Journal of Food Engineering 126 (2014) 98106

Contents lists available at ScienceDirect

Journal of Food Engineering

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

A method for nondestructive prediction of pork meat quality and safety

attributes by hyperspectral imaging technique
Feifei Tao, Yankun Peng
China Agricultural University, National R&D Center for Agro-Processing Equipments, 17 Qinghua East Road, Haidian, Beijing 100083, China

a r t i c l e

i n f o

Article history:
Received 17 April 2013
Received in revised form 7 September 2013
Accepted 11 November 2013
Available online 20 November 2013
Keywords:
Hyperspectral imaging
Pork meat
Quality and safety attributes
Nondestructive detection
Light scattering
Gompertz function

a b s t r a c t
Rapid and nondestructive methods for predicting meat quality and safety attributes are of great concerns
at present. A Hyperspectral imaging technique was investigated for evaluating pork meat tenderness and
Escherichia coli (E. coli) contamination in this study. Totally 31 samples were used for hyperspectral imaging in the spectral range of 4001100 nm. A novel method by Modied Gompertz function was exploited
to extract the scattering characteristics of pork meat from the spatially-resolved hyperspectral images.
Gompertz parameters a, b, e and d which can represent different optical meanings were derived by
curve-tting to the original scattering proles. The tting coefcients were all around 0.99 between
470 and 960 nm, which indicating the effective interpretation by Gompertz function. Multi-linear regression models were established using both individual parameters and integrated parameters, and the
results showed that Gompertz parameter d was superior to other individual parameters for both pork
meat tenderness and E. coli contamination, and the integrated parameter can perform better than individual parameters. The validation results (RCV) by the integrated parameter method were 0.949 and
0.939 for pork meat tenderness and E. coli contamination respectively. The study demonstrated that
hyperspectral imaging technique combined with Gompertz function was potential for rapid determination of pork meat tenderness and E. coli contamination, and so hopefully to provide a promising tool for
monitoring the multiple attributes concerning meat quality and safety.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
With the development of economy and improvement of peoples living standards, meat and meat products have become an
important food item in the human diet. According to the report,
over the past 50 years, the global meat consumption has quadrupled from 70 million tons in 1961 to a current total of 283 million
tons in 2011 (According to Global agricultural). It is well known
that pork meat is one of the most important meat products in peoples food. In 2011, the global production of pork meat was around
101 million tons, which accounted for 34% of the total meat production. However, the great expansion of meat industry and
increasing demand of consumers for high-quality and safe meat
have also produced new challenges for the meat industry.
One challenge facing meat industries is to obtain reliable information on meat quality throughout the production process, which
would ultimately provide a guaranteed quality of meat products
for consumers. The meat producers may incur economic losses if
meat quality is not judged accurately for marketing, as the great
variability in raw meat leads to highly variable products being

Corresponding author. Tel.: +86 10 62737703; fax: +86 10 62737997.

E-mail address: ypeng@cau.edu.cn (Y. Peng).
0260-8774/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfoodeng.2013.11.006

marketed without a controlled level of quality (Damez and Clerjon,

2008; Xing et al., 2007). This problem will be aggravated when the
industry is unable to satisfactorily characterize this level of quality
and cannot therefore market products with a certied quality level,
which is an otherwise essential condition for the survival and
development of any modern industry. Among conventional quality
attributes such as meat color, tenderness, juiciness and water holding capacity, consumer research suggests that tenderness is the
most important element for eating quality and the variation in
meat tenderness will directly affect consumers decision to repurchase (Boleman, 1995; Fonseca et al., 2003). At present, the determination of meat tenderness is mainly via shear force apparatus or
sensory assessment in meat industries. However, these methods
are time-consuming, destructive, and are not suitable to fast-paced
production or processing environment in meat plants (Xing et al.,
2007). It is reported that the lack of fast, reliable and nondestructive methods for determining meat characteristics has been one
of the main obstacles for the development of quality control in
the meat industry (ElMasry et al., 2011b). Therefore, an effective
technique that can rapidly and nondestructively segregate meat
carcasses or cuts based on predicted tenderness is highly desirable
for meat producers.
The widespread and increasing incidences of food-borne diseases have brought meat safety to the forefront of public health

