Endmember extraction used for hyperspectral imagery loss compression

One of the problems limiting the utility of hyperspectral imagery is how to compress the large number of data effectively. The current methods cannot resolve the problem of the contradiction between large compression rate and spectral information veracious reservation, even the best loss compression method can not bring the satisfying result. The paper presented a loss compression method based on the endmember extraction technology, so as to resolve the contradiction between large compression ratio and spectrum preserved accurately. The endmembers were obtained with vertex component analysis (VCA) and the fractions of them were estimated based on the proportion of cosine angle similitude between endmembers and observed spectrum. The endmembers spectrum and fraction were compressed with the lossless compression method and JPEG2000 loss compression method was used for all of the hyperspectral single-band images to increase compression ratio. The experiment on the AVIRIS data shows that compression ratio was increased greatly and the spectra were resumed effectively. When the compression ratio is 50:1, the spectrum angle loss is about 2% for most pixels.