Introduction
3
Fig. I. The hyperspectral cube (source: http://aviris.jpl.nasa.gov/htmllaviris.cube.html). For
a colored version of this figure, see the end of the book
gies have made the hyperspectral imaging technology more readily available.
Much like improvements in spectral resolution, spatial resolution has also been
dramatically increased by the installation of hyper spectral sensors on aircrafts
(airborne imagery), opening the door for a wide array of applications.
Nevertheless, the processing of hyper spectral data remains a challenge since
it is very different from multispectral processing. Specialized, cost effective
and computationally efficient procedures are required to process hundreds
of bands acquiring 12-bit and 16-bit data. The whole process of hyper spectral
imaging may be divided into three steps: preprocessing, radiance to reflectance
transformation and data analysis.
Preprocessing is required for the conversion of raw radiance into at-sensor
radiance. This is generally performed by the data acquisition agencies and the
user is supplied with the at-sensor radiance data. The processing steps involve
operations like spectral calibration, geometric calibration and geocoding, signal to noise adjustment, de-striping etc. Since, radiometric and geometric
accuracy of hyperspectral data vary significantly from one sensor to the other,
the users are advised to discuss these issues with the data providing agencies
before the purchase of data.
Further, due to topographical and atmospheric effects, many spectral and
spatial variations may occur in at-sensor radiance. Therefore, the at-sensor
data need to be normalized in the second step for accurate determination of
the reflectance values in difference bands. A number of atmospheric models
and correction methods have been developed to perform this operation. Since
the focus of this book is on data analysis aspects, the reader is referred to (van
der Meer 1999) for a more detailed overview of steps 1 and 2.
3
Fig. I. The hyperspectral cube (source: http://aviris.jpl.nasa.gov/htmllaviris.cube.html). For
a colored version of this figure, see the end of the book
gies have made the hyperspectral imaging technology more readily available.
Much like improvements in spectral resolution, spatial resolution has also been
dramatically increased by the installation of hyper spectral sensors on aircrafts
(airborne imagery), opening the door for a wide array of applications.
Nevertheless, the processing of hyper spectral data remains a challenge since
it is very different from multispectral processing. Specialized, cost effective
and computationally efficient procedures are required to process hundreds
of bands acquiring 12-bit and 16-bit data. The whole process of hyper spectral
imaging may be divided into three steps: preprocessing, radiance to reflectance
transformation and data analysis.
Preprocessing is required for the conversion of raw radiance into at-sensor
radiance. This is generally performed by the data acquisition agencies and the
user is supplied with the at-sensor radiance data. The processing steps involve
operations like spectral calibration, geometric calibration and geocoding, signal to noise adjustment, de-striping etc. Since, radiometric and geometric
accuracy of hyperspectral data vary significantly from one sensor to the other,
the users are advised to discuss these issues with the data providing agencies
before the purchase of data.
Further, due to topographical and atmospheric effects, many spectral and
spatial variations may occur in at-sensor radiance. Therefore, the at-sensor
data need to be normalized in the second step for accurate determination of
the reflectance values in difference bands. A number of atmospheric models
and correction methods have been developed to perform this operation. Since
the focus of this book is on data analysis aspects, the reader is referred to (van
der Meer 1999) for a more detailed overview of steps 1 and 2.