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

concerns, and microbial hazard was reported to be one of the major

challenges to meat safety (Sofos, 2008). Nowadays, one of the tools
to ensure meat safety in meat plants is to implement Hazard Analysis and Critical Control Point (HACCP) (Vanne et al., 1996). The
HACCP plans in meat plants mainly focus on control measures that
can reduce the likelihood of contamination of meat from microbial
hazards. Whereas, conventional methods for bacteria detection are
labor-intensive, requiring cumbersome pretreatments and long
time for bacteria incubation, and thus it is difcult to give timely
and effective control on contaminated meat when applying HACCP
plans (Ellis and Goodacre, 2001; Ellis et al., 2002, 2005). Therefore,
advanced sensing technologies that can nondestructively inspect
bacteria contamination would greatly reduce the risk of unsafe
meat, and meanwhile the application of HACCP plans in meat
plants further advances the need for rapid and nondestructive
methods in microbiology.
Among current emerging technologies, optic-based methods
were reported to have the greatest potential for online application
(Shackelford et al., 1999; Vote et al., 2003). Therefore, use of optical
methods has been extensively studied and implemented as an
alternative to conventional analytical methods which are destructive and time-consuming. The Near infrared spectroscopy (NIR) has
been investigated to predict beef tenderness, while the correlation
coefcient (R) yielded only reached to 0.610.81 (Park et al., 1998;
Rdbotten et al., 2000; Zhao et al., 2006). Moreover, considerable
inconsistency among the samples of different genders and postslaughter time was also reported for the prediction of beef tenderness using NIR technology (Rdbotten et al., 2000). Recently, NIR
spectroscopy has also been applied to evaluate the bacterial contamination in shredded cabbage (Suthiluk et al., 2008), and detection and discrimination of Escherichia coli (E. coli) ATCC 25922 and
E. coli K12 (Siripatrawan et al., 2010). However, conventional NIR
spectroscopic instruments are commonly considered as pointbased scanning instruments, and thus spectra obtained from NIR
instruments do not give the spatial information of a sample
(Gowen et al., 2008, 2009).
Hyperspectral imaging technique is a new rapidly growing
technique which integrates spectroscopic and imaging techniques
in one system for providing both spectral and spatial information
simultaneously. As a result, each hyperspectral image contains a
large amount of information in a three-dimensional (3-D) form
called hypercube which can be analyzed to characterize the object more reliably than the traditional machine vision (Kumar and
Mittal, 2010) or spectroscopy techniques (Klaypradit et al., 2011;
Quevedo and Aguilera, 2010). Hyperspectral imaging was originally developed for remote sensing (Goetz et al., 1985), and has
currently emerged as a powerful tool for nondestructive assessment of agro-products quality and safety (Gowen et al., 2007).
Considerable research studies have been reported on applying
hyperspectral imaging technique to determine the inner quality
of fruits (Lu and Peng, 2006; Mendoza et al., 2011; Peng and Lu,
2008), detect apple surface defects/contamination (Kim et al.,
2002; Liu et al., 2007; Mehl et al., 2004), identify poultry contaminants (Lawrence et al., 2006; Park et al., 2002, 2006; Yoon et al.,
2011), evaluate meat quality (ElMasry et al., 2011a; Liu et al.,
2010; Qiao et al., 2007; Wu et al., 2012) and detect the bacterial
contamination of meat and vegetables (Barbin et al., 2013; Feng
et al., 2013; Peng et al., 2010, 2011; Siripatrawan et al., 2011;
Tao et al., 2010, 2012a), etc.
Meat is known to be a turbid biological material, which indicating that the interaction with light involves both absorption and
high scattering, and thus the resulting light attenuation can no
longer be adequately described by the BeerLambert law (Qin
and Lu, 2008; Tao et al., 2012b). The advantage of hyperspectral
imaging is to provide the spatial information simultaneously besides providing the spectral information for each pixel in the

99

image, and therefore one approach that can analyze the hyperspectral data based on both spectral and spatial information would definitely represent the sample better. Previous studies have reported
the usefulness of spatially-resolved hyperspectral images (hyperspectral scattering method) for evaluating the quality attributes
of fruits, meat and milk (Peng and Lu, 2007, 2008; Peng et al.,
2010, 2011; Qin and Lu, 2007; Tao et al., 2012a; Wu et al., 2012).
The hyperspectral scattering method is based on the hardware system in which the point light was applied as the illuminant source,
and thus the scattering pattern of the object can be acquired by
resolving the hyperspectral image spatially. Whereas, the reports
on comprehensively evaluating meat quality and safety attributes
by this method are still rare.
In the previous study, we have investigated using the spatiallyresolved hyperspectral images which were analyzed by the 3parameter Lorentzian function to predict pork meat tenderness
and E. coli contamination and reported the potential of hyperspectral scattering method for the nondestructive evaluation of meat
quality and safety attributes (Tao et al., 2012a). Lorentzian function
is commonly used to describe the laser proles and light distribution patterns in optics research (Davis, 1996). While, Gompertz
function curves were reported to have a steeper gradient than
Lorentzian function curves near the saturation area which implies
that Gompertz function would be a more effective function to t
the scattering proles with the steep descending gradient nearby
the saturation area (Peng and Lu, 2007). Therefore, compared to
our previous study, the objectives of this article are to present a
more effective data analysis method for spatially-resolving hyperspectral images by Gompertz function, validate its usefulness on
nondestructive determination of the multiple attributes of pork
meat (taking tenderness and E. coli contamination as an example
in the article) and also compare these results to the ones that analyzed by 3-parameter Lorentzian function. One promising tool for
comprehensively monitoring the multiple attributes of meat quality and safety was provided in the article.

2. Materials and methods

2.1. Hyperspectral imaging system
A hyperspectral imaging system in the spectral range of 400
1100 nm was used to acquire the images of pork meat samples.
Fig. 1 shows the sketch of hyperspectral imaging system used
in this study. The hyperspectral imaging system mainly consisted
of a high-performance back-illuminated 12-bit charge coupled
device (CCD) camera (Sensicam QE, PCO AG, Kelheim, Germany),
an imaging spectrograph (ImSpector V10E, Spectral Imaging
Ltd., Oulu, Finland), an illumination unit (Oriel Instruments, Stratford, USA) equipped with optical ber, a computer supported
with a data acquisition and control software (Camera control
Kit V2.19, the Cooke Corp., Germany). The optical ber was used
to form point light in the imaging system, and the diameter of the
light beam formed was 5 mm, with working power in 150 W and
incidence angle around 15. The system worked in a line scanning
mode, and all scans were obtained at a position of 3 mm (from
the incident light center to the scanned position) off the incident
light center in order to avoid the signal saturation on CCD
detector.
The resolution of the imaging system was spectrally 2.8 nm
with a 0.74 nm interval, and spatially less than 9 lm. The image
generated by this system was of 1376  1040 (spatial  spectral)
pixels, with the binning in the horizontal and vertical directions
of 1, 2, 4, 8 and 1, 2, 4, 8, 16 respectively. Additionally, in order
to minimize the effect of ambient light, the imaging system was
enclosed in a shield box.

