Journal of Food Composition and Analysis 113 (2022) 104724

Contents lists available at ScienceDirect

Journal of Food Composition and Analysis

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

Determination of peroxide value in edible oils based on Digital

Image Colorimetry
Amanda Carolina Souza Andrada Anconi a, Nathália Cristina Silva Brito b, Cleiton
Antônio Nunes b, *
a
Department of Chemistry, Federal University of Lavras, University Campus, Post Office Box 3037, 37200-900 Lavras, Minas Gerais, Brazil
b
Department of Food Science, Federal University of Lavras, University Campus, Post Office Box 3037, 37200-900 Lavras, Minas Gerais, Brazil

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

Keywords: Edible oils are naturally oxidized under heat or light, generating compounds responsible for undesirable flavors
Lipids and tastes, which lead to the depreciation of products that contain them. The first compounds formed in oil
Oxidation oxidation are hydroperoxides, which can be estimated by determining the peroxide value (PV, meq O2/Kg) based
Smartphone
on iodometric titration, requiring large amount of hazardous reagents and presenting difficulties detecting the
Hydroperoxides
endpoint. Thus, this work aimed to develop an economical and non-subjective method to predict the PV in edible
Peroxide index
oils based on Digital Image Colorimetry. First, six standards were prepared with hydrogen peroxide solutions.
After reaction with potassium iodide, the mixtures were imaged using a smartphone camera. The color pa­
rameters from the RGB, HSV, CMYK, L*a*b*, or XYZ systems were calibrated against the meq O2 of standards. A
test set of 50 edible oil samples was used for validation based on the standard method. RMSE and R2 values of
3.21 meq O2/kg and 0.97 were obtained using M from CMYK color system as analytical signal. The amount of
reagents consumed in the proposed method was 98.7% reduced compared to that of the standard method.

1. Introduction and do not directly contribute to the product’s sensory characteristics.

However, the oxidative process promotes fatty acid decomposition,
In recent years, global production and consumption of edible oils forming small and volatile molecules that produce unpleasant sensory
have increased. In the last two decades, production has increased by characteristics known as rancidity (Jorge, 2009).
140%, and the trend is for the sector to continue growing in the coming The first stage of lipid oxidation produces hydroperoxides, which are
years (USDA, 2022). Edible oils contain essential components for human unstable compounds that, upon formation, immediately undergo ho­
nutrition, which act as carriers of fat-soluble vitamins (A, D, E, and K) molytic cleavage to produce peroxyl, alkoxide, and hydroxyl radicals.
and have an energetic function (Wsowicz et al., 2004). Furthermore, These species, in turn, generate aldehydes, ketones, and other dangerous
these foods are considered essential supporters of human health due to and sensorially unpleasant molecules as by-products (Wsowicz et al.,
their high content of unsaturated fatty acids and natural antioxidants 2004). Thus, this process compromises the quality and safety of these
(Belitz et al., 2009). foods and is the main factor influencing shelf life (Jiang et al., 2020).
Edible oils primarily comprise triacylglycerols mainly containing The peroxide value (PV) is commonly used to assess initial oxidation.
oleic (C18:1), linoleic (C18:2), linolenic (C18:3), palmitic (C16:0), and The standard method for PV determination is based on a titration that
stearic (C18:0) acids (Fonseca and Gutierrez, 1974). The double bonds measures the iodine released from potassium iodide due to the reaction
of unsaturated fatty acids, such as linoleic and linolenic acids, are sus­ with hydroperoxides present in the oil samples. To a known amount of
ceptible to oxidation reactions mainly accelerated by light and high oil, excess potassium iodide is added, which reacts with the hydroper­
temperatures (Jorge, 2009; Xie et al., 2019). oxides in the sample. The iodine liberated is titrated with sodium thio­
Lipid oxidation can occur in the production and storage stage and is sulfate, using starch as an indicator. The calculated amount of potassium
the leading cause of the deterioration of edible oils. The fatty acids iodide required to react with the hydroperoxides and convert them to
present in these foods have a high molecular weight and low volatility, iodine is used to determine the PV (Nielsen, 2017). Nonetheless, the

* Corresponding author.
E-mail addresses: [email protected] (A.C.S.A. Anconi), [email protected] (N.C.S. Brito), [email protected] (C.A. Nunes).

