Food Control 131 (2022) 108418

Contents lists available at ScienceDirect

Food Control
journal homepage: www.elsevier.com/locate/foodcont

Fast online estimation of quail eggs freshness using portable NIR

spectrometer and machine learning
Yasmin Lima Brasil , J.P. Cruz-Tirado , Douglas Fernandes Barbin *
Department of Food Engineering, University of Campinas, Rua Monteiro Lobato, 80, Cidade Universitária, Campinas, SP, 13083-862, Brazil

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

Keywords: Quail eggs are one of the main natural sources of essential nutrients, presenting high amounts of protein, an­
Near infrared spectroscopy tioxidants, calcium, iron and phosphorus. However, its quality assessment demands laborious methods and
Chemometrics chemicals, and there is currently no standard method do quantify its freshness. This work aimed to investigate
Shelf life
the performance of a portable NIR spectrometer, in combination with machine learning, to estimate the freshness
Data mining
of quail eggs. Since there is no standard index to classify quail eggs, we compared Haugh Unit (HU), Yolk Index
(YI) and the Egg Quality Index (EQI) as reference methods. Partial Least Squares Regression (PLSR) and Support
Vector Machine Regression (SVMR) were used to build prediction models, and Partial Least Squares-Discriminant
Analysis (PLSDA) and Support Vector Machine Classification (SVMC) for the development of classification
models. For the first time, we demonstrated that EQI, which is a parameter that measures egg freshness according
to the quality of the yolk and the albumen, is the best way to express the freshness of quail eggs. The best
prediction models were obtained for YI and EQI, using SVMR, with RPD = 2.0–2.5 and RER >10, indicating good
predictive capacity. PLSDA and SVMC models showed similar performance, correctly classifying more than 80%
of the samples. The results obtained demonstrate the potential of portable NIR spectrometer for monitoring quail
eggs freshness during storage.

These reactions are accentuated with the increase in storage parameters

1. Introduction such as relative humidity, time and temperature, decreasing egg quality
for consumption due to deterioration of the internal quality, and higher
Eggs are one of the main natural sources of essential nutrients, such rate transfer of CO2 through the shell (Sari et al., 2016). Therefore, it is
as antioxidants, amino acids, proteins, fatty acids, in addition to important to estimate the real freshness of quail eggs for consumers’
numerous vitamins and minerals (Bao et al., 2020). In recent years, quail safety.
(Nothura maculosa) eggs have attracted a lot of attention due to their Egg aging can be estimated using several parameters associated to
nutritional and sensory properties. Compared to chicken eggs, quail eggs the characteristics of egg components. The most important parameters
have a higher nutritional value, with higher amounts of protein, anti­ that describe the quality of eggs during storage are albumen height,
oxidants, calcium, iron and phosphorus (Bao et al., 2020; Sun et al., Haugh unit (HU), pH of albumen and yolk, yolk index (YI) and air
2017). In Brazil, the production of quail eggs was estimated at 3786.95 chamber size (Kumbár et al., 2015). HU is the reference parameter used
billion eggs in 2019, placing the country as the second largest producer to estimate the freshness of chicken eggs, recognized by the Brazilian
in the world (IBGE, 2020). regulation of egg sanitation (BRASIL, 1990). This parameter is calcu­
Immediately after oviposition, aging affects the internal quality of lated using the height of the albumen and the weight of the egg, which
egg due to physical and chemical changes in the albumen and yolk, decreases with storage time due mainly to moisture loss. Although HU
caused by the loss of water and carbon dioxide (CO2) through the shell, has been established for estimate chicken eggs freshness, it is used to
and by osmotic exchanges between the albumen and the yolk through measure the freshness of eggs from other poultry species such as quail
the vitelline membrane (Giunchi et al., 2008). As consequence, there is (El-Tarabany, 2016), duck (Wang et al., 2014), partridge (Günhan &
an increase in egg acidity, liquefaction of the albumen, movement of Kırıkçı, 2017) and pigeon (Chang et al., 2019). Recently, Narushin et al.
liquids between the compartments, an increase in the air layer and (2021) proposed a new parameter called Egg Quality Index (EQI) that is
distension and flaccidity of the yolk membrane (Kumbár et al., 2015). proportional to the HU scale, but includes other features such as yolk

* Corresponding author.
E-mail address: [email protected] (D.F. Barbin).

https://doi.org/10.1016/j.foodcont.2021.108418
Received 31 May 2021; Received in revised form 7 July 2021; Accepted 11 July 2021
Available online 12 July 2021
0956-7135/© 2021 Elsevier Ltd. All rights reserved.
Y.L. Brasil et al. Food Control 131 (2022) 108418

et al., 2011). On the other hand, in recent years, low cost, robust and
Abbreviations ergonomic portable/handled NIR spectrometers were developed,
allowing real-time evaluations, online applications, convenience,
HU Haugh unit obtaining various simultaneous information and simple, fast and
YI Yolk index non-invasive analysis in scientific, industry and field research (de Car­
EQI Egg quality index valho et al., 2020; He et al., 2021; Oliveira et al., 2020; Pereirados et al.,
PLSR Partial least squares regression 2020). Recently, the application of a portable NIR spectrometer was
PLSDA Partial least square discriminant analysis investigated for determination of chicken eggs freshness (Cruz-Tirado
RMSEC Root mean square error of calibration et al., 2021). In this way, portable NIR spectrometers represent an
RMSECV Root mean square error of cross validation attractive and viable alternative for monitoring quail eggs freshness in
RMSEP Root mean square error of prediction the entire production chain.
RPD ratio of standard deviation In the best of our knowledge, there are no reports on the application
RER error reach rate of NIR spectroscopy, bench or portable, to determine quality attributes
R2CAL coefficient of determination of calibration of quail eggs. Therefore, the objective of this work was to investigate the
R2CV coefficient of determination of cross validation performance of a low-cost, portable NIR spectrometer in combination
R2Pred coefficient of determination of prediction with machine learning, to estimate quail egg freshness. In addition, for
the first time, we compared different freshness parameters that best
LV latent variables
MC mean centring explains the aging of quail eggs.
PCA principal component analysis
SG Savitzky Golay 2. Material and methods
SNV standard normal variate
SVMC support vector machine classification 2.1. Egg sampling
SVMR support vector machine regression
One thousand and one hundred (1100) freshly laid quail eggs (one-
day old, 11.34 ± 1.03 g in weight, 3.35 ± 0.16 cm high and 2.69 ± 0.17
cm wide (Fig. 1A) were obtained from local farms in Campinas (São
Paulo, Brazil). Unfertilized eggs were collected from the same group of
quails, which were raised under the same feeding and environmental
conditions. These were carefully transported from the farm to the lab­
oratory, where they were sanitized by immersion in chlorinated water
(50 ppm) at 40 ◦ C, for 1 min, according to the Brazilian regulation of egg
sanitation (BRASIL, 1990), and analysed for the presence of cracks or
damage. Subsequently, the eggs were dried at room temperature for 5
min, and then placed in polyethylene terephthalate (PET) trays. The
samples were stored in a BOD chamber at 50–65% RH and 20 ◦ C, for a
period of forty-one days. One hundred and twenty (1020) eggs were
immediately analysed as a control group for the characteristics of fresh
eggs. After, one hundred (100) eggs were randomly removed from
storage for each day of analysis (0, 4, 7, 10, 14, 17, 21, 24, 32, 41) for
spectra acquisition and determination of quality indexes (HU, EQI and
YI).

