U.S. patent application number 14/010467 was filed with the patent office on 2014-08-21 for hyperspectral image dimension reduction system and method.
The applicant listed for this patent is RAYTHEON COMPANY. Invention is credited to Ian S. Robinson.
Application Number | 20140233857 14/010467 |
Document ID | / |
Family ID | 40266163 |
Filed Date | 2014-08-21 |
United States Patent
Application |
20140233857 |
Kind Code |
A1 |
Robinson; Ian S. |
August 21, 2014 |
HYPERSPECTRAL IMAGE DIMENSION REDUCTION SYSTEM AND METHOD
Abstract
Provided is a method of hyperspectral image dimension reduction.
The method includes receiving a hyperspectral image having a
plurality of pixels. A set of basis vectors is established at least
in part with respect to the spectral vectors of the initial
hyperspectral image. For each pixel of the hyperspectral image, the
spectral vector is read and decomposed, i.e. unmixed, with the
basis vector set to provide at least a reduced dimension vector for
each pixel. Collectively the reduced dimension vectors for each
pixel represent the dimensionally reduced image. A system operable
to perform the method is also provided.
Inventors: |
Robinson; Ian S.; (Redondo
Beach, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
RAYTHEON COMPANY |
Waltham |
MA |
US |
|
|
Family ID: |
40266163 |
Appl. No.: |
14/010467 |
Filed: |
August 26, 2013 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11856588 |
Sep 17, 2007 |
8538195 |
|
|
14010467 |
|
|
|
|
Current U.S.
Class: |
382/191 |
Current CPC
Class: |
G06K 9/0057 20130101;
G06K 2009/00644 20130101; G06K 9/6232 20130101; G06K 9/0063
20130101 |
Class at
Publication: |
382/191 |
International
Class: |
G06K 9/00 20060101
G06K009/00 |
Claims
1-33. (canceled)
34. A method of hyperspectral image dimension reduction,
comprising: receiving a hyperspectral image having a plurality of
pixels; establishing a basis vector (BV) set; for each of the
plurality of pixels: reading a spectral vector from a subject
pixel; unmixing the spectral vector with the BV set resulting in a
set of alpha coefficients not constrained in value, a reduced
dimension vector of the subject pixel being composed of the set of
alpha coefficients; and providing a dimension reduced image
including the reduced dimension vectors for each pixel.
35. The method of claim 34 wherein providing the reduced dimension
vector for the subject pixel includes: weighting basis vectors from
the BV set with the set of alpha coefficients not constrained in
value; combining the weighted basis vectors; and forming a residual
vector based on the combination and the subject pixel.
36. The method of claim 34, wherein the BV set is provided with the
dimension reduced image.
37. The method of claim 34, wherein unmixing the spectral vector
with the BV set further includes providing a residual vector for
the pixel.
38. The method of claim 37, wherein at least the magnitude of the
residual vector is also provided with the dimension reduced
image.
39. The method of claim 36, wherein the basis vector set is
provided with the dimension reduced image.
40. The method of claim 34, wherein the dimension reduction is
performed as a lossy process.
41. The method of claim 34, wherein the dimension reduction is
performed as a lossless process.
42. The method of claim 34, wherein the dimension reduced image
includes an alpha vector and a residual vector for each pixel.
43. The method of claim 34, wherein establishing a basis vector
(BV) set includes receiving at least one pre-determined basis
vector.
44. The method of claim 34, wherein establishing a basis vector
(BV) set includes: selecting an initial pixel; reading the spectral
vector for the initial pixel, the spectral vector set as an initial
basis vector (BV) of the set; and for each of the remaining
plurality of pixels: reading a spectral vector from the pixel;
unmixing the spectral vector with the BV set to determine a
residual vector; and in response to a magnitude of the residual
vector being greater than a threshold, adding the spectral vector
to the BV set.
45. The method of claim 44, wherein the threshold is provided by an
operator along with the hyperspectral image.
46. The method of claim 44, wherein the BV set has a pre-determined
maximum number of members, and in response to an attempt to exceed
the maximum number, optimizing the set to maintain the maximum
number of spectral vectors.
47. The method of claim 46, wherein the pre-determined maximum
number of members is provided by an operator along with the
hyperspectral image.
48. The method of claim 46, wherein the BV set is optimized to
maintain the pre-determined maximum number of members.
49. A method of hyperspectral image dimension reduction,
comprising: receiving a hyperspectral image having a plurality of
pixels; establishing, incrementally, a basis vector (BV) set having
a pre-determined maximum number of members, such that members of
the BV set having lesser residual vector magnitudes are replaced by
basis vectors from pixels yielding greater residual vector
magnitudes; for each of the plurality of pixels: reading a spectral
vector from the pixel; and unmixing the spectral vector with the BV
set to provide a reduced dimension vector for the pixel and the
residual vector magnitude for the pixel; and providing a dimension
reduced image comprising the reduced dimension vectors for each
pixel.
