Journal of Food Engineering 119 (2013) 744749

Contents lists available at SciVerse ScienceDirect

Journal of Food Engineering

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

Analysis of pork adulteration in minced mutton using electronic nose of

metal oxide sensors
Xiaojing Tian a,b, Jun Wang a,, Shaoqing Cui a
a
b

Department of Biosystems Engineering, Zhejiang University, 886 Yuhangtang Road, Hangzhou 300058, China
College of Life Science and Engineering, Northwest University for Nationalities, Lanzhou 730024, China

a r t i c l e

i n f o

Article history:
Received 16 January 2013
Received in revised form 4 April 2013
Accepted 8 July 2013
Available online 15 July 2013
Keywords:
Electronic nose
Pattern recognition
Meat adulteration
Feature extraction

a b s t r a c t
The aims were to detect the adulteration of mutton by applying traditional methods (pH and color evaluation) and the E-nose, to build a model for prediction of the content of pork in minced mutton. An Enose of metal oxide sensors was used for the collection of volatiles presented in the samples. Feature
extraction methods, Principle component analysis (PCA), loading analysis and Stepwise linear discriminant analysis (step-LDA) were employed to optimize the data matrix. The results were evaluated by discriminant analysis methods, nding that step-LDA was the most effective method. Then Canonical
discriminant analysis (CDA) was used as pattern recognition techniques for the authentication of meat.
Partial least square analysis (PLS), Multiple Linear Regression (MLR) and Back propagation neural network (BPNN) were used to build a predictive model for the pork content in minced mutton. The model
built by BPNN could predict the adulteration more precisely than PLS and MLR do.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
Adulteration of meat, involving the replacement of selected
breeds, particular geographical region or particular traditional
method with other cheaper animal proteins and even none meat
proteins (soy proteins), has attracted increasing attention. The
choice of one meat over another can reect the lifestyle, religion,
diet and health concerns. For example, lard, pork and meats not ritually slaughtered are forbidden for Muslims and Jews (Bonne and
Verbeke, 2008). However, cheaper animal protein, take pork as
an example, has been fraudulently used to substitute more expensive animal proteins, like mutton and beef. It requires reliable
methods for the authentication of meat adulteration.
Techniques that have been used in the detection of meat adulteration include molecular biology-based methods, enzyme linked
immunological methods, chromatographic methods and spectroscopy methods. Molecular biology-based methods, such as polymerase chain reaction (PCR), real-time PCR (Rodriguez et al.,
2005), restriction fragment length polymorphism analysis (RFLP)
(Chen et al., 2010), multiplex PCR (Ghovvati et al., 2009) and speciesspecic PCR (Man et al., 2007) have been used in the identication of species and adulteration of meat. Enzyme linked
immunological methods used in meat and meat products had been
reviewed by Asensio et al. (2008). These methods are the most specic and sensitive for species identication. However, they require
Corresponding author. Tel.: +86 571 88982178; fax: +86 571 88982191.
E-mail address: jwang@zju.edu.cn (J. Wang).
0260-8774/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfoodeng.2013.07.004

expensive laboratory equipments, high degree technical expertise

and also suffer from higher false-positive rates. Chromatographic
methods, such as gas chromatography mass spectrometer (GC/
MS), high performance liquid chromatography (HPLC) have been
reported in the differentiation (Nurjuliana et al., 2011) and adulteration of meat (Chou et al., 2007). The requirement for tedious
extractions and long analysis times signicantly limited the widespread use of the chromatographic methods. For spectroscopy
methods, mid-infrared spectroscopy combined with soft independent modeling of class analogies (SIMCA), visible (VIS) and near
infrared reectance spectroscopy (NIRS), Nuclear Magnetic Resonance (NMR) spectroscopy and Fourier transform infrared (FTIR)
spectroscopy were shown to be useful for meat in the detection
and quantication of adulterants (Meza-Marquez et al., 2010; Rohman et al., 2011), origin traceability (Sacco et al., 2005; Zhang et al.,
2008), authentication and identication of meat muscle species
(Cozzolino and Murray, 2004). However, the complex analysis of
testing data requires specialized software and algorithms and it
is difcult for ordinary inspectors to master it. However, there
were few studies on meat adulteration in the view of aroma of
the sample.
Electronic noses are devices with several advantages over other
techniques for analyzing food aroma, such as the small amount of
sample required, speed, simplicity, high sensitivity and good correlation with data from sensory analyses. E-nose is comprised of a
sensor array with broad and partly overlapping selectivity for the
measurement of volatile compounds within the headspace on a
sample, combined with computerized multivariate statistical data

