基于波段与形状特征的遥感影像建筑物分类提取

Building Classification and Extraction from Remote Sensing Image Based on Wave Range and Shape Features

提出一种特征组合的方法实现高分辨率遥感影像中建筑物的分类提取。在图像预处理、多尺度分割后,以分割得到的影像对象为基础,考虑到建筑物与其他地物特征上的差异,构建了基于模糊隶属度函数的由紧致度、最大化差异和绿波图层比率3个特征组成的知识规则。前者为形状特征,后两者为波段特征。实验表明,该方法提取的建筑物精度达到89.02%,总体Kappa系数为0.777 8,优于SVM、KNN和单纯用波段特征分类的方法;克服了采用单一特征提取建筑物的局限性,有效提高了建筑物分类精度,具有一定的应用前景。

To work out an approach based on features assemble to achieve building classification and extraction from high-resolution remote sensing image. After preprocessing, multi-resolution segmentation, based on image object, the differences between the char- acteristics of buildings and other surface features were taken into account and knowledge rules were built based on fuzzy membership function by compactness, max diff, and ratio layer 2. The first one belongs to shape features, and the other two are wave range features. The experiment prove that this method has accuracy as high as 89.02%, and overall Kappa coefficient reaches 0.777 8. This method is better than other methods including SVM, KNN, and the method based on wave range feature. This method overcomes the limitations of the method based on wave range feature, effectively improves the accuracy of building classification, and has a certain application prospect.