欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2016, Vol. 33 ›› Issue (5): 674-678.DOI: 10.7523/j.issn.2095-6134.2016.05.015

• 信息与电子科学 • 上一篇    下一篇

基于动态K均值聚类算法的SAR图像分割

邢涛1, 黄友红2, 胡庆荣1, 李军1, 王冠勇1   

  1. 1 中国航天二院二十三所, 北京 100854;
    2 中国人民解放军驻航天二院二十三所军代表室, 北京 100854
  • 收稿日期:2016-01-11 修回日期:2016-04-05 发布日期:2016-09-15
  • 通讯作者: 邢涛
  • 基金资助:

    国家自然科学基金(61271417)和高分专项青年创新基金(GFZX04060103)资助

SAR image segmentation based on dynamical K-means clustering algorithm

XING Tao1, HUANG Youhong2, HU Qingrong1, LI Jun1, WANG Guanyong1   

  1. 1 No. 23 Institute of the Second Academy of China Aerospace, Beijing 100854, China;
    2 The PLA Office in No. 23 Institute of the Second Academy of China Aerospace, Beijing 100854, China
  • Received:2016-01-11 Revised:2016-04-05 Published:2016-09-15

摘要:

针对SAR图像的分割问题,对K均值聚类算法进行研究.分析动态K均值聚类算法,用聚类样本数的正比函数对该聚类适应度函数进行平均,改进适应度函数的计算.毫米波SAR图像分割实验结果表明,对于城区建筑及路、桥场景的分割,改进后的动态K均值聚类算法和自适应动态K均值聚类算法的分割质量与改进前相同,但是分割时间有一定的减少,改进适应度函数后分割效率得到了提高.

关键词: 合成孔径雷达, 图像分割, 聚类, K均值

Abstract:

We present our study on SAR image segmentation based on K-means clustering. We analyze dynamical K-means clustering algorithms and improve the adaptation degree function computation method which divides the raw adaptation degree function by a direct ratio function of the sample number in clustering. Millimeter SAR image segmentation results verify that, for urban area, road, and bridge scenes segmentation, dynamical K-means clustering algorithm and adaptive dynamical K-means clustering algorithm with the improved adaptation degree function computation method have the same segmentation quality while the segmentation efficiency is higher than before.

Key words: synthetic aperture radar(SAR), image segmentation, clustering, K-means

中图分类号: