foods

Review
Recent Advances in Portable and Handheld NIR Spectrometers
and Applications in Milk, Cheese and Dairy Powders
Yuanyuan Pu 1,2 , Dolores Pérez-Marín 2 , Norah O’Shea 1, * and Ana Garrido-Varo 2

1 Teagasc Food Research Centre, Food Chemistry and Technology Department, Moorepark, Fermoy, Co. Cork,
Ireland; Yuanyuan.Pu@teagasc.ie
2 Department of Animal Production, Faculty of Agriculture & Forestry Engineering, Campus Rabanales,
University of Cordoba, Nacional IV-Km 396, 14071 Cordoba, Spain; dcperez@uco.es (D.P.-M.);
pa1gavaa@uco.es (A.G.-V.)
* Correspondence: norah.oshea@teagasc.ie

Abstract: Quality and safety monitoring in the dairy industry is required to ensure products meet
a high-standard based on legislation and customer requirements. The need for non-destructive,
low-cost and user-friendly process analytical technologies, targeted at operators (as the end-users)
for routine product inspections is increasing. In recent years, the development and advances in
sensing technologies have led to miniaturisation of near infrared (NIR) spectrometers to a new era.
The new generation of miniaturised NIR analysers are designed as compact, small and lightweight
devices with a low cost, providing a strong capability for on-site or on-farm product measurements.
Applying portable and handheld NIR spectrometers in the dairy sector is increasing; however, little
information is currently available on these applications and instrument performance. As a result,





 this review focuses on recent developments of handheld and portable NIR devices and its latest
Citation: Pu, Y.; Pérez-Marín, D.;
applications in the field of dairy, including chemical composition, on-site quality detection, and safety
O’Shea, N.; Garrido-Varo, A. Recent assurance (i.e., adulteration) in milk, cheese and dairy powders. Comparison of model performance
Advances in Portable and Handheld between handheld and bench-top NIR spectrometers is also given. Lastly, challenges of current
NIR Spectrometers and Applications handheld/portable devices and future trends on implementing these devices in the dairy sector
in Milk, Cheese and Dairy Powders. is discussed.
Foods 2021, 10, 2377. https://
doi.org/10.3390/foods10102377 Keywords: near infrared spectroscopy; handheld sensor; on-site measurement; dairy composition;
milk; cheese; powder; authentication issues
Academic Editor:
Ana M. Vivar-Quintana

Received: 6 August 2021

1. Introduction
Accepted: 17 September 2021
Published: 8 October 2021 The consumption of dairy products is expected to increase world-wide, particularly
in developing countries where there is an increase in population as well as in family
Publisher’s Note: MDPI stays neutral incomes [1]. To meet the increasing demand for high quality dairy products, rapid mea-
with regard to jurisdictional claims in surements utilised for achieving greater process understanding, quality control and safety
published maps and institutional affil- assurance are required in every stage of dairy manufacturing, from on-farm milking, to
iations. production, and finally to the storage of end-products.
In the past decades, several destructive (i.e., solvent extraction, liquid and gas chro-
matographic techniques, rheological techniques) and non-destructive (i.e., near infrared,
mid-infrared, front face fluorescence spectroscopic techniques) analytical technologies
Copyright: © 2021 by the authors.
have been implemented in the dairy sector for determination of physicochemical param-
Licensee MDPI, Basel, Switzerland.
eters (i.e., protein, fat, viscosity) of various products [2–7]. Near infrared spectroscopy
This article is an open access article (NIRS) technology is one of the most promising spectroscopic analytical technologies in
distributed under the terms and the agri-food field [8], and has gained increasing interest in the dairy industry. Compared
conditions of the Creative Commons to traditional wet-chemistry for routine measurement of chemical composition, the advan-
Attribution (CC BY) license (https:// tages of NIRS technology include its rapid and non-destructive testing, minimal sample
creativecommons.org/licenses/by/ preparation, less labour intensive nature and no-chemical consumption [9].
4.0/).

Foods 2021, 10, 2377. https://doi.org/10.3390/foods10102377 https://www.mdpi.com/journal/foods

Foods 2021, 10, 2377 2 of 23

The application of NIRS in the dairy sector dates back to the late 1970s, where it was
used to measure low moisture products such as milk powders [10]. Rodriguez-Otero, et al.
reviewed the application of NIRS for the determination of major components (i.e., fat, pro-
tein, and moisture) of dairy products between 1978 and 1997 [11]. The authors pointed out
in the conclusion that the use of fibre optic probes in the production line had great potential
in future dairy applications. Giangiacomo and Cattaneo summarised the application of
NIRS in different areas along the dairy manufacturing chain, and highlighted some current
and potential applications of NIR-based analyses in milk coagulation process, powder
authenticity and dairy product classification [10]. A comprehensive review regarding the
application of NIR analysis on dairy products was written by Frankhuizen [9,12], providing
new updates on the use of NIR to measure major (i.e., moisture, fat, protein, lactose) and
minor (i.e., salt, pH, water-soluble primary amines) constituents in different dairy products
which included liquid milk, milk powders, casein and caseinates, butter and cheese. Hol-
royd summarised the application of NIRS in milk and milk products from 2008 to 2012 [13],
providing a table with the assignment of important NIR bands for different dairy products
(cheese, liquid milk, milk powder). Holroyd also summarised new trends and applications
of NIRS in the dairy industry [14], which included understanding the inter-instrument
variability, evaluating the performance of the same calibration being used across different
locations, and the non-targeted approach for quality assurance. More recently, De Marchi,
et al. reviewed the period from 2013 to 2017 [15], focusing on the difficulties to predict
chemical components such as fatty acids, minerals and volatile compounds, as well as
sensory attributes and ripening time in cheese.
From the reviews and publications mentioned above, NIRS has demonstrated its
feasibility for the analysis of a variety of dairy products. The number of publications related
to “portable or handheld NIR” has shown an increasing trend since 2006 (Figure 1i). Also,
the application of portable or handheld NIR in the “Food Science Technology” category
has the second greatest number of publication (Figure 1ii). However, within the “Food
Science Technology” category, the number of publications regarding to dairy applications
is still limited. Publications only begin to appear in 2017, as indicated in Table 1.
To the best of the author’s knowledge, information on the application of portable and
handheld NIR analysers in the dairy sector has yet to be reviewed. Therefore, this review
aims to provide an overview targeted at dairy processors, researchers, technologists and
engineers regarding portable and miniaturised NIR analysers and their potential in the
dairy industry. Driven by the concept of “precision dairy farming” and “quality by design
(QbD)”, transportable instruments i.e., small versions of laboratory NIR spectrometers [29],
are being increasingly used [30–33]. However, this review, except for comparative purposes,
will only focus on small-size and light-weight handheld devices. Firstly, the evolution of
handheld or portable NIR devices for agri-food applications is introduced. Secondly, the
state-of-the-art of handheld or portable NIR instrumentation reported in literature for dairy
applications is briefly described. Thirdly, the applications of NIRS in three major dairy
products, i.e., milk, cheese, and dairy powders is reviewed. Lastly, challenges and future
trends on the use of handheld and portable NIR devices in the dairy sector is discussed.
 DLP mm
technology
Weight: 35 g; Cheese
Consumer Size: Liquid milk
SCiO 740–1070 Reflectance - - - [25,27,28]
Physics (Israel) 3.15*9.5*27.5 (commercial
Foods 2021, 10, 2377 mm UHT milks) 3 of 23

(i)
200
Number of Publications

150
100
50
0
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
Foods 2021, 10, x FOR PEER REVIEW 4 of 21

Year

(ii)
210
Number of Publications

180
150
120
90
60
30
0

Web of Science Categories

Figure 1.1.Publication
Figure Publicationrecords from
records fromthethe
WebWeb of Science database,
of Science using the
database, search
using the words
search“NIR,
wordsportable” or “NIR, handheld”
“NIR, portable” or “NIR,
or “NIR, hand-held”
handheld” from 2000 from
or “NIR, hand-held” to 20202000(Data accessed
to 2020 on 29 October
(Data accessed 2020). (i)
on 29 October Number
2020). of publications
(i) Number in each
of publications in year.
each
(ii)
year.Number of publications
(ii) Number for thefor
of publications first
theten Web
first tenofWeb
Science categories.
of Science categories.

To the best of the author’s knowledge, information on the application of portable and
handheld NIR analysers in the dairy sector has yet to be reviewed. Therefore, this review
aims to provide an overview targeted at dairy processors, researchers, technologists and
engineers regarding portable and miniaturised NIR analysers and their potential in the
dairy industry. Driven by the concept of “precision dairy farming” and “quality by design
(QbD)”, transportable instruments i.e., small versions of laboratory NIR spectrometers
[29], are being increasingly used [30–33]. However, this review, except for comparative
purposes, will only focus on small-size and light-weight handheld devices. Firstly, the
evolution of handheld or portable NIR devices for agri-food applications is introduced.
Secondly, the state-of-the-art of handheld or portable NIR instrumentation reported in
literature for dairy applications is briefly described. Thirdly, the applications of NIRS in
three major dairy products, i.e., milk, cheese, and dairy powders is reviewed. Lastly,
challenges and future trends on the use of handheld and portable NIR devices in the dairy
sector is discussed.

2. Current Description of Handheld or Portable NIR Devices for Agri-Food

Applications
It is outside the scope of this review to describe the current technologies used for
portable spectroscopy. Several recent reviews cover in detail the technological
developments that have taken place in the last two decades in this area [29,34–37].
However, in the present review, it is important to briefly mention the main characteristics
of the portable spectrometers that have been scientifically evaluated in dairy products
(Table 1).
Several portable and on-site NIR instruments have been available since the late
1990ss and have been used mainly in the research domain for foods such as fruit and
Foods 2021, 10, 2377 4 of 23

Table 1. Specifications of handheld or portable NIR devices reported in literature with applications in the dairy sector.

