INFRARED SPECTROSCOPY / Near-Infrared 415
See also: Infrared Spectroscopy: Near-Infrared. Photo- Harrick NJ (1967) Internal Reflection Spectroscopy. New
acoustic Spectroscopy.
York: Harrick Scientific Corporation.
Messerschmidt RG and Harthcock MA (eds.) (1988) In-
frared Microspectroscopy: Theory and Applications.
Further Reading
New York: Dekker.
Mirabella FM and Harrick NJ (1985) Internal Reflection
Chalmers JM and Dent G (1997) Industrial Analysis with
Spectroscopy: Review and Supplement. New York:
Vibrational Spectroscopy. Cambridge: Royal Society of
Harrick Scientific Corporation.
Chemistry.
Siesler HW and Holland-Moritz K (1980) Infrared and
Chalmers JM and Griffiths PR (eds.) (2002) Handbook of
Raman Spectroscopy of Polymers, Practical Spectro-
Vibrational Spectroscopy, vol. 2. Chichester: Wiley.
scopy Series, vol. 4. New York: Dekker.
Coleman PB (ed.) (1993) Practical Sampling Techniques
Willis HA, van der Maas JH, and Miller RGJ (eds.) (1987)
for Infrared Analysis. Boca Raton, FL: CRC Press.
Laboratory Methods in Vibrational Spectroscopy, 3rd
Colthup NB, Daly LH, and Wiberley SE (1990) Introduc-
edn. Chichester: Wiley.
tion to Infrared and Raman Spectroscopy, 3rd edn.
San Diego, CA: Academic Press.
Near-Infrared
C N G Scotter, Cherington, UK
Mid-IR and NIR spectra are generated via appro-
priate instrumentation by electromagnetic energy ab-
& 2005, Elsevier Ltd. All Rights Reserved.
sorption between 400 and 12 500 nm. Subsets of the
wavelength range are employed by instruments with
Introduction
detectors and energy sources appropriate for obtain-
ing spectra in their particular range. The mid-IR
Since the 1970s, near-infrared (NIR) spectroscopy
range is about 2500 25 000 nm, the NIR range
has been applied as a qualitative and quantitative
about 1100 2500 nm, and the visible range about
analytical tool in the industrial fields of agriculture,
400 800 nm. In the mid-IR range it is usual to refer to
food, chemical and textile production, and pharma-
1
wavenumber or frequency in cm according to the
ceuticals. The technique is capable of making rapid,
formula wave number (cm 1) ź 107/wavelength (nm).
nondestructive multicomponent analyses of organic
The mid-IR region provides fundamental absorp-
materials in a complex background matrix. In situ-
tion data, whereby for a specific bond pair the energy
ations where inorganic materials are dissolved in po-
transfer from one level to another occurs at a unique
lar solvents an indirect NIR measurement can be
frequency. It is for this reason that the mid-IR region
obtained. If the concentration of an analyte exceeds
has been used to analyze the bond pairs in organic
0.1% then it is likely that an acceptable analytical
1
molecules. It is the 3600 1200 cm region that
error will be achieved. NIR spectroscopy is essen-
gives rise to the overtones in the NIR part of the
tially an applied instrumental technique not often
spectrum. The NIR region thus consists of overtones
used in a pure research context, unlike the related
and combination bands that result in broad and/or
mid-infrared (mid-IR) spectroscopy, the spectra from
overlapping absorption peaks. It is for this reason
which can be interpreted to identify specific organic
that NIR spectroscopy employs more statistical,
bond pairs, e.g., CO, OH, NH. This article discusses
mathematical approaches to the analysis of organic
the physical theory underpinning the use of NIR,
constituents, compared with the approaches used in
NIR calibration methodology, and types of instru-
mid-IR spectroscopy.
mentation. Examples are given of the application of
Thus, the two regions, although instrumentally
the technique in the industrial contexts noted above.
and chemometrically distinct, are theoretically parts
of the same vibrational and rotational spectroscopic
NIR Theory  Molecular Spectroscopy
model. Figure 1 illustrates the basic models for
Two main areas of physical theory are relevant to the energy absorption by covalently bonded atoms of
application and understanding of NIR: first, the dissimilar mass. A dipole moment (produced when
chemical principles derived from quantum physics atoms of dissimilar mass are bonded) is a basic
and molecular spectroscopy, and, second, the physi- condition for absorption to occur. The H H bond,
cal principles used to relate the spectra to transmis- for example, does not produce mid-IR or NIR
sion or diffuse reflectance of electromagnetic energy. absorbance. When molecules are subjected to an
416 INFRARED SPECTROSCOPY / Near-Infrared
multiples of the fundamental frequency. Greater
amounts of energy are required to displace mole-
cules to higher overtones and the numbers of mole-
cules displaced is in inverse proportion to the
Harmonic
overtone number. Transitions occur most frequently
potential
from the ground state (n ź 0, see Figure 1) to the first
level, and are increasingly infrequent for the higher
levels. Consequently, higher overtone absorptions are
much weaker than the fundamental.
Anharmonic potential
Combination bands are also encountered in NIR
(Morse curve)
spectra. These result from energy transitions invol-
ving two or more different vibrational modes for the
The second overtone
same functional group. For example, CH2 groups
from the energy transition
= 3 have two symmetrical vibrational modes, a symmet-
between = 0 and = 3
ric stretching mode, and a corresponding in-plane
= 2 The first overtone from
bending, or scissoring. The combination band result-
the energy transition
ing from these two vibrational modes is seen at
between = 0 and = 2
= 1
2320 nm. A final factor that has been identified as
The fundamental vibration
complicating NIR spectra even further is that of
from the energy transition
Fermi resonance. In this condition, unalike sets of
between = 0 and =1
= 0
bonds, e.g., the O C O bending mode and the sym-
metrical C O stretching mode, occur at approxi-
mately the same frequency and a repulsion occurs.
