Stand-off detection of solid targets with diffuse

reflection spectroscopy using a high-power

mid-infrared supercontinuum source

Malay Kumar,1,* Mohammed N. Islam,1,2 Fred L. Terry, Jr.,1 Michael J. Freeman,2

Allan Chan,3 Manickam Neelakandan,3 and Tariq Manzur4
1
Department of Electrical Engineering and Computer Science, University of Michigan, 1301 Beal Avenue,
Ann Arbor, Michigan 48109, USA
2
Omni Sciences, Inc., 9477 N. Territorial Road, Suite 120, Dexter, Michigan 48130, USA
3
United States Army CERDEC I2WD, Aberdeen Proving Ground, Maryland 21005, USA
4
Naval Undersea Warfare Center (NUWC, DIVNPT), 1176 Howell Street,
Building #1319, Room #318, Newport, Rhode Island 02841, USA
*Corresponding author: [email protected]

Received 7 November 2011; accepted 19 December 2011;

posted 2 February 2012 (Doc. ID 157700); published 14 May 2012

We measure the diffuse reflection spectrum of solid samples such as explosives (TNT, RDX, PETN), fer-
tilizers (ammonium nitrate, urea), and paints (automotive and military grade) at a stand-off distance of
5 m using a mid-infrared supercontinuum light source with 3.9 W average output power. The output
spectrum extends from 750–4300 nm, and it is generated by nonlinear spectral broadening in a 9 m long
fluoride fiber pumped by high peak power pulses from a dual-stage erbium-ytterbium fiber amplifier
operating at 1543 nm. The samples are distinguished using unique spectral signatures that are attrib-
uted to the molecular vibrations of the constituents. Signal-to-noise ratio (SNR) calculations demonstrate
the feasibility of increasing the stand-off distance from 5 to ∼150 m, with a corresponding drop in SNR
from 28 to 10 dB. © 2012 Optical Society of America
OCIS codes: 300.6340, 320.6629.

1. Introduction ing explosives, fertilizers, and paint coatings. We

We demonstrate the detection of solid targets at a observe unique spectral fingerprints in the near- and
stand-off distance of 5 m with diffuse reflection mid-IR wavelength regions of the reflection spectrum
spectroscopy using a mid-IR supercontinuum (SC) and develop a correlation-function-based algorithm
light source. The SC output spectrum extends from to distinguish between the samples. The demon-
0.75 to 4.3 μm, has a maximum time-average output strated SC light source has excellent beam quality
power of 3.9 W, and is obtained by pumping a 9 m due to the single-mode fiber output, covers multiple
long ZBLAN (ZrF4 -BaF2 -LaF3 -AlF3 -NaF) fluoride fi- atmospheric transmission windows, and is all-fiber
ber with the output of a dual-stage erbium-ytterbium integrated with no moving parts or free space optics.
fiber amplifier. The SC source is used to obtain the Thus, it is well suited for spectroscopy applications
reflection spectra of a wide range of samples includ- in a wide range of fields such as defense, homeland
security, remote sensing, and geology.
Diffuse reflection spectroscopy is a widely used
1559-128X/12/152794-14$15.00/0 technique for both qualitative and quantitative ana-
© 2012 Optical Society of America lysis of IR active samples. The most common method

2794 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

involves the use of a conventional Fourier transform sensing, and explosives detection at distances of
infrared (FTIR) instrument fitted with an attenuated up to 470 m [7].
total reflection or diffuse reflectance measurement Laser-induced breakdown spectroscopy (LIBS) is
accessory [1]. While the above technique covers the another widely used technique for stand-off detection
entire near-IR to far-IR spectral range of 0.8 to of explosives at distances of up to 50 m [8]. A high-
20 μm, the substance to be analyzed must be confined energy laser pulse is used to create a plasma on the
to the sample chamber of the instrument, thereby sample surface, and the characteristic atomic emis-
limiting its use to offline measurements in a labora- sion lines of the constituent elements are analyzed
tory environment. Recent advances in mid-IR fiber to determine the sample composition. However, the
technology have enabled the use of fiber probe acces- high beam intensity required for the breakdown lim-
sories for use with FTIR instruments. Fiber-optic re- its the spot size on the sample to a few millimeters
flectance spectroscopy uses fluoride or chalcogenide and may cause thermal degradation of the samples.
fiber bundles that enable portable field analysis by While LIBS offers high sensitivity, the selectivity
allowing the light to be transmitted and collected of the technique is reduced due to contamination
outside the sample chamber [2,3]. However, poor from atmospheric gases. A different technique, called
SNR due to the lamp light source and fiber loss re- photofragmentation laser-induced fluorescence (LIF),
strict the distance of the sample from the fiber probe uses an initial fragmentation of the explosive mole-
tip to a few centimeters. cule followed by LIF detection of characteristic frag-
The need for stand-off detection of samples re- ments [9]. The technique is used to detect the NO
quires the transition from lamp-based sources to molecule and is thus limited to explosives containing
high-power and high-spatial-coherence laser sources. the nitro group. Photothermal spectroscopy [10] is an-
For the detection of gases, lasers with a very low line- other novel method that has been recently shown to
width are ideal due to the narrow spectral absorption detect the presence of explosive residues at a 1 m dis-
features of the gas. The differential absorption light tance using tunable quantum cascade lasers. As the
detection and ranging (DIAL) [4] technique uses two laser is tuned to the peak of an absorption feature of
laser wavelengths (either discrete or tunable) to the sample of interest, a thermal camera is used to
measure the concentration of gases at distances of measure the rise in temperature of the sample. The
up to a few kilometers. The technique works by mea- limitation of this technique is similar to DIAL; i.e.,
suring the relative difference in the backscattered each sample of interest requires a different laser spe-
signal from the two lasers—one that is tuned to the cifically tuned to the absorption wavelength.
center of an absorption peak, while the other is tuned In comparison, our technique uses a spatially co-
outside the peak. However, each target gas requires herent beam from a SC laser to measure the diffuse
the use of different laser sources in the respective reflection spectra of solid samples. The SC laser com-
absorption bands or a widely wavelength tunable bines the advantages of a broad-bandwidth light
source, such as an optical parametric amplifier/ source, such as a lamp, with that of a conventional
oscillator. monochromatic laser. The 3000 nm wide output spec-
Remote Raman spectroscopy is a popular techni- trum covers multiple fundamental and overtone mo-
que for identification of solid and liquid samples at lecular vibration absorption bands, and it allows the
stand-off distances. The sample is illuminated with measurement of a wide range of samples. On the
a high-energy pulse from a visible or ultraviolet other hand, the high output power and single-mode
YAG laser, and the emitted Raman spectrum unique beam quality enable simple stand-off measurements.
to the sample is detected with a gated charge-coupled In particular, we measured the reflection spectra of
device camera. While this technique offers a simple several samples at a distance of 5 m and show the
low-complexity design with excellent overall selectiv- feasibility of extending the range to ∼150 m using
ity, some samples are difficult to analyze due to fluor- SNR calculations. The reflection spectra obtained
escence caused by the intense pump beam. Another using our technique show good agreement with those
potential drawback is the need for extremely sensi- reported in the literature using conventional meth-
tive detection with large collection optics due to ods such as an FTIR or a spectrophotometer.
the inherent weak nature of the Raman effect. How- This paper is organized as follows. Section 2
ever, the sensitivity can be increased using resonance describes the experimental setup for stand-off reflec-
Raman spectroscopy [5], in which the probe laser wa- tion spectroscopy, as well as the details of the mid-IR
velength tuned near an electronic transition within SC laser used for the experiments. In Section 3 we
the molecule gives rise to significant enhancement describe the experimental results obtained with
of certain Raman lines. Coherent anti-Stokes Raman three different sample sets. The first set consists of
spectroscopy [6] is another variation to improve sen- 4%–8% explosives deposited on fused silica crystals,
sitivity, though at the cost of a big increase in system and the second set consists of miscellaneous samples
complexity. It is based on a nonlinear conversion of such as ammonium nitrate, urea, gypsum, and pine
three laser beams into a coherent laserlike Raman wood. The third set consists of aluminum blocks
beam of high intensity in the anti-Stokes wavelength painted with different automotive and military
region. Stand-off Raman spectroscopy has success- paints. The repeatability of the reflectance measure-
fully been used in the fields of geology, chemical ments and comparison of the stand-off spectra with

