Analytica Chimica Acta 638 (2009) 1–15

Contents lists available at ScienceDirect

Analytica Chimica Acta

journal homepage: www.elsevier.com/locate/aca

Review

A 21st century technique for food control: Electronic noses

Miguel Peris a,∗ , Laura Escuder-Gilabert b
a
Departamento de Química, Universidad Politécnica de Valencia, 46071 Valencia, Spain
b
Departamento de Química Analítica, Universitat de Valencia, C/Vicente Andrés Estellés s/n, E-46100 Burjasot,
Valencia, Spain

a r t i c l e i n f o a b s t r a c t

Article history: This work examines the main features of modern electronic noses (e-noses) and their most important
Received 28 November 2008 applications in food control in this new century. The three components of an electronic nose (sample
Received in revised form 3 February 2009 handling system, detection system, and data processing system) are described. Special attention is devoted
Accepted 3 February 2009
to the promising mass spectrometry based e-noses, due to their advantages over the more classical gas
Available online 12 February 2009
sensors. Applications described include process monitoring, shelf-life investigation, freshness evaluation,
authenticity assessment, as well as other general aspects of the utilization of electronic noses in food
Keywords:
control. Finally, some interesting remarks concerning the strengths and weaknesses of electronic noses
Electronic noses
Food analysis
in food control are also mentioned.
Gas sensors
© 2009 Elsevier B.V. All rights reserved.
Mass spectrometry

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Why an electronic nose in food control? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3. Description and structure of an e-nose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3.1. Sample handling system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3.2. Detection system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.3. Data processing system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4. Applications of electronic noses in food analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.1. Process monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2. Shelf-life investigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.3. Freshness evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.4. Authenticity assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Abbreviations: 4EP, 4-ethylphenol; 4EG, 4-ethylguaiacol; ANN, artificial neural network; APC, aerobic plate count; API, atomic pressure ionization; BAW, bulk acoustic
wave; BP-ANN, back-propagation artificial neural network; BP, backpropagation; CA, cluster analysis; CART, classification and regression tree; CP, conducting polymer;
CP-ANN, counterpropagation artificial neural network; DFA, discriminant factor analysis; DHS, dynamic headspace; DOE, design of experiments; e-nose, electronic nose;
EC, electrochemical sensor; FCM, fuzzy C means; FDA, factorial discriminant analysis; FSGDA, forward step-wise general discriminant analysis; GA, genetic algorithm;
GC, gas chromatography; GC–MS, gas chromatography–mass spectrometry; HS, headspace; HS-MS, head-space mass spectrometry; IMS, ion mobility spectrometry; INDEX,
inside-needle dynamic extraction; k-NN, k-nearest neighbors; KOSM, Kohonen self-organizing map; LDA, linear discriminant analysis; LVQ-NN, learning vector quantisa-
tion neural network; MDA, multiple discriminant analysis; MIMS, membrane introduction mass spectrometry; MLP, multilayer percepton; MLR, multiple linear regression;
MOS, metal oxide semiconductors; MOSFET, metal oxide semiconductor field effect transistor; MS, mass spectrometry; MSE-nose, mass spectrometry based electronic
nose; p-AV, anisidine value; P&T, purge and trap; PARAFAC, parallel factor analysis; PCA, principal component analysis; PCR, principal component regression; PLS, partial
least squares; PLS-DA, partial least square-discriminant analysis; PNN, probabilistic neural network; PR, pattern recognition; PTR, proton transfer reaction; PV, peroxide
value; QDA, quadratic discriminant analysis; QLSR, quadratic least squares regression; QMB or QCM, quartz microbalances; REP-PCR, repetitive extragenic palindromic
polymerase chain reaction; SAW, surface acoustic wave; SBSE, stir bar sorptive extraction; SHS, static headspace; SIMCA, soft independent modelling of class analogy;
SLDA, stepwise linear discriminant analysis; SOM, self-organizing map; SPME, solid-phase microextraction; SPR, surface plasmon resonance; SVM, support vector machine;
TBARS, thiobarbituric acid reactive substances; TDNN, time-delay neural networks; TSM, thickness shear mode; VOCs, volatile organic compounds; WPTER, wavelet packet
transform for efficient pattern recognition.
∗ Corresponding author. Tel.: +34 963877007; fax: +34 963877349.
E-mail address: mperist@qim.upv.es (M. Peris).

0003-2670/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.aca.2009.02.009
2 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

4.5. Other food quality control studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

5. Conclusions and future trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1. Introduction of the analysis can be greatly improved by adopting an appropriate

sampling technique.
Electronic noses (e-noses) are instruments which mimic the To introduce the volatile compounds present in the headspace
sense of smell. These devices are typically array of sensors used (HS) of the sample into the e-noses detection system, several sam-
to detect and distinguish odors precisely in complex samples and pling techniques have been used in literature [3,4,9]:
at low cost [1,2]. These features make e-noses very useful for diverse
applications in the food, cosmetic and pharmaceutical industry as 1. Static headspace (SHS) technique consists of placing the sam-
well as in environmental control or clinical diagnostics. In fact, ple in a hermetically sealed vial and then, once equilibrium has
in the past decade, many papers have appeared in the literature been established between the matrix and the gaseous phase,
describing the use of electronic noses, most notably in the field of sampling the HS. Sample temperature, equilibration time, vial
food control [3–8]. This paper will focus on the use of electronic size and sample quantity are the main parameters that have to
noses in quality control and process operations in the food indus- be optimized. Because of the poor repeatability of manual HS
try. The principles behind the design of this type of devices and injection, it is recommended that an automatic HS sampler be
the most relevant contributions in food control along the present used. In some applications a vapor-flow system has been used,
century are also discussed. and has provided better control than manual headspace injection
of the operating temperature and the amount of analyte that is
2. Why an electronic nose in food control? introduced into the detector.
2. Purge and trap (P&T) and dynamic headspace (DHS) techniques
The complexity of most food aromas make them difficult to be have been used in some applications to increase sensitivity, since
characterized with conventional flavor analysis techniques such as they provide a pre-concentration of volatile compounds. In these
gas chromatography or gas chromatography olfactometry. Never- systems, the volatile components are purged by a stream of inert
theless, sensory analysis by a panel of experts is a costly process gas and trapped onto an adsorbent. In the case of P&T, the gas
since it requires trained people who can work for only relatively flow is injected through the sample, whereas, in the case of DHS,
short periods of time; additional problems such as the subjectivity only the HS is purged with the gas. The constant depletion of
of human response to odors and the variability between individ- the HS leads to a displacement of the equilibrium in favor of
uals are also to be considered. Hence, the need of an instrument the desorption of these molecules from the matrix. The trapped
such as the electronic nose, whose strengths include high sensitiv- molecules are subsequently desorbed by heating and introduced
ity and correlation with data from human sensory panels for several into the detection system. Apart from the choice of trap, the
specific applications in food control. Because they are easy to build, main parameters to optimize are the temperature of the sam-
cost-effective and as they provide a short time of analysis, electronic ple, the equilibration time, the flow rate of the extractor gas and
noses are becoming more and more popular as objective automated the purge time of the HS.
non-destructive techniques to characterize food flavors. However, 3. Solid-phase microextraction (SPME) is a user-friendly pre-
there is much research still to be done especially with regard to concentration method. The principle involves exposing a silica
sensors technology, data processing, interpretation of results and fiber covered with a thin layer of adsorbent in the HS of the sam-
validation studies. ple in order to trap the volatile components onto the fiber. The
adsorbed compounds are then desorbed by heating and intro-
3. Description and structure of an e-nose duced into the detection system. Apart from the nature of the
adsorbent deposited on the fiber, the main parameters to opti-
An electronic nose is a machine that is designed to detect and mize are the equilibration time, the sample temperature and the
discriminate among complex odors using a sensor array. The sen- duration of extraction. This technique has a considerable con-
sor array consists of broadly tuned (non-specific) sensors that are centration capacity and it is very simple because, unlike P&T or
treated with a variety of odor-sensitive biological or chemical mate- DHS, it does not require especial equipment.
rials. An odor stimulus generates a characteristic fingerprint (or 4. Stir bar sorptive extraction (SBSE) is a magnetic bar coated with
smellprint) from the sensor array. Patterns or fingerprints from polymers, which can be held in the HS for sampling. Its load-
known odors are used to construct a database and train a pattern ing capacity is much higher than that of SPME. Even though it
recognition system so that unknown odors can subsequently be has been developed only recently, SBSE is a promising extraction
classified and identified [2]. This is the classical concept of an e- technique when very high sensitivity is required.
nose; however, in recent years, as discussed below, the classical 5. Inside-needle dynamic extraction (INDEX) is also a pre-
sensor types used for e-noses have been enhanced and comple- concentration technique [10]. The INDEX needles contain an
mented by other technologies introduced in this field. Nevertheless absorbing polymer phase very much like a fixed bed. The volatile
and in a broader sense, electronic nose instruments are composed compounds are forced through the needle by repeated aspi-
of three elements, namely: (i) a sample handling system, (ii) a detec- ration/ejection motions of the syringe plunger. The potential
tion system, and (iii) a data processing system. All of them will be advantage of this system compared to SPME is its mechan-
described thereafter. ical robustness and the possibility of increasing the amount
of absorbing polymer as well as the surface area available for
3.1. Sample handling system adsorbing volatile compounds.
6. Membrane introduction mass spectrometry (MIMS) is a sam-
Sample handling is a critical step affecting the analysis by e- ple handling system used in mass spectrometry (MS) based
noses whose importance is very often ignored. Anyway, the quality e-noses. This technique allows the direct introduction of specific
 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
compounds of a liquid or gas sample into a mass spectrometer.
A thin membrane is installed between the sample and the ion
source of a mass spectrometer in such a way that some com-
pounds dissolve in the membrane, diffuse through it and, finally,
evaporate directly into the ion source [11].
Although, a priori, any sampling headspace technique can be
used as the sample-handling part of an e-nose, the choice must be
made with care and take into account the type of sample and the
method specifications required. SHS is the most common technique
because it is very simple to use, however, for some applications,
the SHS technique has the drawback of low sensitivity because the
volatile compounds are not pre-concentrated. On the other hand,
pre-concentration systems improve the sensitivity making detec-
tion easier and extracting semi-volatiles which otherwise would
not be detected. However, they introduce a supplementary step in
the method, which increases the time of analysis. Moreover, analyt-
ical artifacts (memory effects, bleeding or irreversible adsorption)
are generated in some cases. In this respect, the pre-concentration
media must be carefully chosen. For instance, graphite carbon traps
show a high capacity of adsorption for polar compounds; never-
theless, this adsorption can be sometimes irreversible and may
produce an undesirable retention of large amounts of water. On
the contrary, the porous polymer traps have a lower adsorption
capacity but present the advantage of having less affinity for water.
3.2. Detection system
The classical e-nose, consisting of an array of gas sensors as
detection system, is the most common approach, although new
technologies such as MS and ion mobility spectrometry (IMS) have
recently entered this field. Their principles and technology will be
briefly described in this section.
A chemical sensor is a device capable of converting a chemical
quantity into an electrical signal that is related to the concentration
of specific particles such as atoms, molecules, or ions in gases or
liquids [2]. The types of sensors that can be used in an e-nose need
to respond to molecules in the gas phase, which are typically volatile
organic compounds (VOCs) with different relative molar masses [2].
Several major categories of sensor arrays, have been involved in the
development of e-noses and each will be briefly described. Further
details about the types of sensors discussed here can be found in
literature [2,12–16].
1. Piezoelectric (also called gravimetric or acoustic) sensors, based
on propagation of acoustic waves produced by piezoelectrical
materials (i.e. quartz or LiNbO3 ) in a multilayer structure. Sur-
face acoustic wave (SAW) and bulk acoustic wave (BAW) are the
most common of those acoustic sensors. BAW sensors are also
commonly referred to as thickness shear mode (TSM) sensors or
quartz crystal microbalances (QCM or QMB).
2. Electrochemical sensors, including amperometric, potentiomet-
ric and chemiresistive or conductimetric sensors. Among them,
chemiresistive sensors such as metal oxide semiconductors
(MOS, also called Taguchi) and conducting polymer (CP) are
widely used to make arrays for gas and odor measurements. Sev-
eral potentiometric gas sensors using metal oxide semiconductor
field effect transistors (MOSFET) have been also developed and
used in e-noses technology.
3. Optical, such as optical fibers, as well as the more traditional
absorbance, reflectance, luminescence and surface plasmon res-
onance (SPR) techniques.
4. Calorimetric or thermal sensors, in which the heat of a chemical
reaction involving the analyte is monitored with a transducer
such as a thermistor.
3
A subdivision of the sensor grouping is the biosensors. These
incorporate a biological sensing element positioned close to the
transducer to give a reagentless sensing system specific for the tar-
get analyte [13]. This ensures specificity of the biological molecules
for target species.
At first, e-noses tended to be based on an array of gas sensors
of the same type, but practical experience has shown that often
this does not produce enough information for many real-world
problems. Increasingly, the tendency is to combine different types
of gas sensors to produce hybrid systems. However, this involves
more complex electronics and it is then necessary to normalize or
standardize the different sensor outputs [4].
Despite the considerable number of applications of e-noses
based on gas sensors that appear in food analysis literature, much
development is still required before can reach their full potential.
Gas sensors-based e-noses functioning is affected by several prob-
lems, such as sensor poisoning, profile masking by some major
constituents of the sample (ethanol for example), the strong influ-
ence of moisture, and the non-linearity of signals [11]. To overcome
some of these problems, mass spectrometry applications to food
aroma profiling are growing in literature [4,11].
MS-based e-noses are referred to as mass sensors or, sometimes,
new-generation electronic noses. This new instrument introduces
the volatile compounds into the ionization chamber of a MS instru-
ment (usually a quadrupole mass spectrometer) without prior
chromatographic separation. Each fragment ion (m/z ratio) of the
mass spectrum obtained acts as a “sensor” and its abundance is
equivalent to the sensor signal. Therefore, the number of sen-
sors in MS-based e-noses is variable, readily modifiable and, in
most cases high. Moreover, these “sensors” provide chemical infor-
mation about the sample. So, information about what kinds of
compounds are responsible for the differences between samples
can be obtained from the ion-fragmentation patterns.
Most applications to date based on MS-based e-noses have
focused on qualitative analysis. Nevertheless, some recent works
have demonstrated the potential of these instruments in quan-
tification tasks with a high sample throughput and have been
reviewed by Pérez Pavón et al. [11]. This review also affords sev-
eral strategies to correct long-term signal instability associated
with both headspace generation and the response of the mass
detector.
It must be mentioned that MS-based e-noses are promising
instruments for the analysis of alcoholic beverages because ethanol
does not cause saturation problems in the MS, unlike in gas sen-
sors. However, the ethanol fragment ions are much more abundant
than the fragment ions of the other volatile compounds. As a conse-
quence, when the chemometric analysis is performed, samples may
be differentiated by their ethanol content alone. This problem can
easily be solved if the fragment ions corresponding to ethanol are
not included in the fragment-ion range selected for the MS analysis
[4].
Alcoholic beverages is one of the examples where MS-based
e-noses have shown better performance than the classical gas sen-
sor based e-noses, but the former group is still the most common
approach used in the literature for many tasks; this could be due to
the price/performance relationship of MS-based e-noses. Briefly, a
comparison can be made between these two detection systems.
MS-based e-noses clearly overcome gas sensors in terms of: (i)
adaptability, by selecting the optimum set of fragment ions; (ii)
sensitivity, by rejecting fragment ions from potentially interfer-
ing components such as ethanol or water and (iii) versatility, since
they can be applied to a wide range of food samples. However, gas
sensors still have some advantages that have to be remarked: (i)
easier to use compared to the complex MS devices; (ii) lower price
and maintenance; (iii) possibility of on-line utilization due to their
portability. A good approach could then be to consider gas sensor-
4 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