Interatomic distance
One type of bond energy level is lowered and the
other type gains energy. In the case of carbon dioxide
Figure 1 Hooke s law and Morse curves as functions of energy
1
and interatomic distance. this results in bands at 1388 and 1286 cm instead
of one strong and one weak line at 1340 and
1
1330 cm , respectively. Although relatively obvious
external energy source they acquire the potential for in the mid-IR region, Fermi resonance phenomena
energy changes and the amount of energy that can be are usually hidden beneath the broad overlapping
imparted from a source is dependent upon the peaks in the NIR.
wavelength, according to the formula: Although NIR absorbance peaks are generally
broad, spectral assignment is possible. A number of
authors have discussed bond assignment in the NIR
E ź hcn
%
region and tables and lists of assignments of partic-
ular bonds and vibrational modes have been pub-
where E is the energy, h is Planck s constant
34
(6.626 10 J s), c is the speed of light, and n is lished.
%
1
the wave number in cm .
Transmission and Diffuse Reflectance
When organic molecules absorb energy from an
external source, e.g., a tungsten or halogen (while
When NIR radiation is passed through a non-
tight) lamp, some molecules will move from the
scattering organic medium, the intensity of absorp-
ground state ðn ź 0Þ to the next energy level ðn ź 1Þ
tion can be described in terms of transmittance, T:
(see Figure 1). The combined vibrational and rota-
tional energy changes involved in the energy level
T ź I=I0
change constitute the absorption process.
Figure 1 illustrates the simplest theoretical treat-
where I is the intensity of the emergent radiation and
ment for the energy absorption behavior of covalent
I0 is the incident energy. For absorption spectra, I can
bonds. Hooke s law curve corresponds to bonds
be expressed in terms of the Beer Lambert relation-
vibrating in simple harmonic motion. In practice,
ship:
covalent bond vibrations are anharmonic and follow
log10ðI0=IÞ Åºlog10ð1=TÞ Åºkcl ź A
the potential energies depicted in the Morse curve,
also illustrated in Figure 1. It is this phenomenon of
anharmonicity that allows the overtones or harmonics where A is the absorbance, k is the molar ab-
to occur in the NIR region. Thus, the frequencies of sorptivity, c is the concentration of the absorbing
the overtones are slightly less than whole number molecular species, and l is the pathlength of energy
Energy
INFRARED SPECTROSCOPY / Near-Infrared 417
through the sample. For a fixed pathlength the is appropriate and repeatable sample presentation.
absorbance is thus directly proportional to the con- Instrument manufacturers provide a plethora of
centration of the absorbing species. sampling devices that attempt to optimize presentation
In the case of samples that produce scatter in criteria for specific sample types.
transmission or diffuse reflectance spectra, a number One approach to establishing a calibration sample
of factors corrupt the linearity of the Beer Lambert set is to scan a large number of samples that repre-
absorbance concentration relationship. Sample scat- sent the potential population to be predicted and to
tering of radiation results in an alteration of the pro- submit the spectra to an analysis that establishes
portion of absorbed and reflected radiation so that which combination of samples shows the largest
pathlength becomes another unknown in the Beer spectral variability overall. This smaller subset of
Lambert relationship. Particle size, particle shape, samples can then be analyzed by the laboratory
crystalline form, bulk, density, and the nature of the method. In practice, it will seldom be possible to
pore space (filled with air, water, or oil) are all establish calibrations that will optimally predict all
variables that dictate the effective pathlength of the future samples. It is therefore helpful to include new
radiation. Sample surfaces also reflect specular samples in the calibration that are spectrally different
energy that has not interacted with molecular struc- from the original calibration set. The standard error
tures. This form of energy has an overall effect on of calibration (SEC) is used as an indication of the
spectra, contributing primarily to the curvilinearity efficacy of a calibration.
of the spectral baseline. It is equally important that a calibration is
A number of attempts have been made to describe tested against a set of samples different from
both theoretically and practically diffuse reflectance those of the calibration set. They should ideally
and scattering functions to enable a linear relation- be separately collected but they must embody the
ship to be established between absorbance (A), ex- chemical and physical features encompassed by
pressed as log10(1/R) where R is the reflectance, and the calibration set. It is the standard error of predic-
molecular concentration. Perhaps the commonest re- tion or performance (SEP), which defines the success
lationship encountered is that ascribed to Kubelka of a calibration. It is important, however, to relate
and Munk, who established nine assumptions and 16 the SEP to the overall range of the analyte in ques-
variables. These can be simplified to the Kubelka tion.
Munk function, namely: The factors listed below are commonly responsible
for increasing the prediction error:
ð1 RNÞ2 K kc
ź ź
2RN S S *
faulty laboratory analysis,
*
sample heterogeneity,
where RN is the reflectance for a sample of ef-
*
lack of repeatability in the packing and presenta-
fectively infinite depth, K is a constant proportional
tion of the sample,
to the kc term in the Beer Lambert relationship, S is
*
instrumental malfunction.
1
a scattering coefficient usually taken as d where d
is the diameter of the particles in a densely packed
The standard deviation for the residuals is due to
scattering medium, k is the molar absorptivity, and c
differences between the laboratory analytical values
is the concentration of the absorbing analyte.
and the NIR predicted values for samples in the cali-
bration set (SEC) or in a test set (SEP).