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2795

that obtained using an FTIR instrument are also ∼105 V∕A and connected to a lock-in amplifier setup
described in this section. Section 4 describes a corre- for high sensitivity detection. The use of a chopper
lation algorithm for distinguishing between the dif- and lock-in amplifier enables low noise detection
ferent samples on the basis of their unique reflection by reducing the noise bandwidth in the measure-
spectra. The various performance metrics of the sys- ment. In our experiment, the chopper frequency is
tem, their limitations, and scope for further improve- set to 400 Hz and the lock-in time constant is set
ment are discussed in Section 5. Finally, we to 100 ms corresponding to a noise bandwidth
summarize the results in Section 6. of ∼1 Hz.
The reflection spectrum of a given sample is ob-
2. Experimental Setup tained using the following procedure. First, a refer-
ence scan is obtained using an “Infragold” diffuse
A. Stand-Off Detection Using Diffuse Reflectance reflectance target as the sample. Infragold from Lab-
Spectroscopy sphere, Inc., is an electrochemically plated diffuse
The experimental setup for a reflection-spectroscopy- gold metallic coating that has >95% reflectivity from
based stand-off detection system is shown in Fig. 1, 1 to 20 μm, and a nearly ideal Lambertian scattering
while details about the SC light source are described profile. Although the SC short wavelength edge is
later in Subsection 2.B. First, the diverging SC out- 750 nm, the scans we obtain start at 1200 nm due
put from the ZBLAN fiber is collimated to a 1 cm dia- to the rapid fall-off of the grating efficiency as well
meter beam using a 25 mm focal length, 90° off-axis, as detector responsivity at shorter wavelengths.
gold-coated, parabolic mirror. To reduce the effects of The scan from 1200 to 4200 nm is comprised of three
chromatic aberration, refractive optics are avoided in different scans, each with a different long-pass filter
the setup. All focusing and collimation are done rotated into place via a circular stepper motor. A si-
using metallic mirrors that have almost constant re- licon filter (1100 nm long pass) is used to scan from
flectivity and a focal length over the entire SC output 1200 to 2100 nm, a germanium filter (1800 nm long
spectrum. The sample is kept at a distance of 5 m pass) to scan from 1800 to 3400 nm, and a 3000 nm
from the collimating mirror, which corresponds to long pass filter to scan from 3100–4200 nm. The
a total round-trip path length of 10 m before reaching three different sections are then stitched together
the collection optics. A 12 cm diameter silver-coated using the overlap regions to form a single continuous
concave mirror with a 75 cm focal length is kept scan from 1200 to 4200 nm. Despite the use of three
20 cm to the side of the collimation mirror. The mir- long-pass filters in conjunction with a stitching algo-
ror is used to collect a fraction of the diffusely re- rithm, we obtain highly reproducible spectra for each
flected light from the sample and focus it into the measurement. The time taken for the complete scan
input slit of a monochromator. Thus, the beam is in- is about 150 s and is limited by the scanning mono-
cident normally on the sample, but detected at a re- chromator. A more detailed discussion regarding the
flection angle of tan−1
0.2∕5 ∼ 2.3°. Appropriate long acquisition time is contained in Subsection 5.D.
wavelength pass filters mounted in a motorized ro- After obtaining a reference scan, the Infragold tar-
tating filter wheel are placed in the beam path before get is replaced with the sample of interest, and a
the input slit to avoid contribution from higher wave- second scan is obtained using the same procedure de-
length orders from the grating (300 groves∕mm, scribed above. Because the beam passes through the
2 μm blaze). The output slit width is set to 2 mm, same optics in each case, the wavelength dependence
corresponding to a spectral resolution of 10.8 nm, of the system due to the SC shape, detector response,
and the light is detected by a 2 mm × 2 mm liquid- and grating response is identical for both the refer-
nitrogen-cooled (77 K) indium antimonide (InSb) ence and sample scans. Thus, dividing the sample
detector. The detected output is amplified using scan by the reference scan gives the reflectance spec-
a transimpedance preamplifier with a gain of trum of the sample normalized to a Lambertian re-
flector kept at the same distance. Some samples
cannot be oriented in the vertical plane normal to
the collimated SC beam due to their powdered nat-
ure. To solve this problem, a fold mirror inclined at
45° is added close to the sample to bend the beam 90°
and allow the sample to be placed horizontally on
the table.
After the reference and sample scans are acquired,
postprocessing of the data is performed in MATLAB.
The percentage sample reflectance is obtained by
Fig. 1. Experimental setup for SC-based stand-off diffuse reflec-
dividing the sample scan by the Infragold reference
tion spectroscopy. Collimated light from the SC source is incident
normally on the sample 5 m away, and diffusely reflected light at a
scan, followed by smoothening of the reflectance
reflection angle of 2.3° is collected by a concave mirror. Low noise curve using a five-point Savitzky–Golay filter [11].
detection of the signal is achieved with a lock-in amplifier con- Finally, the three outlier data points at 1530, 1540,
nected to a liquid-nitrogen-cooled InSb detector at the monochro- and 1550 nm are deleted from the final reflectance
mator output. curve due to the saturation of the lock-in amplifier

2796 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

Fig. 2. Experimental setup for mid-IR SC generation in ZBLAN fiber. Light from a 1543 nm telecom laser diode is amplified to ∼19 kW
peak power in two stages of cladding pumped erbium-ytterbium fiber amplifiers. The output is mechanically coupled into an 8 μm core
diameter and 9 m long ZBLAN fiber to produce a broadband SC spectrum.