X. Tian et al. / Journal of Food Engineering 119 (2013) 744749

processing tools to give an odor ngerprint of the sample. Although

this technique does not allow the identication of compounds and
has a high detection limit in comparison with GCMS, it has been
successfully used in processing monitoring, shelf-life investigation,
freshness evaluation and authenticity assessment in a wide range
of food products, including meat products. In the previous works
on adulteration detection, the electronic nose was shown to be
able to discriminate adulteration of oil (Cosio et al., 2006; Haddi
et al., 2011; Hai and Wang, 2006; Man et al., 2005), wines (Penza
and Cassano, 2004) and meat. For meat discrimination, studies
were done on freshness evaluation (Musatov et al., 2010), processing methods evaluation (Limbo et al., 2010), meat products differentiation and authenticity assessment (Nurjuliana et al., 2011). For
the identication and differentiation of pork for halal authentication, pork and pork sausage from beef, mutton and chicken meat
were studied by E-nose (Garca et al., 2006).
However, most of the research on meat adulteration is mainly
focused on the differentiation and classication of species of meat,
with few studies performed on aroma differentiation. In addition,
for most studies on authentication using E-nose, data used for
analysis were the sensor conductance at a particular time, for
example, 15 s, 30 s, 42 s, 45 s, 60 s, etc. (Gmez et al., 2006,
2007; Yu and Wang, 2007). In this study, sensors conductance at
different collections times were analyzed and feature extraction
methods were used to optimize data set.
The potential use of E-nose for detection of pork adulteration in
minced mutton was investigated in this work. The objectives of
this study were: (1) to investigate the use of an E-nose combined
with pattern recognition methods to detect the presence of pork
in minced mutton, (2) to build a model for the prediction of pork
content in minced mutton, (3) to optimize the feature extraction
methods, and (4) to develop a rapid method for detection of pork
adulterated in minced mutton.

2. Materials and methods

745

illuminant and calibrated with a white plate. The means were used
for analysis.
2.3. The electronic nose (E-nose)
A PEN2 E-nose (portable electronic nose II, Airsense Corporation, Germany) was used to obtain the chemical ngerprint of
the samples. The basic system, which has been described in previous researches (Hai and Wang, 2006; Zhang et al., 2007), consisted
of a sampling apparatus, a detector unit containing the sensor array and pattern recognition software (Win Muster v.1.6) for data
recording and processing. The sensor array was composed of 10
different metal oxide sensors positioned into a small chamber.
Each sensor has a certain degree of afnity towards specic chemical or volatile compounds, and the nomenclature and characteristics of the sensors used are as follows: W1C (S1), sensitive to
aromatics; W5S (S2), sensitive to nitrogen oxides; W3C (S3), sensitive to ammonia, aromatic molecules; W6S (S4), sensitive to
hydrogen; W5C (S5), sensitive to methane, propane, and aliphatic
non-polar molecules; W1S (S6), sensitive to methane; W1W (S7),
sensitive to sulfur-containing organics; W2S (S8), sensitive to
broad alcohols; W2W (S9), sensitive to aromatics, sulfur- and chlorine-containing organics; W3S (S10), sensitive to methane and
aliphatic.
The experimental conditions for E-nose are given as follows:
10 g of the minced mixed meat was placed in a beaker of 250 ml
at the temperature of 25 C 3 C, and the beaker was sealed by
plastic for a headspace generation time of 30 min. The headspace
generation was carried out to increase the volatile compounds
from the meat sample. Before one sample was detected by E-nose,
the sensors were cleaned with the ow of fresh dry air, so that the
sample can be tested. Thereafter, the sensors were exposed to sample volatiles and the changes in sensors responses were acquired
by the data acquisition system (Winmuster). During the sampling
process, the sample gas was transferred into the sensor chamber
at a ow rate of 200 ml min1 and the collection time was 80 s at
an interval of 1 s.