NIR Calibration Development
Mathematical Data Treatment
The most important first step in the practical appli-
cation of NIR is appropriate sample selection. To Although the theoretical models noted above attempt
take advantage of the potential rapidity and accuracy to satisfy well-established physical principles, in prac-
of analysis that NIR spectroscopy can offer, the NIR tice most NIR workers start the calibration equation
instrument must be calibrated against an acceptable, development process with the orthodox log10 1/R or
accurate, and reproducible laboratory method for the log10 1/T spectra. Although calibrations have been
analyte or analytes of interest. The selection of a generated using the  draw log reciprocal data, it has
sample set for NIR scanning and laboratory analysis been found in most cases that some form or forms of
should be, as far as practically possible, represen- mathematical data treatment prior to regression anal-
tative of the material that the calibration will en- ysis improves a calibration substantially.
counter when it is used to predict the targeted analyte Mathematical approaches that have been em-
or analytes. Another critical aspect of NIR analysis ployed to obviate instrumental noise and other
418 INFRARED SPECTROSCOPY / Near-Infrared
physical effects in favor of chemical features in NIR Standard normal variate (SNV) and detrending For
spectra include: a given set of similar samples the reflectance spectra
will vary in their slope and curvilinearity according
Smoothing functions The simplest smoothing func-
to the particle size distribution between samples,
tion employed is to average the absorbance (log 1/R
differences in crystallinity, and differences in the
or log 1/T) values over a segment of the spectrum.
chemically nonspecific scatter at the sample surfaces.
Many other smoothing algorithms are available to
The SNV function for a 700 wavelength spectrum
smooth noisy spectra and reduce the numbers of
may be expressed as:
wavelengths in spectra. It should be noted, however,
"# 1=2
that a high signal-to-noise ratio is characteristic of n
X
SNVn ź yn y ðyn yÞ2 ðn 1Þ
% %
NIR spectra and smoothing algorithms are primarily
1
used as an initial data reduction step.
n ź 1; y; 700
Derivative functions Derivative functions (strictly
where SNVn ðn ź 1; y; 700Þ are the individual SNV
termed differencing) provide a plot of the rate of
terms for each of the 700 wavelengths, yn are the
change of the slope in the spectrum. The average
700 log 1/R values, and y is the mean of the 700 log 1/R
%
slope is defined over a spectral segment that is fol-
values.
lowed by the omission of several wavelengths
A dominant feature of NIR diffuse reflectance
(a gap). The gap may be zero. The process is repeated
spectra is the increase in absorbance values from
along the spectrum to its final wavelength. This finite
1100 to 2500 nm. This trend is curvilinear for densely
difference technique is usually combined with a
packed samples. A second-degree polynomial func-
smoothing algorithm. The size of the gap and
tion has proved to be an adequate model to linearize
segment may be varied. This technique has the
the spectral baseline.
effect of removing parallel baseline shift due to particle
The list of NIR spectral transformations is not ex-
size differences in the samples, but it retains slope
haustive but it exemplifies the physical features that
differences between spectra, and thus retains the
are perceived to interfere with the chemical absorb-
basic peak height molecular species concentration
ance features and the approaches used to overcome
relationship. Derivatized spectra contain lobes and
the perceived interference in diffuse reflectance NIR
valleys and inflected peaks and are therefore difficult
spectroscopy.
to interpret. Up to four orders of derivative have
None of the above approaches optimizes the rela-
been employed with NIR spectra. Trial and error and
tionship between NIR absorbances and analyte for a
experience with gap and segment sizes provide
range of sample types. Derivative transformations
optimal solutions for calibrations and predictions.
have been found to be generally useful when stepwise
multiple linear regression (SMLR) techniques are
Multiplicative scatter correction (MSC) The model
used. When multidimensional statistics are em-
is based partly upon theoretical considerations of
ployed, e.g., partial least-squares (PLS), principal
radiation scatter and partly upon empirical observa-
component regression (PCR), or neural nets, it has
tions of spectra.
been observed in some cases that the untransformed
If x1; x2; y; xk are the absorbance values of a sam-
log 1/R data can perform just as well in correlation
ple and x1; x2; y; xk are the average absorbance values
% % %
coefficient and error terms as in any kind of trans-
for the set of samples. The regression model used is:
formation. It is considered that in some cases physi-
xk ź a þ bxk þ ek
%
cal manifestations of the sample contained in the
spectra provide valid and useful discriminant data.
The coefficients a and b are estimated by least
squares over the wavelength range and the scatter- Fourier and Wavelet Transforms
corrected spectrum is obtained by subtracting a and
In the past, the large amount of spectral data gene-
dividing by b, i.e.,
rated by NIR instruments challenged the ability of
computers to provide computations within a reason-
xknew źðxkold aÞ=b
able time frame. Mathematical techniques, therefore,
e is a term accounting for  error caused by, for which offered a reduction of the raw data, but with
example, random measurement noise. MSC also minimum loss of information, were often employed.
requires a linearity transform to ensure a linear The decomposition of 1000 spectral data points to
relationship between absorbance values at wave- 100 Fourier transform coefficients provides a great
lengths absorbing for the analyte of interest. saving in computational time with very little loss of
INFRARED SPECTROSCOPY / Near-Infrared 419
spectral information. Large memory computers have It is often suggested that slope and bias corrections
meant, however, that NIR workers can derive func- should be made to computed regression models, i.e.,
tional calibrations or discriminant functions using all an adjustment be made to the regression line so that
the spectral data. With the advent of spectral imaging it can be extrapolated through the zero laboratory/
datasets, and very large sample databases, the need NIR prediction values (bias or intercept). This pro-
has reemerged for efficient data reduction tech- cedure will also adjust the slope term in the equation.
niques. One of the most recent, and promising tech- The most recent view of this technique is that it
niques is wavelet transformation. This technique should be used with utmost caution, in that if such
employs various wave functions that can be com- adjustments are apparently required to improve the
bined to represent a spectrum. Wavelet transforms statistical results then it probably indicates that the
have not been used widely so far, but there is hope original model requires improvement or that labora-
that they may provide a powerful way of data re- tory or NIR measurements are at fault.
duction without loss.
