无人工样本的SVM遥感图像分类方法

Remote Sensing Image Classification Method Based on SVM

针对遥感图像监督分类方法需要人工提取训练样本的缺陷,提出一种模糊K均值聚类(FCM)提取训练样本、支持向量机(SVM)进行分类的方法。首先用FCM进行初步分类得到隶属度矩阵并判断每个样本的类别号;然后根据隶属度矩阵提取每类样本中密集程度较高的样本作为训练样本;最后用SVM对样本进行训练、再次分类。该方法克服了SVM算法需要人工样本的缺点,改善了传统非监督分类算法的性能,UCI标准数据库Iris数据和遥感数据样本的实验结果证明了该方法的可行性。

In order to solve the problem of supervised classification method for remote sensing image in training samples＇ manual extraction,a remote sensing image classification method is proposed ,which uses fuzzy K-means clustering algorithm（FCM） to extract training samples and makes classification by support vector machine （SVM）. First of all ,this algorithm conduct preliminary classification to obtain membership degree matrix and judge category number of each sample using FCM. Then ,it extracts training samples which intensive in every cluster according to the membership degree matrix. Finally, train the samples and conduct second classification using SVM. The improved algorithm overcomes the drawback of SVM in training samples＇ extraction;greatly improves the classification performance of traditional unsupervised classification method. Standard UCI database of Iris data and remote sensing data sample experimental results prove the feasibility of this method.