https://doi.org/10.1016/j.jfca.2022.104724
Received 23 February 2022; Received in revised form 18 May 2022; Accepted 30 June 2022
Available online 2 July 2022
0889-1575/© 2022 Elsevier Inc. All rights reserved.
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

accuracy of this method depends on the precise timing of protection of

the mixture from light and can be subjective due to variable analyst
experience in detecting the endpoint (Mehta et al., 2015). In addition, an
aggravating factor is that oils and fats are soluble only in organic sol­
vents, requiring large amounts of these chemicals, which results in
increased costs as well as hazards to health and the environment.
Recently, new proposals have been discussed to measure oxidation in
edible oils. Infrared spectroscopy associated with chemometric tools has
been very useful (Rohman, 2017; Cebi et al., 2017). However, this
technique requires expensive equipment, and several spectral
pre-treatments are necessary to develop models, making analysis chal­
lenging. Raman spectroscopy has also been used to assess oxidation Fig. 1. Apparatus used for the acquisition of digital images without interfer­
ence from external lighting.
parameters in oil samples (Hu et al., 2019). Low-resolution Raman
spectroscopy was used, combined with chemometric methods, to eval­
uate olive oil sample oxidative stability. The PV was studied as a primary In summary, the available methods have some gaps, such as high
oxidation parameter in addition to secondary parameters. The results consumption of reagents and subjectivity in detecting the titration
showed that the technique could be relevant for evaluating olive oil endpoint, also dependence on expensive equipment when using instru­
oxidation status since the oxidation parameters were quickly deter­ mental determinations. Then, this work aimed to contribute to the
mined in a non-destructive manner (Guzmán et al., 2011). However, this advancement in this area by developing a method based on digital image
technique was indirect, as it required a reference and official method­ data captured with a smartphone camera to determine the PV in edible
ologies to calibrate the instrument. The changes in electrical conduc­ oils. Furthermore, the proposal aims to simplify the standard method­
tivity (EC) values of the aqueous phase during the reaction of ology for PV determination, replacing the titration with a cheap and
hydroperoxides presented in oil samples with potassium iodide (KI) accessible dispositive, making it non-subjective and reducing the anal­
were used to estimate PV. Nevertheless, it required a standard calibra­ ysis time, chemicals, and waste generated in the analytical process.
tion method, and large amounts of organic solvents, leading to envi­
ronmental pollution concerns (Yang et al., 2014). Electronic nose and 2. Materials and methods
discriminant analyses were used to assess the degree of oxidation in
edible oils using the American Oil Chemists’ Society (AOCS) Official 2.1. Sampling
Method as a reference. The authors showed that it was possible to
differentiate oxidized oil from non-oxidized oil samples using this Samples of 14 vegetable edible oils of different brands obtained from
method (Xu et al., 2016). However, the results provided a purely qual­ local retail outlets were analyzed: 7 extra virgin and virgin olive oils
itative analysis and were unsuitable for stringent quality control. In this (Andorinha, La Rambla Olivenza, Fátima, and Quinta da Boa Vista), 5
sense, developing simple, practical analytical methods with low eco­ refined oils, namely, sunflower (Liza), canola (Salada), maize (Mazola
nomic and environmental costs would provide valuable alternatives for and Sinhá) and soybean (Cocamar) oils, 1 avocado oil (EPAMIG), and 1
evaluating edible oil deterioration. olive and soybean oil combination (Maria). Mixtures of these samples