Multidimensional Regression Techniques
Developing the Regression Equation
There are a number of multidimensional, multi-
Until recently, it was most common to use SMLR to variate techniques that are capable of overcoming
develop a calibration model. A condition for the use many of the effects that corrupt calibrations, without
of the technique is that the relationship between ab- necessarily using data pretransformations. These
sorbance expressed as log10 1/R or log10 1/T and the techniques also involve wavelength data compres-
laboratory measurement should be very close to linear. sion, resulting in a decrease of computational time.
The specular reflectance and particle size effects seen Equations are also derived from functions of the
in spectra discussed below also affect the optimal whole spectrum. PLS and PCR are the most com-
reduction of the SEC and SEP. monly used of these techniques. Both technique
When the laboratory value is plotted against the derive linear functions from the wavelength data
NIR predicted value for the calibration sample set it (latent variables in PLS, principal components in PCR).
may well be noted that some points lie well away Typically, a few of the functions in relationship to the
from the computed regression line. This will, of sample number will provide sufficient data to gene-
course, reduce the correlation between laboratory rate coefficients for good calibrations equations.
and NIR data and increase the SEC or SEP. These In PCR, the components are derived by extracting
samples may be outliers. The statistic hi describes the the maximum amount of residual variation from the
 leverage or effect of an individual sample upon a spectral data. Each principle component is ortho-
regression. If a particular value of hi is exceeded this gonal to the next (they have 0 correlation). In this
may be used to determine an outlier sample. Evalu- way the wavelength intercorrelation problem noted
ation criteria for selecting outliers, however, are above is alleviated. PLS functions are derived by in-
somewhat subjective so there is a requirement for terrelating the laboratory value variability and the
expertise in multivariate methods to make outlier NIR spectral variability and latent variables have low
selection effective. intercorrelation. Although a number of writers point
Intercorrelation (or multicollinearity) phenomena to relative advantages for these techniques, in prac-
also reduce the effectiveness of a calibration. First, tice, used properly, they provide similar solutions.
there can be intercorrelation between the constitu- Fourier transform techniques are also used for cali-
ents in the samples. For example, water and fat in bration. Once again the wavelengths are compressed
some meats are negatively correlated and it is pos- into a relatively few (Fourier) coefficients. In com-
sible to obtain the same calibration wavelengths for mon with PLS and PCR, the adjacent Fourier func-
both components although the two components do tions (coefficients) have a low intercorrelation.
not absorb in the same band. The three techniques of multidimensional calibra-
Particle size differences between samples cause all tion noted above also generate functions that are
calibration coefficients to change because of the valuable for the chemical and physical interpretation
baseline shift of the spectra. Furthermore, adjacent of spectra.
wavelengths are also highly correlated, so it is Principle components regression, for instance,
difficult for the SMLR algorithm to  focus on a single generates  loadings for each principal component
combination of wavelengths that will optimize the (PC) that may be plotted against wavelength, to
terms in the calibration. There are also correlations illustrate the importance of specific wavelength
between areas in the spectrum that represent over- regions to the regression analysis. The peaks and
tones of the same covalent bond. troughs of the plot correspond to high positive or
420 INFRARED SPECTROSCOPY / Near-Infrared
low negative loadings on the ordinate axis and are represents a spectral pattern that is generated by
indicative of the degree of influence that the specular (mirror-like, no interaction with molecular
wavelength regions, encompassed by the peaks or bonds) reflection and particle size differences be-
troughs, have on the final model. Since the loadings tween samples. Figure 5 shows a series of leaf tea
are calculated from spectral variation in the whole reflectance spectra where the spectra are very similar
sample set, the loadings against wavelength plot re- in shape but the spectra are spaced up and down the
flects spectral features in the dataset. Figure 2 shows ordinate axis. This is a typical spectral response to
the loadings against wavelength plot for the second samples with particle size differences.
PC derived from a set of orange juice samples. In this Fourier analysis offers the process of self-
plot, the peak at 2100 nm is homologous with a peak deconvolution that is capable of resolving under-
in the orange juice spectra at 2100 nm (see Figure 3), lying peaks from broad overlapping bands. Fourier
which in turn is a characteristic peak in the spectrum self-deconvolution has the advantage over derivati-
of sucrose, and sucrose is a major varying constituent zed spectra that there is no shift or averaging effect
of the orange juice samples from which the PC was and the self-deconvoluted spectrum is directly com-
derived. Figure 3 illustrates an orange juice spectrum. parable with the log/reciprocal wavelength scale.
In Figure 4, the first PC loadings are plotted against Two other techniques of calibration equation com-
wavelength for the orange juice reflectance data putation have recently been used by NIR workers,
referred to above. In this case, the plot shows a weak locally weighted regression (LWR) and artificial
 smoothed orange juice spectrum (cf. Figure 3). This neural nets.
