You are on page 1of 5

Journal of Food Engineering 275 (2020) 109848

Contents lists available at ScienceDirect

Journal of Food Engineering


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

Estimation of eye formation in Swiss-type cheese using


geometrical measurements
�lez, Eliana Budelli *, Nicola
Mariana Gonza �s P�erez, Patricia Lema
Instituto de Ingeniería Química, Facultad de Ingeniería, Universidad de la República, Ave. Julio Herrera y Reissig 565, 11300, Montevideo, Uruguay

A R T I C L E I N F O A B S T R A C T

Keywords: Quality of eye-forming cheeses is strongly related to eyes’ size and their distribution. Therefore, it is interesting
Cheese to determine when eyes start to form, their volume fraction and their size, shape and distribution during
Ripening ripening. Currently, destructive methods are used to assess the aforementioned parameters. In this work, we
Non-destructive monitoring
propose a non-destructive method based on geometrical measurements to estimate some parameters associated
with eye formation. During ripening, eye formation was linked to changes in mass, diameter and height of the
cheese wheel. In addition, every five days, one wheel was cut and photographs were taken to estimate eye
distribution. These images were used as reference to assess eyes’ volume fraction during ripening. Models were
developed to predict eyes’ volume fraction from external geometrical measurements and cheese mass. Results are
promising for industrial application as good correlation between eye volume estimation using image analysis and
geometrical changes was obtained.

1. Introduction cheese as well as to determine their size and distribution (Lukinac


et al., 2018). X-ray based imaging has been used to monitor the in­
For Swiss type cheeses, the presence of eyes (gas holes) is an fluence of lactic acid bacteria in eye formation (Guggisberg et al.,
important quality parameter expected by consumers. Usually this type 2013), growth kinetics and spatial distribution of eyes (Huc et al.,
of cheese is displayed in stores cut in half to show internal character­ 2014a; Kraggerud et al., 2009). More sophisticated imaging techniques
istics such as size, quantity and distribution of eyes. During ripening, like magnetic resonance imaging and computed tomography were also
cheese develops its flavor, firmness and aroma as well as eyes (Fox used to study eye formation and its relationship with the
et al., 2017; Guggisberg et al., 2015). Depending on the variety of manufacturing process (Rosenberg et al., 1992; Musse et al., 2014; Lee
cheese, this stage may take from a few weeks to several months. et al., 2012; Huc et al., 2014c). From the image processing point of
Therefore, a nondestructive technique capable of monitoring eyes’ view, a great effort has been devoted to develop different techniques
growth during cheese ripening has industrial relevance. and algorithms to achieve a better interpretation of the images of
Nowadays, at industrial scale, cheese is qualitatively evaluated by cheese eyes and its structure (Schuetz et al., 2013, 2016; Huc et al.,
the sound made when hit with a small hammer, or by taking a small 2014b; Grenier et al., 2016; Caccamo et al., 2004).
cylindrical sample. Some authors present studies on the relationship Revision of recent literature shows that it is important to monitor
between acoustic response and the changes in texture or structure of eye growth in Swiss type cheeses. However, there is not a practical
cheese during ripening (Conde et al., 2008a; Conde et al., 2008b; technique to do the aforementioned task at an industrial scale. In this
Nassar et al., 2010). Acoustic techniques can be applied at an industrial work, we propose the use of the external geometry and mass of the
scale, but currently they do not allow determining size, quantity or cheese to determine when eyes appear in cheese and estimate their
distribution of eyes. Ultrasound imaging was studied as a non-invasive total volume relative to the cheese volume. Results showed that the
technique to detect eyes in cheese and to estimate their size and shape following of geometry, mass and bulk density of a cheese wheel pro­
(Eskelinen et al., 2007). However, attenuation does not allow inspec­ vides information about the formation of eyes and their volume.
tion of the samples beyond 5–7 cm depth. On the other hand, several
imaging techniques have been studied to monitor growth of eyes in

* Corresponding author.
E-mail address: ebudelli@fing.edu.uy (E. Budelli).

https://doi.org/10.1016/j.jfoodeng.2019.109848
Received 11 July 2019; Received in revised form 15 October 2019; Accepted 27 November 2019
Available online 29 November 2019
0260-8774/© 2019 Elsevier Ltd. All rights reserved.
M. Gonz�
alez et al. Journal of Food Engineering 275 (2020) 109848

