Focal Point Review

Applied Spectroscopy
2017, Vol. 71(7) 1403–1426
! The Author(s) 2017
A Review of the Principles and Applications Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
of Near-Infrared Spectroscopy to DOI: 10.1177/0003702817709299
journals.sagepub.com/home/asp
Characterize Meat, Fat, and Meat Products

Nuria Prieto1, Olga Pawluczyk2, Michael Edward Russell Dugan1,

and Jennifer Lynn Aalhus1

Abstract
Consumer demand for quality and healthfulness has led to a higher need for quality assurance in meat production.
This requirement has increased interest in near-infrared (NIR) spectroscopy due to the ability for rapid, environmentally
friendly, and noninvasive prediction of meat quality or authentication of added-value meat products. This review includes
the principles of NIR spectroscopy, pre-processing methods, and multivariate analyses used for quantitative and qualitative
purposes in the meat sector. Recent advances in portable NIR spectrometers that enable new online applications in the
meat industry are shown and their performance evaluated. Discrepancies between published studies and potential sources
of variability are discussed, and further research is encouraged to face the challenges of using NIRS technology in com-
mercial applications, so that its full potential can be achieved.

Keywords
Near-infrared spectroscopy, portable instruments, online, meat, fat, quality
Date received: 28 February 2017; accepted: 19 April 2017

Introduction meat, NIR spectroscopy was not used for on-/inline

When consumers purchase meat, price and quality are purposes until 1996, when Isaksson et al.5 used a conveyor
important to their decisions; however, an increasing NIR instrument located at the outlet of the meat grinder
number of consumers also factor in health implications of to predict the fat, moisture, and protein content of
meat consumption when making a purchase.1,2 This ground beef.
demand for quality and healthfulness has led to a higher Near-infrared spectroscopy measures the absorption of
need for quality assurance in meat production. To meet electromagnetic radiation including wavelengths from 750
this demand, new objective quality control methods are to 2500 nm. The NIR spectra include broad bands that arise
required as traditional means of analysis are lengthy, require from absorptions in overlapping wavelengths. The absorp-
toxic solvents and reagents, and can have high costs. For tions measured by NIR spectroscopy correspond mostly to
these reasons, the development of rapid, environmentally overtones and combinations of vibrational modes involving
friendly, and noninvasive methods for either prediction of C–H, O–H, and N–H chemical bonds.6 Recording the
meat quality or authentication of added-value meat prod- electromagnetic radiation absorbed from those molecular
ucts has become a priority in the last years. In this regard, bonds in the NIR wavelengths produces spectra which are
near-infrared (NIR) spectroscopy could be considered as a unique to a sample acting as a ‘‘fingerprint’’. The collected
rapid and cost-effective alternative. spectrum includes data related to the chemical and physical
Although NIR spectroscopy is referred to as a new tech-
nology, it was discovered in 1800 when Herschel found that 1
Lacombe Research and Development Centre, Agriculture and Agri-Food
dispersion of electromagnetic waves beyond the visible Canada, AB, Canada
2
range of the spectrum could be observed by using a P&P Optica Inc, Kitchener, ON, Canada
series of thermometers with blackened bulbs.3 However,
Corresponding author:
it was not until the 1960s when major developments Nuria Prieto, Lacombe Research and Development Centre, 6000 C&E
occurred with NIR spectroscopy technology, allowing its Trail, Lacombe, Alberta, T4L 1W1, Canada.
application in various aspects of animal production.4 For Email: nuria.prietobenavides@agr.gc.ca
1404 Applied Spectroscopy 71(7)

Figure 1. Near-infrared spectra collected with a portable instrument on intact meat samples from several species.
Source: Agriculture and Agri-Food Canada–Lacombe.

properties of organic molecules in the sample and, there- and long start-up times. In contrast, light emitting diodes
fore, important information on sample composition. (LED) have recently emerged in the NIR field as they
Examples of NIR spectra collected on intact meat from address several of the problems with QTH light sources.
several species using a portable instrument are presented Due to the higher efficiency of LED illumination systems,
in Figure 1. power requirements are lower for the same brightness.
Recently, efforts have focused on applying this technol- Furthermore, excess heat production is not an issue with
ogy to meat analyses. Several comprehensive reviews7–9 LED illumination sources. However, due to the costliness of
have summarized studies reporting the capabilities of NIR LED, they have not been widely adopted in NIR spectrom-
spectroscopy in the area of meat quality. However, this eters.10 Beam splitter systems in NIR spectrometers output
technology still has some limitations and online applications single-color light translated from multi-color input. These
under industrial environments remain challenging. systems use light filters in discrete-wavelength spectropho-
This review provides a summary of recent uses of NIR tometers, or interferometer and gratings in continuous
spectroscopy in the meat sector, emphasizing papers pub- spectrum NIR instruments.
lished after 2010. Discrepancies between published studies When spectra are collected, NIR radiation interacts
and potential sources of variability are discussed, and fur- with the sample and the energy may be absorbed,
ther research is encouraged to face the challenges of using transmitted, or reflected. As such, various modes of meas-
this technology in commercial field applications, so that its urements can be applied in NIR spectroscopy designed to
full potential can be achieved. suit different uses. These measurement modes include:
transmittance, interactance, transflectance, diffuse trans-
Fundamentals of Near-Infrared mittance, and diffuse reflectance, with the latter two meth-
ods being applied most often.11 The choice of measurement
Spectroscopy is dependent on sample characteristics; for example, the
The NIR spectrometers include a light source, beam split- phase (i.e., solid or liquid), translucency, and by the size
ter system (wavelength selector), sample detector, optical of the particles.
detector, and data processing/analyzing system. These parts Sample detectors within the NIR spectrometer differ by
can have different properties and should be selected based spectral response, speed of response, and the minimum
on their intended use to provide an effective and consistent threshold for detectable radiant power. Most commonly,
instrument. The most commonly used NIR radiation NIR spectrometers utilize single and multi-channel photon
sources are quartz–tungsten–halogen (QTH) lamps; this is detectors. Single-channel detectors use lead salt semi-
due to the low cost and high intensity radiation in the NIR conductors such as lead sulfide (PbS, 1100–2500 nm),
wavelengths where the spectral output is continuous. indium gallium arsenide (InGaAs, 800–1700 nm or
However, QTH lamps present several problems in indus- extended range up to 2500 nm), and silicon detectors
trial application due to their low energy efficiency, heat (400–1100 nm). Multi-channel detectors utilize diode
generation, temperature sensitivity, vibration sensitivity, arrays or charge-coupled devices (CCDs).12 The use of
Prieto et al. 1405

Figure 2. Real-time collection of NIR spectra at the rib-eye from beef using a portable instrument. Source: Agriculture and Agri-Food
Canada–Lacombe.

Figure 3. Real-time collection of NIR spectra from the inner layer of pig subcutaneous fat using a portable instrument.70

the different sample detectors varies according to sample fiber optic cable in portable instruments is now being used
properties: liquid samples are often analyzed in glass or for online application to evaluate the quality of meat, fat,
quartz chambers of different sizes, while diffuse reflection and meat products (Figures 2 and 3). Several of these port-
carrier accessories are generally used for solid samples. able NIR spectrometers are now available for purchase and
Many sample detectors or chambers are constrained for vary in cost and designated application. See dos Santos
use in laboratory settings; however, recently developed et al.8 for an excellent review of the different specifications
portable handheld devices are more compact, have simpli- for portable NIR instruments most often used in scientific
fied use, designs that are more robust, and lower cost. research and potential applications in the agri-food sector.
These developments enable use of NIR spectroscopy tech- In the last few years, NIR spectrometers using micro-
nology in a greater variety of applications. In this regard, the electro-mechanical systems (MEMS) technology have been
1406 Applied Spectroscopy 71(7)

information about chemical properties from samples; this

processing is referred to as chemometrics. Chemometric
methods have been described in detail in previous
literature16 and it is not the aim of this review to cover
chemometric details; however, a brief summary is provided
here. Smoothing is used to reduce instrumental noise or
background information and de-trending techniques are
usually performed to reduce the effects of accumulating
data sets from a trend. Derivatives are commonly used to
reduce insignificant baseline signals from samples.17 Pre-
processing methods that involve normalizations, such as
multiple scatter correction (MSC) and standard normal
variate (SNV), have been used to correct path length
Figure 4. Fully packaged micro-electro-mechanical systems effects, scattering effects, source or detector variations,
(MEMS)-based NIR spectrometer, approximate size is 6 in  4 and some instrumental sensitivity effects.18 More recently,
in  1 in (15.3 cm  10.2 cm  2.5 cm).14 orthogonal signal correction (OSC), direct orthogonal
signal correction (DOSC), and orthogonal wavelet correc-
tion (OWAVEC) have been developed to reduce light-
developed. The combination of MEMS with mathematic scattering effects, among other types of interference.19
transformations, such as digital transform spectroscopy Net analyte signal (NAS) is comparable to OSC insofar as
(DTS), produces powerful spectrometers with several both algorithms are orthogonal to concentration matrix
clear advantages. The MEMS chips equipped with fixed dif- with spectral matrix. However, NAS has attracted more
fraction grating are rapidly programmable and the use of a attention in chemometrics, as it can also be used in add-
high-contrast pixilated optical reflector working as a tun- itional multivariate calibrations, such as sensitivity, selectiv-
able spectral filter, combined with a single detector, can ity, signal-to-noise ratio, and limit of detection for the
reduce equipment costs while eliminating detector noise. spectral figures.20 Wavelet transforms (WT) has become
The MEMS-based NIR spectrometers are also particularly a popular pre-processing method today due to the rapid
useful in in situ applications due to the reduction in moving compression of data sets in order to remove the noise and
parts and robustness in design.13 These design advantages background signals.21
and the high-resolution collection allow the use of MEMS- Following mathematical pre-processing, a prediction
based NIR spectrometers in environments with high levels model must be built, referred to as the calibration. The
of vibration and variable temperatures where other designs calibration is a regression model allowing prediction of
might not be suitable. Therefore, MEMS technology repre- chemical properties based on spectral data. Usually, NIR
sents a paradigm shift for industrial applications of NIR spectra do not have specific wavelengths that directly cor-
spectroscopy, introducing new opportunities for spectro- relate with distinct sample properties; this is due to over-
scopic sensors (Figure 4).14 For instance, Pügner et al.15 lapping bands in the spectra resulting from various chemical
have recently developed a novel hybrid-integrated MEMS and physical attributes of complex samples. As such, using a
scanning grating spectrometer to estimate chemical com- least squares regression (LSR) based univariate calibration
ponents in food products with low power consumption (a is ineffective, as a direct relationship of the property to be
few milliwatts) and a volume of only 2.1 cm3, which could predicted and the measured signal is required. Therefore, a
potentially be implemented in smartphones. multivariate calibration is needed; this is accomplished by
Advances in computing power and statistical programs using the data from all or some wavelengths as spectral
have allowed NIR spectroscopy to output data almost in predictor variables. Numerous multivariate approaches
real time. However, the raw spectra collected from NIR have been used for building reliable calibration models for
spectrometers contain background data and noise besides quantitative analysis in food. Early calibration methods from
the sample data. The absorption signals are often relatively the 1960s and 1970s were conducted using multiple linear
weak compared with interference in the interaction regression (SMLR),12 using a discrete number of wave-
between light and particles. Interferences that introduce lengths from filter instruments. However, spectral measure-
noise in collected spectra can include: water absorption ments have high linear relations between variables, resulting
bands, scattering effects, instrumental noise, sample com- in an issue called multicollinearity. With the emergence of
plexity, extraneous light sources, and matrix/environmental computer science in the 1980s and the introduction of
effects. Additionally, distinct chemical molecules can pre- monochromators to record full range spectra, more
sent absorption peaks that overlap in several regions sophisticated approaches to solve the multicollinearity
along spectra. As such, mathematical pre-processing of problem were developed. These consist in developing
spectral data is required in order to obtain valuable orthogonal (uncorrelated) linear combinations from
Prieto et al. 1407