2.2. Spectra acquisition

The NIR spectra were acquired using a portable spectrometer (DLPR

NIRscanTM Nano, Texas Instruments, USA), in absorbance mode
(900–1700 nm) with 4 nm intervals and 20 repetitions for each spec­
Fig. 1. Quail egg variability (A), egg region for spectral acquisition (B) and
trum, using a halogen lamp (10 W) and an InGaAs detector. Quail eggs
measurement region of physical parameters (C). H: height of the thick albumen were placed in direct contact to the spectrometer. For each egg, three
(mm), h: height of the yolk (mm), L: width of the yolk (mm). acquisitions were made, at different points: bottom region (R1), central
(equatorial) region (R2) and top region (R3) (Cruz-Tirado et al., 2021)
height, being more representative to express chicken egg aging. (Fig. 1B), to obtain a better representation of the samples. In this way, a
These methods are destructive, time-consuming and require trained total of 3060 spectra were acquired (1020 spectra × 3 points of analysis:
analysts. Therefore, for moving to called Industry 4.0, the food industry R1, R2 and R3). The data matrices were created using the mean spectra
needs innovative and non-destructive techniques for simple, fast and (R1 + R2 + R3), and separately for each region (R1, R2 and R3), to
economical determination of chemical and physical compounds in food. compare the best region for spectral measurement and evaluate whether
Near-infrared spectroscopy (NIRS), associated with chemometric tools, the addition of more measurements would improve the performance of
meets these requirements, finding several applications for food prod­ prediction and classification models. Therefore, in total, four databases
ucts, such chemical and physical analysis, quality control, process (1020 spectra for each data set) were used to calibrate and validate the
monitoring and shelf life estimation in the food industry as an alterna­ models.
tive to conventional analytical techniques (Grassi & Alamprese, 2018;
McVey et al., 2021; Mendez et al., 2019; Nolasco Perez et al., 2018; 2.3. Egg quality indexes
Nolasco-Pérez et al., 2019; Walsh et al., 2020). Previous works reported
the efficiency of NIRS to estimate HU and air chamber size of chicken 2.3.1. Haugh unit (HU)
eggs (Aboonajmi & Abbasian Najafabadi, 2014; Giunchi et al., 2008; Lin The HU is measured as a function of the height of the thick albumen
and the weight of the egg. Eggs were initially weighed individually on a

2
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 2. Average value of the Haugh Unit (HU), Yolk Index (YI) and Egg Quality Index (EQI) and experimental observations of quail eggs stored for 41 days at 20 ◦ C.

semi-analytical scale (±0.0001 g), then gently broken on a flat glass

Yolk height
surface where the height of the albumen was measured using a stainless YI (%) = *100 Eq. (2)
Yolk width
steel analogue caliper (±0.01 mm). Measurements were made from two
points on the thick albumen (~10 mm from the yolk) (Fig. 1C). HU
2.3.3. Egg quality index (EQI)
values were calculated using Eq. (1) developed by Haugh (1937) for
The egg quality index (EQI) was estimated according to the egg
chicken eggs:
weight and the height of the thick albumen and yolk, according to Eq.
HU = 100x log(H + 7.57 − 1.7xW 0.37 ) Eq. (1) (3) proposed by Narushin et al. (2021):
√̅̅̅̅̅̅
where H is the average height of the thick albumen (millimeters) and W 100H Hh
EQI = 100 x Log √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Eq. (3)
is the weight of the egg (grams). 0.5W − 0.5)h + (0.4W + 16.5 − 1.3h3 )H

2.3.2. Yolk index (YI) where H is the average height of the thick albumen (millimeters), h is the
The YI is an indicator of the spherical nature of the yolk. After average of the yolk height (millimeters) and W is the weight of the egg
measuring the height of the thick albumen (H), still on the flat glass (grams).
surface the height (h) and width (L) of the yolk were measured using a
stainless steel analogue micrometre (±0.01 mm). The measurements 2.4. Data analysis
were made from two points (Fig. 1C), and based on the average values
obtained, the yolk index was estimated using Eq. (2) (Sharp & Powell, The spectral data analysis was performed in four main steps: pre-
1930): treatment of the data, exploratory analysis, modelling and model per­
formance evaluation. Data analysis was entirely performed using PLS-
Toolbox (Eigenvector Research, Inc.,Manson, WA, USA) for Matlab

3
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 3. Raw NIR spectra (A), pre-processed spectra (SNV + 1st derivative – SG) (B), score plot based on storage time (C), score plot based on freshness grading (fresh
or stale) (fresh and stale) (D) and loading plot (E).