based e-noses for screening purposes and MS-based sensors for
confirmatory purposes.
Finally, as commented above, IMS is also an emerging approach
in e-noses technology. This technique can be used to detect and
characterize organic vapors in air, which involves the ionization of
molecules and their subsequent drift through an electric field. Anal-
ysis is based on analyte separations resulting from ionic mobilities
rather than ionic masses. Compared to MS, the virtual sensor array
is not given by discrete mass/charge ratios, but by the signal inte-
gration over definable time intervals. IMS has the ability to separate
ionic species at atmospheric pressure; therefore a major advantage
is that it is possible to have smaller analytical units, lower power
requirements, lighter weight and easier use [17–19].
3.3. Data processing system
As mentioned in the previous section, pattern recognition (PR)
techniques are used for data processing of responses generated by
each sensor. The potential advantages of such an approach include
the reduction in complexity of the sensor coating selection, the abil-
ity to characterize complex mixtures without the need to identify
and quantify individual components. In this section, we will briefly
review the concepts behind this method.
Pattern recognition is a decision vector used to classify species
based on a series of measurements (a pattern) on that species.
Generally, a matrix is formed from the patterns for a number of
species and then a decision vector which divides the pattern into
an assigned binary classification is calculated based on standard
experiments. This is then used to classify unknown patterns. The
success of PR techniques can be enhanced or simplified by suitable
prior treatment of the data such that feature selection and feature
extraction are important approaches [20]. The former identifies and
selects those features present in the analytical data, which may be
important to calibration. Feature extraction changes the dimen-
sionality of the data and generally refers to processes combining
original variables to provide new and better ones [20].
The PR methods may be divided into supervised and non-
supervised methods although a combination of both can be used.
The major unsupervised technique is principal component analy-
sis (PCA) while artificial neural network (ANN) is the best-known
supervised technique. Briefly, PCA is a linear feature extraction tech-
nique which reduces dimensionality of data with a minimum loss
of information. This is achieved by projecting the data onto fewer
dimensions that are chosen to exploit the relationships between the
variables, so that the maximum amount of information is retained
in the smallest number of dimensions. This technique allows the
similarities and differences between samples to be better assessed.
On the other hand, the ANNs are computer programs based on a
simplified model of the brain; they reproduce its logical operation
using a collection of neurone-like entities forming networks to per-
form processing. ANN programs are multipurpose and with suitable
training, a single program could solve several problems.
There are a significant number of PR algorithms used for of
e-noses data processing and continuous developments are being
made in this sense. However, a full discussion about data analysis
for e-nose systems is out of the aim of this review. For this purpose,
the reader is referred to the review reported by [21].
Prior to PR analysis, it is usual to perform pre-processing of the
data [20]. The main aims of this stage are:
1. To reduce the amount of data which are irrelevant to the study.
2. To enhance sufficient information within the data to achieve the
desired goal.
3. To extract the information in, or transform the data to, a form
suitable for further analysis.
Probably the most common method of pre-processing spec-
tral data is normalization. This may involve either scaling each
spectrum in a collection so that the most intense band in each
spectrum is a constant value or the spectra may be normalized to
constant area under the absorption or emission curve. More com-
plex approaches involve developing a covariance matrix between
variables and extracting the eigenvectors and eigenvalues. Eigen
analysis will provide a set of variables, which are linear combina-
tions of the original variables. This has the effect of reducing the
dimensionality of the data and making the analysis simpler [20].
An important consideration is the statistical design of experi-
ments (DOE) to ensure that data obtained are valid. This is largely
because of the use of too few samples with very different char-
acteristics. One-at-a-time experiments, i.e. those where only one
variable is changed at a time, are inefficient. Thus, the statistical
DOE is of paramount importance, especially when many variables
are involved. DOE techniques allow the experimenter to identify
the input variables which affect the output of any process [22]. DOE
can speed up process and product development as well as improve
existing processes and products. Furthermore, DOE can also be used

Table 1
Main applications of e-noses in food process monitoring.

Sample Type of study Sample handling system Detection system Data processing Ref.
algorithm

Wine-must Discrimination between fermentation stages SHS—pervaporation A32S AromaScan: 32 CP PCA [24]
Iberian hams (‘Montanera’) Spoiling during the curing process SHS 16 Tin-oxide thin films PCA, PNN [25]
Milk fermented with Discrimination between genotype strains INDEX Smart Nose® : MS PCA [26]
Lactobacillus casei strains
used in Gruyère cheese
Milk fermented with Discrimination between odor intensity scores SHS FOX 3000: 12 MOS PCA [27]
Lactococcus lactis strains
Australian red wines Spoilage caused by Brettanomyces yeast SHS HP4440: MS PCA, PLS, SLDA [28]

Australian red wines Spoilage caused by Brettanomyces yeast SPME (for MOS) FOX 3000: 12 MOS PLS [29]
SHS (for MS) HP4440: MS

Tomato cv. Cencara Dehydration processes of tomato slices SHS Air Sense: 10 MOS PCA [30]

Mangoes (Mangifera indica L.)
Discrimination between harvest maturities SHS FOX 4000: 18 MOS DFA [31]
within a ripenning stage
Discrimination between ripenning stages
within a maturity stage
Discrimination between fruit varieties

Black tea Estimation of optimum fermentation time SHS 8 MOS TDNN, SOM [32]

For details see Abbreviations.

