地球信息科学学报 ›› 2014, Vol. 16 ›› Issue (5): 769-775.doi: 10.3724/SP.J.1047.2014.00769

• • 上一篇    下一篇

基于局部空间信息KFCM的遥感图像聚类算法

吴一全1,2,3(), 沈毅1, 陶飞翔1   

  1. 1. 南京航空航天大学电子信息工程学院,南京 210016
    2. 农业部农业信息技术重点实验室,北京 100081
    3. 江西省数字国土重点实验室,南昌 330013
  • 收稿日期:2013-07-27 修回日期:2013-12-27 出版日期:2014-09-10 发布日期:2014-09-04
  • 作者简介:

    作者简介:吴一全(1963-),男,江苏启东人,教授,博士生导师,研究方向为遥感图像处理与分析、目标检测与识别等。E-mail:nuaaimage@163.com

  • 基金资助:
    农业部农业科研杰出科技人才基金和农业部农业信息技术重点实验室开放基金项目(2013001);国家自然科学基金项目(60872065);江西省数字国土重点实验室开放基金项目(DLLJ201412);江苏高校优势学科建设工程资助项目

An Algorithm of Remote Sensing Image Clustering Based on Kernel Fuzzy C-Means with Local Spatial Information

WU Yiquan1,2,3,*(), SHEN Yi1, TAO Feixiang1   

  1. 1. College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
    2. Key Laboratory of Agricultural Information Technology, Ministry of Agriculture, Beijing 100081, China
    3. Jiangxi Province Key Laboratory for Digital Land, East China Institute of Technology, Nanchang 330013, China
  • Received:2013-07-27 Revised:2013-12-27 Online:2014-09-10 Published:2014-09-04
  • Contact: WU Yiquan E-mail:nuaaimage@163.com
  • About author:

    *The author: CHEN Nan, E-mail:fjcn99@163.com

摘要:

针对模糊C均值(Fuzzy C-Means, FCM)算法,不能有效地对夹杂噪声的遥感图像聚类的问题,本文提出了一种基于局部空间信息核模糊C均值(Kernel Fuzzy C-Means, KFCM)的遥感图像聚类算法。首先,运用核函数将遥感图像的所有像元映射到高维特征空间,通过非线性映射优化遥感图像的有用特征;然后,根据相邻像元之间的相关性,利用一种空间函数重新定义像元的模糊隶属度,将像元的局部空间信息引入到FCM算法中,并在高维特征空间中使用这种基于局部空间信息的FCM算法对像元聚类。由于引入了像元的局部空间信息,算法可以直接应用于原始遥感图像,不需要滤波预处理。大量实验结果表明,本文提出的基于局部空间信息KFCM的遥感图像聚类算法具有较强的抗噪能力,可得到较好的同质区域,优于现有的FCM算法、模糊局部信息C均值(Fuzzy Local Information C-Means, FLICM)算法及KFCM算法。

关键词: 遥感图像, 图像聚类, 模糊C均值算法, 核模糊C均值, 局部空间信息

Abstract:

Aiming at the problem that the fuzzy c-means (FCM) algorithm cannot effectively segment remote sensing images with noise, an algorithm of remote sensing image clustering based on Kernel Fuzzy C-Means (KFCM) clustering with local spatial information is proposed in this paper. Firstly, all pixels of a remote sensing image are mapped into a high-dimensional feature space through the kernel function. Different contributions of each feature vector to the clustering results are fully taken into consideration as well. Thus the influence of noise on the clustering results is greatly reduced and the high-dimensional non-clustered data can be divided nonlinearly. Then, the useful features of the remote sensing image are optimized by non-linear mapping. Next, according to the correlation between adjacent pixels, a space function is used to redefine the fuzzy membership of the pixels. Additionally, the local spatial information of pixels is introduced into the FCM algorithm and the pixels are clustered within the high-dimensional feature space by applying the above-mentioned FCM algorithm based on local spatial information. Accordingly, the clustering results are more accurate. Because of the introduction of local spatial information of pixels, the proposed algorithm can be directly applied to the original remote sensing image without filtering preprocesses and its robustness is adequately strong. A large number of experiments are performed and the results show that the proposed remote sensing image clustering algorithm based on KFCM with local spatial information has stronger noise reduction capabilities and can obtain better homogeneous regions. Therefore, the clustering effect of remote sensing image can be further improved. It is superior to the existing algorithms of remote sensing image clustering such as FCM algorithm, Fuzzy Local Information C-Means (FLICM) algorithm and KFCM algorithm. The proposed algorithm lays a good foundation for the next step of high-spatial-resolution remote sensing image processing.

Key words: remote sensing image, image clustering, fuzzy c-means algorithm, kernel fuzzy c-means clustering, local spatial information