100

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

Fig. 1. The sketch of hyperspectral imaging system.

2.2. Experimental procedures

2.2.1. Sample preparation
Fresh pork meat (Longissimus dorsi muscle) was purchased from
a local market earlier on the day of the experiment and transported
to the lab under refrigeration in 30 min. In order to avoid the effect
from initial contamination, the surface pieces in 1-cm thickness
were removed from all pork meat chops under sterile procedures.
Then, all samples were trimmed to the size of 9  5  2.5 cm
(length  width  thickness) uniformly.
The generic E. coli was provided by College of Food Science and
Nutritional Engineering, China Agricultural University. The bacteria used were rst activated by duplicate transfers and incubation
in Nutrient Broth (Aoboxing Bio-tech Co. Ltd., Beijing, China) for
24 h at 37 C respectively, and then delivered into 0.85% sterilized
saline solutions to achieve series of bacterial suspensions. Pork
meat samples with different E. coli loads were articially contaminated by submerging into different concentrations of E. coli suspensions accordingly. Additionally, in order to ensure the
adhesion of E. coli cells on pork meat samples, the samples were
placed in the super clean bench for a certain time uniformly.

the resulting images were of 688  520 (spatial  spectral) pixels.

All images obtained were saved in TIFF format for further analysis.
2.2.4. Tenderness measurement
After acquisition of hyperspectral images, the measurement of
pork meat tenderness was performed. The testing method was referred to NY/T 1180-2006. Briey, the digital thermometer was
xed at the center of the sample and then the sample was placed
in 80 C water bath for heating. As soon as the center temperature
reached to 70 C, the sample was taken out of the water bath for
cooling. After that, 3 cores with the diameter of 1.27 cm were obtained from each sample in parallel to the longitudinal direction of
muscle bers for shear force measurement. Digital meter with
WarnerBratzler shear accessory (C-LM3B, Northeast Agricultural
University, China) was used to measure the shear force which
was regarded as the reference value of meat tenderness (Honikel,
1998). The speed for apparatus probe was set at 5 mm/s. The repetitions of tenderness were averaged for each sample and the values were expressed in the unit of Newton (N).
2.3. Data analysis method

2.2.2. Microbiological test

The cell counts of different E. coli suspensions were determined
immediately after the sample preparation. Three original or diluted
concentrations for each suspension were selected to be incubated
on Nutrient Agar (Aoboxing Bio-tech Co. Ltd., Beijing, China) and
two repetitions were performed on each concentration. All plates
were incubated at 37 C for 48 h and the results were expressed
in Log CFU/mL. The data of two repetitions was averaged and then
used for further analysis.
2.2.3. Acquisition of hyperspectral images
In order to eliminate the effect from dark current of the imaging
system, dark image was acquired before imaging for each sample.
The dark images were obtained by covering the camera lens using
its opaque cap. The sample surface with no visible fat or connective
tissue was selected for imaging, and the scanning lines were all
kept parallel to the longitudinal orientation of pork meat. Four
positions were selected from each sample for imaging, and four
images were automatically averaged to one by setting the working
parameter from the camera control software. That means, actually
16 hyperspectral images were acquired from each sample. Additionally, 2  2 binning was performed on the original images and

2.3.1. Gompertz function tting to scattering proles

The Gompertz function is a sigmoid function, and can be considered as an asymmetric logistic model. It is reported that the Gompertz function has been widely used for describing the growth of
bacteria, cells and organisms (Chowdhury et al., 2007; Khamis
et al., 2005; Vandepitte et al., 1995). The original Gompertz function is commonly written in Eq. (1), with N0, A, a as the function
parameters and t as the time variable in a period studied (Mueller
et al., 1995).

f t N0 exp



A1  expat

Peng and Lu (2007) proposed using different forms of modied

Gompertz functions to describe the scattering proles of apples,
and proved that useful information can be extracted by the modied Gompertz functions to evaluate the inner quality of fruits. The
Gompertz functions after modication were reported to be
appropriate for describing the descending scattering proles (Peng
and Lu, 2007). Moreover, other than the original Gompertz
function, the parameters from the modied Gompertz functions
can represent specic optical scattering meaning respectively.

101

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

Therefore, the modied Gompertz function with four parameters,

which was shown in Eq. (2) was exploited to extract the optical
information of pork meat in this study.

Rwi awi bwi 1  e expewi dwi z

where R is the light intensity, in CCD Count; z is the distance from

the detected position to the light incident center, in mm; a is the
asymptotic value of light intensity; b is related to the upper value
of estimated light intensity at the light incident center; e is the full
scattering width at the inection point; d is the slope around the
inection point; wi represents the designated wavelength in the
spectral range of 4001100 nm in accordance with i = 1, 2,
3, . . . N; and N is the number of wavelengths.
Based on the least-squares principle, the scattering proles of
pork meat derived from hyperspectral images were tted by Gompertz function in Matlab 7.0 (Mathworks, Natick, USA) to give the
function parameters a, b, e and d. The four individual parameters
can represent different optical scattering characteristics of samples
due to their particular meanings in Eq. (2). The tting results were
evaluated by the correlation coefcient (R) between the original
scattering prole and the tting curve. The higher R is, the more
effective the tting is.
2.3.2. Model calibration and validation
First, stepwise discrimination regression method was applied to
determine the optimal wavelengths using Gompertz parameters a,
b, e and d. Stepwise discrimination regression method is a standard
procedure for variable selection, which involves both backward
elimination and forward addition to determine the pertinent variables (Wu et al., 2012). In this method, the number of predictors
retained in the nal model was determined by the levels of significance assumed for inclusion and exclusion of predictors from the
model. Stepwise discrimination regression method was performed
between the reference values of pork meat tenderness, E. coli contamination and the corresponding Gompertz parameters in Matlab
7.0 (Mathworks, Natick, USA) to select the optimal wavelengths.
Then, parameters at optimal wavelengths were used to establish
Multi-linear regression (MLR) models respectively. Eq. (3) was employed to establish prediction models.