50. A system for hyperspectral image dimension reduction,
comprising: one or more processors configured to: receive a
hyperspectral image having a plurality of pixels; establish a basis
vector (BV) set; for each of the plurality of pixels: read a
spectral vector from a subject pixel; unmix the spectral vector
with the BV set resulting in a set of alpha coefficients not
constrained in value, a reduced dimension vector of the subject
pixel being composed of the set of alpha coefficients; and provide
a dimension reduced image including the reduced dimension vectors
for each pixel.
51. A system for hyperspectral image dimension reduction,
comprising: one or more processors configured to: receive a
hyperspectral image having a plurality of pixels; establish,
incrementally, a basis vector (BV) set having a pre-determined
maximum number of members, such that members of the BV set having
lesser residual vector magnitudes are replaced by basis vectors
from pixels yielding greater residual vector magnitude; for each of
the plurality of pixels: read a spectral vector from the pixel; and
unmix the spectral vector with the BV set to provide a reduced
dimension vector for the pixel and the residual vector magnitude
for the pixel; and provide a dimension reduced image comprising the
reduced dimension vectors for each pixel.
Description
FIELD
[0001] This invention relates generally to the field of
hyperspectral image processing, and in particular to a system and
method for hyperspectral image dimension reduction such as may be
desired for data transmission or simplified processing.
BACKGROUND
[0002] A black and white photograph of an object or a geographic
area is a two dimensional construct of the actual image or
area--for each X and Y coordinate in the image there is a single
value blackness or whiteness of that particular image spot. As
human beings, the eye can perceive useful information about objects
or areas based on the differences between black, white, and the
shades of gray.
[0003] Color photographs add more visual information, but for most
purposes the color information that is represented is tied to the
visual spectrum. For each X and Y coordinate in the image there is
an approximation of the visual color spectrum of that particular
image spot created through the blending of three color values, such
as for example Red Green and Blue.
[0004] Multi spectral sensing systems such as the Landsat Thematic
Mapper remote imager and weather satellites produce images with a
few relatively broad wavelength bands. The imager may capture a
visual spectrum image and also one in infrared, but still they are
limited in their ability to perceive information that may otherwise
be present in a different part of the spectrum.
[0005] Hyperspectral sensors, on the other hand, collect image data
across dozens if not hundreds of spectral bands, combining the
technology of spectroscopy and remote imaging. The measurements
captured by hyperspectral sensors make it possible to derive a
contiguous spectrum for each image pixel. In other words for each X
and Y coordinate of a pixel, rather than a single value for a gray
or visible color, there is a third dimension--a vector, providing
distinct information for that particular pixel across the large
spectrum of wavelengths.
[0006] As different materials will reflect wavelengths of visible
and invisible light selectively, analysis of the contiguous
spectrum permits finer resolution and greater perception of
information contained in the image, by separating and evaluating
different wavelengths. For example, inorganic materials such as
minerals, chemical compositions and crystalline structures control
the shape of the spectral curve and the presence and positions of
specific absorption bands.
[0007] The spectral reflectance curves of healthy green plants also
have a characteristic shape that is dictated by various plant
attributes, such as the absorption effects from chlorophyll and
other leaf pigments. Leaf structure varies significantly between
plant species, and can be affected by plant stress. Therefore
species type, plant stress and canopy state can all affect near
infrared reflectance measurements, which are captured by
hyperspectral sensors.
[0008] In addition, for a given pixel, a combination of different
materials, e.g., biological, chemical, mineral, will provide a
composite signal. Upon analysis, and through comparison to known
signal waveforms it is frequently possible to derive the presence
of materials within a pixel, and therefore appreciate a detection
granularity that is greater than the actual pixel resolution.
[0009] Hyperspectral sensors providing hyperspectral imaging can
therefore be beneficially applied in a wide array of practical
applications. Examples of such uses include aid in the detection of
chemical or biological weapons, bomb damage assessment of
underground structures, drug production and cultivation, as well as
the detection of friend or foe troops and vehicles beneath foliage
or camouflage.
[0010] With each pixel having information for a spectrum of
wavelengths, it is not surprising that hyperspectral images are
represented by vast quantities of data. As the hyperspectral
sensors are often located remotely, such as in satellites or in
aircraft, the transmission of the data can be an issue as the
volume of data can be enormous and transmission of such data to
ground stations typically involve a radio transmission having a
limited frequency bandwidth.
[0011] More specifically, the technical increases in hyperspectral
sensors to capture images with greater spatial content, increased
resolution, and/or across greater wavelength spectrums has raised
the total data volume to the point where data latency is a problem
and a limiting factor in the distribution of near real time
information. When the imaged area corresponds to a disaster area,
such as coast lines ravaged by a tsunami, a combat zone, or a
terrorist strike upon a target, real time transfer and processing
is key to the prevention of loss of life and property.