 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
to reduce the amount of effort needed in an experiment by elimi-
nating redundant observations. Thus, the main advantages of DOE
are [23]:
1. Fewer experiments are needed because several factors are varied
simultaneously.
2. The judicious choice of factor settings reduces the number of
experimental runs.
3. More information per experiment is generated.
4. Applications of electronic noses in food analysis
A search of the recent, relevant literature shows that there are
five major categories of use for electronic noses in food control.
These are (i) process monitoring, (ii) shelf-life investigation, (iii)
freshness evaluation, (iv) authenticity assessment and (v) other
quality control studies. The most important contributions to each
of these in the present century will be described.
4.1. Process monitoring
Several successful applications of the electronic noses to the
monitoring of flavor and/or aroma components along a food pro-
duction process have been published. The main features of such
applications are summarized in Table 1.
Aroma production along a grape must fermentation has been
monitored during bioconversion [24]. In this study, the musca-
tel aroma was chosen because the profile formed as a result
of yeast metabolism is complex, being composed of many com-
pounds. These differ from each other in concentration, chemical
and organoleptic properties and contribute to the overall muscatel
aroma. This, therefore, was a challenging project. A commercially
available e-nose (A32S AromaScan) consisting of 32 organic con-
ducting polymer-based sensors was used. Data analysis was carried
out by PCA. The authors found that without sample pretreatment,
the nose could only detect ethanol production, while for small
quantities of muscatel a selective enrichment step was needed.
Once this was done, the electronic nose was able to discriminate
samples based on aroma content. However, enrichment increases
the time and effort required for analysis, which could prove to be a
major disadvantage to this approach. Additionally, there is the prob-
lem that there is competition between the aroma compounds and
ethanol for detection by the sensor. As ethanol is present in higher
concentration, it will be detected preferentially to the aroma.
A further complication is that ethanol may also interfere during
the headspace sampling. Ethanol acts as a co-solvent in the aqueous
wine-must matrix and so the activity coefficient of the hydrophobic
aroma compounds in the aqueous-phase is lowered, resulting in a
decreased partitioning into the sample headspace [24]. This can
lead to erroneous results with the electronic nose.
García et al. [25] have made use of an e-nose to identify spoiled
Iberian hams during the curing process. The sensors involved were
tin-oxide semiconductor thin films, some of which were doped with
metal catalysts such as Cr and In. A good discrimination (success
rate of 100%) of two types of Iberian hams (spoiled and unspoiled)
was obtained through the statistical methods of PCA and proba-
bilistic ANN (PNN).
E-noses have been also applied to bioprocess monitoring where
microbiological processes are involved in food production, i.e. to
screen the aroma generation of lactic acid bacteria strains in the
production of cheese and other fermented dairy products. In a paper
by Marilley et al. [26], PCA inspection of a MS-based e-nose (Smart
Nose® ) data allowed the discrimination between 7 different geno-
types strains of Lactobacillus casei isolated from Gruyère cheeses.
This classification of strains based on the production of volatile
5
compounds was in conformity with the classification obtained
with the repetitive extragenic palindromic polymerase chain reac-
tion (REP-PCR) molecular method. Gutiérrez-Méndez et al. [27]
used a MOS based e-nose (FOX 3000) to screen the aroma gen-
eration of Lactococcus lactis strains isolated from different dairy,
non-dairy and industrial sources for their potential use in starter
cheese cultures. PCA evaluation of e-nose data showed a clear sep-
aration of 4 sample groups based on their odor intensity scores
(yogurt-like and Fresco cheese-like), but not based on the isolation
source.
During the winemaking process, unpleasant organoleptic taints
arise from Brettanomyces yeasts spoilage. The two main compo-
nents of the taint are 4-ethylphenol (4EP) and 4-ethylguaiacol
(4EG). The existing procedures to monitor spoilage due to Bret-
tanomyces/Dekkera sp. are time-consuming and expensive, making
it difficult for winemakers to monitor their wines at all stages of pro-
duction. Consequently, there is a need for a rapid and cost-effective
screening method to monitor the levels of 4EP and 4EG in wine.
In this way, Cynkar et al. [28] used a MS-based e-nose (HP4440)
together with PCA and stepwise linear discriminant analysis (SLDA).
On the other hand, Berna et al. [29] compared the performance of
a MOS sensor based e-nose (FOX 3000) and a MS-based e-nose.
GC–MS was used for quantification and prediction purposes. Fol-
lowing ethanol removal and SPME sample handling, the limits of
detection of a MOS based e-nose were determined as 44 ␮g L−1
for 4EP and 91 ␮g L−1 for 4EG (values significantly lower than the
reported human sensory thresholds). Partial least sequares (PLS)
regression of MOS based e-nose signals against known levels of
4EP and 4EG in 46 Australian red wines showed that such device
was unable to identify Brettanomyces spoilage reliably because of
the response of the gas sensors to inter-sample variation in VOCs
other than 4EP and 4EG. Conversely, the MS-based e-nose (SHS
sample handling without ethanol removal but selecting a window
scan excluding ethanol derived ions) was capable of reliably esti-
mating concentrations of 4EP higher than 20 ␮g L−1 and good PLS
correlations were obtained between estimates of 4EP and 4EG con-
centrations with the concentrations determined by conventional
GC–MS.
Pani et al. [30] used a MOS based e-nose (Air Sense) for
monitoring the changes in aroma profile of tomato slices during
air dehydration processes. Two kinds of samples (untreated and
osmodehydrated in corn syrup) were studied. E-nose data analysis
by means of PCA was able to characterize the process aromatic fin-
gerprint, which could be helpful to understand and parameterize
the degradative events caused by dehydration.
Lebrun et al. [31] undertake a study to discriminate between
mango fruit maturity by volatiles using a FOX 4000 e-nose (com-
prising 18 metallic oxide sensors) and GC. Three different mango
fruit varieties (Mangifera indica L.) were harvested at different
maturities and at different sizes. Immediately after harvest (green)
or after 1 week of ripening at room temperature (ripe), fruit were
homogenized or left intact and evaluated by the e-nose or by GC
for aroma and other volatiles as well as for soluble solids and acids.
Volatile data from the different harvest maturities and ripening
stages were discriminated by using discriminant factor analysis
(DFA). Both the e-nose and GC were able, in most cases, to sepa-
rate fruit from different harvest maturities (at both the green and
ripe stages) as well as discriminate green from ripe fruit and fruit
from the different varieties within a maturity stage. Solids and acids
data indicated that later harvest maturities resulted in sweeter fruit
and later-harvested fruit had a different volatile profile from earlier-
harvested fruit. These results demonstrate the benefit that could be
obtained if a hand-held e-nose device were developed that could
determine optimal harvest maturity for mangoes on the tree by
the volatiles emitted, or an e-nose device that could be used as a
screening tool on fruit after harvest.
6 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

Table 2
Main applications of e-noses in food shelf-life investigation.

Sample Type of study Sample Detection system Data processing Ref.

handling algorithm
system

Pinklady apples Discrimination between ripening stages SHS 21 MOS PCA, ANN, PLS [33]
Estimation of fruit quality parameters

Apples Post-harvest ripening SHS 12 QMB Radial plots [34]

Both qualitative (type of apple) and
quantitative identification
Tomatoes (Lycopersicon Discrimination between ripening states SHS (for Libra LibraNose: 5 QMBs PCA [35]
esculentum Mill.) Nose)
MS
Discrimination between cultivars SPME (for MS)

Jonagold apples Discrimination between shelf-life durations SPME (for both LibraNose: 7 QMBs PCA, PLS [36]
e-noses)
Discrimination between storage conditions MS
Prediction of firmness and days of shelf-life

Tomatoes ‘Heatwave’ Discrimination between ripeness states SHS PEN 2: 10 MOS PCA, LDA, PLS [37]
(Lycopersicum
esculentum)
‘Zaojin Jiaogan’ mandarins Discrimination between storage shelf-life SHS PEN 2: 10 MOS PCA, LDA, PLS [38]
Discrimination between storage treatments
Prediction of fruit quality characteristics

Heatwave tomatoes Discrimination between storage shelf-life SHS PEN 2: 10 MOS PCA, LDA, PLS [6]
(Lycopersicum times during two storage treatments
esculentum)
Prediction of fruit quality characteristics

Peaches (Prunus persica L.) Discrimination between cultivars and SHS PEN 2: 10 MOS PCA, LDA, CART [39]
between ripenning states during shelf-life
Crescenza cheese Definition of the threshold of the shelf-life at SHS Model 3320: 10 MOFSET + 12 MOS PCA, CA, LDA [40]
different storage temperatures
Milk Determination of shelf-life SHS FOX 4000: 18 MOS PCA, vectors [42]
norm analysis
Correlation with bacteria growth

Extra virgin olive oils Evaluation of the oxidative status at different SHS Model 3320: 10 MOFSET + 12 MOS PCA, LDA [43]
storage conditions
Refined raspeed oil Evaluation of lipid autoxidation under SHS FOX 4000: 18 MOS PCA, PLS [44]
different storage conditions

For details see Abbreviations.

During black tea manufacturing, tealeaves pass through a fer-
mentation in which the grassy smell is transformed into a floral
smell. Optimum fermentation is extremely crucial in deciding the
final quality of finished tea and it is very important to terminate the
fermentation process at the right time. Bhattacharya et al. [32] pre-
sented a study on real-time smell monitoring of black tea during the
fermentation process using an e-nose (8 MOS sensors array) as well
as prediction of the correct fermentation time. Different time-delay
neural networks (TDNNs) and self-organizing map (SOM) methods
for the prediction of optimum fermentation were used and both
the methods appear to be suitable for the purpose. However, the
combined SOM- and TDNN-based prediction algorithm proved to
be the better alternative as the computational complexity is rela-
tively less. The results showed excellent promise for the instrument
to be used for the on-line prediction of optimum fermentation time
by the industry.
4.2. Shelf-life investigation
Many e-nose applications in the literature (Table 2) are devoted
to monitor the ripening process of fruits and other vegetables
during their shelf-life period (from harvest until consumption)
[6,31,33–39]. Monitoring and controlling ripeness is a very impor-
tant issue in fruit and vegetables management since it is a very
important quality indicator for consumers. Many methods to mon-
itor fruit ripeness have already been proposed but they are not
useful for packinghouses and most of them require the destruc-
tion of samples used for analysis. Therefore, predictions of shelf-life
ripeness state are mainly based on practical experience. Leaving
these critical decisions to subjective interpretation implies that
large quantities of fruit and vegetables reach consumers markers
in poor condition [33]. In this framework, e-noses have proved to
be promising tools for fruit ripeness assessment.
Brezmes et al. [33] used an e-nose, consisting of 21 MOS sensors
array, to assess the ripeness state of pinklady apples through their
shelf-life. In order to evaluate the e-nose performance, fruit qual-
ity indicators, such as firmness, starch index and acidity, were also
obtained to compare results from both techniques. Pinklady apples
were harvested at their optimal date so that e-nose and fruit qual-
ity measurements could be performed on the fruit samples during
their ripening process. A PCA analysis did not show any cluster-
ing behavior that might be attributed to ripening. On the other
hand, fuzzy art, an unsupervised ANN classification algorithm,
showed a tendency to classify measurements regarding to their
self-life period. Finally, good correlation coefficients were obtained
between e-nose signals and classical fruit quality parameters (firm-
ness and acidity) by means of PLS modelling, thus indicating that
e-nose signals are related to the ripening process of apples.
In the paper by Herrmann et al. [34] an e-nose based on arrays
of differently coated QMB has been used to discriminate between
VOCs formed during the post-harvest ripening of apples. The com-
pounds monitored were aldehydes and esters. The relative ratios
of these compounds change during post-harvest ripening and this
allows them to be analyzed by PR methods. This is due to the
 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
formation of characteristic patterns of sensor responses. During
the ripening of apples, trans-2-hexenal can serve as an indicator
compound because its concentration increases significantly. It was
found that the detection limit of trans-2-hexenal was 20 mL m−3 .
Both qualitative (type of apple) and quantitative identification were
possible. However, the correlation between the vapor concentration
and sensor response is linear within a limited concentration range.
It was found that the shape of the curve is similar to the Langmuir
adsorption model, thereby limiting the range, which could be used.
The performance of a QMB coated by modified metallo-
porphyrins and related compounds based e-nose (LibraNose) and
a MS-based e-nose for tomato aroma profiling was evaluated by
Berna et al. [35]. SPME headspace sampling combined with GC were
used as a reference method. In the first experiment, the changes in
tomato aroma profiles of two different cultivars were monitored
during shelf-life (days 1, 8, 12 and 19). The score plot of PCA for the
e-nose measurements showed a slight shift along the first princi-
pal component corresponding to an increasing number of days of
shelf-life. However, the tomato aroma profiles measured on days
1 and 8 could not be discriminated by the e-nose. In contrast, MS-
nose score plots indicated an evident change in aroma profile with
shelf-life. The loading plots of the PCA revealed that mainly three
discriminant mass to charge ratios are responsible of the observed
shelf-life effect in both cultivars. These fragments indicate that
␤-phellandrene, 6-methyl-5-hepten-2-one, 1-nitropentane and 2-
methylbutanol undergo important concentration changes during
shelf-life of tomatoes. In the second experiment, the aroma differ-
ences between cultivars were analyzed by means of the electronic
nose systems. A clear distinction between cultivars based on MS-
nose measurements was obtained; however, based on QMB e-nose
measurements it was difficult to discriminate between tomato vari-
eties in two dimensions.
The potential of a Libra Nose e-nose and a MS-based e-nose to
monitor changes in apple fruit volatiles during shelf-life has been
studied by Saevels et al. [36]. These techniques were compared with
a traditional technique to measure volatiles, GC–MS. Apples were
stored for 8 months at three different storage conditions and the
volatile profile changes were followed subsequently over a period of
15 days. Analysis of the PCA score plot for the e-nose measurements
showed no storage history effect and only very little shelf-life effect.
In contrast, the MSE-nose and GC–MS score plots clearly indicated
the presence of both shelf-life and storage history trend. Moreover,
the volatile profile changes during shelf-life depended on the stor-
age history. The loading plots of the PCA of the GC–MS data revealed
which volatiles are important to differentiate between storage con-
ditions and which ones are important during ripening on the shelf.
PLS models based on the three data sets to relate firmness and days
of shelf-life with the volatile production of apples were built based
on the three data sets. It was found that the models based on the
e-nose data had worse prediction performance than those based on
the MSE-nose data.
The possibility of exploiting information on behavior aroma to
assess fruit ripening stage has been the cornerstone of the work
of Hernández-Gómez et al. [37]. The objective of this study was to
evaluate the capacity of e-noses to monitor the change in volatile
production of ripeness states for tomato, using a specific e-nose
device with 10 different metal oxide sensors (portable e-nose,
PEN 2). PCA and linear discriminant analysis (LDA) were used to
investigate whether the e-nose was able to distinguishing among
different ripeness states (unripe, half-ripe, full-ripe and over-ripe).
The loadings analysis was used to identify the sensors responsible
for discrimination in the current pattern file. The results prove that
the PEN 2 e-nose could differentiate among the ripeness states of
tomato. The electronic nose was able to detect a clearer difference in
volatile profile of tomato when using LDA analysis than when using
PCA analysis. Using LDA analysis, it was possible to differentiate and
7
to classify the different tomato maturity states, and this method was
able to classify 100% of the total samples in each respective group.
Some sensors in e-nose have the highest influence in the current
pattern file for PEN 2 e-nose. A subset of a few sensors in e-nose
can be chosen to explain all the variance. This result could be used
in further studies to optimize the number of sensors.
Hernández-Gómez et al. [38] also evaluated the capacity of PEN
2 e-nose in monitoring the change in volatile production of man-
darins during different storage treatments, as well as their storage
shelf-life. They found that storage shelf-life of mandarin was better
distinguished utilizing LDA than PCA. Both were efficient to clas-
sify mandarins with the same storage time in its respective groups,
but not to separate the mandarins by different storage times. The
PLS correlation between the measured and predicted values of fruit
quality attribute (such as compression force, acidity, and soluble
solid content) showed poor to reasonable prediction performance
by means of e-nose signals. According to the results obtained, only
a few sensors of the e-nose device are required to get a good per-
formance.
In a posterior work [6], monitoring of tomato storage shelf-
life during two storage treatments (plastic folded bags and carton
boxes) using PEN 2 e-nose was studied. Measurements spanned
over 12 days after harvest (at 0, 3, 6, 9 and 12 days shelf-life). By PCA
and LDA, e-nose could clearly discriminate storage times of tomato
in carton box, but not in the case of folded bag packing where groups
overlapped lightly. On the other hand, as in the previous work, the
PLS correlations between the measured and predicted values of
fruit quality attribute showed poor prediction performance on the
base of signals of e-nose sensors.
Benedetti et al. [31] evaluated the capability of a PEN 2 e-nose
to classify four Prunus persica L. peach cultivars and to assess their
ripening stage during shelf-life. Firstly, PCA and LDA were carried
out on peaches using the e-nose sensors responses collected the day
after harvest and both techniques were able to distinguish among
cultivars. Secondly, PCA applied on the e-nose data (collected dur-
ing shelf-life from harvest until senescence) revealed that only one
sensor (W5S) was relevant to differentiate peaches during ripeness
on the basis of their shelf-life, and classified peaches as unripe, ripe
and over-ripe. Classification and regression tree (CART) analysis
classified samples in each respective group with cross-validation
error rate of 4.87% and established limit values of the W5S to split
off the three classes. The performance of the e-nose was compared
with the results of classical and non-destructive techniques such
as ethylene measurement and color evaluation, frequently used to
assess the ripening stage of climacteric fruit.
As shown in Table 2, in addition to the evaluation of fruit and
vegetables ripeness states, other e-noses applications to shelf-life
investigation have been performed in cheese [40,41], milk [42] and
oil samples [43,44].
The shelf-life of a traditional soft cheese was measured by
Benedetti et al. [40] making use of a commercial e-nose (model
3320, Applied Sensor Laboratory Emission Analyser). A clear dis-
crimination between “fresh”, “aged”, and “very aged” samples was
obtained using PCA, cluster analysis (CA) and LDA statistical tech-
niques. The predictive ability of the LDA classification model was
confirmed by considering a new set of cheese samples purchased
at the beginning of their commercial life and analyzed until their
expiration date.
The combination of e-nose technology and image analysis was
utilized by Riva and Mannino [41] to evaluate the shelf-life of min-
imally processed cicorino as well as Italian cheese. For cicorino, a
good correlation was achieved between traditional microbiological
parameters and both color change, measured by image digital-
ization, and e-nose headspace data. In the case of the cheese, a
satisfactory correlation was obtained between proteolysis param-
eters commonly used to monitor cheese ripening and image
8 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

