中分辨率遥感影像分类方法研究

Study on moderate resolution remote sensing image classification method

为了提高中分辨率遥感影像的分类精度,综合使用独立分量分析法(ICA)与纹理特征应对分类特征进行获取,使用相关系数分析、灰度差异分析确定了独立分量图层、纹理图层和NDVI图层的特征组合方式。在同样使用非监督分类方法的前提下,将该图层组合方式与常用的其他两种组合方式的分类结果进行对比后,该图层组合可获得更好的类别分离性,总体分类精度更高,达到87%,Kappa系数0.84。过程中发现,基于均方差的纹理图层进行ICA处理后,地物在图层组中的灰度差异极小,对分类工作没有贡献。

To improve the classification （ICA） and texture feature are applied to accuracy using moderate resolution satellite image, independent component analysis create classification features, and correlation coefficient analyses and difference comparison of grey value are used to define the feature combination we used which includes ICA layers, texture layers and NDVI layer. On the premise of using the same unsupervised classification method, the feature combination we used can get better class separability and more higher classification accuracy after comparing with other two feature combinations, its totally accuracy can be reach 87%, and Kappa coefficient is 0. 84. During the studying process, we found that in the result image of texture feature after ICA process, the differences between different land objects are very small, it shown that the result image had no contribu tion to the improvement of classification accuracy.