Hyperspectral Imagery
April 1st, 2004
Stefan A. Robila
robilas@mail.montclair.edu
www.csam.montclair.edu/~robila/RSL/
Source: http://nis-www.lanl.gov/~borel/
Hyperspectral Remote Sensing
a remote sensing technology
seeing characteristics not recognized by the human eye
10 -11 10 -8 10 -6 10
Gamma Rays Ultraviolet Infrared Radio
Non-Imaging Instruments
(example: FieldSpec Hand Held Spectroradiometer)
sensor obtains data (amount of light per wavelength)
computer software displays recorded spectrum
analyze spectral signature
Imaging systems
Scanning radiometers
Passive system
Produces digital images
Imaging systems
Scanning radiometers
Mirror scans across-track (swath)
Imaging Systems
Scanning radiometers
2-D image formed by platform
forward motion
CCD
CCD arrays
Passive system
Line or block of CCDs instead
of scanning mirror
Senses entire swath (or block)
simultaneously
Hyperspectral Remote Sensing
In the Visible-NIR range, water ice and dry ice give characteristic spectral
curves, as shown here:
Hyperspectral Data Analysis
General Approach:
Develop Spectral Library
Construct spectral curve for relatively "pure" materials
Specific reflectance peaks and absorption troughs are
read from these curves.
Compare to lab spectra (mixture analysis)
Mixtures of two or even three different materials can be
identified as the components of the compound spectral
curve.
Hyperspectral Data Analysis
Spectral Libraries:
Sets of hundreds of measured spectra for components likely to
be encountered in the study area.
Spectral Angle
The distance measure used for spectral screening.
Band 2
(x,y) y
Band 1
For two pixel vectors x and y, the spectral angle is computed as:
n
x ,y x i yi
(x, y) cos-1 cos1 i 1
1/ 2
x 2 y 2 1/ 2
x 2 y 2
n n
i i
i 1 i 1
Hyperspectral Data Analysis
Pure Pixel Analysis
Find relatively pure pixels
Pixel Purity Index (PPI)
Pure spectra are spectral endmembers
Endmembers
Spectral characteristics of an image that represent classes of
interest
Usually assigned based on lab spectra
Can be done manually
Hyperspectral Data Analysis
where S is the nxm matrix of spectra (s1, .., sm) of the individual materials (also
called endmembers), a is an m-dimensional abundance vector and w is the
additive noise vector.
a
i 1
i 1
Hyperion
Future Hyperspectral Sensors
Data volume
Cost
Difficulty of analysis
Spectral Libraries
More complex