Commercial Wavelength Measurement Light Wavelength Weight

Manufacturer Detector Dairy Applications Reference
Name (nm) Mode Source Selector and Size
Liquid Milk
Thermo Fisher Scientific Reflectance
MicroPhazir 1600–2400 Tungsten MEMS InGaAs Weight: 1.2 kg (raw milk from [16]
Inc. (USA) Transmittance
individual cows)
VIAVI Solutions Inc.
MicroNIR (USA), formerly known Two tungsten
1128–2162 Reflectance LVF InGaAs Weight: <60 g Milk powder [17]
2200 as JDS Uniphase lamps
Corporation, USA
Liquid milk
VIAVI Solutions Inc. Reflectance; Two integrated (pasteurized retail [18,19]
MicroNIR Weight: 64 g
(USA), formerly known 950–1650 Transmittance; vacuum LVF InGaAs milks, UHT milks)
1700/1700ES Size: 45 × 50 mm
as JDS Uniphase Transflectance tungsten lamps
Cheese [20]
Corporation, USA
Polychromix Inc. Liquid Milk
Phazir
(USA), sold to Thermo 1600–2400 Reflectance - MEMS InGaAs 1.7 kg (raw milk from [21]
1624
Fisher Scientific in 2010 individual cows)
Liquid milk
Spectral Engines Two tungsten Weight: 15 g;
NIRONE 1100–2500 Transmittance MEMS InGaAs (raw milk from [22]
(Finland) vacuum lamps Size: 25 × 25 × 17.5 mm
individual cows)
X-NIR Dinamica Generale 950–1800 Reflectance - - Weight: 1.6 kg; Cheese [23]
Liquid Milk
Si-Ware Systems Spectral resolution:16 nm; (raw milk from
NeoSpectra (Egypt/Europe/ 1350–2500 Reflectance Three lamps MEMS photodetector Weight: 17 g; individual cows; [24,25]
USA) Size: 32 × 32 × 22 mm commercial
UHT milks)
Based on Texas
Tungsten- Weight: 87 g;
NIR-S-G1 InnoSpectra Co., (Taiwan) 750–1700 Reflectance Instruments InGaAs Milk powder [26]
Halogen Size: 76 × 82 × 27 mm
DLP technology
Cheese

- - - Weight: 35 g; Liquid milk

SCiO Consumer Physics (Israel) 740–1070 Reflectance [25,27,28]
Size: 3.15 × 9.5 × 27.5 mm (commercial
UHT milks)
Foods 2021, 10, 2377 5 of 23

2. Current Description of Handheld or Portable NIR Devices for Agri-Food Applications

It is outside the scope of this review to describe the current technologies used for
portable spectroscopy. Several recent reviews cover in detail the technological develop-
ments that have taken place in the last two decades in this area [29,34–37]. However, in the
present review, it is important to briefly mention the main characteristics of the portable
spectrometers that have been scientifically evaluated in dairy products (Table 1).
Several portable and on-site NIR instruments have been available since the late 1990s
and have been used mainly in the research domain for foods such as fruit and vegetables.
However, these instruments did not achieve widespread adoption [38]. At the time,
these instruments had some limitations, for example, limited wavelength range (less than
1100 nm), low resolution, poor spectra reproducibility, limited optical window size, the
lack of a compartment cell for liquid samples, and the requirement to be connected to an
external computer.
Typically bench-top at-line NIR spectrometers are based on optical setups with diffrac-
tion grating. Therefore, the main disadvantage for on-site analysis is the presence of mobile
parts, their large size and the high price of the bench-top instrument. The strategies to
reduce the size of spectrometers are mainly done via the light splitting and detecting
components. By the first decade of the 21st century, advances in handheld and micro
NIR instrumentation have been made due to the rapid progress on sensing technologies
such as Linear Variable Filters (LVF) or micro-electro-mechanical systems (MEMS) and its
integration with micro-optics (MOEMS). The precise dimensions and alignment of MEMS
devices, combined with the mechanical stability that comes with miniaturisation, make
optical MEMS sensors well suited to a variety of challenging measurements [39].
Based on the type of detectors, portable spectrometers can be classified into two
categories (array-detector and single-detector instruments). For miniaturised NIR spec-
trometers, cost and power consumption are major drivers. Therefore, single element
detectors are preferred in miniaturised NIR spectrometers; however, the disadvantages of
this type of detector is that the spectra obtained are noisier than that obtained by standard
InGaAs detectors (with a 1700 nm cut-off), and the detector requires cooling [29,35,40,41].
A summary of the main specifications of handheld or portable NIR devices reported in
the literatures is listed in Table 1, including applications in the dairy sector. The following
paragraphs will briefly describe the main features of these instruments.

2.1. Phazir™ Instrument

In 2005, the Phazir™ handheld NIR spectrometer was launched by Polychromix (now
marketed by Thermo Fisher Scientific). This instrument was presented at the 2006 Pittcon
conference held in the USA and was claimed to be the first MEMS based handheld NIR
spectrometer [42]. Micro-electro-mechanical systems are small parts that integrate electrical
and mechanical components into a semiconductor chip for wavelength selection [43,44].
When MEMS is integrated with micro-optics, it is called a micro-opto-electro-mechanical
system (MOEMS). These are capable of sensing or manipulating optic signals at a chip-size
scale based on the combination of optical, electrical and mechanical technologies [39].
Phazir™ has a programmable MEMS chip, equipped with a fixed diffraction grating
and combined with a single detector and a digital transform spectrometer engine, which
allows a reduction in size and in the cost of the equipment. This architecture (Figure 2)
offers potential advantages over conventional detector-array-based designs, because the
use of digital-transform spectroscopy (DTS) not only improves the signal to noise ratio
associated with modulation of the light source, but also creates an inherent insensitivity to
stray light.
Phazir™ appeared on the market in two different configurations, covering 1100–1700 nm,
with a single element InGaAs detector, and a resolution of 6 nm per pixel, or covering
1600–2400 nm (Phazir1624), with a single element InGaAs detector, a resolution of 8 nm per
pixel, and with two-stage cooling [29]. Next the MicroPhazir™ was launched, and was only
available in the highest wavelength range. MicroPhazir™ consists of the same instrument
Foods 2021, 10, 2377 6 of 23

design as Phazir™, but with some differences, including a smaller size and lighter weight,
it contains an internal reference, and it has an information screen to display the operation
state of the device [45]. This instrument not only allows the user to scan solids but also
liquids using different adaptors (Figure 2ii). Two published papers on dairy products using
Foods 2021, 10, x FOR PEER REVIEW 6 of 21
the MicroPhazir™ were found in the literature (see Table 1). Despite the reduction in the
size of the device, this instrument is still heavy (approx. 1.2 kg) for a handheld analyser.

(i) (ii)

(iii) (iv)
Figure 2.
Figure 2. Handheld
Handheld MEMS Phazir™
Phazir spectrometer. (i) The sideview
TM
sideview ofof the Phazir
Phazir™ TM spectrometer;
spectrometer;
(ii) Microphazir TM with a liquid cuvette adaptor (images are originally from [46]); (iii) Schematic of
(ii) Microphazir™ with a liquid cuvette adaptor (images are originally from [46]); (iii) Schematic of a
a post-dispersive
post-dispersive instrumental
instrumental design
design incorporating
incorporating MEMSMEMS technology
technology (image
(image is originally
is originally from
from [46]);
[46] ); (iv) MEMS diffraction grating, multi-layer MEMS in reflective state (image is originally from
(iv) MEMS diffraction grating, multi-layer MEMS in reflective state (image is originally from [47–49]).
[47–49]).
2.2. The MicroNIR Spectrometer
2.2. The MicroNIR Spectrometer
A pocket-size NIR spectrometer called “MicroNIR” (developed and commercialised
previously by JDSU
A pocket-size NIR Corporation, Milpitas,
spectrometer calledCA, USA and currently
“MicroNIR” (developed by VIAVI Solutions Inc.,
and commercialised
Scottsdale, AZ, USA) was presented at the PittCon ® annual conference in 2012 [49]. The
previously by JDSU Corporation, Milpitas, CA, USA and currently by VIAVI Solutions
MicroNIR spectrometer
Inc., Scottsdale, AZ, USA) contains a light source,
was presented at the spectra
PittConcollection optics, electronics
® annual conference in 2012 and
[49].
detector in a package
The MicroNIR that weighs
spectrometer 58 g and
contains it is powered
a light by a USB
source, spectra connection.
collection The electronics
optics, MicroNIR
owes its smallin
and detector size and weight
a package thattoweighs
a thin-film
58 glinearly
and it isvariable
powered filter
by(LVF)
a USBtechnology
connection. as The
the
dispersive element, as shown in Figure 3. The LVF is a dielectric thin-film
MicroNIR owes its small size and weight to a thin-film linearly variable filter (LVF) bandpass filter
with a coating
technology wedged
as the in oneelement,
dispersive direction.asDepending
shown in Figureon the3.thickness
The LVFofisthe coating, only
a dielectric thin-
light
film bandpass filter with a coating wedged in one direction. Depending on athe
with a specific wavelength can pass through. The coating thickness in LVF can be
thickness
customised to select
of the coating, onlywavelength
light with range on-demand.
a specific wavelength can pass through. The coating
MicroNIR was first marketed with two different
thickness in a LVF can be customised to select wavelength configurations: covering the standard
range on-demand.
InGaAs detector was
MicroNIR wavelength range 950–1650
first marketed with two nm (MicroNIR 1700) as well as the
different configurations: extended
covering the
InGaAs detector wavelength range 1150–2150 nm (MicroNIR 2200). For both solutions,
standard InGaAs detector wavelength range 950–1650 nm (MicroNIR 1700) as well as the
the MicroNIR incorporates an uncooled detector. Due to power consumption limitations,
extended InGaAs detector wavelength range 1150–2150 nm (MicroNIR 2200). For both
the cooler alone would have far exceeded the 2.5 W allowable power usage of this USB-
solutions, the MicroNIR incorporates an uncooled detector. Due to power consumption
powered device [50,51]. A study undertaken by VIAVI Solutions confirmed that the average
limitations, the cooler alone would have far exceeded the 2.5 W allowable power usage of
noise level of the MicroNIR 2200 was 3–5 times higher than that of MicroNIR 1700 [52]. The
this USB-powered device [50,51]. A study undertaken by VIAVI Solutions confirmed that
original and the current MicroNIR spectrometers and their optical design and operating
the average noise level of the MicroNIR 2200 was 3–5 times higher than that of MicroNIR
principle are illustrated in Figure 3.
1700 [52]. The original and the current MicroNIR spectrometers and their optical design
and operating principle are illustrated in Figure 3.
 Foods
Foods 10,10,
2021,
2021, 2377PEER REVIEW
x FOR 7 of7 of
21 23

(i) (ii) (iii)

(iv) (v)
Figure 3. The
Figure 3. Thehandheld
handheldMicroNIR spectrometer.
MicroNIR (i) and
spectrometer. (i,ii)(ii)
TheThe MicroNIR
MicroNIR 1700
1700 spectrometerwith
spectrometer withand
and without a holder; (iii) The MicroNIR Pro and the transmission fixtures; (iv) Optical
without a holder; (iii) The MicroNIR Pro and the transmission fixtures; (iv) Optical design design of of
MicroNIR working in diffuse reflection and transmission modes; (v) Working principle
MicroNIR working in diffuse reflection and transmission modes; (v) Working principle of a linear of a linear
variable filter (LVF) component. (The images are originally from [50–52], copyright permission
variable filter (LVF) component. (The images are originally from [50–52], copyright permission has
has been granted).
been granted).

