|
White
Papers
Reprints
and Technical Papers from Applied Analysis Inc.
Click on a title to view an abstract
of the research paper.
Some papers are available as links to other sites and some
are available for download in Adobe PDF format.
"IMAGINE
Subpixel Classifier" Applied Analysis
Inc., May 2003
"IMAGINE
Subpixel Classifier Compared to Linear Spectral Unmixing"
Applied Analysis Inc., May 2003
"Regional Ecosystem Analysis: Puget Sound Metropolitan
Area," American
Forests, 7/25/98.
"Regional Ecosystem Analysis: Chesapeake Bay Region and Baltimore-Washington
Corridor," American
Forests, 3/10/99.
"An Evaluation of the Utility of Sub-Pixel Analysis of
Thematic Mapper Imagery for the Spruce Beetle Outbreak on
the Manti-LaSal National Forest,"
J. Johnson, P. Greenfield, and A. Steve Munson, published
June 23, 1998.
"Automated Scene-Derived Normalization of Spectral Imagery for Subpixel
Classification,"
R. Huguenin, M. Wang, M. Karaska, and K. Roberts; submitted
for presentation at SPIE International Symposium on Optical
Science, Engineering and Instrumentation, July 1998.
"Utilizing Subpixel Spectral Identification Schemes to Address Emerging
Applications Areas,"
C. Erdman, R. Huguenin, and L Scarff; SPIE Vol. 3119.
"Subpixel Classification of Bald Cypress and Tupelo Gum Trees in Thematic
Mapper Imagery,"
R. Huguenin, M. Karaska, D. Van Blaricom, and J. Jensen; Photogrammetric
Engineering & Remote Sensing Vol. 63, pp. 717-725,
June 1997.
"Adaptation of the AASAP (IMAGINE Subpixel Classifier) Analysis Software
for Automated Bathymetry Mapping," R. Huguenin, E. Boudreau, and M. Karaska; presented
at the ERIM Fourth International Conference on Remote Sensing
for Marine and Coastal Environments, Orlando, Florida, 17-19
March 1997.
"Nonparametric Classification of Subpixel Materials in Multispectral
Imagery," E. Boudreau,
R. Huguenin, M. Karaska; SPIE Vol. 2758, 1996.
"Subpixel Analysis Process Improves Accuracy of Multispectral Classifications,"
R. Huguenin, Earth Observation Magazine, July 1994.
"The Silicate Component of Martian Dust," R. Huguenin, Copyright 1987 by Academic Press, Inc.
0019-1035/87.
"Intelligent Information Extraction from Reflectrance Spectra: Absorption
Band Positions," R. Huguenin
and J. Jones; Journal
of Geophysical Research Vol. 91, No. B9, pp. 9585-9598,
August 10, 1986.
"Automated
Subpixel Photobathymetry and Water Quality Mapping,"
R. Huguenin, M. Wang, R. Biehl, S. Stoodley, and J. Rogers,
Photogrammetric
Engineering & Remote Sensing Vol. 70, No. 1, January
2004,
pp. 111-123.
"AVIRIS
Measurements of Chlorophyll, Suspended Minerals, Dissolved
Organic Carbon and Turbidity in the Neuse River, NC,"
M.
Karaska, R. Huguenin, J. Beacham, M. Wang, J. Jensen, and
R. Kaufmann, Photogrammetric Engineering & Remote Sensing
Vol. 70, No. 1, January 2004, pp. 125-133.
"Regional
Ecosystem Analysis: Puget Sound Metropolitan Area," American Forests, 7/25/98.
"Regional Ecosystem Analysis: Chesapeake Bay Region
and Baltimore-Washington Corridor," American Forests, 3/10/99.
Projects
Overview
AMERICAN FORESTS conducted a Regional Ecosystem Analysis of
the Puget Sound area and a Regional Ecosystem Analysis of
Chesapeake Bay Region and Baltimore-Washington corridor to
determine how the landscape has changed over time and assess
the value of the areas' ecology.
