基于影像认知和地学理解的面向对象分类研究

以遥感影像认知和地学理解为主要分析视角,在图像多尺度分割的基础上,充分挖掘目标地物的光谱特征、形状特征、纹理特征和语义特征信息,明确对象的特征信息与地物之间的对应关系。在此基础上,合理选择目标地物的分类特征,建立分类规则,实现研究区地物的逐级分层分类。结果表明:所选特征能够很好地实现目标地物的信息提取,并具有明确的地学意义,便于理解。与传统的基于像素的最大似然法分类相比较,该方法分类精度有明显提高。

A Study on Object-oriented Remote Sensing Image Classification based on Image Cognition and Geographical Understanding

Taking remote sensing image cognition and geographical understanding as the main analytic perspective,this paper firstly explored the characteristics of the image objects spectral,shape,textural and semantic by the multi-scale segmentation,to determine the corresponding relations between the characteristic information of the object and the ground.And then,this study rationally selected some classification features and built classification rules.Finally,multiple level classifications were carried out hierarchically to extract information of the interested objects in the study area.The result showed that the selected features in classification rules not only can effectively extract information,but also had the explicit geographical significance.Compared with the traditional pixel-based maximum likelihood classification methods,the classification accuracy of the new method was improved significantly.