2.1. Meat samples

2.4. Optimization of sensor array and signal processing
All the mutton samples detected by E-nose and used for determination of physical properties were obtained from logistics center
for agricultural products of Hangzhou, and the pork samples were
obtained from Wal-Mart Stores in Hangzhou, China, at the day they
were slaughtered. Before experimental process, fat and connective
tissue were removed, and the meat samples were frozen at 18 C.
The adulterated mutton was made by blending the frozen mutton with pork at levels of 0%, 20%, 40%, 60%, 80% and 100% by
weight, respectively. The adulterated meat was minced for 2 min
by mincer. The mixed meat was brought to room temperature before detection.
2.2. Physiochemical analysis
2.2.1. pH measurement
pH was measured by a pH meter (PB-10, Sartorius, Germany)
using the method of GB/T9695.5 (2008). The experiment was completed by three duplicates for each sample. The pH was expressed
as the mean of three replicates.
2.2.2. Color analysis
Three samples from each treatment were randomly selected to
evaluate their color. Color (CIE tristimulus system, L, a and b values) of the minced meat samples was measured using a Minolta
CM-700d/600d spectrophotometer (Konica Minolta Sensing, Inc.,
Osaka, Japan) with 10 standard observer and D65 daylight

Containing 10 sensors with different sensitivity, the E-nose

gives a data set of 800 (10 sensors  80 s of detecting time) for
each, with a total of 120 samples. The multidimensional signals
of the E-nose required some data pretreatment before statistical
analysis was performed. Feature extraction and selection was done
by method of Stepwise discriminant analysis (Step-LDA), Principle
component analysis (PCA) and loading analysis, the effects were
reviewed by comparison with the original data and data set containing one particular time using three discriminant analysis
methods.
The ability of E-nose in identication of adulteration was analyzed by multivariate data analysis. As a supervised method,
Canonical discriminant analysis (CDA) and Bayes discriminant
analysis (BDA) were used for data visualization and identication
of adulteration according to the content of pork. Partial least
square analysis (PLS), using cross-validation, was performed to
study the predictive capacity of E-nose for the content of pork Multiple Linear Regression (MLR) is one kind of statistical technique
that use several explanatory variables to predict the outcome of
a response variable. The goal of MLR is to model the relationship
between the E-nose signals and the content of pork. The model creates a relationship in the form of a straight line (linear) that best
approximates all the individual data points. Back propagation neural network (BPNN), famous for its ner and more complex classications, a commonly employed and most intensely studied neural
network, was employed to study the predictive capacity of E-nose

746

X. Tian et al. / Journal of Food Engineering 119 (2013) 744749

for the content of pork. The predictive capacity of these three

methods was compared to nd the best predictive model.
For data analysis, the SAS version 8 (SAS Institute Inc., Gary,
USA) was used.

3. Results and discussion

3.1. Physical properties of minced samples
The consumption of glycogen before one animal was slaughtered would lead to higher nal pH, which causes darker muscle
color. The pH of meat could reect the freshness of meat and also
affect the color of meat. The color of meat is an important index of
eating quality. It is determined by the color and luster of muscles
and fat, and varies with the specie, age and gender of the animal.
These physical properties of meat might be useful in the adulteration detection of meat.
With the help of pH meter and spectrophotometer, these physical parameters can be digitalized, without the subjective inuence
of sensory evaluation member. As shown in Figs. 1 and 2, the pH of
mutton was higher than that of pork. For higher content of pork,
the color of meat was brighter, the value of lightness increased
with the content of pork adulterated. This is in accordance with
the color of meat by Nan (2003).

