-situ observation ,which shows that the overall accuracy is 89.2%.