signal level close to the seed laser wavelength of of the ZBLAN fiber are angle-cleaved to reduce back
1543 nm. For certain samples, the reflectance curve reflections, alignment of the two fibers is done under
is transformed to a pseudoabsorbance curve using a microscope to ensure the correct rotational orienta-
the transformation A  log
1∕R for better visualiza- tion of the fiber ends for maximum coupling.
tion of the features of interest. At the output of the ZBLAN, we measure 3.9 W of
average output power (with a 50% duty cycle modu-
B. Mid-IR SC Generation in ZBLAN Fiber lation) and a SC spectrum extending from 750–
A block diagram of the mid-IR SC source used for the 4300 nm (Fig. 3) at the −20 dB level. The output
above experiments is shown in Fig. 2. First, a pulse spectrum from 750 to 1750 nm is measured using
generator is used to produce 0.5 ns pulses at a 2 MHz a calibrated optical spectrum analyzer, while the
repetition rate from a 1543 nm distributed feedback spectrum from 1200 to 4300 nm is measured using
(DFB) seed laser diode. Next, the light is amplified to a monochromator setup described in Subsection 2.A.
a peak power of ∼19 kW using dual-stage cladding The latter is corrected for both the detector and grat-
pumped Er-Yb fiber amplifiers, pumped by high- ing response and then stitched together with the
power 976 nm laser diodes. A 0.8 nm wide narrow- short wavelength spectrum to obtain a continuous
band filter centered at the seed wavelength is placed curve from 750 to 4300 nm. Finally, the y axis is con-
between the two gain stages to remove the out-of- verted to an absolute scale representing the SC spec-
band amplified spontaneous emission from the pre- tral power density in dBm∕nm (mW∕nm) by scaling
amplifier stage. To prevent heat-induced damage the spectrum to ensure that the area under the curve
to the power amplifier, the gain fiber is cooled by represents the measured average power. Table 1
wrapping it in a spiral groove machined on a cylind- shows the power in different wavelength bands of the
rical-shaped copper heat sink. In addition, the entire SC measured by using appropriate long wavelength
system is modulated at 500 Hz with a 50% duty cycle pass (LP) filters and a powermeter. Because the en-
to further reduce the thermal load. At the highest tire SC output is modulated with a 50% duty cycle,
976 nm pump power of 52 W (104 W without modu- the average power during the on time is 2× the va-
lation), the 1543 nm signal output from the power lues listed in the table. The broad SC is generated
amp is 9.6 W, corresponding to a pump-to-signal ef- by an interplay of nonlinear effects, such as modula-
ficiency of 18.5%. Finally, the output from the power- tion instability and stimulated Raman scattering,
amp is coupled into a 9 m ZBLAN fiber (8∕125 μm, while the long wavelength edge is limited by the
0.27 NA) via a mechanical fiber alignment stage. Be- inherent material absorption loss of the ZBLAN
cause both the output of the power amp and the input glass [12].
A unique feature of the SC system is the ability to
scale the average output power, while maintaining
the same spectral shape and, hence, the same

Table 1. Supercontinuum Output Power in Different

Spectral Regionsa

Output Power
Pump power (976 nm) 52.0 W
Power amp output (1543 nm) 9.6 W
ZBLAN output (entire SC) 3.9 W
1800 nm LP 2.7 W
2500 nm LP 2.0 W
3000 nm LP 1.3 W
3500 nm LP 0.7 W
3800 nm LP 0.5 W
Fig. 3. SC output from 9 m ZBLAN fiber: spectrum spanning
a
750–4300 nm with 3.9 W average output power. Output has The values are measured using a powermeter and
>1 mW∕nm
0 dBm∕nm spectral power density in the wave- long-pass filters. Because the SC output is modulated
length range from 2000 to 4200 nm. The broad SC is generated with a 50% duty cycle pattern, the average power
by a mixture of nonlinear effects such as modulation instability during the on time of the SC is 2× the values listed
and stimulated Raman scattering. in the table.

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2797

 nitrate and urea are common fertilizers, but are also
often used as explosives. These samples are ground
to a fine powder in a mortar and pestle and filled to a
depth of ∼5 mm in a shallow glass container. We also
measure the reflection spectrum of a 10 cm diameter
×0.5 cm thick gypsum (CaSO4 :2H2 O) disk and a
5 cm × 5 cm × 0.5 cm piece of pine wood, because
these samples are relevant in the remote sensing
community (minerals and vegetation).
The final set of samples is selected to distinguish
between commercial automotive and military vehicle
paints based on their reflection signatures. We ob-
tained red, black, and green acrylic-based spray
Fig. 4. Power scalability of SC output power. The plot shows a
paints from an auto supply store and sprayed three
linear increase in SC average output power with the seed laser coats on different areas of a sanded aluminum block
pulse repetition rate. The corresponding 976 nm pump power used to make our automotive paint samples. The sample
in the power amp stage is listed next to the data points. The SC of the military paint consisted of an aluminum block
was operated at the highest repetition rate of 2 MHz for all the coated with a chemical agent resistant coating
sample measurements. (CARC) green paint, and was loaned to us by
General Dynamics—Advanced Information Systems
fractional distribution of power in different wave- (Ypsilanti, Mich.).
length bands. We accomplish this by increasing the
repetition rate of our seed laser along with a propor- 3. Experimental Results
tional increase in the 976 nm pump power, while
maintaining a constant peak power going into the A. Sample Set #1—NESTT—TNT, RDX, PETN,
ZBLAN fiber [13]. Figure 4 shows the linear increase Potassium Nitrate
in the SC average output power with an increase in The chemical structure and molecular formula of the
the seed laser repetition rate, with the corresponding four NESTT samples are shown in Fig. 5, while the
976 nm power amp pump power labeled next to the absorbance spectra obtained using our SC source are
data point. The results shown in Fig. 3 and Table 1 shown in Fig. 6. For each sample, the positions of
correspond to our highest power results obtained at the strongest/unique peaks have been labeled for
the 2 MHz repetition rate and 52 W pump power. A clarity, and they are in good agreement with FTIR
further increase in output power was limited by the measurements reported in the literature [15–17].
available pump power, and we expect the results to The common broad absorption feature around
scale up with higher pump power and thermal man- 2720 nm present in all four samples arises due to
agement of the critical splice points and power amp the O–H stretching vibration of hydroxyl groups from
gain fiber. absorbed water in the fused silica host.
TNT belongs to a class of compounds known as ni-
C. Sample Preparation troaromatics, in which the carbon directly attached
Three sets of solid samples are chosen to demon- to the nitro (NO2 ) group is part of an aromatic ring.
strate the stand-off diffuse reflection spectra mea- The strongest peaks in the spectrum observed at
surement capability of our system. The first set 3230 and 3270 nm are due to the fundamental
consists of “Non-Hazardous Explosives for Security C–H stretching vibrations in the aromatic ring.
Training and Testing” (NESTT) manufactured by the The remaining peaks from 3300 to 3500 nm are also
XM Division of VanAken International. These sam- due to the C–H stretch, but from the alkyl group. We
ples contain small amounts of explosives deposited also observe sharp but weak features between 2200
on an inert fused silica powder substrate. The result- and 2600 nm, which arise due to the combination
ing products are nonflammable, shock-insensitive, between the C–H stretch and C–H bend vibrations.
and easily transportable, while retaining the unique Finally, the features between 3600 and 3850 nm are
chemical signature of the explosive that makes them
detectable. NESTT samples have previously been
used as substitutes for pure explosives in stand-off
optical detection using Raman spectroscopy [14].
In our experiments, we use the following samples—
trinitrotoluene (TNT), research department explo-
sive (RDX), pentaerythritol tetranitrate (PETN), and
Fig. 5. Chemical structure and formula of NESTT samples—
potassium nitrate. The TNT, RDX, and potassium ni- TNT, RDX, PETN, potassium nitrate. TNT belongs to the nitroaro-
trate NESTT samples have 8% (by weight) explo- matics class, RDX to nitramines, and PETN to nitrate esters. The
sives, while the PETN sample has 4%. fundamental and overtone vibrations of different functional
The second sample set consists of ammonium groups are responsible for the unique spectral features observed
nitrate, urea, gypsum, and pine wood. Ammonium in each sample.