spectral variables (components or factors) instead of using of the calibration models may not always be compatible
the original data to establish regression equations. between instruments; therefore, this possibility is suitable
Algorithms such as principal component regression (PCR) only when the same spectral data sets can be used in
or partial least squares regression (PLSR) provide more different environments and instruments. Methods for
stable and reliable regression equations and predictions,12 transferring calibration models to make the spectra less
since the variables are compressed through removal of dependent of instrumentation variation can be found in
unrelated and unstable data, such as noise and redundancy, the review of Cen and He.10 Zamora-Rojas et al.22
while retaining most of the meaningful information. demonstrated the successful transfer of calibration
Principal component regression is conducted by using prin- models, where large calibration data sets collected pre-
cipal component analysis (PCA) of the spectral variables for viously were used with new NIR spectroscopy devices
data compression, and subsequently LSR between selected better suited to in situ analysis. Nevertheless, despite
principal components (PC) and the reference values. Partial possibilities for several methods of calibration transfer,
least squares regression differs from PCR by using the there is always a loss of precision when compared to
information of both spectroscopic and parameter variables the original models.
analyzed in samples. Nevertheless, increases in calibration Calibration models are usually assessed for their
database size have highlighted the need for analyzing predictive ability using coefficient of determination (R2),
nonlinear relationships using calibration methods. To over- root mean square error of cross-validation (RMSECV), or
come intrinsic nonlinearities within data sets, nonlinear prediction (RMSEP) depending on the type of validation
regression approaches such as local regression methods undertaken,23 and ratio of performance deviation
(e.g., locally weighted regression [LWR]) or artificial (RPD).24,25 The equations for root mean square error
neural networks (ANN) have been recently applied. (RMSE) and RPD are defined in Eqs. 1 and 2:
Qualitative analyses can also be applied to NIR spectral
data. Qualitative methods are referred to as pattern recog- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
nition methods in which individual samples categorized 1X n
_
RMSE ¼ ð y  yi Þ 2 ð1Þ
prior to analysis are classified based on collected spectra. n i¼1 i
Several methods can be used for qualitative analyses includ-
ing: ANN, cluster analysis (CA), discriminant partial least
SD
square (DPLS), K-nearest neighbors (KNN), linear discrim- RPD ¼ ð2Þ
inant analysis (LDA), PCA, soft independent modeling of RMSE
class anthology (SIMCA), and support vector machine where n is the number of samples in the calibration set, yi
_
(SVM). A comprehensive review of qualitative methods as represents the measured responses, yi is the estimated
well as their applications can be found in Cen and He.10 responses obtained through cross-validation or external
Once the calibration has been performed, a validation validation, and SD corresponds to the standard deviation
process is required to ensure successful calibration models. of reference values from calibration or prediction set.
The validation method commonly used is the internal or Since its first introduction in a peer-reviewed publication
cross-validation, where the same data set used for calibra- in 1993,26 RPD is gradually becoming widely used for quick
tion is used in the validation. However, external validations assessment of an NIR spectroscopy calibration model due
using a separate and independent sample set (test set) pro- to its non-dimensionality. The calculated RMSE should be
vide more reliable and relevant estimates of the future pre- considerably less than the SD and ideally the RPD should be
diction ability of the model.9 Hence, it becomes very 5 or higher.27 Low RPD values may in part be due to a low
important to select a representative set of samples provid- variability in reference values or a large RMSE relative to
ing the largest information for the calibration data set, since SD.28 Values of over 20 have been observed for the predic-
it is of critical importance that this data set represents as tion of some factors, such as moisture content of whole-
much variation as possible that will be encountered in kernel corn (maize). However, due to complications caused
future samples. Despite the external validation being the by sample preparation, sample presentation, difficulties with
most desirable option, its application may not be always reference testing, and characteristically low variability in a
possible; especially in research studies where biological sample set, RPD values of  3.0 may be difficult to obtain. In
samples are involved and the number of animals available these circumstances the NIR spectroscopy may still have
per experiment is limited. value for use as an analytical tool in both scientific and
Developing an adequate calibration data set can be dif- industrial contexts. Table 127 provides RPD ranges which
ficult and expensive, as such calibrations are thought to be a correspond to appropriate application of NIR spectroscopy
major challenge with NIR spectroscopy. Hence, sharing of for samples which have more complex physical structures,
model libraries and transferring multivariate calibration and in prediction of more functional attributes such as meat
models is important in order to provide large databases texture, where no ‘‘classical’’ absorbers are directly asso-
with, likely, a higher variability. Nevertheless, the transfer ciated with them.
1408 Applied Spectroscopy 71(7)

Table 1. Degrees of merit for the ratio of performance in a single sample; increasing the scans per sample or using a
deviation (RPD) to the application of NIR spectroscopy.27 larger surface area for each scan could also help to improve
precision. These results agree with Prieto et al.31 and
RPD value Classification Application
Balage et al.,32 who reported low NIR spectroscopy pre-
0.0–1.9 Very poor Not recommended dictability for intramuscular fat content in beef and pork,
2.0–2.4 Poor Rough screening respectively, when spectra were collected on intact muscle
2.5–2.9 Fair Screening using fiber optic devices from 350–1800 nm and
3.0–3.4 Good Quality control 400–1395 nm, respectively.
3.5–4.0 Very good Process control In order to overcome the lack of homogeneity and the
4.1þ Excellent Any application
limited spectral range, Prieto et al.33 tested NIR spectros-
copy potential to predict chemical composition in beef
where the spectra were collected (400–2498 nm) with
benchtop equipment on homogenized meat samples.
Application of Near-Infrared Spectroscopy Despite using homogenized samples, NIR spectroscopy
to Characterize Meat Fat and Meat was only suitable for rough screening purposes for the
Products moisture, protein, and intramuscular fat content, probably
due to the limited number of cattle used in the study (i.e.,
Prediction of Chemical Components only 63 steers were used and fed diets containing either
Evaluation of meat composition is essential due to its rela- sunflower or flaxseed in order to modify meat fatty acid
tionship with overall quality and palatability characteristics, (FA) profiles). Prevolnik et al.34 tested NIR spectroscopy
wholesomeness, and impact on consumer health. Many stu- (400–2500 nm) on homogenized samples to predict chem-
dies exploring NIR spectroscopy application for the analysis ical composition in a diverse set of raw meats and meat
of chemical composition of meat have been reported over products including several pork muscles and muscles
the past few years and a selection of studies (Table 2) are obtained from different species. These authors demon-
discussed below for individual meat chemical components. strated the remarkable ability of NIR spectroscopy to pre-
The ability to predict protein, fat, and moisture content dict moisture, protein, and intramuscular fat content over
in meat samples using NIR spectroscopy has been covered the diverse sample set. However, unreliable predictability
in previous reviews.7,29 Nevertheless, even today, variable was observed for moisture and protein content when only
results are found in different studies. Liao et al.30 used a the pig longissimus dorsi muscle was considered. These
conveyor spectrometer to scan (350–1100 nm) intact slices results showed again the importance of a wide range of
of M. longissimus dorsi under simulated online conditions variability of the components to enable their accurate pre-
and found a great variability in the results depending on the diction by NIR spectroscopy. Indeed, when Su et al.35
method used for spectra pre-processing. These authors tested NIR spectroscopy (1000–1800 nm) to predict the
observed that the multiplicative scatter correction (MSC) main chemical components in homogenized beef samples
in conjunction with a first derivative provided the most with a wide range of variability, NIR spectroscopy was
accurate calibration models. These pre-processing proced- reported to have an outstanding ability to predict moisture,
ures eliminated the negative impact of the translational protein, and intramuscular fat content.
error independent of wavelength in the reflectance spectra. Hence, the results from the previous studies show that
Such errors are caused by varying sample thicknesses that NIR spectroscopy has potential to replace existing wet
can greatly affect the calibration procedure. Nevertheless, chemistry methods to successfully estimate meat chemical
despite the mathematical treatment applied to the spectra, composition. Nevertheless, in order to obtain prediction
visible–near-infrared (Vis-NIR) spectroscopy was suitable equations accurate enough to be used in quality and pro-
only for rough screening purposes for moisture content cess control, both a wide range of chemical composition
and unsatisfactory for protein and intramuscular fat content variability and sample homogenization are required. These
predictions, according to the criteria established by constraints can hamper the implementation of NIR spec-
Williams.24,25 These results could have been due to the troscopy online, where further improvement is still neces-
small range of values for protein content and the lack of sary before industrial use can be considered.
homogeneity in intact meat.30 Thus, these authors indicated Meat is a major contributor of lipids to the human diet.
that predictability of Vis-NIR spectroscopy could be Consumers have become more interested in the fat com-
improved by increasing sample numbers and the variability position of meat due to nutritional guideline recommenda-
of muscles. Additionally, predictions could be improved by tions for reducing total fat and saturated fatty acids (SFA)
utilizing NIR spectral data from wavelengths above intake, while increasing consumption of polyunsaturated
1100 nm; this region contains spectral data important for fatty acids (PUFA).36 Hence, there is a growing demand
the prediction of chemical components.29 Due to the lack for fast and efficient alternative methods to monitor meat
of homogeneity, repetitive scans were not totally replicated FA profiles. In this regard, several studies have tested the
Table 2. Summary of applications of NIR spectroscopy to predict chemical components in meat (sorted by species and publication year).

Calibration and prediction performance

Spectrometer, spectral
Prieto et al.

Sample range (nm) Multivariate

Species presentation Attributes Acquisition mode analysis R2CAL SECV RPD Reference

Pig Intact IMF (g/kg) EPP2000 PLSR 0.28 1.03y 1.3 Balage et al. (2015)32
400–1395
Reflectance
Pig Intact Moisture (%) USB4000 PLSR 0.42–0.83 0.78–1.34 1.16–2.01 Liao et al. (2010)30
Protein (%) 350–1100 0.07–0.82 0.40–0.74 1.01–1.74
IMF (%) Reflectance 0.16–0.85 0.08–0.15 1.03–1.77
Beef Ground SFA (mg/g meat) NIRSystems6500 MPLSR 0.93* 108.8 3.82 Mourot et al. (2015)44
MUFA (mg/g meat) 400–2500 0.93* 116.0 3.68
PUFA (mg/g meat) Reflectance 0.59* 28.71 1.54
CLA (mg/g meat) 0.79* 2.38 2.14
Omega-3 (mg/g meat) 0.37* 8.32 1.25
Omega-6 (mg/100 g meat) 0.58* 22.09 1.52
Beef Ground Moisture (%) NIRSystems6500 PLSR 0.90 0.60 2.13 Prieto et al. (2014)33
Protein (%) 400–2498 0.85 0.48 2.10
IMF (%) Reflectance 0.86 1.08 2.01
SFA (mg/g meat) 0.97 1.05 4.54
MUFA (mg/g meat) 0.96 1.15 4.11
PUFA (mg/g meat) 0.44 0.21 1.18
CLA (mg/g meat) 0.83 0.04 2.28
Omega-3 (mg/g meat) 0.61 0.12 1.36
Beef Ground Moisture (%) SupNIR-1500 PLSR 0.92–0.997 1.22–4.95 2.63–10.69 Su et al. (2014)35
Protein (%) 1000–1800 0.92–0.99 0.70–1.41 2.71–5.46
IMF (%) Reflectance 0.998–0.998 0.99–1.20 14.22–17.37
Beef Intact IMF (mg/100 g meat) LabSpec2500 PLSR 0.43/0.75 1029/477 1.1/1.9 Prieto et al. (2011)31
(AA/LIMx) SFA (mg/100 g meat) 350–1800 0.40/0.68 405/235 1.1/1.7
MUFA (mg/100 g meat) Reflectance 0.44/0.75 452/240 1.1/1.9
PUFA (mg/100 g meat) 0.16/0.64 16/17 1.0/1.5
Omega-3 (mg/100 g meat) 0.43/0.12 8.1/9.0 1.1/1.0
Omega-6 (mg/100 g meat) 0.73/0.45 18/21 1.4/1.1
Lamb Intact IMF (%) FieldSpec GA-PLSR 0.71* 1.60y 1.77 Pullanagari et al. (2015)39
SFA (mg/100 g meat) 350–2500 0.61* 192.21y 1.61
MUFA (mg/100 g meat) Reflectance 0.62* 168.72y 1.56
PUFA (mg/100 g meat) 0.71* 27.86y 2.11
1409

(continued)
 1410

Table 2. Continued

Calibration and prediction performance

Spectrometer, spectral
Sample range (nm) Multivariate
Species presentation Attributes Acquisition mode analysis R2CAL SECV RPD Reference

