Abstract: Compressed sensing (CS) provides a new method for data acquisition. Ahyperspectral images CS
methodology is proposed in this paper. In the proposed framework, hyperspectral images are divided into
several groups, and each group consists of a reference band followed by some common bands. Random measurements of the individual spectral bands are obtained using block CS independently. In image reconstruction,the reference bands are reconstructed with the smoothed projected Landweber algorithm, and the common bands with a new reconstruction algorithm. The algorithm is implemented as follows: 1) Obtain predicted values of the common bands using the spectral bidirectional prediction. 2) Calculate measurement differences using block observation on the predicted values. 3) Reconstruct the images and their corresponding prediction differences in an iterative fashion. This method can improve reconstruction quality as it has fully considered the spectral and spatial correlations. Experimental results reveal that reconstruction performance of the proposed method is substantially superior to that by applying 2-D image reconstruction independently and that of a multiple-vector CS variant method.  

Key words:  hyperspectral image, compressed sensing (CS), block observation, spectral bidirectional prediction, reconstruction