Hyperspectral Sensors and Applications
29
and spaceborne sensors. By comparing spectra to those retrieved from the
imaging spectrometers, these noise influences can be largely removed and the
data can be converted from units of radiance to that of scale surface reflectance.
The latter represents a simple standardization that allows comparison of spatially and/or temporally diverse data.
The second reason is that field and laboratory spectra can be collated into
spectral libraries which can be useful for a number of purposes. For example,
field spectra of certain mineral assemblages can be used to train the classification of geologically diverse areas and have been used frequently to derive
geological maps (Kruse and Hauff 1991; Kruse et al. 1993).
Spectral libraries can also playa key role in selecting endmembers for subsequently spectral unmixing of hyperspectral data. This process decomposes
individual pixels into their reflectance components based on a knowledge of
and comparison with the spectral characteristics of known target materials.
These known materials, also referred to as endmembers, are then used as references to decompose the pixels. For a more extensive description of this process
see (Adams and Smith 1986; Elvidge 1990; Gaddis et al. 1993; Aber et al. 1994;
Kupiek and Curran 1995; Levesque et al. 2000). Endmembers for spectral decomposition can be extracted by locating "pure" targets in the image through
manual or automated techniques. Such an approach assumes that such pure
targets exist in the image, which may be the case for very fine « 1-2 m) spatial
resolution images but becomes less likely as the spatial resolution coarsens
(esp. mixed pixels). An alternative approach therefore is to use spectrallibraries generated from field-based spectroscopy or even from the finer spatial
resolution data themselves. Much of the theory related to spectral endmembers
has originated from minerals exploration applications with pure targets that
do not vary greatly. However, it should be noted that the spectral signatures of
vegetation are dynamic in spectral-, spatial-, and temporal-space, and should
cautiously be constructed into spectral libraries.
1.S
Software for Hyperspectral Processing
The datasets originating from hyperspectral sensors are complex and cannot
be adequately analyzed using the more traditional image processing packages.
Significant research has therefore been carried out to develop algorithms that
can be incorporated into commercially available software. At the time of writing
two of the more commonly used packages for hyperspectral image processing
have been produced by Research Systems Inc. (RSI), U.S.A. and PCI Geomatics,
Canada. Other examples of software that state hyperspectral capability are
listed in Table 1.5. It should be noted that many of the algorithms used for
hyperspectral imaging applications presented in this book are of an advanced
nature and are still under research and development.
29
and spaceborne sensors. By comparing spectra to those retrieved from the
imaging spectrometers, these noise influences can be largely removed and the
data can be converted from units of radiance to that of scale surface reflectance.
The latter represents a simple standardization that allows comparison of spatially and/or temporally diverse data.
The second reason is that field and laboratory spectra can be collated into
spectral libraries which can be useful for a number of purposes. For example,
field spectra of certain mineral assemblages can be used to train the classification of geologically diverse areas and have been used frequently to derive
geological maps (Kruse and Hauff 1991; Kruse et al. 1993).
Spectral libraries can also playa key role in selecting endmembers for subsequently spectral unmixing of hyperspectral data. This process decomposes
individual pixels into their reflectance components based on a knowledge of
and comparison with the spectral characteristics of known target materials.
These known materials, also referred to as endmembers, are then used as references to decompose the pixels. For a more extensive description of this process
see (Adams and Smith 1986; Elvidge 1990; Gaddis et al. 1993; Aber et al. 1994;
Kupiek and Curran 1995; Levesque et al. 2000). Endmembers for spectral decomposition can be extracted by locating "pure" targets in the image through
manual or automated techniques. Such an approach assumes that such pure
targets exist in the image, which may be the case for very fine « 1-2 m) spatial
resolution images but becomes less likely as the spatial resolution coarsens
(esp. mixed pixels). An alternative approach therefore is to use spectrallibraries generated from field-based spectroscopy or even from the finer spatial
resolution data themselves. Much of the theory related to spectral endmembers
has originated from minerals exploration applications with pure targets that
do not vary greatly. However, it should be noted that the spectral signatures of
vegetation are dynamic in spectral-, spatial-, and temporal-space, and should
cautiously be constructed into spectral libraries.
1.S
Software for Hyperspectral Processing
The datasets originating from hyperspectral sensors are complex and cannot
be adequately analyzed using the more traditional image processing packages.
Significant research has therefore been carried out to develop algorithms that
can be incorporated into commercially available software. At the time of writing
two of the more commonly used packages for hyperspectral image processing
have been produced by Research Systems Inc. (RSI), U.S.A. and PCI Geomatics,
Canada. Other examples of software that state hyperspectral capability are
listed in Table 1.5. It should be noted that many of the algorithms used for
hyperspectral imaging applications presented in this book are of an advanced
nature and are still under research and development.