https://doi.org/10.52973/rcfcv-e33268
Received: 15/05/2023 Accepted: 06/06/2023 Published: 25/06/2023
1 of 6
Revista Científica, FCV-LUZ / Vol. XXXIII, rcfcv-e33268, 1 – 6
ABSTRACT
Among the various methods available to determine the storage time
of frozen meat, including analyses based on physical and chemical
properties, sensory analysis, particularly color changes, is an
important aspect of meat acceptability for consumers. In this study,
an articial neural network (ANN) was employed to predict the storage
time of the meat based on the CIELAB color space, represented by the
Lab* (L*), (a*), and (b*) values measured by a computer vision system
at two–month intervals over a period of up to one year. The ANN
topology was optimized based on changes in correlation coecients
(R
2
) and mean square errors (MSE), resulting in a network of 60 neurons
in a hidden layer (R
2
= 0.9762 and MSE = 0.0047). The ANN model's
performance was evaluated using criteria such as mean absolute
deviation (MAD), MSE, root mean square error (RMSE), R
2
, and mean
absolute error (MAE), which were found to be 0.0344, 0.0047, 0.0687,
0.9762, and 0.0078, respectively. Overall, these results suggest that
using a computer vision–based system combined with articial
intelligence could be a reliable and nondestructive technique for
evaluating meat quality throughout its storage time.
Key words: Beef meat; ANN modeling; color parameters; storage
time
RESUMEN
Entre los diversos métodos disponibles para determinar el tiempo
de almacenamiento de la carne congelada, incluidos los análisis
basados en propiedades físicas y químicas, el análisis sensorial,
en particular los cambios de color, es un aspecto importante de la
aceptabilidad de la carne por parte de los consumidores. En este
estudio, se empleó una red neuronal articial (ANN) para predecir
el tiempo de almacenamiento de la carne con base en el espacio
de color CIELAB, representado por los valores Lab* (L*), (a*) y (b*)
medidos por un sistema de visión articial a intervalos de dos meses
durante un período de hasta un año.La topología ANN se optimizó
en función de los cambios en los coecientes de correlación (R
2
)
y los errores cuadráticos medios (MSE), lo que resultó en una red
de 60 neuronas en una capa oculta (R
2
= 0,9762 y MSE = 0,0047). El
rendimiento del modelo ANN se evaluó utilizando criterios como
desviación absoluta media (MAD), MSE, error cuadrático medio
(RMSE), R
2
y error absoluto medio (MAE), que resultaron ser 0,0344;
0,0047; 0,0687; 0,9762 y 0,0078, respectivamente. En general, estos
resultados sugieren qu’el uso de un sistema basado en vision por
computadora combinado con inteligencia articial podría ser una
técnica conable y no destructiva para evaluar la calidad de la carne
durante su tiempo de almacenamiento.
Palabras clave: Carne de res; modelado ANN; parámetros de color;
tiempo de almacenamiento
Storage Time prediction of Frozen Meat using Articial Neural Network
modeling with Color values
Predicción del tiempo de almacenamiento de carne congelada usando modelado de redes
neuronales articiales con valores de color
Saliha Lakehal
1
* , Brahim Lakehal
2
1
University of Batna1, Institute of Veterinary Science and Agricultural Sciences, Department of Veterinary Sciences. Batna, Algeria.
2
University of Batna2, Institute of Hygiene and Industrial Security. Batna, Algeria.
*Corresponding author: saliha.lakehak@univ–batna.dz
Artificial Neural Network for Predicting Storage Time of Frozen Meat / Lakehal and Lakehal _______________________________________
2 of 6
INTRODUCTION
Across the world, meat occupies a central place in diet. It is used
in the preparation of a variety of dishes, from the most traditional
to the most modern. It is associated with moments of pleasure and
celebration, with family or friends. In this regard, freezing meat is
a common and widespread practice in most families and is part of
their preservation habits. However, even when frozen, meat is not
an inert material; it is subject to changes in organoleptic properties
(texture, taste, appearance), nutritional properties (oxidation of lipids
and proteins), or structural changes during the freezing process itself
[1].Sensory analysis provides a fast and reliable non–destructive
alternative to analyze meat quality and shelf life. The main sensory
attributes analyzed in such analysis are texture, avor, and color.
When determining shelf life during sensory analysis, color proling
is of particular importance [2].
Color analysis is a vital component in assessing food quality, and it
can be performed through sensory evaluation by trained inspectors
or using instrumental methods such as colorimeters. However, the
subjectivity of human inspectors in evaluating color can lead to
discrepancies between observers. To overcome this limitation, the
International Commission on Illumination (CIE) proposed a standard
color space known as CIELAB in 1976 [3]. This color space denes
colors in terms of three coordinates: L* for brightness, a* for the
red–green component, and b* for the yellow–blue component. L*
ranges from 0 to 100, while a* and b* range from positive to negative.