Lamb Intact/Ground SFA (g/100 g meat) NIRSystems6500 MPLSR 0.52/0.98* 0.55/0.13 1.43/6.78 Guy et al. (2011)37
MUFA (g/100 g meat) 400–2500 0.46/0.98* 0.59/0.14 1.34/6.81
PUFA (g/100 g meat) Reflectance 0.45/0.89* 0.08/0.04 1.32/3.00
CLA (g/100 g meat) 0.41/0.84* 0.02/0.01 1.32/2.46
Omega-3 (g/100 g meat) 0.35/0.78* 0.02/0.01 1.21/2.08
Omega-6 (g/100 g meat) 0.37/0.83* 0.07/0.04 1.23/2.43
Chicken Intact SFA (% of total FA) LabSpec2500 MPLSR 0.16 1.66 1.17 De Marchi et al. (2012)40
MUFA (% of total FA) 350–1830 0.41 2.58 1.36
PUFA (% of total FA) Reflectance 0.39 3.35 1.29
Omega-3 (% of total FA) 0.28 0.26 1.27
Omega-6 (% of total FA) 0.39 3.10 1.29
Chicken Ground SFA FoodScan MPLSR 0.09/0.92 1.51/53.6 0.66/4.86 Riovanto et al. (2012)41
(% of total FA / MUFA 850–1050 0.59/0.98 2.29/70.1 1.46/4.73
mg FA.100 g–1 meat) PUFA Transmittance 0.44/0.65 3.15/62.0 1.34/2.84
Omega-3 0.39/0.39 0.39/5.0 1.38/1.86
Omega-6 0.41/0.67 2.76/57.9 1.41/2.88
Chicken Freeze-dried ground SFA NIRSystems6500 MPLSR 0.29/0.95 1.23/0.87y 1.2/4.0 Zhou et al. (2012)42
(% of total FA / MUFA 400–2500 0.96/0.94 3.29/1.12y 1.6/4.2
g FA. kg–1 meat, DM) PUFA Reflectance 0.96/0.97 2.74/0.83y 1.9/2.6
Omega-3 0.86/0.95 0.24/0.08y 2.1/2.5
Omega-6 0.97/0.98 3.38/0.79y 1.6/2.5
Several species Ground Moisture (%) NIRSystems6500 PLSR 0.91–0.99 0.35–1.38y 1.2–5.0 Prevolnik et al. (2010)34
Protein (%) 400–2500 0.10–0.93 0.39–0.88y 1.1–4.5
IMF (%) Reflectance 0.96–0.99 0.16–2.14y 4.1–10.1
*Coefficient of determination of cross-validation.
y
Standard error of prediction.
AA, Aberdeen Angus crossbred; LIM, Limousin crossbred; DM, dry matter basis; IMF, intramuscular fat; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids;
CLA, conjugated linoleic acid; PLSR, partial least squares regression; MPLSR, modified partial least squares regression; GA-PLSR, genetic algorithm based partial least squares regression; R2CAL, coefficient of
determination of calibration; SECV, error standard of cross-validation; RPD, ratio of performance deviation.
Applied Spectroscopy 71(7)
Prieto et al. 1411

potential of NIR spectroscopy to predict the FA compos- the organic matter at very low concentration. Despite the
ition in meat from different species (Table 2, only FA groups above limits, De Marchi et al.40 indicated that the applica-
are shown). Guy et al.37 reported NIR spectroscopy pre- tion of an online spectroscopic analysis could be useful for
diction models were improved by using ground versus improving the FA composition of meat through genetic
intact non-ground muscle samples from lambs (R2 of selection programs; however, further investigation is
cross-validation: 0.29 to 0.98 versus 0.05 to 0.53 for indi- required to confirm this application in chicken. In contrast,
vidual and groups of FA, respectively). This is likely the much higher NIR spectroscopy predictability for FA profile
result of heterogeneity of intact muscle. While the ability (expressed in absolute concentrations) was observed when
to collect NIR spectra from intact non-ground muscle spectra were collected on ground chicken breast,41 with
would be advantageous from a practical perspective, the exceptional NIR spectroscopy ability to predict SFA and
heterogeneous nature of muscle probably limits the ability MUFA, and suitable for screening purposes for PUFA and
to generate accurate calibration models for FA contents n-6 FA content. Although the NIR spectroscopy predict-
based on NIR spectra. Additionally, the type of milling treat- ability for individual FA is not shown in this review, these
ment has a marked effect on the quality of the FA predic- authors reported that the NIR transmission spectroscopy
tion models. For instance, Guy et al.37 ground and finely showed the best prediction performances for the major FA
powdered the samples in liquid nitrogen, obtaining a very which were well represented in chicken meat (oleic, lino-
homogeneous muscle sample, and this produced much leic, palmitic, and stearic acids with R2 of 0.98, 0.70, 0.94,
better results than those obtained on samples minced and 0.79, respectively). Some individual PUFA were the
using simple commercial choppers.38 Also, the liquid nitro- hardest to predict probably due to the low content of
gen treatment reduced lipolysis, peroxidation, and break- PUFA in chicken meat samples. Several reasons have been
down caused by heat generated with the grinding process. indicated in the literature for the lack of NIR spectroscopy
Guy et al.37 also observed that the accuracy of prediction success to predict PUFA. The similarities in their NIR
models varied with the FA content: predictions were absorption pattern due to the presence of identical func-
satisfactory for FA groups or individual FA present at tional groups could make difficult the accurate estimation of
medium-to-high concentrations (total SFA, cis, and total these minor components. The greater number of double
monounsaturated FA (MUFA), 16:0, 18:0, 18:1
9 cis), bonds in PUFA result in less C-H bonds that can be
but lower for FA generally found in meat at low or very detected in the NIR region, potentially contributing to
low concentrations (18:1
9 trans, 18:2 omega (n)-6, lower prediction accuracy of PUFA by NIR spectroscopy.31
20:3 n-6, 20:5 n-3, 22:5 n-3, 22:6 n-3, total n-3 PUFA). As Moreover, the main sources of PUFA in meat are phospho-
major constituents are more easily predicted by NIR spec- lipids which are found in plasma and intracellular mem-
troscopy than compounds with low concentrations, these branes, as opposed to triacylglycerols, which are found in
authors suggested that increasing the presence of some discrete highly concentrated lipid droplets.
minor FA through dietary treatments would improve accur- Riovanto et al.41 and Zhou et al.42 also evaluated the
acy of the prediction models. Low prediction accuracies potential of NIR spectroscopy to predict the FA profile in
were also observed by Pullanagari et al.39 for individual chicken breast with each FA expressed as a percentage of
and groups of FA when NIR spectra were collected on total fat. The authors from both studies observed that FA
intact longissimus lumborum from lambs. Although meat were predicted at lower levels when they were expressed
heterogeneity can originate the portion of muscle scanned as a percentage of FA compared to absolute concentration
is not representative of total FA variation of that muscle, (mg FA per 100 g meat, g FA per kg meat, respectively).
these authors attributed the lack of success to the large Correlating the NIR spectroscopy spectral data with abso-
standard errors of laboratory . Hence, Pullanagari et al.39 lute concentration of FA should be more accurate than
suggested that NIR spectroscopy prediction ability might be using proportions, since NIR absorbance depends on the
further enhanced by improving the efficiency of fat extrac- quantity of molecular bonds in the organic matrix.31,38
tion for reference samples. Hence, Riovanto et al.41 and Zhou et al.42 corroborated
The effect of sample pre-processing treatment on NIR that the unit of measurement used to express reference
spectroscopy predictability for FA content has also been data can influence the NIR spectroscopy prediction per-
observed in chicken meat. De Marchi et al.40 reported a formance. Regarding spectra pre-processing, both studies
limited ability of NIR spectroscopy to predict both individ- showed that derivatives combined with certain scatter cor-
ual and groups of FA on intact breasts. Besides the hetero- rection methods such as MSC and SNV and detrend (SNV-
geneity of the samples, these authors indicated that another D) provided the best predictions and recommended that,
aspect that could affect NIR spectroscopy prediction per- for each FA calibration equation, the ideal mathematical
formance would be the fat content of the samples. Chicken pre-treatment should be identified to maximize the per-
breast muscle has a low-fat content; as such the FA profile formances. Riovanto et al.41 and Zhou et al.42 concluded
is more difficult to estimate due to the limited ability of NIR that freeze drying process and/or milling treatment are time
spectroscopy to detect compounds that are contained in consuming and do not permit the application of NIR
1412 Applied Spectroscopy 71(7)

spectroscopy as an online analytical technique, but they chemical structure and functional groups (mainly –CH2–).
seem necessary to guarantee accurate prediction These similar absorbances can make the prediction of FA
responses. Freeze drying is used to avoid water interfer- with low concentrations difficult, as their effect on final
ence and to increase FA concentrations; however, this spectra is relatively minor and their associated peaks can
process also increases cost of analysis. High NIR spectros- be obscured by other major FA with similar functional
copy predictability can also be obtained when other pre- groups. Further research aimed at improving prediction
treatments are applied to meat, for example, prediction accuracies of FA profiles from intact meat samples needs
performances have been found to be much higher for fat to be conducted. Nevertheless, from industry perspective,
extracts rather than intact samples.43 use of NIR spectroscopy as a noninvasive and rapid analyt-
Prieto et al.31 used NIR spectroscopy online in the abat- ical technique for FA screening purposes might be of
toir to predict the FA profile of beef. These authors enough value for meat processors and manufacturers, and
scanned the intact rib-eye area using a fiber optic contact this possibility would need to be explored for use online
probe and observed different NIR spectroscopy predictabil- under production conditions.
ity between breeds, despite animals being managed, fed, and
slaughtered in a similar manner. Prieto et al.31 hypothesized Prediction of Technological Parameters and
that NIR spectroscopy was more accurate in predicting the
FA profile of Limousin crossbred carcasses due to smaller
Sensory Attributes
adipocytes than those of Aberdeen Angus crossbred, as Technological parameters such as water holding capacity,
adipocyte size could vary spatial arrangements of intramus- color, and pH are important meat quality characteristics
cular fat deposits in meat. Later, Prieto et al.33 and Mourot that correlate with the sensory appreciation of meat by
et al.44 tested the NIR spectroscopy ability using benchtop consumers.46 Several studies have tested the NIR spectros-
equipment to predict FA content of homogenized beef and copy ability to quickly predict technological parameters in
found higher NIR spectroscopy predictability than Prieto meat. As shown in Table 3, Vis-NIR spectroscopy was suit-
et al.31 on intact beef, which agrees with the results previ- able for screening purposes for L*47 and L*, a*, and b* color
ously indicated for other species. Prieto et al.33 and Mourot values32 when longissimus dorsi samples from pigs were
et al.44 reported excellent NIR spectroscopy prediction scanned intact. However, Kapper et al.47 did not find reli-
equations for the content of SFA and MUFA, and suitable able Vis-NIR spectroscopy predictions for a* and b* color
for screening purposes for conjugated linoleic acid (CLA). values in pork intact samples. Similar to the suggestion
Nevertheless, unsuitable predictions were found for total made above for prediction of chemical components, these
PUFA, n-3, and n-6 FA content. These authors attributed authors indicated increasing the scanning area could have
the unsatisfactory predictions of PUFA to either an insuffi- improved their prediction equations. However, scanning
cient variability in the data (probably because of their struc- more area would likely increase scan time, which is limiting
tural function in membrane phospholipids) or a low for practical application. In a parallel study under
concentration. As Azizian and Kramer45 mentioned, a min- production plant conditions, Kapper et al.48 scanned
imum threshold of FA content appears to be required to (833–2500 nm) intact the longissimus dorsi from pigs with
detect PUFA, which is not always reached. Nevertheless, a contact probe of approximately 4.5 cm2, covering the loin
Prieto et al.33 indicated that the high NIR spectroscopy area of the sample. Nevertheless, NIR spectroscopy pre-
predictability for SFA and MUFA contents would allow dictability for L* color value was even lower than that
for calculation of the PUFA concentration as the difference reported at laboratory scale using benchtop equipment,47
between the total FA predicted by NIR spectroscopy which could be due to the lack of visible region in the
(R2 ¼ 0.97, RPD ¼ 4.70) and the sum of SFA and MUFA spectra. Additionally, Kapper et al.48 indicated that each
contents. sample was only scanned once by the NIR spectroscopy
While scanning of intact meat samples would be ideal in device, suggesting scanning the sample multiple times may
making NIR spectroscopy a rapid and low-cost analytical have improved the results. Again, increasing time required
technique to predict FA profile, the current calibration to scan each sample would limit the use of NIR spectros-
and prediction performances are still not adequate for copy under online conditions. In studies with beef, Prieto
practical use. When homogenized samples are used, et al.33 found unsatisfactory predictions for L*, a*, and b*
robust calibrations are obtained when data sets with wide color values when Vis-NIR spectroscopy spectra were col-
variability are available, but this condition is difficult to lected on ground samples. Grinding samples prior to NIR
reach for FA that are found in a constant proportion. spectra collection could have increased the rate of meat
Additionally, sample grinding is a time-consuming step, discoloration, resulting in reflectance differences between
and the commercial value of the beef and pork loin/chicken the visible region from the spectra and the objective color
breast would be decreased if their integrity were lost due measurements. Additionally, for that study, there was an
to analytical processing. Another limiting factor is the simi- approximately 5 h time period between objective color
lar absorption patterns of different FA due to similarities in measurement and NIR spectra collection. This delay likely
Table 3. Summary of applications of near infrared spectroscopy to predict technological and sensory characteristics in meat (sorted by species and publication year).

Calibration and prediction performance

Prieto et al.

