Note: Descriptions are shown in the official language in which they were submitted.
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
Method And Apparatus For Gas Detection Based On Spectral Spatial
Misregistration
FIELD OF THE INVENTION
The current invention relates to the field of spectral imaging, and in
particular to the
area of spectral processing for the detection of gases.
BACKGROUND OF THE INVENTION
There are many areas that require detection and identification of gases by
remote
sensing. Some examples include monitoring of pollution from ground stations by
airborne
sensors and by satellites, measuring trace gas constituents of the atmosphere,
assessing the
state of the environment, monitoring industrial effluents, detecting harmful
gases, such as
carbon monoxide, and anesthetic gases in the respiratory air of a patient.
Patent number: U.S. patent 6545278 to Franois Mottier and Scott Bruce for Gas
Discriminating Gas Detector System and Method teaches an overview of gas
detection
systems based on measurement of absorption of electromagnetic radiation by a
gas of
interest. This patent teaches a gas detection system for detecting and
measuring
concentration of gases in a chamber containing an air/gas mix by providing an
optical light
source through the air/gas mixture to one or more light detectors and
measuring the
absorption by the air/gas mixture of light in the bands individually
associated with the
detection channels.
Electromagnetic radiation is subject to absorption and scattering by the
atmosphere
and surfaces on the path from the radiation source to a radiation sensor. An
example in the
atmosphere of the Earth is absorption by ozone, carbon dioxide and water. The
gases in the
atmosphere of the earth selectively impede the passing of electromagnetic
radiation. Energy
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
in certain wavelengths is allowed to pass through almost unchanged while other
wavelengths
are almost totally blocked.
A spectral band, or spectral region, is a well-defined, continuous wavelength
range in
the spectrum of reflected or radiated electromagnetic energy. A spectral image
is an image of
one spectral band of a scene taken with a sensor that is sensitive to
electromagnetic energy in
that spectral band. Each image is also referred to as a band. A multispectral
image is an
image which contains data from a plurality of spectral images. A well known
multispectral
(or multi-band image) is a RGB color image, consisting of a red, a green and a
blue image,
each of them taken with a sensor sensitive to a different band (wavelength).
While the three
band RGB example is well known, and is included in the definition of
multispectral, in the
field of the current invention, the term multispectral is conventionally used
for data
containing from tens to hundreds of bands. Another type of multispectral
imaging is known
as hyperspectral imaging (HSI). Hyperspectral data may contain hundreds to
thousands of
bands. Hyperspectral imaging can also be defined by the manner in which the
data is
collected. Hyperspectral data is a set of contiguous bands (usually by one
sensor) whereas
multispectral data is a set of optimally chosen spectral bands that are
typically not contiguous
and can be collected from multiple sensors.
U.S. patent 7301148 to Timothy J. Johnson for Method and System for Remote
Detection of Gases teaches systems that use active sources of infrared (IR)
radiation, such as
Fourier Transform Infrared (FTIR) and teaches passive IR techniques. A system
is provided
that includes at least one extended source of broadband infrared radiation and
a spectrally
sensitive receiver positioned remotely from the source. The source and the
receiver are
oriented such that a surface of the source is in the field of view of the
receiver. The source
includes a heating component thermally coupled to the surface, and the heating
component is
2
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
configured to heat the surface to a temperature above ambient temperature. The
receiver is
operable to collect spectral infrared absorption data representative of a gas
present between
the source and the receiver.
U.S. patent 6750453 to Loren D. Nelson and Martin J. O'Brien for Methods of
and
apparatus for detecting low concentrations of target gases in the free
atmosphere teaches an
apparatus that includes a source that directs broadband modulated light into a
region the free
atmosphere in which target gas may be present. A gas correlation radiometer
responds to
light transmitted through the region. Separate radiometer channels respond to
a single beam
of light after transmission through the region. A beam splitter separates the
beam into two
beams, one directed into each of the channels. The two channels separately and
simultaneously respond to a respective one of the light beams for separately
and
simultaneously generating signals that together indicate whether the target
gas is in the free
atmosphere.
U.S. patent 4676642 to Herbert A. French for Apparatus and method for remote
sensing of gases, vapors, or aerosols teaches a remote sensor comprising a
means to measure
the change in temporal coherence of light of a selected narrow waveband when
it interacts
with the gas. The light can be provided by a laser source or filtered sun
light.
Improvements to conventional methods of spectral imaging for gas detection are
focused on avoiding, reducing, or compensating for optical effects that affect
the accuracy of
the spectral image. An example of this approach is described by Francesco
Dell'Endice, ET.