The adoption of the CIELAB color space allows for a more reliable and
objective evaluation of food matrix colors. It is particularly useful for
detecting color changes during storage and processing, making it a
crucial tool for quality control in the food industry.
To predict meat storage time, mathematical models such as Logistic,
Baranyi, modied Gompertz, square–root, Arrhenius model, interaction
models, and generic models are used to analyze changes in bacterial
growth with temperature uctuations, according to Hansen et al. [4].
These mathematical models are precise and effective in forecasting
meat storage time. However, in recent years, articial neural networks
(ANNs) have become increasingly popular in predicting changes that
occur in meat quality and evaluation. For example, Zhu et al. [5] used
a neural network to predict various qualities of dry–cured ham based
on protein degradation. Similarly, Kaczmarek and Muzolf–Panek [6]
used the ANN modeling technique to simulate variations in TBARS
levels in the intramuscular lipid fraction of raw beef enriched with
plant extracts. Additionally, Xu et al. [7] presented a neural network–
based approach to anticipate changes in the quality of frozen shrimp
(Solenocera melantho). In their recent work, Kaczmarek and Muzolf–
Panek [8] used predictive models to monitor changes in the levels of
the thiol group (SH) in raw and thermally processed ground chicken
meat that had been enriched with selected plant extracts during
storage at different temperatures. Other researchers, including
Taheri–Garavand et al. [9] and Lalabadi et al. [10], have also used
various ANN models to analyze the quality, production optimization,
and sensory freshness of various food products. However, the use
of computer vision systems as a non–invasive method for quality
control of meat during its conservation is relatively new. For this
reason, the aim of this research is to ascertain the potential of color
values as a reliable method in the evaluation of frozen meat quality
and to develop an ANN–based model for predicting meat storage
time based on color values.
MATERIALS AND METHODS
Samples preparation
One hundred twenty samples weighing approximately 600 g were
selected from the beef biceps femoris muscle slaughtered at the
Municipal Slaughterhouse of Batna, in North Eastern Algeria, and the
beef used was less than two years old. These 120 samples were taken
24 h after the slaughter. The fresh meat samples were analyzed on the
same day, while the remaining samples were divided into six portions
corresponding to the six freezing periods (2, 4, 6, 8, 10, and 12 months),
with each period containing 20 samples. Noting that these samples
have been vacuum packed in bags made of polyamide and polyethylene
using vacuum packaging machine (Sealer Machine, China) and frozen at
–23 ± 1.5°C (CRF–NT64GF40, Condor, Algeria). Temperature monitoring
throughout the frozen storage period was conducted three times a day
using a thermometer (TIA 101, China). Prior to photography, each frozen
sample was thawed in a refrigerator (CRF–NT64GF40, Condor, Algeria)
at a cool and constant temperature of 4 ± 0.6°C for 24 h.
The computer vision system for image capture consisted of a box
and two lamps located 50 cm above the samples at an angle of 45°,
whose purpose was to obtain a uniform light intensity on the sample.
A digital camera (Canon DS126621,China) was also placed on top at a
distance of 30 cm from the sample. To reduce the background light,
the inner walls of the box are covered with an opaque black cloth [11].
Using Adobe Photoshop CS3, the color's clarity (L*), redness (a*), and
yellowness (b*) were numerically assessed (FIG.1).
Application of Articial Neural Networks for modeling
In order to create and utilize ANNs, certain characteristics were
chosen. While there are numerous types of ANNs to choose from,
a multilayer perceptron (MLP) was selected. FIG. 2 illustrates the
schematic representation of MLP networks consisting of three layers:
the input layer, hidden layer(s), and output layer [12]. Neurons in the
input layer display three color variables for frozen/thawed meat. The
output layer, which contains time storage, is complemented by one or
more neurons in the hidden layer. The number of nodes in these layers
is determined by trial and error, so there is no xed rule for how many
hidden layers or neurons are necessary (FIG. 2). The MATLAB interface
was used during the design phase and optimization. During this study,
well–known variable statistical indicators, namely R, MSE, MAD MAE,
and RMSE, were utilized to evaluate the network's eciency. To ensure
a good t between model approaches and target data points, the R
value was employed. Meanwhile, MSE is a dependable measure to
assess the accuracy of a developed procedure, especially when it
comes to predicting errors for an external set of samples. Additionally,
the MAD can be applied as a scale measure to account for individual
differences and elucidate any correlations. Finally, both MAE and
RMSE are used as evaluation metrics in prediction tasks to assess
the accuracy of the predictions. Lower values of both MAE and RMSE
indicate better prediction accuracy.