2. Materials and methods and 420-pixels wide, conversion to grayscale and contrast enhancing by
histogram stretching. Eyes segmentation was performed by binarizing
2.1. Ripening conditions the images using a suitable threshold for each image. Fig. 2B shows an
example of binarization results of Fig. 2A.
Eight cheese wheels, Emmental type obtained from a local provider
(Granja Naturalia, Colonia, Uruguay), were monitored during ripening
2.4. Model to estimate eyes volume from image analysis
for 30 days. At the beginning of ripening, cheese wheels dimensions
were 15 cm in height and 25 cm in diameter, whereas wheels were 8 kg
Total eye volume (Veyes ) was estimated following two assumptions:
on average. Cheeses were kept in humidity-temperature controlled
1) Cheese wheels are axisymmetric; 2) There are no preferential angles
chambers, at 18� C � 2� C and 75 %RH � 5 %RH. Twice a week,
for eye area relative to any cheese section that contains cheese central
cheeses were turned over following manufacturer’s procedure. Five days
axis (from now on this plane will be referred to as cut section). There­
a week, geometrical measurements and mass of all cheese wheels were
fore, every cut section has the same relative eyes area. To estimate eyes
registered. Every five days a wheel was cut in half to register optical
volume from the available images, cheese wheels were approximated as
images as described in section 2.3. Also, samples of cheese without holes
solids of revolution. Based on the hypothesis that a cheese wheel is a
were taken from wheels’ center to estimate cheese matrix density (see
revolution object and that every cut section has statistically the same
section 2.2). Eye formation was evaluated in six different stages, one
eyes area (Huc et al., 2014a), eyes’ reconstruction comes as half toroid.
sample for the first five stages and three replicates for the final stage.
Equation (2) shows the eyes volume estimation using the toroid’s model,
where Ai is pixel area previously segmented as eye, ri is pixel coordinates
2.2. Mass and geometry determination relative to cheese central axis and CF is a conversion factor from pixels to
m3.
Two physical parameters were registered: mass and geometrical di­
mensions of the wheels. Mass (Mexp ) was measured using a scale with a X
N
Veyes ¼ CF � π � ðAi * ri Þ (2)
20 g precision (Ohaus SD35, Ohaus Corporation, USA). Geometrical
i¼1
dimensions were measured to the millimeter. Wheel diameter, wheel
maximum height and minimum height were chosen as the most signif­ The proposed model (toroid model) was validated estimating cheese
icant geometrical dimensions to measure. Fig. 1 shows a schematic wheel mass using equation (3) and contrasting the results with the
representation of the mentioned dimensions. experimental values. In equation (3), the volume of cheese matrix is
Geometrical parameters shown in Fig. 1 were used to approximate calculated by subtracting the volume of eyes (Veyes ) to the measured
cheese wheel volume (Vcw ) as a cylinder with a spherical cap on top volume of the cheese wheel (Vcw ). Then, using cheese matrix density
according to Equation (1). D is cheese diameter, h minimum height and (ρch ), an estimation of cheese wheel mass was obtained (Mest ). The
H maximum height. Cheese wheel volume from Equation (1) was con­ parameter ρch is used only for validation purposes.
trasted with a second estimation obtained by assuming the cheese wheel �
Mest ¼ ρch � Vcw Veyes (3)
is a solid of revolution, which results of rotating the plane (section)
captured in the images. This second estimation involves cheese wheel
diameter, area of the cheese section registered in photographs and the 3. Results and discussion
conversion ratio between pixels to world dimensions.
3.1. Validation of eyes volume model
� � � � �2 !
D2 π 3
Vcw ¼ π h þ H h � D2 þ H h (1)
4 6 4 Eyes volume was estimated using image analysis on the basis of the
model proposed in section 2.4. Validation of the model proposed in
An estimation of the cheese wheel density ρcw (bulk density) was section 2.4 was performed as explained in such section. Table 1 shows
calculated using the experimental mass (Mexp ) and the volume approx­ the results of experimental and estimated values of cheese wheels’ mass.
imation (Vcw ). This parameter includes both the cheese matrix and the Estimation of cheese mass calculated by using Equation (3) was higher
holes. The density of the cheese matrix itself, ρch was estimated using than experimental values for most cases. Relative error between esti­
Archimedes’ method on a small piece of cheese without holes. mated and experimental mass was always lower than 10% (Table 1).
This proved that toroid model gives a good estimate of eye volume.
2.3. Optical image acquisition and processing
3.2. Fitting of an evolution law to estimate eyes volume
Optical color-images were registered using a digital camera (15
megapixels, 35 mm wide lens). After cutting cheeses in half, both halves
Eyes volume obtained using the toroid’s model (Section 2.4) was
were photographed with white background and under white fluorescent
normalized by cheese wheel volume to obtain eyes volume fraction
light. Digital images were processed using commercial software (MAT­
(Reyes ). Eyes volume fraction was fitted to a microbial growth model
LAB, The Mathworks Inc., USA).
Digital image preprocessing included resizing to 256-pixel height

Fig. 2. (A) Color photograph of a cheese half. (B) Binarized image of (A) where
cheese matrix is on white, and central axis of the cheese is in dotted line. (For
Fig. 1. (A) Photograph of cut cheese wheel with dimensions measured. (B) interpretation of the references to color in this figure legend, the reader is
Ideal representation of (A). referred to the Web version of this article.)