R2019a (Mathworks, Natick, USA). Spectra were pre-processed for light calibration set (70% data) for the construction and internal cross-
scattering (derived from different data acquisition) and baseline validation (using 10-fold cross-validation), and (2) external prediction
correction. Therefore, after different pore-processing combinations, the set (30%) used to validate the predictive capacity of the models.
best approach was the combination of mean centre (MC), the standard
normal variate (SNV) and 1st derivative (Savitzky Golay - SG smoothing, 2.4.1. Exploratory analysis
11 points window, 2nd polynomial order). Regression models were To reduce the size of the data set and obtain an overview of the
developed to predict EQI and HU indexes for quail eggs. Classification variation between samples, principal component analysis (PCA) was
models were developed in order to offer a quick screening of egg used as an exploratory analysis. This technique consists of linear trans­
freshness. However, regression models were needed to identify whether formation of an original set of variables, initially correlated with each
the sample is near the expiring date. For both prediction and classifi­ other, into a smaller set of non-correlated variables, called principal
cation models, the data sets were divided into two subsets: (1) components (PCs), without losing significant information. PCA was

4
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 4. PLSR models to predict HU, IG and EQI in the three different measurement points (R1: bottom, R2: middle, R3: top). Grey circle are calibration samples, and
red diamonds are validation samples. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

performed on pre-processed spectral data to evaluate the variation of regression (SVMR) were built, based on pre-processed spectra, to predict
quail egg quality during storage. In addition, the outliers (anomalous egg quality parameters (HU, YI and EQI). The choice of the number of
samples) were identified and eliminated using the values of Qresiduals and latent variables (LVs) that should be included in the PLSR models was
T2 Hotellings. made based on the smallest mean square error of the cross-validation
(RMSECV), with a maximum of 12 LVs. PLSR is the most popular
2.4.2. Prediction of egg quality parameters regression model for building NIR prediction model, which is based on
Partial least squares regression (PLSR) and support vector machines the linear regression relationship between NIR data and dependent

5
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 5. SVMR models to predict HU, IG and EQI in the three different measurement points (R1: bottom, R2: middle, R3: top).

variables. SVMR is a supervised and nonparametric kernel-based base function (RBF), and their optimization based on the parameters c
learning algorithm (Vapnik, 2013), which uses training vector to de­ (cost) and ε (epsilon) with a grid search, based on the smallest RMSECV
fines the best sparse deterministic regression relationship between the of a 10-fold cross validation.
NIR data and the response values (Grelet et al., 2020). In contrast to The reliability and robustness of the regression models were evalu­
PLSR, SVMR is able to model nonlinear relations, and it also can effi­ ated by the calculation of the root of the mean square error of calibra­
ciently work with high dimensional input vectors (Wang et al., 2014). In tion, cross-validation and prediction (RMSEC, RMSECV and RMSEP), of
this study, SVMR was used after compression of the data by PLS using 10 the error range rate (RER), provided from the ratio between the cali­
latent variables. The SVMR models were developed using a radial kernel bration interval (interval) and the RMSEP (Eq. (4)), and the residual

6
Y.L. Brasil et al. Food Control 131 (2022) 108418

Table 1
Performance of the PLSR and SVMR models for prediction of the Haugh Unit (HU), Yolk Index (YI) and Egg Quality Index (EQI) of quail eggs.
Model Index Spectra origin RMSEC RMSECV RMSEP R2CAL R2CV R2PRED RPD RER

PLSR HU R1 3.90 4.15 4.12 0.68 0.63 0.65 1.70 7.11

R2 4.02 4.23 4.18 0.66 0.63 0.64 1.67 7.01
R3 3.73 3.99 4.24 0.71 0.67 0.65 1.65 6.92
Mean 3.70 3.91 3.90 0.71 0.67 0.69 1.79 7.50
YI R1 3.71 3.97 3.89 0.76 0.72 0.76 2.00 10.25
R2 3.46 3.80 3.79 0.79 0.75 0.77 2.05 10.53
R3 3.58 3.82 3.73 0.78 0.75 0.77 2.08 10.67
Mean 3.45 3.65 3.64 0.80 0.77 0.79 2.13 10.96
EQI R1 10.94 11.68 11.31 0.72 0.68 0.70 1.92 8.25
R2 11.23 11.93 11.41 0.71 0.67 0.70 1.91 8.18
R3 10.49 11.17 11.95 0.75 0.72 0.70 1.82 7.81
Mean 10.54 11.16 11.02 0.75 0.72 0.72 1.97 8.47
SVMR HU R1 2.36 3.80 3.76 0.89 0.70 0.71 1.86 7.79
R2 2.66 3.65 3.70 0.85 0.72 0.73 1.89 7.92
R3 3.02 3.67 3.69 0.81 0.72 0.72 1.90 7.94
Mean 3.12 3.74 3.71 0.79 0.70 0.73 1.88 7.90
YI R1 3.12 3.75 3.75 0.83 0.76 0.78 2.07 10.63
R2 2.59 3.57 3.36 0.89 0.78 0.81 2.31 11.87
R3 2.95 3.66 3.60 0.85 0.77 0.79 2.15 11.06
Mean 3.11 3.52 3.37 0.83 0.79 0.81 2.30 11.82
EQI R1 6.40 10.31 9.93 0.91 0.76 0.78 2.19 9.40
R2 7.11 10.17 10.03 0.89 0.77 0.79 2.17 9.30
R3 7.26 10.12 10.15 0.88 0.77 0.77 2.14 9.19
Mean 6.81 9.42 9.26 0.90 0.80 0.82 2.35 10.08

prediction deviation (RPD), as the ratio between the standard deviation

TP
of the calibration set (DPcal) and the average values of the RMSEP (Eq. Sensibility (%) = x100 Eq. (6)
(TP + FN)
(5)). RER values > 10 are indicative of models with a good estimate
(Páscoa et al., 2013). RPD values < 1.5 indicate that the model cannot be TN
used, RPD = 1.5–2.0 states that the model can distinguish between high Specificity (%) = x100 Eq. (7)
(TN + FP)
and low values, RPD = 2.0–2.5 evolution a model with approximate
prediction capacity, RPD = 2.5–3.0 indicates a good model for predic­ (TP + TN)
tion, and RPD >3 requiring excellent predictive capacity of the model Accuracy (%) = x100 Eq. (8)
Total
(Saeys et al., 2005).
(FP + FN)
RER =
range
Eq. (4) Error (%) = x100 Eq. (9)
RMSEP Total