2.3. Other
2.3. Other Miniaturised
Miniaturised NIRNIRSpectrometers
Spectrometers
AnAn instrument
instrumentinitially
initiallydesigned
designedfor for forage analysis at
forage analysis ataafarm
farmlevel
levelisisthethe X-NIRTM
X-NIR™ (Di-
(Dinamica Generale, Milan, Italy). The NIR unit and the computation
namica Generale, Milan, Italy). The NIR unit and the computation unit of the spectrometer unit of the
spectrometer
are integrated areinto
integrated
a singleinto a single(in
instrument instrument (in which
total 1.6 kg) total 1.6
is kg) which
heavier is heavier
than than
MicroPhazir™
MicroPhazir TM (1.2
(1.2 kg). This kg). Thiscan
instrument instrument can betoconnected
be connected a computer to via
a computer
USB portvia orUSB
using port or
a WIFI
using a WIFI connection,
connection, and it iswith
and it is equipped equipped with a touchscreen
a touchscreen display for display for easy
easy access of allaccess of
functions
allusing fingertips.
functions using An application
fingertips. note on using
An application thison
note instrument
using thistoinstrument
predict drytomatter,
predictprotein
dry
and fat
matter, contentand
protein of cheese has been
fat content published
of cheese [23], details
has been publishedare shown in Table
[23], details are1.shown in
Table 1. Compared to an array detector, the price for a single detector is much lower and in an
attempt
Comparedto further
to anreduce the hardware
array detector, costs,for
the price new developments
a single detector is aremuch
focusedloweron and
systems
in
anwith singletodetectors
attempt (costs under
further reduce $5000). Therefore,
the hardware costs, newduring the past 5–6
developments are years
focused several
on
companies
systems withhave launched
single detectors NIR spectrometers
(costs under $5000). that Therefore,
fit into the during
palm ofthe a hand
past[29,36,37,53].
5–6 years
Four of
several these miniaturised
companies have launched instruments have been used
NIR spectrometers that for
fit research in dairy
into the palm of products.
a hand
They are the NeoSpectra-Micro, NIRONE, Innospectra NIR-S-G1 and
[29,36,37,53]. Four of these miniaturised instruments have been used for research in dairy SCiO (Table 1). The
four NIR spectrometers cover different spectral ranges between
products. They are the NeoSpectra-Micro, NIRONE, Innospectra NIR-S-G1 and SCiO 740 nm and 2500 nm and
differ in wavelength selection and detection element. Figure 4 provides
(Table 1). The four NIR spectrometers cover different spectral ranges between 740 nm and an image of the
instruments,
2500 nm and differthe most recent versions
in wavelength of them,
selection and also provides
and detection element. aFigure
schematic view ofan
4 provides the
optical
image features.
of the instruments, the most recent versions of them, and also provides a schematic
view of the optical features.
NIR-S-G1 uses Texas Instruments Digital Light Projector (DLP®) Technology. DLP
based spectrometers replace the traditional linear array detector with a Digital
Micromirror Device (DMD) for wavelength selection and a single point detector
[41,53,54]. Figure 4 displays the instruments and the schematic configuration for a DLP
and a single detector.
Foods 2021, 10, 2377 8 of 23
Foods 2021, 10, x FOR PEER REVIEW 8 of 21

(i) (ii) (iii) (iv)

(v) (vi) (vii) (viii)

(ix) (x) (xi)

FigureFigure 4. Commercially availablehandheld

handheld NIR NIRdevices. (i) DLP, the NIR
the Scan Nano optical optical[55]
Nano engine ; (ii) The NIR-
TM
4. Commercially available devices. (i) DLP, NIR Scan™ engine [55]; (ii) The
S-G1 sensor based on DLP technology [56] ; (iii) NeoSpectra Micro monolithic MEMS Michelson
NIR-S-G1 sensor based on DLP technology [56]; (iii) NeoSpectra Micro monolithic MEMS Michelson Interferometer; and Interferometer; and (iv)
NeoSpectra Scanner [57]; (v) NIRONE sensor and (vi NIRONE device (which contains the NIRONE sensor) [58]; (vii)
(iv) NeoSpectra Scanner [57]; (v) NIRONE sensor and (vi) NIRONE device (which contains the NIRONE sensor) [58];
SCiO spectrometer and (viii) SCiO cup [59]; (ix) Spectrometer using a DLP DMD and single element detector; and (x) DLP
(vii) SCiO spectrometer and (viii) SCiO cup [59]; (ix) Spectrometer using a DLP DMD and single element detector; and
DMD with a close-up of mirror array [54]; (xi) Michelson interferometer on a MEMS chip [60].(All images in Figure 4 are
(x) DLP DMD with
originally fromathe
close-up of mirror
above cited papers,array [54]; permission
copyright (xi) Michelson interferometer
has been on a MEMS
granted. Courtesy chip [60]. for
Texas Instruments (AllFigure
images in
Figure 4(i),
4 are originally
(ix), from the above cited papers, copyright permission has been granted. Courtesy Texas Instruments for
(x) and (xi)).
Figure 4i,ix–xi).
Fourier Transform (FT)-NIR spectrometers provide several advantages over classical
dispersive
NIR-S-G1spectrometers such as greater
uses Texas Instruments robustness
Digital for industrial
Light Projector (DLPuses® ) Technology.
and better DLP
resolution
based and spectra
spectrometers replace reproducibility. However,
the traditional linear array miniaturization
detector with a Digitalof a FT-NIR
Micromirror
spectrometer is more challenging. It is widely recognised that MEMS
Device (DMD) for wavelength selection and a single point detector [41,53,54]. Figure 4 technology advances
the development
displays the instruments of portable
and the NIRschematic
instrumentation. The firstfor
configuration MEMS
a DLP FT-NIR
and asensor,
single thedetector.
NeoSpectra, was put onto the market by Si-Ware Systems. This sensor uses a patented
Fourier Transform (FT)-NIR spectrometers provide several advantages over classical
MEMS chip (that consists of a Michelson interferometer) and a single element extended
dispersive spectrometers such as greater robustness for industrial uses and better resolution
InGaAs detector to implement a Fourier Transform approach between 1350 and 2500 nm
and[29].
spectra reproducibility. However, miniaturization of a FT-NIR spectrometer is more
Another FT-NIR sensor (The NIRONE Sensor) was launched by Spectral Engines
challenging. It istechnology
using a similar widely recognised
to that usedthat MEMSSystems.
by Si-ware technology advances
However, the development
the NIRONE sensor
of portable NIR instrumentation. The first MEMS FT-NIR
consists of a Fabry-Perot interferometer instead of a Michelson interferometer sensor, the NeoSpectra,
[58] for the was
putFourier
onto the market by
Transform. Si-Ware
Finally, the Systems. This sensor
SCiO (Consumer usesCity,
Physics, a patented
Israel) usesMEMS chip (that
a silicon
consists of a Michelson
photodiode array (PDA) interferometer)
detector sensitive andfrom
a single element
around 740 nm extended
to 1070 InGaAs
nm. It was detector
to implement
designed to be a Fourier Transform approach
used by non-NIR-experts between
and it can 1350using
be operated and 2500 nm [29].
an Android app Another
via
Bluetooth
FT-NIR [59].
sensor (The NIRONE Sensor) was launched by Spectral Engines using a similar
technology to that used by Si-ware Systems. However, the NIRONE sensor consists of
2.4. A Note about
a Fabry-Perot Software
interferometer instead of a Michelson interferometer [58] for the Fourier
MostFinally,
Transform. handheldtheand
SCiOportable NIR analysers
(Consumer Physics,come with software
Tel-Aviv, which
Israel) uses allows simple
a silicon photodiode
data treatments such as the spectra plotted, use of first and second derivative for
array (PDA) detector sensitive from around 740 nm to 1070 nm. It was designed to be used spectra
preprocessing, PLS regression and prediction of unknown samples. However, in order to
by non-NIR-experts and it can be operated using an Android app via Bluetooth [59].
develop robust models (particularly in high moisture materials), other algorithms (i.e., use
2.4. A Note about Software
Most handheld and portable NIR analysers come with software which allows simple
data treatments such as the spectra plotted, use of first and second derivative for spectra
preprocessing, PLS regression and prediction of unknown samples. However, in order
to develop robust models (particularly in high moisture materials), other algorithms (i.e.,
Foods 2021, 10, 2377 9 of 23

use of repeatability files, non-linear regression methods, pattern recognition methods,

cloning/standardization) are highly recommended.
It should be noted that these software packages allow spectral data to be exported
to other chemometrics or multivariate data analysis software (i.e., UNSCRAMBLER, PLS
Toolbox, MATLAB, etc.). This is a good option for scientists to explore other meth-
ods/algorithms to improve model performance. However, in real applications, a problem
arises as one may not be able to import the MATLAB model back into the portable instru-
ment software. In addition, chemometric software i.e., UNSCRAMBLER or MATLAB are
expensive software packages, which are not feasible to be included with low-cost portable
instruments. Some handheld NIR analyser providers (i.e., SCIO, TELLSPEC) offer cloud
computing services, which enables the spectra to be saved in the cloud for data processing
later using tools already available on the provider’s website. This cloud-based solution
is of limited value for developing a robust model for real world applications. Another
problem associated with this type of handheld NIR instrument is the cost to protect the
data (i.e., spectral data, reference data, calibration models that latter can be used in a real
life application) which can be more expensive than a portable instrument itself. Therefore,
data security (i.e., ownership of stored data in the cloud) and data protection (i.e., data
protected from being exploited by the handheld NIR analyser providers for use in other
applications) should be carefully considered.
Another option offered to consumers is the possibility to predict unknown samples by
using the models already incorporated on the app. In the opinion of the authors, this is not
advisable as NIRS will always bring up a prediction value, however, it is difficult to know
the reliability of the prediction. For example, NIR scientists and practitioners use solid
multivariate statistics (i.e., based in Mahalanobis distance) to identify if the spectrum of an
unknown sample falls into the calibration population. These type of data quality checks
should be included as part of cloud computing services offered by these handheld NIR
analyser providers.

3. Applications in the Dairy Sector

Applying portable and handheld NIR analysers in the dairy sector is very promising,
as they can provide more flexibility to dairy operators for quick quality inspection of
the milk supply to final products. The use of portable and handheld NIR devices in the
pharmaceutical [61] or agri-food (i.e., fruits and vegetables, meats, feedstock) areas has been
previously reviewed [35,44]. In the work of dos Santos, Lopo, Páscoa and Lopes [44], the
authors summarised the applications of portable NIR spectrometers in the food industry
including dairy product analysis, however only three research studies were mentioned in
the review which were related to dairy (milk).
The application of portable and handheld NIR analysers in the dairy sector has become
a hot research topic in recent years. Since most of the current portable and handheld
NIR devices are not specifically designed for dairy applications, the studies and research
available in literature are mainly focused on evaluating NIR performance and exploring its
potential applications in the dairy industry.

3.1. Milk
Dairy products such as yogurt, cream, cheese and milk powders are made from
milk. As a result, raw milk quality and safety play an important role in the production of
other milk-based products. This section of the review introduces the use of portable and
handheld NIR spectrometers for measuring major and minor chemical constituents of milk
as well as to in the detection of milk adulteration.