0.80
0.60
0.40
0.20
0.00
-0.20
-0.40
-0.60
1100 1300 1500 1700 1900 2100 2300 2500
Wavelength (nm)
Figure 2 Second principal component loadings versus wavelength plot of a set of 25 components used for canonical variates
analysis of orange juice NIR reflectance spectra.
8.36
0.90
5.42
0.72
2.49
0.54
-0.45
0.37
-3.38
0.19
-6.32
0.01
1100 1300 1500 1700 1900 2100 2300 2500
1100 1300 1500 1700 1900 2100 2300 2500
Wavelength (nm)
Wavelength (nm)
Figure 3 log 1/R spectrum of orange juice dried onto a glass
Figure 4 First principal component loadings versus wavelength
fiber disk.
plot for a set of orange juice reflectance spectra.
Loadings
2
log 1/
R
×
10
Loadings
INFRARED SPECTROSCOPY / Near-Infrared 421
1.05 characterized as authentic. The question asked was
 Can NIR spectra be used to discriminate between
the three categories of juice concentrate? That is,
0.86
country A, country B (authentic), and country B
(adulterated).
0.67
The technique chosen to perform the analysis was
canonical variates analysis. Briefly, the method ab-
0.48
stracts functions from combinations of PCs. It has
been found with large sample sets (D100 or more)
0.29
that up to 25 PCs give optimal predictive ability. It is
unwieldy to select the predictively important analytes
0.10 from 25 PC dimensions. For this reason the
1100 1300 1500 1700 1900 2100 2300 2500
PC-derived canonical varieties are computed wherein
Wavelength (nm)
the dimensions available are one less than the number
Figure 5 log 1/R spectra of tea leaf samples showing the
of ascribed characteristics or groups. In the example
 stacking of similarity shaped spectra along the ordinate axis.
noted above, therefore, the two CVs describe the 25
PC dimensions, since there are three groups in the
LWR has been recommended to counteract the analysis. In this example, a calibration discriminant
common inherent nonlinearity existing between labora- analysis was calculated using 71 samples, and 23
tory values and the spectral absorbances. Essentially, prediction samples were used to test the calibration.
segments of the NIR versus laboratory value regre- Figure 6 illustrates the two-dimensional CVs cali-
ssion computed are used to give a much better ap- bration and prediction.
proximation to linearity than a regression line model One problem that is evident from this analysis is
for the whole dataset. that samples that lie outside the defined circles (95%
of the samples sample points should lie within the
Neural Networks group circles providing each group has the same vari-
ance), may (1) belong to one of the defined groups or
Neural networks have been used, most effectively, as
(2) belong to a population not defined in the cali-
a way of establishing robust NIR calibrations. It is
bration. At present no satisfactory method exists to
the calibration method of choice for a whole cereal
make an objective decision about such  outlier sam-
grain, transmission instrument, which is being used
ples. One approach to help with this problem is to
worldwide for many of the analyses required for esti-
employ cluster analysis techniques. In this case sam-
mating grain quality. An artificial neural network
ple spectra are allotted to specific groups in a hier-
program performs many iterations in order to estab-
archical fashion, beginning with a division of the
lish the optimum solution required from large, com-
sample set into two groups, then four, and so on.
plex datasets.
This type of analysis requires decisions to be made
about appropriate distance and similarity functions,
Discriminant Analysis
which are required to perform the computations.
NIR instruments may not only be calibrated for There are many such functions and only experience
quantitative analysis, but they may also be  trained of their use leads to valid analysis and samples that
for qualitative purposes. This process is usually may be truly outside the group populations will at
termed discriminant analysis. The criteria noted some level of the analysis be included in a group.
previously for establishing quantitative calibrations
with minimum prediction error are equally applica-
A Nonregression-Based Technique
ble to discriminant calibration sets.
Spectral matching algorithms are also employed Comparison analysis using restructured near-infrared
using multidimensional mathematics. Such tech- and constituent data (CARNAC) can be used to
niques are often used in quality assurance programs provide quantitative or discriminant analysis for
where a spectrum of a known material is compared large databases. In the current version of the tech-
with incoming raw materials to check its purity or nique, CARNAC D, wavelet transforms are used to
authenticity. For example, samples of orange juice reduce the spectral information, and then the spec-
concentrate were collected to represent the types trum of the sample requiring a particular analyte
available from two countries. In addition, samples measurement or discriminant comparison (trans-
that were known to be adulterated were obtained formed in the same way) is compared with the  cali-
from one of the countries. All other samples were bration database. The sample giving the closest
log 1/
R
422 INFRARED SPECTROSCOPY / Near-Infrared
8
6
Country B
(authentic)
4
2
0
-2
Country A
-4
Country B
(adulterated)
-6
-8 -6 -4 -2
0 246
First canonical variate
Figure 6 Canonical variates calibration and prediction sample plot of NIR diffuse reflectance spectra characterized as: m, D, country A;
K, Jcountry B (authentic); and  , &, country B (adulterated). Solid symbols, calibration samples; hollow symbols, prediction samples.
match gives the solution to the required analyte or each sample spectrum obviating the inclusion of
discriminant identification. purely instrumental contributions to the sample
spectrum. Single-beam instruments, using a detector
array set to 451 to the sample surface, must have a
Types of NIR Instrumentation
reference material to scan, the spectrum of which is
Until the late 1980s there were only two basic types of
subtracted from the sample scan. The full-wave-
NIR instrument, those employing discrete wavelength
length scanning instruments are more expensive than
filters and wavelength scanning instruments using
filter instruments and are in general used for cali-
holographic gratings. Both types employ high-intensity
bration development and applied research.
white light sources and one or two types of detector
In the late 1980s and into the new millennium in-
depending upon the wavelength range being used.
strumental approaches to the NIR region have in-
For the NIR region (1100 2500 nm) a lead sulfide
creased substantially and some of these new
detector provides appropriate sensitivity and in the
approaches are noted briefly below.
higher-energy visible/far visible region from 400 to
1100 nm less sensitive silicon detectors are used.