DPcal where TP = true positive, TN = true negative, FP = false positive and FN

RPD = Eq. (5) = false negative.
RMSEP

2.4.3. Egg freshness classification 2.5. Statistical analysis

Partial least squares discriminant analysis (PLSDA) and the support
vector machine classification (SVMC) were explored in order to assess Analysis of variance (ANOVA) followed by the Tukey test (p < 0.05)
the efficiency of the portable NIR in classifying quail eggs according to were performed to compare the averages of quality parameters of quail
their freshness (fresh and stale), based on HU and EQI values. The choice eggs during storage. The software used to perform the statistical calcu­
of the number of latent variables (LVs) that should be included in the lations was Statistica version 8.0 (Statsoft, Oklahoma, USA).
PLSDA models was made based on the smallest mean square error of the
cross-validation (RMSECV), with a maximum of 12 LVs. The classifica­ 3. Results and discussion
tion by SVMC obtains an optimal hyperplane for which the margin of
separation between the different classes is maximum. In this study, 3.1. Freshness of eggs during storage
SVMC was compacted using 10 PCs. The SVMC models were developed
using a radial kernel base function (RBF), and their optimization based The evolution of quail eggs freshness was calculated considering
on the parameters c (cost) and ε (epsilon) with a grid search, based on quality parameters such as the Haugh unit (HU), yolk index (YI) and egg
the smallest RMSECV of a 10-fold cross validation. quality index (EQI), as well as experimental observations (Fig. 2). Due to
The performance of the classification models was evaluated by the the lack of previous references on the shelf life of quail eggs, experi­
sensitivity, specificity, precision and error rates for the calibration/ mental observations, such as observation of the size of the air layer and
cross-validation and prediction sets. Sensitivity (Eq. (6)) and speci­ the density reduction caused by weight loss, were essential to verify
ficity (Eq. (7)) can be described as the model’s ability to correctly whether the eggs were fresh or stale.
classify positive and negative samples, respectively. The precision (Eq. Fig. 2 shows that the HU decreases significantly (Tukey’s test, p <
(8)) and error (Eq. (9)) rates are statistical parameters that provide a 0.05) only after 14 days of storage, showing a decrease of only 21.55% at
global value to measure the model’s performance and error. These are the end of the storage period. Akpinar et al., (2015) observed a decrease
calculated using the ratio between the number of samples correctly of 13.03% for quail eggs stored at 25 ◦ C for 35 days. Kumbár et al.,
classified and misclassified, regardless of the class, and the total number (2015) also reported a small change of 4.38% in HU values, for quail
of samples, respectively. eggs stored for 42 days at 4 ◦ C. During the storage of eggs, the pH of the
albumen increases due to CO2 losses through the shell, which cause the
hydrolysis of the amino acid chains, which release water that was linked

7
Y.L. Brasil et al. Food Control 131 (2022) 108418

Table 2 (Fig. 2), reinforcing the idea that they can more accurately describe the
Performance of PLSDA and SVMC models for classification of quail eggs ac­ freshness of quail eggs stored at 20 ◦ C.
cording to their freshness using Haugh Unit (HU) and Egg Quality Index (EQI). On the other hand, EQI has a high correlation with HU, which means
Model Index Spectra Error SEN SPE ACC that the values defined in the HU scale to classify the freshness of
origin (%) (%) (%) (%) chicken eggs are practically the same for EQI (Class AA: EQI ≥ 72; Class
PLSDA HU R1 Cal 12.1 87.8 88.1 87.9 A: EQI = 71–60; Class B: EQI <60; or EQI ≥ 70 for fresh eggs and EQI
CV 14.5 85.8 85.3 85.5 <70 for stale eggs). Therefore, analysing Fig. 2, we can define that eggs
Pred 15.0 89.6 80.5 85.0 stored until the 14th day can be considered fresh. Previous work has
R2 Cal 10.7 88.1 90.5 89.3
shown that chicken eggs stored at 20 ◦ C remained fresh until the 14th
CV 13.8 85.5 87.0 86.2
Pred 14.8 90.2 80.3 85.3 day (Cruz-Tirado et al., 2021; Liu et al., 2020). In addition, still based on
R3 Cal 8.5 91.1 91.8 91.5 the correlation between EQI and HU, it is possible to redefine the HU
CV 12.3 87.5 87.9 87.7 scale specifically for quail eggs, with HU ≥ 80 for fresh eggs (related to
Pred 17.7 81.3 83.2 82.3 day 14) and HU < 80 for stale eggs.
Cal 7.7 92.8 91.8 67.5
Mean CV 10.1 90.2 89.7 89.9
Pred 14.6 89.4 81.5 85.4 3.2. Spectral and exploratory analysis
EQI R1 Cal 11.7 90.6 86.0 88.3
CV 15.1 87.9 81.9 84.9 Fig. 3 shows the raw spectra (Fig. 3A), pre-processed spectra by SNV
Pred 18.1 92.0 71.8 81.5
+1st S-G derivative (Fig. 3B), PCA analysis based on storage time
R2 Cal 11.9 92.6 83.6 88.1
CV 14.8 89.4 81.0 85.2 (Fig. 3C), PCA analysis based on freshness grading (fresh or stale)
Pred 20.5 92.4 66.7 79.3 (Fig. 3D) and loadings plot (Fig. 3E). PCA analysis was carried out in two
R3 Cal 8.2 93.7 89.9 91.8 forms: (1) to observe the spectra variations of quail eggs samples during
CV 10.8 92.2 86.2 89.1 storage, which include nine classes (or days), and (2) freshness grading
Pred 16.8 86.8 79.5 83.2
(fresh and stale).
Cal 8.3 93.3 90.1 91.8
Mean CV 10.8 92.0 86.4 89.1 Fig. 3A and B shows the graphs of the raw and pre-processed mean
Pred 15.0 89.4 80.7 85.0 spectra. The difference in the baseline between fresh and stale eggs
SVMC HU R1 Cal 11.0 88.3 89.7 90.6 observed in the raw spectra (Fig. 3A), corresponds to the change in light
CV 17.4 81.5 83.8 82.6
scattering properties, probably associated with the air chamber and the
Pred 17.5 82.4 82.7 82.5
R2 Cal 12.5 89.6 85.5 89.8 degradation of the inner film of quail eggs. The regions of 938 nm and
CV 15.5 85.7 83.3 84.5 the peak around 1334 nm, are associated with the C–H elongation,
Pred 17.6 86.7 78.2 82.5 characteristic of saturated fatty acids (fat) present in the yolk (Hourant
R3 Cal 9.6 91.9 89.0 92.8 et al., 2000). The absorption band at 1188 nm have been described as
CV 15.5 85.7 83.3 84.5
the second overtone of C–H absorption of pure fatty acids containing cis
Pred 15.7 83.1 85.6 84.3
Cal 8.4 93.8 89.3 93.8 double bonds (Sato et al., 1991), which could be associated with the
Mean CV 13.6 89.0 83.7 86.4 higher content of oleic acid in quail egg yolk (Arantes da Silva et al.,
Pred 14.7 88.4 82.1 85.3 2009; Choi et al., 2001). The peak observed at 989 and 1400 nm is
EQI R1 Cal 17.8 95.0 69.4 85.1
associated with the O–H stretching and elongation, respectively, refer­
CV 18.9 94.0 68.1 81.1
Pred 20.2 90.0 69.5 79.8
ring to water. The difference in absorbance in this region may be linked
R2 Cal 6.0 98.3 89.6 96.6 to the loss of moisture through the pores of the shell, as well as the
CV 15.8 92.2 76.2 84.2 structural change of proteins in quail eggs during storage (Cruz-Tirado
Pred 20.4 92.6 66.7 79.6 et al., 2021). The peaks around 1334 and 1643 nm are related to the
R3 Cal 13.0 96.8 77.2 88.8
second aromatic C–H elongation overtone, mainly related to CH2 and
CV 17.9 96.0 68.3 82.2
Pred 19.0 92.5 69.4 81.0 CH3 (Kaufmann et al., 2019). The increase in absorbance for stale quail
Cal 10.9 95.3 82.9 91.9 eggs is probably associated with the release of carbon dioxide by the
Mean CV 12.3 94.0 81.5 87.8 shell (Jiang et al., 2018).
Pred 17.4 93.8 71.4 82.6
Fig. 3 shows the score plot for spectral data according to storage day,
SEN: sensitivity; SPE: specificity; ACC: accuracy. with PC1 explains 83.74% and PC2 explains 6.59%. It is possible to
observe as samples were moving from positive to negative scores on PC1
to large protein molecules, mainly ovomucine, resulting in liquefaction with the increasing of storage day. In addition, the freshest samples (Day
of thick albumen (de Moraes Garcia et al., 2015). Therefore, if we use the 0) and oldest sample (Day 41) were characterized for negative scores on
HU scale as established for chicken eggs, all quail eggs would be clas­ PC2. The egg samples from Day 14 is overlapped in the centre of PCA
sified as fresh, which is not consistent with experimental observations score plot, indicating a transition phase from fresh (0, 4, 7 days) to stale
(Fig. 2). As for the yolk index (YI), it is possible to observe a significant (17, 21, 24, 32, 41) grading. Based on the physicochemical analyses
decrease (Tukey’s test, p < 0.05) during storage. This is due to the (Fig. 2), it is possible to consider that the samples with more than 14
increased permeability and weakening of the vitelline membrane, days of storage are stale. Therefore, Fig. 3D was plotted using threshold
facilitating the displacement of excess water from the albumen to the of 14 days for clustering fresh (≤14 days) and stale (>14 days).
yolk, making it flabby and flattened (Marinho, 2011), which is visible in Although not clearly observed, the majority of fresh eggs are located on
digital images of boiled eggs (Fig. 2). Similarly, the egg quality index positive PC1 scores, while stale eggs are located in negative PC1 scores.
(EQI) decreased with storage time (Fig. 2). Due to the fact that EQI is a The loadings plot in Fig. 3E shows that most of the peaks and valleys that
function of both the height of the albumen and yolk, this decreasing was contributed for sample clustering are those associated with water, fat
probably caused by the degradation of the yolk, instead of changes in the and aromatic compounds. This indicates that the spectral information
albumen, which is supported both by the experimental observations collected with the portable NIR is strongly associated with the variations
(Fig. 2) and by the evolution of the HU and YI parameters. Therefore, it in the composition of the quail eggs during storage, and can be used for a
is possible to conclude that the storage period is a crucial factor affecting screening of eggs according to their freshness.
quail eggs quality. However, it should be noted that YI and EQI were
more significantly influenced by the storage time, when compared to HU