In recent years, analytical methodologies developed to extract were also used to expand the range of peroxide values, resulting in 50
chemical information based on Digital Image Colorimetry (DIC) have different samples (Table S1).
been highlighted. These approaches provide results quickly, accurately,
and inexpensively (Peng et al., 2017; Pessoa et al., 2017). In this sense, 2.2. Determination of the PV by the DIC method
smartphone cameras are an exciting trend since these devices are
portable and have a fast and straightforward operation without The standard methodology for PV determination measures the iodine
requiring expensive equipment (Roda et al., 2016; Nelis et al., 2020). In released from potassium iodide due to the reaction with hydroperoxides,
general, digital images are obtained directly from the sample or after a according to the AOCS method (Nielsen, 2017). The oil sample is solu­
colorimetric chemical reaction. If the value of the color parameters bilized in chloroform and acetic acid and reacted with KI to form iodine,
varies proportionally with an analyte concentration, it can be used as an which is titrated using starch as indicator. The proposed method aimed
analytical response (Rateni et al., 2017). to simplify this methodology by replacing the titration step with DIC.
The decomposition of a digital image into its color parameters gen­ Firstly, analytical curves were built from six calibration standards
erates numerical matrices that describe the distribution of different containing aqueous solutions with different concentrations of 30%
tones in pixels delimited by a region of interest (ROI) (Rezazadeh et al., hydrogen peroxide (Êxodo Científica) (Table S2). To this, in a 1 cm glass
2019). For example, the image is broken down into red, green, and blue cuvette, 0.5 mL of hydrogen peroxide solutions (Table S2) were mixed
parameters in the RGB color system, quantified from 0 to 255. The RGB with 1.2 mL of acetic acid, and 0.1 mL of a freshly prepared saturated
parameters can also be converted into the other color systems. When potassium iodide solution was added and mixed. The mixture was
used as an analytical signal, this information allows the building of placed in a closed apparatus (Fig. 1), where it was kept protected from
regression models in which successive data from several samples are light for 1 min, assuming a brownish color which was imaged.
calibrated against the values of a chemical property obtained by a Secondly, to validate the proposed method, 50 oil samples (Table S1)
reference method (Diniz, 2020). had their PV determined by the titration method (Nielsen, 2017). The
Recent papers involving digital image-based methods for edible oil results were expressed in milliequivalent of active oxygen per kilogram
analysis describe the quantification of adulterants in extra virgin olive of oil (meq O2/kg). Then, these same samples were analyzed by the DIC
oil (Karagiorgos et al., 2017; Hakonen; Beves, 2018; Song et al., 2020), method employed to the standards, replacing the hydrogen peroxide
the determination of free radical scavenging activity (Vucane et al., with 0.2 g of each oil sample, diluted in 1.2 mL of acetic acid and 0.8 mL
2020), the determination of parameters of the bleaching process (Per­ of chloroform.
egonchaya et al., 2020), determination of the iodine index (Peamaroon As chloroform is required for the complete solubilization of the oil
et al., 2021) and monitoring of the oxidation reaction during reaction samples, a biphasic mixture is formed after KI aqueous solution addition.
with Fenton reagent (Singkhonrat et al., 2019), indicating that methods Then, while the calibration standards had 1.8 mL of a monophasic
based on digital images represent a promising technique for monitoring colored solution, the validation samples had only the aqueous phase
the quality of edible oils. (bottom) colored by iodide, with 1.3 mL. To equalize the difference, the

