MKFCM算法在遥感影像分类中的应用研究  

Remote Sensing Image Classification Based on Multi-core Fuzzy C-means Clustering

作　　者：张成才[1] 李飞 王艳梅[1] 罗蔚然 ZHANG Chengcai;LI Fei;WANG Yanmei;LUO Weiran(College of Environment & Water Conservancy, Zhengzhou University, Zhengzhou 450001, China)

机构地区：[1]郑州大学水利与环境学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2020年第3期20-25,共6页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(51739009);河南省科技攻关项目(182102210017)。

摘　　要：模糊C均值聚类(FCM)算法对遥感影像进行分类时,没有考虑像元间相关性的问题,为增加像元间相关性信息,提出了多核模糊C均值聚类(MKFCM)算法,即将多个核函数引入到FCM算法中,依据像元在特征空间中的分布特点,自动筛选出核函数间的最佳权重组合,并以中高分辨率Landsat8影像和高分辨率Pleiades影像为研究数据。通过对FCM、KFCM和MKFCM算法分类结果的精度分析可知,对于不同分辨率的遥感影像,MKFCM都能更好地区分具有相似光谱信息的像元,而且无论是单个类别的精度还是整体的分类精度都高于FCM和KFCM算法,这为精确提取区域土地覆盖信息提供了有效途径。Fuzzy C-Means clustering(FCM)algorithm could not consider the correlation between pixels when classifying remote sensing images.In order to increase the correlation information between pixels,a multi-core fuzzy C-means clustering(MKFCM)algorithm was proposed,in which multiple kernel functions were introduced into FCM algorithm.According to the distribution characteristics of pixels in feature space,MKFCM algorithm automatically filtered out the optimal weight combination among kernel functions.The medium-high resolution Landsat8 images and high resolution Pleiades images were selected as research data,and by analyzing the accuracy of classification results of FCM、KFCM and MKFCM algorithm,the results showed that MKCM algorithm could do better in distinguish pixels with similar spectral information for different resolution remote sensing images.Whether the accuracy of the single category or the overall,classification accuracy with MKCM algorithm was higher than that with FCM or KFCM algorithm,which provided an effective way to extract the regional land cover information accurately.

关 键 词：遥感影像 多核 模糊C均值 核函数 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]