8
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 6. PLSDA models for classification of quail eggs according to their freshness: confusion matrix for Haugh Unit (HU) and Egg Quality Index (EQI).

3.3. Prediction of egg quality indexes general, the SVMR models showed better performance than PLSR
models, which leads us to suggest that perhaps the relationship between
The prediction performance of egg quality indexes (HU, YI and EQI) the NIR spectra and the parameters HU, YI and EQI tends to be
using PLSR and SVMR models are shown in Fig. 4 and Fig. 5, respec­ non-linear instead of linear. Thus, linear methods may not completely
tively. In this study, the performance of the models was preferably show the relationship between the NIR spectra and the freshness of quail
evaluated based on error (RMSECV and RMSEP) and related variables eggs. The best performance of PLSR was achieved for data set from mean
(RER and RPD), since R2 is dependent on the range of the data set spectra of three measurement points (R1 + R2 + R3) for all quality in­
(Davies & Fearn, 2006). The best models are considered those that have dexes (Table 1). In contrast, the models built using the SVMR algorithm
lower values of RMSECV and RMSEP and higher values of RPD and RER, showed the best performance in the R3, R2 and mean spectra for HU, IG
as these measurements are made to quantitatively verify the precision of and EQI predictions, respectively (Table 1). Based on the RPD and RER,
the predictive capacity of the chemometric models (Conzen, 2006). In for the PLSR only the models for estimating the YI (RPD = 2.00–2.13 and

9
Y.L. Brasil et al. Food Control 131 (2022) 108418

Fig. 7. SVMC models for classifying quail eggs according to their freshness: confusion matrix for Haugh Unit (HU) and Egg Quality Index (EQI).