parameters of hue and hole growth. Kinetic analysis of data from
both studies was used to define a stability time for both foodstuffs
and to predict shelf-life at different storage temperatures.
Labreche et al. [42] have resorted to the use of an e-nose (FOX
4000) to determine the shelf-life of milk. It was stored at room
temperature and at a constant temperature of 5 ◦ C. Samples of each
of the stored milk were analyzed by the e-nose at different times.
The authors present a mathematical method that suggests some
interesting time-events (found in the literature on milk shelf-life)
that correspond to significant ones in the milk evolution or ageing.
In the work of Cosio et al. [43], the oxidation of extra virgin olive
oils was considered at different storage periods and conditions. The
oxidation is usually evaluated by applying an accelerated thermoxi-
dation, while in this case real storage conditions were used. In order
to study the differences of the storage situations, multivariate sta-
tistical analysis was applied on classical chemical determinations
(acidity, peroxide value, and UV absorbance at 232 and 270 nm),
electronic nose and electronic tongue responses. Results showed
how the e-nose (model 3320, Applied Sensor Lab Emission Anal-
yser) was enough to define the extra virgin olive oil oxidation and
appeared to be able to describe the different storage conditions,
while classical chemical parameters and electronic tongue were not
relevant. In fact, the classification model built by means of LDA gave
an equal classification performance by considering all the variables
or just the electronic nose sensor responses. Compared to classi-
cal methods, this new approach could represent an alternative and
innovative tool for faster and cheaper evaluation of extra virgin oil
oxidation.
The work by Mildner-Szkudlarz et al. [44] was aimed at assess-
ing the effectiveness of the FOX 4000 e-nose to monitor off-flavor
associated with raspeed oil lipid autoxidation as a supplementary
tool to human sensory panel assessment. Refined rapeseed oil was
subjected to an accelerated storage test for 12 days at 60 ◦ C and to
an ambient temperature storage test in which it was stored in retail
plastic bottles for up to 6 months. A specialized panel examined six
odor attributes of samples. Also, SPME–GC–MS analyses and chem-
ical analyses of oil samples were run. PCA of e-nose data samples
stored at an elevated temperature was related to PCA of sensory
analysis, and similarities in sample clustering were observed. How-
ever, for samples stored at room temperature, the human panel
showed greater sensitivity than the e-nose. On the other hand,
prediction models based on PLS of e-nose data were able to pre-
dict the sensory quality changes during storage at elevated and
room temperature. Peroxide values (PV) and anisidine values (p-
AV) measured by means of wet analysis were well predicted on
the basis of e-nose and sensory analysis data for both storage tem-
peratures. Moreover, PLS analyses performed on electronic and
human nose data with particular volatile compounds (measured by
SPME–GC–MS) gave significant correlations. This paper points out
the applicability of the e-nose technology to verify sensory and ran-
cidity changes during storage is promising in routine quality control
of oils.
4.3. Freshness evaluation
Freshness is another important quality property in the food
industry. Since a number of different VOCs are generated during
storage of foods the electronic noses have shown their potential in
predicting freshness or spoilage of different food raw material and
products. In particular, foods where significant release of volatiles
occur during storage due to rapid degradation by bacterial pro-
cesses, such as fish [45–49], oysters [50], shrimps [51], eggs [52],
soybean curds [53] and meats [54,55] (Table 3).
Most freshness investigations with electronic noses have
involved studies with fish and fish products. In one of these, Di
Natale et al. [45] used the measurements of two e-noses, based
on different sensor technologies and sampling methodologies, to
detect freshness of cod-fish fillets. One of the e-noses used con-
sisted on an array of eight thickness shear mode resonators coated
with various kinds of metaloporphyirns (LibraNose), and the other
one (FreshSense) is based on five electrochemical sensors each ori-
ented towards a certain gas (CO, H2 S, NO, SO2 and NH3 ). E-noses
data were analyzed by means of partial least square-discriminant
analysis (PLS-DA). Over a period of 17 storage days, the two sen-
sor systems showed different resolution, while the integration of
both e-noses improved the performances allowing an almost com-
plete evaluation of the freshness of samples. Trimethylamine and
total volatile basic nitrogen, measured with conventional tech-
niques (flow injection analysis-gas diffusion method), showed a
non-monotonic behavior, that induces the possibility of large errors
in freshness estimation.

Table 3
Main applications of e-noses in food freshness evaluation.

Sample Type of study Sample handling Detection system Data processing Ref.
system algorithm

Cod-fish fillets Discrimination between storage periods SHS LibraNose: 8 TSM PLS-DA [45]
FreshSense: 5 EC

Fresh Atlantic salmon Spoilage classification of salmon fillets SHS AromaScanTM : 32 CPs MDA [46]
(Salmo salar) during storage at different temperatures
Cold smoked Atlantic Spoilage/freshness classification during SHS FishNose (GEMINI): 6 MOS PLS, PCA [47]
salmon (Salmo salar) storage at different temperatures
Alaska pink salmon Spoilage classification under different SHS Cyranose 320TM : 32 PCA, FSGDA [48]
(Oncorhynchus storage conditions thin-film carbon-black
gorbuscha) polymer sensors
Fresh tilapia fillets Discrimination between storage times of SHS e-Nose 4000: 12 CPs DFA [49]
(Oreochromis fillets under different treatments
niloticus)
Oysters (Cassostrea Prediction models for odor changes in SHS EEV model 4000: 12 CPs DFA [50]
virginica) shucked oysters
Eggs Establishment of freshness categories SHS 4 Tin-oxide sensors PCA, FCM, SOM, [52]
ANN
Predicion of the state of freshness

Ground beef Rancidity detection SHS FOX 3000: 12 MOS QLSR [54]
Beef and sheep meats Spoilage classification and bacteriological SHS 6 Tin-oxide sensors PCA, SVM, PLS [55]
parameters prediction

For details see Abbreviations.