AL., in the paper "Scene-based method for spatial misregistration detection in
hyperspectral
imagery," Applied Optics 46, 2803-2816 (2007). The paper describes how
hyperspectral
imaging sensors suffer from spatial misregistration, an artifact that prevents
the accurate
3
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
acquisition of the spectra. The paper proposes a method for estimating spatial
misregistration,
identifying misalignments, and suggestions are given to correct for spatial
misregistration.
Whereas conventional approaches in the art are focused on eliminating optical
effects,
which are considered a hindrance to attempts to provide gas identification,
the current
invention uses an optical effect of the spectral imaging process, namely,
spectral
misregistration, in an innovative method to provide detection and
identification of gases by
remote sensing.
SUMMARY
According to the teachings of the present embodiment there is provided a
method for
remote identification of at least one gas including: sampling a plurality of
spectral images of
a scene wherein each spectral image is sampled at a different wavelength;
providing a
reference spectral image; generating a spatial displacement expression by
detecting the
spatial misregistration in at least one region of the spectral images between
the reference
spectral image and at least one of the plurality of spectral images; providing
at least one
reference spatial displacement expression corresponding to at least one gas;
and
implementing at least one identification process to identify at least one gas,
the identification
process employing the generated spatial displacement expression and the at
least one
reference spatial displacement expression.
In an optional embodiment, the reference spectral image is one of the sampled
spectral images. In another optional embodiment, the method further includes
determining
the concentration of the at least one gas. In another optional embodiment, the
at least one
reference spatial displacement expression is provided from a library of
reference spatial
4
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
displacement expressions. In another optional embodiment, the spectral images
are of the
entire the scene. In another optional embodiment, the spectral images are of
an area of the
scene. In another optional embodiment, the spectral images are of a pixel of
the scene. In
another optional embodiment, the library is a general-purpose library. In
another optional
embodiment, the library is adapted for the specific atmospheric conditions of
the scene. In
another optional embodiment, the library is adapted for the specific gas
concentrations of the
scene.
According to the teachings of the present embodiment there is provided a
system for
remote identification of at least one gas including: a spectral image sampling
device
configured to sample a plurality of spectral images of a scene wherein each
spectral image is
sampled at a different wavelength; a processing system including at least one
processor,
operationally connected to the spectral image sampling device, configured to:
provide a
reference spectral image; generate a spatial displacement expression by
detecting the spatial
misregistration in at least one region of the spectral images between the
reference spectral
image and at least one of the plurality of spectral images; provide at least
one reference
spatial displacement expression corresponding to at least one gas; and
implement at least one
identification process to identify at least one gas, the identification
process employing the
generated spatial displacement expression and the at least one reference
spatial displacement
expression.
In an optional embodiment, the reference spectral image is one of the sampled
spectral images. In another optional embodiment, the system is further
configured to
determine the concentration of the at least one gas. In another optional
embodiment, the
system is operationally connected to a library of reference spatial
displacement expressions,
the library of reference spatial displacement expressions providing the at
least one reference
5
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
spatial displacement expression. In another optional embodiment, the spectral
images are of
the entire the scene. In another optional embodiment, the spectral images are
of an area of
the scene. In another optional embodiment, the spectral images are of a pixel
of the scene. In
another optional embodiment, the system is operationally connected to a
multispectral
imaging device, the multispectral imaging device providing the spectral
images. In another
optional embodiment, the system is operationally connected to .a hyperspectral
imaging
device, the hyperspectral imaging device providing the spectral images.. In
another optional
embodiment, the library is a general-purpose library. In another optional
embodiment, the
library is adapted for the specific atmospheric conditions of the scene. In
another optional
embodiment, the library is adapted for the specific gas concentrations of the
scene.
BRIEF DESCRIPTION OF FIGURES
FIGURE 1 is a flowchart of a method for gas detection based on spectral
spatial
misregistration.
FIGURE 2 is a plot of the x and y displacement for each spectral band.
FIGURE 3 is a diagram of a system for gas detection based on spectral spatial
misregistration.
DETAILED DESCRIPTION
The principles and operation for gas detection based on spectral spatial
misregistration according to the present embodiment may be better understood
with reference
to the drawings and the accompanying description. A hyperspectral image
consists of a great
6
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
number of spectral images. Although these spectral images are obtained
simultaneously, due
to a number of optical effects the hyperspectral bands are not co-registered.
Identification of
the spatial misregistration can be done using an innovative method of subpixel
image
registration between spectral bands. This method generates a mapping of the
gas absorption
features that can be used for gas detection and identification.
In the context of this document, the term wavelength is used to refer to a
particular
wavelength or a band of wavelengths around the particular wavelength value.