3.1.1. Major Milk Composition

Kalinin, et al. proposed the use of a portable NIR spectrophotometer (BIKAN-K,
weight of 4.6 kg, measuring in transflectance mode in the wavelength range of 800–1080 nm)
for the measurement of fat, total protein and lactose in liquid milk [62]. Sixty-five milk
Foods 2021, 10, 2377 10 of 23

samples were prepared by mixing several ingredients (i.e., cream, skim milk, lactose)
followed by a homogenisation and a standardisation process, resulting in a milk fat content
ranging from 1.5–5.15%, total protein 2.2–4.0% and lactose 4.0–5.8%. Calibration models
developed using partial least square regression (PLSR) gave a correlation coefficient of
calibration (r2 c ) of 0.984, 0.962 and 0.878 for fat, protein and lactose, with a root mean
square error for cross validation (RMSECV) of 0.078%, 0.080% and 0.081%, respectively.
The authors evaluated the performance of a two-channel portable NIR spectrometer for
predicting fat, casein and whey protein in liquid milk [63]. One channel in the spectrometer
was used for the acquisition of the transmission spectra and the other channel was used
for acquisition of backscattering spectra for each sample. PLSR calibration models based
on transmission spectra/backscattering spectra only, or the combination of both were
developed. This study demonstrated that prediction models using the combined spectra
gave the best result, having a correlation coefficient of prediction (r2 p ) of 0.88, 0.89 and 0.91
for fat, casein and whey protein, with a root mean square error of prediction (RMSEP) of
0.08%, 0.13% and 0.07%, respectively.
A handheld MicroPhazir NIR device (Thermo Fisher Scientific., Waltham, MA, USA)
was evaluated to develop calibration models for protein, fat and solids-non-fat (SNF) of raw
milk [16]. The MicroPhazir spectrometer had a scanning window 4 mm in diameter, which
results in a sampling area of 0.13 cm2 . A liquid adaptor was required for spectral acquisition
of liquid milk using this device. Two types of cuvettes, C1 (a reusable quartz cuvette with
a 1 mm pathlength for transmission measurement) and C17 (a reusable quartz cuvette
with a 17 mm pathlength with an aluminium attached to the other side of the cuvette for
transflectance measurement), were assayed in the study to evaluate the spectra quality. The
authors confirmed that using the C17 liquid adapter for sampling and setting 80 scans per
sample provided the best spectral repeatability and reproducibility. Calibration models
based on the modified partial least square (MPLS) method were developed on 482 raw
milk samples (444 milk samples as a calibration set and 38 milk samples as an external
validation set). The models had a coefficient of determination for calibration (R2 c) of 0.97,
0.76 and 0.61 for fat, protein and SNF, and a standard error of prediction (SEP) of 0.126%,
0.124% and 0.221%. The authors also evaluated the possibility of transferring the above
calibration models to another MicroPhazir device using 10 selected milk samples to build a
standardization procedure. Studies demonstrated the capability of sharing calibration data
(through a simple calibration transfer procedure) between different handheld MicroPhazir
analysers [16].
Four different handheld NIR sensors (NIRONE 1.4, NIRONE 2.0, NIRONE 2.5 in
transmission mode and NIRONE 2.0 in reflectance mode, from Spectral Engines, Finland)
for measuring fat, protein and lactose of 252 raw milk samples were tested [22]. The PLSR
models developed had a prediction error of 0.54–0.58% for fat, 0.30–0.35% for protein
and 0.16–0.19% for lactose. The prediction error of NIRONE sensors were compared to
international ICAR (The Global Standard for Livestock Data) recommendations for on farm
milk analysers (i.e., Tec5 cooled InGaAs spectrometer, AFI milk on-line analyser). The data
indicated that NIRONE sensors had a higher prediction accuracy for fat measurement but
a lower prediction accuracy for protein and lactose. Amr, et al. investigated the use of a
handheld MEMS-based Fourier Transform NIR spectrometer to monitor the fat content in
milk [64]. Milk samples with different fat concentrations (0.1%–6.5%) were prepared by
mixing skimmed milk with full cream. The spectrometer was operated in transmission
mode within a wavelength range of 1300–2500 nm, the milk sample were inserted into
a 1 mm pathlength quart cuvette for acquisition of transmission spectra. A calibration
model was developed based on the principal component regression (PCR) method, giving
a maximum error in concentration prediction of 0.5%. The authors highlighted that the
result was acceptable and could be used to determine the categorisation of milk type, i.e.,
skimmed, low fat, medium fat or full cream milk.
Two miniaturised spectrometers (SCiO and NeoSpectra) were compared for the analy-
sis of commercial milk purchased from Spain, Italy and Switzerland [25]. Both handheld
Foods 2021, 10, 2377 11 of 23

spectrometers could provide a rapid and reliable analysis for the prediction of fat content
(in the range of 0.1–3.7%) and milk sample classification (i.e., skimmed, semi-skimmed and
whole milk). Since the SCiO (740–1070 nm) and NeoSpectra (1350–2558 nm) cover different
spectral regions, a data-fusion PLS model was developed using the spectral data from both
sensors, which provided a better fat prediction result than the PLS models developed using
each individual sensor.

3.1.2. Minor Milk Composition

Other components in milk, for example, fatty acid profiles, were evaluated using a
handheld NIR device (Phazir 1624, 1600–2400 nm) with an opaque liquid cup for spec-
tral acquisition in transflectance mode [21]. Reference data for the fatty acids profile of
108 raw milk samples were determined by gas chromatography mass spectrometry (GC-
MS) method. Calibration was carried out by combining spectral pre-treatments and PLSR
modelling. Good results were obtained for predicting polyunsaturated fatty acid (PUFA),
monounsaturated fatty acid (MUFA), linoleic acid and caproic acid, with a coefficient
of determination for prediction (R2 p ) that varied between 0.87–0.92, demonstrating the
potential of using handheld NIR devices for rapid on-site monitoring of these fatty acids in
raw milk at farm level.

3.1.3. Milk Adulteration

Besides chemical composition measurements, handheld NIR analysers were also used
to evaluate the capability of detecting milk adulterants. Santos, et al. utilised a handheld
NIR device (MicroPhazir, Thermo Fisher Scientific, Waltham, MA, USA) to detect bovine
milk (purchased from a local supermarket) and milk spiked with 6 different adulterants
(i.e., tap water, whey, hydrogen peroxide, synthetic urine, urea and synthetic milk) at
different concentrations ranging from 3 to 50% v/v [65]. Soft independent modelling of
class analogy (SIMCA) method was used to classify the control milk (no adulteration) and
adulterated milk samples. It was demonstrated from the results that MicroPhazir provided
a correct classification rate (CCR) of 0% for control milk and a CCR of 56% for adulterated
milk. These results were compared to a handheld mid infrared (MIR) device (4200 Flex
scan, Agilent Technologies Inc., Santa Clara, CA, USA) and a portable MIR system (Cary
630 FTIR spectrometer, Agilent Technologies Inc., USA), showing that both MIR devices
provided better performance than the MicroPhazir NIR device (R2 c = 0.92) for classification
and quantification of milk adulterants.
A low-cost digital NIR photometer prototype was proposed by Moreira, et al. for the
detection of milk adulteration with water [66]. This prototype was a portable and micro-
controlled device, which used three LEDs focused at three wavelengths (970 nm, 1200 nm
and 1450 nm); the chosen wavelengths are related to the absorption of water molecules.
The diluted milk samples contained 0–25% added water. To measure the percentage of
added water, a linear relationship was found between the amplitude of the output signals
at each wavelength and the amount of added water, with a mean absolute error of <1%.
Liu, et al. compared the performance of a handheld NIR analyser (MicroNIR 1700, VIAVI
Solution Inc., Scottsdale, AZ, USA) with a benchtop FT-NIR spectrometer (NIRFlex N-500,
Buchi, Flawil, Switzerland) for authentication of organic (37 milk samples) and non-organic
(50 milk samples including 36 conventional milks and 14 pasture milks) milk samples
bought from a local supermarkets [18]. Partial least square discriminant analysis (PLSDA)
method was used for milk classification (organic vs. non-organic). A similar classification
rate was obtained using the handheld device (CCR = 73%) and the benchtop instrument
(CCR = 78%). Both instruments gave a CCR of 89% for differentiation of the 37 organic and
36 conventional milks. However, based on the fatty acids in the milk samples obtained by
the gas chromatography method, the PLSDA classification model had a CCR = 100% to
classify organic and non-organic milks.
Similarly, de Lima, et al. investigated the same handheld NIR analyser (MicroNIR
1700) and a benchtop FT-NIR instrument (Perkin Elmer, City, USA) to classify 41 lactose-
Foods 2021, 10, 2377 12 of 23

free ultra-high-temperature (UHT) milk samples (two skimmed milk, 33 semi-skimmed

milk and 6 whole milk) with 30 regular UHT milk samples (including six skimmed milk,
six semi-skimmed milk and 18 whole milk) [19]. Classification models developed using
PLSDA or genetic algorithm linear discriminant analysis (GA-LDA) gave a CCR of 100%,
for both the handheld and benchtop NIR instruments, indicating the feasibility of using
the ultra-compact handheld NIR device for a quick and precise discrimination between
regular and lactose-free milk on-site or in the field.

3.2. Cheese
Cheese can be classified as fresh cheese (unripened cheese) and aged cheese (ripened
cheese). Most of the cheeses around the world are ripened. During the cheese maturation
process, changes in several physical and chemical properties (i.e., composition, pH, tex-
ture and flavour) occur [67]. As a result, real-time monitoring of cheese compositions is
important to ensure final cheese quality [68].
In recent years reports and publications on the use of handheld NIR analysers in
cheese applications have become available. Stocco, et al. compared the accuracy and biases
of a portable NIR spectrometer (LabSpec2500, ASD Inc., Boulder, CO, USA) with two
different benchtop NIR spectrometers (NIRSystem 5000, FOSS Analytical A/S, Denmark,
1100–2498 nm, measuring in reflectance mode; FoodScan, FOSS Analytical A/S, Denmark,
850–1048 nm, measuring in transmittance mode) for predicting chemical attributes (i.e., dry
matter, ash, protein, lipids, water-soluble nitrogen), pH, texture (hardness and working
shear force) and the colour of 37 different types of cheese [69]. This research demonstrated
that all three NIR instruments had a good prediction performance for the chemical compo-
sition, while they were less accurate for pH and texture parameters. The authors also found
that the portable one performed better in predicting all the quality attributes compared to
the other two benchtop instruments; this is mainly due to the instrument technology of the
handheld spectrometer used.
Similarly, three different NIR instruments were evaluated for the determination of fat
and dry matter of 160 fresh Swiss cheese blocks (size of 35 × 28 × 12 cm, weight of 12 kg)
during the cheese manufacturing process [20]. These instruments were (A) a handheld
MicroNIR 1700 (VIAVI Solutions Inc, Scottsdale, AZ, USA) operating in reflectance mode,
with a spectral range of 908–1676 nm and a spectral resolution of 6.2 nm, (B) a prototype
NIR instrument operating in interaction mode (the light can penetrate in depth down to
1–2 cm), with a spectral range of 760–1040 nm and a spectral resolution of 20 nm, (C) a
NIR imaging system (QVision 500, TOMRA Sorting Solutions, Belgium) which collected
15 spectral images from 760 to 1040 nm, with a spectral resolution of 20 nm. Calibration
models were developed using PLSR. Results indicated that both NIR instrument (A) and
(B) performed equally well for predicting the dry matter content of cheese. However,
instrument (A) had a higher prediction accuracy for fat content in comparison to the
instrument (B). Instrument (C) had the lowest prediction accuracy for both parameters (dry
matter and fat), possibly due to the lower signal-to-noise ratio of the system.
Ma, Babu and Amamcharla evaluated a low-cost handheld NIR device (SCiO, Con-
sumer Physics, Israel) for predicting total protein and intact casein in 49 Cheddar cheeses
(35 samples as calibration and 14 samples as validation) using PLSR [28]. The intact
casein ranged from 14 to 23 g/100 g of cheese and the total protein ranged from 20 to
25 g/100 g of cheese. Different spectral pre-processing and wavelength selection strategies
were applied to improve model performance. Results showed that all models had an
RMSEP of 0.91–1.58 g/100 g of cheese for intact casein prediction, while an RMSEP of
0.62–0.88 g/100 g of cheese for total protein prediction. This study indicated that the
low-cost SCiO sensor had the potential for rapid and on-site quantification of intact casein
and total protein in cheddar cheese. Verena Wiedemair utilised the same NIR sensor (SCiO)
to evaluate its performance for the prediction of fat and moisture content of 46 cheese
samples [27]. The cheese samples were also grated in order to investigate the impact of
cheese physical status (whole pieces versus grated cheese) using a NIR measurement.
Foods 2021, 10, 2377 13 of 23