[0012] In addition, for some applications there is great advantage
to the data being processed at the location where the data is
collected and for the processing results and/or selected subsets of
the data to be transmitted to users. Given the size of the data set
corresponding to a hyperspectral image, in many instances localized
processing may significantly tax localized resources and/or lead to
an even greater delay in transmitting the data to a user.
[0013] One option to increase transmission rate is to increase the
usable communications bandwidth. Such an increase typically
requires additional power at the transmission source For existing
satellites and even future satellites where power sources are
limited, increasing bandwidth is not ideal. Another option to
increase transmission rate is to actually reduce the volume of data
that is being transmitted. A common measure for such reduction is
compression, which may be performed in one of two forms--lossless
and lossy.
[0014] A related function is data dimensionality reduction (DIMRED)
DIMRED is similar to compression in that the volume of data is
reduced. It differs from compression in that the intent of DIMRED
is to enable further processing of the data (e.g. to detect
targets, to measure changes in successive images of the same scene)
without decompressing the data set. As with compression, however,
contemporary methods for data dimension reduction suffer drawbacks
that are often quite undesierable.
[0015] Typically, prior art methods DIMRED methods include forming
a data set of many fewer spectral bands by merging or selecting a
subset of spectral bands of a hyperspectral data cube or the use of
principal components analysis (PCA) to create basis vectors. In the
latter case the eigenvectors and corresponding eigenvalues of the
hyperspectral data set are computed. B basis vectors are selected
by choosing the B eigenvectors with the B largest eigenvalues.
Other techniques have been employed to reduce scene dimension based
on linear unmixing.
[0016] It is well known that most hyperspectral images can be well
represented by the linear superposition of the spectra of the
mixture of materials found in each spatial pixel. As such, if the
spectra for those materials are known then a linear algebra
technique known as unmixing can be applied to find the abundance of
each spectrum in each pixel. The abundance or umnixing coefficients
can then be used to represent the information content of each
pixel.
[0017] However, and quite importantly, most of the time the
constituent spectra of the scene are not known a priori. Some prior
art algorithms attempt to compute these pure material spectra from
the spectra of mixed pixels in the scene. Other prior art has found
mathematical basis functions that can be superimposed to represent
each pixel. Examples of such methods include: the Orassis algorithm
that fits an optimal convex hull around the high dimensionality
data trying to find the endmembers or vertices of the hull; the
Nfindr technique by Winter attempts to find the spectrally purest
pixels and make them endmembers; and the Max D technique from RIT
which is similar to NFINDR. The commercial tool ENVI offers various
versions of the Principal Components Analysis (PCA).
[0018] The pixel purity and endmember techniques have degraded or
limited performance when pure pixels are not found in the scene and
offer less efficient reduction/compaction than Principal Components
and the present invention. They also do not control either the
average pixel error or the worst case pixel error; the latter being
very important if subsequent processing is going to detect
anomalies.
[0019] The Residual correlation method from BAE applies endmember
selection criteria that are subjective and rely on lab or a priori
spectra. ENVI and other processing tools offer analysts the option
to select pixels from a scene manually to be used as endmembers.
Analyst selections are subjective and do not control the errors in
a predictable or optimal fashion.
[0020] An overview of many of these techniques can be found in the
following references. Winter, Michael E., "N-FINDR: an Algorithm
for Fast Autonomous Spectral End-member Determination in
Hyperspectral Data", Proc of SPIE Vol 3753, Imaging Spectrometry V
(Descour and Shen editors) pp 266-277, 1999; Keshava, Nirmal, A
Survey of Spectral Unmixing Algorithms, LINCOLN LABORATORY JOURNAL
VOLUME 14, NUMBER 1, 2003; Robinson Ian S., et al., "Target
Detection Using HSI Systems: An Update Of TRW Results ", in Imaging
Spectrometry IV, S Shen and M R Descour, Eds., SPIE Vol 3438, San
Diego, Calif., July 1998; and McGregor, Dan. et al., Comparison of
Hyperspectral Sub-Pixel Detection With and Without A Priori
Knowledge of Target Features, 1998 IEEE Aerospace Conf. Proc., Vol.
5, Snow Mass Colo., March 1998.
[0021] Moreover, PCA analysis is generally effective in creating
basis vectors that span the scene in a mean-square sense. However,
PCA does not directly control the error levels of fidelity of
individual spectra. Thus it is possible, and even likely, to incur
a large error in a small fraction of the scene, especially when and
where the pixels are uncommon. This error can be extremely
detrimental when the detection problem is for spectrally unique
objects that sparsely populate the scene. In addition, PCA analysis
typically requires high overhead including computation and
inversion of the spectral covariance matrix.
[0022] Lossless data compression is a class of data compression
algorithms that permits the extraction of the original data from
the compressed data to be complete, i.e., the restored data is
exactly the same as the original data. Lossless algorithms
typically are categorized according to the type of data that they
are designed to compress, such as text, images and sound. Zip and
gzip file formats are common lossless applications for data, and
PNG and GIF are frequently used for images.