Statistical analysis
Statistical analysis was performed on the observed values using
variance analysis (ANOVA) with SPSS software version 22 (IBM SPSS
Statistics v22). The means were compared using the Tukey method.
The difference was considered signicant if the probability (P<0.05).
Otherwise, the difference was considered insignicant (P≥0.05).
FIGURE 1. Utilization of Adobe Photoshop CS3 software for image analysis
FIGURE 2. Schematic of the training process of the ANN
______________________________________________________________________Revista Cientifica, FCV-LUZ / Vol. XXXIII, rcfcv-e33268, 1 – 6
3 of 6
RESULTS AND DISCUSSION
Color evaluation
The color of meat has a signicant inuence on its appearance and
market acceptability [13]. As shown in FIG. 3A, the L* value decreased
signicantly after 4 months of storage (P<0.05). Similar results have
been reported by Muela et al. [14] and Muela et al. [15], where the L*
values tended to decrease with prolonged storage time. The content
and chemical state of myoglobin are important factors in determining
meat color [15, 16]. Increased sensitivity to myoglobin oxidation
during freezing could darken the meat color, resulting in decreased L*
values [17, 18]. Other factors may also inuence meat color, especially
after long periods of storage, by promoting marked darkening with
decreased luminance (L*), such as dehydration of the meat surface
layer during long–term storage [20] or temperature uctuation during
the storage process [21].
By observing FIG. 3B, it was able to deduce that the (a*) values
showed a signicant decrease after 10 months of frozen storage
(P<0.05), which is consistent with the ndings of Hansen et al. [4]
who noted a decrease in (a*) values after 30 months of frozen storage.
Medić et al. [22], on the other hand, discovered decreasing (a*) values
up to the 4th month of frozen storage, but an increase at the 6th
month. Alonso et al. [23] suggested that denaturation of myoglobin
during freezing could be the cause of the decrease in (a*) values. In
contrast to the current results, Daszkiewicz et al. [24] did not detect
any differences in (a*) values of frozen Kamieniec sheep (Ovis aries)
meat samples for a period of 12 months.
The b* value was lowest in fresh samples and increased signicantly
with time, starting from the 6th month (P<0.05), reaching its highest
value at 12 months of storage (FIG.3C). Studies have reported similar
results, where meat (b*) value increased as the freezing period became
longer, such as Hansen et al. [4]; Vieira et al. [25]; Fernandes et al.[19] and
Coombs et al. [26]. It may be related to the accumulation of metmyoglobin
and the increase in lipid oxidation as freezing time increases[13, 27].
ANN prediction model
Constructing an ANN model requires careful consideration when
determining its topology. In this particular investigation, after
many tests of different models, : it was considered considered the
multilayer perceptron (MLP) with 60 hidden layers. FIG.4 shows the
symbolic notation of the ANN–optimized model. The optimal number
of neurons required to maximize R
2
and minimize MSE was identied
through careful evaluation. R
2
and MSE values varied with the number
of neurons up to 50, but adding 55 and 60 neurons produced the same
R
2
value of 0.97. Similarly, MSE values varied independently of the
number of neurons. A high correlation coecient of 0.97 indicates
a signicant correlation between the variables used in the model
development and optimization process (FIG. 4). The lowest MSE values
were obtained with 60 and 55 neurons, while neuron 1 had the highest
MSE of 0.070. Although the main goal of optimizing the topology is to
maximize R
2
and minimize MSE, the highest R
2
and lowest MSE were
FIGURE 3. Effect Effect of time freezing on meat colorof freezing on meat color
FIGURE 4. The variations in MSE and R
2
values in reacting to variations in the number of neurons in the hidden layer
Artificial Neural Network for Predicting Storage Time of Frozen Meat / Lakehal and Lakehal _______________________________________
4 of 6
achieved with 60 neurons, with the MSE value of 60 neurons being
slightly lower than that of 55 neurons.
Previous studies have also shown similar results with a lower
number of neurons. For example, Zhang et al. [28] reported that MSE
values reached their minimum after adding 11 nodes to the hidden layer
when constructing an ANN model for the hydrolysis of Newfoundland
shrimp waste. The MSE values remained stable when adding additional
neurons. In another study, Liu et al. [7] used an ANN model to predict
changes in the quality of rainbow trout (Oncorhynchus mykiss) llets.