The same samples were also measured using a benchtop NIR instrument (NIRFlex N-
500, Buchi, Flawil, Switzerland) for comparison purposes. Results demonstrated that
the PLSR calibration models developed had R2 p values over 0.93. For fat content, both
instruments had a similar prediction accuracy (R2 p ~0.99 and RMSEP ~0.08%) for grated
cheese. However, the SCiO sensor (R2 p = 0.98, RMSEP = 1.19%) performed better than the
benchtop instrument (R2 p = 0.94, RMSEP = 1.90%) for the prediction of fat in whole pieces
of cheese, as the whole pieces of cheese can be spatially inhomogeneous. For moisture
content, both instruments gave a similar prediction accuracy for whole cheese, while the
benchtop instrument (R2 p = 0.96, RMSEP = 0.93%) performed better than the SCiO device
(R2 p = 0.93, RMSEP = 1.71%) for the prediction of moisture in grated cheese. Marinoni, et al.
purchased 12 portable NIR spectrometers (XNIR, Dinamica Generale, Italy) for evaluating
the prediction performance of dry matter (in the range of 64.18–70.16%), fat (in the range of
23.98–32.90%), protein (in the range of 29.29–36.24%), fat to dry matter ratio (in the range
of 35.85–48.71), and protein to dry matter ratio (in the range of 43.36–54.20) of Italian hard
cheese (Grana Padano) [23]. A total number of 195 slices of protected designation of origin
(PDO) Grana Padano cheese that had 6–13 months ripening and were an average weight
of 4 kg, were sampled by the consortium from several dairies located in the Po Valley.
Samples were scanned with a portable instrument applied directly on the entire slice. After
the removal of the rind (first 6 mm from the outside), the cheese was ground and scanned
with a benchtop FT-NIR NIRFlex N-500 (Buchi Italia, Cornaredo, Italy). Calibration models
were developed using a training set of 116 samples and evaluated with a validation set
of 74 samples. The spectra were also acquired at three different temperatures (10, 16 and
25 ◦ C) to minimise the effect of temperature on the model performance. Calibration models
based on the cheese paste spectra had a good prediction accuracy for fat, protein, fat to dry
matter ratio, and protein to dry matter ratio, with an R2 P of 0.907, 0.832, 0.902, 0.870 and an
RMSEP of 0.461%, 0.396%, 0.602, 0.580. The results were comparable to those obtained with
a benchtop FT-NIR instrument (RMSEP = 0.54%, 0.49%, 0.67, 0.47 for fat, protein, fat-to-dry
matter ratio, protein-to-dry matter ratio). However, the PLSR model yielded a relatively
poor performance for dry matter (R2 p = 0.616 and RMSEP = 0.654%), possibly because there
is a moisture gradient inside the cheese which affects the sample homogeneity that could
have an impact on the performance of the calibration model. Predictive models based on
the cheese rind spectra were built for fat to dry matter ratio and protein to dry matter ratio,
which had a R2 P value close to 0.6. Nevertheless, this study proved the use of small and
cost-effective portable NIR devices for predicting cheese composition at batch level, which
is important for small scale factories who produce Grana Padano cheese.

3.3. Dairy Powders

Dairy powders including skimmed milk powder, whole milk powder and milk/whey
protein powders can be used as ingredients in baking, recombined food products and infant
formula [70]. The quality and safety of dairy powders should be monitored on a regular
basis to ensure they have a consistent composition and functionality and meet a high safety
standard, especially for infant formula production. The NIRS technology in dairy powders
is mainly used for compositional measurements and powder adulteration detection.

3.3.1. Powder Composition

Kalinin et al. applied a portable NIR device (BIKAN-CP, equipped with a fibre-
optic and heat-resistant probe) to measure moisture, fat and protein of a dried milk mix-
ture [62]. The NIR device operated in reflectance mode and in the wavelength region of
1050–1670 nm. The optimum PLSR models obtained had a correlation coefficient value of
0.945, 0.981, 0.842 and a RMSECV value of 0.25%, 1.87% and 3.62% for moisture, protein
and fat, respectively. Another type of portable NIR instrument (FieldSpec Pro, Analytical
Spectral Devices, Inc., Boulder, CO, USA) was studied by Wu, et al. [71], to investigate
its performance for the simultaneously measurement of fat, protein and carbohydrate in
350 infant milk powder samples, using the short-wave NIR with the spectral region of
Foods 2021, 10, 2377 14 of 23

800–1050 nm. Spectral pre-treatments combined with PLSR or least squares-support vector
machine (LS-SVM) were employed for model development. The models yielded good
prediction accuracy, with an R2 p of 0.98 achieved for the three components. The authors
also investigated the wavelength assignments for fat, protein and carbohydrates based
on the loading weights and regression coefficients derived from the models. For example,
the wavelength 861 nm and 897 nm were assigned to the third overtone of C-H stretching
vibration of carbohydrate; the wavelength 968 nm was assigned to the second overtone
of O-H stretching present in water; and the wavelength of 1033 nm was assigned to the
second overtone of N-H stretching present in fat.
Kong, et al. utilised the same NIR device (FieldSpec Pro) but using different wave-
length regions (350–1075 nm) to detect irradiation doses of irradiated milk powders, as
irradiation technology are commonly used to destroy microorganisms in foods [72]. How-
ever, the use of gamma-irradiation may have an effect on protein functionality due to
possible oxidation of amino acids and formation of protein free radicals. A total of 150 milk
powders were irradiated with gamma-rays at different doses (0–6 kGy) and stored in a
controlled environment (temperature = 25 ± 1 ◦ C, humidity = 70%) for seven days prior
to the NIR spectra being taken. The best model obtained in the study had a correlation
coefficient of 0.97 and an RMSEP of 0.844 kGy, indicating the feasibility of portable NIR
devices in quantification of irradiation dose in milk powders.

3.3.2. Powder Adulteration

Studies on evaluating the use and sensitivity of NIRS integrated with chemometrics
for rapid determination of melamine and other possible adulterants in milk powders has
increased since the milk scandal incident that occurred in China in 2008 [73–75].
Henn, et al. compared two miniaturised NIR spectrometers, a MicroNIR 2200 (from
VIAVI Solutions, Scottsdale, AZ, USA) and a MicroPhazir (from Thermo Fisher Scientific,
Waltham, MA, USA) with a benchtop NIR instrument (NIRFlex N-500, Buchi, Flawil,
Switzerland) for melamine detection in milk powder (infant formula) [17]. The adulterated
milk powders were prepared by adding pure melamine (purity ≥ 99%) into off-the-shelf
infant formula powders, to reach a final adulteration concentration from 0% to 5.5% which
was added at 0.5% increments. The best PLSR calibration model achieved by each instru-
ment had a similar RMSEP value of 0.28 g/100 g (melamine/milk powder), 0.33 g/100 g
and 0.27 g/100 g for the NIRFlex N-500, the MicroPhazir and the MicroNIR 2200 instrument.
The authors further calculated the limit of detection (LOD) interval [LODmin , LODmax ]
for the three spectrometers using the calculation proposed by Allegrini and Olivieri [76].
The three instruments had a LOD interval of [0.20, 0.30] g/100g, [0.28, 0.54] g/100 g and
[0.44, 1.15] g/100 g, indicating that the benchtop instrument was more sensitive for de-
tecting low concentrations of components like melamine, and the benchtop spectrometer
had a wider wavelength region and a more sophisticated instrument design. The au-
thors highlighted that additional information (i.e., the LOD interval) should be taken into
consideration when evaluating the robustness of models.
Karunathilaka, et al. investigated two benchtop FT-NIR spectrometers (from the man-
ufacturers Bruker and PerkinElmer) and a handheld NIR device (Phazir 1624, Polychromix
Inc., Wilmington, MA, USA) for non-targeted detection of contaminations in commercial
non-fat-dried-milk powders [77]. Eleven potential adulterants were considered in the
study, which can be categorised into four groups: (1) low molecular weight, nitrogen-rich
compounds (i.e., melamine); (2) plant-based proteins (i.e., soy protein); (3) inorganic salts
(i.e., calcium carbonate); and (4) non-fat solids (i.e., sucrose). Classification models based
on the SIMCA method demonstrated that the benchtop FT-NIR instruments yielded a
higher sensitivity and specificity for authentication of milk powder when compared to
the portable device, as the portable device had a narrower spectral range and a lower
spectral resolution.
Zinia Zaukuu, et al. compared a benchtop NIR spectrometer (MetriNIR, Development
and Service Co., Hungary, wavelength region of 750–1700 nm with a spectral resolution
Foods 2021, 10, 2377 15 of 23

of 2 nm) and a handheld spectrometer (NIR-S-G1, InnoSpectra Co., Taiwan, wavelength

region of 700–1700 nm with a spectral resolution of 3 nm) to detect low concentration of
nitrogen-based adulterants in whey protein powder [26]. Using four different adulterants,
urea (U), melamine (M), glycine (G) and taurine (T), 15 types of adulterated whey protein
powders were prepared, by adding one single adulterant (U, G, T, M), two adulterants (GT,
GU, GM, TU, TM, UM) or more than two adulterants (GTU, GTM, GUM, TUM, GTUM)
to the whey protein powders. The final adulteration level was from 0.5 to 3 (%, w/w) with
an increment of 0.5%. Three sets of NIR spectra were acquired, dataset 1 was obtained
by placing the powder samples in an optical glass cuvette and scanned by the benchtop
instrument; dataset 2 was obtained by placing the powder samples in an optical glass
cuvette and scanned using the handheld device; dataset 3 was obtained by placing powder
samples in a low density polyethylene zip-lock bag and scanned using the handheld device.
PLSR was used to quantify the adulteration concentration and linear discriminant analysis
(LDA) was used to classify adulterants in different scenarios. Overall, the benchtop NIR in-
strument performed the best in predicting these nitrogen-based adulterants. The handheld
instrument was not sensitive in detecting the lowest concentration (0.5%). However, for
other concentrations, reliable results were achieved, proving the advantages of handheld
devices for on-site quality screening of whey protein powders though glass cuvettes or
plastic bags.