Calibration Transfer
Detector geometry is important to optimize the
collection of diffuse reflectance energy in favor of NIR instruments of all kinds differ in there optical
specular reflectance energy that contains no chemical and operational properties. Instruments of the
absorbance information. Two approaches are same kind and make have small differences, but
commonly used. One uses an integrating sphere that these produce differences in spectra that are sufficient
provides a double-beamlike instrument, whereby the to erode the accuracy of a calibration if it is trans-
arrangement of the detectors within the reflective formed from one instrument to another. There has
sphere enables a reference and a sample beam to be been, and there continues to be, an interest in ways of
employed. The reference spectrum is subtracted from transforming calibrations, such that, when they are
Second canonical variate
INFRARED SPECTROSCOPY / Near-Infrared 423
transferred from one instrument to another they re- has been their application to at-, on-, or in-line anal-
tain their predictive accuracy in the new instrument. ysis for factory processes. Once a calibration has
been developed, the NIR technique does not require
Transmission through Solids
chemicals and is capable of rapid, multicomponent,
noninvasive analysis. Two features of NIR instru-
Using the higher-energy visible region, instruments
mental development are most significant in this con-
have been made to provide transmission spectra of
text. The first is the increasing use of optical fibers,
solid samples that have been satisfactorily calibrated
which enables the instrument to be multiplexed and/
for kernel hardness of gain, meat components, and
or remote from the line. Optical fibers are also used
tenderness of peas, for example.
within instruments to optimize the light path geometry.
The second feature is the development of nonmoving
Photodiode Array Instruments
part or solid-state spectrometers that are light,
Photodiode array (PDA) type of instrument is solid
rugged, long-lasting, small, and up to 10 times less
state (no moving parts) and because of the electronic
expensive than their equivalents with moving parts.
switching of the diodes providing the NIR illumina-
As has been noted, this type of instrument can also be
tion, it is very rapid scanning, compared with other
developed to be handheld.
instrument types. Such a construction lends itself to
miniaturization, and the result is physically robust,
enabling it to be used in many situations where other
Industrial Applications of NIR
instrument types could not be used, including situ-
Technology
ations where a hand-held analyzer is required. At
Applications in Agriculture and the Animal Feed
present the wavelength range is limited between 500
Industries
and 1100 nm (other instruments can provide wave-
lengths up to 2400 nm). The spectra from a PDA
In the agriculture and animal feed industries NIR
instrument also tend to be noisier than spectra from
instruments, and in particular discrete filter instru-
other instruments. Currently, this type of NIR ana-
ments, have found the widest application in the
lyzer is being used as an integral measurement system
analysis of grain and flour. Protein, moisture, hard-
in a cereal combine harvester. It is also being used
ness, and baking quality are all examples of analysis
experimentally to estimate quality parameters of
that have been performed by NIR instrumentation,
whole fruit on the tree. The uses for this NIR tech-
both off- and online. Several countries, including the
nology are increasing rapidly.
USA and Canada, have adopted NIR protein analysis
as national standards to gauge the level of payment
NIR Image Analysis
for wheat. The protein method was approved by the
AACC (American Association of Cereal Chemists) in
Developments are currently taking place in using the
1983.
visible and NIR region for video imaging purposes.
Experiments have been carried out to image meat sur- In the feed and forage industries NIR methods
have been used extensively. NIR factory-based quality
faces and the textural appearance of bread crumbs.
A leading pharmaceutical manufacturer has devel- control schemes have been implemented which
enhance the economy with which feed components
oped an NIR-based image analysis technique to
are used, and the quality and consistency of the final
image the ingredients of capsules and tablets to check
product. Moisture, crude protein, and acid detergent
there even distribution in a formulated product.
fiber analyses of feeds are being steered through the
Diffuse reflected NIR radiation from the surface of
AOAC (Association of Official Analytical Chemists)
the sample is collected by a series of imaging optics
and passes through an NIR tunable filter prior to for- validation process. NIR is also used widely in the US
tobacco industry for measuring nicotine and other
ming an image on an IR focal plane array detector. The
process parameters both on- and offline.
filter is continuously stepped through a predetermined
spectral interval and an image at each wavelength is
Applications in the Food and Beverage Industry
stored. The resulting data hypercube is analyzed by
proprietary principal component-based software.
It is perhaps the food and beverage industry that
presents the most difficult challenge to both off- and
online NIR analysis. The materials are organically
Industrial Uses of NIR Instrumentation
complex, and often need to be analyzed in the pres-
NIR spectroscopy is an application-driven technique ence of a high percentage of moisture, the broad,
and perhaps one of the greatest motive influences in highly absorbing bands of which, in the NIR region,
the development of the many kinds of NIR instrument tend to obscure the lesser absorbing constituents.