3.2. Sensor array response to meat odor

The typical responding of meat samples were shown in Fig. 3.
The typical responding of mutton, pork, adulterated meat containing 20% and 60% of pork were shown. For one sample, each curve
represents a sensor transient. The sensor response was given as
G/G0, where G and G0 express the resistance of a sensor in clean
air and in detecting gas, respectively. After an initial period of
low and stable conductivity, the conductivities for sensors S2, S6
and S8 increased sharply and then stabilized after collection time
of 60 s, while the conductivities of sensors S1, S3 and S5 dropped
sharply and then stabilized after collection time of 40 s; for the rest
of the sensors, S4, S7, S9 and S10, the conductivity increased slowly
in the detection time. The sensor signal stabilized generally and
was considered to be used in analysis of electronic nose.
As shown in Fig. 3, for different content of pork, the response
tendency of the sensors was quite similar, except that the conductivity of sensors S2, S6 and S8 varied with each other. The response
of S2 and S6 met at 75 s for 20% of pork, at 70 s for 60% of pork and
at 45 s for mutton, while they paralleled with each other for pork.

Fig. 1. pH of the minced mutton adulterated with different content of pork.

Fig. 2. Color of the minced mutton adulterated with different content of pork.

3.3. Feature extraction and selection by loading analysis, stepwise

discriminant analysis and principle component analysis
The number of data for each sample available was 800, one
point per second over 80 s with 10 sensors. However, for one single
sample, the high correlation between the data corresponding to
different points of the same sensors required a reduction in the
number of matrix dimension; three points per sensor were selected as representatives at the following times: 40 s, 60 s and
75 s. In this way, more data were included form the ascending zone
of the recording responses. Thus, 30 original variables were obtained for each sample. For feature extraction and selection, Loading analysis, stepwise discriminant analysis and principle
component analysis were used.
The stepwise discriminant analysis was carried out by SAS on
this original data matrix to reduce the number of the data. The
selection of the variable starts with the largest classication
weight (the sher weight). The procedure was repeated with the
unselected variables, and the Signicance Level to Enter of 0.15
and Signicance Level to Stay of 0.15 were chosen. As a result, 10
variables were selected for further analysis. As shown in Table 1,
the sensors selected were, S5, S8 at 40 s, S1, S3, S5, S6 at 60 s,
and S2, S7, S9, S10 at 75 s. Sensors with small inuence on the
identication process were switched off for further analysis.
By means of SAS V8, the PCA procedure was performed on the
data set containing 30 original variables. In order to retain most
information of the original data, 99.0% cumulative calibrated variance was acquired. Thus, the rst 11 principle components, which
contain more than 99.0% cumulative variance, were selected for
further analysis.
Loading analysis was performed, and the loading plot of loading
factors for meat was shown in Fig. 4. The relative importance of the
sensors in the array was implied. The loading factors associated to
the rst two principal components for each sensor were represented. There are sensors that have identical loading parameters
and their effect for analysis might be represented by just one sensor. As shown in Fig. 4, sensors S1, S3 and S5 have similar loading
factor at 40 s, 60 s and 75 s. S6 and S8 have similar loading factor at
40 s, 60 s and 75 s. Both of them could be represented by one sensor. S2 has similar loading factor at 40 s, 60 s and 75 s. S7 at 60 s
and S7 at 75 s have similar loading factor, they could be represented by one response time. In this way, the variables were reduced from 30 to 14. And the variables were S2, S5, S6 and S7 at
75 s, S7 at 40 s, and S4, S9, S10 at 40 s, 60 s and 75 s.
To validate the methods used in feature extraction and selection, discriminant analysis methods (BDA, CDA with linear discriminant function and quadratic discriminant function) were

747

X. Tian et al. / Journal of Food Engineering 119 (2013) 744749

Fig. 3. Ten sensors response curve to the minced mutton adulterated with different content of pork.

