高光谱图像的分布式压缩感知成像与重构

根据高光谱数据的特点,提出了一种基于像元的分布式压缩采样模型来实现高光谱图像的有效压缩采样与重构。搭建了能实现该模型的压缩采样光谱成像系统,并研究了用于该系统成像的重构算法。在图像采集阶段,将高光谱数据分为参考像元和压缩感知像元;地面像元的辐射能通过棱镜进行谱带分离,再利用数字微镜器件实现谱带的线性编码。对压缩感知像元进行低采样率的线性编码,对参考像元进行采样率为1的线性编码。压缩采样数据重构时,不再采用传统方法直接重构高光谱数据,而是利用线性混合模型将重构高光谱数据转换成端元提取和丰度估计,然后根据重构的端元和丰度恢复原数据。对比实验表明,在压缩采样数据为总数据的20%时,重构的平均信噪比提高了10dB。所设计的成像系统应用压缩感知理论减少了采集的数据量,采样方式简单,可应用于星载或机载的高光谱压缩感知成像。

Distributed compressive sensing imaging and reconstruction of hyperspectral imagery

According to the characteristics of high spectral data, a distributed compressed sampling model based on pixels was proposed to realize the efficient compressive sampling and reconstruction. A spectral imaging system based on distributed compressed sampling was established and a reconstruction algorithm for this system was investigated. In the image acquisition stage, the hyperspectral data were divided into key pixels and compressive sensing pixels. The ground pixels were separated along the spectral direction by a prism. Then, the linear encoding between the spectral bands was realized by a digital micro-mirror device. The compressive sensing pixels were coded with a low sampling rate, and the key pixels were coded by a sampling rate of 1. In the reconstruction of the compressive sampled data, the traditional compressive sensing reconstruction methods which recover hyperspectral data directly were abandoned. However, the linear mixed models were used to convert the hyperspectral data reconstruction into an endmember extraction and an abundance estimation, then, the hyperspectral data were recovered by using the extracted endmember and estimated abundance. The comparison experiments show that the reconstruction average signal noise rate by proposed algorithm is improved about 10 dB when the used data are 20% that of total data. The system is suitable for the spaceborne or airborne hyperspectral compressive sensing imaging for its less data collected and simple sampling method.