电子学报 ›› 2017, Vol. 45 ›› Issue (10): 2368-2374.DOI: 10.3969/j.issn.0372-2112.2017.10.009

• 学术论文 • 上一篇    下一篇

面向目标检测基于稀疏表示的波段选择方法

唐意东, 黄树彩, 薛爱军   

  1. 空军工程大学防空反导学院, 陕西西安 710051
  • 收稿日期:2016-11-30 修回日期:2017-04-05 出版日期:2017-10-25
    • 通讯作者:
    • 唐意东
    • 作者简介:
    • 黄树彩,男,陕西西安人,1967年出生,教授,博士生导师,主要研究方向为模式识别与智能信息处理.E-mail:hsc1967@126.com
    • 基金资助:
    • 国家自然科学基金 (No.61273275)

Sparse Representation Based Band Selection for Hyperspectral Imagery Target Detection

TANG Yi-dong, HUANG Shu-cai, XUE Ai-jun   

  1. Air and Missile Defense College, Air Force Engineering University, Xi'an, Shaanxi 710051, China
  • Received:2016-11-30 Revised:2017-04-05 Online:2017-10-25 Published:2017-10-25
    • Supported by:
    • National Natural Science Foundation of China (No.61273275)

摘要: 随着高光谱成像技术的发展,日益提高的光谱分辨率在提高目标检测和识别能力的同时,其较高的数据维度和较大的数据量也为数据分析和处理带来了很大的挑战.波段选择作为一种有效提高处理效率的技术受到广泛关注,但却鲜有专门针对目标检测设计的方法.针对上述问题,本文在分析约束能量最小化(CEM)检测算法特点的基础上,提出了一种面向目标检测,基于稀疏表示的波段选择方法.该方法首先基于数据的对称KL散度分布情况,将原始高光谱数据划分为若干波段子空间.然后在各子空间内稀疏重构检测结果,利用选择波段与稀疏向量非零项的一一对应关系,通过求解最优化问题实现波段选择.实验结果验证了该方法的有效性.

关键词: 波段选择, 高光谱图像, 稀疏表示, 目标检测, 子空间划分

Abstract: With the development of hyperspectral imaging technology,the raising spectral resolution improves the ability of target detection and classification.But its great data size and high data dimension also bring challenge to analysis and processing.As a dimensionality reduction technology,band selection (BS) plays an important role in the pre-processing of hyperspectral imagery (HSI).However,few BS algorithms are specially designed for target detection.In this paper,based on analyzing the character of constrained energy minimization (CEM) algorithm,a sparse representation based band selection method (TD-SRBBS) is proposed for HSI target detection.The symmetric Kullback-Leibler divergence is defined for subspatial partition,which makes the original HSI dataset some subset.Sparse reconstruct the detection result in each subset,and then band selection can be implemented based on the one-to-one correspondence between selected bands and nonzero elements of sparse vector.The experiments on real hyperspectral data demonstrate the effectiveness of TD-SRBBS.

Key words: band selection, hyperspectral imagery, sparse representation, target detection, subspatial partition

中图分类号: