领域知识辅助下基于多尺度与主方向纹理的遥感影像土地利用分类

Classification of Land-use Based on Remote Sensing Image Texture Features with Multi-scales and Cardinal Direction Inspired by Domain Knowledge

基于灰度共生矩阵(GLCM)的纹理特征在影像空间分析中具有重要作用,提出了一种在领域空间知识辅助下构建GLCM多尺度窗口与主方向权值的方法,从而提高纹理特征的有效性,并解决影像土地利用分类中存在的不确定性问题。为此,根据人类目视解译的特点,对GIS与RS数据进行集成计算:首先,在图像配准的基础上,利用经典的GIS空间数据挖掘算法,渐近式地提取领域形态知识;接着,采用关联分析法建立其与GLCM构造因子之间的响应机制,并设计了基于地类形状指数的多尺度窗口建立算法,以及基于地类主方向分布指数的方向权值测度算法。试验结果表明,领域形态知识与GLCM空间因子之间具有强相关关系,该方法提取出的纹理特征可以描述复杂地物的空间意义,算法复杂度低,性能优越,有效提高了影像土地利用分类的精度。

Texture features based on grey level co-occurrence matrix （GLCM）are effective for image analysis,and this paper proposed a new method to construct GLCM with multi-scales and cardinal direction factors inspired by domain knowledge,in order to improve the performance of texture features and solve the uncertainty problems in image classification of land-use.By simulating the process of human visual interpretation,an integrated computation pattern of GIS and RS data were performed.Firstly,on the basis of image registration, some classic GIS spatial data mining algorithms were employed to asymptotically extract domain morphological knowledge;Next,under the responding mechanism derived from correlated analysis,an algorithm for establishing GLCM multi-scale windows that can match categories one by one,an algorithm for determining GLCM weighted cardinal direction windows that can describe observation orientation were designed based on relevant morphology indexes.Experimental results indicate that, there is a strong correlation between domain morphological knowledge and GLCM construction factors,meanwhile,with lower computational complexity,the new method can extract stable texture features to describe actual spatial meanings of complex objects,thereby improve the image classification accuracy of land-use.