A regional
level analysis was conducted of three satellite images spanning
a 24-year period from 1972 to 1996. Landsat Multispectral
and Thematic Mapper images were used to study several thousand
square miles of the watersheds.
The Ecosystem
Analyses uses Geographic Information System (GIS) technology
to measure the changing structure of the landscape and analyze
the scientific and engineering implications of the change.
Neighborhood level computer models were developed using CITYgreen
software, American Forests’ GIS application for calculating
ecosystem benefits. The models represent five typical neighborhood
landscapes and measure the effects of these landscapes on
storm water and air quality.
The purpose
of this project is to document the value of tree-covered landscapes
to urban areas. Furthermore, it provides urban decision makers
with the information and tools they need to measure the value
of natural landscapes and incorporate more trees into future
development.
Please link me to "American Forests Publications."
"An Evaluation of the Utility of Subpixel Analysis of Thematic Mapper
Imagery for the Spruce Beetle Outbreak on the Manti-LaSal National
Forest," J. Johnson,
P. Greenfield, and A. Steve Munson, published June 23, 1998.
Abstract
Large area mapping and monitoring of forest pest and disease
infestations is typically conducted using aerial sketch mapping,
and where necessary, ground survey. Both techniques have limited
utility in wilderness areas where ground access is difficult
and aerial mapping is too costly. The Forest Health Technology
Enterprise Team (FHTET) in cooperation with Forest Health
Protection (FHP), the Manti-LaSal National Forest and the
Remote Sensing Applications Center (RSAC) investigated the
utility of subpixel processing for analysis of Landsat Thematic
Mapper (TM) imagery of a spruce beetle outbreak.
The study
area was on a portion of the Wasatch Plateau area on the Manti-LaSal
NF in east-central Utah. Three dates of imagery were acquired
and processed using Imagine Subpixel Classification software
and the results were compared with existing ground survey
data and aerial sketch map data.
The subpixel
analysis successfully detected spruce mortality, but did not
distinguish between mortality due to spruce beetles versus
other mortality, both pest and non-pest in the study area.
Subpixel analysis can be an effective supplement to other
means of forest health monitoring in species and situations
where the geographic extent of the outbreak is too large for
standard aerial sketch mapping techniques to adequately document
and where the impacted species hold their needles for long
periods of time following attack.
Please
link me to An Evaluation of the Utility of Subpixel Analysis of Thematic Mapper
Imagery for the Spruce Beetle Outbreak on the Manti-LaSal
National Forest."
"Automated Scene-Derived Normalization of Spectral
Imagery for Subpixel Classification," R. Huguenin, M. Wang, M. Karaska, and K. Roberts, submitted
for presentation at SPIE International Symposium on Optical
Science, Engineering and Instrumentation, July 1998.
Abstract
Changing illumination and atmospheric conditions hamper the
automated analysis of spectral imagery. Applied Analysis Inc.
developed an Environmental Correction module as part of its
Subpixel Classifier software. This module derives atmospheric
and sun angle correction factors directly from an image without
the use of predictive models. Subpixel occurrences of dark
and bright surface features are used to characterize atmospheric
radiance, atmospheric attenuation and sensor transfer functions.
A significant
component of each pixel used to derive this information can
be unwanted surface reflectance from sun glint, sky illumination,
or other solar-illuminated terrain materials. These spectral
contributions distort the accurate assessment of atmospheric
radiance, atmospheric attenuation and sensor transfer functions.
By working at a subpixel level, the Subpixel Classifier
software is able to more accurately derive these factors,
resulting in improved environmental correction.
Download "Automated Scene-Derived
Normalization of Spectral Imagery for Subpixel Classification." 
"Utilizing Subpixel Spectral Identification Schemes
to Address Emerging Applications Areas," C. Erdman, R. Huguenin, and L Scarff; SPIE Vol.
3119.
Abstract
The process of extracting information from hyperspectral imagery
datasets provided by newer sensor systems can be enhanced
through a combination of unique spectral processing algorithms.