2
M. Gonz�
alez et al. Journal of Food Engineering 275 (2020) 109848

Table 1 water evaporation and CO2 release (Riahi et al., 2007). Water loss is
Results for the validation of toroid model to estimate eyes volume fraction in usually the most relevant phenomenon in mass loss, and it can be
cheese as explained in Section 2.4. modelled as a diffusion process from cheese to the environment (Simal
Toroid model validation with mass estimation et al., 2001). Thus, a first order differential equation was used to model
Cheese Experimental mass (kg) Mass from Eq. (3) (kg) Difference (%)
remaining mass and the result is shown in Equation (5). In Equation (5) t
is time in days, to is 2519 days and τ is 547 days. The values of to and τ
I 9.00 9.70 7.76
indicate remaining mass for this system changes slowly in comparison
II 7.60 8.20 7.95
III 8.72 9.55 9.54 with ripening time. The total absolute error between experimental
IV 8.96 9.48 5.80 remaining mass and the values obtained with Equation (5) was 1.09%.
V 8.56 8.26 3.53
VI 7.90 8.21 3.91 M% ¼ e ðt toÞ=τ
(5)
There are no reported models for maximum height and bulk density
based on the fact that eyes are consequence of gas produced by microbial evolution during cheese ripening. For the mentioned reason, no models
fermentation inside cheese (Fox et al., 2017). A logistic model for mi­ were used to fit the experimental data of those parameters.
crobial growth was chosen. This model fits an environmental restrained
population growth as it is the case for microorganisms in cheese (Kareva
and Karev, 2018). Equation (4) corresponds to a second order system 3.4. Relationship between eyes volume fraction and physical parameters
with time delay, which describes logistic model equations. Error be­
tween estimated eyes volume fraction and values obtained from the After validating the toroid model employed to estimate eyes volume
model is less than 0.005. Parameters that fitted the data were: K1 ¼ from image analysis (cf. Table 1), Reyes was compared to percentual
remaining mass, maximum height, and bulk density. For each of the
5:97 � 10 4 , τ ¼ 3:88 days, K2 ¼ 0:40, τ1 ¼ 12:37 days, τ2 ¼
mentioned parameters, a model was fitted to experimental data. Having
0:0246 days, to ¼ 16:33 days.
a model to predict Reyes based on an external parameter as mass or ge­
0 1
8 t
ometry is advantageous as these external parameters are easy to
>
>
>
Reyes ¼ K1 � @eτ 1A for t < 20 days measure.
<
A model of Reyes as a function of percentual remaining mass was
0 0 1, 1
>
> deducted by replacing ripening time in Equation (4) by performing a
>
: variable substitution of time for remaining mass. Fig. 5 shows experi­
ðt toÞ ðt toÞ
Reyes ¼ K2 � @1 @τ2 � e τ2 τ1 � e τ1 A ðτ2 τ1 ÞA for t > 20 days
mental data in circles and the model that results of the mentioned
(4) mathematical operation in continuous line.
Knowing the relationship between remaining mass and eyes volume
Fig. 3A shows eyes volume fraction plotted against ripening time
fraction can be useful when monitoring cheese ripening in the industrial
(squares) and the curve of a microbial growth kinetics model (contin­
scale, as mass is a simple parameter to measure. The difference between
uous line). Fig. 3B shows images of cheese wheels cut in half during
experimental and estimated data was lower than 0.03.
ripening, which correspond to experimental points in Fig. 3A.
Fig. 6 shows the relationship between eyes volume fraction and
normalized maximum height (relative to day 1) from day 10 and
3.3. Mass and geometry measurements beyond. In square markers, experimental data is presented, whereas in
continuous line is the simplest model that best fitted data. In this case,
Fig. 4 shows results of remaining mass (Fig. 4A), maximum height there was not a previous physical model to predict the evolution of
(Fig. 4B) and bulk density (Fig. 4C) during ripening time, all as relative height over time. For this reason, results are fitted using a simple
percent to their value at day 1. Normalized values of remaining mass exponential model.
(M%), maximum height and bulk density were selected as the most Relative maximum height and eyes volume fraction ðReyes Þ from day
significant parameters to monitor eyes formation. Minimum height and 10 and beyond were fitted to a first order exponential model. The result
wheel diameter fluctuated around a mean value and were not suitable to of the mentioned fitting is shown in Equation (6), where H is percentual
detect eye formation nor monitor cheese ripening. maximum height relative to day 1, K ¼ 0:34, Ho ¼ 96:4% and τ ¼
Remaining mass, maximum height and bulk density changed 22:1%. Error between fitted and experimental data was less than 0.01.
significantly with ripening time. Bulk density and remaining mass �
Reyes ¼ K � 1 eðH HoÞ=τ (6)
decreased approximately 33% and 5% respectively, while maximum
height increased 30%. Remaining mass decreases as consequence of Bulk density normalized by cheese matrix density (ρN ) is related to