2798 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

Fig. 6. Absorbance spectra of NESTT sample set #1—TNT, RDX, PETN, potassium nitrate. The strongest absorption features in TNT,
RDX, and PETN are in the 3200–2500 nm band and arise due to the fundamental aromatic and aliphatic C–H stretch. The spectral
features around 3600 nm in potassium nitrate arise due to the first overtone of the N–O asymmetric stretch. The common broad feature
at 2720 nm in all four samples is due to the O–H stretch from absorbed water in the fused silica host.

due to the first overtone of the nitro group symmetric predominant spectral features in the gypsum
stretch. (CaSO4 :2H2 O) reflectance occur due to the funda-
RDX belongs to the nitramines class containing mental as well as combination bands of the water
the N–NO2 bond and also has multiple features in molecule, and the locations are consistent with pre-
the 3200–3500 nm band due to the C–H stretch vi- viously reported values [18]. These valleys are pre-
brations. The two strongest peaks in this region sent at 1450, 1750, 1940, and 2860 nm. In addition
are at 3250 and 3330 nm. The spectrum also contains to the features from water, we also observe small dips
the C–H combination bands from 2200 to 2600 nm in the spectrum at 2220, 2260, and 2480 nm, which
and very weak nitro group overtones from 3600 to arise due to the first overtone of the S–O bending vi-
3850 nm. bration. Finally, the valley at 3970 nm occurs due to
PETN is classified as a nitrate ester containing the the first overtone of the —O–S–O stretching vibration
C–O–NO2 bond, and its reflection spectrum is char- of the sulfate (SO2−
4 ) ion.
acterized by a triplet of peaks at 3310, 3350, and The pine wood spectrum comprises bands due to
3440 nm due to the C–H stretch vibration from its main constituents—cellulose, lignin, and water.
the aliphatic groups. The C–H combination band is Poli et al. [19] used fiber-optic reflectance spectro-
also present from 2200 to 2600 nm, while the peak scopy to obtain the reflection spectrum of wood and
around 3900 nm corresponds to the first overtone identify the peak assignments. As is the case in the
of symmetric stretching vibration of the O–NO2 gypsum reflection spectrum, the valleys at 1450,
bond. We will show later in Section 4 that the bands 1920, and 2860 nm are attributed to water. The dip
from 3200 to 3500 nm and 2200 to 2600 nm will be at 2100 nm is due to the first overtone of the C–O
used to distinguish between the three explosive asymmetric stretch, the one at 2270 nm is due to
samples due to the unique positions and shapes of the combination band of O–H and C–H, and the one
various peaks. at 2490 nm is due to the combination band of C–H
Potassium nitrate, being an inorganic compound, and C–O. Finally, the broad feature around 3450 nm
does not contain any absorption features due to is due to the C–H stretching vibration.
the C–H bond present in the other three samples. In- The ammonium nitrate (NH4 NO3 ) spectrum has
stead, the unique spectral feature for this sample is a three prominent features in the near-IR region,
pair of peaks at 3590 and 3650 nm, which arise due to which were also observed by Canal et al. [20] using
the first overtone of the asymmetric N–O stretching a halogen light source. The dip at 1270 nm is due to
vibration of the nitrate ion (NO3 ). the combination of N–H stretching and N–H bending
B. Sample Set #2—Gypsum, Pine Wood, Ammonium vibrations, while the dip at 1570 nm is due to the first
Nitrate and Urea overtone of N–H stretch. The doublet at 2050 and
The reflection spectrum for the four samples is 2140 nm are possibly due to the second overtone of
shown in Fig. 7. While the absorption features show the N–H bending vibrations, while the fundamental
up as peaks in the absorbance spectrum, they are N–H stretch appears as a broad feature around
present as valleys in the reflection spectrum. The 3000 nm.

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2799

Fig. 7. Reflection spectra of sample set #2—gypsym, pine wood, ammonium nitrate, urea. The main spectral features in gypsum arise due
to fundamental and combination bands of water. The pine wood spectrum has absorption features from C–H, O–H, and C–O stretches
attributed to cellulose, lignin, and water. The N–H stretch and N–H bending vibrations are responsible for absorption dips in ammonium
nitrate, while the C–O stretch produces absorption features in urea.

Urea
NH2 2 CO has two amide (––NH2 ) groups Because all four paints contain a variety of organic
joined by a carbonyl (C═O) functional group. The compounds, we observe strong features between
absorption line at 1490 nm occurs due to the third 3200 and 3500 nm from the C–H stretch and from
overtone of the C═O stretching vibration, while 2200 to 2600 nm due to the C–H stretch and C–H
the line at 1990 nm is due to the second overtone bend combination band. In addition, the 2850 to
of the same [21]. 3150 nm band in automotive red and green paints
has absorption features due to the N–H stretch from
C. Sample Set #3—Automotive and Military Paints the acrylic-melamine base. However, the primary
In our final set of samples, the objective is to distin- difference between the automotive and CARC paint
guish between commercial automotive paints and is the presence of a strong dip between 1200 and
military-grade CARC paint based on the differences 1850 nm in the latter. Using the Lawrence Berkeley
in their reflection spectra. The obtained reflection National Laboratory Pigment Database [22], we
spectra of the three automotive and one military found out that the above feature could be attributed
paint samples are shown in Fig. 8. Unlike the pre- to the absorption from Cobalt Chromite—a green
vious sample sets, the paints consist of a complex pigment found in CARC-green. The measured spec-
mixture of many different chemicals, and, hence, it trum of CARC-green is in good agreement with that
is not possible to identify individual absorption lines. reported in the literature [23]. While the features
However, we can make some broad observations re- due to Cobalt Chromite were the main difference be-
garding the spectral features of the paints. tween the paints in our small sample set, the above
The difference in reflectance levels of the four green pigment might not work well as a universal
paints can primarily be attributed to the surface fin- discriminator between all military and automotive
ish of the paint coats. While the auto-red has a glossy paints. A more detailed study with a much larger
finish, the auto-green, auto-black, and CARC-green sample set is required to determine the optimum
paints have a more matte finish. Because our reflec- wavelength bands for distinguishing between the
tion measurement is performed at an angle of only two types of paints.
2.3° away from the specular direction, we expect a
higher reflection from the glossy paint. It is impor- D. Repeatability of Reflection Measurements
tant to note that we expect different reflectance le- The repeatability of the sample reflectance measure-
vels for the same sample with a different surface ments is directly related to the amplitude stability of
finish and at different measurement angles. Thus, the SC source. We measure the magnitude of the
in Section 4, we will demonstrate a sample identifi- SC fluctuations by directly coupling the SC output
cation algorithm that is only sensitive to the shape to the monochromator input, and recording the
and feature positions of the spectrum and not the output versus time for a given wavelength. The lock-
absolute reflectance level. in amplifier time constant is kept at 100 ms (the

2800 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

Fig. 8. Reflection spectra of sample set #3—automotive and military CARC paints. The strong absorption from 3200 to 3500 nm in all
paints is due to the C–H stretch, while that from 2850 to 3150 nm in the auto-red and auto-green is due to the N–H stretch from the acrylic-
melamine base. The unique feature of the CARC-green paint is a broad absorption between 1200 and 1850 nm due to the green pigment
Cobalt Chromite.

