摘要
压缩是高光谱遥感(hyperspectral remote sensing)图像的一个重要研究领域.文中充分考虑了高光谱遥感图像的谱间相关性较强而空间相关性相对较弱的特点,采用了自适应波段选择降维方法与基于神经网络的矢量量化方法相结合的方法对高光谱遥感图像进行压缩.首先采用自适应波段选择(Adaptive band selection)的谱间压缩方法,通过自适应地选择信息量大并且与其他波段相关性小的波段来降低高光谱数据量.然后对降维后图像在空间进行小波变换并进行矢量量化,最后对量化后数据进行自适应算术编码.实验结果表明,谱间压缩能够保留信息丰富的波段,同时计算复杂度大大降低;基于神经网络的SOFM算法及其改进算法取得较好的空间压缩效果,实现了对高光谱遥感图像的有效压缩.
Compression is an important research field of hyperspectral remote sensing images. Proposed is a method to compress hyperspectral remote images that combines a dimensional reduction algorithm of adaptive band selection with an algorithm using codebook design that is based on a self organization feature maps （SOFM）. First, a dimensional reduction algorithm of adaptive band selection was developed to reduce dimensions by adaptively selecting a great deal of information and low correlative bands. Second, wavelets transformations and vector quantization were applied to reduced the images. Finally, using vector quantization, adaptive arithmetic coding was applied to the data. Experimental results show that spectrum compression can contain significant information bands while greatly decreasing computing complications. The SOFM algorithm based on a neural network and an improved algorithm have a good effect on space compression of hyperspectral remote sensing images.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2006年第3期447-452,共6页
Journal of Harbin Engineering University
关键词
高光谱遥感图像
矢量量化
神经网络
自适应波段选择
图像压缩
hyperspectral remote sensing image
vector quantization
neural network
adaptive band selection
image compression
