Hyperspectral data compression based upon the principal component analysis

The paper is aimed at developing an algorithm of hyperspectral data compression that combines small losses with high rate. relies on a principal component analysis and method exhaustion. components are singular vectors initial signal matrix, which found by the A retrieved matrix formed in parallel. process continues until required retrieval error attained. described detail input output parameters specified. Testing performed using AVIRIS (Airborne Visible-Infrared Imaging Spectrometer). Three images differently looking sky (clear sky, partly clouded overcast skies) analyzed. For each image, testing for all spectral bands set from water-vapour absorption excluded. Retrieval errors versus rates presented. formulas include root mean square deviation, noise-to-signal ratio, structural similarity index, relative deviation. It shown decrease more than order magnitude if gas disregarded. reason weak signals measured great errors, leading to dependence between spectra different spatial pixels. cosine distance pixels suggested be used assess image compressibility.