Y b0

m
X
bi Pwi

i1

where Y is the prediction value of tested sample, and refers to pork

meat tenderness and E. coli contamination respectively; P is the value of Gompertz parameter at selected wavelengths; b0 is constant;
bi is the regression coefcient, in which m represents the number of
selected wavelengths; and wi is the selected wavelengths. Lastly,
the method of full cross validation was conducted to evaluate the
performance of different models. Full cross validation method
works by omitting one observation (test data) once a time, recalculating the model using the remaining data (training data), and then
predicting the omitted observation. This process is then repeated
with different subsets, until each observation in the dataset is used
once as the test data. The value of variances was calculated for each
performance, and nally the correlation coefcient (RCV) and standard error of cross validation (SECV) were calculated accordingly.
The values of RCV and SECV were used to estimate the model

performance in the study. The higher RCV and the lower SECV is,
the better predictability the model has.
Additionally, the method of integrated parameter which was
described in the previous study (Tao et al., 2012a) was performed
to develop prediction models in this study. The method is based
on the supposition that the extracted parameters a, b, e and d
at the same wavelength represent complementary information
of the tested sample. Briey, the four Gompertz parameters extracted from the same wavelength are integrated as a whole
when determining the optimal variables. That is to say, when
determining the optimal variables, the parameters a, b, e and d
which are derived from the same wavelength are of the same
possibility for being selected to develop the prediction models
or not.
3. Results and discussion
3.1. Results of reference methods
The reference tests of pork meat tenderness and E. coli contamination were performed according to the methods described above,
and the statistical results are shown in Table 1. From this table, it
can be observed that a wide range of pork meat samples were
tested, with the tenderness ranging from 13.26 to 84.99 N and
E. coli load varying from 2.62 to 8.90 Log CFU/mL. Therefore, the
prediction models developed in this study covering the wide range
data can be expected to give relatively better results than narrow
data coverage for online application.
3.2. Spatially-resolved hyperspectral images
The hyperspectral image of pork meat in 3-D format is shown
in Fig. 2. The hyperspectral image contains three dimensions of
information within the tested sample, namely the axis of light
intensity, wavelength and distance. As depicted in Fig. 2, the
two axes that perpendicular to the axis of light intensity represent the spectral and spatial information respectively. The spectral range is between 400 and 1100 nm, and the spatial distance
is from 25 to 25 mm with the incident light center at 0 mm.
The hyperspectral image contains large amounts of information
of the pork meat sample. From Fig. 2, we can see that there will
be a reectance spectrum at each spatial point in the scanned
line, and similarly a light scattering prole at each wavelength
from 400 to 1100 nm.
As hyperspectral data contains 3-D information of the object,
the hypercube should rst be unfolded and restructured into a
2-D matrix in order to apply chemometric techniques. The method of spatially-resolving hyperspectral image was applied in this
study. The spatially-resolved image which refers to the scattering
proles at different wavelengths can be achieved by resolving the
3-D hyperspectral image along the spectral axis. Fig. 3 shows the
scattering proles of pork meat at the wavelengths of 600, 700
and 800 nm. From this gure, it can be observed that the scattering proles of pork meat at different wavelengths were conspicuously similar in shape just with different light intensities, so it
is reasonable to employ an equation to analyze their optical
pattern.

Table 1
Statistical results of reference method for pork meat tenderness and E. coli contamination.
Pork meat properties

Number of samples

Minimum

Maximum

Mean

Standard deviation

Tenderness (N)
E. coli contamination (Log CFU/mL)

31
31

13.26
2.6

84.99
8.9

31.67
6.0

16.82
1.6

102

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

Fig. 2. The hyperspectral image of pork meat in 3-D format.

Fig. 4. Results of Gompertz function tting to the scattering prole of pork meat at
550 nm.

Fig. 3. The scattering proles of pork meat at different wavelengths.

3.3. Gompertz function tting to scattering proles

As depicted in Fig. 3, the scattering proles of pork meat were
symmetrical to the incident light center, therefore only halves of
the scattering proles were used in the following data analysis.
In order to reduce the signal to noise ratio (SNR), two pre-processing methods of Savitsky-Golay smoothing and Median lter were
performed on the original hyperspectral images. The tting results
were compared among the groups of No pre-process, SavitskyGolay and Median lter, while no obvious difference was observed. The tting results of No pre-process and Savitsky-Golay
smoothing at 550 nm were shown in Fig. 4, and it can be seen that
the tting curves were almost coinciding. Thus, the scattering proles were analyzed by Gompertz function without pre-processing
in the following section.
Fig. 5 shows the tting results to the scattering prole of pork
meat by Gompertz function between 400 and 1100 nm. From this
gure, it can be seen that the tting coefcients were pretty high in
the middle, while inferior at the ends of the spectral range. That
was primarily due to the low light signals which resulted in lower
SNR at the ends of the spectral range. In the spectral range of 470
960 nm, the tting coefcients by Gompertz function were all
around 0.99 for every sample, which indicating the effective interpretation by Gompertz function to the scattering proles, so the

Fig. 5. Fitting results by Gompertz function in the spectral range of 4001100 nm.