They built an ANN model and observed the lowest MSE values and
highest R
2
values using only 6 neurons.
The ANN model developed in this study includes 60 neurons in the
hidden layer. The model's performance is evaluated using several
measures, such as MAD, MSE, RMSE, R
2
, and MAE. All of which are
measures of error or precision. The values of these measures for the
test data are presented in TABLE I.
______________________________________________________________________Revista Cientifica, FCV-LUZ / Vol. XXXIII, rcfcv-e33268, 1 – 6
5 of 6
The results show that the developed ANN model has been successfully
validated according to performance criteria, indicating that the model's
predictions are in good agreement with the observed data.
CONCLUSION
Using color values (L*, a*, and b*) to predict the storage time of
frozen meat through an articial neural network has been proven
to be an accurate method. The results of the study validate the
developed ANN model, as it has shown good agreement with observed
data based on performance criteria. Unlike physico–chemical and
microbiological analysis, color measurements are non–destructive, less
time–consuming, and do not require preliminary training for evaluation.
Hence, this study utilized color measurements to estimate the storage
time of frozen meat for a period of 12 months. It is important to note
that the combination of ANN models and color parameters as inputs
has great potential to accurately and quickly predict the storage time of
meat, providing a reliable method for predicting the shelf life of meat.
Conicts of interest
The authors declare no competing interests.
BIBLIOGRAPHIC REFERENCES
[1] Zhang Y, Ertbjerg P. On the origin of thaw loss: Relationship
between freezing rate and protein denaturation. Food Chem.
2019;299:125104. doi: https://doi.org/9w9h
[2] Beltrán JA, Bellés M. Effect of freezing on the quality of meat.
Encycl. Food Secur. Sustain. 2018; 493–7. doi: https://doi.org/kgdb.
[3] Kori R, Le Dréau Y, Antinelli JF, Valls R, Dupuy N. CIEL*a*b* color
space predictive models for colorimetry devices – Analysis of
perfume quality. Talanta. 2013; 104:58–66. doi: https://doi.org/
f4zswt
[4] Hansen E, Juncher D, Henckel P, Karlsson A, Bertelsen G, Skibsted
LH. Oxidative stability of chilled pork chops following long term freeze
storage. Meat Sci. 2004; 68:479–84. doi: https://doi.org/fckgtk
[5] Zhu N, Wang K, Zhang SL, Zhao B, Yang JN, Wang SW. Application
of articial neural networks to predict multiple quality of dry–
cured ham based on protein degradation. Food Chem. 2021;
344:128586. doi: https://doi.org/kgdd
[6] Kaczmarek A, Muzolf–Panek M. Article predictive modeling
of changes in TBARS in the intramuscular lipid fraction of
raw ground beef enriched with plant extracts. Antioxidants.
2021;10(5):730. doi: https://doi.org/kgdf
[7] Xu Z, Liu X, Wang H, Hong H, Luo Y. Comparison between the
Arrhenius model and the radial basis function neural network
(RBFNN) model for predicting quality changes of frozen shrimp
(Solenocera melantho). Int. J. Food Prop. 2017; 20:2711–23. doi:
https://doi.org/kgdg
[8] Kaczmarek A, Muzolf–Panek M. Prediction of thiol group changes
in minced raw and cooked chicken meat with plant extracts–
kinetic and neural network approaches. Anim. 2021; 11(6):1647.
doi: https://doi.org/kgdh
[9] Taheri–Garavand A, Fatahi S, Shahbazi F, de la Guardia M. A
nondestructive intelligent approach to real–time evaluation of
chicken meat freshness based on computer vision technique.
J. Food Process Eng. 2019; 42:1–10. doi: https://doi.org/kgdj
[10] Mohammadi Lalabadi H, Sadeghi M, Mireei SA. Fish freshness
categorization from eyes and gills color features using multi–
class articial neural network and support vector machines.
Aquac. Eng. 2020; 90:102076. doi: https://doi.org/kgdk
[11] Tomasevic I, Tomovic V, Ikonic P, Lorenzo–Rodriguez JM, Barba
FJ, Djekic I, Nastasijević I, Stajić S, Živković D. Evaluation
of poultry meat colour using computer vision system and
colourimeter: Is there a difference? Br. Food. J. 2019; 121:1078–
87. doi: https://doi.org/kgdm
[12] Lakehal B, Dibi Z, Lakhdar N, Dendouga A. Electrical equivalent
model of intermediate band solar cell using PSpice. Sadhana.