The first technique we describe is a unique method for extracting
the relevant bands within a hyperspectral dataset;
this set of optimized bands will provide the greatest potential
for discriminating specified materials of interest. The second
process, subpixel spectral identification, uses the results
from the subset of hyperspectral bands to further refine and
distinguish between specific materials of interest, improving
classification accuracy and diminishing false alarms. Comparison
results produced using the full hyperspectral bandset, a six-band
selection chosen based on thematic-mapper band centers, and
the optimized bandset are presented for a test scene
using HYDICE hyperspectral imagery.
Download "Utilizing Subpixel Spectral
Identification Schemes to Address Emerging Applications Areas."
"Subpixel Classification of Bald Cypress and Tupelo
Gum Trees in Thematic Mapper Imagery," R. Huguenin, M. Karaska, D. Van Blaricom, and J. Jensen;
Photogrammetric Engineering & Remote Sensing Vol.
63, pp. 717-725, June 1997
Abstract
A subpixel spectral analytical
process was used to classify Bald Cypress and Tupelo Gum wetland
in Landsat Thematic Mapper imagery in Georgia and South Carolina.
The subpixel process enabled the detection of Cypress and
Tupelo trees in mixed pixels.
Two hundred
pixels were field verified for each tree species to independently
measure errors of omission and commission. The cypress total
accuracy was 89 percent and the tupelo total accuracy was
91 percent. Field investigations revealed that both cypress
and tupelo trees were successfully classified when they occurred
both as pure stands and when mixed with other tree species
and water.
In a comparison
with traditional classification techniques (ISO-DATA clustering,
maximum likelihood, and minimum distance) the subpixel classification
of cypress and tupelo yielded improved results. Large areas
of wetland where cypress was heavily mixed with other tree
species were correctly classified by the subpixel process
and not classified by the traditional classifiers.
Download "Subpixel Classification of Bald Cypress and Tupelo Gum Trees in
Thematic Mapper Imagery."
"Adaptation of the AASAP (a.k.a. Subpixel Classifier)
Analysis Software for Automated Bathymetry Mapping," R. Huguenin, E. Boudreau, and M. Karaska; presented
at the ERIM Fourth International Conference on Remote Sensing
for Marine and Coastal Environments, Orlando, Florida, 17-19
March 1997.
Abstract
The Applied Analysis Spectral
Analytical Process (AASAP) has been adapted for automated
bathymetric analysis. AASAP is a multispectral image processing
software module that performs automated subpixel analysis,
i.e.; it is able to detect spectral contributions from materials
of interest that may occupy only small fractions of image
pixels.
It does this
by identifying and removing unwanted spectral contributions
from background materials in the pixels. This provides a means
for automatically identifying and removing terrain and surface
reflection (sky and sun reflection) contributions from water
pixels. It also allows composite depths and bottom materials
within pixels to be resolved into individual components, e.g.,
shallow coral and deep sand. This enables more accurate determinations
of the water column and bottom reflectance characteristics
to be made.
AASAP provides
the additional advantage of automatically calculating atmospheric
correction factors for the scene being processed. This allows
the attenuated bottom radiance to be converted from units
of digital number into units of calibrated reflectance, providing
a means for automatically calculating depth using a standard
regression analysis of logarithmic reflectance.
It also allows
signatures derived in one scene to be ported to other scenes.
The output of the process includes a pixel fraction and depth
for each bottom material per pixel, as well as the mean depth
and confidence for each pixel.
Download "Adaptation of the AASAP
(Subpixel Classifier) Analysis Software for Automated
Bathymetry Mapping."
"Nonparametric Classification of Subpixel Materials
in Multispectral Imagery,"
E. Boudreau, R. Huguenin, M. Karaska; SPIE Vol. 2758,
1996
Abstract
An effective process for the
automatic classification of subpixel materials in multispectral
imagery has been developed. The Applied Analysis Spectral
Analytical Process (AASAP) isolates the contribution of specific
Materials of Interest (MOI) within mixed pixels.