4. Comparison between Handheld and Benchtop Instruments

There are many factors which affect NIR model performance. Williams mentioned
three different error sources [78], that is, those related to the instrument (i.e., signal-to-
nose-ratio, wavelength accuracy), sample (i.e., granulometry, affected by temperature
fluctuations) and operational (i.e., sample preparation, high errors in the reference data,
chemometric method used). Therefore, one of the difficulties when evaluating portable
instruments, is the correct interpretation of model performance, in particular, what are
the reasons or factors that cause the differences in the models between portable and
benchtop instruments.
Some authors have tried to overcome this by developing in parallel (on a portable
instrument and on a benchtop instrument) predictive models with identical samples. In the
case of dairy products, most studies have been done using a low number of samples and
without proper validation. As a result, the interpretation of the results is sometimes very
optimistic, and therefore some publications suggest that portable devices have a similar
performance to benchtop instruments.
Table 2 summarises the research applications where a comparison between both
portable and benchtop instruments were carried out using dairy products. Liu, et al. [18]
and de Lima, et al. [19] compared the MicroNIR 1700 with two different bench top in-
struments, to evaluate the ability to qualitatively analyse liquid milk, for two different
authentication issues (organic vs. non-organic, lactose free vs. regular milk containing
lactose). Both studies concluded that the models developed with the MicroNIR 1700
have a CCR similar to those obtained with the benchtop instruments. However, in both
cases the sample type (bought from the market) and the low number of samples used
make it difficult to believe that the models developed are sufficiently robust to lead to
this this conclusion. For cheese studies, researchers found that the MicroNIR 1700 and
the SCiO portable devices had a similar prediction accuracy to the NIR-Flex bench top
instrument for predicting macro composition (i.e., dry matter, protein, fat) of several types
of cheese [20,23,27]. According to the studies reported on milk powders, current handheld
devices are not as sensitive as benchtop NIR instruments in detecting adulterants in milk
powders in terms of prediction accuracy [17,26,77]. It is also worth noting that based on
the experimental design presented in the mentioned studies, it is not possible to form firm
conclusions as not all the information is given for a more rigorous analysis of the data.
Foods 2021, 10, 2377 16 of 23

Table 2. Comparison of model performance between handheld/portable NIR instruments and benchtop NIR instruments in the dairy products.

Research Objective, Handheld/Portable NIR Spectrometer Benchtop NIR Spectrometer

Reference Predictive Models,
Number and Type of Samples Instrument Multivariate Model Performance Instrument Multivariate Model Performance
MicroNIR 1700
Organic milk authentication, NIRFlex N-500 (Buchi,
(VIAVI Solution Inc.,
[18] N = 37 organic milk and N = 50 PLS-DA CCR = 73% Flawil, Switzerland) PLS-DA CCR = 78%
Scottsdale, AZ, USA)
non-organic retail milks 1000–2500 nm
908–1670 nm
Lactose-free milk authentication, MicroNIR 1700 PLS-DA CCR = 100% FT-NIR Spectrum PLS-DA CCR = 100%
N = 30 regular UHT milk (VIAVI Solution Inc., Frontier (Perkin
[19] SPA-LDA CCR = 80% SPA-LDA CCR = 100%
samples and N = 41 lactose-free Scottsdale, AZ, USA) Elmer, USA)
UHT milks 908–1670 nm GL-LDA CCR = 100% 833–2500 nm GL-LDA CCR = 100%
Moisture: Moisture:
R2 CV = 0.96 R2 CV = 0.83
RMSECV = 2.10% RMSECV = 4.51%
Bayesian regression Protein: Bayesian regression Protein:
(whole spectral R2 CV = 0.88 (whole spectral R2 CV = 0.81
range,350–1830 nm) RMSECV = 1.77% range, 1100–2498 nm) RMSECV = 2.25%
Lipids: Lipids:
R2 CV = 0.85 NIRsystems5000 R2 CV = 0.67
LabSpec2500 (ASD RMSECV = 2.03% (FOSS, Denmark) RMSECV = 3.20%
Moisture, protein, lipids;
[69] Inc., USA) 1100–2498 nm,
N = 197 cheese samples Moisture: Moisture:
350–1830 nm in reflectance;
R2 CV = 0.96 R2 CV = 0.84
grinding the sample
RMSECV = 2.00% RMSECV = 4.39%
Bayesian regression Protein: Bayesian regression Protein:
(common spectral R2 CV = 0.91 (common spectral R2 CV = 0.76
range,1100–1830 nm) RMSECV = 1.59% range, 1100–1830 nm) RMSECV = 2.51%
Lipids: Lipids:
R2 CV = 0.85 R2 CV = 0.69
RMSECV = 2.03% RMSECV = 3.11%
Foods 2021, 10, 2377 17 of 23

Table 2. Cont.

Research Objective, Handheld/Portable NIR Spectrometer Benchtop NIR Spectrometer

Reference Predictive Models,
Number and Type of Samples Instrument Multivariate Model Performance Instrument Multivariate Model Performance
Moisture: Moisture:
R2 CV = 0.96 R2 CV = 0.83
RMSECV = 2.10% RMSECV = 4.51%
Bayesian regression Protein: Bayesian regression Protein:
(whole spectral R2 CV = 0.88 (whole spectral R2 CV = 0.81
range,350–1830 nm) RMSECV = 1.77% range, 350–1830 nm) RMSECV = 2.25%
Lipids: Lipids:
R2 CV = 0.85 FoodScan (FOSS, R2 CV = 0.67
LabSpec2500 (ASD RMSECV = 2.03% Denmark) RMSECV = 3.20%
Inc., USA) 850–1048 nm,
Moisture: Moisture:
350–1830 nm in transmittance;
R2 CV = 0.96 R2 CV = 0.77
grinding the sample
RMSECV = 2.30% RMSECV = 5.26%
Bayesian regression Protein: Bayesian regression Protein:
(common spectral R2 CV = 0.91 (common spectral R2 CV = 0.73
range,850–1050 nm) RMSECV = 1.57% range,850–1050 nm) RMSECV = 2.62%
Lipids: Lipids:
R2 CV = 0.85 R2 CV = 0.64
RMSECV = 2.03% RMSECV = 3.28%
Moisture: Moisture:
R2 P = 0.94 R2 P = 0.94
RMSEP = 1.14% RMSEP = 1.10%
PLSR PLSR
(on whole cheese) Fat: (on whole cheese) Fat:
R2 P = 0.98 R2 P = 0.94
Moisture and fat, N = 46 cheese RMSEP = 1.19% NIRFlex N-500 RMSEP = 1.90%
SCiO (Consumer
samples (Buchi, Flawil,
[27] Physics, Israel); Moisture: Moisture:
(whole and grated) Switzerland);
740–1070 nm R2 P = 0.93 1000–2500 nm R2 P = 0.96
RMSEP = 1.71% RMSEP = 0.93%
PLSR PLSR
(on grated cheese) Fat: (on grated cheese) Fat:
R2 P = 0.99 R2 P = 0.99
RMSEP = 0.82% RMSEP = 0.77%
Foods 2021, 10, 2377 18 of 23

Table 2. Cont.

Research Objective, Handheld/Portable NIR Spectrometer Benchtop NIR Spectrometer

Reference Predictive Models,
Number and Type of Samples Instrument Multivariate Model Performance Instrument Multivariate Model Performance
Dry matter: Dry matter:
R2 P = 0.62 R2 P = N/A
RMSEP = 0.65% RMSEP = 0.71%
X-NIR
PLSR Fat: NIRFlex N-500 PLSR Fat:
(Dinamica Generale
Dry matter, fat and protein; (on grinded cheese R2 (Buchi, Flawil, (on grinded cheese R2 P = N/A
[23] Electronic Solutions P = 0.91
N = 195 Grana Padano cheese pasture) RMSEP = 0.46% Switzerland); pasture) RMSEP = 0.54%
& Sensors, Italy)
1000–2500 nm
950–1800 nm Protein: Protein:
R2 P = 0.83 R2 P = N/A
RMSEP = 0.40% RMSEP = 0.49%
MicroNIR2200
(VIAVI Solution Inc., R2 P = 0.96
PLSR
Scottsdale, AZ, USA); RMSEP = 0.27% NIRFlex N-500
Melamine detection; 1128–2162 nm (Buchi, Flawil, R2 P = 0.96
[17] PLSR
N = 111 milk powder samples MicroPhazir (Thermo Switzerland); RMSEP = 0.28%
Fisher Scientific, R2 P = 0.95 1000–2500 nm
PLSR
Waltham, MA, USA); RMSEP = 0.33%
1600–2400
FT-NIR (Bruker Multi
The authors highlighted Purpose Analyser CCR = 100% for
Phazir 1624 (Thermo hat portable device of (MPA), USA)
Non-targeted milk powder Melamine,
Fisher Scientific, limited utility for 800–2500 nm
[77] authentication; SIMCA SIMCA Aminothiazole,
Waltham, MA, USA) non-targeted detection of
(N > 67) FT-NIR (Perkin Elmer, Biuret at a 0.6–2%
1596–2400 nm adulteration of this food
USA) adulteration level
commodity
1000–2500 nm
Foods 2021, 10, 2377 19 of 23

Table 2. Cont.

Research Objective, Handheld/Portable NIR Spectrometer Benchtop NIR Spectrometer

Reference Predictive Models,
Number and Type of Samples Instrument Multivariate Model Performance Instrument Multivariate Model Performance
Urea: Urea:
R2 P = 0.91 R2 P = 0.92
RMSEP = 0.25% RMSEP = 0.23%
Glycine: Glycine:
R2 P = 0.75 MetriNIR (MetriNIR R2 P = 0.85
NIR-S-G1 PLSR PLSR
Whey protein powders RMSEP = 1.03% Research, RMSEP = 0.82%
(InnoSpectra Co., (common spectral (common spectral
[26] authentication; Development and
Taiwan); Taurine: Taurine:
(N = 819) range 950–1650 nm) Service Co., Hungary); range 950–1650 nm)
750–1700 nm R2 P = 0.85 R2 P = 0.90
950–1650 nm
RMSEP = 1.37% RMSEP = 1.14%
Melamine: Melamine:
R2 P = 0.82 R2 P = 0.86
RMSEP = 0.53% RMSEP = 0.21%
CCR = correct classification rate.
Foods 2021, 10, 2377 20 of 23

5. Challenges and Future Trends

Portable and handheld NIR analysers have gained increasing attention in the dairy
industry as a potential at-line process monitoring tool to be used on site or alongside the
processing line for fast product screening. Miniaturised NIR devices can offer many benefits
to dairy producers in their routine product checks. Nevertheless, the handheld NIR devices
are more of a complementary tool rather than a replacement of benchtop spectrometers.
It is clear that portable, handheld and miniaturised devices are under continuous
development and it is critical that the number of scientific publications related to the
evaluation of new instruments increases. However, these studies should be undertaken
following standardised evaluation protocols which at least consider the minimum number
of samples to use, the evaluation of the spectral repeatability, noise levels, stability of the
optical readings in different environmental conditions, optimisation of sample presen-
tation (especially in the case of liquids and semi-liquids), and proper validation of the
developed models.
One of the challenges for the current handheld NIR instruments is the sensitivity of
the measurements and the applicability. The potential applications of miniaturised NIR
devices can be overly simplified by the suppliers of the technology. As well, the developed
NIR models require ongoing model maintenance, which is a detail that is neglected to
be communicated to the customer. For example, the application models advertised on
the supplier website for agri-food products (e.g., fruit) are not robust enough to model
the wide variability that can exist in the chemical composition of such products, as well
as the influence of the environment and production factors. It should be mentioned that
the software used for rapid data analysis is also required. Some handheld NIR analyser
providers offer a cloud-based solution for easy access, storage and analysis of NIR data.
However, caution should be taken regarding cloud connectivity, data safety and data
protection. In the upcoming years, there will be a growth in the use of open-source tools,
such as R and Python and computation in the cloud. This will allow researchers of portable
instruments to develop their own calibration and classification models, as well as bespoke
apps for mobile/tablet devices for a given product/application.
Despite this, there is an obligation to contribute to the body of knowledge within
NIR spectroscopy regarding the potential of these novel NIRS spectrophotometers. It can
also act as the object (front end) sensing technology as part of an Internet of Things (IoT)
network. Without a doubt, these compact, economical, and miniaturised spectrometers
already represent a new era in NIRS technology.