2015; 40:1473–9. doi: https://doi.org/kgdn
[13] Zhang R, Yoo MJY, Farouk MM, Delgado–Pando G. Quality and
Acceptability of Fresh and Long–Term Frozen In–Bag Dry–Aged
Lean Bull Beef. J. Food Qual. 2019; 2019:e1975264. doi: https://
doi.org/kgdp
[14] Muela E, Monge P, Sañudo C, Campo MM, Beltrán JA. Meat quality
of lamb frozen stored up to 21 months: Instrumental analyses
on thawed meat during display. Meat Sci. 2015; 102:35–40. doi:
https://doi.org/f64fd7
[15] Muela E, Sañudo C, Campo MM, Medel I, Beltrán JA. Effect of
freezing method and frozen storage duration on instrumental
quality of lamb throughout display. Meat Sci. 2010; 84:662–9.
doi: https://doi.org/bpzjvz
[16] Mancini RA, Hunt MC. Current research in meat color. Meat Sci.
2005; 71:100–21. doi: https://doi.org/dks36s
[17] Leygonie C, Britz TJ, Hoffman LC. Impact of freezing and thawing
on the quality of meat: Review. Meat Sci. 2012; 91:93–8. doi:
https://doi.org/fzh375
[18] Farouk MM, Swan JE. Effect of Rigor temperature and frozen
storage on functional properties of hot–boned manufacturing
beef. Meat Sci. 1998; 49:233–47. doi: https://doi.org/bktm5g
[19] Fernandez X, Monin G, Culioli J, Legrand I, Quilichini Y. Effect of
Duration of Feed Withdrawal and Transportation Time on Muscle
Characteristics and Quality in Friesian–Holstein Calves. J. Anim.
Sci. 1996; 74:1576–83. doi: https://doi.org/kgdq
[20] Zhu LG, Brewer MS. Discoloration of fresh pork as related to
muscle and display conditions. J. Food Sci. 1998; 63:763–7. doi:
https://doi.org/bw7g8s
TABLE I
Assessing the Effectiveness of Articial Neural Networks (ANN)
MAE R
2
RMSE MSE MAD
0.0078 0.9762 0.0687 0.0047 0.0344
MAE: Mean absolute Error, R
2
: Correlation Coecient, RMSE: Residual Mean Squared
Errors, MSE: Mean Squared Errors, MAD: Mean Absolute Deviation.
Artificial Neural Network for Predicting Storage Time of Frozen Meat / Lakehal and Lakehal _______________________________________
6 of 6
[21] Li X, Zhang Y, Li Z, Li M, Liu Y, Zhang D. The effect of temperature
in the range of − 0.8 to 4°C on lamb meat color stability. Meat
Sci. 2017; 134:28–33. doi: https://doi.org/kgdr
[22] Medić H, Djurkin Kušec I, Pleadin J, Kozačinski L, Njari B, Hengl
B, Kušec G. The impact of frozen storage duration on physical,
chemical and microbiological properties of pork. Meat Sci. 2018;
140:119–27. doi: https://doi.org/gdhwf9
[23] Alonso V, Muela E, Tenas J, Calanche JB, Roncalés P, Beltrán
JA. Changes in physicochemical properties and fatty acid
composition of pork following long–term frozen storage. Eur.
Food Res. Technol. 2016; 242:2119–27. doi: https://doi.org/kgds
[24] Daszkiewicz T, Purwin C, Kubiak D, Fałkowska M, Kozłowska E,
Antoszkiewicz Z. Changes in the quality of meat (Longissimus
thoracis et lumborum) from Kamieniec lambs during long–term
freezer storage. Anim. Sci. J. 2018; 89:1323–30. doi: https://doi.
org/kgdt
[25] Vieira C, Diaz MT, Martínez B, García–Cachán MD. Effect of
frozen storage conditions (temperature and length of storage)
on microbiological and sensory quality of rustic crossbred beef
at different states of ageing. Meat Sci. 2009; 83:398–404. doi:
https://doi.org/dthjs3
[26] Coombs CEO, Holman BWB, Collins D, Friend MA, Hopkins DL.
Effects of chilled–then–frozen storage (up to 52 weeks) on lamb
M. longissimus lumborum quality and safety parameters. Meat
Sci. 2017; 134:86–97. doi: https://doi.org/kgdw
[27] Estévez M. Protein carbonyls in meat systems: A review. Meat
Sci. 2011; 89:259–79. doi: https://doi.org/dxcw6j
[28] Zhang K, Zhang B, Chen B, Jing L, Zhu Z, Kazemi K. Modeling
and optimization of Newfoundland shrimp waste hydrolysis
for microbial growth using response surface methodology and
articial neural networks. Mar. Pollut. Bull. 2016; 109:245–52.
doi: https://doi.org/gfkr59
