Remote sensing image classification based on neighbor sample density and membership weighted FCM algorithm

Xiaofang Liu,Binbin He

DOI: https://doi.org/10.19650/j.cnki.cjsi.2011.10.013

2011-01-01

Abstract:Fuzzy C-means (FCM) algorithm has an equal partition trend for data sets. To overcome this disadvantage, several weighted FCM (WFCM) algorithms are formed by introducing the neighbor distributing characteristics of samples themselves in the standard FCM algorithm, they are neighbor sample density WFCM (NSD-WFCM), neighbor sample membership WFCM (NSM-WFCM) and neighbor sample density and membership WFCM (NSDM-WFCM). These algorithms are applied to a set of remote sensing images for classification. Classification results show that these WFCM algorithms (NSD-WFCM, NSM-WFCM and NSDM-WFCM) improve the classification accuracy of the FCM algorithm by 5.67%, 7.50% and 11.17%, and the Kappa coefficient by 8.50%, 11.25% and 16.75%, respectively. Experimental results indicate that these WFCM algorithms improve the clustering performance to some extent, and significantly improve the unsupervised classification ability of remote sensing images compared with the FCM algorithm; moreover, the classification performance of the NSM-WFCM algorithm is superior to that of the NSD-WFCM algorithm, and the NSDM-WFCM algorithm has the best classification performance in the three algorithms.