RER = 10.25–10.96), and for the SVMR only the models to estimate YI best RPD value reached 1.90 using SVMR model based on spectra from
(RPD = 2.07–2.31 and RER = 10.63–11.87) and EQI (RPD = 2.14–2.35 R3 region. Based on RPD scale, the model can be considered for
and RER = 9.19–10.08) indicate a good correlation between the values screening between higher and lower values (Saeys et al., 2005). During
predicted by the model and the real (experimental) values. Figs. 3 and 4 storage, R3 is located in the egg region of greatest contact with the
show the best PLSR and SVMR models, separately (R1; R2; R3) and albumen (top), making the light from the NIR sensor, passing through
based on the mean (R1 + R2 + R3), respectively. the pores of the egg shell, stay in direct contact with the albumen during
HU regression models showed the lowest R2P values (0.63–0.73) and the period evaluated. In contrast, regions R2 and R1 are the region
the higher RMSEP (3.69–4.12) for both PLSR and SVMR model (Fig. 1). where the albumen liquefies during storage causing the yolk to approach
This was probably because the R2 is dependent on the rank of the the shell (R2) and to increase the air chamber (R1) (Fig. 2).
reference values and HU had the lowest reference range. In addition, the For YI, the SVM and PLS models presented similar performance with

10
Y.L. Brasil et al. Food Control 131 (2022) 108418

RPD ranging from 2.0 to 2.5 and RER> 10, which suggests a model with 4. Conclusion
reasonable prediction capacity. However, the best PLSR model (RPD =
2.13) required spectral information from the three regions, while the We demonstrate that the combination of NIR spectroscopy with
best SVMR model (RPD = 2.30) required only spectra fom R2 region, chemometric tools provides a fast, non-invasive and high-performance
optimizing the time to collect and process spectral information. This system for monitoring internal quality parameters and assessing the
may be related to the fact that during storage the yolk approaches the freshness of quail eggs during storage. EQI is a better parameter to
shell (Fig. 2), allowing a better interaction with the NIR, and conse­ represent quail egg freshness when compared to HU. Hence, we suggest
quently obtaining more representative spectral information in R2. The the utilization of EQI to determine quail egg freshness, and also that HU
egg spectral profile (Fig. 3B) shows characteristics bands associated with scale for quail eggs to be restructured as HU ≥ 80 for fresh eggs and HU
the lipid compounds present in the yolk. < 80 for old eggs. The regression models with greater predictive ca­
Regarding the EQI, the SVMR models (RPD = 2.35 and RER = 10.08) pacity and robustness were obtained for YI and EQI, using only one
showed a better predictive capacity, when compared to the PLSR models spectral region (R2) when developed through SVMR. In contrast, for
(RPD = 1.97 and RER = 8.47). These results indicate that the SVMR PLSR models it was necessary to use spectral information of three points
model has a reasonable prediction capacity for EQI values. The best (R1 + R2 + R3). The best PLSDA and SVMC models for HU and EQI
models, for both algorithms, were obtained using the mean spectra of correctly classified more than 80% of the samples, due to the reconfi­
the three regions. A better performance for the EQI regression models, guration of the HU scale proposed in this study. Finally, it can be
compared to the HU models, may be related to the fact that the spectral concluded that the physical and chemical properties of quail eggs are
information of the NIR presents representative absorption peaks asso­ closely related to the storage period, and that NIR spectroscopy is a
ciated with the yolk components (Fig. 3B). promising alternative for assessing the freshness of quail eggs, replacing
As stated by (Dong et al., 2019), the eggshell thicknesses at the conventional destructive methods.
bottom (R1), middle (R2) and top (R3) are different, so that each region
would have an influence on the spectra. In addition, the composition of CRediT authorship contribution statement
egg white and yolk, and the composition of egg cuticle, change with
storage time (Samiullah & Roberts, 2014). In R2 region, when the light Yasmin Lima Brasil: Conceptualization, Methodology, Software,
falls on the pores of the shell and comes into contact with the cuticle, Validation, Formal analysis, Investigation, Writing – original draft. J.P.
which in turn is in contact with albumen, the spectrum can better detect Cruz-Tirado: Conceptualization, Methodology, Software, Data cura­
the freshness of the egg. On the other hand, the air chamber is located at tion, Writing – review & editing. Douglas Fernandes Barbin:
the bottom of the egg (R1), with the diameter and height increasing Conceptualization, Resources, Writing – review & editing, Supervision,
during the storage. Therefore, they have a high correlation with the Project administration, Funding acquisition.
freshness of the egg (Liu et al., 2020). Cruz-Tirado et al. (2021) also
reported that the better predictive capacity for chicken eggs was reached
for HU regression models based on mean spectra of three measurements Declaration of competing interest
using SVMR algorithm.
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
3.4. Egg freshness grading the work reported in this paper.

Table 2 shows the performance of the PLSDA and SVMC classifica­ Acknowledgment
tion models according to the error rates, sensitivity, specificity and
precision for each spectral region separately (R1; R2; R3) and for the This research was subsidized by the Coordenação de Aperfeiçoa­
mean spectra (R1 + R2 + R3). The confusion matrices for classification mento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001
results of the PLSDA and SVMC models into “fresh” (HU ≥ 80; EQI ≥ 70) and São Paulo Research Foundation (FAPESP) (project number 2020/
and “stale” (HU < 80; EQI < 70) are shown in Fig. 6 and Fig. 7, 09198–1, 2018/02500–4, 2015/24351–2). Yasmin Lima Brasil thanks
respectively. the FAPESP grant, grant nº 2019/11896–1. We are also grateful to our
The PLSDA and SVMC models for classifying quail eggs into “fresh” academic colleagues for their contributions.
and “stale”, defined by HU and EQI values, showed similar performance
(Table 2). This indicates that the restructuring of the HU scale (based on
References
the correlation with EQI) (Narushin et al., 2021), proposed in this study
can properly express quail egg freshness, which can be used for classi­ Aboonajmi, M., & Abbasian Najafabadi, T. (2014). Prediction of poultry egg freshness
fication purposes. The best detection region for the construction of the using vis-nir spectroscopy with maximum likelihood method. International Journal of
PLSDA and SVMC models was observed in the mean spectra. This Food Properties, 17(10), 2166–2176. https://doi.org/10.1080/
10942912.2013.784330
behavior is probably related to the fact that the medium spectra are Akpinar, G. C., Canogullari, S., Baylan, M., Alasahan, S., & Aygun, A. (2015). The use of
more representative to describe the changes in the characteristics of all propolis extract for the storage of quail eggs. The Journal of Applied Poultry Research,
the components of the quail eggs (air chamber, albumen and yolk). The 24(4), 427–435. https://doi.org/10.3382/japr/pfv043
Arantes da Silva, W., Naiverti Elias, A. H., Aricetti, J. A., Sakamoto, M. I.,
best PLSDA and SVMC model was able to correctly classify 110/133 Murakami, A. E., Marques Gomes, S. T., Visentainer, J. V., Evelázio de Souza, N., &
(82.7%) and 107/129 (82.9%) ‘fresh’ and 101/114 (87.8%) and Matsushita, M. (2009). Quail egg yolk (Coturnix coturnix japonica) enriched with
101/115 (85, 4%) ‘old’ for HU, while 143/160 (89.4%) and 150/174 omega-3 fatty acids. Lebensmittel-Wissenschaft und -Technologie- Food Science and
Technology, 42(2), 660–663. https://doi.org/10.1016/j.lwt.2008.08.005
(86.2%) in ‘new’ and 71/88 (80.7%) and 60/70 (85, 7%) ‘old’ were Bao, Z., Kang, D., Li, C., Zhang, F., & Lin, S. (2020). Effect of salting on the water
classified for EQI (Figs. 6 and 7), respectively. These models achieved migration , physicochemical and textural characteristics , and microstructure of
rates of sensitivity and specificity greater than 80.0% and error rates less quail eggs. Lebensmittel-Wissenschaft & Technologie, 132(1), 109847. https://doi.org/
10.1016/j.lwt.2020.109847
than 20.0% for both classes (Table 2). Most of the classes classified
BRASIL. (1990). Ministério da Agricultura, Pecuária e Abastecimento. Portaria n◦ 1, de 21 de
incorrectly (Figs. 6 and 7) belong to the transition group (days 14 and fevereiro de.
17), as these have similar HU and EQI values, but belong to different de Carvalho, L. M., Madruga, M. S., Estévez, M., Badaró, A. T., & Barbin, D. F. (2020).
classes. Therefore, the PLSDA and SVMC models can be used to identify Occurrence of wooden breast and white striping in Brazilian slaughtering plants and
use of near-infrared spectroscopy and multivariate analysis to identify affected
and discriminate quail fresh eggs fresh from stale eggs, which is very chicken breasts. Journal of Food Science, 85, 3102–3112. https://doi.org/10.1111/
useful for the egg products industry. 1750-3841.15465