same value as the setup for sample measurements) Thus, the error bar on the reflectance measurements
and corresponds to an effective noise bandwidth of can be equated to the percentage standard deviation
∼1 Hz. Figure 9 shows the variation of the SC output (root mean square error) of the normalized SC
at 4000 nm over a time period of 300 s. The plot has output.
been normalized to a mean of 100%, and we obtain a
E. Comparison of SC and FTIR Measurements
standard deviation of 0.16%. While the plot is shown
for a wavelength of 4000 nm, we obtain similar va- In order to verify the validity of our measurements,
lues across the entire SC spectrum. Next, the laser we compared the reflection spectrum obtained using
is attenuated by 10 dB using a neutral density filter our setup with that obtained using a diffuse reflec-
and the experiment is repeated. We find that the tance accessory in an FTIR spectrometer. As an ex-
standard deviation is almost identical to the previous ample, Fig. 10 shows a comparison of the powdered
case, confirming that the SC fluctuations can be mod- urea reflection spectrum obtained using the two
eled as a fixed root-mean-square percentage fluctua- methods. There is good agreement in the shape
tion superimposed on the mean SC signal level. and wavelength position of the various spectral fea-
tures. However, there is a difference in the absolute
reflectance levels of the two curves. We attribute the
difference in scale to the different measurement geo-
metries of the two setups. The SC-based setup de-
scribed in Subsection 2.A measures the reflection
spectrum of the sample at a fixed angle of 2.3° from
the normal. However, the FTIR accessory uses a con-
cave mirror that collects light over almost the entire
hemisphere above the sample. Thus, the latter meth-
od produces a directionally integrated spectrum,
compared to a single direction spectrum from the for-
mer. The shape and positions of the absorption lines
depend mostly on the chemical composition of the
sample, and hence do not change significantly be-
tween the two methods. However, the magnitude
of light reflected along different directions depends
on the sample texture and, thus, is responsible for
Fig. 9. Variation of SC output at 4000 nm over 300 s. With the
signal normalized to a mean of 100%, we obtain a standard devia-
the scale difference between the two spectra.
tion of 0.16%. The standard deviation remains unchanged even
4. Sample Identification Algorithm
when the signal is attenuated by 10 dB. Thus, the SC fluctuations
can always be expressed as a fixed root-mean-square percentage of This section describes an algorithm that demon-
the mean SC output, regardless of the absolute signal level. strates the selectivity of our technique, i.e., the ability

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2801

Fig. 10. Comparison of urea reflection spectrum obtained using the SC-based setup with that obtained using an FTIR with a diffuse
reflection accessory. There is good agreement in the shape and peak positions of the two spectra. The difference in absolute scale between
the two plots is attributed to the difference in reflection measurement geometries of the two setups.

to distinguish one sample from another based on the correlation. The value of r is independently calcu-
unique spectral features of each sample. As an exten- lated for each selected wavelength band using the
sion, the algorithm can also be used to identify an un- following equation:
known sample by comparing it against a precompiled
library of relevant samples. A flow chart describing Pn
the different steps of the algorithm is shown in Fig. 11. i1
X i − X
Y i − Y
r  q

































Pn q
































:
Pearson’s correlation coefficient r always lies be- 2 Pn 2

X i − X i1
Y i − Y
tween −1 and 1, with 1 indicating perfect positive i1

correlation and −1 indicating perfect negative

We test the above algorithm on each of the three

sample sets. For each sample in a given set, one or
two wavelength bands containing a unique spectral
signature are chosen. Tables 2, 3, and 4 show the re-
sults of the correlation between different samples in
sample set #1, #2, and #3, respectively. As expected,
all the diagonal elements of the tables are equal to 1
(the definition of autocorrelation), while the others
are below 0.9, thus confirming the ability to distin-
guish different samples within a set. Table 3 only
contains a single row for CARC paint, as the objective
was to distinguish the military paint from the auto-
motive paints and not the difference between the
different color automotive paints themselves.
The advantage of the correlation-based algorithm
we have chosen is the insensitivity to the absolute
level of the sample reflectance. The correlation value
only depends on the shape and position of the spec-
tral features, and it can be used to compare samples
measured using different reflection geometries,
as mentioned in Subsection 3.E. While the above
algorithm demonstrates proof of concept for distin-
guishing between distinct samples using multiple
wavelength bands, more sophisticated algorithms
[24] have been used by the remote sensing commu-
Fig. 11. Flow chart describing the algorithm to demonstrate se-
lectivity between two samples. The algorithm uses the correlation
nity for distinguishing between geological samples
coefficient in selected spectral bands to quantify the degree of si- with similar compositions. In addition to the detec-
milarity or dissimilarity between two reflection spectra. If the tion of pure samples, algorithms using derivative
mean correlation coefficient over the selected bands >0.9, the spectroscopy [25] and multivariate curve resolution
two samples are said to be a match. [26] have also been demonstrated in the literature

2802 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

 Table 2. Correlation Table for Sample Set #1: NESTT—TNT, RDX, PETN, and Potassium Nitrate

Test Samples
Reference Sample Band 1 Band 2 TNT RDX PETN Potassium Nitrate
TNT 2200–2600 3200–3500 1.00 0.47 0.05 0.01
RDX 2200–2600 3200–3500 0.47 1.00 −0.05 −0.33
PETN 2200–2600 3200–3500 0.05 −0.05 1.00 0.03
Potassium Nitrate 3500–3700 — −0.81 −0.46 −0.38 1.00

Table 3. Correlation Table for Sample Set #2: Ammonium Nitrate, Urea, Gypsum, and Pine Wood

Test Samples
Reference Sample Band 1 Band 2 Ammonium Nitrate Urea Gypsum Pine Wood
Ammonium Nitrate 1400–1770 1780–2470 1.00 0.34 −0.25 0.24
Urea 1340–1620 3980–4150 −0.24 1.00 0.15 0.67
Gypsum 1820–2070 3730–4070 −0.16 −0.23 1.00 −0.11
Pine Wood 1300–1840 1850–2200 0.62 0.67 0.39 1.00

Table 4. Correlation Table for Sample Set #3: Military CARC-Green and Automotive Black, Red, and Green Paints

Test Sample
Reference Sample Band 1 Band 2 CARC-Green Auto-Black Auto-Red Auto-Green
CARC-green 1200–1850 2850–3150 1.00 0.21 −0.27 0.17

for the detection of trace amounts in the presence of is calculated by multiplying the optical power inci-
interfering species. dent on the detector with the detector responsivity,
followed by the transimpedance gain of the preampli-
5. Discussion fier. Using the actual values of the various terms in
The limitations and scope for further improvement of our system, the analysis is performed below.
the different aspects of the SC-based diffuse reflec- Detection wavelength, λ  4000 nm (assumption)
tion spectroscopy technique are discussed in this SC spectral power density, PSC  1 mW∕nm
section. Mirror 1 reflection, Rm1  95%
A. Signal-to-Noise Ratio of Reflection Measurements Chopper transmission, T ch  50%
Mirror 2 reflection, Rm2  95%
We begin by performing a detailed signal-to-noise ra- Sample reflectivity, Rs  10% (assumption)
tio (SNR) estimate of our system, which determines Sample to collection mirror distance, d (variable
the smallest relative percentage change in sample parameter)
reflectivity that can be measured at a given stand- Collection mirror radius, r  6 cm
off distance. The signal voltage measured at the Spectrometer output slit bandwidth, Δλ  10.8 nm
lock-in amplifier can be computed by determining Grating efficiency, η  27%
the power reaching the InSb detector after diffuse InSb detector responsivity, D  1.9 A∕W
reflection from the sample. Our analysis assumes a Preamp transimpedance gain, G  1.07 × 105 V∕A.
detection wavelength of 4000 nm and a sample re-
flectivity of 10%. First, the SC output spectral power Signal voltage measured at the lock-in amplifier:
density at the detection wavelength is determined
from Fig. 3 in Subsection 2.B. Next, the power den- V signal  PSC × Rm1 × T ch × Rm2 × Rs ×
π∕d2  ×
πr2 
sity reaching the sample is calculated after reflection
× Δλ × η × D × G
from the collimating mirror, transmission through
the chopper, and a second reflection from the turn  0.096∕d2 . (1)
mirror next to the sample. Assuming the sample to
be a Lambertian reflector, the amount of light cap- The total noise floor of the measurement is a combi-
tured by the concave collection mirror placed a given nation of the detector noise, SC fluctuations, pream-
distance from the sample can also be calculated. plifier noise, and lock-in amplifier noise. Because
Finally, after taking into account the grating each noise term is independent, the total noise can
efficiency and the monochromator output slit band- be computed from the square root of the sum of the
width, the power incident on the detector can be de- squares of the various terms. The detector noise term
termined. The voltage signal on the lock-in amplifier is calculated by multiplying the shot noise current

