一种模糊C均值聚类的影像分类方法

An Image Classification Method Based on Fuzzy C-means Clustering

针对传统的模糊C均值聚类(fuzzy c-means clustering,FCM)后处理方法往往不能有效划分较优类别的问题,提出了一种聚类结果明确化的新方法,命名为邻域加权隶属度和(neighboring weighted membership grade sum,NMS)方法。方法增加了邻域信息的使用,采用了阈值加权和反距离加权处理聚类结果,并以多种类型遥感影像为测试实例,进行了不同方法的影像分类对比研究。结果表明,分类结果全局精度比最大隶属度方法平均提高约8%,Kappa系数平均提高11%;同时噪声图斑数量下降,图斑具有更好的完整性;新方法对具体分类问题的使用更具灵活性与普适性。

In order to solve the problem that the traditional post-processing method of fuzzy c-means clustering（FCM）cannot effectively classify the remote sensing image,in this paper,a new method for clarifying clustering results is proposed,which is called neighboring weighted membership grade sum（NMS）method.The method increases the use of neighborhood information and adopts threshold weighting and inverse distance weighting.Through various types of remote sensing images for testing,different methods of image classification have been studied.Results show that the whole classification accuracy and Kappa coefficient of the new mothed is 8% and 11% higher than that of the maximum membership,respectively;the noise figure spot number of classification results decreases,and the integrity of map spot is better;the new method is more flexible and universal for the use of specific classification problems.