optical information in this spectral range was processed for further

analysis.
3.4. Prediction models using Gompertz parameters
3.4.1. Gompertz parameters
As it was mentioned above that Gompertz function can t the
scattering proles of pork meat accurately between 470 and
960 nm, therefore the extracted parameters a, b, e and d can represent the information within the hyperspectral image effectively.
Gompertz parameters a, b, e and d from 31 pork meat samples
are shown in Fig. 6. From this gure, it can be seen that the four
parameters are in different scales, with parameter b in the biggest
scale and parameter d in the smallest one.
The simple correlations between Gompertz parameters and the
reference values of pork meat tenderness, E. coli contamination
were calculated, and the results showed that all the four parameters a, b, e and d related to pork meat tenderness positively, and
parameters a, b and d related to E. coli contamination negatively
between 470 and 960 nm. This suggested that the tougher pork
meat was, the more photons would be scattered.
3.4.2. Prediction models for pork meat tenderness
Stepwise discrimination method was rst performed to determine the optimal variables using individual Gompertz parameters

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

103

Fig. 6. Gompertz parameters extracted from pork meat samples: (a) parameter a, (b) parameter b, (c) parameter e, (d) parameter d.

for model construction. The optimal wavelengths determined

using parameters a, e and d were shown in Table 2 respectively,
while no optimal variables were determined for parameter b. Then,
MLR models were established using the selected parameters for
predicting pork meat tenderness. The prediction model using
parameter a, as an example model, is shown in Eq. (3).

T 35:235  0:165  P583nm 0:263  P625nm  0:490

 P632nm 0:455  P645nm  0:253  P923nm  0:283
 P931nm

where T is the predicted tenderness, in N; and P is the value of

parameter a at assigned wavelengths. The calibration result of this
model is shown in Fig. 7(a), with RC of 0.937 and SEC of 6.570. The
method of full cross validation was performed to evaluate the model performance, and the validation result is depicted in Fig. 7(b).
By performing similar procedures, MLR models using parameters e and d were also calibrated and validated. The calibration
and validation results were summarized in Table 3. In comparison,
the model developed using Gompertz parameter d performed better than the other individual parameters. This is in agreement with
our previous study that the combined parameter (ba)/c from

Table 2
Optimal wavelengths determined for pork meat tenderness using individual
parameters.
Gompertz parameter

Optimal wavelengths (nm)

a
e

583, 625, 632, 645, 923, 931

625, 649, 770, 777, 810, 865
656, 698, 701, 721, 742, 822

Lorentzian function can give a better prediction result than other

parameters investigated (Tao et al., 2012a), as both of them represent the ratio of light attenuation with Gompertz parameter d representing the slope around the inection point and (ba)/c
representing the average ratio through the whole transmitting distance. Previous studies were also reported to explain the principle
underlying the optical scattering method for meat tenderness prediction, and it was concluded that the optical scattering of meat
was subject to the effects of muscle structural properties such as
sarcomere length and collagen content which were also two primary mechanisms controlling meat tenderness (Xia et al., 2006,
2007, 2008a, 2008b).
After developing the prediction models using individual Gompertz parameters, the method of integrated parameters was also
conducted in the study. Similarly, the optimal variables were rst
determined in Matlab software, and the parameters at the wavelengths of 560, 777, 577, 733 and 715 nm were selected. The calibration and validation results of this model were shown in Table 3.
From this table, it can be seen that the model performance can be
improved by using the integrated parameter for model development. The RCV of the model developed using the integrated parameter was 0.949, while the best RCV of the model developed by the
individual Gompertz parameters only reached to 0.933. This is
probably due to that the information within the integrated parameter is more comprehensive, compared to the individual parameters, and thus more effective.
Moreover, comparing the models developed by Gompertz
parameters to 3-parameter Lorentzian parameters in our previous
study, it can be observed that the ones developed using Gompertz
parameters can give better prediction results not only for individual
parameters, but also for integrated parameter (Tao et al., 2012a). The

104

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

100
Rc = 0.937
SEc=6.570

Predicted value (N)

Predicted value (N)

100
75
50
25

Rcv = 0.890
SEcv=7.996

75
50
25
0

0
0

25

50

75

100

25

50

75

100

Reference value (N)

Reference value (N)

(b)

(a)

Fig. 7. Modeling results using parameter a for pork meat tenderness: (a) calibration result, (b) validation result.

Table 3
The results of MLR models for pork meat tenderness using Gompertz parameters.
Gompertz parameter

RC

SEC

RCV

SECV

a
e

0.937
0.951
0.957
0.995

6.570
5.789
5.436
2.796

0.890
0.916
0.933
0.949

7.996
6.824
6.117
5.702

d
Integrated

correlation coefcients of the model validation were 0.890, 0.916

and 0.933 using Gompertz parameters a, e and d, which were all
better than the RCVs of 0.831 and 0.860 by Lorentzian parameters
a and b. The RCV of model developed by the Gompertz integrated
parameter was 0.949, while it only reached to 0.742 by the 3parameter Lorentzian integrated parameter.
3.4.3. Prediction models for E. coli contamination
Similar procedures were conducted to analyze the relationship
between the E. coli counts of suspensions and the hyperspectral
images. Table 4 shows the optimal variables determined by the
stepwise discrimination method using Gompertz parameters a, e,

Table 4
Optimal wavelengths determined for E. coli contamination using individual
parameters.
Optimal wavelengths (nm)

a
e

491, 497, 505, 509, 565, 572, 602

485, 515, 518, 521, 560, 628, 721
625, 754, 791, 829, 865, 868

Predicted value (Log CFU/mL)

10
Rc = 0.890
SEc=0.843

8
6
4
2
2

Reference value (Log CFU/mL)

(a)

T 3:492  0:0348  P 491nm  0:039  P497nm 0:061

 P505nm  0:007  P509nm  0:067  P565nm 0:033
 P572nm 0:013  P602nm

10

where E is the E. coli counts of suspensions, in log CFU/mL; and P is

the value of parameter a at assigned wavelengths. Full cross validation was performed to evaluate the model performance. The calibration and validation results of this model are shown in Fig. 8.
Likewise, the prediction models using Gompertz parameters e and
d were calibrated and validated, and the results were summarized
in Table 5. From the results, it can be observed that the model established by Gompertz parameter d performed better than other
parameters, and this phenomenon was similar to the models developed for tenderness.
Moreover, the method of integrated parameter was conducted
to establish the prediction model for detecting E. coli contamination. The selected wavelengths were 565, 666, 478, 554 and
482 nm in the descending sequence of correlation coefcients.