Author Contributions: Writing—original draft preparation, Y.P.; writing—review and editing, N.O.,
D.P.-M. and A.G.-V.; supervision, N.O., D.P.-M. and A.G.-V. All authors have read and agreed to the
published version of the manuscript.
Funding: Yuanyuan Pu has received funding from the Research Leaders 2025 programme co-funded
by Teagasc (the Irish Agriculture and Food Development authority) and the European Union’s Hori-
zon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement
number 754380.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No new data were created or analyzed in this study. Data sharing is
not applicable to this article.
Acknowledgments: The authors wish to acknowledge the funding agencies as stated above.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the
writing of the manuscript.
Foods 2021, 10, 2377 21 of 23

References
1. Williams, P.C. Variables affecting near-infrared reflectance spectroscopic analysis. In Near-Infrared Technology in the Agricultural and
Food Industries; Williams, P.C., Norris, K., Eds.; American Association of Cereal Chemists: St. Paul, MN, USA, 1987; pp. 143–168.
2. OECD; FAO. OECD-FAO Agricultural Outlook 2019–2028; OECD: Paris, France; FAO: Quebec City, QC, Canada, 2019. [CrossRef]
3. Karoui, R.; De Baerdemaeker, J. A review of the analytical methods coupled with chemometric tools for the determination of the
quality and identity of dairy products. Food Chem. 2007, 102, 621–640. [CrossRef]
4. Pi, F.; Shinzawa, H.; Ozaki, Y.; Han, D. Non-destructive determination of components in processed cheese slice wrapped with a
polyethylene film using near-infrared spectroscopy and chemometrics. Int. Dairy J. 2009, 19, 624–629. [CrossRef]
5. Ozbekova, Z.; Kulmyrzaev, A. Fluorescence spectroscopy as a non destructive method to predict rheological characteristics of
Tilsit cheese. J. Food Eng. 2017, 210, 42–49. [CrossRef]
6. Wang, Y.; Ding, W.; Kou, L.; Li, L.; Wang, C.; Jurick, W.M. A Non-destructive method to assess freshness of raw bovine milk using
FT-NIR spectroscopy. J. Food Sci. Technol. 2015, 52, 5305–5310. [CrossRef] [PubMed]
7. Liu, J. Terahertz spectroscopy and chemometric tools for rapid identification of adulterated dairy product. Opt. Quantum Electron.
2017, 49, 1–8. [CrossRef]
8. Karoui, R.; Dufour, É.; De Baerdemaeker, J. Front face fluorescence spectroscopy coupled with chemometric tools for monitoring
the oxidation of semi-hard cheeses throughout ripening. Food Chem. 2007, 101, 1305–1314. [CrossRef]
9. Shenk, J.S.; Workman, J.J., Jr.; Westerhaus, M.O. Chapter 17 Application of NIR Spectroscopy to Agricultureal Products. In
Handbook of Near-Infrared Analysis, 3rd ed.; Burns, D.A., Ciurczak, E.W., Eds.; CRC Press: Boca Raton, FL, USA, 2007.
10. Frankhuizen, R. Chapter 22—NIR Analysis of Dairy Products. In Handbook of Near-Infrared Analysis, 2nd ed.; Burns, D.A.,
Ciurczak, E.W., Eds.; Taylor & Francis: London, UK, 2001; pp. 609–641.
11. Giangiacomo, R.; Cattaneo, T.M.P. Analysis of Dairy and Eggs. In Near-Infrared Spectroscopy in Agriculture; American Society of
Agronomy, Inc.: Madison, WI, USA; Crop Science Society of America, Inc.: Madison, WI, USA; Soil Science Society of America,
Inc.: Madison, WI, USA, 2004; pp. 559–597.
12. Rodriguez-Otero, J.L.; Hermida, M.; Centeno, J. Analysis of Dairy Products by Near-Infrared Spectroscopy: A Review. J. Agric.
Food Chem. 1997, 45, 2815–2819. [CrossRef]
13. Frankhuizen, R. Chapter 20—NIR Analysis of Dairy Products. In Handbook of Near-Infrared Analysis, 3rd ed.; Burns, D.A.,
Ciurczak, E.W., Eds.; CRC Press: Boca Raton, FL, USA, 2007; pp. 415–437.
14. Holroyd, S.E. The Use of near Infrared Spectroscopy on Milk and Milk Products. J. Near Infrared Spectrosc. 2013, 21, 311–322.
[CrossRef]
15. Holroyd, S.E. The use of NIRS in the dairy industry: New trends and applications. NIR News 2017, 28, 22–25. [CrossRef]
16. Wolffenbuttel, R. MEMS-based Optical Mini- and Microspectrometers for the Visible and Infrared Spectral Range. J. Micromech.
Microeng. 2005, 15, S145. [CrossRef]
17. de la Roza-Delgado, B.; Garrido-Varo, A.; Soldado, A.; González Arrojo, A.; Cuevas Valdés, M.; Maroto, F.; Pérez-Marín, D.
Matching portable NIRS instruments for in situ monitoring indicators of milk composition. Food Control 2017, 76, 74–81. [CrossRef]
18. Henn, R.; Kirchler, C.G.; Grossgut, M.-E.; Huck, C.W. Comparison of sensitivity to artificial spectral errors and multivariate LOD
in NIR spectroscopy—Determining the performance of miniaturizations on melamine in milk powder. Talanta 2017, 166, 109–118.
[CrossRef]
19. Liu, N.; Parra, H.A.; Pustjens, A.; Hettinga, K.; Mongondry, P.; van Ruth, S.M. Evaluation of portable near-infrared spectroscopy
for organic milk authentication. Talanta 2018, 184, 128–135. [CrossRef]
20. de Lima, G.F.; Andrade, S.A.C.; da Silva, V.H.; Honorato, F.A. Multivariate Classification of UHT Milk as to the Presence of
Lactose Using Benchtop and Portable NIR Spectrometers. Food Anal. Methods 2018, 11, 2699–2706. [CrossRef]
21. Eskildsen, C.E.; Sanden, K.W.; Wubshet, S.G.; Andersen, P.V.; Øyaas, J.; Wold, J.P. Estimating dry matter and fat content in
blocks of Swiss cheese during production using on-line near infrared spectroscopy. J. Near Infrared Spectrosc. 2019, 27, 293–301.
[CrossRef]
22. Llano Suárez, P.; Soldado, A.; González-Arrojo, A.; Vicente, F.; de la Roza-Delgado, B. Rapid on-site monitoring of fatty acid
profile in raw milk using a handheld near infrared sensor. J. Food Compos. Anal. 2018, 70, 1–8. [CrossRef]
23. Uusitalo, S.; Aernouts, B.; Sumen, J.; Hietala, E.; Utriainen, M.; Frondelius, L.; Kajava, S.; Pastell, M. Comparison of milk analysis
performance between NIR laboratory analyser and miniaturised NIR MEMS sensors. In Proceedings of the ICAR Conference,
Prague, Czech Republic, 16–21 June 2019; pp. 111–115.
24. Marinoni, L.; Stroppa, A.; Barzaghi, S.; Cremonesi, K.; Pricca, N.; Meucci, A.; Pedrolini, G.; Galli, A.; Cabassi, G. On site
monitoring of Grana Padano cheese production using portable spectrometers. In Proceedings of the 18th International Conference
on Near Infrared Spectroscopy, Copenhagen, Denmark, 11–15 June 2017; pp. 85–90.
25. Si-Ware-Systems. Raw Milk Analysis; Si-Ware-Systems: Tokyo, Japan, 2019.
26. Riu, J.; Gorla, G.; Chakif, D.; Boqué, R.; Giussani, B. Rapid Analysis of Milk Using Low-Cost Pocket-Size NIR Spectrometers and
Multivariate Analysis. Foods 2020, 9, 1090. [CrossRef] [PubMed]
27. Zinia Zaukuu, J.-L.; Aouadi, B.; Lukács, M.; Bodor, Z.; Vitális, F.; Gillay, B.; Gillay, Z.; Friedrich, L.; Kovacs, Z. Detecting Low
Concentrations of Nitrogen-Based Adulterants in Whey Protein Powder Using Benchtop and Handheld NIR Spectrometers and
the Feasibility of Scanning through Plastic Bag. Molecules 2020, 25, 2522. [CrossRef] [PubMed]
Foods 2021, 10, 2377 22 of 23