 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
A quality assessment of salmon fillets under various storage con-
ditions was reported by Du et al. [46]. The fillets were stored at −20,
4 and 10 ◦ C for 14 days and examined by the AromaScan e-nose for
changes in bacteria and histamine over time. Comparison with sen-
sory panel evaluations showed that this approach might be valuable
in evaluating fish quality.
Quality changes of cold smoked salmon from 4 different smoke-
houses in Europe were monitored by a prototype MOS sensors array
system, the FishNose [47]. Samples were stored in different pack-
aging (vacuum and modified atmosphere) for up to 4 weeks under
controlled storage conditions at 5 and 10 ◦ C. Quality criteria based
on sensory attributes (sweet/sour, off, and rancid odor), and total
viable counts and lactic acid bacteria counts were established and
used for classification of samples based on the responses of the
FishNose. The responses of the gas-sensors correlated well with
sensory analysis of spoilage odor and microbial counts suggest-
ing that they can detect volatile microbially produced compounds
causing spoilage odors in cold-smoked salmon during storage. The
system is therefore ideal for fast quality control related to fresh-
ness evaluation of smoked salmon products. PLS regression models
based on samples from single producer showed better performance
than a global model based on products from different producers to
classify samples of different quality.
The ability of a portable hand-held e-nose (Cyranose 320TM ,
composed of 32 individual thin-film carbon-black polymer sensors)
in detecting spoilage of salmon under different storage conditions
(at 14 ◦ C and in slush ice) was investigated by Chantarachoti et al.
[48]. As a result of these experiments, a predictive model may be
developed for spoilage of whole Alaska pink salmon by analyzing
belly cavity odors using the e-nose. This could be easily extended
to other types of fish.
The ability of an electronic nose and machine vision system to
classify tilapia fillets based on their odors and colors was studied
by Korel et al. [49]. Fresh tilapia (Oreochromis niloticus) fillets were
treated with different amounts of sodium lactate and stored at 1.7
and 7.2 ◦ C for 12 days. Trained panelists and an e-nose compris-
ing 12 CPs sensors (e-Nose 4000) evaluated odors and a machine
vision system measured fillet colors. Correct classification based
on experimental variables (percent lactate, microbial load, sensory
scores, storage time, and temperature) by DFA was poor for color
data alone, acceptable for e-nose data alone, and excellent with
these data combined.
Tokusoglu and Balaban [50] proposed the combination of e-nose
technology and computer vision for objective evaluation of oysters
freshness. Odor and color changes in oysters (Crassostrea virginica)
stored at 1.8 and 7 ◦ C were measured every 3 days for up to 13
days with an e-nose (model 4000; EEV Inc.), a computer vision sys-
tem, and sensory panels. Electronic nose and odor sensory data was
analyzed using DFA discriminant function analysis. Correct classi-
fication rates of 100% were obtained to group e-nose data by days
and sensory scores at each temperature. When all temperature and
days data were pooled, DFA predicted sensory scores based on e-
nose readings with 94% accuracy. Oyster colors were analyzed using
the 512 color block scheme and eight colors were found to be signif-
icant. The correlations of these colors, as well as the average L* a* b*
values with time and with color sensory data suggested prediction
of color quality.
An e-nose with 12 CPs sensors was used to measure odors of
raw shrimp treated with different chemicals [51]. Headless shell-on
pink shrimp (Pandalus jordani) were treated with different amounts
bleach, phosphates and sulfites and stored at 2 ◦ C for 48 h. Odors
were evaluated by sensory panels and the e-nose; moreover aerobic
plate counts were performed. DFA results showed that the e-nose
could discriminate differences in odor due to chemicals present in
shrimp. The correct classification rates for bleach, phosphate and
sulfite treated shrimp were 92.7%, 95.8%, and 99.2%, respectively.
9
Dutta et al. [52] used a tin-oxide sensors based e-nose system to
analyse the state of freshness of eggs. The sensor system comprised
four inexpensive commercial tin-oxide odor sensors responding to
different gases such as cooking vapors, ammonia, hydrogen sulfide
and organic solvents (i.e. alcohol, toluene, xylene, etc.). Measure-
ments were taken from the headspace of four sets of eggs over a
period of 20–40 days, two ‘types of egg data’ being gathered using
the e-nose; one type of ‘data’ related to eggs without a hole in the
shells and the other type of ‘data’ related to eggs wherein tiny holes
in the shells were made. PCA, fuzzy C means, self-organizing maps
and 3D scatter plots were used to define regions of clustering in
multisensor space according to the state of freshness of the eggs.
These were correlated with the ‘use by date’ of the eggs. Then four
ANN supervised classifiers (multilayer perceptron, learning vector
quantisation, probabilistic neural network and radial basis function
network) were used to classify the samples into the three observed
states of freshness. The best results suggested the ability to predict
egg freshness into one of three states with up to 95% accuracy. This
shows good potential for commercial exploitation.
The effectiveness of an e-nose at determining the freshness of
soybean curd was investigated by Youn and Noh [53]. Soy curds
were stored at temperatures sequentially fluctuating between 4 and
10 ◦ C. Kinetic data was subjected to principal component analysis
in order to predict freshness, on the basis of production of volatile
compounds. The difference between predicted and actual freshness
of soy curd was less than 8.9%.
McElyea et al. [54] used a FOX 3000 e-nose to determine changes
in lipid oxidation and microbial load of ground beef throughout
simulated retail display. Aerobic, vacuum and CO2 mixing treat-
ments were used to determine their impact on e-nose responses,
lipid oxidation and microbial characteristics. After grinding and
mixing, ground beef was stored under simulated retail display, and
analyzed at days 0, 1, 2, 3, 6, and 10. Analyses included thiobarbituric
acid reactive substances (TBARS, a widely used procedure to esti-
mate lipid oxidation of meat), aerobic plate count (APC) and e-nose
characteristics. The e-nose detected changes in ground beef lipid
and microbial stability as did conventional TBARS and APC mea-
sures. Therefore, the e-nose may hold promise for rapid detection
of meat freshness and safety.
El Barbri et al. [55] used an e-nose system containing an array
of six tin-oxide gas sensors for the quality control of red meat. E-
nose and bacteriological measurements were performed to analyze
samples of beef and sheep meat stored at 4 ◦ C for up to 15 days.
First, PCA and support vector machine (SVM) based classification
techniques were used to investigate the performance of the e-nose
system in the spoilage classification of red meats. PCA (a linear tech-
nique) could be used for spoilage classification of beef meat but not
in the case of sheep meat. A very good success rates in the classi-
fication of spoiled or unspoiled beef and sheep meats (98.81% and
96.43%, respectively) were obtained when (non-linear) SVM were
employed. On the other hand, PLS regression models showed good
correlation coefficients between the e-nose signals and bacteriolog-
ical for beef and sheep meats. According to these results, an e-nose
system can become a simple, fast and non-destructive alternative
tool to bacterial analysis for shelf-life determination (i.e. quality
assessment) and spoilage classification (safety assessment) of red
meats.
4.4. Authenticity assessment
There is no doubt that, in terms of identification, alcoholic bever-
ages provide the best-known example of electronic noses. However,
other foods, which have been analyzed for identification purposes,
include olive oil, cheese, honey, vegetable oil, fruit juices and vine-
gars [56] (Table 4).
10 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

Table 4
Main applications of e-noses in food authenticity assessment.

Sample Type of study Sample handling Detection system Data processing Ref.
system algorithm

Tequila, whisky, vodka Discrimination between the four DHS FOX 4000: 18 MOS PCA [57]
and red wine types of beverages
Discrimination of wines from
different regions
Italian wines Recognition and quantitation of SHS 4 Thin-film MOS PCA, BP-ANN [58]
adulterations
Spanish wines Classification of wines varieties SHS, P&T, SPME 16 Tin-oxide sensors PCA, PNN [59]
Spanish wines Classification of wines varieties SHS 8 Tin-oxide sensors PCA, PNN [60]
Spanish wines Classification of wines varieties SHS ZnO SAW PCA, PNN [61]
Spanish wines Classification of wines varieties, SHS MS PCA, SIMCA [62]
origins and ageing
Virgin olive oils Detection of adulterations SHS FOX 3000: 12 MOS LDA, QDA, ANN [63]
Extra virgin olive oils Discrimination between SHS MS PCA, SLDA, LDA [64]
geographical origins
Extra virgin olive oils Discrimination between SHS Model 3320: 10 MOFSET + 12 MOS PCA, CP-ANN [65]
geographical origins
Orange juices Discrimination between SHS FOX 3000: 12 MOS PCA, FDA [66]
geographical origins
Emmental cheese Discrimination between SHS Smart Nose® : MS PCA [67]
geographical origins
Swiss unifloral honeys Discrimination between SHS, SPME, Smart Nose® : MS PCA, DFA [10]
botanical origin of honey INDEX
‘Aceto Balsamico Classification of different aged SHS MS PARAFAC, PCA, [68]
Tradizionale di products SIMCA, WPTER
Modena’

For details see Abbreviations.

Among applications to alcoholic beverages authenticity assess-
ment, successful applications to differentiation of wines on the
basis of geographical origin and grape variety have been reported in
the literature, as well as for the recognition of adulterations [57–62].
French researchers have coupled GC with an e-nose to identify
alcoholic beverages [57]. As discussed previously, high concentra-
tions of ethanol tend to affect the detection of aroma compounds
by an electronic nose. However, pretreatment of samples before
GC by dehydration and de-alcoholization solved this problem. A
semi-dynamic headspace at controlled temperature was used for
pretreatment of samples and also to standardize aroma composi-
tion. Tequila, whisky, vodka and red wine were analyzed and four
compounds responsible for off-flavor in red wine were detected.
PCA allowed discrimination between the four types of beverages
and wines from different regions of France were also distinguished.
Penza and Cassano [58] fabricated a set of four thin-film (WO3 )
metal oxide sensors, surface-activated by Pt, Au, Pd, Bi metal cata-
lysts, for recognition some adulterated Italian wines (two white,
four red, two rosè from different denominations of origin and
vintages) added with methanol, ethanol or another same-color
primary wine. Multivariate analysis including PCA and BP-ANNs
(back-propagation artificial neural networks) were used to identify
both the adulteration of wines and to determine the added content
of adulterant agent of methanol or ethanol up to 10% volume. The
cross-validated ANNs provided the highest achieved percentage of
correct classification of 93% and the highest achieved correlation
coefficient of 0.997 and 0.921 for predicted vs. true concentration
of methanol and ethanol adulterant agent, respectively. It must be
noted that since the aim of this work was to determine adulterants
(alcohols and other same-color wine compounds) at high concen-
tration levels, SHS sample handling system was chosen due to its
high sensitivity towards this compounds.
A comparison among several sampling systems usually
employed in an e-nose is performed in the work of Lozano et al. [59]
to improve the performance of this instrument for wine discrim-
ination. Sampling handling methods studied were SHS, P&T and
SPME. A 16 dopated tin-oxide sensors array was home-prepared.
These e-noses were developed in order to discriminate five different
Spanish wines coming from different grape varieties and elabo-
ration processes (Malvar, Airén, Garnacha, Tempranillo and Aged
Tempranillo). Linear techniques as PCA and non-linear ones as PNN
were used for pattern recognition. Although the highest response
was obtained by the SHS method, due to the amount of ethanol
extracted by this method, the best discrimination was achieved
with the two other methods. Using P&T and SPME, the separation
of different clusters with PCA and the classification performed with
PNN was major (100% and 95.8% success rates for SPME and P&T,
respectively). The choice of the technique of extraction could be
determined by the final application of the prototype: SHS and P&T
could be more suitable for in situ and portable systems whereas
SPME could be used in laboratory.
In a posterior work Aleixandre et al. [60] developed a portable
e-nose containing two different micromachined resistive sensor
arrays, one with a polysilicon integrated heater and the other with
a platinum one (each sensor array contains four tin-oxide sensors).
This device was also tested with four different wines coming of
Madrid region (Malvar, Airén, Garnacha and Tempranillo). The PCA
plots showed a slight overlapping of the responses for the sensor
array with platinum heater and an 88% of classification result was
obtained with a PNN. With the sensor array with polysilicon heater
a good separation was achieved by the PCA and the PNN classifi-
cation success was 100%. Results were comparable to others done
with non-portable e-noses and the same wines [59,61].
A MS-based e-nose combined with PCA and SIMCA (soft inde-
pendent modelling of class analogy) was successfully applied by
Martí et al. [62] for the differentiation and classification of wines
according to different oenological parameters (variety, origin and
ageing). SHS sample handling system was used and ethanol interfer-
ence was avoided by selection of the proper fragment-ion range for
the MS. The main advantages of the proposed methodology are the
minimum sample preparation required and the speed of analysis
(10 min sample−1 ). The study was conducted in three sample sets:
(i) for the study of varietal wines (Cabernet sauvignon, Tempranillo
and Merlot wines) samples from different Catalonian wine growing
zones and three vintages were chosen; (ii) the samples chosen in
the study of wine origin differentiation were from Priorat and Terra
 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
Alta (most of them blends of two varieties). (iii) In the study of age-
ing differentiation, the wines chosen were from Ribera del Duero (a
very important Spanish wine growing zone) where the commercial
wines are classified according to their ageing process, by a regu-
latory council, into four different administrative categories: Young,
Crianza, Reserva and Gran Reserva wines.
Another major application area of e-noses is authenticity assess-
ment of olive oils. Successful applications of e-nose technology for
the differentiation of olive oils on the basis of geographical origin
and to detected adulterations have been reported [63–65].
Cerrato Oliveros et al. [63] used FOX 3000 e-nose for the
detection of adulterations of virgin olive oil with sunflower oil
and olive-pomace oil. Prior to different supervised PR treatments,
feature selection techniques were employed to choose a set of opti-
mally discriminant variables. Multivariate chemometric techniques
such as LDA, quadratic discriminant analysis (QDA) and ANN were
applied. Excellent results were obtained in the differentiation of
adulterated and non-adulterated olive oils (prediction percentages
>95%, >85% for ANNs) and it was even possible to identify the type of
oil used in the adulteration. Satisfactory results were also obtained
as regards classification of the samples as a function of the percent-
age of adulteration.
In a posterior work the same authors [64] used an MS-based
e-nose for the discrimination of 105 extra virgin olive oils coming
from five different Mediterranean areas. Chemometric analysis of
data was performed using PCA, stepwise LDA and LDA. The 93.4%
of samples resulted to be correctly classified, the 90.5% correctly
predicted by cross-validation procedure and the 80.0% of an exter-
nal test set were correctly assigned. It is important to underline
that the headspace generation temperature was optimum at low
values. Indeed, high extraction temperatures might promote oxi-
dation and degradation reactions, whose results could confuse the
delicate and light aroma of Ligurian olive oils with those of the other
Mediterranean olive oils.
An electronic nose (model 3320 Applied Sensor Lab Emission
Analyser) and an electronic tongue, in combination with multi-
variate analysis, have been used by Cosio et al. [65] to verify the
geographical origin and the uniqueness of specific extra virgin olive
oils. The olive oil samples belong to a small production, located in
the lake of Garda (north of Italy) and distinguished with a Euro-
pean Protected Denomination of Origin trademark since 1998. In
order to obtain a complete description of oil samples, free acidity,
peroxide value, ultraviolet indices, and phenol content were also
determined. The dataset included 36 Garda oils and 17 oils from
other regions. Two classification models were built by means of
counterpropagation artificial neural networks (CP-ANN) in order to
separate Garda and not-Garda oils, as follows: first, by using all the
chemical variables and sensor signals; second, by using electronic
tongue sensors; finally, by using four selected electronic nose sen-
sors. All the models were also tested with 19 commercial olive oil
samples. ANNs provided very satisfactory results and indicated the
electronic nose as the most appropriate tool for the characterization
of the analyzed oils.
Other authenticity studies successfully conducted with e-noses
have been carried out for the determination of the geographical
origins of Valencia orange juices [66] and Emmental cheese [67],
for the differentiation of unifloral honey samples [10] as well as to
discriminate ‘Aceto Balsamico Tradizionale di Modena’ on age basis
[68].
Steine et al. [66] employed a FOX 3000 e-nose for the dif-
ferentiation of 49 pure Valencia orange juices from 5 different
origins representative of the main culture areas of citrus accord-
ing to their volatile organic fractions. An experimental design was
used to set analytical factors, which control the generation of the
SHS (oven temperature, headspace generation time and injection
volume). Data obtained from e-nose were statistically treated by
11
PCA and factorial discriminant analysis (FDA) and a good discrim-
ination of the juices according to the origin of oranges used was
achieved. A calibration of the sensors was then performed to build
a database aiming to classify the juices according to the origin of
oranges involved. New supplementary samples were then analyzed
to assess the efficiency of the database. This technique coupled with
other ones such as high-pressure liquid chromatography and gas
chromatography may be a new tool for the investigation of adul-
teration detection due to the authentication of the origin of raw
materials employed in the orange juice processes.
The volatile compounds of Emmental cheese of different origins
were investigated by Pillonel et al. [67] to check their suitabil-
ity as markers of geographic origin. A total of 20 Emmental
cheese samples from different European regions and countries
were analyzed using P&T-GC followed by flame ionization and
mass spectrometry. All regions were easily differentiated by using
compounds occurring only in the corresponding region or by com-
bining a few compounds by PCA. Further analyses were carried out
using a MS-based e-nose (Smart Nose® ). PCA achieved 90% and
91% of correct classifications for the Emmental from Switzerland
and other regions, respectively. By discriminating each non-Swiss
region one at a time from Switzerland, correct classification rates
of 90–100% were obtained for Switzerland and 83–100% for the
other regions. The use of a trained classification technique, such
as linear discriminant analysis or differential factor analysis, in
combination with a larger database, should improve the discrimi-
nation. Moreover, the pre-concentration of the headspace before
injecting it into the MS detector should enhance the sensitivity
of the analysis and hence the potential of this analytical tech-
nique.
Smart Nose® e-nose has been also applied [10] to the control
of the authenticity of the botanical origin of honey. PCA and DFA
models were built based on groups of samples identified as typical
unifloral honey by a classical method (a combination of sensory,
pollen and physicochemical analysis). Swiss unifloral honeys of the
following types were analyzed: acacia, chestnut, dandelion, lime, fir
and rape. Three different sampling modes were tested: SHS, SPME
and INDEX. The last two showed interesting abilities to extract
volatile components in a higher concentration and, most impor-
tant, heavier compounds than SHS. The best classification, under
the sampling conditions used, was provided by the SPME sam-
pling mode. This method proved to be fast, reliable and powerful
for this type of task. A good correlation was found between the
present approach and the classical one for the determination of the
botanical origin of honey.
Cocchi et al. [68] have made an attempt to develop an analytical
procedure in the context of quality control, authenticity assessment
and as well as process monitoring of typical food. To this aim a
MS-based e-nose coupled with multivariate data analysis tools was
used to furnish a clear visualization and a suitable interpretation of
the ageing process of ‘Aceto Balsamico Tradizionale di Modena’, as
well as for classification of different aged products. Data obtained
were first analyzed by parallel factor analysis (PARAFAC) to visualise
the ‘natural’ grouping of vinegar samples and to inspect producers
similarity/dissimilarity. On the basis of the PARAFAC results a rea-
sonable class partition with respect to ageing was accomplished
and both LDAs and SIMCA were applied as classification tools. Fur-
thermore, discrimination was improved by using feature selection
in the wavelet domain through WPTER (wavelet packet transform
for efficient pattern recognition) algorithm.
4.5. Other food quality control studies
In this section, we will consider quality aspects other than those
which were covered in the preceding categories since they cannot
be easily included in them (Table 5).
12 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15