Table 5
The result of MLR models for E. coli contamination using Gompertz parameters.
Gompertz parameter

RC

SEC

RCV

SECV

a
e

0.890
0.918
0.946
0.995

0.843
0.733
0.583
0.274

0.828
0.863
0.916
0.939

1.021
0.876
0.658
0.639

d
Integrated

Predicted value (Log CFU/mL)

Gompertz parameter

d and no wavelengths were determined using parameter b. MLR

model was established using parameter a and shown in Eq. (4).

10
Rcv= 0.828
SEcv=1.021

8
6
4
2
2

10

Reference value (Log CFU/mL)

(b)

Fig. 8. Modeling results using parameter a for E. coli contamination: (a) calibration result, (b) validation result.

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

The prediction model was calibrated and validated, and the results
were shown in Table 5. The point discussed above that the model
performance can be improved by applying the method of integrated parameter was also observed here. Additionally, the model
also performed better than the one that established by the integrated parameter from 3-parameter Lorentzian. The RCV of the
model developed by the integrated Gompertz parameter for
E. coli contamination prediction was 0.939, while it only reached
to 0.841 by the integrated Lorentzian parameter (Tao et al., 2012a).
Overall, comparing the prediction results in Table 3 to Table 5,
we can see that the model for predicting E. coli contamination by
individual/integrated Gompertz parameters was not so good as
that for pork meat tenderness. The RCVs of the models developed
using Gompertz parameters a, e and d were 0.828, 0.863 and
0.916 respectively for the prediction of E. coli contamination, which
were inferior to those for the prediction of pork meat tenderness
with RCVs of 0.890, 0.916 and 0.933. However, the difference between the models developed by the integrated Gompertz parameter for the prediction of pork meat tenderness and E. coli
contamination was not so obvious, with the RCVs of 0.939 and
0.949 respectively (Tao et al., 2012a).
4. Conclusion
The study demonstrated that hyperspectral imaging technique
combined with Gompertz function could be a rapid and nondestructive tool for prediction of pork meat tenderness and E. coli
contamination. Gompertz function can t the scattering proles
of pork meat effectively, with the tting coefcients all around
0.99 between 470 and 960 nm. Among the models developed using
individual parameters, the model developed by parameter d is
superior to other parameters for both pork meat tenderness and
E. coli contamination. The method of integrated parameter was also
implemented to improve the model performance, and better validation results were obtained with RCV of 0.949 and 0.939 for pork
meat tenderness and E. coli contamination respectively.
The proposed method has high potential to provide a real-time
detection system for evaluating pork meat tenderness and E. coli
contamination, and hopefully to become a powerful tool for comprehensively monitoring meat quality and safety attributes in the
future. Further research requires more precise prediction equation,
accurate hardware and software system development for industrial application.
Acknowledgements
The authors gratefully acknowledge the Special Fund for
Agro-scientic Research in the Public Interest Program (Project
No. 201003008) and the National Science and Technology Support
Program (Project No. 2012BAH04B00) for supporting this research.
References
Barbin, F.D., ElMasry, G., Sun, D.W., Allen, P., Morsy, N., 2013. Non-destructive
assessment of microbial contamination in porcine meat using NIR
hyperspectral imaging. Innovative Food Science and Emerging Technologies
17, 180191.
Boleman, 1995. Consumer evaluation of beef of known tenderness levels. In: Proc.
41st International Congress of Meat Science and Technology. San Antonio,
Texas, USA: Elsevier Science, pp. 594595.
Chowdhury, R.B., Chakraborty, R., Chaudhuri, U.R., 2007. Validity of modied
Gompertz and Logistic models in predicting cell growth of Pediococcus
acidilactici H during the production of bacteriocin pediocin AcH. Journal of
Food Engineering 80 (4), 11711175.
Damez, J.L., Clerjon, S., 2008. Meat quality assessment using biophysical methods
related to meat structure. Meat Science 80, 132149.
Davis, C.C., 1996. Lasers and Electro-optics: Fundamentals and Engineering.
Cambridge University Press, New York, NY.

