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 ›› 2019, Vol. 36 ›› Issue (2): 267-274.DOI: 10.7523/j.issn.2095-6134.2019.02.015

• 信息与电子科学 • 上一篇    下一篇

一种基于谱聚类算法的高光谱遥感图像分类方法

杨随心1,2, 耿修瑞1,2, 杨炜暾1,2, 赵永超1,2, 卢晓军3   

  1. 1. 中国科学院电子学研究所 中国科学院空间信息处理与应用系统技术重点实验室, 北京 100190;
    2. 中国科学院大学, 北京 100049;
    3. 中国国际工程咨询公司, 北京 100048
  • 收稿日期:2018-01-05 修回日期:2018-03-30 发布日期:2019-03-15
  • 通讯作者: 杨随心
  • 基金资助:
    高分5号应用共性关键技术项目(30-Y20A28-9004-15/17)和国家重大科研仪器研制项目(41427805)资助

A method of hyperspectral remote sensing image classification based on spectral clustering

YANG Suixin1,2, GENG Xiurui1,2, YANG Weitun1,2, ZHAO Yongchao1,2, LU Xiaojun3   

  1. 1. Key Laboratory of Spatial Information Processing and Application System Technology of CAS, Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China;
    2. University of Chinese Academy of Sciences, Beijing 100049, China;
    3. China International Engineering Consulting Corporation, Beijing 100048, China
  • Received:2018-01-05 Revised:2018-03-30 Published:2019-03-15

摘要: 结合K-means算法和谱聚类方法的优点,提出一种新的高光谱图像聚类方法。该方法在对高光谱图像数据进行特征降维的基础上,采用K-means算法对图像进行粗聚类处理,然后采用谱聚类方法对粗聚类结果进行较高精度的聚类。与K-means聚类算法相比,该方法有效提高了高光谱图像聚类的分类精度。对模拟数据和真实的高光谱数据的对比实验表明,相对于K-means和谱聚类方法,该方法具有良好的聚类性能。

关键词: 高光谱图像, 聚类, 谱聚类, K均值聚类

Abstract: As common unsupervised clustering methods, K-means and spectral clustering methods have some disadvantages and limitations in clustering hyperspectral remote sensing image. Aiming at these problems, a new clustering method of hyperspectral image is proposed in this study. In this method, based on the feature reduction dimension of hyperspectral image data, K-means algorithm is first used to make rough clustering of images. Then spectral clustering method is used to cluster the results of coarse clustering with high precision. Compared with K-means clustering algorithm, this method effectively improves the classification accuracy of hyperspectral image clustering. Experiments on simulated data and real hyperspectral data show that this method has good clustering performance compared with K-means and spectral clustering methods.

Key words: hyperspectral images, clustering, spectral clustering, K-means

中图分类号: