A new approach to hyperspectral data compression using rational function approximation for spectral response curve fitting

Regarding to enormous data volumes of hyperspectral sensors containing hundreds of spectral bands and their very high between-band correlation, compression of this data type is an interesting issue for researchers. Since spectral information of hyperspectral image cube is more crucial than its spatial information, compression techniques must be able to preserve this information. In this paper a rational fraction function approximation approach is considered for spectral response curve fitting of each pixel of hyperspectral image. Coefficients of numerator and denominator are saved and considered as new features for signal representation. Results show that the proposed method provides good compression rates and the original data can be reconstructed in a good way. In addition, our method is applied to each pixel of hyperspectral individually and parallel implementation of it is possible.