Fig. 3. (A) Reyes evolution during maturation time in days. Squares (▪▫) represent eyes volume fraction and continuous line ( ) the microbial kinetics model that fit
the data. (B) Images of cheeses’ sections at different maturation days.

3
M. Gonz�
alez et al. Journal of Food Engineering 275 (2020) 109848

Fig. 4. (A) Percentual remaining mass. (B) Percentual maximum height relative to day 1. (C) Percentual bulk density relative to day 1. All measurements are plotted
over maturation time in days. All error bars correspond to the standard deviation of the measurements over the remaining cheese wheels.

Fig. 6. Eyes volume fraction as a function of percentual maximum height


relative to day 1. Experimental data: –▪▫–. Fitting of eyes volume fraction and
maximum height: –.
Fig. 5. Eyes volume fraction as a function of remaining mass relative to day 1.
Experimental data: ●. Fitting of eyes volume fraction and remaining weight: –.

eyes volume fraction in cheese by the following explanation. Assuming


that gas contained inside cheese eyes has negligible mass relative to
cheese wheel mass, Reyes should be approximately equal to 1 ρN . In
Fig. 7, results of plotting Reyes as function of 1 ρN are shown in square
markers, whereas linear fitting of data is shown in continuous line.
Equation (7) presents the parameters of the linear model obtained by
least squares method, with a r2 ¼ 0:9658 and a root mean square error of
0.023.
Reyes ðρN Þ ¼ 0:964 � ð1 ρN Þ 0:0679 (7)

4. Conclusions

A method to estimate eyes volume in Swiss type cheese using


external parameters was developed. From measurements of percentual
remaining mass, normalized maximum height and normalized bulk
Fig. 7. Eyes volume fraction as function of bulk density relative to cheese
density of a cheese wheel, models were deducted, and they allowed
density. Experimental data: ▪▫. Fitting of eyes volume fraction and bulk density
estimation of eyes volume fraction from the aforementioned measure­
relative to cheese density: –.
ments. In order to determine eyes volume, analysis of photographs of
cheese wheels cut in half was used in a model described by Equation (2).