2
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

meq O2 in 1.3 mL of the standard solutions was calculated (according to

Eq. 1) and used in the calibrations. The meq O2 contained in each
standard is available in Table S2.
meq O2 (in 1.8 ml)
meq O2 (in standards) = 1.3 × (1)
1.8
Finally, the PV in the oil samples was calculated by dividing the meq
O2 obtained from the analytical curve by the sample mass (0.0002 kg),
according to Eq. (2).
/
meq O2
PV(meq O2 kg) = (2)
oil mass (kg)

2.3. Acquisition of digital images

An apparatus of dimensions 6 cm × 12 cm × 8 cm (width × length ×

height), developed in our previous work (De Carvalho and Nunes,
2021), was used to block the effect of external light under analysis
(Fig. 1). The inside was made of Ethylene-vinyl acetate (EVA), which, as
it is a matte material, does not affect the reflection of the flash. A cuvette
holder, made of the same material, was positioned inside the box, at the
bottom, and centered. The outside of the box was covered with card­
board. A removable lid was installed to facilitate the exchange of sam­
ples. An elastic band was placed on the outside to fix the smartphone,
and the camera was fitted into a hole of approximately 2 cm in diameter,
aligned with the support for the cuvette containing the samples at
8.5 cm.
Digital images were captured with a Moto C Plus 7.0 smartphone
with an 8 MP camera (f/2.2, 1.12 µm) with a resolution of 3264 × 2448
pixels and a focal distance of 3.50 mm positioned in the hole of the
apparatus so that the samples were illuminated only by the light from
the smartphone flash. Fixed-focus, flash-on, white balance, and single
image mode options were selected to capture the images. This exact
configuration was maintained for all photographs to standardize the
images. The digital images acquired from the colorimetric reaction were Fig. 2. Biphasic solution after reaction of KI with hydroperoxides in edible oil.
The green square is the ROI used to obtain the color parameters.
saved in JPEG format and decomposed into their RGB, HSV, CMYK,
L*a*b*, and XYZ color systems using the free application Color Piker.
The procedures were performed in duplicate, and the arithmetic means by analyzing their leverages and Student residues with a limit of ± 2.5
of color parameters were used as descriptors. to prevent the influence of anomalous samples during prediction.
The results for the oil samples obtained by titration and DIC methods
were compared in terms of meq O2 and PV values. For comparison based
2.4. Statistical analysis
on meq O2, the PV values obtained by the titration method were
multiplied by the mass of the sample (in kg). For comparisons based on
The color parameters (from RGB, HSV, CMYK, L*a*b*, or XYZ) were
PV, the meq O2 obtained by the DIC method was divided by the sample
singly calibrated by linear regression against meq O2 of the standards.
mass (Eq. 2).
The standard deviation of 15 blanks (sdb) was used to calculate the limit
The calculations were carried out using GNU Octave version 5.2.0
of detection (LOD) and limit of quantification (LOQ). The blank
(Eaton et al., 2019), Chemoface version 1.64 (Nunes et al., 2012), and
comprised oil samples solubilized in chloroform and acetic acid before
Microsoft Excel version 2007.
the colorimetric reaction. The analytical curve was described as y = bx
+ a, in which y was the predicted response (meq O2) and x was the
3. Results and discussion
instrumental response (color parameter). Then, LOD and LOQ were
given by:
The standard procedure for PV determination is based on the reac­
sdb tion of potassium iodide with hydroperoxides present in oils, which
LOD = 3.3 × (3)
1/a results in the formation of iodine by iodide oxidation. In calibration
standards, the addition of 0.5 mL of peroxide solution, 1.2 mL of acetic
sdb acid, and 0.1 mL of aqueous KI resulted in 1.8 mL of a unique phase
LOQ = 10 × (4)
1/a solution entirely colored by iodine formed, for which a 20 × 20-pixel
centralized on the solution was used to extract the color parameters to
The quality of the models was evaluated by the root mean square
calibrate the method based on digital images. On the other hand, for
error (RMSE) and the coefficient of determination (R2) for calibration
validation samples, besides acetic acid, 0.8 mL of chloroform was
(RMSEc, R2c) and prediction (RMSEp, R2p) as well as the mean relative
necessary for the complete dilution of the oil, forming a biphasic mixture
error (MRE). In addition, to ensure not only that a good correlation
(Fig. 2) after the addition of KI aqueous solution. The produced iodine,
coefficient in the external validation has been achieved but also that the
responsible for the solution’s color change, remains soluble in the
measured and predicted values are congruent, the statistical parameter
aqueous phase. Thus, the bottom phase was the region of interest (ROI),
R2m was also obtained. R2m values greater than 0.5 are considered
for which a 20 × 20-pixel centralized on the bottom phase was used to
acceptable (Mitra et al., 2010).
extract the color parameters to validate the proposed method.
Outliers were detected and removed from the validation sample set

3
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

Fig. 3. Color scale obtained according to the amount of meq O2 for standards.