The term gas
includes any substance with relatively low density and viscosity with the
ability to diffuse
readily, as well as all gaseous elements, compounds or gas-borne droplets or
particles.
Hyperspectral sensors collect information as a set of images. Each image
represents a
wavelength of the electromagnetic spectrum and is also known as a spectral
band. This
sequence of images can be combined to form a three dimensional hyperspectral
cube as a
way to handle the data for processing and analysis. The unprocessed (raw)
spectral images
can be combined to form a hyperspatial cube with or without various levels of
processing. In
one implementation, the unprocessed spectral images are processed to register
them and then
combined into a hyperspatial cube. In an alternative implementation having the
same result,
the unprocessed images that have been combined into a hyperspatial cube are
processed to
register the images and then re-combined into a new hyperspatial cube.
Image registration is the process of overlaying two or more images in correct
alignment. Registration techniques are classified according to their nature
and various
techniques are known in the art, for example: area-based registration and
feature-based
registration. Each pixel in the hyperspectral cube is characterized by three
indices (x,y,),).
Thus for each pixel (x,y) in the spatial domain a complete spectrum allows us
to obtain
signatures characterizing the scene imaged. The accuracy of these signatures
depends on
7
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
many factors. One optical effect of hyperspectral collection is one spectral
band (?) not
corresponding spatially to another spectral band. This effect is known as
spatial
misregistration, and corresponds to the displacement which would be required
to achieve
registration between the corresponding regions of the images. In this
description, the
calculated spatial misregistration between a reference band and a spectral
band is referred to
as spatial displacement expression. There is a correlation between the spatial
misregistration
and atmospheric transmission. Spatial misregistration is particularly
prominent in images of
bands near the absorption bands of atmospheric gases.
Referring now to the drawings, FIGURE 1 is a flowchart of a method for gas
detection based on spectral spatial misregistration. Sampling spectral images
of a scene is
shown in block 100. Choosing one of the spectral images as a reference
spectral image is
shown in block 104. The spectral images are processed to detect spatial
misregistration,
shown in block 108. The spatial misregistration is used to generate a spatial
displacement
expression, shown in block 110. In block 112, a reference spatial displacement
expression is
provided. In block 114 at least one identification process uses the spatial
displacement
expressions to identify the gas.
Sampling spectral images of a scene is shown in block 100. The actual number
of
images can vary depending on the specific implementation of the method. At
least two
images are normally used, with hyperspectral imaging providing hundreds to
thousands of
images. A reference spectral image is provided, shown in block 104. In one
implementation,
the reference image is one of the sampled spectral images. Choosing the
reference image can
be done manually or directed by an automatic process. If one of the spectral
images is of a
band with no gas features present, it is preferable to use this spectral image
as the reference
8
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
image to provide improved processing. The other spectral images may contain at
least one
spectral band of at least one gas. In an optional implementation, the
reference image can be a
calculated mean position.
The spectral images are processed in block 108 to detect spatial
misregistration
between the reference spectral image and the spectral images of the other
bands. Note that
processing of the spectral image is not limited to an entire scene. The
spectral image can be
the entire frame, a block (an area), or even a pixel from a scene. The process
of detecting
spatial misregistration finds the x,y spatial displacement, at a subpixel
level, in a given
spectral band (source) relative to a spectral reference band. This x,y spatial
displacement
(dx,dy) for each spectral band is the spatial misregistration. FIGURE 2 is a
plot of the x and
y displacement for each spectral band using this method. The procedure is
repeated twice,
once to estimate dx and once to estimate dy. The first calculation is of all
the subpixel
displacements for all the rows of the source relative to the reference using
the normalized
correlation interpolated for subpixel accuracy. Next is obtaining an average
row
displacement using the average of all rows whose correlation coefficient is
above a threshold.
The same is done for the column disparity calculation after transforming and
resampling the
image using the average row displacement. Using an area-based registration
technique
facilitates a robust process using a "greedy" algorithm that tries to use as
much of the image
as possible and is less dependent on elements that may disappear in different
areas of the
spectrum.
The detected spatial misregistration from block 108 is used by block 110 to
generate a
spatial displacement expression. The spatial displacement expression is a set
of numbers, or
a corresponding function, that correspondingly correlates to the calculated
spatial
misregistration between a reference band and a spectral band. Note that the
spatial
9
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
displacement expression is not limited to being generated at the same location
or immediately
prior to performance of the subsequent analysis of the displacement
expression, and can also
be provided from another source, previously calculated and provided to the
process, or
generated by another means.
Providing a reference spatial displacement expression is shown in block 112.
