Recent advances in sensor technology have led to the development of hyperspectral sensors capable of collecting imagery containing several hundred bands over the spectrum. However, the increase in the number of bands is both a blessing and a curse. The large number of bands provides the opportunity for more materials to be discriminated by their respective spectral response. However, this large number of bands is the characteristic which leads to complexity in analysis techniques. The techniques described in the following sections are those which are widely used by the USGS, NASA's Jet Propulsion Laboratory, ENVI, and others. There are, however, other methods and algorithms to extract information from hyperspectral sensors.

One difficulty in working with hyperspectral data is to understand the differences associated with working in n-dimensional space. One must be careful about using two or three dimensional conceptual truths as a basis for conclusions in higher dimensional spaces [Langrebe, 1997].

Radiometric Correction

Hyperspectral imaging sensors collect radiance data from either airborne or spaceborne platforms which must be converted to apparent surface reflectance before analysis techniques can take place. Atmospheric correction techniques have been developed that use the data themselves to remove spectral atmospheric transmission and scattered path radiance. There are seven gases in the Earth's atmosphere that produce observable absorption features in the 0.4 - 2.5 micron range. They are
Approximately half of the 0.4 - 2.5 micron spectrum is affected by gaseous absorption is illustrated below in Figure 1. For this reason, the ATREM 3.0 (Atmosphere Removal Program) developed by the Center from the Study of Earth from Space (CSES) at the University of Colorado can be used to remove the effects of the atmosphere from AVIRIS or HYDICE imagery. ATREM is available via anonymous ftp at cses.colorado.edu from the pub/atrem directory.


Solar Spectrum with Atmospheric Absorbtions



The ATREM software was developed to determine the scaled surface reflectance from hyperspectral imagery from both AVIRIS and HYDICE sensors. The atmospheric scattering used by ATREM is modeled after the MODTRAN 5S code. The ATREM software assumes that the surface is horizontal and has a Lambertian reflectance. If topography is known, then the scaled surface reflectance can be converted into real surface reflectance.

The ATREM model is a good approximation to radiometric correction of the imagery. However, calibration of the ATREM surface reflectance to in situ measurements should improve the final results. A by-product from the ATREM software is an image of the columnar water vapor which was removed from the input hyperspectral data. The two figures below represent an AVIRIS frame prior to the ATREM correction and a water vapor scene removed from an AVIRIS scene which was acquired over the Kennedy Space Center on March 23, 1996. The images show a significant amount of water vapor removed from the imagery which causes attenuation of the upwelling radiance.

Original AVIRIS data over KSC (Bands 20, 29, 40) Columnar water vapor image removed from AVIRIS data using ATREM program.


Minimum Noise Fraction (MNF) Transformation

While hyperspectral imagery is capable of providing a continuous spectrum ranging from 0.4 to 2.5 microns (in the case of AVIRIS) for a given pixel, it also generates a vast amount of data required for processing and analysis. Due to the nature of hyperspectral imagery (i.e. narrow wavebands), much of the data in the 0.4-2.5 micron spectrum is redundant.

A minimum noise fraction (MNF) transformation is used to reduce the dimensionality of the hyperspectral data by segregating the noise in the data. The MNF transform is a linear transformation which is essentially two cascaded Principal Components Analysis (PCA) transformations. The first transformation decorrelates and rescales the noise in the data. This results in transformed data in which the noise has unit variance and no band to band correlations. [ENVI] The second transformation is a standard PCA of the noise-whitened data.

For this particular example, an AVIRIS frame over the Kennedy Space Center was radiometrically corrected using ATREM and a MNF tranformation was performed on the ATREM-corrected imagery. In this particular frame, the first 14 eigenvectors of the MNF transformation contain coherent information which can be used for further processing.

Eigenvectors 1, 2, & 3 of MNF Transform Data Eigenvectors 6, 9, & 12 of MNF Transform Data



Pixel Purity Index

The Pixel Purity Index (PPI) is a processing technique designed to determine which pixels are the most spectrally unique or pure. Due to the large amount of data, PPI is usually performed on MNF data which has been reduced to coherent images. The most spectrally pure pixels occur when there is mixing of endmembers. The PPI is computed by continually projecting n-dimensional scatterplots onto a random vector. The extreme pixels for each projection are recorded and the total number of hits are stored into an image. These pixels are excellent candidates for selecting endmembers which can be used in subsequent processing.