[0023] While in principle, any general-purpose lossless compression
algorithm can be used on any type of data, many lossless algorithms
are unable to achieve significant compression, especially on data
that is not of the form they were designed to compress. There is no
guaranteed compression factor for lossless compression. In certain
cases the output from such a lossless algorithm may actually be
larger then the original.
[0024] Lossy data compression is a class of data compression
algorithms where compressing data and then decompressing provides a
new data stream that may well be different from the original, but
close enough to be useful for one or more applications. Lossy
methods are frequently employed rather than lossless methods
because the compressed file size is far smaller while still meeting
the requirements of the application.
[0025] Common examples of lossy compressed data files are JPEG
image files and MP3 audio files. JPEG is designed specifically to
provide images that respond well to the human eye. With JPEG, the
image is converted from RGB (Red Green Blue) into the YCbCr color
space, wherein Y represents the brightness of a pixel, the Cb and
Cr respectively represent chrominance as split into blue and red
components.
[0026] As the human eye sees more detail in the Y component of
brightness then in the Cb (blue) and Cr (red), JPEG downsamples the
Cb and Cr elements. JPEG also performs a normalized,
two-dimensional type II discrete cosine transform to further reduce
the data. MP3 compression likewise removes data that is not likely
to be perceived by the human user. However, the removal of data in
both JPEG and MP3 compression methods does result in new files that
although acceptable to human users may not be acceptable for
computer processing and analysis. More specifically, JPEG, MP3 and
other lossy compression methods may well remove information that is
valuable and desired, but potentially not realized to be so until
processing and analysis is attempted.
[0027] As hyperspectral image data is quite large due to the
wavelength spectrum captured, most lossless compression methods
have not proven to be truly effective in balancing compression
desirability while maintaining important data. This is especially
true where a factor involved in the compression method is unrelated
to the data of the image itself, and/or where the method is highly
dependent on statistical repetition of the data and the ability to
represent a statistical group of pixels with a token
representative.
[0028] As indicated above, with hyperspectral images it is very
often possible to infer information from a single pixel so as to
determine the presence of materials within the area of a pixel,
e.g., a camouflaged tank or missile below a canopy of foliage, even
when the detected element is smaller than the resolution of the
pixel. When a compression system removes data from the pixel, this
desirability for material detection may be inadvertently
hampered.
[0029] Hence there is a need for an image dimension reduction
system and method suitable for use with hyperspectral images that
overcome one or more of technical problems noted above.
SUMMARY
[0030] This invention provides a system and method for
hyperspectral image dimension reduction.
[0031] In particular, and by way of example only, according to one
embodiment of the present invention, provided is a method of
hyperspectral image dimension reduction, including: receiving a
hyperspectral image having a plurality of pixels and establishing a
basis vector (BV) set For each of the plurality of pixels, the
spectral vector is read from the pixel. The spectral vector is then
decomposed with the BV set to provide a reduced dimension vector
for the pixel. The provided output is a dimension reduced image
consisting of the reduced data vectors for each pixel.
[0032] In at least one alternative embodiment, provided is a method
of hyperspectral image dimension reduction, including: receiving a
hyperspectral image having a plurality of pixels and establishing a
basis vector (BV) set. The BV set is established by selecting an
initial pixel and reading the spectral vector for the initial pixel
as an initial basis vector (BV) of the set For each of the
remaining plurality of pixels taken in turn, each spectral vector
is read and unmixed with the BV set to determine a residual vector.
In response to the magnitude of the residual vector being greater
than a threshold, adding the spectral vector to the BV set. With
the BV set so established, then returning to read the spectral
vector from the pixel once again. In this pass the spectral vector
is unmixed with the BV set to provide an alpha vector and a
residual vector for the pixel. The provided output is a dimension
reduced image consisting of at least the alpha vector for each
pixel.
[0033] In yet another embodiment, provided is a system for
hyperspectral image dimension reduction. The system includes an
interface operable to receive at least a hyperspectral image having
a plurality of pixels; a spectrum reader operable to read a
spectral vector for each pixel; a pool operable to provide a basis
vector (BV) set having at least one basis vector; a decomposer
operable to decompose the spectral vector of each pixel with at
least one basis vector and provide a result; and an evaluator in
connection with the pool and decomposer and operable in at least a
first instance to compare the result to a threshold value, and
operable in at least a second instance to provide an alpha vector
and a residual vector for each pixel.