105

Ellis, D.I., Goodacre, R., 2001. Rapid and quantitative detection of the microbial
spoilage of muscle foods: current status and future trends. Trends in Food
Science and Technology 12, 414424.
Ellis, D.I., Broadhurst, D., Kell, D.B., Rowland, J.J., Goodacre, R., 2002. Rapid and
quantitative detection of the microbial spoilage of meat by Fourier transform
infrared spectroscopy and machine learning. Applied and Environment
Microbiology 68 (6), 28222828.
Ellis, D.I., Broadhurst, D., Clarke, S.J., Goodacre, R., 2005. Rapid identication of
closely related muscle foods by vibrational spectroscopy and machine learning.
Analyst 130 (12), 16481654.
ElMasry, G., Iqbal, A., Sun, D.W., Allen, P., Ward, P., 2011a. Quality classication of
cooked, sliced turkey hams using NIR hyperspectral imaging system. Journal of
Food Engineering 103 (3), 333344.
ElMasry, G., Sun, D.W., Allen, P., 2011b. Non-destructive determination of waterholding capacity in fresh beef by using NIR hyperspectral imaging. Food
Research International 44 (9), 26242633.
Feng, Y.Z., ElMasry, G., Sun, D.W., Scannell, A.G.M., Walsh, D., Morcy, N., 2013. Nearinfrared hyperspectral imaging and partial least squares regression for rapid
and reagentless determination of Enterobacteriaceae on chicken llets. Food
Chemistry 138, 18291836.
Fonseca, S., Wilson, I.J., Horgan, G.W., Maltin, C.A., 2003. Slow ber cluster pattern in
pig longissimus thoracis muscle: implications for myogenesis. Journal of Animal
Science 81 (4), 973983.
Goetz, A.F.H., Vane, G., Solomon, J.E., Rock, B.N., 1985. Imaging spectrometry for
earth remote sensing. Science 228, 11471153.
Gowen, A.A., ODonnell, C.P., Cullen, P.J., Downey, G., Frias, J.M., 2007. Hyperspectral
imaging-an emerging process analytical tool for food quality and safety control.
Trends in Food Science and Technology 18, 590598.
Gowen, A.A., ODonnell, C.P., Cullen, P.J., Bell, S.E.J., 2008. Recent applications of
chemical imaging to pharmaceutical process monitoring and quality control.
European Journal of Pharmaceutics and Biopharmaceutics 69 (1), 1022.
Gowen, A.A., Taghizadeh, M., ODonnell, C.P., 2009. Identication of mushrooms
subjected to freeze damage using hyperspectral imaging. Journal of Food
Engineering 93 (1), 712.
Honikel, K.O., 1998. Reference methods for the assessment of physical
characteristics of meat. Meat Science 49, 447457.
Khamis, A., Ismail, Z., Haron, K., Mohammed, A.T., 2005. Nonlinear growth models
for modeling oil palm yield growth. Journal of Mathematics and Statistics 1 (3),
225233.
Kim, M.S., Lefcourt, A.M., Chao, K., Chen, Y.R., Kim, I., Chan, D.E., 2002. Multispectral
detection of fecal contamination on apples based on hyperspectral imagery.
Part I: Application of visible and near-infrared reectance imaging. Transactions
of the ASAE 45 (6), 20272037.
Klaypradit, W., Kerdpiboon, S., Singh, R.K., 2011. Application of articial neural
networks to predict the oxidation of menhaden sh oil obtained from fourier
transform infrared spectroscopy method. Food Bioprocess Technology 4 (3),
475480.
Kumar, S., Mittal, G.S., 2010. Rapid detection of microorganisms using image
processing parameters and neural network. Food and Bioprocess Technology 3
(5), 741751.
Lawrence, K.C., Windham, W.R., Park, B., Heitschmidt, G.W., Smith, D.P., Feldner, P.,
2006. Partial least squares regression of hyperspectral images for contaminant
detection on poultry carcasses. Journal of Near Infrared Spectroscopy 14 (4),
223230.
Liu, Y., Chen, Y.R., Kim, M.S., Chan, D.E., Lefcourt, A.M., 2007. Development of simple
algorithms for the detection of fecal contaminants on apples from visible/near
infrared hyperspectral reectance imaging. Journal of Food Engineering 81 (2),
412418.
Liu, L., Ngadi, M.O., Prasher, S.O., Garipy, C., 2010. Categorization of pork quality
using Gabor lter-based hyperspectral imaging technology. Journal of Food
Engineering 99 (3), 284293.
Lu, R., Peng, Y., 2006. Hyperspectral scattering for assessing peach fruit rmness.
Biosystems Engineering 93 (2), 161171.
Mehl, P.M., Chen, Y.R., Kim, M.S., Chan, D.E., 2004. Development of hyperspectral
imaging technique for the detection of apple surface defects and
contaminations. Journal of Food Engineering 61 (1), 6781.
Mendoza, F., Lu, R., Ariana, D., Cen, H., Bailey, B., 2011. Integrated spectral and image
analysis of hyperspectral scattering data for prediction of apple fruit rmness
and soluble solids content. Postharvest Biology and Technology 62, 149160.
Mueller, L.D., Nusbaum, T.J., Rose, M.R., 1995. The gompertz equation as a predictive
tool in demography. Experimental Gerontology 30 (6), 553569.
Park, B., Chen, Y.R., Hruschka, W.R., Shackelford, S.D., Koohmaraie, M., 1998. Near
infrared reectance analysis for predicting beef longissimus tenderness. Journal
of Animal Science 76 (8), 21152120.
Park, B., Lawrence, K.C., Windham, W.R., Buhr, R.J., 2002. Hyperspectral imaging for
detecting fecal and ingesta contaminants on poultry carcasses. Transactions of
ASABE 45 (6), 20172026.
Park, B., Lawrence, K.C., Windham, W.R., Smith, D., 2006. Performance of
hyperspectral imaging system for poultry surface fecal contaminant
detection. Journal of Food Engineering 75 (3), 340348.
Peng, Y., Lu, R., 2007. Prediction of apple fruit rmness and soluble solids content
using characteristics of multispectral scattering images. Journal of Food
Engineering 82 (2), 142152.
Peng, Y., Lu, R., 2008. Analysis of spatially resolved hyperspectral scattering images
for assessing apple fruit rmness and soluble solids content. Postharvest
Biology and Technology 48 (1), 5262.