Table 1 Matrix of food products and components and characteristics analyzed offline and online by NIR
Food product
Barley - - × - - - - - - - - - - × × - - - × - - - - - × - - - - - - - - × - - - - - - - - - - -
Beer - × - - - - - - - - - - - - - - - - - - - - - - × - - - ż - - - - - - - - - - - - - - - -
Beet sugar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ż - ż - - -
Biscuit - - - - - - - - - - - - × - - - - - - - - - - - ż - - - - - - - - - - - - - - × - - - - -
Bread - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - × - - - - - - -
Breakfast cereal - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Cake mixes - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - - - × - - - - -
Cheese - - - - - - - - - - - - ż - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Chocolate - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - - - × - - - - -
Cocoa - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - × - - - - - - -
Coffee - - - - × - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - × -
Dough - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - - - × - - - - -
Edible oils - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - -
Fish - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Flour - - - × - - - × - × - - - - - ż - - - - - - - - × - - - - - - - - ż - - - - × - - - - - ×
Hops × - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - -
Ice cream - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Liquors - × - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - × -
Maize - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - -
Malt - - - - - - - - - - - - - - × - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Margarine - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - -
Meat - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × × - × - - - - - - - -
Milk products - - - × - - × - - - - - ż - - - - × - - - - × - ż - - - - × - - - × × - - - - - - - - × -
Nondairy creamer - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Oats - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Oilseed - - - - - - - - × - - - × - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Packaging laminate - - - - - - - - - - - - - - - - - - - - - - - - - - - × - - × × - - - - - - - - - - - - -
Pasta - - - - - - - - - - - × × - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Peas - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - × - - - - - - - - × - -
Peppers - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Potato - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - -
Pulses - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - × - - - - - - -
Rice - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - × - - - -
Salad cream - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Sausages - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - × - - - - - - - - - - -
Spices - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - × - - - - - - - - - - - -
Tea - - - - × - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - -
Tobacco - - - - - - - - - - - - - - - - - - - - - - - - ż ż - - - - - - - - - - - - - - - - - - -
Toffee - - - - - - - - - - - - × - - - - - - - - - - - × - - - - - - - - - - - - - - × - - - - -
Tomatoes - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - - - - - -
Triticale (rye) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - -
Wheat - - × - - - - - - - - - - × - - × - - - × - - - × - - - - - - - - × - × - - - - - - - - -
Wine × × - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - × - - - - -
Yogurt - - - - - - - - - - - - × - - - - - - - - - - - - - - - - - - - - × - - - - - - - - - - -
-, data not known; ×, offline analysis; ż, online analysis.
r

,
e
e

s
n
-Glucan
Acids,
Alcohol
Amino acids
Ash
Caffeine
Capsaicin
Casein
Cellulose
Chlorophyll
Color
Dietary fiber
Egg
Fat/oil
Fungal spores

Gluten
Hardness
Heat treatment
Hot water extract
Howard mold count
Insect
Iodine value
Lactose
Methionine
Moistur
Nicotin
Nitrogen (total)
Nylon
Original gravity
Pasteurization
Polypropylene
Polythene
Proportions in a mixture
Protei
Salt
Sedimentation volume
Soya bean flour
Starch
Starch damage
Sucrose
Sulfur
Sugars
Tenderomete
Total solids
Water absorption
INFRARED SPECTROSCOPY / Near-Infrared 425
Table 2 Overall quality parameters measured by NIR in food
Pharmaceuticals and Medical Applications
and animal feed
Discriminant analysis of NIR spectra was used in
Product Quality parameter
1986 to assay the level of lincomycin in a pharma-
Cheese Ripeness ceutical formulation. This was the first NIR analysis
Forage Digestibility
to be accepted by this US Food and Drug Adminis-
Fruit, vegetables Maturity
tration (FDA). NIR is now used to measure the sali-
Fruit juices Authenticity
cylic acid content of aspirin and pharmaceutical
Meat Heat treatment
companies use discriminant NIR procedures to check
Milk Pasteurization
Peas Tenderometer readings incoming raw materials for drug production. Sample
Seaweed (edible) Quality
identification can be achieved using at-line NIR fiber
Spices Gamma irradiation
optic systems.
Tea Taster-defined quality
Antibiotic production in fermentations can now be
Wheat Seed viability
monitored by NIR. A highly significant development
in the medical field is the use of a handheld non-
In spite of the challenging nature of NIR analysis
invasive NIR-based filter instrument for the moni-
of food, applications of the technique are most com- toring of blood glucose levels for diabetics. Research
mon and diverse in this industry. Table 1 lists NIR
is currently being undertaken to explore the feasibil-
food applications both off- and online. Table 2 lists
ity of measuring human body temperature using
those applications where NIR has been used to anal- NIR. Another potential medical application being
yze for broad, relatively indefinable quality charac- examined is the measurement of oxygen levels in
teristics in food.
brain blood using a fiber optic probe placed on the
skull. In the medical laboratory context, user-friendly
Whole Fruit and Vegetable Analysis
fecal fat analysis is performed using NIR.