BRIEF DESCRIPTION OF THE DRAWINGS
[0034] FIG. 1 illustrates a simplified diagram of a hyperspectral
imaging system including an embodiment of the hyperspectral image
dimension reduction system of the present invention;
[0035] FIG. 2 is an exemplary hyperspectral image as may be
dimensionally reduced in accordance with at least one embodiment of
the present invention;
[0036] FIG. 3 is block diagram of a hyperspectral image dimension
reduction system in accordance with at least one embodiment of the
present invention;
[0037] FIG. 4 is a high level block diagram presenting at least one
method of hyperspectral image dimension reduction in accordance
with at least one embodiment of the present invention;
[0038] FIG. 5 is a more specific block diagram of a method of
hyperspectral image dimension reduction in accordance with at least
one embodiment of the present invention; and
[0039] FIG. 6 is a block diagram of a computer system in accordance
with one or more embodiments
DETAILED DESCRIPTION
[0040] Before proceeding with the detailed description, it is to be
appreciated that the present teaching is by way of example only,
not by limitation. The concepts herein are not limited to use or
application with a specific type of problem solving system. Thus,
although the instrumentalities described herein are for the
convenience of explanation, shown and described with respect to
exemplary embodiments, it will be appreciated that the principles
herein may be applied equally in other types of image dimension
reduction systems and method.
[0041] FIG. 1 illustrates an exemplary image dimension reduction
system ("IDRS") 100 incorporated as part of a hyperspectral image
system 102 provided by a satellite 104. The satellite 104 is tasked
to image a geographical area 106, and in so doing receives light
108 from a single ground-resolution cell. To generalize the basic
elements provided by a hyperspectral image system 102 and the
general method of operation, the light 108 is received by scan
mirrors 110 and or other optics that direct the light through at
least one dispersing element 112.
[0042] The dispersing element separates wavelengths and provides
them to imaging optics 114 which in turn focus the wavelengths upon
an array of detectors 116 arranged to capture data information
across a spectrum of wavelengths. The hyperspectral data is then
processed by IDRS 100 for transmission 118 to a ground station
120.
[0043] As is further described below, the degree of dimension
reduction and classification as lossy or lossless is an adjustable
feature of IDRS 100. Further, the transmitted dimension reduced
data 118 is suitable for immediate processing without requiring
reconstruction of the original data, or an approximation of the
original data.
[0044] FIG. 2 is a further representation of the image 200, and
specifically images? 200i.about.n, acquired simultaneously in many
different adjacent wavelength bands As indicated, the image is
comprised of pixels arranged in an X-Y coordinate system. Under
appropriate circumstances, alternative coordinate systems may be
employed, however X-Y is represented herein for ease of
illustration and discussion, and not by way of limitation. Each
pixel 202 has a spectral vector 204. The spectral vector contains
at least some number of spectral measurements of the energy
upwelling from that pixel. As indicated in the enlarged section
206, of row X' the spectral vector 204A of the initial pixel 202A
may well be different from the spectral vectors 204B.about.204n for
pixels 202B.about.202n.
[0045] As is also shown in FIG. 2, a basis vector (BV) set 208 is
provided, and is shown to have an initial basis vector 210. In
contrast to the limitations of the principal component analysis
(PCA) methods discussed above, IDRS 100 permits control of the
error levels and/or fidelity of individual spectra, and as such is
highly advantageous when the detection problem is for spectrally
unique objects that sparsely populate the scene. In addition, the
present invention as embodied in IDRS 100, derives basis vectors
using no a priori knowledge. The basis vectors span the background
materials and (statistically sparse) targets in the scene. The
basis vectors may be endmembers (e.g. pixel spectra) or may have no
physical interpretation. The reduced dimensionality data can
control maximum error on each pixel. The number of basis vectors is
minimized for a given maximum error level on a per pixel basis.
User/Analyst selected spectra can be directly added to the basis
vector set. Spectra collected or measured a priori can be added to
the basis vector set. The reduced dimensionality scene can be
processed using all hyperspectral (vector-type) processing. The
computation of the reduced dimensionality scene is faster than
prior art techniques with comparable levels of reduction. The
determination and/of establishment of the BV set 208 and at least
the initial basis vector 210 is further discussed below.
[0046] The method of hyperspectral image dimension reduction 300 is
presented at a high level in FIG. 3. It is understood and
appreciated that the method need not be performed in the order
herein described, but rather is presented for ease of discussion
and illustration in accordance with at least one embodiment.
[0047] Moreover, in at least one embodiment, a hyperspectral image
is provided, block 302. A BV set 208 is developed with respect to
the hyperspectral image, block 304. Then, with the BV set
established, each pixel is decomposed with the BV set to provide a
plurality of alpha vectors, block 306. These alpha vectors are
provided as the reduced data set of the output scene, block
308.
[0048] FIG. 4 is a high level block diagram of the system
architecture of the IDRS 100, in accordance with at least one
embodiment. IDRS 100 may be implemented on a computer having
typical components, such as a processor, memory, storage devices,
and input and output devices. During operation, IDRS 100 may be
maintained in active memory for enhanced speed and efficiency. In
addition, in at least one embodiment, IDRS 100 may be operated on a
computer network and may utilize distributed resources.