The models obtained from the L*, X, and Y parameters had the worst
Table 1
performances, with mean relative errors for the prediction above 44%,
Linearity between RGB, HSV, CMYK, Lab, and XYZ
worse fit to the linear model, compared to the others (R2 of 0.78, 0.68,
color parameters and the meq O2 of the standards.
and 0.75, respectively) and non-existence of congruence between pre­
Color Component R2 dicted and measured values, mainly for X and Y (R2m of 0.29 and 0.30
R 0.85 respectively). Thus, the L* parameter and the XYZ system were
G 0.99 considered unsuitable for predicting the amount of meq O2 in vegetable
B 0.47
oil samples. Despite the RGB color system being the most used in the
H 0.98
S 0.47 specialized literature, the model with the G parameter did not have the
V 0.81 best performance in predicting the meq O2 in the samples. This model
C 0.00 had an R2 of 0.86 and a high RMSE (0.0021 meq O2), representing a
M 0.98 mean relative error for prediction of 31.16%. Even so, this parameter
Y 0.47
K 0.81
was kept in the next step to confirm its low predictive quality compared
L 0.99 to the other chosen models.
a 0.96 The meq O2 values predicted by the models in the above step were
b 0.63 converted to PV (dividing by the mass of oil in kg) and compared to PV
X 0.98
determined by titration to validate the proposed method. The PV of
Y 0.94
Z 0.74 samples ranged from 4.3 to 63.6 meq O2/kg. The results are shown in
Table 3.
Remarkably, there was a relative poor correlation between the
The standards (after reaction with potassium iodide) presented measured PVs and PVs estimated by the model based on the G param­
colors varying from brownish-yellow (for low meq O2) to reddish-brown eter, noticed by the lowest R2 (0.87), higher RMSE (9.93 meq O2/kg),
(for high meq O2) (Fig. 3). The range spanned from 1.816 × 10-4 to and mean relative error of 31.16%. The parameter a* also performed
1.227 × 10-2 meq O2. poorly, with R2 of 0.92 and RMSE of 7.25 meq O2/kg, producing a mean
In color systems, each parameter contains a fraction of information relative error of 22.76%. On the other hand, excellent results were
that could describe the predicted response. Thus, it is crucial to select an produced by the models using H and M parameters. The models per­
appropriate color parameter to obtain a better performance of the pre­ formed similarly, with R2 of 0.96 and 0.97 and R2m of 0.98 for both.
diction models. The analysis of the relationship between the meq O2 in However, the results indicated a better relationship between the
the standards and the color parameters revealed the remarkable line­ measured and predicted PVs to the model obtained for the M parameter,
arity of the data (Table 1). with lower values of RMSE (3.21 meq O2/kg) and RE (10.23%), in
The color parameters with a linear correlation higher than 90% (R2 ≥ addition to the excellent congruence between these data (R2m = 0.98).
0.90) were used to build calibration models. Thus, calibration models Thus, the model built from the M parameter of the CMYK color system
were built using G data from RGB, H data from HSV, M data from CMYK, was considered the best for predicting PV in vegetable oil samples
L* and a* data from L*a*b*, X and Y data from XYZ color systems as among those presented in this study, with LOD and LOQ of 0.00017 meq
descriptors to predict the meq O2 values of edible oil samples. According O2 and 0.00050 meq O2, respectively. The graphical representations of
to Table 2, the analytical curves showed an excellent fit to the linear the analytical curve, meq O2 prediction, and PV prediction for this
model, with R2c value varying from 0.94 to 0.99 and RMSEc from model are shown in Fig. 4.
0.00044 to 0.001 meq O2. Comparable results were obtained by Singkhonrat and coauthors
The analytical curves were employed to predict meq O2 in the oil (2019) in their research on monitoring lipid peroxidation by the Fenton
samples. The results (Table 2) showed a good predictive ability for the
models built from H, M, and a* parameters, with R2p ranging from 0.92 Table 3
to 0.97, RMSEp ranging from 0.00067 to 0.0015 meq O2, and R2m from Performance of predictive models for PV determination in edible oils.
0.87 to 0.92. The models with H and M parameters were similar, with
G H M a
RMSEp of 0.00073 and 0.00067 meq O2 and R2p of 0.96 and 0.97,
respectively. Among these two models, a better performance was ob­ RMSE 9.93 3.40 3.21 7.25
R2 0.87 0.96 0.97 0.92
tained for the M parameter of the CMYK system, with a higher R2p and a
R2 m 0.91 0.98 0.98 0.95
lower RMSEp, which represented a mean relative error of 10%. MRE % 31.16 10.64 10.23 22.76
Furthermore, the high R2m (0.91) in this model attested to the
Note: RMSE is expressed in meq O2/kg
congruence between the measured and predicted values.

Table 2
Performance of predictive models for meq O2 determination using color parameters with a R2 of calibration ≥ 0.90.
G H M L a X Y

RMSEc 0.00047 0.00062 0.00063 0.00044 0.00080 0.00065 0.001

R2c 0.99 0.98 0.98 0.99 0.96 0.98 0.94
RMSEp 0.0021 0.00073 0.00067 0.0037 0.0015 0.0052 0.003
R2p 0.86 0.96 0.97 0.78 0.92 0.68 0.75
R2 m 0.67 0.92 0.91 0.59 0.87 0.29 0.3
MRE % 31.42 10.92 10.03 55.36 22.44 77.81 44.89
N 50 49 48 50 48 50 50

Note: RMSE is expressed in meq O2; N: number of samples used to test the models.

