IEEE Geoscience and Remote Sensing Magazine - June 2019 - 109

These systems are associated with the high-temperature,
vibrational states of the combustion products and strong
emission that permits convenient discrimination against
the ambient background [41], [42]. Typical IR intensities of
intercontinental ballistic missile (ICBM)-class engines, e.g.,
are on the order of several hundred to several thousand
W/sr-μm [40], [43]. As a missile rises in the atmosphere, fuel
and combustion products continue to burn when mixed
with atmospheric oxygen, a process known as afterburning.
The IR intensity from large ICBM plumes is approximately
1-10 MW/sr-μm at peak intensity, i.e., at altitudes below
40 km [43], [44].
Combustion plume species emit in the MWIR (3-5-µm)
wavelengths and are principally composed of water, carbon
dioxide, cobalt, sulfur dioxide (SO2), nobelium, and nitrogen oxide [40], [43]. Theoretically, there is a relationship
between the depth of the spectral absorption features and
the concentration of the chemical species. Thus, although
thermal emission within the plume may help detect and
track the missile, the spectra of the combustion species and
their distribution may help determine the type and velocity
of the rocket [39]. The former (type) is based on the detection of the species and their concentrations, while the latter (velocity) is based on the spectral retrievals at different
atmospheric altitudes in a given time frame.
Advanced research in IR signature analysis has garnered
significant interest in the defense communities affiliated
with all of the applications discussed in this article, including
the characterization of combustion species and early warning systems of the BMDS. As a consequence, their outcomes
are controlled by security restrictions leading to rare and sporadic publications. Nevertheless, the HSI research conducted
at the U.S. Army Research Laboratory for detecting ballistic
missiles is summarized in [45]. The IR spectral characteristics of rocket exhaust plumes under varying motor operating
conditions were reported by [42], [47] and [48]. Wang et al.
proved that the radiation of the plume and its spectral distribution vary for different solid propellants with different
energy characteristics [50]. Rhoby et al. found that the high
spectral resolution of the Telops' FIRST imager is essential for
diagnosing combustion plumes and analyzing their tomography [48], [53]. Lang et al. [42] studied the spectral distribution characteristics of a rocket plume and the effect of the
missile's high-speed motion on the spectral recognition. Niu
et al. [49] developed a model that computes the IR irradiative signatures of rocket plumes and investigates IR spectral
radiation characteristics acquired by an MWIR hyperspectral imager. In [47] and [52], modeling of the IR signature
of rocket plumes was reported and validated using in-flight
measurements, demonstrating that the relation between the
IR irradiance signature and rocket types is not evident. Understanding the combustion processes during flight requires
accurate measurements, radiative transfer models, and computational fluid dynamics models. Understanding computational fluid dynamics models is a challenging task that
combines rocket and engine type, propellant composition,
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

chamber and nozzle parameters, burn time, flight mechanics, and atmospheric conditions [41], [43], [44].
DETECTING WEAPONS OF MASS DESTRUCTION
Today, (inter)national defense organizations face a range of
complex challenges and threats to their security, from both
state and nonstate actors who promote the proliferation of
WMDs, which comprise chemical, biological, radiological, and nuclear (CBRN) weapons. However, the acronym
CBRN is misleading because CBRN hazards are different,
depending on their production, effects, operational consequences, and legal issues [54].
In this article, we focus on the chemical elements of
CBRN weapons. Chemical warfare agents (CWAs) and hazardous materials (hazmats) are gases, liquids, or solids that
have a direct, toxic effect on people, other living organisms,
or the environment. NATO [55] lists the chemical substances that may be used to manufacture chemical weapons; an
analysis of this list shows that most of these precursors are
commonly used in ordinary civilian industrial applications. These dual-use activities lead to the understanding
that chemical agents, whether in the form of weapons employed by terrorists, rogue states, or toxic spills, pose significant risks that compel defense and civil security agencies
(CSAs) to detect and identify these threats [56]. Accurately
identifying and tracking airborne toxins is crucial for combatting the use of chemical gases as weapons, preventing
fatalities due to accidental leakage of hazmats, and avoiding atmosphere contamination. Therefore, under the activities framework of defense and CSAs, efforts are canalized
into three realms: the presence of CWAs, the proliferation
of chemical warfare, and the presence and concentration of
hazmat spills.
Stand-off chemical detection is a separate machinelearning problem in HSI for two main reasons. The first is
related to the characteristic of the soft target, a Gaussian
distribution from the source, and the low level of concentration in the clutter. The second is related to the spectral region of choice. Most chemical species, except homonuclear
diatomic species, can be detected and quantified because of
their unique IR spectral properties in the wavelength region
of 3-13 µm [57]. Most HSI demonstrators in this application are operated in this atmospheric transmission window.
Chemical plume detection requires the exploitation of both
the spectral signatures of the target compounds and radiometric and temperature differences between the plume and
the background scene. Nevertheless, this spectral region is
dominated by thermal self-emission, which allows the HSI
sensor to perform chemical and hazmat detection and classification regardless of illumination conditions [57].
The presence of a chemical spill in the thermal hyperspectral scene can be explained as a three-layer model: the
medium (e.g., atmosphere, soil, or water), chemical plume,
and background [18]. Because each element contributes to
the total radiance in a given pixel, it is necessary to first extract the spectrum of the spill to obtain its retrievals (i.e., the
109

IEEE Geoscience and Remote Sensing Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2019