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2803

density (from the dark current) with the transimpe- Using the generalized equations (1)–(6), the pre-
dance gain of the preamplifier. As was shown in dicted variation of the SNR with sample distance
Subsection 3.D, the SC fluctuations are proportional is shown in Fig. 12. The top curve is calculated for
to the SC output power. Thus, the noise term can be the actual experimental noise bandwidth of 1 Hz,
modeled as a fixed root-mean-square percentage while the middle and bottom curves are for higher
fluctuation with respect to a mean signal level. Final- noise bandwidths of 10 Hz and 100 Hz, correspond-
ly, we also include contributions to the noise from the ing to smaller lock-in time constants of 10 and 1 ms,
transimpedance preamplifier and lock-in amplifier respectively. We observe that the curves have two dif-
voltage noise. However, we will show later that the ferent shapes in two distinct regions. At sample dis-
last two terms are much smaller compared to the de- tances below 10 m, the curve is flat, corresponding to
tector and SC noise. Hence, they can be neglected for the limiting case 1. We observe a constant SNR ∼ 28,
further analysis. After computing the total voltage 23, and 18 dB, respectively, for the top, middle, and
noise density, the total voltage noise in the measure- bottom curves. We call this region the SC fluctuation
ments is calculated by multiplying the noise density noise limited region. However, at longer distances
with the square root of the detection bandwidth (as corresponding to case 2, there is a constant drop in
set by the lock-in amplifier). the SNR with increasing distance. In this region,

InSb detector dark current; I det  36 μA
manufacturer spec

Detector dark current shot noise current density; I det 
2eI det 1∕2 pA∕Hz1∕2
Detector dark current shot noise voltage density; edet  idet × G  363 nV∕Hz1∕2 (2)

SC noise density; eSC  0.0016  V signal nV∕Hz1∕2
from Subsection 3:D (3)

Lock-in amplifier noise; elia  7 nV∕Hz1∕2
manufacturer spec

Detector preamp noise; epa  20 nV∕Hz1∕2
manufacturer spec
   1∕2
Total noise floor density; etot  e2det  e2SC  e2lia  e2SC ∼ e2det  e2SC (4)

Lock-in detection bandwidth; B  1 Hz the SNR is detector noise limited and the SNR falls
1∕2 by 6 dB for every doubling of the sample distance.
Total noise floor; V noise  etot × B (5) Finally, we verify the calculations by experimen-
tally measuring the SNR at different distances
and comparing it to the predicted value. At a wave-
SNR
linear  V signal ∕V noise length of 4000 nm, the reflected signal from a 10%
SNR
dB  10 × log
V signal ∕V noise . (6) reflectivity sample is acquired for a period of 5 min,
and the ratio of the mean to standard deviation is
defined as the SNR. The available space in our lab
The SNR equation can be further simplified for two limited the maximum sample distance to 5 m. The
limiting cases. At a critical sample distance effect of a longer sample distances is emulated by
dc ∼ 20.5 m, eSC  edet —using Eqs. (1), (2), and (3). keeping the sample at the maximum physical dis-
Case 1: At distances d ≪ dc, eSC ≫ edet , and tance of 5 m and adjusting an aperture opening in
etot ∼ eSC . front of the collection mirror to reduce the collected
Thus, SNR  V signal ∕
0.0016 V signal   625  28 dB light by the required 1∕d2 factor. The experimentally

constant. measured results for a 1 Hz noise bandwidth are
Case 2: At distances d ≫ dc, edet ≫ eSC , and plotted using square markers in Fig. 10, and they
etot ∼ edet . are found to be in good agreement with the theoreti-
Thus, SNR 
0.096∕d2 ∕
363e − 9  2.6e5∕d2 (in- cally predicted curve. Future improvements in the
versely proportional to the square of the sample short distance SNR might be accomplished using
distance). a ratiometric dual-beam detection scheme [27] to

2804 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

 measurements up to a distance of 150 m. If the time
constant is reduced from 100 ms (1 Hz noise band-
width) to 10 ms (10 Hz bandwidth) or 1 ms (100 Hz
noise bandwidth), the distance is reduced to 85 or
45 m, respectively, according to Fig. 12. The system
sensitivity at large distances can be improved
using two methods. First, as discussed earlier in
Subsection 2.B, the SC output power may be scaled
up further using higher 976 nm pump power and a
corresponding increase in pulse repetition rate. Sec-
ond, because the noise equivalent power for a photo-
diode is directly proportional to the square root of the
Fig. 12. Variation in predicted system SNR versus sample stand- active area, the detector noise contribution may be
off distance. At small distances <10 m, the noise is dominated by reduced by choosing a detector with a smaller area.
the SC fluctuations and the SNR is independent of distance. At
large distances >20 m, the detector noise term is dominant and C. System Selectivity
SNR is inversely proportional to the square of the sample distance.
The selectivity of a system is defined as the ability to
distinguish between various samples based on their
reduce the effect of the SC fluctuations, and achieve spectral signatures. The conventional “fingerprint”
detector noise limited performance. Another noise region of a sample is defined as 6.5 to 25 μm, as it
term that we have not accounted for is the noise contains many of the strong fundamental bending
due to beam motion and sample nonuniformity. Be- and stretching vibrations. However, we have shown
cause our setup was mounted on a vibration-free op- in Section 3 that the SC spectral coverage from
tical table, we did not observe any changes in 1.2 to 4.2 μm also contains unique spectral features
spectrum due to the relative movement of the beam due to various overtone, combination, and funda-
with respect to the sample surface. However, further mental bands. The selectivity of the system was
work needs to be performed to determine the effect of demonstrated in Section 4, where the correlation-
motion on a field portable system with much larger function-based algorithm was used to quantitatively
stand-off demonstrate the dissimilarity between different
distances. samples using bands within the 3000 nm wide reflec-
B. System Sensitivity tion spectra.
In this paper, we have demonstrated the measure-
The system sensitivity for a given SNR is defined as
ment of the reflectance spectrum of pure samples or
being equal to the minimum band depth (Fig. 13)
those diluted in high concentrations (4%–8%) in a
that can be measured. Under the assumption that
nonabsorbing matrix. However, further research
a band depth ∼3 standard deviations above the noise
needs to be performed on the detection of trace
floor can be measured, the system sensitivity ex-
amounts of relevant samples mixed with other sam-
pressed as a percentage is equal to 300∕SNR. As
ples to determine the selectivity of the technique in
an example, the SNR at 5 m is 625, corresponding
the presence of interfering species. We believe that
to a system sensitivity of 0.48%. Assuming a cut-
an improvement in the system selectivity may be
off value of SNR  10 dB (minimum band depth of
achieved by using a light source with a wider spectral
30%), our current system can be used to make
bandwidth that covers more absorption features of a
sample. An SC source using a 2 cm long fluoride fiber
has recently been demonstrated with a spectrum ex-
tending to 6.3 μm and average output power of
∼20 mW [28]. While the above light source might in-
crease system selectivity due to the wider band-
width, it would also produce a decrease in system
sensitivity due to the lower output power.
D. Acquisition Time
The feasibility of using a stand-off technique in a
real-world field application is strongly dependent on
the data acquisition time. Because the lock-in ampli-
fier time constant in our setup is 100 ms, the acquisi-
tion time should ideally only be ∼0.5 s (five time
constants for the signal to settle). However, the need
to rotate the grating and step through each wave-
Fig. 13. Definition of absorption band depth. Ra is the reflectance length means that a complete scan from 1200 to
value at the bottom of an absorption dip, while Rb is the reflectance 4200 nm takes ∼150 s (300 data points 10 nm
value outside it. apart). The replacement of the current scanning