4
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

proposed method was less expensive, and less chemical waste was
produced.
To physically understand the linearity between the color parameters
and the peroxide content (in terms of meq O2), one must understand the
meaning of each color parameter and how it varies in the calibration
standards. In this context, it should be considered that when the white
light falls on the sample, part is absorbed, and part is reflected, the latter
being captured by the smartphone camera.
The RGB system consists of an additive color model, in which the
linear combination of the three primary colors, red (R), green (G), and
blue (B), form the other colors (Firdaus et al., 2014). As seen in Fig. 3,
the color of the solution formed after KI addition ranged from
brownish-yellow to reddish-brown. According to Table S3, such color­
ations were formed by R and G reflection, and their intensities had a
negative variation, decreasing when the peroxide concentration
increased. Although red was the color that most contributed to the color
formation of standards due to higher values of the R parameter, the
variation of G was more relevant, which can justify its higher correlation
with peroxide concentration. On the other hand, the B parameter
remained practically null. Thus, blue was the absorbed color that did not
contribute to the resultant coloring and, therefore, was negligibly
detected by the smartphone camera.
The CMYK system is based on a subtractive linear combination to
form the secondary colors of RGB (Fernandes et al., 2020). For example,
the parameter M (magenta) is formed by red and blue primary colors,
and its complementary color is green. This fact supports the negative
variation of the G parameter and simultaneous positive variation of the
M parameter with increased peroxide content in the standards
(Table S3). Furthermore, as there was practically no variation in B,
corresponding to the blue color, the increase in the M can also be related
to the red reflection. Together, these considerations support the good
description of the reaction medium color by the M parameter.
As well as M, the H parameter (hue angle) also shows an excellent
linear correlation with the reaction medium color. This parameter de­
scribes the dominant spectral color component (such as red, green, or
blue), whose value ranges from 0◦ to 360◦ . Each angle corresponds to
one color: 0◦ is red, 120◦ is green, and 240◦ is blue (Fan et al., 2021). As
shown in Table S3, the H parameter values ranged from 59◦ (yellowish)
to 22◦ (reddish) along with standards, which corroborates the color
variation from brownish-yellow to reddish-brown with the increase of
the peroxide content.
The luminosity (L*) varies from 0 to 100 in the L*a*b* system, with
0 corresponding to black and 100 to white. The L* parameter had its
value decreased, which is justified by the solutions’ darkening (in a
grayscale image) with increasing peroxide concentration. Despite its
excellent linearity with the peroxide content, this parameter varies on
grayscale and alone is not efficient in describing color, which may
explain the low predictive performance of the model based on L*.
Already, the parameter a* varies from negative (green) to positive (red)
values (Leão et al., 2007). Then, the increase of a* indicates an increase
in the red proportion in color, which is consistent with the color change
Fig. 4. A – Example of analytical curve based on M from CMYK. B – Graphical in the reaction medium.
comparison of estimated meq O2 determined in a set of edible oils vs. meq O2 The parameters of the XYZ color system can be obtained from a
predicted by model. C – Graphical comparisons of PV measured in the set of mathematical transformation of RGB, where Y is related to the bright­
edible oils vs. PV estimated by model.
ness of a color (Fernandes et al., 2020). Therefore, it justifies this
parameter decreases along with peroxide content in standards, which
reaction using DIC. Over the PV range of 3–14 meq O2/kg, the authors less lightness (in a grayscale image) can be related to the increase of
found a good correlation between the proposed method and the UV–Vis color intensity. On the other hand, X is a mix of three RGB curves chosen
method (R2 = 0.98). However, this study did not include the reduction to be nonnegative and, in combination with Y, can describe color vari­
of reagents used and waste generated in the analysis. Peamaroon and ations from yellowish to reddish (Ibraheem et al., 2012). However, Y or
coauthors (2021) developed a DIC method based on a smartphone X alone could not produce models with good predictive capacity,
camera to determine iodine value in vegetable oils, using the Wijs probably due to interdependence in describing the color variation of the
iodometric titration method as reference. The authors also replaced the reaction medium. Finally, the same reasoning can be extended to un­
titration step with DIC, reducing the amount of chemicals. An analytical derstand the prediction performance of each model built from the color
curve was set up to predict iodine value in vegetable oils, and the results parameters of the studied colorimetric systems.
showed an RSD value between 3.2% and 8.6% and an R2 of 0.99. The The standard method for PV determination (Nielsen, 2017) employs

5
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

Table 4 Finance code 001.