Table 1
The STEPDISC procedure stepwise selection.
Step

Number In

Entered

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

1
2
3
4
5
6
7
8
9
10
9
10
9
10
9
10

S5
S2-60
S6-60
S3
S1
S7-75
S3-60
S5-60
S8
S1-60

Removed

S3
S9-75
S1
S2-75
S2-60
S10-75

Partial R-square

F value

Pr > F

Wilks Lambda

Pr < Lambda

Average squared canonical correlation

Pr > ASCC

0.9533
0.8707
0.4988
0.4076
0.6311
0.2225
0.213
0.2757
0.1605
0.1747
0.0603
0.1237
0.0635
0.119
0.0181
0.1133

465.35
152.23
22.29
15.28
37.64
6.24
5.84
8.15
4.05
4.45
1.35
2.97
1.42
2.84
0.39
2.68

<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
0.0021
0.001
0.2505
0.0152
0.2218
0.0192
0.8563
0.0253

0.04670696
0.00603757
0.00302594
0.00179249
0.00066125
0.00051415
0.00040465
0.00029309
0.00024606
0.00020307
0.00021609
0.00018935
0.00020219
0.00017812
0.00018141
0.00016086

<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001

0.19065861
0.36440033
0.39701149
0.4549528
0.54188591
0.54814386
0.56269654
0.57913335
0.58464976
0.5921713
0.58448351
0.60806873
0.60708783
0.60942244
0.60843243
0.61918425

<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001

employed. The results obtained for each of the tasks were summarized in Table 2. It can be seen that the variable selection processes
afforded good results as regards to classication. The misclassied
were lower than 10.83% in most cases. Compared with data set
containing one point per sensor, 40 s, 60 s and 75 s, respectively,
the results obtained with data set containing three points per sensor were greatly improved especially for step-LDA, with one sample misclassied using CDA. This comparison showed that the
results were always improved through combining more
information.
The results obtained by different feature extraction methods
were also compared in Table 2. The best method was found to be
step-LDA for reducing the misclassied samples to the lowest
number of 1 using CDA, which might be explained by the elimination of the redundant signals of the original data set. However, the
results were worsened by loading analysis and PCA, as important
signals in discriminating the groups of meat were omitted. In

Fig. 4. Loading analysis related to PC1 and PC2.

748

X. Tian et al. / Journal of Food Engineering 119 (2013) 744749

Table 2
Result of discriminant analysis.
40 s

60 s

75 s

Original data 40s, 60s and 75s

Step-LDA

First 7 PCs

Loading analysis

Number of factors

10

10

10

30

10

11

14

CDA-linear discriminant function

Number of misclassied
Percentage of misclassied

10
8.33%

13
10.83%

13
10.83%

5
4.17%

4
3.33%

7
5.83%

8
6.67%

CDA-quadratic discriminant function

Number of misclassied
10
Percentage of misclassied
8.33%

10
8.33%

13
10.83%

2
1.67%

1
0.83%

7
5.83%

10
8.33%

Bayes discriminant analysis

Number of misclassied
Percentage of misclassied

6
5.00%

7
5.83%

4
3.33%

5
4.17%

7
5.83%

7
5.83%

5
4.17%

general, the best feature extraction method was step-LDA selection

technique.
Hence, the sensor array data processed by step-LDA, S5 and S8
at 40 s, S1, S3, S5 and S6 at 60 s, S2, S7, S9 and S10 at 75 s were chosen to carry on the pattern recognition analysis.
3.4. Adulteration classication of meat samples
The results of CDA with linear discriminant function using variables optimized by step-LDA were shown in Fig. 5 and Table 2. The
CAN1 and CAN2 explained 92.44% of the total variance with the value of 100%. So, it can be inferred that the rst two CANs can give
most information of optimized data set. The analysis of variance
and the Partial Wilks Lambda test showed that 10 optimized sensors response varied signicantly for different pork content in
minced mutton. After applying the CDA method, the samples could
be grouped into 6 distinct clusters according to their pork content,
except that group of 40% overlapped with group of 20% and 60%,
with 4 misclassied samples. Two discriminant functions showed
good separation in the direction of CAN 1 among 5 groups of
minced meat containing mutton. Whats more, the 100% pork samples scattered in the area of CAN2 lower than 5, far from the other
groups. The other samples containing mutton scatted in the area of
CAN2 higher than 5, and CAN2 increased with decreasing of pork
content. With different content of mutton, the minced mixed meat
all had the smell of mutton with diverse intensity, which could be
used in discrimination of adulteration.
3.5. Rapid characterization of pork content in minced mutton
In order to establish relationship between the E-nose and pork
content in minced mutton and classify meat samples according to

Fig. 5. Result of CDA based on step-LDA.

