基于Kmeans与SVM结合的遥感图像全自动分类方法*

遥感图像分类方法通常采用监督的学习算法,它需要人工选取训练样本,比较繁琐,而且有时很难得到;而非监督学习算法的分类精度通常很难令人满意.针对这些缺陷,提出一种基于K-means与支持向量机(SVM)结合的遥感图像全自动分类方法.首先使用K-means聚类算法对样本进行初始聚类,根据每类中样本数及其稀疏程度选取一些点作为标记的学习样本训练SVM分类器,然后用SVM对原始数据重新分类.Iris数据和遥感数据的实验结果均验证了新方法的有效性.

Automated remote sensing image classification method based on Kmeans and SVM

The supervised learning algorithm was usually used for remote sensing image classification, but its training samples need to be chosen by manual, which was boring and sometimes even difficult. However, in unsupervised learning algorithm classification result was often not satisfactory. According to these limitations, an automated remote sensing image classification method of combining K-means algorithm with SVM. In new method, at first K-means algorithm was used to cluster original data points, and then according to the number and sparse degree of points in each class, some points as labeled samples were chosen to train SVM, at last SVM was utilized to reclassify original data points. Experimental results for Iris data and remote sensing data verify the validity of the proposed method.