Spectrometer
Sample spectral range (nm) Multivariate
Species presentation Attributes Acquisition mode analysis R2CAL SECV RPD Reference
y
Pig Intact pH EPP2000 PLSR 0.80 0.11 2.1 Balage et al. (2015)32
Color L* 400–1395 0.88 2.02y 2.3
Color a* Reflectance 0.82 0.61y 2.2
Color b* 0.80 1.07y 2.1
WBSF (N) 0.48 5.51y 1.2
Pig Intact pH NIRSystems6500 MPLS 0.39 0.2y 1.3 Kapper et al. (2012)47
Color L* 400–2498 0.76 2.3y 2.0
Color a* Reflectance 0.54 1.2y 1.4
Color b* 0.48 1.3y 1.5
Drip loss (%) 0.80 0.8y 1.9
Pig Intact pH Matrix-F FT-NIR MPLS 0.37–0.65 0.1y 1.1–1.3 Kapper et al. (2012)48
Color L* 833–2500 0.48–0.67 2.8–3.6y 1.1–1.8
Drip loss (%) Reflectance 0.58–0.76 0.7–1.1y 1.5–2.1
Pig Intact pH USB 4000 PLSR 0.79–0.86 0.10–0.16 1.52–2.40 Liao et al. (2010)30
350–1100
Reflectance
Beef Ground pH NIRSystems6500 PLSR 0.73 0.09 1.14 Prieto et al. (2014)33
Color L* 400–2498 0.80 1.27 1.69
Color a* Reflectance 0.71 1.45 1.25
Color b* 0.77 0.77 1.56
Shear force 16 d (kg) 0.81 0.62 1.70
Beef Intact/ pH LabSpec2500 PLSR 0.62/0.42* 0.10/0.13 1.70/1.31 De Marchi et al. (2013)49
Ground Color L* 350–1800 0.70/0.55* 1.97/2.39 1.87/1.54
Color a* Reflectance 0.73/0.52* 1.37/1.82 1.89/1.43
Color b* 0.60/0.41* 1.33/1.64 1.59/1.29
Ageing loss (%) 0.15/0.12* 1.32/1.33 1.09/1.08
Cooking loss (%) 0.38/0.12* 3.02/3.50 1.23/1.06
WBSF (N) 0.34/0.13* 9.39/10.74 1.24/1.09
Beef Intact MORSf (N) NIRSystems6500 PLSR 0.56–0.86 3.15–4.15 1.13–1.50 Yancey et al. (2010)54
WBSF (kg) 400–2498 0.42–0.80 0.65–0.73 1.05–1.18
Tenderness Reflectance 0.47–0.86 0.57–0.71 1.10–1.38
Overall impression 0.43–0.89 0.39–0.52 1.06–1.40
1413

(continued)
1414 Applied Spectroscopy 71(7)

De Marchi et al. (2011)50

WBSF, Warner-Bratzler shear force; MORSf, Meullenet-Owens razor shear force; PLSR, partial least squares regression; MPLSR, modified partial least squares regression; R2CAL, coefficient of determination
contributed to the lower reliability of those NIR spectros-
copy predictions, as the oxidation states of myoglobin pig-
ments in meat would have changed between measures, as
well as color. Although more accurate predictions were
Reference reported on intact than in homogenized meat samples by
De Marchi et al.,49 as expected for color values, Vis-NIR
spectroscopy calibration models for the prediction of L*,
a*, and b* color values on intact samples did not meet the
requirements for screening purposes according to
Williams.24,25 Unlike previous studies in beef, De Marchi
et al.,50 using portable NIR spectroscopy equipment in
the range of 350–1800 nm, found Vis-NIRS prediction suit-
RPD

1.42
1.40
2.14
2.82
2.14
1.57
1.20 able for screening purposes for a* and b* values on intact
Calibration and prediction performance

chicken breast, although unreliable prediction was observed

for L* value. Hence, the ability of NIR spectroscopy to
predict L*, a*, and b* has been found to be variable
between studies, and this variability could be attributed
SECV

to various factors including instrument capabilities and set-

0.09
1.73
0.29
1.16
1.00
1.88
3.18

tings, number and types of samples, different species, stat-

istical methods adopted, different ranges of wavelength in
spectra collection, and local conditions (industry or labora-
tory) among others.
Regarding pH, Balage et al.32 and Liao et al.30 found NIR
R2CAL

0.50*
0.48*
0.77*
0.86*
0.49*
0.58*
0.17*

spectroscopy suitable for rough screening purposes for

intact pork samples, whereas other authors have reported
unreliable predictions for intact pork and chicken, and for
both intact and homogenized beef.33,47–50 As indicated by
Multivariate

Prieto et al.,29 the limited predictive ability of NIR spectros-

analysis

PLSR

copy models for pH could be caused by the low variation in

measures for this trait. Additionally, pH values vary
of calibration; SECV, error standard of cross-validation; RPD, ratio of performance deviation.

between samples and locations in the same muscle depend-

ing on the marbling.51 This affects the repeatability of the
spectral range (nm)
Acquisition mode

reference method and jeopardizes the NIR spectroscopy

Spectrometer

predictability, especially when pH measurements and NIR

LabSpec2500

Reflectance
350–1800

spectra collection have been performed in different muscle

locations. It is common practice to avoid the incision cre-
ated from the pH-probe when collecting NIR spectra after
pH measurements; this results in NIR spectra collection
from a different muscle area. De Marchi et al.49 observed
a better prediction of pH when collecting spectra on intact
Cooking loss (%)
Thawing loss (%)

rather than homogenized samples; probably due to a lack of

information about the muscle structure (light scattering
*Coefficient of determination of cross-validation.
WBSF (N)
Attributes

properties from intact muscle tissue) when scanning the

Color b*
Color a*
Color L*

samples after grinding.52 Additionally, pH measures from

intact meat are mainly from lean tissue, since effort is usu-
pH

ally taken to avoid large intramuscular fat depositions when

measuring pH. However, measures of pH obtained from
Standard error of prediction.
presentation

ground samples contain information from both lean and

fat, so pH values from the same muscle could vary accord-
Table 3. Continued

Sample

Intact

ing to sample preparation. These could be the reasons why

Prieto et al.33 did not find accurate NIR spectroscopy pre-
dictions for pH measured on intact meat when spectra
were collected on homogenized samples.
Chicken
Species

Near-infrared spectroscopy has been tested to predict

other technological traits such as water holding capacity
y
Prieto et al. 1415

(measured as cooking, drip, thawing, or ageing losses). quantify moisture levels, protein quality, and possibly the
Acceptable NIR spectroscopy equations for screening pur- presence of flavor compounds. Additionally, characteristics
poses have been reported for drip loss in beef47,48 and such as protein density, sarcomere length, and collagen
thawing loss in poultry meat.50 Conversely, De Marchi cross-linking that influence tenderness might be detected
et al.49,50 did not find reliable predictions for ageing and/ using NIR spectroscopy. Hence, Yancey et al.54 concluded
or cooking losses in beef and poultry meat. It is well known that due to the ability to measure traits beyond tenderness
that NIR spectroscopy cannot directly predict ageing and and owing to the nondestructive manner through which
cooking losses, but it may through the association of water measures can be obtained, Vis-NIRS may be a better alter-
holding capacity with water, fat (especially intramuscular), native to traditional shear tenderness in predicting tender-
and protein wavelengths. Nevertheless, the heterogeneity ness and overall impression from a consumer panel.
of meat samples29 and the low repeatability of measuring Despite recent improvements in NIR spectroscopy tech-
water holding capacity of meat53 have been indicated as nology having increased its potential,47,48 published results
possible causes for the limited ability of NIR spectroscopy show NIR spectroscopy still has a limited ability to predict
to predict ageing and cooking losses. the technological and sensorial quality of meat.
De Marchi et al.49 and Prieto et al.33 reported limited Nevertheless, NIR spectroscopy has shown some ability
NIR spectroscopy ability to predict shear force of beef for screening or classification of meat based on some of
when spectra were collected on homogenized meat sam- these characteristics. However, it remains unknown if these
ples. Again, this could be attributed to homogenization classifications would be successful for online application in
prior to spectra collection, as clearly grinding will destroy meat processing plants. As there is a need within the meat
the muscle structure and fiber arrangements. Nevertheless, processing industry for further market segmentation to
NIR spectra collected from intact samples also did not yield fulfil consumer desires for high quality products, efforts
accurate predictions in beef, pig, and poultry meat.32,49,50,54 to control all the factors influencing the spectral data and
Shear force measures obtained from the same muscle can the precision of reference methods are needed to improve
have a high variability due to muscle heterogeneity, thus the predictive/screening performance of NIR models for
making NIR spectroscopy prediction for this parameter dif- technological and sensory attributes of meat.
ficult. Nevertheless, Balage et al.32 observed a classification
model that correctly categorized 72% of pork samples into
Prediction of Carcass Fat Quality
tender and tough classes using Vis-NIR spectroscopy. These
authors stated that a technology able to discriminate In recent years, due to dietary recommendations put forth
tender pork with such accuracy might be useful for indus- by the World Health Organization,56 several strategies to
trial applications, as it could allow the high-quality meat to enhance FA composition of animal products destined for
be rapidly identified for differentiated product lines. human consumption (e.g., modulating genetics and diet) are
Additionally, the potential of NIR spectroscopy for online being investigated.57,58 However, the increased concentra-
classification of beef carcasses for longissimus tenderness tion of dietary unsaturated FA can have negative effects on
has been shown by Shackelford et al.,55 who suggested that fat quality. Porcine fat quality, for example, is an important
this technology may allow for tenderness-based beef mer- factor for economic value, human nutrition, and taste.
chandising systems. Increased concentrations of unsaturated FA can lead to
Yancey et al.54 used Vis-NIR spectroscopy regression soft fat, processing problems, reduced quality and shelf
equations for predicting consumer panel responses for ten- life of processed pork products, and an inability to meet
derness and overall impression. These authors found better fresh pork export specifications.59 Consequently, soft fat is
results when a second order derivative was applied to the deemed undesirable in multiple countries, such as
spectra, where the percentage of variance explained by the Canada,60 the United States,61 and the European Union.62
model was over 85% for both tenderness and overall In fact, soft fat is the main cause for downgrading and
impression. Visible NIRS was more successful in the pre- lowered price in Japan due to its reduced sliceability and
diction of consumer responses for tenderness and overall lower processing quality,63 particularly in bacon manufac-
impression than shear methods, Meullenet–Owens razor turing. Beyond issues with fat softness, increased contents
shear (MORS), and Warner–Bratzler shear force (WBSF). of unsaturated FA can also reduce oxidative stability and
While the Vis-NIRS equations had similar accuracies in pre- negatively affect the flavor of meat.64
dicting tenderness and overall impression, the MORS and Analyses to evaluate fat quality using standard methods,
WBSF methods were better to estimate tenderness such as gas chromatography, demand a high level of tech-
(R2 ¼ 0.38–0.58) than overall impression (R2 ¼ 0.15–0.37). nical expertise, are expensive, time-consuming, and require
Shear techniques are unable to predict aspects of overall toxic solvents and reagents. Since NIR spectroscopy is
impression such as juiciness and flavor. However, NIR rapid, relatively easy to use, and requires no chemical
values are obtained from vibration modes of chemical use, several authors (Table 4) have evaluated its potential
bonds within samples;6 therefore, it is better able to to predict the FA composition in pig carcasses. In addition
Table 4. Summary of applications of NIR spectroscopy to predict carcass fat and meat product quality (sorted by product and publication year). 1416
Spectrometer Calibration and prediction performance
spectral range
Sample (nm) Acquisition Multivariate
Product presentation Attributes mode analysis R2 CAL SECV RPD Reference
y
Pork Intact Iodine value LabSpec4 PLSR 0.95 1.03 3.61 Prieto et al. (2016)70
subcutaneous fat SFA (%) 350–2500 0.88 0.82y 2.55
MUFA (%) Reflectance 0.85 0.90y 2.18
PUFA (%) 0.94 0.62y 3.54
Omega-3 (%) 0.93 0.36y 3.09
Omega-6 (%) 0.89 0.62y 2.54
Pork Intact Iodine value NIRSystem6500 PLSR 0.90/0.87 1.66/1.80 2.54/2.42 Prieto et al. (2014)67
subcutaneous fat (Cold/Warm) SFA (%) 400–2498 0.89/0.86 1.11/1.37 2.68/2.27
MUFA (%) Reflectance 0.86/0.82 1.11/1.23 2.17/1.96
PUFA (%) 0.89/0.86 0.97/1.08 2.47/2.16
Omega-3 (%) 0.82/0.80 0.25/0.26 2.08/2.04
Omega-6 (%) 0.86/0.83 0.94/1.03 2.18/2.02
Pork Intact Iodine value NitFom iPLSR 0.83 1.44 2.35 Sorensen et al. (2012)71
subcutaneous fat 1100–2200
Transmittance
Pork Small pieces/ Iodine value FoodScan MPLSR 0.98 0.57 6.8 Gjerlaug-Enger et al. (2011)66
subcutaneous fat Melted fat SFA (%) 850–1050 0.98 0.38 7.1
in a microwave MUFA (%) Transmittance 0.95 0.45 4.2
PUFA (%) 0.98 0.28 6.5
Pork Intact (Longitudinal/ SFA (%) Matrix-F FT-NIR PLSR 0.85/0.80 1.7/1.9y 2.33/2.08 Perez-Juan et al. (2010)68
subcutaneous fat Transversal cuts) MUFA (%) 909–2500 0.92/0.88 1.2/1.2y 3.99/3.99
PUFA (%) Reflectance 0.77/0.74 1.6/1.4y 1.94/2.21
Pork dry-cured Ground SFA (%) NIRSystem6500 MPLSR 0.94 0.98 2.63 Fernandez-Cabanas et al. (2011)74
sausages MUFA (%) 400–2498 0.78 1.47 1.45
PUFA (%) Reflectance 0.83 0.88 1.58
Thai steamed Intact Moisture (%) MPA FT-NIR PLSR 0.92–0.98/ 0.79–2.67/ 1.46–4.92/ Ritthiruangdej et al. (2011)73
pork sausages (Plastic casing/ 800–2500 0.98–0.98 0.76–2.05y 2.08–5.62
Non-plastic casing) Protein (%) Reflectance 0.65–0.96/ 0.68–1.33/ 1.54–3.0/
0.44–0.97 0.57–1.78y 1.15–3.41
Fat (%) 0.87–0.91/ 1.71–2.31/ 1.99–2.69/
0.90–0.94 1.51–1.99y 2.31–3.06
Ash (%) 0.61–0.70/ 0.19–0.20/ 1.13–1.20/
0.19–0.99 0.11–0.27y 1.05–2.45
Carbohydrate (%) 0.93–0.94/ 2.22–4.67/ 1.17–2.45/
Applied Spectroscopy 71(7)

0.92–0.96 2.15–2.65y 1.90–2.34

(continued)
 Prieto et al.