PCA based on step-LDA

Table 3
Comparison of three predictive models built on the responses of the E-nose optimized
by step-LDA.
Methods

PLS
MLR
BPNN

Calibration

Validation

R2

RMSEC (%)

R2

RMSEP (%)

0.9609
0.9609
0.9886

6.72
6.72
3.78

0.9092
0.91
0.9762

10.94
10.94
5.26

their content of pork, analytical methods of partial least square

(PLS), Multiple Linear Regressions (MLR) and Back propagation
neural network (BPNN) were used and the results were compared
to nd the best prediction model.
Here, the data set optimized by step-LDA was used. All variables
left in the models are signicant at the 0.05 level. The samples used
for training and testing were randomly selected before the model
was built. Data set containing meat samples (96 for calibration
and 24 for validation) were used to predict the content of pork
adulterated into minced mutton. The correlation coefcient and
root mean square error between predicted and experimental values were used to evaluate the performance of the model. The larger
R2 and lower RMSE lead to better calibration model.
For BPNN, the experimental design was completely randomized
with each sample as an experimental unit. The architecture of the
articial neural network chosen was N  (2N + 1)  M three-layer
back-propagation. The input layer was designed as 10 neurons
according to the sensor array optimized by step-LDA, 1 hidden
neuron; the output layer had 6 neurons for different pork content
in minced mutton.
The results obtained by the training model of BPNN showed
that all the training and testing samples were correctly classied
according to their content of pork. And the correlations between
the observed and predicted content of pork in minced mutton were
higher than 0.97 (Table 3) both for the training and testing sets.
The low errors of prediction and the high correlation of the BPNN
model suggested that the E-nose can be successfully applied in
the determination of the adulteration detection of pork in minced
mutton.
PLS is a multivariate projection method for modeling the relationship between dependent variables and independent variables.
Leave-one-out technique was applied. The accuracy was estimated
using the parameters (the correlation coefcient and root mean
square error between predicted and experimental values) obtained
from the tted equation. As shown in Table 3, good correlations of
calibration were found between E-nose data and content of pork
with a determination of coefcient R2 = 0.9609. When the model
was applied to predict the other 24 samples, similar results could
be found, good prediction results for content of pork with the
coefcient R2 of 0.9092 was obtained. The PLS-E-tongue methods
have been proved to be suitable for prediction of chemical param-

X. Tian et al. / Journal of Food Engineering 119 (2013) 744749

eters (Apetrei et al., 2007; Rudnitskaya et al., 2009). So, pork content in minced mutton could be determined simultaneously by
PLS using the E-nose data in this work.
The MLR algorithm establishes a model that describes the relationship between sensor signals and the pork content. The predictive models for pork content in minced mutton were given as
follow:

Pork content in minced mutton 20:07 7:73  S540s  1:08

 S840s 21:92  S160s
 51:23  S360s 12:73
 S560s  1:32  S660s
 2:59  S275s 9:86
 S775s  6:38  S975s
 8:08  S1075s
2

