面向目标检测的高光谱图像压缩技术  

Hyperspectral Image Compression Employing Target Detection

作　　者：宋世刚[1] 粘永健[2] 李纲[2] 

机构地区：[1]国防科技大学信息系统与管理学院,湖南长沙410073 [2]国防科技大学电子科学工程学院,湖南长沙410073

出　　处：《计算机技术与发展》2009年第11期1-3,158,共4页Computer Technology and Development

基　　金：国家自然科学基金资助项目(60572135)

摘　　要：提出了一种基于面向目标检测的高光谱图像压缩算法。该算法利用主成分分析对高光谱图像进行降维,引入虚拟维数对高光谱图像的本征维数进行估计,在估计结果基础上确定降维后的主成分数,并采用SPIHT算法对保留的主成分进行有损压缩。同时,虚拟维数可以实现对图像中端元数目的有效估计,继而采用基于无监督正交子空间投影的端元提取算法提取各目标端元,利用算术编码对目标端元的位置进行无损压缩,解码端可以利用获得的端元位置信息对解压缩后的主成分进行目标检测。实验结果表明,该算法在获得较高压缩性能的同时,能够有效检测出图像中的目标信息。Presents a new lossy compression method for hyperspectrai image, which employs target detection, principal components analysis is used for dimensionality reduction of original data volumes. Virtual dimensionality is introduced to estimate the number of endmembers and principal components. The selected principal components after principal components analysis are compressed by SPIHT algo- rithm. Endmembers are extracted by using unsupervised orthogonal subspace projection method, which are compressed by arithmetic coding losslessly. At the decoder side, the targets can be detected from the decompressed principal components according to the endmembemrs information. Experimental results on AVIRIS data show that the proposed algorithm provides better compression performance,as well as efficient detection for target information.

关 键 词：高光谱图像 有损压缩 主成分分析 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]