高光谱图像压缩感知投影与复合正则重构  

Compressed Sensing Projection and Compound Regularizer Reconstruction for Hyperspectral Images

作　　者：冯燕[1] 贾应彪[1,2] 曹宇明[1] 袁晓玲[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710129 [2]韶关学院计算机科学学院,广东韶关512005

出　　处：《航空学报》2012年第8期1466-1473,共8页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(61071171);国家自然科学基金重点项目(60736007)~~

摘　　要：压缩感知理论提供了一种新的数据获取和压缩思路,能有效地把计算负担从编码端转移到解码端。高光谱数据具备数据稀疏性、空间相关性和谱间相关性,结合这3类先验知识,提出了一种基于复合正则化的高光谱图像压缩感知投影与重构方法。该方法的编码端只需要一个简单的投影操作;在重构算法实现中,基于变量分裂的思想,把具备多个正则项的优化问题转化成多个简单的优化问题,并用迭代方式求解。实验结果表明,本文算法在高光谱图像重构上能获得更高的峰值信噪比和更好的重构效果。该方法具备极低的编码复杂度,适用于资源受限的机载和星载高光谱成像平台。Compressed sensing has proposed a new mechanism for data acquisition and compression, which can shift heavy computational loads from encoders to decoders. The hyperspectral images have sparse/compressible representations on some orthonormal bases and are of spectral and spatial correlations. According to the prior information of hyperspectral images,a novel hyperspectral compressed sensing projection and reconstruction method via compound regularizers is proposed. At the encoder, it only needs a simple projection. In the Implementation of the reconstruction algorithm, the problem of compound regularizers is turned into dealing with a few simple optimization problems by applying the variable-splitting method and is solved by iteration. Experimental results show that the proposed algorithm is able to reconstruct the hyperspectral images more efficiently than the current algorithms. Our method has very low decoding complexity and it is suitable for severely resource-constrained spaceborne and airborne remote sensing platforms.

关 键 词：压缩感知 高光谱图像 数据压缩 测量 重构 凸优化 

分 类 号：V243.5[航空宇航科学与技术—飞行器设计] TP751.1[自动化与计算机技术—检测技术与自动化装置]