[0049] As shown in FIG. 4, for at least one embodiment, IDRS 100
includes an interface 400, a pool 402, a spectrum reader 404, a
decomposer 406 and an evaluator 408. The interface is operable to
receive information such as a hyperspectral image, a user provided
basis vector, and/or other information such as for example, but not
limited to, whether to perform the dimension reduction as a lossy
or lossless operation. The interface may receive information
directly from a user through an associated input device, or
directly from a system component, such as the detectors 116 shown
in FIG. 1.
[0050] With respect to the issue of image reduction as a lossless
operation, it is understood that the resulting data may not qualify
in general terms as compressed data, for the additional components
added to the compressed element may in fact result in a data set
that is substantially the same size as the original data set.
However, it is possible for the compressed portion of the data to
be sent, processed or otherwise utilized and to deliver other data
elements at a later time. Moreover, it is possible to utilize a
compressed lossy version of the data for immediate transfer to the
ground, and still acquire the additional data elements so as to
achieve lossless version at a later date. For these possible
benefits, this option is intended to be provided to users of the
system.
[0051] The pool 402 is in communication with the interface 400 and
is operable to receive basis vectors from the interface, if
provided by a user or built in as a default. More generally, the
pool 402 is operable to provide at least one basis vector, and more
commonly a set of basis vectors for use in dimensionally reducing
the hyperspectral image. The spectrum reader 404 is operable to
read and provide the spectral vector for each pixel of the
hyperspectral image.
[0052] The decomposer 406 is coupled to both the pool 402, so as to
receive the BV set, and the spectrum reader 404, so as to receive
the spectral vector of a given pixel. The decomposer 406 is
operable to decompose the spectral vector with the BV set and
provide a result. In at least one embodiment the decomposer 406 is
an unmixer. The evaluator 408 is in communication with at least the
pool 402 and the decomposer 406. In a first instance, the evaluator
is operable to compare the result, e g., the magnitude of a
residual vector, with a threshold. If the magnitude of the residual
vector is over a threshold value then the evaluator 408 will direct
the spectral vector be added to the pool 402, and more specifically
the BV set. In a second instance, the evaluator 408 is operable to
provide an alpha vector, a residual vector, and/or the magnitude of
the residual vector for each pixel.
[0053] It is of course understood and appreciated that the
components/elements of the IDRS 100 as identified above may be
implemented with subcomponents or otherwise further subdivided.
Each element has been illustrated as an identifiable element for
ease of discussion and illustration. In addition, as indicated
above, in at least one embodiment, IDRS 100 is implemented in a
computer environment, wherein the interface 400, the pool 402, the
spectrum reader 404, the decomposer 406 and the evaluator 408 are
implemented as routines, and/or objects rendered in a variety of
different forms of code and instructions as may be preferred for
different computer systems and environments.
[0054] FIG. 5 provides a more specific block diagram of the method
presented in FIG. 3. More specifically, FIG. 5 provides greater
detail for the operations of developing the BV set, block 304, and
generating the alpha vectors, block 306. Again it is understood and
appreciated that the order of the steps herein described are
provided for ease of illustration and discussion and not by way of
limitation.
[0055] More specifically, as shown in FIG. 5, in at least one
embodiment the process of developing the BV set commences with
analysis of whether a seed basis vector has been provided, decision
500. Stated simply, the basis vector set provides a collection of
spectral vectors by which the spectral vectors of all pixels within
the hyperspectral image are evaluated. Whereas in at least one
embodiment, the evaluation and comparison is based entirely upon
data already present in the hyperspectiral image itself, in at
least one embodiment at least one basis vector is provided by an
operator or is read from a default file. Such a basis vector may,
for example, correspond to the spectral vector of a particular
material known to exist or expected not to exist, within the
geographic area from which the hyperspectral image was formed.
[0056] Where one or more seed vectors are provided as initial basis
vectors, the seed basis vector(s) is received and added to the
basis vector set 210, block 502. To develop the reset of the BV
set, an initial pixel (P) is selected, block 504. Generally, and in
at least one embodiment, the initial pixel is selected to be the
first pixel in the top left or bottom left corner, the pixels to be
compared in an incremental step pattern across each row X.
[0057] Where a seed vector is not provided, decision 500, an
initial pixel is immediately selected, block 506. The spectral
vector for this initial pixel is then read by the spectrum reader
and this initial spectral vector is used as the initial basis
vector for the further development of the BV set, block 508. In at
least one embodiment, the selection of the initial pixel to
determine the initial basis vector is a random selection. In at
least one alternative embodiment, the initial pixel is selected to
be either the top left or bottom left corner pixel.
[0058] Moreover, if a seed basis vector(s) is provided then it is
treated as if it was the determined basis vector from an initial
pixel and the process immediately commences with unmixing and
testing as discussed below. If a seed basis vector is not provided,
then a pixel is selected and its spectral vector is adopted as the
initial basis vector, and then the process commences with unmixing
and testing as set forth below.