4
M. Gonz�
alez et al. Journal of Food Engineering 275 (2020) 109848

The model to estimate eyes volume was validated by estimating cheese Guggisberg, D., Schuetz, P., Winkler, H., Amrein, R., Jakob, E., Frohlich-Wyder, M.-T.,
et al., 2015. Mechanism and control of the eye formation in cheese. Int. Dairy J. 47,
wheel mass and contrasting it with the measured values.
118–127.
Developed models for external parameters fit the data accurately, Huc, D., Challois, S., Monziols, M., Michon, C., Mariette, F., 2014. Spatial
and they show it is possible to predict fraction of eyes volume inside a characterisation of eye-growing kinetics in semi-hard cheeses with propionic acid
cheese wheel without any destructive assay. This non-destructive tech­ fermentation. Int. Dairy J. 39, 259–269.
Huc, D., Mariette, F., Challois, S., Barreau, J., Moulin, G., Michon, C., 2014. Multiscale
nique could estimate eyes volume in cheese without requirement of investigation of eyes in semi-hard cheese. Innov. Food Sci. Emerg. Technol. 24,
expensive equipment neither trained personnel once models are fitted. 106–112.
However, parameters of the mentioned models must be estimated be­ Huc, D., Roland, N., Grenier, D., Challois, S., Michon, C., Mariette, F., 2014. Influence of
salt content on eye growth in semi-hard cheeses studied using magnetic resonance
forehand since they depend on the cheese type, temperature, relative imaging and CO2 production measurement. Int. Dairy J. 35, 157–165.
humidity, among other internal and external factors. Future work in­ Kareva, I., Karev, G., 2018. From experiment to theory: what can we learn from growth
volves a series of steps to improve the results of this work. curves? Bull. Math. Biol. 80, 151–174.
Kraggerud, H., Wold, J.P., Hoy, M., Abrahamsen, R.K., 2009. X-ray images for the control
Funding of eye formation in cheese. Int. J. Dairy Technol. 62, 147–153.
This work was supported by “Espacio Interdisciplinario-Universidad Lukinac, J.L., Juki�c, M.J., Mastanjevi�c, K.M., Lu�can, M.L., 2018. Application of computer
de la República.” vision and image analysis method in cheese-quality evaluation: a review. Ukrainian
Food J. 7 (2), 192–214.
Lee, K., Uegaki, K., Nishii, C., Nakamura, T., Kubota, A., Hirai, T., et al., 2012. Computed
References tomographic evaluation of gas hole formation and structural quality in Gouda-type
cheese. Int. J. Dairy Technol. 65, 232–236.
Caccamo, M., Melilli, C., Barbano, D.M., Portelli, G., Marino, G., Licitra, G., 2004. Musse, M., Challois, S., Huc, D., Quellec, S., Mariette, F., 2014. MRI method for
Measurement of gas holes and mechanical openness in cheese by image analysis. investigation of eye growth in semi-hard cheese. J. Food Eng. 121, 152–158.
J. Dairy Sci. 87, 739–748. Nassar, G., Lefbvre, F., Skaf, A., Carlier, J., Non gaillard, B., No^el, Y., 2010. Ultrasonic
Conde, T., C�arcel, J., García-P�erez, J., Benedito, J., 2008. Non-destructive analysis of and acoustic investigation of cheese matrix at the beginning and the end of ripening
manchego cheese texture using impact force deformation and acoustic impulse- period. J. Food Eng. 93, 1–13. https://doi.org/10.1016/j.jfoodeng.2009.06.029.
response techniques. J. Food Eng. 82, 238–245. Riahi, M.H., Trelea, I.C., Leclercq-Perlat, M.-N., Picque, D., Corrieu, G., 2007. Model for
Conde, T., Mulet, A., Clemente, G., Benedito, J., 2008. Detection of internal cracks in changes in weight and dry matter during the ripening of a smear soft cheese under
Manchego cheese using the acoustic impulse-response technique and ultrasounds. controlled temperature and relative humidity. Int. Dairy J. 17, 946–953.
J. Dairy Sci. 91, 918–927. Rosenberg, M., McCarthy, M., Kauten, R., 1992. Evaluation of eye formation and
Eskelinen, J.J., Alavuotunki, A.P., Haeggstrom, E., Alatossava, T., 2007. Preliminary structural quality of Swiss-type cheese by magnetic resonance imaging. J. Dairy Sci.
study of ultrasonic structural quality control of Swiss-type cheese. J. Dairy Sci. 90, 75, 2083–2091.
4071–4077. Schuetz, P., Guggisberg, D., Jerjen, I., Frohlich-Wyder, M.T., Hofmann, J., Wechsler, D.,
Fox, P.F., Guinee, T.P., Cogan, T.M., McSweeney, P.L.H., 2017. Fundamentals of Cheese et al., 2013. Quantitative comparison of the eye formation in cheese using
Science. Springer, New York, USA. radiography and computed tomography data. Int. Dairy J. 31, 150–155.
Grenier, D., Laridon, Y., Le Ray, D., Challois, S., Lucas, T., 2016. Monitoring of single eye Schuetz, P., Guggisberg, D., Fr€ ohlich-Wyder, M.T., Wechsler, D., 2016. Software
growth under known gas pressure: magnetic resonance imaging measurements and comparison for the analysis of cheese eyes in X-ray computed tomography. Int. Dairy
insights into the mechanical behavior of a semi-hard cheese. J. Food Eng. 171, J. 63, 117–12.
119–128. Simal, S., Sanchez, E.S., Bon, J., Femenia, A., Rossello, C., 2001. Water and salt diffusion
Guggisberg, D., Frohlich-Wyder, M.-T., Irmler, S., Greco, M., Wechsler, D., Schuetz, P., during cheese ripening: effect of the external and internal resistances to mass
2013. Eye formation in semi-hard cheese: X-ray computed tomography as a non- transfer. J. Food Eng. 48, 269–275.
invasive tool for assessing the influence of adjunct lactic acid bacteria. Dairy Sci.
Technol. 93, 135–149.

You might also like