Table 5
Other applications of e-noses in food quality control studies.

Sample Type of study Sample handling system Detection system Data processing Ref.
algorithm

Olive oils
Virgin olive oils Discrimination between quality grades SHS FOX 3000: 12 MOS PCA [69]
Virgin olive oils Discrimination between qualities grades SHS 8 CPs KSOM [70]
Virgin olive oils Qualitative and quantitative information SHS MS SIMCA, PLS [7]
about negative and positive sensory
attributes
Wines
Italian dry red wines Prediction of sensorial descriptors DHS PEN 2: 10 MOS GA [71]
Wines Correlation with sensorial descriptors SHS and P&T 16 Thin film tin-oxide PLS [72]
and GC/MS profiles sensors
Fruits and vegetables
Oranges and apples Evaluation of post-harvest quality SHS LibraNose: 7 TSMs PCA, PLS, PLS-DA [73]
Peaches and nectarines from Evaluation of the sensorial features LibraNose: 7 TSMs PCA, LVQ-NN [74]
several cultivars typical of each class
“Xueqing” pears Quality indices prediction (firmness, SHS 8 MOS MLR, ANN [75]
soluble solids content and pH)
“Xueqing” pears Quality indices prediction (firmness, SHS 8 MOS MLR, PCR, PLS [76]
soluble solids content and pH)
Apricots (Prunus armeniaca) Discrimination between varieties SHS-GC FOX 4000: 18 MOS PCA [77]
Apples Discrimination between cultivars SHS 14 Tin-oxide gas PCA, PLS, BP-ANN [78]
sensors
Longjing green teas Discrimination between different quality SHS PEN 2: 10 MOS LDA, PCA [80]
grades
Onions (Allium cepa) Influence of edaphic factors on bulbs Sample pretreatment-SHS A32S AromaScan: 32 PCA [81]
quality CPs
Other food quality studies
Hams Discrimination of different types of hams SHS 16 Tin-oxide thin film PCA, PNN [82]
sensors
Chinese vinegars Identification of several commercial SHS 9 Doped nano-ZnO BP-ANN with k-NN [83]
vinegars thick film sensors
Several kind of foods Discrimination between foods, e-nose SHS Moses II: 7 QMB, 8 PCA [84]
sensors selection MOS, 4 EC
Apples Differentiation between kinds of apple SPME 8 Micro-SAW Radial plots and [85]
oscilators visual inspection
of signals
Pineapples Discrimination between ripe and unripe
pineapples
Sugar Discrimination between sugar with
off-flavors and pure sugar

For details see Abbreviations.