[0059] With an initial basis vector so determined, the process
increments to select the next pixel (P), block 510. Generally, and
in at least one embodiment, it is understood and appreciated that
the pixels will be incremented through in a left to right iteration
for each row X.
[0060] The process now continues with the reading of the spectral
vector of pixel P, block 512. The spectral vector of the selected
pixel is then decomposed with the members of the developing BV set.
In at least one embodiment the decomposition is performed by
unmixing the spectral vector with the members of the BV set.
[0061] Generally stated, unmixing is understood and appreciated to
be, in essence, an algebraic approach to decomposing the spectral
vector into its component parts. For example, a spectral vector
from a given pixel may comprise the spectrum elements indicative of
grass, aluminum and sand, and thus be a composite spectral vector.
When this vector is unmixed with vectors representative of aluminum
and sand, the spectral vector of grass will be revealed. The
unmixing process provides alpha coefficients representative of the
abundance of the basis vectors and will provide a residual vector.
The vector that is unmixed is an algebraic combination of the basis
vectors, weighted in some fashion by the alpha coefficients plus
the residual vector.
[0062] In at least one embodiment, the magnitude of the residual
vector is then determined, and is understood and appreciated to be
derived from the square root of the dot product of the residual
vector with itself, otherwise known as Euclidean distance provided
by the equation:
.parallel.x.parallel.:= {square root over (.chi..sub.1.sup.2+ . . .
+.chi..sub.n.sup.2)}.
where x=[x.sub.1, x.sub.2, . . . , x.sub.n].
[0063] The magnitude of the residual vector is then evaluated with
a threshold value, decision 516. In at least one embodiment the
threshold value is provided by an operator. The threshold value may
also be a pre-determined value that is hard coded into the method.
If the magnitude is over the threshold, decision 516, the spectral
vector is added to the BV set, block 518. In some embodiments,
there may be a pre-determined maximum number of members of the BV
set. When the total number of members is below this pre-determined
maximum there is of course no issue and the new spectral vector is
simply added to the BV set, decision 520.
[0064] If, however, the maximum number of members has been reached,
the members of the BV set are optimized, block 522. Various
optimization methods may be employed and in general the type of
optimization strategy adopted is determined by an operator. These
methods may include, but are not limited to, optimization to
selection of the maximum spectral vectors (e.g., the spectral
vectors having the greatest magnitudes), optimization to select the
greatest range between the spectral vectors, optimization to select
the spectral vectors closest to a spectral vector mean,
optimization to select the spectral vectors closest to a spectral
median.
[0065] In at least one embodiment, the residual magnitude is stored
when a basis vector is added to the BV set. When the maximum number
for the BV set is reached, the magnitude of a new residual must be
larger than the smallest residual computed from the existing
members of the BV set. If it is not, the new basis vector is not
added. If it is, then the new basis vector is added and the basis
vector that previously had the smallest residual is removed.
[0066] As indicated above, in at least one embodiment the user is
permitted to provide seed basic vectors. In at least one such
embodiment where the user has also set a maximum number for the BV
set, the seed basis vector(s) are treated as special and maintained
as core elements of the BV set. In one embodiment, this is achieved
by tagging the seed vectors such that they are not considered in
the elimination evaluation process. Moreover, in at least one
embodiment the maximum number of BV set members does not include
the seed basis vector(s). In an alternative embodiment, the maximum
number of BV set members may indeed include the seed basis
vector(s), but they are simply not included in the evaluation for
removal process. As such, key factors that are indicated by the
seed vector(s) are maintained within the BV set.
[0067] Having resolved the question of whether to add the residual
vector to the BV set, and whether to optimize the BV set, the
method continues to select the next pixel, decision 524. If indeed
more pixels remain then the method will increment to the next
pixel, block 526. It should be understood and appreciated that,
although the incremental process is intended to be orderly, the
type of incrementing--left to right, right to left, top to bottom,
bottom to top, or random--is largely immaterial. The key element is
that each pixel of the hyperspectral image is being evaluated so as
to determine an appropriate set of basis vectors that are, in at
least one embodiment, inherently tied to hyperspectral image
itself.
[0068] With the BV set so established, the method now continues
with the generation of the alpha vectors, block 306. As in the
development of the BV set, the generation of the alpha vectors
begins with the selection of an initial pixel, block 528. The
spectral vector is then read from the selected pixel, block 530.
The spectral vector is then decomposed with the BV set, block 532.
In at least one embodiment the decomposition is performed as
unmixing, as discussed above. It is to be realized that for each
member of the BV set there will be a corresponding value, an
unmixing coefficient. There will also be a residual vector.
