In this paper we define a spectral plot corresponding to one pixel in the scene ofthe 3-dimensional data cube from an imaging spectrometer as a vector for Vector Quantization（VQ）. Then each such vector is encoded into binary codes by the Spectal-Feature-Based CodingScheme （SFBCS）. The codevector searching is performed with the binary codes got by the SFBCS.In this way not only the processing time is greatly sped up （30 times faster than normal VQ whenthe codebook size is 256

and 43 times faster than normal VQ when codebook size is 4096）

but alsothe precision of the reconstructed data is increased. The way of constituting vectors in this papercan well preserve spectral feature

and can make good use of the correlation in spectral domain aswell. With this proposal scheme compression ratios greater than 192: 1

PSNR’s of the coded hyperspectral sequences exceeding 45. 2dB can be achieved.