In the
case of known gases, the spatial displacement expression can be provided from
a library or
other reference source. The library may be a general-purpose library or may be
adapted for
the specific atmospheric conditions to which the method is being implemented.
Adaption of
the gas properties for the specific atmospheric conditions and/or gas
concentrations can be
done using known techniques such as MODTRAN simulation software.
In block 114 matching techniques are preferably applied to compare each
generated
spatial displacement expression with reference spatial displacement
expressions. The
identification process tries to identify a best match gas or combination of
gases that
correspond to the generated spatial displacement expression for each area
(frame, block, or
pixel). This processing can be performed by well-known techniques such as the
Spectral
Angle Mapper (SAM) classification or other techniques that are known in the
art. A
threshold is then typically applied for each gas to give an indication of what
gases are present
in that area (frame, block, or pixel).
The identification block 114 may optionally include other algorithms. In one
optional implementation, block 114 detects the presence of gas. For detection
of the presence
of a gas, it may be sufficient to processes the generated spatial displacement
expression,
without the need to provide a reference spatial displacement expression for
comparison. In
this case, a technique such as threshold comparison can be used.
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
In another optional implementation, the identification block 114 may include
an
algorithm for determining the concentration of a gas. In certain applications,
particularly in
controlled environments where a specific target gas is to be detected,
concentration derivation
may be performed for a predetermined gas without a prior detection algorithm.
Based on the
classification output and/or the magnitude of the spatial movement, the path
concentration
can be estimated. In this case, the technique may or may not use a reference
spatial
displacement expression, or the technique may use other information to provide
a
concentration of the gas.
In one implementation, this method enables identification of a gas from a
single frame
without need for time analysis. This is particularly important when there is a
slow dispersion,
and can reduce false alarms. In another implementation, this method has been
shown to work
particularly well when the background is inhomogeneous. Conventional detection
schemes
either use the time analysis of the spectrum or assess the background when
homogenous. The
background is usually inhomogeneous in urban areas and airborne scenarios.
Referring now to the drawings, FIGURE 3 is a diagram of a system for gas
detection
based on spectral spatial misregistration. A spectral image sampling device
300A or
optionally a library of spectral images 300B are operationally connected to a
processing
system 302 containing one or more processors 304 configured for spectral image
processing
306 and identification 308.
The spectral image-sampling device 300A is configured to sample a plurality of
spectral images of a scene wherein each spectral image is sampled at a
different wavelength.
The actual number of spectral images sampled can vary depending on the
specific
11
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
implementation of the system. In one implementation, a multispectral imaging
device
samples a few to tens to a hundred spectral images of a scene. In another
implementation, a
hyperspectral image capture device can supply between hundreds and thousands
of images
for processing.
In an optional implementation, the spectral images can be provided from a
library of
spectral images 300B. In another optional implementation, the spectral images
can be
provided from one source or more than one source. If the spectral images are
provided from
different sources, the spectral images may require pre-processing or
additional processing to
compensate for differences between the sources and/or to achieve overall
registration of the
images to allow subsequent detection of localized misregistration. One example
is that a
different reference may be needed for different spectral ranges if the sensor
contains separate
imaging devices. In such cases, a translation transformation is typically not
a sufficient
mapping between devices. Other possible sources for providing spectral image
will be
obvious to one skilled in the art.
The spectral images are provided to a processing system 302 containing one or
more
processors 304. The processors are configured to process the spectral images
306.
Processing includes choosing one of the spectral images as a reference image,
then generating
a spatial displacement expression by detecting the spatial misregistration in
at least one
region of the spectral images between said reference spectral image and at
least one of said
plurality of spectral images.
The generated spatial displacement expression is provided to an identification
module
308. At least one reference spatial displacement expression is also provided
to the
identification module. This reference spatial displacement expression can be
provided from a
library or other sources as will be obvious to one skilled in the art. The
library may be a
12
CA 02717937 2010-09-08
WO 2009/125413 PCT/IL2009/000416
general-purpose library, or adapted for the specific atmospheric conditions
for which the
system is used. The library can also be adapted for the specific gas
concentrations for which
the system is being used. The identification module implements at least one
identification
process to identify at least one gas, the identification process using the
generated spatial
displacement expression with at least one reference spatial displacement
expression. In an
optional implementation, the processors may additionally or alternatively be
configured with
an algorithm for determining the concentration of a gas. Note that the
provided spectral
images and the provided reference spatial displacement expressions can be of
an entire scene,
such as a frame, part of an image, known as a block, or a pixel.
It will be appreciated that the above descriptions are intended only to serve
as
examples, and that many other embodiments are possible within the scope of the
present
invention as defined in the appended claims.
13
