Spatio-spectral hybrid compressive sensing of hyperspectral imagery

The amount of data typically captured with hyperspectral imaging systems measuring the light reflected by the Earth surface in hundreds or thousands of spectral bands is very large. The huge size of hyperspectral data cube has motivated the development of compressive sensing (CS) techniques for hyperspectral imagery. In this letter, we proposed an efficient CS scheme, spatio-spectral hybrid CS, to fully exploit the high degree of correlation of hyperspectral data based on linear mixture model. The main contribution of this letter lies in (1) rephrasing the CS acquisition of hyperspectral data as a spatial and spectral hybrid random measurement problem and (2) proposing a recovery approach to estimate both the endmember signatures and abundance fractions matrix (and thus the whole data set) from the compressed measurements instead of solving underdetermined problem of standard CS reconstruction. In a series of experiments with real data, we show that the proposed scheme can achieve significant reconstruction performance. In addition, as a by-product, endmember signatures and their corresponding abundance fractions are obtained directly.