基于可变类FCM算法的多光谱遥感影像分割

为了自动确定多光谱遥感影像中地物目标类别数，该文提出一种基于可变类模糊C均值(Fuzzy C-Means, FCM)的多光谱遥感影像分割方法。首先定义像素与聚类的非相似性测度并据此构建目标函数，而后通过求解目标函数得到最优模糊隶属度和聚类中心。其次，研究模糊因子与影像地物目标类别数的关系，并通过定义划分熵(Partition Entropy, PE)指数优选模糊因子，选择PE指数值稳定收敛后所对应的最小模糊因子值为最优模糊因子，根据模糊因子与类别数的关系得到最优类别数，从而实现了影像的可变类分割。最后，利用提出算法分别对合成和真实多光谱遥感影像进行分割实验，实验结果表明，提出算法不仅能自动确定影像的最优类别数，还能获得较好的分割结果，为实现自动确定遥感影像中地物目标类别数提供新方法。

Multispectral Remote Sensing Image Segmentation Based on FCM Algorithm with Unknown Number of Clusters

In order to automatically determine the number of clusters in multispectral remote sensing image segmentation, Fuzzy C-Means (FCM) algorithm with unknown number of clusters is proposed. First of all, a new dissimilarity measure between a pixel and a cluster is defined. The fuzzy membership function and cluster center are obtained through minimizing the objective function. Then, the relationship between fuzzy factor and the number of clusters is studied. The optimal fuzzy factor is selected by defining the Partition Entropy (PE) index and corresponding to the minimum of fuzzy factor after the convergence of PE values. According to the relationship between the fuzzy factor and the number of clusters, the optimal number of clusters is obtained, and the variable cluster segmentation of the image is realized. The analysis based on segmentation results of synthesized image and real multispectral remote sensing images show that the proposed algorithm can automatically determine the number of clusters and obtain the ideal segmentation results simultaneously. It provides a new method for automatically determine the number of clusters of remote sensing image.