Table 4. Continued

Spectrometer Calibration and prediction performance

spectral range
Sample (nm) Acquisition Multivariate
Product presentation Attributes mode analysis R2 CAL SECV RPD Reference
y
Fermented Intact Moisture (%) Matrix-F FT-NIR PLSR 0.997/0.998 0.675/0.622 19.8/21.6 Collell et al. (2010)77
pork sausages (On-contact/ 830–2500
Remote probe) Reflectance
Water activity 0.991/0.987 0.006/0.007y 9.2/8.3
NaCl (%) 0.981/0.994 0.117/0.116y 6.2/6.2
Smoked and Ground Moisture (%) NIRSystem5000 PLSR 1.00 1.0 6.9 Boschetti et al. (2013)82
dry-cured pork Water activity 1100–2500 1.00 0.4 9.1
NaCl (%) Reflectance 0.86 8.2 2.7
Protein (% DM) 1.00 1.4 2.0
Fat (% DM) 1.00 11.4 2.8
Ash (% DM) 1.00 2.1 7.5
Dry-cured pork Ground Moisture (g/kg) NIRSystem6500 PLSR 0.86–0.90 4.90–6.58 2.18–2.88 Prevolnik et al. (2011)75
Salt (g/kg) 400–2500 0.96–0.97 1.60–2.28 3.12–4.57
Protein (g/kg) Reflectance 0.81–0.84 0.59–0.63 1.82–1.94
Non-protein nitrogen (g/kg) 0.76–0.89 0.38–0.47 1.74–2.06
Proteolysis index (%) 0.72–0.82 0.84–0.97 1.58–1.78
Intramuscular fat (g/kg) 0.88–0.91 3.08–3.91 2.46–2.97
Free amino acids 0.63–0.93 358–656 1.29–2.30
(mg/100 g DM)
Several meat Intact/ NaCl (%) FoodScan MPLSR 0.64–0.98/ 0.07–0.13/ 1.27–5.71/ De Marchi et al. (2017)83
products Ground 850–1050 0.81–0.99 0.08–0.17 1.78–6.56
Transmittance
*Coefficient of determination of cross-validation.
y
Standard error of prediction.
SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; DM, dry matter basis; FT, Fourier transformation; PLSR, partial least squares regression; iPLSR, interval
partial least squares regression; MPLSR, modified partial least squares regression; R2CAL, coefficient of determination of calibration; SECV, error standard of cross-validation; RPD, ratio of performance.
1417
1418 Applied Spectroscopy 71(7)

to using NIR spectroscopy to estimate FA profile, research- replace the need for destructive and time-consuming tech-
ers have attempted to predict iodine value (IV) in pork niques to measure IV chemically. In addition, providing this
subcutaneous fat. The IV is an estimate of the proportion information to pig producers and the feed industry may
of unsaturated FA in a sample and, therefore, correlates to allow for genetic selection and diet formulation aimed at
carcass fat firmness.65 Gjerlaug-Enger et al.66 reported improving fat quality. The use of NIR spectroscopy could
excellent NIR spectroscopy equations to predict propor- also lead to the inclusion of IV in payment systems. For
tions of FA groups and IV from pig fat using benchtop equip- example, Switzerland has successfully implemented IV in
ment scanning 850–1050 nm, whereas Prieto et al.67 their payment system for pork, where the suppliers of fin-
observed that NIR spectroscopy was only suitable for ished pigs receive payment deductions when a carcass has
screening purposes despite using laboratory equipment an IV > 62.72
capable of scanning the full Vis-NIR range (400–2498 nm).
Perez-Juan et al.68 successfully predicted the content of
Prediction of Meat Products Quality
MUFA in pork ham fat scanning (909–2500 nm) longitudinal
and transversal cuts, albeit SFA and PUFA predictions were Regarding the recent use of NIR spectroscopy on meat
only adequate for screening. The better NIR spectroscopy products (Table 4), Ritthiruangdej et al.73 used Fourier
performance in the former study could be because the fat transform (FT)-NIR spectroscopy on spectra collected in
samples were processed before NIR scanning either by cut- reflectance to estimate chemical composition of steamed
ting into small pieces or melting fat in a microwave, whereas pork sausages. Using an external validation, prediction
in the latter study the fat was scanned intact. Fat samples accuracy was highest for moisture followed by protein,
from Gjerlaug-Enger et al.66 were, therefore, more homo- fat, carbohydrate, and ash; with sufficient accuracy for qual-
geneous, and tissue homogeneity is an important factor for ity control for moisture and adequate for screening of pro-
maximizing NIR spectroscopy predictability.29 Similar tein and fat content. The NIR spectroscopy predictability
results were observed by Zamora-Rojas et al.69 when pre- was similar when the sausages were packed and scanned
dicting individual FA (data not shown), who reported better through the packaging. These results indicated that plastic
predictions for individual FA when fat samples were melted casing does not have a large impact on the building of PLS
before scanned in transflectance mode using laboratory models, and that unwanted light scattering effects were
equipment (R2 ¼ 0.95–0.99, SEP ¼ 0.19–0.51% of total FA) adequately reduced from NIR spectra using the MSC treat-
compared to scanning intact adipose tissue (R2 ¼ 0.93–0.98, ment. This is a great advantage for NIR spectroscopy to be
SEP ¼ 0.31–1.25% of total FA). Additionally, NIR spectros- implemented by the industry as a nondestructive method
copy predictability using benchtop equipment was higher for rapid analysis of chemical composition of steamed pork
than when using a handheld micro-electro-mechanical sausages. Ritthiruangdej et al.73 observed different predic-
system (MEMS)-based NIR spectrometer on transverse tion accuracies according to the mathematical treatment
subcutaneous adipose tissue sections scanned directly on used, where prediction results were worse from second
the carcasses (R2 ¼ 0.81–0.87, SEP ¼ 0.55–1.30% of total derivative mathematical pre-processing. According to
FA). However, aforementioned sample preparation proced- these authors, the reason for the apparent breakdown
ures would be impractical for online measurements under could be explained as the non-monotinic amplification scat-
commercial conditions. Therefore, efforts have recently ter substantially induced by the physical structure of saus-
been made to test the ability of NIR spectroscopy under ages. Such scattering is better managed by MSC
real-time conditions. Prieto et al.70 accurately predicted the transformations than second derivatives; the latter appear
IV and some groups of FA in pork subcutaneous fat when to enhance differences in the physical structure of these
portable NIR spectroscopy equipment fitted with a fiber meat products creating irrelevant variation of spectral
optic contact probe was used online in a Canadian federally intensity. The second derivative has been most commonly
inspected abattoir, where spectra were captured directly in used as a pre-treatment method for NIR spectra, and
real time without interrupting carcass processing. A lower sometimes maybe without consideration as to why it may
predictability for IV in pork fat, although still suitable for or may not be suitable. However, the results from
screening purposes, was reported by Sorensen et al.71 who Ritthiruangdej et al.73 suggest that physical structure of
developed an online method based on NIR transmission the samples is important to consider, so that the adequate
spectroscopy and refined by chemometrics at full abattoir mathematical treatment to spectra collected in reflectance
processing speed (approximately 1000 carcasses per hour). is applied.
Hence, the potential of NIR spectroscopy technology to Fernandez-Cabanas et al.74 estimated the FA profile in
quickly and accurately estimate the IV of subcutaneous fat pork dry-cured sausages by NIR spectroscopy and found
on the carcass, immediately after slaughter and without prediction equations suitable for screening purposes for
sample treatment, opens new possibilities for the subse- SFA, but unsatisfactory for unsaturated FA. Sausages have
quent ability to sort pork carcasses according to fat hard- a complex physical matrix composed of meat and fat mix-
ness for marketing purposes. Such an approach would tures obtained from different anatomical regions and
Prieto et al. 1419

potentially different animal species. Hence, developing ade- reported an outstanding NIR spectroscopy predictability
quate NIR calibrations for complex products and non-con- for moisture content and excellent for water activity and
ventional analytical parameters will require a large number NaCl content with both setups. Production of high-quality
of samples in the calibration set attempting to increase the fermented sausages requires strict control during the
variation for the FA studied. For example, adding samples drying and ripening processes; inadequate control during
from animals fed different diets will vary their FA profiles. these processes can cause texture problems including
Nevertheless, these authors indicated that NIR spectros- crust formation.78 Several studies have demonstrated
copy could be useful for the rapid quality control of dry- there is a relationship between textural problems with
cured sausages, allowing for an estimation of the major superficial water activity, moisture, and NaCl contents.79,80
constituents and/or potentially allow classification of saus- Thus, monitoring of these parameters at the surface of the
ages based on FA profiles. This information could also be product online would be useful in order to prevent crust-
used to estimate sausage shelf life and for additional nutri- ing,81 and NIR spectroscopy could be a successful technol-
tional labeling. Alternatively, fats could be pre-screened for ogy for online monitoring of drying processes. Boschetti
their composition prior to adding to the sausage blend, et al.82 reported similar NIR spectroscopy ability to predict
allowing the opportunity to develop sausages with healthful water activity and lower, but still reliable, for NaCl, mois-
FA profiles. ture, fat, and ash content in a smoked and dry-cured pork
Prevolnik et al.75 used Vis and/or NIR range to predict product (Bauernspeck) using a benchtop instrument. More
chemical composition, salt content and free amino acids in recently, De Marchi et al.83 successfully predicted NaCl
dry cured ham, and observed no major differences in the proportion by NIR spectroscopy in a wide range of pro-
prediction of chemical constituents between NIR or the cessed meat products (cured meat, boiled sausages, dry
entire spectral range (Vis-NIR). Indeed, when visible spec- meat, and bacon), showing better predictability when spec-
tra were tested alone, the models had lower prediction tra were collected on ground than on intact meat products
accuracy. When using NIR spectral data, excellent NIR due to the lower percentage of outliers in the former.
spectroscopy prediction equations were obtained for salt From the studies previously mentioned, it is apparent
content and salt percentage in moisture/dry matter, satis- NIR spectroscopy has the potential for accurately predict-
factory for moisture, non-protein nitrogen, intramuscular ing and/or quickly screening different quality attributes of
fat, and total free amino acids, while unsuccessful for pro- meat products. Near-infrared spectroscopy has become a
tein content and proteolysis index. Individual free amino powerful analytical tool especially suited for quantitative
acid predictions had variable success, with comparable estimation or qualitative classification of multi-component
results from external validation. The most accurate predic- systems such as meat products and, therefore, could
tions from this study were obtained for salt content, des- replace chemical methods in quality control processes.
pite the inability of NIR spectroscopy to detect inorganic
substances76 unless they are bound to organic substances. Classification and Identification of Meat and Meat
As such, it is likely that salt (NaCl) content in dry-cured
ham is indirectly predicted from other compounds (e.g.,
Products
correlation coefficient between salt and water content Many consumers place emphasis on non-compositional
was 0.53 in Prevolnik et al.)75 and/or different quality prop- aspects of meat related to quality, such as intrinsic charac-
erties. While NaCl itself is not absorbed in the NIR region, teristics of animals (species, breed), geographical origin,
different dissolved salt concentrations cause wavelength feeding system, or post-mortem strategies. As opportu-
shifts in the spectrum. Unfortunately, the same effect can nities to provide differentiated meat and meat products
be caused by changes to sample temperature;70 hence, with enhanced quality attributes are expanding84 and con-
these models for predicting NaCl are temperature- sumers are willing to pay premiums for such products,
dependent. In the case of proteins, the protein content potential for fraud may exist. In order to guarantee that
was calculated assuming all nitrogen in the sample is in pro- consumers are not being defrauded in the purchase of meat
tein; however, a portion (27%) of nitrogen is not associated and meat products making claims of quality, origin, or spe-
with protein. Thus, this discrepancy between reference cies, authorities require tools to rapidly and successfully
method and spectral data could partly explain the lower distinguish those meat and meat products.
NIR spectroscopy predictability for the protein content. As shown in Table 5, NIR spectroscopy has been used
Based on those results, Prevolnik et al.75 concluded that for species identification purposes. Mamani-Linares et al.85
NIR spectroscopy could replace chemical methods in qual- used NIR spectroscopy to successfully identify cattle, llama,
ity control of dry-cured ham. and horse meat from homogenized meat and meat juice
Collell et al.77 tested the feasibility of NIR spectroscopy samples (89–100% of samples correctly classified).
for predicting parameters related to the drying process of Restaino et al.86 discriminated meat pates according to
fermented sausages, by acquiring spectra through two NIR animal species, with 100% of the beef and pork samples
setups with contact and remote probes. These authors correctly classified. Likewise, Schmurtzler et al.,87 using
Table 5. Summary of applications of NIR spectroscopy to classify meat and meat products (sorted by species and publication year).
1420