Following on from the success with whole cereal
Other Industrial Applications of NIR
grain analysis using NIR transmission technology,
many workers have been using a variety of instru- An NIR method has been developed for measuring the
hardwood content of bleached hardwood and the
ment types to obtain spectra from whole fruit and
lignin content of unbleached hardwood pulp. Work
vegetables, nondestructively. Some success has been
has also been carried out using NIR reflectance spectra
gained, for example, in measuring dry matter and
of hardwoods to discriminate between the different
sugar concentration in this way.
species. Using a Fourier transform NIR instrument
Chemicals and Textiles
carbonate measurements have been made on soil sam-
ples. NIR analysis of forest humus samples has
Quantitative analysis by NIR of petrochemicals dates
provided satisfactory calibrations for microbial basal
back to the 1930s. Since then a range of NIR cali-
respiration, based on the organic polymer content of
brations have been developed, for example, for octane
the humans. NIR analysis of the lake sediments for pH
number, methyl group analysis, and methanol content
has been used to construct the lake water history.
in petroleum. NIR has been used to monitor water,
The examples noted above are only a small frac-
detergent solids, and glycerol in shampoo, and to
tion of the industrial applications of NIR, and the
analyze moisture and lubricant levels on polymer
advent of solid-state instrumentation, will further
films. Process NIR spectrometers have been used to
widen the way in which NIR measurements can be
monitor naphtha composition and NIR instrumenta-
made in an industrial context.
tion has been used to monitor ethylene polymerization.
See also: Chemometrics and Statistics: Statistical
For over a decade NIR quality control has been
Techniques; Multivariate Classification Techniques; Multi-
used in the textile and fiber industries. Perhaps the
variate Calibration Techniques. Food and Nutritional
widest use of NIR has been in the cotton industry.
Analysis: Overview. Fourier Transform Techniques.
Cotton blending, mercerization, and fiber maturity
Fuels: Oil-Based. Infrared Spectroscopy: Overview.
measurement have been used offline for rapid proc-
Pharmaceutical Analysis: Drug Purity Determination.
ess control. The wool industry has also employed
Process Analysis: Overview. Proteins: Foods. Quality
NIR to measure the residual grease after scouring.
Assurance: Quality Control. Textiles: Natural; Synthetic.
Calibrations have been developed for measuring
moisture and heat set temperature in nylon yarn.
Further Reading
Online NIR analysis is being developed for the quality
control of the dyeing procedure for carpet yam and
Burns DA and Ciurczak EW (eds.) (1992) Handbook of
for measuring yarn diameter. Near Infrared Analysis. New York: Dekker.
426 INFRARED SPECTROSCOPY / Photothermal
Davies AMC (2002) Is the end of the NIR  calibration spectroscopic data: Part 1. Data compression. Journal of
problem in sight? NIR NEWS 13(5): 10 12. Near Spectroscopy 11: 3 15.
Davies AMC and Cho RK (2002) Near infrared spec- Osborne BG, Fearn T, and Hindle PH (1993) Practical NIR
troscopy. Proceedings of the 10th International Spectroscopy with Applications in Food and Beverages
Conference, Kyonjgu, Korea. Chichester: NIR Analysis. Harlow: Longman Scientific & Technical.
Publications. Williams P and Norris K (eds.) (1987) Near-Infrared Tech-
Fearn T and Davies AMC (2003) A comparison of Fourier nology in the Agricultural and Food Industries, pp. 1 15.
and wavelet transforms in the processing of near infrared St Paul, MN: American Association of Cereal Chemists, Inc.
Photothermal
S E Bialkowski, Utah State University, Logan, UT, USA
dependence (Figure 1B) with a penetration distance,
ml (m), given by the time-dependent width of the
& 2005, Elsevier Ltd. All Rights Reserved.
Gaussian profile:
This article is a revision of the previous-edition article by
J F Power, pp. 2219 2225, & 1995, Elsevier Ltd.
ml źð2DtÞ1=2 ½2Š
Introduction
Photothermal spectroscopy is a class of optical anal- Åš = 0
ysis methods that measures heat evolved as a conse-
quence of light absorption in an irradiated sample. In
conventional spectrometric methods information is
obtained by measuring the intensities of light trans-
Åš = /2
mitted, reflected, or emitted by the sample. In photo-
thermal spectrometry, spectroscopic information is
obtained by measuring the heat accompanying non-
radiative relaxation. Because of the universality of the
photothermal effect (e.g., heat evolution accompanies
Åš =
essentially all optical absorption), photothermal spec-
Depth, Z (µm)
0 256
troscopy has diverse applications in chemistry, physics,
(A)
biology, and engineering. Some applications and mea-
surements in the analysis of solids are reviewed here.
t = 0.005 s
Theory
When a sample surface is heated with a periodically
modulated beam, a thermal wave is generated which
t = 0.01 s
propagates away from the heated region. The ther-
mal wave is a critically-damped temperature oscilla-
t = 0.05 s
tion that, in a homogeneous material, decays
exponentially with distance from the heated surface
(Figure 1A). The thermal wave damping distance is
the distance at which the temperature attenuates to
Depth, Z (µm)
0 256
1/e of the value observed at the surface. This dam-
(B)
ping distance is controlled by varying the modulation
Figure 1 (A) Spatial dependence of a single-frequency (har-
frequency, o (rad/s), of the radiation source. For a
monically driven) thermal wave. Frequency 20.0 Hz, diffusivity
sample of thermal diffusivity D (m2/s), the thermal
3 1
1 10 m2 s . (B) Spatial dependence of the thermal wave
wave damping distance, m (m), is given by
observed at various times after application of a heat pulse.
3 1
Diffusivity 1 10 m2 s . The instantaneous phase f ź ot.
m źð2D=oÞ1=2 ½1Š
(Reprinted from Power JF (1993) Scanning probes III: Photo-
acoustic and photothermal imaging. In: Morris MD (ed.) Micro-
With impulse heating, the time-dependent tempera-
scopic and Spectroscopic Imaging of the Chemical State,
ture profile below the heated surface has a Gaussian pp. 255 302. New York: Dekker, courtesy of Marcel Dekker, Inc.)