Electronic noses have been extensively used in the quality
evaluation of extra virgin olive oils [7,69,70]. For example, the dis-
crimination between virgin olive oil quality grades for detection
of defects by e-nose technology coupled with SHS analysis has
proved impossible [69]. In contrast, Pioggia et al. [70] developed
a portable, reliable, and flexible instrument for real-time transduc-
tion and processing of chemoresistive conducting polymer-based
sensor array signals. The instrument is also able to automatically
control the SHS sampling system and the experimental parame-
ters. We have also described the assessment of olive oil samples
belonging to three classes: extra virgin, virgin, and defected. To dis-
criminate among data set patterns, a Kohonen self-organizing map
(KSOM) was applied. A minimal misclassification was observed for
extra virgin and virgin olive oils, while a null misclassification for
defected olive as belonging to extra virgin oil was obtained. The
accuracy rates were, respectively, 95.4%, 86.8%, and 96.1%.
An MS-based e-nose was used by López-Feria et al. [7] for
obtaining qualitative and quantitative information about nega-
tive and positive sensory attributes of virgin olive oils (lampante
and extra). In this way, SIMCA allowed the classification of vir-
gin olive oil samples according to their sensory attributes and
PLS modelling, the subsequent determination of the intensity of
these positive/negative attributes. For this purpose, SIMCA and PLS
regression approaches were used for attribute identification and
quantification, respectively. A decision tree by the combination of
both models was generated to provide the joint prediction of the
sensory attributes and their respective intensities by means of a
single output result. The good prediction results obtained when the
decision tree generated were applied to a new set of virgin olive oil
samples reveals its potential applicability in routine analysis owing
to its simplicity, automation, rapidity and reliability.
Ampuero and Bosset [3] reported a review about diverse appli-
cations of e-noses in the investigation of various aspects of dairy
products such as the evaluation of Swiss and Cheddar cheese aroma,
the assessment of the ripening of Pecorino Toscano cheese (ewe’s),
the detection of mould in Parmesan cheese, the classification of
milk by trademark, by fat level and by preservation process, the
classification and the quantification of off-flavours in milk, the eval-
uation of Maillard reactions during heating processes in block-milk,
as well as the identification of single strains of disinfectant-resistant
bacteria in mixed cultures in milk.
A chapter [5] of a monography on odors in the food industry has
been devoted to the application of electronic nose technology to
quality evaluation in the fish industry.
Martí et al. [4] presented an overview of electronic nose appli-
cations in the quality control of alcoholic beverages, with a detailed
description of the e-noses that have been developed so far as well
as a discussion on their advantages and drawbacks. Special atten-
tion is devoted to the comparison between the “classical” devices
based on gas sensors and the modern MS-based instruments.
An electronic nose (PEN 2) and electronic tongue were used,
together with spectrophotometric methods, by Buratti et al. [71]
to predict sensorial descriptors of Italian red dry wines of different
denominations of origin. Genetic algorithms (GAs) were employed
 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
to select variables and build predictive regression models. It was
not possible to build an acceptable regression model for only one
descriptor, sourness. The results obtained demonstrate the possi-
bility of using these techniques to describe and predict a large part
of the selected sensorial information.
Lozano et al. [72] studied the performance of a gas sensor-based
e-nose in the framework of wine quality assessment. The e-nose
used based on an array of 16 thin film tin-oxide sensors was home-
made and home-developed and SHS and P&T sample handling
systems were studied. A total of 28 wines from different regions,
varieties and elaboration processes were tasted by a sensory panel,
analyzed by GC–MS and measured with the e-nose. Sensory scores
for 15 attributes identified by a sensory panel and concentration
of 19 chemical compounds determined by GC–MS were corre-
lated with e-nose responses using PLS regression analysis. Results
obtained showed that e-nose data obtained from wine headspace
allowed to predict wine quality parameters and sensory descriptors
of sensory panel. However, for the prediction of the concentra-
tions of main chemical compounds responsible of wine aroma
determined by GC–MS, only a semiquantitative prediction could be
performed for some of the chemical compounds analyzed. In gen-
eral terms, better results were obtained for the correlation between
sensory panel and e-nose due to the global response of receptors
(human nose and chemical sensors) to the whole aroma, instead of
the chromatograph separation. With respect to the aroma extrac-
tion techniques used in the e-nose better results were obtained with
SHS, which is more similar to the natural process of taste. The P&T
method presented worse prediction ability due to the concentration
and elimination of several chemical compounds in the trap.
The quality of fruit and vegetables has also been studied
by electronic noses in post-harvest quality studies [73–76], for
discrimination between varieties or cultivars [77,78], for their clas-
sification according to their quality and/or origin [79,80] as well as
to study the influence of edaphic factors on quality [81].
Post-harvest quality, e.g. detection of defects of apples and
oranges, was investigated by TSM quartz resonator sensors coated
with different pyrrolic macrocycles [73]. Defects such as meali-
ness (due to over-ripening), skin damage and infections can lead
to changes in aroma. It was found that the electronic nose was sen-
sitive enough to correctly predict the number of defects in these
fruits. In other paper, the same authors [74] used also the same
kind of e-nose together with a sensory panel to define a sensorial
profile of different classes of peaches and nectarines.
Zhang et al. [75,76] investigated the predictability of an e-nose
(comprising 8 metal oxide semiconductors) for “xueqing” pear
quality indices (firmness, soluble solids content and pH). The mul-
tivariate calibration methods, multiple linear regression (MLR),
principal component regression (PCR), ANN and PLS were applied
to predict the quality indices of “xueqing” pear from different pick-
ing dates based on the signal of e-nose. The prediction models for
firmness and soluble solids content indicated a good prediction per-
formance, but for acidity there was a very poor correlation with
the e-nose signal in all cases. All multivariate calibration meth-
ods showed similar performances, although, ANN provided slightly
better predictions (0.93 correlation coefficient value between pre-
dicted and measured values).
In a paper by Solis-Solis et al. [77] discrimination of eight
varieties of apricots (Prunus armeniaca) was intended by several
instrumental techniques. Free aroma compounds from apricots
varieties were discriminated by a FOX 4000 e-nose using PCA. In
a second step, aroma compounds were obtained by liquid–liquid
extraction and SPME, and identified by GC–MS. Aroma compounds
concentrations were statistically analyzed by PCA and FDA. The
extraction techniques were able to differentiate and identify the
varieties without any error. PCA and FDA were able to discriminate
eight varieties of apricots.
13
The potential of an e-nose gas sensor array device and
SPME–GC–MS to the characterization of three apple cultivars
volatiles was studied by Xiaobo and Jiewen [78]. The sensor
array used included 14 tin-oxide gas sensors. The most abundant
VOCs were determined by SPME–GC–MS. PCA, PLS and BP-ANN
were used as pattern recognition techniques. From the PCA score
plots, different apple cultivars could be clearly distinguished by
SPME–GC–MS measurements, while there was slight overlap in
the e-nose measurements. BP-ANN was able to distinguish differ-
ent cultivars based on e-nose responses, with an accuracy of 87%.
PLS results showed poor-moderate correlations between e-nose
responses and SPME–GC–MS profiles.
Bucking et al. [79] compared several methodologies (wet chem-
ical analysis, GC, e-nose and sensory analysis) with regard to their
utility for the classification of peanuts according to their quality
and/or origin. The only approach that enabled such classification
completely was a commercial hybrid gas-sensor array system (e-
nose) based on metal oxide conductors. The wet chemical methods
(such as the acid and the peroxide values, widely used for the
determination of quality of peanuts in routine analysis) were not
able to give an accurate indication of their quality and/or origin.
Regarding to GC methods, the determination of flavour compounds
by means of static headspace GC clearly showed the increase of
hexanal (a product of lipid oxidation) with storage time. Further
experiments and statistical evaluation of data obtained from this
method allowed the prediction of quality and/or origin of peanuts.
Moreover, the coupling of GC with an array of chemical sensors was
used for the detection of and off-flavour, but the method needs fur-
ther refinement. On the other hand, the sensory analysis by means
of a trained sensory panel was only able of distinguish between
peanuts of different origin and/or storage conditions in some cases.
In the paper by Yu et al. [80] a PEN 2 e-nose combined with LDA
analysis was capable of discriminating different quality grades of
green teas (100% correct classification). However, PCA analysis did
not provide satisfactory discrimination results.
The influence of edaphic factors (soil type and sulphur and/or
nitrogen fertilisation) on onion bulb quality was evaluated by
means of an AromaScan e-nose [81]. Onion bulb quality discrim-
inated by the e-nose responded to soil type and N, S treatments.
Therefore additional studies in this field may seek to establish rec-
ommendations ranges for N:S ratios vs. onion quality characteristics
including flavour.
As can be observed in Table 5, in the literature other kinds of food
samples have been studied using electronic noses [82,83]. García et
al. [82] reported a specific application of an e-nose based on a semi-
conductor tin-oxide thin film sensor array combined with PCA and
PNN data analysis for the discrimination of different hams (Fodder,
‘Montanera’, Spoiled and Parma). The success rate of the classifi-
cation of different hams, attending to the signals produced by the
array of sensors, was 100%.
Seventeen commercial Chinese vinegars were analyzed by an
e-nose containing nine doped nano-ZnO thick film gas sensors by
Zhang et al. [83]. The type, raw materials, total acidity, fermentation
method and production area of the vinegars were selected as the
sensory attributes to establish the fingerprints. A 89.7% identifica-
tion accuracy of the vinegars was achieved by means of BP-ANNs
incorporating with k-nearest neighbors (k-NN).
Other studies on food quality control have been mainly cen-
tered in the evaluation of the sensors performance [84,85]. Pardo
et al. [84] evaluated the possibility of using a hybrid e-nose for
detecting contraband food, measured diverse food products with
the Moses II e-nose equipped with three different classes of chem-
ical sensors (seven QMB, eight MOS and four electrochemical cells
(EC)). Data were analyzed both with traditional PCA plots, pointing
out the limits encountered by this technique and via exhaustive
sensor selection. The principal sensor selection results are that:
14 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
(i) the ranking of the sensor type with regard to discrimination
is QMB > EC > MOS; (ii) selected hybrid sensors have much better
performances than selected sensors from any single sensor class;
(iii) sensors selected from the hybrid array also have better perfor-
mances than the complete set of hybrid sensors; and in particular
(iv) a subsets of as few as two sensors (one QMB, one EC cell) give
results similar or better to all 19 sensors.
Barié et al. [85] presented the coupling of a SAW based micro-
nose, with the SPME technique. To prove the practical usage of this
analytical concept in the field of food quality control, different tests
were successful conducted for the differentiation between apple
varieties, ripe and unripe pineapple, and finally the detection of
an off-flavor in sugar. The main advantages of this setup include
its rapidity (short preparation times, short extraction times, short
measuring times), simplicity and the low-costs. Since this device is
easy to handle a broad range of applications is possible, e.g. on-line
measurement of chemical and biochemical processes, supplement
for human sensory experiments and detection of microorganisms
contamination.
It has been reported other miscellaneous e-nose studies related
to food analysis, such as to compare the frying performance of
oils [86], release of volatile compounds into the food matrix from
packaging material, as well as food safety applications [87]. In
the food safety framework, the use of e-nose sensing devices for
rapid and reliable testing pathogenic bacteria contamination in
foods is widespread [87–89]. Detection of food contamination by
using standard microbial plate count methods involves time and
extensive sample preparation. Also, improper sampling of the food
product may give misleading results since the culture-based meth-
ods rely on the site of sampling. The use of electronic based sensing
techniques such as e-nose systems can provide rapid and accurate
means of sensing the incidence of food contaminant bacteria with
little or no sample preparation.
5. Conclusions and future trends
In this work, the most important applications of electronic noses
in food control have been examined. This technique is still being
developed but the advantages are already clear. Strengths of the
electronic nose include high sensitivity and correlation with data