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2805

monochromator with a linear detector array spectro- prints for each sample. Thus, they provide high
graph would reduce the acquisition time by over 2 selectivity, but often at the cost of increased system
orders of magnitude by measuring the entire spec- complexity.
trum in a single shot. By averaging N such shots, The sensitivity to interferents is closely related to
the SNR of the measurements can be improved by the selectivity of the technique. We expect our tech-
a factor of N 1∕2. Spectrographs for the visible and nique to detect trace amounts of a substance in the
near-IR regions using silicon and InGaAs arrays presence of interfering species, provided the absorp-
are widely commercially available. However, the tion feature of the target sample has a strong absorp-
high cost of manufacturing of mid-IR arrays has kept tion cross section in a wavelength region where the
their use largely limited to the defense and astro- interfering species does not absorb strongly [31].
nomical imaging communities [29]. However, because we have not experimentally mea-
sured the reflection spectra under these conditions,
E. Comparison with Other Optical Stand-Off Techniques this field in the comparison table has been left blank.
In this section, we summarize the various perfor- Finally, in the detection time column, we have listed
mance metrics of our system and provide a the value for our technique as 150 s. From a funda-
comparison with other optical stand-off detection mental SNR standpoint, the demonstrated light
techniques. A recent review by Wallin et al. [30] on source has enough output power to perform a sample
laser-based stand-off detection of explosives features reflectance measurement in just 0.5 s. However, as
a comparison table of different optical techniques. discussed earlier, a complete sample scan in our set-
Table 5 has been adapted from the above publication, up takes ∼150 s due to a limitation of the scanning
and an additional row with our SC-based diffused re- monochromator instrument.
flection spectroscopy system has been added to the
top. The different techniques are compared on the 6. Summary
basis of sensitivity, selectivity, sensitivity to interfer- We measured the diffuse reflection spectrum from
ents, stand-off distance, detection time, and system 1200 to 4200 nm of various solid samples at a
complexity. stand-off distance of 5 m, using a mid-IR SC light
Based on the analysis in Subsections 5.A and 5.B, source with 3.9 W average output power and spec-
we can claim the SC-based diffuse reflection spectro- trum spanning 750 to 4300 nm. The SC source was
scopy technique as having “high” sensitivity due to generated by pumping a 9 m long, 8 μm core dia-
the high output power of the SC source combined meter, and 0.27 NA ZBLAN fluoride fiber with
with the low noise detection scheme. However, like ∼19 kW peak power pulses at 1543 nm from a dual-
all the other techniques, the system sensitivity de- stage cladding pumped Er-Yb fiber amplifier system.
creases with increasing distance for a constant detec- The SC average output power was linearly scalable
tion time. We have listed the stand-off distance as with the 976 nm power-amplifier pump power, and
150 m because the SNR drops below the cut-off value the highest demonstrated output was only limited
of 10 dB at larger distances. In terms of selectivity, by the available pump power.
the SC-based diffuse reflection spectroscopy techni- The first sample set comprised NESTT explosives
que is inferior compared to the Raman-based techni- containing between 4%–8% of TNT, RDX, PETN, and
ques, and it is hence listed as “medium.” Because the potassium nitrate on a fused silica substrate. The
SC spectrum does not cover the conventional molecu- second sample set consisted of ammonium nitrate,
lar vibration fingerprint region of the spectrum be- urea, gypsum, and pine wood, while the third sample
yond 6 μm, we are limited to identifying samples set was made of aluminum blocks painted with auto-
based on the weaker overtone and combination motive and military CARC paint. For each sample,
bands or a limited number of fundamental vibration we observed unique spectral features in the reflec-
bands associated with the constituent functional tion spectrum, attributed to the fundamental and
groups. In comparison, the Raman spectroscopy overtone vibration frequencies of the various molecu-
methods usually measure the entire Raman spectral lar bonds in the sample structure. A correlation-
range of ∼3800 cm−1 containing very unique finger- function-based algorithm was used to demonstrate

Table 5. Comparison of Spectroscopic Optical Stand-Off Detection Techniques (Adapted from Wallin et al. [7])a

Molecular Sensitivity to Detection

Method Sensitivity Selectivity Interferents Distance (m) Time (s) Complexity
SC-based DRS High Medium NA ∼150 ∼150 Low
LIBS High Medium Medium to high >45 ∼1 Low to medium
Raman spectroscopy Medium High Low to medium 470 ∼1 Low
RRS High Very high Low >30 ∼10 Medium
CARS Very high High Low >12 ∼10 Very high
PF-LIF Medium Medium Medium ∼30 ∼1 Medium
a
SC DRS, supercontinuum-based diffuse reflection spectroscopy; LIBS, laser-induced breakdown spectroscopy; RRS, resonance Raman
spectroscopy; CARS, coherent anti-Stokes Raman spectroscopy; PF-LIF, photofragmentation followed by laser induced fluorescence.