Large R and low RMSE indicate adequate ts. Table 3 showed the
prediction ability of the E-nose. Similar with the PLS, Table 3 illustrated a linear correlation between the responses of the sensors and
the pork content. When the model was used to predict the prediction set, the results were also high. The MLR model appeared to be
of high ability prediction for percentage of pork adulterated into
minced mutton with R2 higher than 0.93.
In conclusion, excellent prediction ability of E-nose was found
using PLS, MLR and BPNN for prediction of pork content in minced
mutton with R2 higher than 0.9092 and RMSE lower than 10.94%.
The best results were found by BPNN model, with high correlation
(higher than 0.97) for the training and testing subsets.
4. Conclusions
In this paper, an E-nose was used to detect the adulteration of
pork in minced mutton. Multivariate analysis methods were employed to explore the performance of the E-nose in classication
of the adulteration. 120 samples were detected and the signals
were analyzed by three feature extraction methods and pattern
recognition techniques. Step-LDA was proved to be the most effective feature extraction method. It is applicable for the E-nose to detect the adulteration in mutton based on the Step-LDA using CDA.
MLR, PLS and BPNN were used to predict the pork content in
minced mutton precisely. Three methods showed high capacity
in prediction for content of pork in minced mutton. Whats more,
BPNN was the most effective method for the prediction of pork
content. The E-nose had been proved to be a useful authentication
method for meat adulteration detection for its efciency and high
accuracy.
Acknowledgements
The authors acknowledge the nancial support of the Chinese
National Foundation of Nature and Science through Project
31071548,
National
Key
Technology
R&D
Program
2012BAD29B02-4, the Research Fund for the Doctoral Program
of Chinese National Higher Education through Project
20100101110133, and the Fundamental Research Funds for the
Central Universities through Project ZYZ2012071.
References
Apetrei, C., Apetrei, I.M., Nevares, I., del Alamo, M., Parra, V., Rodrguez-Mndez,
M.L., De Saja, J.A., 2007. Using an e-tongue based on voltammetric electrodes to
discriminate among red wines aged in oak barrels or aged using alternative
methods: correlation between electrochemical signals and analytical
parameters. Electrochimica Acta 52 (7), 25882594.