106

F. Tao, Y. Peng / Journal of Food Engineering 126 (2014) 98106

Peng, Y., Tao, F., Li, Y., Wang, W., Chen, J., Wu, J., Dhakal, S., 2010. Rapid detection of
total viable count of chilled pork using hyperspectral scattering technique. In:
Proceedings of SPIE, 7676, 76760K1-K8, Orlando, Florida, USA.
Peng, Y., Zhang, J., Wang, W., Li, Y., Wu, J., Huang, H., Gao, X., Jiang, W., 2011.
Potential prediction of the microbial spoilage of beef using spatially resolved
hyperspectral scattering proles. Journal of Food Engineering 102, 163169.
Qiao, J., Ngadi, M.O., Wang, N., Garipy, C., Prasher, S.O., 2007. Pork quality and
marbling level assessment using a hyperspectral imaging system. Journal of
Food Engineering 83 (1), 1016.
Qin, J., Lu, R., 2007. Measurement of the absorption and scattering properties of
turbid liquid foods using hyperspectral imaging. Applied Spectroscopy 61, 388
396.
Qin, J., Lu, R., 2008. Measurement of the optical properties of fruits and vegetables
using spatially resolved hyperspectral diffuse reectance imaging technique.
Postharvest Biology and Technology 49, 355365.
Quevedo, R., Aguilera, J.M., 2010. Computer vision and stereoscopy for estimating
rmness in the salmon (Salmon salar) Fillets. Food and Bioprocess Technology 3
(4), 561567.
Rdbotten, R., Nilsen, B.N., Hildrum, K.I., 2000. Prediction of beef quality attributes
from early post mortem near infrared reectance spectra. Food Chemistry 69,
427436.
Shackelford, S.D., Wheeler, T.L., Koohmaraie, M., 1999. Tenderness classication of
beef: II, design and analysis of a system to measure beef longissimus shear force
under commercial processing conditions. Journal of Animal Science 77, 1474
1481.
Siripatrawan, U., Makino, Y., Kawagoe, Y., Oshita, S., 2010. Near infrared
spectroscopy integrated with chemometrics for rapid detection of E. coli ATCC
25922 and E. coli K12. Sensors and Actuators B: Chemical 148, 366370.
Siripatrawan, U., Makino, Y., Kawagoe, Y., Oshita, S., 2011. Rapid detection of
Escherichia coli contamination in packaged fresh spinach using hyperspectral
imaging. Talanta 85, 276281.
Sofos, N.J., 2008. Challenges to meat safety in the 21st century. Meat Science 78, 3
13.
Suthiluk, P., Saranwong, S., Kawano, S., Numthuam, S., Satake, T., 2008. Possibility of
using near infrared spectroscopy for evaluation of bacterial contamination in
shredded cabbage. International Journal of Food Science and Technology 43,
160165.
Tao, F., Wang, W., Li, Y., Peng, Y., Wu, J., Shan, J., Zhang, L., 2010. A rapid
nondestructive measurement method for assessing the total plate count on
chilled pork surface. Spectroscopy and Spectral Analysis 30 (12), 34053409.

Tao, F., Peng, Y., Li, Y., Chao, K., Dhakal, S., 2012a. Simultaneous determination of
tenderness and Escherichia coli contamination of pork using hyperspectral
scattering technique. Meat Science 90, 851857.
Tao, F., Tang, X., Peng, Y., Dhakal, S., 2012b. Classication of pork quality
characteristics by hyperspectral scattering technique. In: Symposium
Conducted at ASABE Annual International Meeting, Dallas, Texas, USA.
Vandepitte, V., Quataert, P., Rore, H., Verstraete, W., 1995. Evaluation of the
gompertz function to model survial of bacteria introduced into soils. Soil
Biology and Biochemistry 27 (3), 365372.
Vanne, L., Karwoski, M., Karppinen, S., Sjberg, A.-M., 1996. HACCP-based food
quality control and rapid detection methods for microorganisms. Food Control 7
(6), 263276.
Vote, D.J., Belk, K.E., Tatum, J.D., Scanga, J.A., Smith, G.C., 2003. Online prediction of
beef tenderness using a computer vision system equipped with a BeefCam
module. Journal of Animal Science 81, 457465.
Wu, J., Peng, Y., Li, Y., Wang, W., Chen, J., Dhakal, S., 2012. Prediction of beef quality
attributes using VIS/NIR hyperspectral scattering imaging technique. Journal of
Food Engineering 109, 267273.
Xia, J., Weaver, A., Gerrard, D.E., Yao, G., 2006. Monitoring sarcomere structure
changes in whole muscle using diffuse light reectance. Journal of Biomedical
Optics 11, 04050410405044.
Xia, J., Berg, E.P., Lee, J.W., Yao, G., 2007. Characterizing beef muscles with optical
scattering and absorption coefcients in Vis-NIR region. Meat Science 75, 78
83.
Xia, J., Weaver, A., Gerrard, D.E., Yao, G., 2008a. Heating induced optical property
changes in beef muscles. Journal of Food Engineering 84, 7581.
Xia, J., Weaver, A., Gerrard, D.E., Yao, G., 2008b. Distribution of optical scattering
properties in four beef muscles. Sensing and Instrumentation for Food Quality
and Safety 2, 7581.
Xing, J., Ngadi, M., Gunenc, A., Prasher, S., Gariepy, C., 2007. Use of visible
spectroscopy for quality classication of intact pork meat. Journal of Food
Engineering 82, 135141.
Yoon, S.C., Park, B., Lawrence, K.C., Windham, W.R., Heitschmidt, G.W., 2011. Linescan hyperspectral imaging system for real-time inspection of poultry carcasses
with fecal material and ingesta. Computers and Electronics in Agriculture 79,
159168.
Zhao, J.W., Zhao, J.M., Liu, M.H., 2006. The determination of beef tenderness
using near-infrared spectroscopy. Spectroscopy and Spectral Analysis 26,
640642.