AASAP consists
of a suite of algorithms that perform environmental correction,
signature derivation, and subpixel classification. Atmospheric
and sun angle correction factors are extracted directly from
imagery, allowing signatures produced from a given image to
be applied to other images. AASAP signature derivation extracts
a component of the pixel spectra that is most common to the
training set to produce a signature spectrum and nonparametric
feature space. The subpixel classifier applies a background
estimation technique to a given pixel under test to produce
a residual. A detection occurs when the residual falls within
the signature feature space.
AASAP was
employed to detect stands of Loblolly Pine in a Landsat TM
scene that contained a variety of species of southern yellow
pine. An independent field evaluation indicated that 85% of
the detections contained over 20% Loblolly, and that 91% of
the known Loblolly stands were detected. For another application,
a crop signature from a scene in Texas detected occurrences
of the same crop in scenes from Kansas and Mexico. AASAP has
also been used to locate subpixel occurrences of soil contamination,
wetlands species, and lines of communication.
Download "Nonparametric
Classification of Subpixel Materials in Multispectral Imagery."
"Subpixel Analysis Process Improves Accuracy of Multispectral
Classifications,"
R. Huguenin, Earth Observation Magazine, July 1994.
Synopsis
The reprint provides an overview
of how most classification applications can benefit from the
Subpixel Classifier’s capabilities. Subpixel analysis
is relevant largely because image pixels that contain units
or features of interest are, with rare exception, "mixed pixels,"
i.e.; they contain not only the unit of interest but also
other features that contribute to the spectral qualities of
the pixel.
Two
applications are discussed. The first illustrates the performance
of the Subpixel Classifier in "mixed pixel" environment; the
other illustrates the ease-of-use of the Subpixel Classifier
as a natural extension of conventional multispectral classifiers.
The conclusion is that for most applications, the Subpixel
Classifier "will enable analysts to improve the accuracy of
their current projects by making more complete detections¼ and generate more discriminating
classifications."
Download "Subpixel Analysis Process
Improves Accuracy of Multispectral Classifications."
"The Silicate Component of Martian Dust," R. Huguenin, Copyright 1987 by Academic Press, Inc.
0019-1035/87.
Abstract
Absorption features in telescopic
reflectance spectra of Mars during 1978 were detected and
analyzed. Also detected and analyzed were absorption features
in Mariner 7 Infrared Spectrometer and Mariner 9 Infrared
Interferometer Spectrometer spectra. The Bands Data Analysis
System described by R. L. Huguenin and J. L. Jones (1986,
J. Geophys. Res. 91, 9585-9598) was used.
Atmospheric
CO2 bands were all detected with an average error of 50 cm-1,
providing a test of the sensitivity and accuracy of feature
from the Mars data. Absorption features that were attributed
to H2O ice were detected in the North Polar Cap region, as
well as in regions to the north and east of Hellas basin,
and near the Elysium Montes.
Additional
absorptions were assigned to structural hydroxyl within a
strongly hydrogen-bonded acidic material. Features in the
Mariner 9 spectra suggested that the material may be a silicate.
Hydroxyl stretch fundamentals were deduced to occur at 2661
and 2824 cm-1, consistent with acidic material having strong
hydrogen bonding. In-plane and out-of-plane structural hydroxyl
deformation fundamentals were proposed to occur at 1498 and
909 cm-1, respectively, from which a hydrogen bridge length
of ~2.4 Å was derived.
Si-Ob (bridging
oxygen) and Si-Ot (terminal oxygen) stretch fundamentals were
deduced to occur at 1176 and 995 cm-1, respectively. Si-Ob
and Si-Ot asymmetric bend fundamentals were deduced to occur
near 482 and 397 cm-1, respectively, while a Si-Ob-Si deformation
fundamental was proposed to occur near 697 cm-1. These fundamentals
suggested that the Si-Ob-Si bond angle may be approximately
linear (170o) and that the estimated Si-Ob bridge length may
be ~1.61 Å.