749

Asensio, L., Gonzlez, I., Garca, T., Martin, R., 2008. Determination of food
authenticity by enzyme-linked immunosorbent assay (ELISA). Food Control 19
(1), 18.
Bonne, K., Verbeke, W., 2008. Muslim consumer trust in halal meat status and
control in Belgium. Meat Science 79 (1), 113123.
Chen, S.-Y., Liu, Y.-P., Yao, Y.-G., 2010. Species authentication of commercial beef
jerky based on PCRRFLP analysis of the mitochondrial 12S rRNA gene. Journal
of Genetics and Genomics 37 (11), 763769.
Chou, C.C., Lin, S.P., Lee, K.M., Hsu, C.T., Vickroy, T.W., Zen, J.M., 2007. Fast
differentiation of meats from fteen animal species by liquid chromatography
with electrochemical detection using copper nanoparticle plated electrodes.
Journal of Chromatography B 846 (12), 230239.
Cosio, M.S., Ballabio, D., Benedetti, S., Gigliotti, C., 2006. Geographical origin and
authentication of extra virgin olive oils by an electronic nose in combination
with articial neural networks. Analytica Chimica Acta 567 (2), 202210.
Cozzolino, D., Murray, I., 2004. Identication of animal meat muscles by visible and
near infrared reectance spectroscopy. LWT Food Science and Technology 37
(4), 447452.
Gmez, A.H., Wang, J., Hu, G., Pereira, A.G., 2006. Electronic nose technique potential
monitoring mandarin maturity. Sensors and Actuators B: Chemical 113 (1),
347353.
Gmez, A.H., Wang, J., Hu, G., Pereira, A.G., 2007. Discrimination of storage shelf-life
for mandarin by electronic nose technique. LWT Food Science and Technology
40 (4), 681689.
Garca, M., Aleixandre, M., Gutirrez, J., Horrillo, M.C., 2006. Electronic nose for ham
discrimination. Sensors and Actuators B: Chemical 114 (1), 418422.
GB/T9695.5, 2008. National Standard of the Peoples Republic of China, Meat and
Meat ProductsMeasurement of pH.
Ghovvati, S., Nassiri, M.R., Mirhoseini, S., Moussavi, A.H., Javadmanesh, A., 2009.
Fraud identication in industrial meat products by multiplex PCR assay. Food
Control 20 (8), 696699.
Haddi, Z., Amari, A., Ali, A.O., Bari, N.E., Barhoumi, H., Maaref, A., Jaffrezic-Renault,
N., Bouchikhi, B., 2011. Discrimination and identication of geographical origin
virgin olive oil by an e-nose based on MOS sensors and pattern recognition
techniques. Procedia Engineering 25, 11371140.
Hai, Z., Wang, J., 2006. Electronic nose and data analysis for detection of maize oil
adulteration in sesame oil. Sensors and Actuators B: Chemical 119 (2), 449455.
Limbo, S., Torri, L., Sinelli, N., Franzetti, L., Casiraghi, E., 2010. Evaluation and
predictive modeling of shelf life of minced beef stored in high-oxygen modied
atmosphere packaging at different temperatures. Meat Science 84 (1), 129136.
Man, Y.B.C., Aida, A.A., Raha, A.R., Son, R., 2007. Identication of pork derivatives in
food products by speciesspecic polymerase chain reaction (PCR) for halal
verication. Food Control 18 (7), 885889.
Man, Y.B.C., Gan, H.L., NorAini, I., Nazimah, S.A.H., Tan, C.P., 2005. Detection of lard
adulteration in RBD palm olein using an electronic nose. Food Chemistry 90 (4),
829835.
Meza-Marquez, O.G., Gallardo-Velazquez, T., Osorio-Revilla, G., 2010. Application of
mid-infrared spectroscopy with multivariate analysis and soft independent
modeling of class analogies (SIMCA) for the detection of adulterants in minced
beef. Meat Science 86 (2), 511519.
Musatov, V.Y., Sysoev, V.V., Sommer, M., Kiselev, I., 2010. Assessment of meat
freshness with metal oxide sensor microarray electronic nose: a practical
approach. Sensors and Actuators BChemical 144 (1), 99103.
Nan, Q., 2003. Manual for Meat Industry, rst ed. China Light Industry Press,
Beijing, China, pp. 7274.
Nurjuliana, M., Che Man, Y.B., Mat Hashim, D., Mohamed, A.K.S., 2011. Rapid
identication of pork for halal authentication using the electronic nose and gas
chromatography mass spectrometer with headspace analyzer. Meat Science 88
(4), 638644.
Penza, M., Cassano, G., 2004. Recognition of adulteration of Italian wines by thinlm multisensor array and articial neural networks. Analytica Chimica Acta
509 (2), 159177.
Rodriguez, M.A., Garcia, T., Gonzalez, I., Hernandez, P.E., Martin, R., 2005. TaqMan
real-time PCR for the detection and quantitation of pork in meat mixtures. Meat
Science 70 (1), 113120.
Rohman, A., Sismindari, Erwanto, Y., Man, Y.B.C., 2011. Analysis of pork adulteration
in beef meatball using Fourier transform infrared (FTIR) spectroscopy. Meat
Science 88 (1), 9195.
Rudnitskaya, A., Polshin, E., Kirsanov, D., Lammertyn, J., Nicolai, B., Saison, D.,
Delvaux, F.R., Delvaux, F., Legin, A., 2009. Instrumental measurement of beer
taste attributes using an electronic tongue. Analytica Chimica Acta 646 (12),
111118.
Sacco, D., Brescia, M.A., Buccolieri, A., Jambrenghi, A.C., 2005. Geographical origin
and breed discrimination of Apulian lamb meat samples by means of analytical
and spectroscopic determinations. Meat Science 71 (3), 542548.
Yu, H., Wang, J., 2007. Discrimination of LongJing green-tea grade by electronic
nose. Sensors and Actuators B: Chemical 122 (1), 134140.
Zhang, H., Wang, J., Tian, X., Yu, H., Yu, Y., 2007. Optimization of sensor array and
detection of stored duration of wheat by electronic nose. Journal of Food
Engineering 82 (4), 403408.
Zhang, N., Zhang, D., Li, S., Li, Q., 2008. Preliminary study on origin traceability of
mutton by near infrared reectance spectroscopy coupled with SIMCA method.
Transactions of the Chinese Society of Agricultural Engineering 24 (12), 309
312.