2806 APPLIED OPTICS / Vol. 51, No. 15 / 20 May 2012

system selectivity by quantifying the dissimilarity 13. M. Kumar, C. Xia, X. Ma, V. V. Alexander, M. N. Islam, F. L.
Terry, Jr., C. Aleksoff, A. Klooster, and D. Davidson, “Power
between various spectra based on selected spectral adjustable visible supercontinuum generation using ampli-
features in chosen wavelength bands. fied nanosecond gains-witched laser diode,” Opt. Express
Finally, the variation of the SNR with distance was 16, 6194–6201 (2008).
theoretically calculated, and also verified experimen- 14. J. C. Carter, S. M. Angel, M. Lawrence-Snyder, J. Scaffidi,
tally. For a time constant of 100 ms corresponding to R. E. Whipple, and J. G. Reynolds, “Standoff detection of high
explosive materials at 50 meters in ambient light conditions
∼1 Hz noise bandwidth, we obtained a SNR ∼625 at using a small Raman instrument,” Appl. Spectrosc. 59,
distances <10 m. Assuming a minimum detectable 769–775 (2005).
SNR ∼10, the maximum stand-off distance with 15. S. V. Ingale, P. U. Sastry, A. K. Patra, R. Tewari, P. B. Wagh,
the current SC output power and integration time and S. C. Gupta, “Micro structural investigations of TNT and
PETN incorporated silica xerogels,” J. Sol-Gel Sci. Technol.
was determined to be ∼150 m. Future work on the 54, 238–242 (2010).
system may include faster acquisition time using a 16. A. Banas, K. Banas, M. Bahou, H. O. Moser, L. Wen, P. Yang,
linear detector array spectrograph and measure- Z. J. Li, M. Cholewa, S. K. Lim, and Ch. H. Lim, “Post-blast
ment of trace amounts of a sample to determine the detection of traces of explosives by means of Fourier transform
sensitivity of the technique to interferents. infrared spectroscopy,” Vibrat. Spectrosc. 51, 168–176 (2009).
17. J. Janni, B. D. Gilbert, R. W. Field, and J. I. Steinfeld, “Infra-
We would like to thank our sponsors—Omni red absorption of explosive molecule vapors,” Spectrochim.
Sciences, Inc., the U.S. Army Communications- Acta Part A 53, 1375–1381 (1997).
18. G. Anbalagan, S. Mukundakumari, K. S. Murugesan, and S.
Electronics Research, Development and Engineering Gunasekaran, “Infrared, optical absorption, and EPR spectro-
Center (CERDEC, contract no. W15P7T-10-C-H606), scopc studies on natural gypsum,” Vibrat. Spectrosc. 50,
the Defense Advanced Research Projects Agency 226–230 (2009).
(DARPA), and Naval Air Systems Command 19. T. Poli, O. Chiantore, M. Nervo, and A. Piccirillo, “Mid-IR fiber-
optic reflectance spectroscopy for identifying the finish on woo-
(NAVAIR) PMA-272 for their support. den furniture,” Anal. Bioanal. Chem. 400, 1161–1171 (2011).
20. C. M. Canal, A. Saleem, R. J. Green, and D. A. Hutchins,
References “Remote identification of chemicals concealed behind clothing
1. M. P. Fuller and P. R. Griffiths, “Diffuse reflectance measure- using near infrared spectroscopy,” Anal. Meth. 3, 84–91
ments by infrared Fourier transform spectroscopy,” Anal. (2011).
Chem. 50, 1906–1910 (1978). 21. J. E. Stewart, “Infrared absorption spectra of urea, thiourea,
2. I. Schneider, G. Nau, T. V. V. King, and I. Aggarwal, “Fiber and some thiourea‐alkali halide complexes,” J. Chem. Phys.
optic near-infrared reflectance sensor for detection of organics 26, 248–255 (1957).
in soils,” IEEE Photon. Technol. Lett. 7, 87–89 (1995). 22. Lawrence Berkeley National Laboratory Pigment Database—
3. M. Leona and J. Winter, “Fiber optics reflectance spectroscopy: Cobalt Chromite Green Spinel, http://coolcolors.lbl.gov/
a unique tool for the investigation of Japanese paintings,” LBNL‑Pigment‑Database/paints/G05.html.
Studies Conser. 46, 153–162 (2001). 23. P. J. Kaste, R. G. Daniel, R. A. Pesce-Rodriguez, M. A.
4. B. I. Vasil’ev and O. M. Mannoun, “IR differential-absorption Schroeder, and J. A. Escarsega, “Hydrogen plasma removal
LIDARs for ecological monitoring of the environment,” Quan- of military paints: chemical characterization of samples,”
tum Electron. 36, 801–820 (2006). Report no. A128453, Army Research Laboratory, Aberdeen
5. G. Comanescu, C. K. Manka, J. Grun, S. Nikitin, and D. Proving Ground, 1998.
Zabetakis, “Identification of explosives with two-dimensional 24. F. V. D. Meer and W. Bakker, “CCSM: cross correlogram spec-
ultraviolet resonance Raman spectroscopy,” Appl. Spectrosc. tral matching,” Int. J. Remote Sensing 18, 1197–1201 (1997).
62, 833–839 (2008). 25. T. H. Demetriades-Shah, M. D. Steven, and J. A. Clark, “High
6. H. Li, D. A. Harris, B. Xu, P. J. Wrzesinski, V. V. Lozovoy, and resolution derivatives spectra in remote sensing,” Remote
M. Dantus, “Standoff and arms-length detection of chemicals Sens. Environ. 33, 55–64, 1990.
with single-beam coherent anti-Stokes Raman scattering,” 26. N. B. Gallagher, T. A. Blake, P. L. Gassman, J. M. Shaver, and
Appl. Opt. 48, B17–B22 (2009). W. Windig, “Multivariate curve resolution applied to infrared
7. A. J. Hobro and B. Lendl, “Stand-off Raman spectroscopy,” reflection measurements of soil contaminated with an organo-
Trends Anal. Chem. 28, 1235–1242 (2009). phosphorus analyte,” Appl. Spectrosc. 60, 713–722 (2006).
8. J. L. Gottfried, F. C. D. Lucia, Jr., C. A. Munson, and A. W. 27. M. G. Allen, K. L. Carleton, S. J. Davis, W. J. Kessler, C. E.
Miziolek, “Laser induced breakdown spectroscopy for detec- Otis, D. A. Palombo, and D. M. Sonnenfroh, “Ultrasensitive
tion of explosives residues: a review of recent advances, chal- dual-beam absorption and gain spectroscopy: applications
lenges and future prospects,” Anal. Bionanal. Chem. 395, for near-infrared and visible diode laser sensors,” Appl.
283–300 (2009). Opt. 34, 3240–3249 (1995).
9. V. Swayambunathan, G. Singh, and R. C. Sausa, “Laser 28. G. Qin, X. Yan, C. Kito, M. Liao, C. Chaudhari, T. Suzuki, and
photofragmentation-fragment detection and pyrolysis-laser- Y. Ohishi, “Ultrabroadband supercontinuum generation from
induced fluorescence studies on energetic materials,” Appl. ultraviolet to 6.28 μm in a fluoride fiber,” Appl. Phys. Lett. 95,
Opt. 38, 6447–6454 (1999). 161103 (2009).
10. R. Furstenberg, C. Kendziora, M. Papantonakis, S. V. 29. J. T. Rayner, D. W. Toomey, P. M. Onaka, A. J. Denault, W. E.
Stepnowski, J. Stepnowski, V. Nguyen, M. Rake, and R. A. Stahlberger, W. D. Vacca, M. C. Cushing, and S. Wang,
McGill, “Stand-off detection of trace explosives via resonant “A medium-resolution 0.8–5.5 micron spectrograph and ima-
infrared photothermal imaging,” Appl. Phys. Lett. 93, 224103 ger for the NASA Infrared Telescope Facility,” Publ. Astron.
(2008). Soc. Pac. 115, 362–382 (2003).
11. A. Savitzky and M. J. E. Golay, “Smoothing and differentiation 30. S. Wallin, A. Pettersson, H. Ostmark, and A. Hobro, “Laser
of data by simplified least squares procedures,” Anal. Chem. based standoff detection of explosives: a critical review,” Anal.
36, 1627–1639 (1964). Bioanal. Chem. 395, 259–274 (2009).
12. C. Xia, M. Kumar, O. P. Kulkarni, M. N. Islam, F. L. Terry, Jr., 31. L. A. Averett and P. R. Griffiths, “Mid-infrared diffuse reflec-
M. J. Freeman, M. Poulain, and G. Mazé, “Mid-infrared super- tion of a strongly absorbing analyte on non-absorbing and
continuum generation to 4.5 μm in ZBLAN fluoride fibers by absorbing matrices. Part I: Homogeneous powders,” Appl.
nanosecond diode pumping,” Opt. Lett. 31, 2553–2555 (2006). Spectrosc. 62, 377–382 (2008).

20 May 2012 / Vol. 51, No. 15 / APPLIED OPTICS 2807