The totally
symmetric silicate stretch fundamental was deduced to occur
near 894 cm-1, from which silicate polymerization equivalent
to O/Si = 3.8 ± 0.2 was derived. This is consistent
with the derived Si-Ob-Si bond angle and bridge length. An
in-plane M-OH libration fundamental near 678 cm-1 was derived,
consistent with Ca2+ and/or Mg2+ being the dominant cations
in the vicinity of OH-. An analog compound that has a very
similar set of structural hydroxyl, silicate, and librational
fundamentals is Ca2(HSiO4)OH, with principal differences proposed
to be due to the 0.2 ± 0.2 lower O/Si ratio and the
0.08-Å shorter hydrogen bridge length in the Mars material.
Viking compositional
data suggest that Mr2+, rather than Ca2+, may be the dominant
cation in the Mars material. The differences from the analog
phase properties suggest that the material may be similar
to H-forsterite (Fo79). It is proposed that the silicate component
of the dust may be an incipient alteration (hydrolysis) product
of the olivine-rich ultramafic or mafic material, involving
a process that resulted in minimum loss of mobile cations
and that preserved the high O/Si ratio of the starting material.
Download " The Silicate Component of
Martian Dust"
"Intelligent Information Extraction from Reflectance
Spectra: Absorption Band Positions," R. Huguenin and J. Jones; Journal of Geophysical
Research Vol. 91, No. B9, pp. 9585-9598, August 10, 1986.
Abstract
A multiple high-order derivative
analysis algorithm has been developed that automatically extracts
absorption band positions from reflectance spectra. Absorption
band positions occur where the fifth derivative of the spectrum
equals zero, the fourth derivative has a positive sign, and
the second derivative is negative.
The algorithm
assumes that bands are approximately symmetric about the band
center. Continuum contributions, phase angle effects, and
broad low-frequency calibration errors are suppressed. Overlapping
bands with centers as close as 0.2-0.5W (full band width at
half maximum intensity) can be resolved, as long as bands
have comparable widths and intensities.
If overlapping
bands are dissimilar, band center separations of 0.6-1.0W
are safer limits of resolution. Results are relatively insensitive
to whether constituent bands convolve additively or multiplicatively.
Spectral resolution can be moderately low, requiring only
four to eight data points per W. Errors of derived band centers
are <3%W for separations greater than 0.6-1.0W. For overlapping
bands with widths of a few thousand cm-1 errors would be typically
less than 150 cm-1 from actual band positions.
The band
detection algorithm is sensitive to noise, and data smoothing
is required. The segment length for smoothing (number of points
averaged) needs to be continually adjusted to ~0.5W to minimize
signal distortion. A spectral pattern recognition algorithm,
which statistically characterizes the frequency distribution
of intensity variations in a sliding segment across the spectrum,
can be used to predetect the signal spectrum (low-frequency
components of the sliding intensity distributions) and to
calculate approximate W (predetected W) across the spectrum
using the second derivative.
An intelligent
control algorithm can then continuously locally adjust the
segment lengths for smoothing to 0.5W (predetected W). Smooths
are repeated (typically, 20-30 times) until the high-frequency
components of the sliding intensity variation distributions
across the spectrum are suppressed. A single-pass cubic spline
is applied to the smoothed data. The intelligent control algorithm
then applies the multiple high-order derivative algorithm.
A sliding
segment sixth-order polynomial is fit to the spectrum, with
the length of the segment being continuously locally adjust
to 1.0W (predetected W) across the spectrum. Adjustment of
the segment length to ~1.0W insures that the signal spectrum
is minimally distorted and that weak features are not suppressed.
Derivatives are calculated for the center point of the sliding
segment using the coefficients of the sixth-order polynomial.
The system
has successfully extracted band positions from low-quality
(6% peak-to-peak noise) synthetic spectra with relatively
little degradation of accuracy. Application to natural laboratory
and earth-based telescope spectra displayed good reliability
and consistency. Processing is fully automated, and the same
standardized procedure is followed for all spectra. No continuum
removal or band modeling is needed. The automation of analysis
could potentially significantly increase the efficiency and
yield of information extraction, particularly for high rate
repetitive scan laboratory and synoptic remote sensing spectroscopy
applications.
Download "Intelligent Information Extraction
from Reflectance Spectra: Absorption Band Positions."
|