11
Y.L. Brasil et al. Food Control 131 (2022) 108418

Chang, L., Zhang, R., Fu, S., Mu, C., Tang, Q., & Bu, Z. (2019). Effects of different dietary Marinho, A. L. (2011). Qualidade interna e externa de ovos de codornas japonesas
calcium levels on the performance, egg quality, and albumen transparency of laying armazenados em diferentes temperaturas e períodos de estocagem.
pigeons. Animals, 9(Issue 3). https://doi.org/10.3390/ani9030110 McVey, C., McGrath, T. F., Haughey, S. A., & Elliott, C. T. (2021). A rapid food chain
Choi, S. H., Song, K. T., & Oh, H. R. (2001). Cholesterol contents and fatty acid approach for authenticity screening: The development, validation and transferability
composition of chukar, pheasant, Guinea fowl and quail egg yolk. Asian-Australasian of a chemometric model using two handheld near infrared spectroscopy (NIRS)
Journal of Animal Sciences, 14(6), 831–836. https://doi.org/10.5713/ajas.2001.831 devices. Talanta, 222, 121533. https://doi.org/10.1016/j.talanta.2020.121533
Conzen, J. P. (2006). Multivariate calibration: A practical guide for developing methods in the Mendez, J., Mendoza, L., Cruz-Tirado, J. P., Quevedo, R., & Siche, R. (2019). Trends in
quantitative analytical chemistry. Bruker optik GmbH. Germany: Ettlingen. application of NIR and hyperspectral imaging for food authentication. In Scientia
Cruz-Tirado, J. P., Lucimar da Silva Medeiros, M., & Barbin, D. F. (2021). On-line agropecuaria (Vol. 10, pp. 143–161). scielo.
monitoring of egg freshness using a portable NIR spectrometer in tandem with de Moraes Garcia, E. R., de Ávila, L. R., da Cruz, F. K., de Pedro Souza, R. P.,
machine learning. Journal of Food Engineering, 306, 110643. https://doi.org/ Batista, N. R., Feliciano, W. B., & Arguelo, N. N. (2015). Qualidade de ovos de
10.1016/j.jfoodeng.2021.110643 codornas japonesas: Efeito da idade da ave, temperatura de conservação e período de
Davies, A. M. C., & Fearn, T. (2006). Back to basics: Calibration statistics. Spectroscopy armazenamento. Arquivos de Ciências Veterinárias e Zoologia Da UNIPAR, 18(4).
Europe, 18(2), 31–32. Narushin, V. G., Romanov, M. N., & Griffin, D. K. (2021). A novel Egg Quality Index as an
Dong, X., Dong, J., Li, Y., Xu, H., & Tang, X. (2019). Maintaining the predictive abilities alternative to Haugh unit score. Journal of Food Engineering, 289, 110176. https://
of egg freshness models on new variety based on VIS-NIR spectroscopy technique. doi.org/10.1016/j.jfoodeng.2020.110176. March 2020.
Computers and Electronics in Agriculture, 156, 669–676. https://doi.org/10.1016/j. Nolasco Perez, I. M., Badaró, A. T., Barbon, S., Jr., Barbon, A. P. A., Pollonio, M. A. R., &
compag.2018.12.012. October 2018. Barbin, D. F. (2018). Classification of chicken parts using a portable near-infrared
El-Tarabany, M. S. (2016). Effect of thermal stress on fertility and egg quality of Japanese (NIR) spectrophotometer and machine learning. Applied Spectroscopy, 72(12),
quail. Journal of Thermal Biology, 61, 38–43. https://doi.org/10.1016/j. 1774–1780.
jtherbio.2016.08.004 Nolasco-Perez, I. M., Rocco, L. A. C. M., Cruz-Tirado, J. P., Pollonio, M. A. R., Barbon, S.,
Giunchi, A., Berardinelli, A., Ragni, L., Fabbri, A., & Silaghi, F. A. (2008). Non- Barbon, A. P. A. C., & Barbin, D. F. (2019). Comparison of rapid techniques for
destructive freshness assessment of shell eggs using FT-NIR spectroscopy. Journal of classification of ground meat. Biosystems Engineering, 183, 151–159.
Food Engineering, 89(2), 142–148. https://doi.org/10.1016/j.jfoodeng.2008.04.013 Oliveira, M. M., Cruz-Tirado, J. P., Roque, J. V., Teófilo, R. F., & Barbin, D. F. (2020).
Grassi, S., & Alamprese, C. (2018). Advances in NIR spectroscopy applied to process Portable near-infrared spectroscopy for rapid authentication of adulterated paprika
analytical technology in food industries. Current Opinion in Food Science, 22, 17–21. powder. Journal of Food Composition and Analysis, 87.
https://doi.org/10.1016/j.cofs.2017.12.008 Páscoa, R. N. M. J., Magalhães, L. M., & Lopes, J. A. (2013). FT-NIR spectroscopy as a
Grelet, C., Froidmont, E., Foldager, L., Salavati, M., Hostens, M., Ferris, C. P., tool for valorization of spent coffee grounds: Application to assessment of
Ingvartsen, K. L., Crowe, M. A., Sorensen, M. T., Fernandez Pierna, J. A., antioxidant properties. Food Research International, 51(2), 579–586. https://doi.org/
Vanlierde, A., Gengler, N., & Dehareng, F. (2020). Potential of milk mid-infrared 10.1016/j.foodres.2013.01.035
spectra to predict nitrogen use efficiency of individual dairy cows in early lactation. Pereira, E. V., dos, S., Fernandes, D. D. de S., de Araújo, M. C. U., Diniz, P. H. G. D., &