28. Verena Wiedemair, D.L.; Garsleitner, R.; Dillinger, K.; Huck, C. Investigations into the Performance of a Novel Pocket-Sized
Near-Infrared Spectrometer for Cheese Analysis. Molecules 2019, 24, 428. [CrossRef] [PubMed]
29. De Marchi, M.; Penasa, M.; Zidi, A.; Manuelian, C.L. Invited review: Use of infrared technologies for the assessment of dairy
products—Applications and perspectives. J. Dairy Sci. 2018, 101, 10589–10604. [CrossRef]
30. Crocombe, R.A. Portable Spectroscopy. Appl. Spectrosc. 2018, 72, 1701–1751. [CrossRef]
31. Niemoller, A.; Behmer, D.; Marston, D.; Prescott, B. Online analysis of dairy products with FT-NIR spectroscopy. In Proceedings
of the 11th International Conference on Near Infrared Spectroscopy, Córdoba, Spain, 6–11 April 2003; pp. 607–612.
32. Nguyen, H.N.; Dehareng, F.; Hammida, M.; Baeten, V.; Froidmont, E.; Soyeurt, H.; Niemöeller, A.; Dardenne, P. Potential of
near Infrared Spectroscopy for On-Line Analysis at the Milking Parlour Using a Fibre-Optic Probe Presentation. NIR News 2011,
22, 11–13. [CrossRef]
33. Aernouts, B.; Polshin, E.; Lammertyn, J.; Saeys, W. Visible and near-infrared spectroscopic analysis of raw milk for cow health
monitoring: Reflectance or transmittance? J. Dairy Sci. 2011, 94, 5315–5329. [CrossRef] [PubMed]
34. Pu, Y.-Y.; O’Donnell, C.; Tobin, J.T.; O’Shea, N. Review of near-infrared spectroscopy as a process analytical technology for
real-time product monitoring in dairy processing. Int. Dairy J. 2020, 103, 104623. [CrossRef]
35. Crocombe, R. Handheld Spectrometers in 2018 and Beyond: MOEMS, Photonics, and Smartphones; SPIE: Bellingham, WA, USA, 2018;
Volume 10545.
36. Yan, H.; Siesler, H.W. Hand-held near-infrared spectrometers: State-of-the-art instrumentation and practical applications. NIR
News 2018, 29, 8–12. [CrossRef]
37. Beć, K.B.; Grabska, J.; Siesler, H.W.; Huck, C.W. Handheld near-infrared spectrometers: Where are we heading? NIR News 2020,
31, 28–35. [CrossRef]
38. Rodriguez-Saona, L.; Aykas, D.P.; Borba, K.R.; Urtubia, A. Miniaturization of optical sensors and their potential for high-
throughput screening of foods. Curr. Opin. Food Sci. 2020, 31, 136–150. [CrossRef]
39. Saranwong, S.; Kawano, S. Commercial Portable NIR Instruments in Japan. NIR News 2005, 16, 27. [CrossRef]
40. Solgaard, O.; Godil, A.A.; Howe, R.T.; Lee, L.P.; Peter, Y.; Zappe, H. Optical MEMS: From Micromirrors to Complex Systems. J.
Microelectromech. Syst. 2014, 23, 517–538. [CrossRef]
41. Wang, J.; Zheng, B.; Wang, X. Strategies for high performance and scalable on-chip spectrometers. J. Phys. Photonics 2020,
3, 012006. [CrossRef]
42. Ma, Y.B.; Babu, K.S.; Amamcharla, J.K. Prediction of total protein and intact casein in cheddar cheese using a low-cost handheld
short-wave near-infrared spectrometer. LWT 2019, 109, 319–326. [CrossRef]
43. Polychromix. Polychromix Launches the Phazir NIR Handheld Digital Transform Spectrometer; Polychromix: Woburn, MA, USA, 2006.
44. Crocombe, R.A. MEMS technology moves process spectroscopy into a new dimension. Spectrosc. Eur. 2004, 16, 16–19.
45. dos Santos, C.A.T.; Lopo, M.; Páscoa, R.N.M.J.; Lopes, J.A. A Review on the Applications of Portable Near-Infrared Spectrometers
in the Agro-Food Industry. Appl. Spectrosc. 2013, 67, 1215–1233. [CrossRef] [PubMed]
46. Thermo Fisher. microPHAZIR Analyzer. Available online: https://assets.thermofisher.com/TFS-Assets/CAD/Product-Guides/
microPHAZIR-User-Manual.pdf (accessed on 25 May 2021).
47. Geller, Y. Using MEMS Technology for Cost Effective Recycling of Plastics; SPIE: Bellingham, WA, USA, 2007; Volume 6466.
48. Day, D.R.; Butler, M.A.; Smith, M.C.; McAllister, A.; Deutsch, E.R.; Zafiriou, K.; Senturia, S.D. Diffractive-MEMS implementation
of a Hadamard near-infrared spectrometer. In Proceedings of the 13th International Conference on Solid-State Sensors, Actuators
and Microsystems, Seoul, Korea, 5–9 June 2005; Digest of Technical Papers. Transducers 05. Volume 1242, pp. 1246–1249.
49. Senturia, S.D. MEMS-Enabled Products: A Growing Market Segment. 2005. Available online: https://www.pharmaceuticalonline.
com/doc/mems-enabled-products-a-growing-market-0002 (accessed on 25 May 2021).
50. Ianm. Nir Products at Pittcon 2012. Available online: https://www.impublications.com/content/nir-products-pittcon-2012
(accessed on 28 October 2020).
51. Nada, A.O.B.; Charles, A.H.; Donald, M.F.; Fred, J.V.M.; Marc, K.v.G.; Frank, P.; Heinz, W.S. Miniature near-infrared (NIR)
spectrometer engine for handheld applications. In Next-Generation Spectroscopic Technologies; SPIE: Bellingham, WA, USA, 2012.
52. Friedrich, D.M.; Hulse, C.A.; von Gunten, M.; Williamson, E.P.; Pederson, C.G.; O’Brien, N.A. Miniature near-infrared spectrome-
ter for point-of-use chemical analysis. In Photonic Instrumentation Engineering; International Society for Optics and Photonics:
Bellingham, WA, USA, 2014; p. 899203.
53. VIAVI Solutions Inc. MicroNIR Spectrometer1700 vs. 2200: Comparison and Case Studies; VIAVI Solutions Inc.: Scottsdale, AZ,
USA, 2019.
54. Texas-Instruments. DLP® NIRscan™ Nano Evaluation Module. Available online: https://www.ti.com/tool/DLPNIRNANOEVM#
order-start-development (accessed on 27 May 2021).
55. Nelson, P. DLP Technology for Spectroscopy. Texas Instruments White Paper. 2016. Available online: https://www.ti.com/
lit/wp/dlpa048a/dlpa048a.pdf?ts=1618578980095&ref_url=https%253A%252F%252Fwww.google.com%252F (accessed on 20
September 2021).
56. Perrella, G. Texas Instruments DLP® NIRscan™ Nano Evaluation Module (EVM) Optical Design Considerations; Texas Instruments
Incorporated: Dallas, TX, USA, 2016.
57. InnoSpectra. Available online: http://www.inno-spectra.com/en/product (accessed on 24 May 2021).
58. Si-Ware-Systems. Available online: https://www.si-ware.com/our-offerings/ (accessed on 24 May 2021).
Foods 2021, 10, 2377 23 of 23

59. Spectral-Engines. Nirone Sensor S. Available online: https://www.spectralengines.com/products/nirone-sensors (accessed on

24 May 2021).
60. SCiO. Available online: https://shop.consumerphysics.com/collections/researcher-and-developer-kits (accessed on 24
May 2021).
61. Si-Ware-Systems. The “Neo” Way for FT-IR. Available online: https://www.si-ware.com/how-it-works/mems-ft-ir/ (accessed
on 24 May 2021).
62. Deidda, R.; Sacre, P.-Y.; Clavaud, M.; Coïc, L.; Avohou, H.; Hubert, P.; Ziemons, E. Vibrational spectroscopy in analysis of
pharmaceuticals: Critical review of innovative portable and handheld NIR and Raman spectrophotometers. TrAC Trends Anal.
Chem. 2019, 114, 251–259. [CrossRef]
63. Kalinin, A.; Krivtsun, V.; Krasheninnikov, V.; Sadovskiy, S.; Denisovich, H.; Yurova, H. Calibration Models for Multi-Component
Quantitative Analyses of Dairy with the Use of Two Different Types of Portable near Infrared Spectrometer. J. Near Infrared
Spectrosc. 2008, 16, 343–348. [CrossRef]
64. Kalinin, A.; Krasheninnikov, V.; Sadovskiy, S.; Yurova, E. Determining the Composition of Proteins in Milk Using a Portable near
Infrared Spectrometer. J. Near Infrared Spectrosc. 2013, 21, 409–415. [CrossRef]
65. Amr, M.; Sabry, Y.M.; Khalil, D. Near-infrared optical MEMS spectrometer-based quantification of fat concentration in milk. In
Proceedings of the 2018 35th National Radio Science Conference (NRSC), Cairo, Egypt, 20–22 March 2018; pp. 409–416.
66. Santos, P.M.; Pereira-Filho, E.R.; Rodriguez-Saona, L.E. Application of Hand-Held and Portable Infrared Spectrometers in Bovine
Milk Analysis. J. Agric. Food Chem. 2013, 61, 1205–1211. [CrossRef]
67. Moreira, M.; França, J.A.D.; Filho, D.D.O.T.; Beloti, V.; Yamada, A.K.; França, M.B.D.M.; Ribeiro, L.D.S. A Low-Cost NIR Digital
Photometer Based on InGaAs Sensors for the Detection of Milk Adulterations With Water. IEEE Sens. J. 2016, 16, 3653–3663.
[CrossRef]
68. Upreti, P.; McKay, L.L.; Metzger, L.E. Influence of Calcium and Phosphorus, Lactose, and Salt-to-Moisture Ratio on Cheddar
Cheese Quality: Changes in Residual Sugars and Water-Soluble Organic Acids During Ripening. J. Dairy Sci. 2006, 89, 429–443.
[CrossRef]
69. Upreti, P.; Metzger, L.E. Influence of Calcium and Phosphorus, Lactose, and Salt-to-Moisture Ratio on Cheddar Cheese Quality:
pH Changes During Ripening. J. Dairy Sci. 2007, 90, 1–12. [CrossRef]
70. Stocco, G.; Cipolat-Gotet, C.; Ferragina, A.; Berzaghi, P.; Bittante, G. Accuracy and biases in predicting the chemical and physical
traits of many types of cheeses using different visible and near-infrared spectroscopic techniques and spectrum intervals. J. Dairy
Sci. 2019, 102, 9622–9638. [CrossRef]
71. Augustin, M.A.; Margetts, C.L. Powdered Milk|Milk Powders in the Marketplace. In Encyclopedia of Food Sciences and Nutrition,
2nd ed.; Caballero, B., Ed.; Academic Press: Oxford, UK, 2003; pp. 4694–4702.
72. Wu, D.; He, Y.; Feng, S. Short-wave near-infrared spectroscopy analysis of major compounds in milk powder and wavelength
assignment. Anal. Chim. Acta 2008, 610, 232–242. [CrossRef]
73. Kong, W.W.; Zhang, C.; Liu, F.; Gong, A.P.; He, Y. Irradiation dose detection of irradiated milk powder using visible and
near-infrared spectroscopy and chemometrics. J. Dairy Sci. 2013, 96, 4921–4927. [CrossRef] [PubMed]
74. Lu, C.; Xiang, B.; Hao, G.; Xu, J.; Wang, Z.; Chen, C. Rapid Detection of Melamine in Milk Powder by near Infrared Spectroscopy.
J. Near Infrared Spectrosc. 2009, 17, 59–67. [CrossRef]
75. Chen, H.; Tan, C.; Lin, Z.; Wu, T. Detection of melamine adulteration in milk by near-infrared spectroscopy and one-class partial
least squares. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 2017, 173, 832–836. [CrossRef] [PubMed]
76. Chen, S. Sham or shame: Rethinking the China’s milk powder scandal from a legal perspective. J. Risk Res. 2009, 12, 725–747.
[CrossRef]
77. Allegrini, F.; Olivieri, A.C. IUPAC-Consistent Approach to the Limit of Detection in Partial Least-Squares Calibration. Anal. Chem.
2014, 86, 7858–7866. [CrossRef] [PubMed]
78. Karunathilaka, S.R.; Yakes, B.J.; He, K.; Chung, J.K.; Mossoba, M. Non-targeted NIR spectroscopy and SIMCA classification
for commercial milk powder authentication: A study using eleven potential adulterants. Heliyon 2018, 4, e00806. [CrossRef]
[PubMed]