针对聚类算法在应用中分割速度慢、抑制噪声能力弱等问题, 本文提出一种基于核模糊C-均值(Kernel Fuzzy C-means, KFCM)和融合期望最大化(EM)算法混合聚类的遥感图像分割。首先给原始KFCM算法引入隐含变量来对像素预定义类别, 然后利用EM算法评价预定义的类别是否最优, 以此完成对遥感图像的聚类分割。在利用EM算法进行评价时, 对KFCM引入空间邻域信息, 采用惯性权重对其初始化参数进行优化增强算法效率。与传统的聚类分割方法进行比较, 研究结果表明, 该方法速度快、效果好、精度也能满足应用要求, 具有较高的应用价值。

Aiming at the problem that the clustering algorithm is slow in the application and weak in noise suppression, this article proposes a remote sensing image segmentation based on improved Kernel Fuzzy C-means (KFCM) and EM algorithm hybrid clustering. Firstly, the implicit variables are introduced into the original KFCM algorithm to predefine the pixels, and then the EM algorithm is used to evaluate whether the predefined categories are optimal, so as to complete the clustering of the remote sensing images. When using the EM algorithm for evaluation, the KFCM introduces the spatial neighborhood information and the inertia weight is used to optimize the initialization parameters to enhance the efficiency of the algorithm. Compared with the traditional clustering methods, the results show that the method in this article is fast and effective, and the precision can meet the application requirements and has high application value.