Spectrometer
Sample spectral range (nm) Multivariate
Species Research purpose presentation Acquisition mode analysis Correctly classified (%) Reference

Pork Discrimination of Intact LabSpec4 PLS-DA Lacombe: 94/ Duroc: 95/ Iberian: Prieto et al. (2015)90
enhanced quality pork 350–2500 100 (aged 2 days)
Reflectance Lacombe: 94/ Duroc: 98/ Iberian:
100 (aged 14 days)
ME: 97/ Non-ME: 99
(aged 2 days)
ME: 94/ Non-ME: 95
(aged 14 days)
Aged 2 days: 94/ aged 14 days: 97
Control/ Canola/ Flaxseed: n/a
(aged 2 and 14 days)
BC: 54/ Non-BC: 57 (aged 2 days)
BC: 54/ Non-BC: 53 (aged 14 days)
Pork Classification of pig Intact MEMS-NIR PLS-DA Acorn: 94/ Feed: 96/ Recebo: 61 Zamora-Rojas
carcasses based 1600–2400 et al. (2012)91
on feeding regime Reflectance
Pork Discrimination of pork Intact Antaris II FT-NIR LDA 89 Chen et al. (2011)94
storage time 1000–2500 KNN 92
(from 1 to 6 days) Reflectance BP-ANN 96
Pork Pork meat classification Intact FieldSpec DISCRIM RFN: 70–88 Monroy et al. (2010)93
based on five quality 350–2500 RSE: 57–83
groups Reflectance PFN: 67–84
PSE: 65–100
Beef Discrimination of dark Intact LabSpec4 PLS-DA Normal: 95 Prieto et al. (2014)92
cutters 350–2500 Dark cutters: 95
Ground Reflectance Normal: 88
Dark cutters: 90
Lamb Classification of lamb Intact DA7200 PLS-DA Pastoral and agricultural Sun et al. (2012)89
meat based on 950–1650 region: 100
geographical origins Reflectance Specific regions: 89
LDA Pastoral vs. agricultural
region: 100
Specific regions: 75
(continued)
Applied Spectroscopy 71(7)
 Prieto et al.

Table 5. Continued

Spectrometer
Sample spectral range (nm) Multivariate
Species Research purpose presentation Acquisition mode analysis Correctly classified (%) Reference

Pork þ veal Analysis of pork adulter- Intact NIRFlex N-500 PCA Genuine vs. 50% adulteration: 100 Schmurtzler et al.
ation in veal sausages 1659–1825 SVM Genuine vs. 40% adulteration: 100 (2015)87
Reflectance Genuine vs. 30% adulteration: 100
MicroPhazir GP 4.0 Genuine vs. 20% adulteration: 100
1656–1802 Genuine vs. 10% adulteration: 100
Reflectance
Several species Identification of cattle, Ground NIRSystem6500 PLS-DA Beef: 100/95 Mamani-Linares et al.
llama, and horse meat meat/Meat juice 400–2500 Llama: 95/100 (2012)85
Reflectance Horse: 89/95
Several species Discrimination of meat Ground NIRSystem6500 PCA Beef: 100 Restaino et al. (2011)86
pates according to the 1100–2500 Pork: 100
animal species Reflectance Binary mixtures: 72
MEMS, micro-electro-mechanical systems; PLS-DA, partial least square discriminant analysis; LDA, linear discriminant analysis; KNN, K-nearest neighbors; BP-ANN, back propagation artificial neural
network; DISCRIM, discriminant procedure in SAS; SVM, support vector machines; PCA, principal component analysis; ME, moisture enhanced; BC, blast chilled; n/a, not applicable; RFN, reddish-pink, firm,
and non-exudative; RSE, red, soft, and exudative; PFN, pale, firm, and non-exudative; PSE, pale, soft, and exudative.
1421
1422 Applied Spectroscopy 71(7)

three setups (laboratory, industrial, and on-site), showed groups (RFN: reddish-pink, firm, and non-exudative; RSE:
that NIR spectroscopy can be used to detect the presence red, soft, and exudative; PFN: pale, firm, and non-exudative;
of pork in veal sausage with a contamination level as low as and PSE: pale, soft, and exudative) using different validation
10%. This application could be used for Halal or Kosher approaches, where the highest percentages of samples cor-
verification purposes, which are increasingly becoming rectly classified were in the range of 83–100. Studies con-
mandatory. Due to religious concerns, the presence of ducted to date, therefore, clearly show the potential of
pork derivatives in food products is a serious matter, as using Vis-NIR spectroscopy in beef and pork for classifica-
some faiths forbid consumption of foods containing pork tion purposes.
or its derivatives.88 Another recent application of NIR spectroscopy has
Another application of NIR spectroscopy has been to been in the area of meat shelf life. Chen et al.94 used FT-
classify meat based on geographical origin. Sun et al.89 NIR spectroscopy and different algorithms to discriminate
reported that NIR spectroscopy, together with the applica- pork based on storage time. These authors observed that
tion of partial least squares discriminant analyses (PLS-DA) the performance of a back propagation artificial neural net-
and LDA, correctly classified 100% of lamb meat from both work (BP-ANN) model was superior to LDA and KNN,
pastoral and agricultural region samples. Additionally, 88.9% with discrimination rates of the BP-ANN model of 99.26%
pastoral and 75% agricultural samples were correctly iden- and 96.21% in the training and prediction sets, respectively.
tified to the five individual regions where samples were Hence, these authors concluded that the discrimination
obtained. Hence, these authors concluded that NIR spec- rates obtained using FT-NIR spectroscopy combined with
troscopy combined with chemometrics could effectively BP-ANN classification algorithm demonstrated this tech-
and rapidly distinguish lamb meat by geographical origin. nique could potentially be used to classify pork based on
Prieto et al.90 using PLS-DA based on Vis-NIR spectra, storage time and freshness accordingly.
observed that NIR spectroscopy could successfully classify Although sometimes the visual examination of NIR spec-
pork according to pig breed, moisture enhancement of tra cannot discriminate between authentic and adulterated
loins, and ageing duration. However, Vis-NIRS technology meat products,95,96 the application of NIR spectroscopy
was not able to differentiate pork samples based on diet or together with statistical packages and multivariate data ana-
carcass chilling process in that study. That lack of success of lysis techniques such as discriminant analysis (e.g., PLS-DA)
NIR spectroscopy to classify pork based on feeding regime has improved the understanding of the optical properties of
is in disagreement with that shown by Zamora-Rojas meat. This has allowed classification of samples without
et al.,91 who found NIR spectroscopy technology was chemical information.97 The NIR spectroscopy can there-
able to successfully classify Iberian pig carcasses by feeding fore be used for initial screening in the food chain to guar-
regime (>90% of the sample classifications were correct). antee the quality/authenticity of meat and meat products
Differences between the results of these studies could be for consumers; thus, more costly and time-consuming
attributed to diet treatments (grass and acorns fed free- methods would only be required for samples that do not
range pigs, acorns and grass supplemented with compound pass initial NIR spectroscopy screening.
feeds in an outdoor system, and compound feeds using an
intensive feeding system91 versus animals fed a normal
Conclusion and Future Outlook
Canadian commercial diet, a high-oleic diet, or a high lino-
lenic diet90), the diet durations (over 60 days versus The number of published studies on the application of NIR
3 weeks), or which tissues were scanned (subcutaneous spectroscopy in the area of meat science has been increas-
fat versus meat). Additionally, Prieto et al.90 used canola ing in recent years, as this technique has the advantages of
and flaxseed diets formulated to increase FA beneficial being applied rapidly without further sample preparation.
for human health. Hence, it was expected the largest Many of the studies surveyed in this review show NIR spec-
differences between diets would be in fat tissue. In that troscopy can be a powerful analytical tool especially suited
study, the meat samples scanned had around 2.6% of intra- for quantitative estimation of quality attributes in intact/
muscular fat (2.70% in the Control, 2.53% in the Canola, homogenized adipose tissue and multicomponent systems
and 2.57% in the Flaxseed pork samples), which could in such as meat products and, therefore, could have the
part explain why Vis-NIR spectroscopy classification was potential to replace chemical methods in quality control
unsuccessful. processes. Near-infrared spectroscopy could also have a
Additionally, Vis-NIR spectroscopy has shown the ability role in increasing consumer confidence in meat and final
to classify beef and pork into quality grades. Prieto et al.92 meat products by confirming integrity, particularly identify-
successfully discriminated dark cutters from normal beef ing enhanced quality meat and confirming authenticity of
using homogenized meat scanned with benchtop equipment species and geographical origin. Although NIR spectroscopy
and intact meat scanned using portable Vis-NIR spectros- can successfully predict chemical composition of ground
copy (88–95% of samples were correctly classified). meat, limited ability has been shown for unprocessed
Monroy et al.93 classified pork meat into four quality meat samples, where prediction performances for meat
Prieto et al. 1423