from human sensory panels for several applications. It might have
other advantages regarding mobility, price, and ease of use. There-
fore, it has the potential to enter our daily life far away from
well-equipped chemical laboratories and skilled specialists.
Most of the applications of electronic noses represent limited
feasibility studies with concurrent poor validation especially in
terms of reproducibility and predictive ability. There are very few
long-term studies, which indicate excellent reproducibility with-
out the need for extensive calibration and mathematical analyzes
of the sensor readings. Furthermore, the successful use of electronic
noses is application-specific which may limit their use.
Electronic noses have been proposed as fantastic instruments
which could solve almost any problem concerned with odor. How-
ever, the reality is that there is much research still to be done
especially with interpretation of the results. Although differences
can be measured by these devices, there is not a perfect match
with sensory testing or they may only apply to one set of samples,
which cannot be extrapolated to other samples. In spite of this diver-
gence from the intended simplicity, the products obtained are still
by far less expensive than analytical systems and have the poten-
tial for cost-engineering when adapted to one special task. On the
other hand, more emphasis will need to be placed on the method
of obtaining headspace samples as each method will have a dif-
ferent mix of volatile components. There is also the requirement
for considerable method development for each application. Finally,
sensor arrays and pattern recognition techniques tend to predict
the quality of a sample without providing hard data with respect to
composition and concentration.
A major issue with gas sensors is their sensitivity to humidity.
It is well-documented that water vapor affects measurements by
electronic noses and manufacturers of these instruments have been
forced to issue specific operating procedures. However, electronic
noses using conductive polymers are more sensitive to differences
in moisture in samples than other gas sensors. The use of non-gas
sensor systems such as MS in combination with multivariate anal-
ysis can replace those now widely used. MS-based e-noses have
also showed better performance (in terms of speed, selectivity and
sensitivity) than gas sensor-bases e-noses in alcoholic beverage
research, due to the masking effect of ethanol. To overcome this
problem of gas sensors sample handling systems, such as DHS and
SPME (instead of the classical SHS) have shown to decrease the
ethanol (and water) content and increase the concentration of the
other volatile compounds. Nevertheless, the speed and simplicity,
two of the main advantages of e-noses, are compromised.
There are also a few disadvantages associated with the data pro-
cessing techniques (although continuous developments are being
made in this sense). For example, one of the main problems with
using an ANN is knowing when optimal network parameters have
been found. As data sets become less well-behaved, the training
becomes more difficult and the class prediction becomes unsatis-
factory. PCA is also sensitive to drift in data. In this respect, it should
be mentioned that there is also instrumental drift and, thus, a stable
calibrating standard is essential.
Regardless of these concerns, the future for the electronic nose
appears to be promising as it can fulfill niche analyses. This is
because research and development activities are continuing apace
in several laboratories around the world. Even the early instruments
have performed well for some applications and it is believed that
the newer prototypes will advance the field further. Some of these
exciting developments include the artificial olfactory microsys-
tem (e-mucosa) that replicates the basic structure the mammalian
olfactory system providing spatial and temporal chemosensory
information.
Acknowledgements
L. Escuder-Gilabert is grateful to the Consellería de Edu-
cación de la Generalitat Valenciana for the financial support
(APOSTD/2007/062).
References
[1] J.R. Stetter, W.R. Penrose, Sens. Update 10 (2002) 189.
[2] T.C. Pearce, S.S. Schiffman, H.T. Nagle, J.W. Gardner (Eds.), Handbook of Machine
Olfaction: Electronic Nose Technology, Wiley-VCH Verlag GmbH & Co. KGaA,
Weinheim, 2003.
[3] S. Ampuero, J.O. Bosset, Sens Actuators B: Chem. 94 (2003) 1.
[4] M.P. Martí, R. Boqué, O. Busto, J. Guasch, TrAC-Trend. Anal. Chem. 24 (2005) 57.
[5] G. Ólafsdóttir, K. Kristbergsson, Electronic-nose technology: application for
quality evaluation in the fish industry, in: X. Nicolay (Ed.), Odors in the Food
Industry, Springer US, New York, USA, 2006, pp. 57–74.
[6] A. Hernández-Gómez, J. Wang, G. Hu, A. García-Pereira, J. Food Eng. 85 (2008)
625.
[7] S. López-Feria, S. Cárdenas, J.A. García-Mesa, M. Valcárcel, J. Chromatogr. A 1188
(2008) 308.
[8] F. Korel, M.Ö. Balaban, Electronic nose technology in food analyisis, in: S. Otles
(Ed.), Handbook of Food Analysis Instruments, CRC Press, Taylor & Francis
Group, Boca Raton, FL, USA, 2008.
[9] C. Pérès, F. Begnaud, L. Eveleigh, J.L. Berdagué, TrAC-Trend. Anal. Chem. 22
(2003) 858.
[10] S. Ampuero, S. Bogdanov, J.O. Bosset, Eur. Food Res. Technol. 218 (2004) 198.
[11] J.L. Pérez Pavón, M. del Nogal Sánchez, C. García Pinto, M.E. Fernández Laespada,
B. Moreno-Cordero, A. Guerrero Peña, TrAC-Trend. Anal. Chem. 25 (2006) 257.
[12] R.W. Cattrall, Chemical Sensors, Oxford University Press, Oxford, 1997, pp. 1–2.
[13] E.A.H. Hall, Biosensors, Open University Press, Milton Keynes, Buckingham,
1990, pp. 1–10.
 M. Peris, L. Escuder-Gilabert / Analytica Chimica Acta 638 (2009) 1–15
[14] K.C. Persaud, P.J. Travers, Arrays of broad specificity films for sensing volatile
chemicals, in: E. Kress-Rogers (Ed.), Handbook of Biosensors and Electronic
Noses: Medicine, Food and Environment, CRC Press, Boca Raton, 1997, pp.
563–589.
[15] D.L. García-González, R. Aparicio, Grasas Aceites 53 (2002) 96.
[16] D. James, S.M. Scott, Z. Ali, W.T. O’Hare, Microchim. Acta 149 (2005) 1.
[17] Graseby Ionics, 2002. Available from: http://www.clu-in.org/products/
site/camp/graseby.htm (accessed 13 October 2008).
[18] F. Röck, N. Barsan, U. Weimar, Chem. Rev. 108 (2008) 705.
[19] United States Patent 6100689: Ion Mobility Sensors and Spectrometers
having a Corona Discharge Ionization Source (2008). Available from:
http://www.freepatentsonline.com/6100698.html (accessed 13 October 2008).
[20] M.J. Adams, Chemometrics in Analytical Spectroscopy, Royal Society of Chem-
istry, Cambridge, 1995, pp. 70-79.
[21] S.M. Scott, D. James, Z. Ali, Microchim. Acta 156 (2007) 183.
[22] Connecticut Quality Council, in: Design of experiments (DOE): The Sta-
tistical Tool of Improvement, 2003. Available from: http://ctqualitycouncil.
org/classes/doe.htm.(accessed 29 January 2009).
[23] D.L. Massart, B.G.M. Vandeginste, L.M.C. Buydens, S. de Jong, P.J. Lewi, J.
Smeyers-Verbeke, Handbook of Chemometrics and Qualimetrics, Elsevier, Ams-
terdam, 1997.
[24] C. Pinheiro, C.M. Rodrigues, T. Schäfer, J.G. Crespo, Biotechnol. Bioeng. 77 (2002)
632.
[25] M. García, M. Aleixandre, M.C. Horrillo, Electronic nose for the identification
of spoiled Iberian hams, in: Spanish on Conference Electron Devices, 2005, pp.
537–540.
[26] L. Marilley, S. Ampuero, T. Zesiger, M.G. Casey, Int. Dairy J. 14 (2004) 849.
[27] N. Gutiérrez-Méndez, B. Vallejo-Cordoba, A.F. González-Córdova, G.V. Nevárez-
Moorillón, B. Rivera-Chavira, J. Dairy Sci. 91 (2008) 49.
[28] W. Cynkar, D. Cozzolino, B. Dambergs, L. Janik, M. Gishen, Sens. Actuators B:
Chem. 124 (2007) 167.
[29] A.Z. Berna, S. Trowell, W. Cynkar, D. Cozzolino, J. Agric. Food Chem. 56 (2008)
3238.
[30] P. Pani, A.A. Leva, M. Riva, A. Maestrelli, D. Torreggiani, J. Food Eng. 86 (2008)
105.
[31] M. Lebrun, A. Plotto, K. Goodner, M.N. Ducampa, E. Baldwin, Postharvest Biol.
Technol. 48 (2008) 122.
[32] N. Bhattacharya, B. Tudu, A. Jana, D. Ghosha, R. Bandhopadhyaya, M. Bhuyan,
Sens. Actuators B: Chem. 131 (2008) 110.
[33] J. Brezmes, E. Llobet, X. Vilanova, J. Orts, G. Saiz, X. Correig, Sens. Actuators B:
Chem. 80 (2001) 41.
[34] U. Herrmann, T. Jonischkeit, J. Bargon, U. Hahn, Q.Y. Li, C.A. Schalley, E. Vogel, F.
Vögtle, Anal. Bioanal. Chem. 372 (2002) 611.
[35] A.Z. Berna, J. Lammertyn, S. Saevels, C. Di Natale, B.M. Nicolaï, Sens. Actuators
B: Chem. 97 (2004) 324.
[36] S. Saevels, J. Lammertyn, A.Z. Berna, E.A. Veraverbeke, C. Di Natale, B.M. Nicolai,
Postharvest Biol. Technol. 31 (2004) 9.
[37] A. Hernández-Gómez, G. Hu, J. Wang, A. García-Pereira, Comput. Electron. Agr.
54 (2006) 44.
[38] A. Hernández-Gómez, J. Wang, G. Hu, P. García-Pereira, LWT-Food Sci. Technol.
40 (2007) 681.
[39] S. Benedetti, S. Buratti, A. Spinardi, S. Manninoa, I. Mignani, Postharvest Biol.
Technol. 47 (2008) 181.
[40] S. Benedetti, N. Sinelli, S. Buratti, M. Riva, J. Dairy Sci. 88 (2005) 3044.
[41] M. Riva, S. Mannino, Italian Food Beverage Technol. 42 (2005) 11.
[42] S. Labreche, S. Bazzo, S. Cade, E. Chanie, Sens. Actuators B: Chem. 106 (2005)
199.
[43] M.S. Cosio, D. Ballabio, S. Benedetti, C. Gigliotti, Food Chem. 101 (2007) 485.
[44] S. Mildner-Szkudlarz, H.H. Jeleń, R. Zawirska-Wojtasiak, Eur. J. Technol. 110
(2008) 61.
[45] C. Di Natale, G. Olafsdottir, S. Einarsson, E. Martinelli, R. Paolesse, A. D’Amico,
Sens. Actuators B: Chem. 77 (2001) 572.
[46] W.X. Du, C.M. Lin, T. Huang, J. Kim, M. Marshall, C.I. Wei, JFS: Food Microbiol.
Saf. 67 (2002) 307.
[47] G. Olafsdottir, E. Chanie, F. Westad, R. Jonsdottir, C.R. Thalmann, S. Bazzo, S.
Labreche, P. Marcq, F. Lundby, J.E. Haugen, JFS S: Sens. Nutr. Qualities Food 70
(2005) S563.
[48] J. Chantarachoti, A.C.M. Oliveira, B.H. Himelbloom, C.A. Crapo, D.G. Mclachlan,
JFS S: Sens. Nutr. Qualities Food 71 (2006) S414.
15
[49] F. Korel, D.A. Luzuriaga, M.Ö. Balaban, JFS: Sens. Nutr. Qualities Food 66 (2001)
1018.
[50] O. Tokusoglu, M. Balaban, J. Shellfish Res. 23 (2004) 143.
[51] D.A. Luzuriaga, F. Korel, M.Ö. Balaban, J. Aquat. Food Product Technol. 16 (2007)
57.
[52] R. Dutta, E.L. Hines, J.W. Gardner, Meas. Sci. Technol. 14 (2003) 190.
[53] A.R. Youn, B.S. Noh, Food Sci. Biotechnol. 14 (2005) 437.
[54] K.S. McElyea, F.W. Pohlman, J.F. Meullenet, S. Suwansri, Arkansas Animal Science
Department Report 2003, AAES Research Series, vol. 509, 2003, pp. 32–35.
[55] N. El Barbri, E. Llobet, N. El Bari, X. Correig, B. Bouchikhi, Sensors 8 (2008) 142.
[56] L.M. Reid, P.D. O’Donnell, G. Downey, Trends Food Sci. Technol. 17 (2006) 344.
[57] J.A. Ragazzo, P. Chalier, J. Crouzet, C. Ghommidh, Identification of alcoholic bev-
erages by coupling gas chromatography and electronic nose, in: A.M. Spanier, F.
Shahidi, T.H. Parliament, C. Mussinan, C.T. Ho, E. Tratras Contis (Eds.), Food Fla-
vors and Chemistry, Royal Society of Chemistry, Cambridge, 2001, pp. 404–411.
[58] M. Penza, G. Cassano, Anal. Chim. Acta 509 (2004) 159.
[59] J. Lozano, J.P. Santos, J. Gutiérrez, M.C. Horrillo, Sens. Actuators B: Chem. 126
(2007) 616.
[60] M. Aleixandre, J. Lozano, J. Gutiérrez, I. Sayago, M.J. Fernández, M.C. Horrillo,
Sens. Actuators B: Chem. 131 (2008) 71.
[61] J. Lozano, M.J. Fernández, J.L. Fontecha, M. Aleixandre, J.P. Santos, I. Sagayo, T.
Arroyo, J.M. Cabellos, F.J. Gutiérrez, M.C. Horrillo, Sens. Actuators B: Chem. 120
(2006) 166.
[62] M.P. Martí, O. Busto, J. Guasch, J. Chromatogr. A 1057 (2004) 211.
[63] M.C. Cerrato Oliveros, J.L. Pérez Pavón, C. García Pinto, M.E. Fernández Laespada,
B. Moreno Cordero, M. Forina, Anal. Chim. Acta 459 (2002) 219.
[64] C. Cerrato Oliveros, R. Boggia, M. Casale, C. Armanino, M. Forina, J. Chromatogr.
A 1076 (2005) 7.
[65] M.S. Cosio, D. Ballabio, S. Benedetti, C. Gigliotti, Anal. Chim. Acta 567 (2006)
202.
[66] C. Steine, F. Beaucousin, C. Siv, G. Peiffer, J. Agric. Food Chem. 49 (2001) 3151.
[67] L. Pillonel, S. Ampuero, R. Tabacchi, J.O. Bosset, Eur. Food. Res. Technol. 216
(2003) 179.
[68] M. Cocchi, C. Durante, A. Marchetti, C. Armanino, M. Casale, Anal. Chim. Acta
589 (2007) 96.
[69] F. Lacoste, F. Bosque, R. Raoux, OCL-Oléagineux, Corps Gras, Lipides 8 (2001) 78.
[70] G. Pioggia, M. Ferro, F. Di Francesco, IEEE Sens. J. (2007) 237.
[71] S. Buratti, D. Ballabio, S. Benedetti, M.S. Cosio, Food Chem. 100 (2007) 211.
[72] J. Lozano, J.P. Santos, T. Arroyo, M. Aznar, J.M. Cabellos, M. Gil, M.C. Horrillo,
Sens. Actuators B: Chem. 127 (2007) 267.
[73] C. Di Natale, A. Macagnano, E. Martinelli, R. Paolesse, E. Proiettia, A. D’Amico,
Sens. Actuators B: Chem. 78 (2001) 26.
[74] C. Di Natale, A. Macagnano, E. Martinelli, E. Proietti, R. Paolesse, L. Castellari, S.
Campani, A. D’Amico, Sens. Actuators B: Chem. 77 (2001) 561.
[75] H. Zhang, J. Wang, S. Ye, Anal. Chim. Acta 606 (2008) 112.
[76] H. Zhang, J. Wang, S. Ye, J. Food Eng. 86 (2008) 370.
[77] H.M. Solis-Solis, M. Calderon-Santoyo, P. Gutierrez-Martinez, S. Schorr-Galindo,
J.A. Ragazzo-Sanchez, Sens. Actuators B: Chem. 125 (2007) 415.
[78] Z. Xiaobo, Z. Jiewen, Food Chem. 107 (2008) 120.
[79] M. Bucking, J.E. Haugen, H. Steinhart, Flavour Research at the Dawn of the
Twenty-First Century, 2003, pp. 528–531.
[80] H. Yu, J. Wang, H. Zhang, Y. Yu, C. Yao, Sens. Actuators B: Chem. 128 (2008) 455.
[81] L. Abbey, D.C. Joyce, J. Aked, B. Smith, C. Marshall, Ann. Appl. Biol. 145 (2004)
41.
[82] M. García, M. Aleixandre, J. Gutiérrez, M.C. Horrillo, Sens. Actuators B: Chem.
114 (2006) 418.
[83] Q. Zhang, S. Zhang, C. Xie, C. Fan, Z. Bai, Sens. Actuators B: Chem. 128 (2008)
586.
[84] M. Pardo, L.G. Kwong, G. Sberveglieri, K. Brubaker, J.F. Schneider, W.R. Penrose,
J.R. Stetter, Sens. Actuators B: Chem. 106 (2005) 136.
[85] N. Barié, M. Bücking, M. Rapp, Sens. Actuators B: Chem. 114 (2006) 482.
[86] E. De Marco, M. Savaresea, C. Parisinia, I. Battimo, S. Falco, R. Sacchia, Eur. J. Lipid
Sci. Technol. 109 (2007) 237.
[87] J.E. Haugen, Electronic noses in food analysis, in: Rouseff, Cadwallader (Eds.),
Headspace Analysis of Food and Flavors, Theory and Practice, Kluwer Aca-
demic/Plenum Publishers, New York, 2001, pp. 43–57.
[88] S. Balasubramanian, S. Panigrahi, C.M. Logue, C. Doetkott, M. Marchello, J.S.
Sherwood, Food Control 19 (2008) 236.
[89] U. Siripatrawan, Sens. Actuators B: Chem. 128 (2008) 435.