[0069] For the given pixel, the unmixing coefficients represent an
alpha vector for the given pixel. It is further understood and
appreciated that the relative magnitude of dimension reduction is
approximately the difference between the number of elements in a
spectral vector originally present in the hyperspectral image and
the number of members in the applied BV set For example, if there
were originally one-hundred spectral bands present in the
hyperspectral image, and twenty basis vectors in the BV set, the
dimension reduction represented by the alpha vectors would be about
eighty or a factor of five.
[0070] The process is continued for each pixel, as indicated by the
query to determine the presence of more pixels, decision 536, and
the increment to the next pixel, block 538 and the return to read
the spectral vector for the selected pixel, block 530. Moreover,
all of the pixels in the hyperspectral image scene are unmixed with
the BV set so as to provide vectors having g coefficients
(corresponding to the number of members in the BV set) and a
residual vector for each X-Y location.
[0071] When the generation of alpha vectors has been accomplished
for all pixels of the hyperspectral image, the resulting set of
alpha vectors is provided as the output scene, namely a
dimensionally reduced hypercube. Because of the application of the
BV set, which is derived from the spectral vectors of each pixel,
to each pixel, the resulting information of each alpha vector has
approximately the same information content as the original pixel.
More specifically, the reduced alpha vector is still representative
of the composite spectral frequencies present in the pixel, such a
combination of grass, sand and metal as present in the original
spectral vectors is still represented in the alpha vector.
Furthermore, and quite advantageously, the dimensionally reduced
image provided by the alpha pixels is suitable for processing
without regeneration of the original image. Further, the maximum
loss of information on any pixel is minimized compared to any other
dimensionality reduction process. This is crucial when the image is
processed for subtle changes, the presence of small targets or
anomalies.
[0072] Decompression of the output image can be accomplished with
the use of the BV set. As such, in at least one embodiment the BV
set is provided as output with the dimension reduced image. The
residual vectors, if left untreated are understood to be of the
same size as the original data set. Calculating the magnitude for
each residual vector greatly reduces the volume of the data once
again.
[0073] Indeed, in at least one embodiment the magnitude of each
residual vector is also determined and provided as output in
addition to dimensionally reduced image represented by the alpha
vectors. Moreover, it is understood and appreciated that in varying
embodiments, the method 300 may provide: the alpha vectors, the
magnitude of the residual vectors, the BV set, and combinations
thereof. As such, it is understood that the method is adaptable to
provide compression and dimension reduction options that range from
lossy to lossless.
[0074] FIG. 6 is a high level block diagram of an exemplary
computer system 600 as may be used to provide IDRS 100, and/or
embodiments of method 300. Computer system 600 has a case 602,
enclosing a main board 604. The main board has a system bus 606,
connection ports 608, a processing unit, such as Central Processing
Unit (CPU) 610, and a memory storage device, such as main memory
612, hard drive 614, and CD/DVD Rom drive 616.
[0075] Memory bus 618 couples main memory 612 to CPU 610. A system
bus 606 couples hard drive 614, CD/DVD Rom drive 616, and
connection ports 608 to CPU 610. Multiple input devices may be
provided, such as for example a mouse 620 and keyboard 622.
Multiple output devices may also be provided, such as for example a
video monitor 624 and a printer (not shown).
[0076] Computer system 600 may be a commercially available system,
such as a desktop workstation unit provided by IBM, Dell Computers,
Gateway, Apple, Sun Micro Systems, or other computer system
provider. Computer system 600 may also be a networked computer
system, wherein memory storage components such as hard drive 614,
additional CPUs 610 and output devices such as printers are
provided by physically separate computer systems commonly tied
together in the network. Those skilled in the art will understand
and appreciate the physical composition of components and component
interconnections comprising computer system 600, and select a
computer system 600 suitable for the schedules to be established
and maintained.
[0077] When computer system 600 is activated, preferably an
operating system 626 will load into main memory 612 as part of the
boot strap startup sequence and ready the computer system 600 for
operation. At the simplest level, and in the most general sense,
the tasks of an operating system fall into specific
categories--process management, device management (including
application and user interface management) and memory
management.
[0078] In such a computer system 600, the CPU 610 is operable to
perform one or more of the scheduling embodiments described above.
Those skilled in the art will understand that a computer-readable
medium 628 on which is a computer program 630 for adding activities
to a schedule may be provided to the computer system 600. The form
of the medium 628 and language of the program 630 are understood to
be appropriate for computer system 600. Utilizing the memory
stores, such as for example one or more hard drives 614 and main
system memory 612, the operable CPU 602 will read the instructions
provided by the computer program 630 and operate to perform the
method of hyperspectral image dimension reduction 300 as described
above.
[0079] Changes may be made in the above methods, systems and
structures without departing from the scope hereof It should thus
be noted that the matter contained in the above description and/or
shown in the accompanying drawings should be interpreted as
illustrative and not in a limiting sense. The following claims are
intended to cover all generic and specific features described
herein, as well as all statements of the scope of the present
method, system and structure, which, as a matter of language, might
be said to fall therebetween.
* * * * *