J4 ›› 2009, Vol. 36 ›› Issue (3): 512-516.

• 研究论文 • 上一篇    下一篇

期望最大化聚类的高光谱亚像素目标检测

刘德连;王博;张建奇   

  1. (西安电子科技大学 技术物理学院,陕西 西安  710071)
  • 收稿日期:2008-02-28 修回日期:2008-03-31 出版日期:2009-06-20 发布日期:2009-07-04
  • 通讯作者: 刘德连
  • 基金资助:

    国家自然科学基金资助(60777042)

Hyperspectral subpixel target detection approach based on expectation-maximization cluster

LIU De-lian;WANG Bo;ZHANG Jian-qi   

  1. (School of Technical Physics, Xidian Univ., Xi'an  710071, China)
  • Received:2008-02-28 Revised:2008-03-31 Online:2009-06-20 Published:2009-07-04
  • Contact: LIU De-lian

摘要:

针对高光谱目标检测中复杂背景的影响,提出一种基于期望最大化聚类的亚像素目标检测方法,利用背景分解来描述复杂背景.首先,采用期望最大化聚类法实现高光谱图像的背景分解.然后,将背景子空间模型应用于分解得到的场景.由于分解得到的场景更加单一,因此该方法更适合于复杂背景下的亚像素目标检测.将提出的方法应用于实际的高光谱图像,实验结果表明这种方法具有更好的检测性能.

关键词: 遥感, 高光谱, 亚像素目标, 目标检测, 图像分割

Abstract:

Background is a key interferene in target detection. To avoid the interferene of complex background, an expectation-maximization cluster based approach to subpixel detection in hyperspectral is presented that incorporates background segmentation to model complex background. First, the expectation-maximization cluster model is employed to segment whole background into homogenous regions. Then the adaptive matched subspace detection algorithm (AMSD) is applied in each homogenous region. Since the segmented regions are more homogenous than the whole complex background, our new approach can have a better performance. Experimental result with a real hyperspectral image has proved the validity of the new approach.

Key words: remote sensing, hyperspectral, subpixel target, target detector, image segmentation

中图分类号: 

  • TN219
Baidu
map