quality attributes are still not accurate enough for practical spectroscopy is not sensitive to the mineral content).
use. Nevertheless, from an industry perspective, use of NIR Therefore, the combination of NIR spectroscopy with
spectroscopy as a noninvasive, chemical-free, and rapid ana- other detection techniques, such as dual energy X-ray
lytical technique for screening purposes might be of enough absorptiometry (DEXA), would allow estimation of not
value for meat processors and manufacturers, and would only chemical components and meat quality attributes but
need further exploration for online implementation. also whole carcass composition (fat, lean, and bone tissue
The application of NIR spectroscopy has known advan- content). Next-generation NIR spectrometers may also
tages (lower costs; rapid, in situ, and nondestructive ana- have potential to be implemented in smartphones. As a
lyses; multi-parameter estimation; and environmental consequence, future research for quality control applica-
friendliness). However, there are several drawbacks and tions in carcass, meat and meat products will likely focus
challenges associated with using this technology, such as on using NIR spectroscopy combined with other nondes-
sample presentation, which may become a crucial issue tructive technologies and the implementation of portable,
when scanning intact meat samples due to their heterogen- low-cost next-generation NIR instruments.
eity and high absorbance of water in the infrared region.
Collection of several spectra from the same sample, in Acknowledgments
order to increase the scanned area, could partly solve The authors thank Mr Jordan Roberts and Ms Karen Joy for their
this issue. Nevertheless, this practice can hamper the technical assistance during the preparation of this review.
online implementation of NIR spectroscopy, where scan-
ning time is a limiting factor for industrial applications. Conflict of Interest
Hence, further improvements of NIR spectrometers (e.g., The authors report there are no conflicts of interest.
equipped to scan larger areas to reduce the sampling error)
are still necessary before industrial implementation can be Funding
considered. This would also have to be achieved without
This research received no specific grant from any funding agency in
significant added cost to the equipment. Although measure- the public, commercial, or not-for-profit sectors.
ment costs are low using the NIR spectroscopy technique,
instrument costs are high meaning practical applications References
may still be restricted by cost. Researchers and analysts
1. Z. Kallas, C.E. Realini, J.M. Gil. ‘‘Health Information Impact on the
are therefore looking for sensitive wavelengths in the NIR Relative Importance of Beef Attributes Including Its Enrichment with
region representing characteristics of food products, allow- Polyunsaturated Fatty Acids (Omega-3 and Conjugated Linoleic
ing for development of simpler and more specialized instru- Acid)’’. Meat Sci. 2014. 97(4): 497–503.
ments at a lower cost. If successful, the applications of NIR 2. C.E. Realini, Z. Kallas, M. Pérez-Juan, et al. ‘‘Relative Importance of
Cues Underlying Spanish Consumers’ Beef Choice and Segmentation,
spectroscopy may become more widely used and popular in
and Consumer Liking of Beef Enriched with n-3 and CLA Fatty Acids’’.
many meat industries. Nevertheless, when reducing the Food Qual. Prefer. 2014. 33: 74–85.
wavelength range the possibility of assessing several attri- 3. W. Herschel. ‘‘Experiments on the Refrangibility of the Invisible Rays
butes simultaneously is also constricted. of the Sun’’. Philos. Trans. R. Soc. 1800. 90: 284–292.
Some studies reported in this review show potential for 4. K.H. Norris, J.R. Hart. ‘‘Direct Spectrophotometric Determination of
Moisture Content of Grain and Seeds’’. Paper presented at:
portable instruments; however, most research has been
International Symposium on Humidity and Moisture in Liquids and
performed under laboratory conditions or in research abat- Solids. Washington, DC; 1963. Pp. 19–25.
toirs. Therefore, it is difficult with the current data to evalu- 5. T. Isaksson, B.N. Nilsen, G. Togersen, et al. ‘‘On-Line, Proximate
ate the real performance of portable NIR spectroscopy Analysis of Ground Beef Directly at a Meat Grinder Outlet’’. Meat
devices in industrial settings. Online measurements at Sci. 1996. 43(3–4): 245–253.
6. B.G. Osborne, T. Fearn, P.H. Hindle. Near Infrared Spectroscopy in
industrial levels remain challenging due to major influences
Food Analysis. Harlow, UK: Longman Scientific and Technical, 1993.
of temperature fluctuations or moving samples; therefore, 7. J. Weeranantanaphan, G. Downey, P. Allen, et al. ‘‘A Review of Near
further work is still required to refine these instruments so Infrared Spectroscopy in Muscle Food Analysis: 2005–2010’’. J. Near
they can be used easily under production conditions. Infrared Spectrosc. 2011. 19: 61–104.
Hence, development of a lower cost and a more rugged 8. C.A.T. dos Santos, M. Lopo, R.N.M.J. Páscoa, et al. ‘‘A Review on the
Applications of Portable near-Infrared Spectrometers in the Agro-
NIR instrument is necessary to integrate this technology in
Food Industry’’. Appl. Spectrosc. 2013. 67(11): 1215–1233.
the meat production process. 9. J.U. Porep, D.R. Kammerer, R. Carle. ‘‘On-Line Application of Near
The combination of techniques using different detection Infrared (NIR) Spectroscopy in Food Production’’. Trends Food Sci.
methods is a significant perspective for the use of NIR Technol. 2015. 46(2, Part A): 211–230.
spectroscopy in meat and meat product industries, since 10. H. Cen, Y. He. ‘‘Theory and Application of near Infrared Reflectance
Spectroscopy in Determination of Food Quality’’. Trends Food Sci.
it would be probably more cost effective and reduce the
Technol. 2007. 18(2): 72–83.
time needed for traditional (i.e., multiple and separate) ana- 11. H. Huang, H. Yu, H. Xu, et al. ‘‘Near Infrared Spectroscopy for on/in-
lyses. Additionally, it is an alternative way to overcome Line Monitoring of Quality in Foods and Beverages: A Review’’. J. Food
some of the limitations of techniques (e.g., NIR Eng. 2008. 87(3): 303–313.
1424 Applied Spectroscopy 71(7)

12. P. Berzaghi, R. Riovanto. ‘‘Near Infrared Spectroscopy in Animal 34. M. Prevolnik, M. Škrlep, D. Škorjanc, et al. ‘‘Application of near
Science Production: Principles and Applications’’. Ital. J. Anim. Sci. Infrared Spectroscopy to Predict Chemical Composition of Meat
2009. 107(8): 39–62. and Meat Products’’. Tehnologija Mesa. 2010. 51(2): 133–142.
13. R.F. Wolffenbuttel. ‘‘Mems-Based Optical Mini-and 35. H. Su, K. Sha, L. Zhang, et al. ‘‘Development of near Infrared
Microspectrometers for the Visible and Infrared Spectral Range’’. J. Reflectance Spectroscopy to Predict Chemical Composition with a
Micromech. Microeng. 2005. 15(7): S145. Wide Range of Variability in Beef’’. Meat Sci. 2014. 98(2): 110–114.
14. R. Crocombe. ‘‘MEMS Technology Moves Process Spectroscopy into a 36. J.D. Wood, I.R. Richardson, G.R. Nute, et al. ‘‘Effects of Fatty Acids on
New Dimension’’. Spec. Eur. 2004. 16(3): 16–19. Meat Quality: A Review’’. Meat Sci. 2003. 66: 21–32.
15. T. Pügner, J. Knobbe, H. Grüger. ‘‘Near-Infrared Grating Spectrometer 37. F. Guy, S. Prache, A. Thomas, et al. ‘‘Prediction of Lamb Meat Fatty
for Mobile Phone Applications’’. Appl. Spectrosc. 2016. 70(5): Acid Composition Using near-Infrared Reflectance Spectroscopy
734–745. (NIRS)’’. Food Chem. 2011. 127(3): 1280–1286.
16. Å. Rinnan, F. v.d. Berg, S.B. Engelsen. ‘‘Review of the Most Common 38. V. Sierra, N. Aldai, P. Castro, et al. ‘‘Prediction of the Fatty Acid
Pre-Processing Techniques for near-Infrared Spectra’’. Trends Anal. Composition of Beef by near Infrared Transmittance Spectroscopy’’.
Chem. 2009. 28(10): 1201–1222. Meat Sci. 2008. 78(3): 248–255.
17. M.J.E. Savitzky, Golay. ‘‘Smoothing and Differentiation of Data by 39. R.R. Pullanagari, I.J. Yule, M. Agnew. ‘‘On-Line Prediction of Lamb Fatty
Simplified Least Squares Procedures’’. Anal. Chem. 1964. 36(8): Acid Composition by Visible Near Infrared Spectroscopy’’. Meat Sci.
1627–1639. 2015. 100: 156–163.
18. R.J. Barnes, M.S. Dhanoa, S.J. Lister. ‘‘Standard Normal Variate 40. M. De Marchi, R. Riovanto, M. Penasa, et al. ‘‘At-Line Prediction of
Transformation and De-Trending of near-Infrared Diffuse Reflectance Fatty Acid Profile in Chicken Breast Using near Infrared Reflectance
Spectra’’. Appl. Spectrosc. 1989. 43(5): 772–777. Spectroscopy’’. Meat Sci. 2012. 90(3): 653–657.
19. R. ‘‘Leardi. ‘‘Chemometric Methods in Food Authentication’’. In: D.- 41. R. Riovanto, M. De Marchi, M. Cassandro, et al. ‘‘Use of near Infrared
W. Sun (ed.) Modern Techniques for Food Authentication. Boston, Transmittance Spectroscopy to Predict Fatty Acid Composition of
USA: Elsevier/Academic Press, 2008, pp.585–616. Chicken Meat’’. Food Chem. 2012. 134(4): 2459–2464.
20. K. Lorber, B.R. Faber, Kowalski. ‘‘Net Analyte Signal Calculation in 42. L.J. Zhou, H. Wu, J.T. Li, et al. ‘‘Determination of Fatty Acids in Broiler
Multivariate Calibration’’. Anal. Chem. 1997. 69(8): 1620–1626. Breast Meat by near-Infrared Reflectance Spectroscopy’’. Meat Sci.
21. J. Yoon, B. Lee, C. Han. ‘‘Calibration Transfer of near-Infrared Spectra 2012. 90(3): 658–664.
Based on Compression of Wavelet Coefficients’’. Chemometr. Intell. 43. A.S. Ripoche, Guillard. ‘‘Determination of Fatty Acid Composition of
Lab. 2002. 64(1): 1–14. Pork Fat by Fourier Transform Infrared Spectroscopy’’. Meat Sci. 2001.
22. E. Zamora-Rojas, D. Pérez-Marı́n, E. De Pedro-Sanz, et al. ‘‘Handheld
58: 299–304.
Nirs Analysis for Routine Meat Quality Control: Database Transfer
44. B.P. Mourot, D. Gruffat, D. Durand, et al. ‘‘Breeds and Muscle Types
from at-Line Instruments’’. Chemometr. Intell. Lab. 2012. 114: 30–35.
Modulate Performance of Near-Infrared Reflectance Spectroscopy to
23. M. ‘‘Westerhaus, J.J. Workman, J.B. Reeves, III, et al. ‘‘Quantitative
Predict the Fatty Acid Composition of Bovine Meat’’. Meat Sci. 2015.
Analysis’’. In: C.A. Roberts, J.J. Workman, J.B. Reeves (eds) Near-
99: 104–112.
Infrared Spectroscopy in Agriculture. Madison, WI: American
45. H. Azizian, J. Kramer. ‘‘A Rapid Method for the Quantification of Fatty
Society of Agronomy Inc, 2004, pp.133–174.
Acids in Fats and Oils with Emphasis on Trans Fatty Acids Using
24. P.C. Williams, K. Norris. Near- Infrared Technology in the Agricultural
Fourier Transform Near Infrared Spectroscopy (FT-NIR)’’. Lipids.
and Food Industries, 2nd ed. Eagan, MN: American Association of
2005. 40(8): 855–867.
Cereal Chemists, 2001.
46. E. Huff-Lonergan, T. Baas, M. Malek, et al. ‘‘Correlations Among
25. P.C. Williams. ‘‘Near-Infrared Technology — Getting the Best out of
Selected Pork Quality Traits’’. J. Anim. Sci. 2002. 80(3): 617–627.
the Light’’. In: S. Lawrence, P. Warburton (eds) In: A Short Course in
47. C. Kapper, R.E. Klont, J.M.A.J. Verdonk, et al. ‘‘Prediction of Pork
the Practical Implementation of Near Infrared Spectroscopy for User.
Nanaimo, Canada: PDK Projects, Inc, 2008. Quality with Near Infrared Spectroscopy (NIRS): 1. Feasibility and
26. P.C. Williams, D.C. Sobering. ‘‘Comparison of Commercial near Robustness of Nirs Measurements at Laboratory Scale’’. Meat Sci.
Infrared Transmittance and Reflectance Instruments for Analysis of 2012. 91(3): 294–299.
Whole Grains and Seeds’’. J. Near Infrared Spectrosc. 1993. 1: 25–32. 48. C. Kapper, R.E. Klont, J.M.A.J. Verdonk, et al. ‘‘Prediction of Pork
27. P. Williams. ‘‘Tutorial: The RPD Statistic: A Tutorial Note’’. NIR News. Quality with Near Infrared Spectroscopy (NIRS) 2. Feasibility and
2014. 25: 22–26. Robustness of Nirs Measurements under Production Plant
28. G. Tøgersen, J.F. Arnesen, B.N. Nielsen, et al. ‘‘On-Line Prediction of Conditions’’. Meat Sci. 2012. 91(3): 300–305.
Chemical Composition of Semi-Frozen Ground Beef by Non-Invasive 49. M. De Marchi, M. Penasa, A. Cecchinato, et al. ‘‘The Relevance of
Nir Spectroscopy’’. Meat Sci. 2003. 63: 515–523. Different near Infrared Technologies and Sample Treatments for
29. N. Prieto, R. Roehe, P. Lavı́n, et al. ‘‘Application of near Infrared Predicting Meat Quality Traits in Commercial Beef Cuts’’. Meat Sci.
Reflectance Spectroscopy to Predict Meat and Meat Products 2013. 93: 329–335.
Quality: A Review’’. Meat Sci. 2009. 83: 175–186. 50. M. De Marchi, M. Penasa, M. Battagin, et al. ‘‘Feasibility of the Direct
30. Y.-T. Liao, Y.-X. Fan, F. Cheng. ‘‘On-Line Prediction of Fresh Pork Application of near-Infrared Reflectance Spectroscopy on Intact
Quality Using Visible/near-Infrared Reflectance Spectroscopy’’. Meat Chicken Breasts to Predict Meat Color and Physical Traits’’. Poult.
Sci. 2010. 86(4): 901–907. Sci. 2011. 90(7): 1594–1599.
31. N. Prieto, D.W. Ross, E.A. Navajas, et al. ‘‘Online Prediction of Fatty 51. G. Elmasry, D.F. Barbin, D.-W. Sun, et al. ‘‘Meat Quality Evaluation by
Acid Profiles in Crossbred Limousin and Aberdeen Angus Beef Cattle Hyperspectral Imaging Technique: An Overview’’. Crit. Rev. Food Sci.
Using near Infrared Reflectance Spectroscopy’’. Anim. 2011. 5(01): Nutr. 2012. 52(8): 689–711.
155–165. 52. N. Prieto, S. Andrés, F.J. Giráldez, et al. ‘‘Ability of near Infrared
32. J.M. Balage, S. da. Luz e Silva, C.A. Gomide, et al. ‘‘Predicting Pork Reflectance Spectroscopy (Nirs) to Estimate Physical Parameters of
Quality Using Vis/Nir Spectroscopy’’. Meat Sci. 2015. 108: 37–43. Adult Steers (Oxen) and Young Cattle Meat Samples’’. Meat Sci. 2008.
33. N. Prieto, Ó. López-Campos, J.L. Aalhus, et al. ‘‘Use of near Infrared 79: 692–699.
Spectroscopy for Estimating Meat Chemical Composition, Quality 53. J. Brøndum, L. Munck, P. Henckel, et al. ‘‘Prediction of Water-Holding
Traits and Fatty Acid Content from Cattle Fed Sunflower or Capacity and Composition of Porcine Meat by Comparative
Flaxseed’’. Meat Sci. 2014. 98(2): 279–288. Spectroscopy’’. Meat Sci. 2000. 55: 177–185.
Prieto et al. 1425