Journal of Dairy Science, 103(5), 4435–4445. https://doi.org/10.3168/jds.2019- Maciel, M. I. S. (2020). Simultaneous determination of goat milk adulteration with
17910 cow milk and their fat and protein contents using NIR spectroscopy and PLS
Günhan, Ş., & Kırıkçı, K. (2017). Effects of different storage time on hatching results and algorithms. Lebensmittel-Wissenschaft & Technologie, 127(April), 109427. https://doi.
some egg quality characteristics of rock partridge (A. graeca) (management and org/10.1016/j.lwt.2020.109427
production). Poultry Science, 96(6), 1628–1634. https://doi.org/10.3382/ps/ Saeys, W., Mouazen, A. M., & Ramon, H. (2005). Potential for onsite and online analysis
pew443 of pig manure using visible and near infrared reflectance spectroscopy. Biosystems
Haugh, R. R. (1937). The Haugh unit for measuring egg quality. United States Egg and Engineering, 91(4), 393–402. https://doi.org/10.1016/j.biosystemseng.2005.05.001
Poultry Magazine, 43, 522–555. Samiullah, S., & Roberts, J. R. (2014). The eggshell cuticle of the laying hen. World’s
He, X., Zhao, T., Shen, F., Liu, Q., Fang, Y., & Hu, Q. (2021). Online detection of naturally Poultry Science Journal, 70(4), 693–708. https://doi.org/10.1017/
DON contaminated wheat grains from China using Vis-NIR spectroscopy and s0043933914000786
computer vision. Biosystems Engineering, 201, 1–10. https://doi.org/10.1016/j. Sari, M., Tilki, M., & Saatci, M. (2016). Genetic parameters of egg quality traits in long-
biosystemseng.2020.11.001 term pedigree recorded Japanese quail. Poultry Science, 95(8), 1743–1749. https://
Hourant, P., Baeten, V., Morales, M. T., Meurens, M., & Aparicio, R. (2000). Oil and fat doi.org/10.3382/ps/pew118
classification by selected bands of near-infrared spectroscopy. Applied Spectroscopy, Sato, T., Kawano, S., & Iwamoto, M. (1991). Near infrared spectral patterns of fatty acid
54(8), 1168–1174. http://as.osa.org/abstract.cfm?URI=as-54-8-1168. analysis from fats and oils. Journal of the American Oil Chemists Society, 68(11),
IBGE. (2020). Pesquisa da Pecuária Municipal. https://agenciadenoticias.ibge.gov.br/ 827–833. https://doi.org/10.1007/BF02660596
media/com_mediaibge/arquivos/b25372bebfb621f8c789c4fda346d1a9.pdf. Sharp, P. F., & Powell, C. K. (1930). Decrease in interior quality of hens’ eggs during
Jiang, S., Zhu, T., Jia, D., Yao, J., & Jiang, Y. (2018). Effect of egg freshness on their storage as indicated by the yolk. Industrial & Engineering Chemistry, 22(8), 908–910.
automatic orientation. Journal of the Science of Food and Agriculture, 98(7), https://doi.org/10.1021/ie50248a031
2642–2650. https://doi.org/10.1002/jsfa.8757 Sun, C., Liu, J., Yang, N., & Xu, G. (2017). Egg quality and egg albumen property of
Kaufmann, K. C., Favero, F. de F., de Vasconcelos, M. A. M., Godoy, H. T., Sampaio, K. A., domestic chicken , duck , goose , Turkey , quail , and pigeon. Poultry Science, 98(10),
& Barbin, D. F. (2019). Portable NIR spectrometer for prediction of palm oil acidity. 4516–4521. https://doi.org/10.3382/ps/pez259
Journal of Food Science, 84(3), 406–411. https://doi.org/10.1111/1750-3841.14467 Vapnik, V. (2013). The nature of statistical learning theory. Springer science & business
Kumbár, V., Trnka, J., Nedomová, Š., & Buchar, J. (2015). On the influence of storage media.
duration on rheological properties of liquid egg products and response of eggs to Walsh, K. B., Blasco, J., Zude-Sasse, M., & Sun, X. (2020). Visible-NIR ‘point’
impact loading - Japanese quail eggs. Journal of Food Engineering, 166, 86–94. spectroscopy in postharvest fruit and vegetable assessment: The science behind three
https://doi.org/10.1016/j.jfoodeng.2015.05.030 decades of commercial use. Postharvest Biology and Technology, 168, 111246. https://
Lin, H., Zhao, J., Sun, L., Chen, Q., & Zhou, F. (2011). Freshness measurement of eggs doi.org/10.1016/j.postharvbio.2020.111246
using near infrared (NIR) spectroscopy and multivariate data analysis. Innovative Wang, S., Chen, W., Zhang, H. X., Ruan, D., & Lin, Y. C. (2014). Influence of particle size
Food Science & Emerging Technologies, 12(2), 182–186. https://doi.org/10.1016/j. and calcium source on production performance, egg quality, and bone parameters in
ifset.2011.01.008 laying ducks. Poultry Science, 93(10), 2560–2566. https://doi.org/10.3382/ps.2014-
Liu, L., Yang, R., Luo, X., Dong, K., Huang, X., Song, H., Gao, H., Li, S., & Huang, Q. 03962
(2020). Omics analysis of holoproteins and modified proteins of quail egg. Food
Chemistry, 326(April), 126983. https://doi.org/10.1016/j.foodchem.2020.126983

12