54. J.W.S. Yancey, J.K. Apple, J.F. Meullenet, et al. ‘‘Consumer Responses 73. P. Ritthiruangdej, R. Ritthiron, H. Shinzawa, et al. ‘‘Non-Destructive
for Tenderness and Overall Impression Can Be Predicted by Visible and Rapid Analysis of Chemical Compositions in Thai Steamed Pork
and near-Infrared Spectroscopy, Meullenet–Owens Razor Shear, and Sausages by Near-Infrared Spectroscopy’’. Food Chem. 2011. 129(2):
Warner–Bratzler Shear Force’’. Meat Sci. 2010. 85(3): 487–492. 684–692.
55. S.D. Shackelford, T.L. Wheeler, D.A. King, et al. ‘‘Field Testing of a 74. V.M. Fernández-Cabanás, O. Polvillo, R. Rodrı́guez-Acuña, et al. ‘‘Rapid
System for Online Classification of Beef Carcasses for Longissimus Determination of the Fatty Acid Profile in Pork Dry-Cured Sausages
Tenderness Using Visible and near-Infrared Reflectance by Nir Spectroscopy’’. Food Chem. 2011. 124(1): 373–378.
Spectroscopy’’. J. Anim. Sci. 2012. 90(3): 978–988. 75. M. Prevolnik, M. Škrlep, L. Janeš, et al. ‘‘Accuracy of near Infrared
56. World Health Organization (WHO). ‘‘Diet, Nutrition and the Spectroscopy for Prediction of Chemical Composition, Salt Content
Prevention of Chronic Diseases’’. In: Report of the Joint WHO/FAO and Free Amino Acids in Dry-Cured Ham’’. Meat Sci. 2011. 88(2):
Expert Consultation – Technical Report Series 916. Geneva, 299–304.
Switzerland: World Health Organization Geneva, 2003. 76. T.Van Kempen. ‘‘Infrared Technology in Animal Production’’. World’s
57. N.D. Scollan, D. Dannenberger, K. Nuernberg, et al. ‘‘Enhancing the Poult. Sci. J. 2001. 57(01): 29–48.
Nutritional and Health Value of Beef Lipids and Their Relationship 77. C. Collell, P. Gou, P. Picouet, et al. ‘‘Feasibility of near-Infrared
with Meat Quality’’. Meat Sci. 2014. 97(3): 384–394. Spectroscopy to Predict Aw and Moisture and Nacl Contents of
58. M.E.R. Dugan, P. Vahmani, T.D. Turner, et al. ‘‘Pork as a Source of Fermented Pork Sausages’’. Meat Sci. 2010. 85(2): 325–330.
Omega-3 (N-3) Fatty Acids’’. J. Clin. Med. 2015. 4(12): 1999–2011. 78. J.A. Ordóñez, L.dela Hoz. ‘‘Embutidos Crudos Curados. Tipos.
59. S.N. Carr, P.J. Rincker, J. Killefer, et al. ‘‘Effects of Different Cereal Fenómenos Madurativos. Alteraciones’’. In: S.M. Bejarano (ed.)
Grains and Ractopamine Hydrochloride on Performance, Carcass Enciclopedia de la Carne y de los Productos Cárnicos. Cáceres,
Characteristics, and Fat Quality in Late-Finishing Pigs’’. J. Anim. Sci. Spain: Martı́n & Macı́as, 2001, pp.1063–1090.
2005. 83(1): 223–230. 79. J. Ruiz-Ramı́rez, J. Arnau, X. Serra, et al. ‘‘Relationship between Water
60. P. Soladoye, P. Shand, J. Aalhus, et al. ‘‘Review: Pork Belly Quality, Content, NaCl Content, Ph and Texture Parameters in Dry-Cured
Bacon Properties and Recent Consumer Trends’’. Can. J. Anim. Sci. Muscles’’. Meat Sci. 2005. 70(4): 579–587.
2015. 95(3): 325–340. 80. X. Serra, J. Ruiz-Ramı́rez, J. Arnau, et al. ‘‘Texture Parameters of Dry-
61. United States Department of Agriculture. United States Standards for Cured Ham M. Biceps Femoris Samples Dried at Different Levels as a
Grades of Pork Carcasses. Washington, DC: USDA, 1985, pp.1–11. Function of Water Activity and Water Content’’. Meat Sci. 2005.
62. V. Russo. ‘‘Carcass and Pork Quality: Industrial and Consumer 69(2): 249–254.
Requirements’’. Paper presented at: Pig Carcass and Meat Quality 81. J. Ruiz-Ramı́rez, X. Serra, J. Arnau, et al. ‘‘Profiles of Water Content,
Meeting. Reggio Emilia, Italy; 1988. Pp. 3–22. Water Activity and Texture in Crusted Dry-Cured Loin and in Non-
63. M. Irie, K. Ohmoto. ‘‘Studies on Physical Characteristics and Crusted Dry-Cured Loin’’. Meat Sci. 2005. 69(3): 519–525.
Techniques for Evaluation in Porcine Soft Fat’’. Jpn. J. Swine Sci.1982. 82. L. Boschetti, M. Ottavian, P. Facco, et al. ‘‘A Correlative Study on Data
19: 165–170. from Pork Carcass and Processed Meat (Bauernspeck) for Automatic
64. F. Shahidi. ‘‘Assessment of Lipid Oxidation and Off-Flavour Estimation of Chemical Parameters by Means of near-Infrared
Development in Meat and Meat Products’’. In: F. Shahidi (ed.) Flavor Spectroscopy’’. Meat Sci. 2013. 95(3): 621–628.
of Meat and Meat Products. New York, NY: Springer US, 1994, 83. M. De Marchi, C.L. Manuelian, S. Ton, et al. ‘‘Prediction of Sodium
pp.247–266. Content in Commercial Processed Meat Products Using near Infrared
65. J. Eggert, M. Belury, A. Kempa-Steczko, et al. ‘‘Effects of Conjugated Spectroscopy’’. Meat Sci. 2017. 125: 61–65.
Linoleic Acid on the Belly Firmness and Fatty Acid Composition of 84. M.S. Honeyman, R. Pirog, G. Huber, et al. ‘‘The United States Pork
Genetically Lean Pigs’’. J. Anim. Sci. 2001. 79(11): 2866–2872. Niche Market Phenomenon’’. J. Anim. Sci. 2006. 84(8): 2269–2275.
66. E. Gjerlaug-Enger, J. Kongsro, L. Aass, et al. ‘‘Prediction of Fat Quality 85. L.W. Mamani-Linares, C. Gallo, D. Alomar. ‘‘Identification of Cattle,
in Pig Carcasses by near-Infrared Spectroscopy’’. Anim. 2011. 5(11): Llama and Horse Meat by Near Infrared Reflectance or Transflectance
1829–1841. Spectroscopy’’. Meat Sci. 2012. 90(2): 378–385.
67. N. Prieto, B. Uttaro, C. Mapiye, et al. ‘‘Predicting Fat Quality from Pigs 86. E. Restaino, A. Fassio, D. Cozzolino. ‘‘Discrimination of Meat Patés
Fed Reduced-Oil Corn Dried Distillers Grains with Solubles by near According to the Animal Species by Means of near Infrared
Infrared Reflectance Spectroscopy: Fatty Acid Composition and Iodine Spectroscopy and Chemometrics Discriminación De Muestras De
Value’’. Meat Sci. 2014. 98(4): 585–590. Paté De Carne Según Tipo De Especie Mediante El Uso De La
68. M. Pérez-Juan, N.K. Afseth, J. González, et al. ‘‘Prediction of Fatty Acid Espectroscopia En El Infrarrojo Cercano Y La Quimiometria’’. CyTA
Composition Using a Nirs Fibre Optics Probe at Two Different – J. Food. 2011. 9(3): 210–213.
Locations of Ham Subcutaneous Fat’’. Food Res. Int. 2010. 43(5): 87. M. Schmutzler, A. Beganovic, G. Böhler, et al. ‘‘Methods for Detection
1416–1422. of Pork Adulteration in Veal Product Based on Ft-Nir Spectroscopy
69. E. Zamora-Rojas, A. Garrido-Varo, E. De Pedro-Sanz, et al. for Laboratory, Industrial and on-Site Analysis’’. Food Control. 2015.
‘‘Prediction of Fatty Acids Content in Pig Adipose Tissue by near 57: 258–267.
Infrared Spectroscopy: At-Line Versus in-Situ Analysis’’. Meat Sci. 88. J.M. Regenstein, M.M. Chaudry, C.E. Regenstein. ‘‘The Kosher and
2013. 95(3): 503–511. Halal Food Laws’’. Compr. Rev. Food Sci. Food Saf. 2003. 2(3):
70. N. Prieto, M.E.R. Dugan, M. Juárez, et al. ‘‘Real Time Prediction of 111–127.
Backfat Composition and Iodine Value by Portable Near Infrared 89. S. Sun, B. Guo, Y. Wei, et al. ‘‘Classification of Geographical Origins
Spectroscopy in a Diverse Population of Pigs’’. Poster presented at: and Prediction of
13c and
15n Values of Lamb Meat by near
69th Reciprocal Meat Conference. San Angelo, TX, USA; Jun 19–22 Infrared Reflectance Spectroscopy’’. Food Chem. 2012. 135(2):
2016. 508–514.
71. K.M. Sørensen, H. Petersen, S.B. Engelsen. ‘‘An On-Line Near-Infrared 90. N. Prieto, M. Juárez, I.L. Larsen, et al. ‘‘Rapid Discrimination of
(Nir) Transmission Method for Determining Depth Profiles of Fatty Enhanced Quality Pork by Visible and Near Infrared Spectroscopy’’.
Acid Composition and Iodine Value in Porcine Adipose Fat Tissue’’. Meat Sci. 2015. 110: 76–84.
Appl. Spectrosc. 2012. 66(2): 218–226. 91. E. Zamora-Rojas, D. Pérez-Marı́n, E. De Pedro-Sanz, et al. ‘‘In-Situ
72. M. Muller, M.R.L. Scheeder. ‘‘Determination of Fatty Acid Iberian Pig Carcass Classification Using a Micro-Electro-Mechanical
Composition and Consistency of Raw Pig Fat with near Infrared System (Mems)-Based near Infrared (Nir) Spectrometer’’. Meat Sci.
Spectroscopy’’. J. Near Infrared Spectrosc. 2008. 16: 305–309. 2012. 90(3): 636–642.
1426 Applied Spectroscopy 71(7)

92. N. Prieto, Ó. López-Campos, R.T. Zijlstra, et al. ‘‘Discrimination of 95. G. Downey. ‘‘Authentication of Food and Food Ingredients by Near
Beef Dark Cutters Using Visible and Near Infrared Reflectance Infrared Spectroscopy’’. J. Near Infrared Spectrosc. 1996. 4: 47–61.
Spectroscopy’’. Can. J. Anim. Sci. 2014. 94(3): 445–454. 96. L. Murray, I. Aucott, Pike. ‘‘Use of Discriminant Analysis on Visible and
93. M. Monroy, S. Prasher, M.O. Ngadi, et al. ‘‘Pork Meat Quality near Infrared Reflectance Spectra to Detect Adulteration of Fishmeal
Classification Using Visible/near-Infrared Spectroscopic Data’’. with Meat and Bone Meal’’. J. Near Infrared Spectrosc. 2001. 9: 297.
Biosyst. Eng. 2010. 107(3): 271–276. 97. T. Naes, T. Isaksson, T. Fearn, et al. A User Friendly Guide to
94. Q. Chen, J. Cai, X. Wan, et al. ‘‘Application of Linear/Non-Linear Multivariate Calibration and Classification. Chichester, UK: NIR
Classification Algorithms in Discrimination of Pork Storage Time Publications, 2002.
Using Fourier Transform near Infrared (Ft-Nir) Spectroscopy’’.
LWT- Food Sci. Technol. 2011. 44(10): 2053–2058.