Reagents employed in the standard method and in the DIC method for PV
determination. Appendix A. Supporting information
Reagent Standard method (mL) DIC method (mL)

Acetic acid 18 1.2

Supplementary data associated with this article can be found in the
Chloroform 12 0.8 online version at doi:10.1016/j.jfca.2022.104724.
KI solution 0.5 0.1
Water 30 0 References
1% starch solution 0.5 0
Sodium thiosulfate solution variable 0 Belitz, H.-D., Grosch, W., Schieberle, P., 2009. Edible fats and oils. Food Chemistry.
Total (except sodium thiosulfate) 61 2.1 Springer, Berlin Heidelberg, pp. 602–630.
Cebi, N., Yilmaz, M.T., Sagdic, O., Yuce, H., Yelboga, E., 2017. Prediction of peroxide
value in omega-3 rich microalgae oil by ATR-FTIR spectroscopy combined with
a mixture of 12 mL of chloroform and 18 mL of acetic acid, in addition to chemometrics. Food Chem. 225, 188–196.
0.5 mL of saturated KI solution, 30 mL of water, 0.5 mL of starch solu­ De Carvalho, T.C.L., Nunes, C.A., 2021. Smartphone-based method for the determination
of chlorophyll and carotenoid contents in olive and avocado oils: An approach with
tion, and variable volumes of sodium thiosulfate solution. On the other calibration transfer. J. Food Compos. Anal. 104.
hand, the proposed method based on digital images had these volumes Diniz, P.H.G.D., 2020. Chemometrics-assisted color histogram-based analytical systems.
considerably reduced or set to zero (Table 4), which represented a J. Chemom. 34 (12), 1–27.
Eaton, J.W., Bateman, D., Hauberg, S., Wehbring, R., 2019. GNU Octave Version 5.2.0
reduction of approximately 98.7% considering all reagents, except so­ Manual: a High-level Interactive Language for Numerical Computations. Samurai
dium thiosulfate solution. Media Limited.
Given the above results, the proposed method reliably predicts the Fan, Y., Li, J., Guo, Y., Xie, L., Zhang, G., 2021. Digital image colorimetry on smartphone
for chemical analysis: a review. Measurement 171, 108829.
primary oxidation in edible vegetable oil samples. Furthermore, using Fernandes, G.M., Silva, W.R., Barreto, D.N., Lamarca, R.S., Gomes, P.C.F.L., Petruci, J.F.
digital images as a replacement for titration enables the economical S., Batista, A.D., 2020. Novel approaches for colorimetric measurements in
savings of reagents and waste. Finally, the results were not affected by analytical chemistry - a review. Anal. Chim. Acta 1135, 187–203.
Firdaus, M.L., Alwi, W., Trinoveldi, F., Rahayu, I., Rahmidar, L., Warsito, K., 2014.
the subjectivity introduced when judging the endpoint of the titration, Determination of chromium and iron using digital image-based colorimetry.
and this proposed method is simpler and less expensive than the stan­ Procedia Environ. Sci. 20, 298–304.
dard method, making it affordable for both laboratories and industry. Fonseca, Н, Gutierrez, L.E., 1974. Composição em ácidos graxos de óleos vegetais e
gorduras animais. An. Esc. Super. Agric. Luiz Queiroz 31, 485–490.
Guzmán, E., Baeten, V., Fernández Pierna, J.A., García-Mesa, J.A., 2011. Application of
4. Conclusion low-resolution Raman spectroscopy for the analysis of oxidized olive oil. Food
Control 22 (12), 2036–2040.
Hakonen, A., Beves, J.E., 2018. Hue parameter fluorescence identification of edible oils
The results presented in this work demonstrated that the PV in edible
with a smartphone. ACS Sens. 3 (10), 2061–2065.
oils could be easily and rapidly determined by a method based on digital Hu, R., He, T., Zhang, Z., Yang, Y., Liu, M., 2019. Safety analysis of edible oil products via
images obtained with a smartphone. The amount of meq O2 was cali­ raman spectroscopy. Talanta 1 (191), 324–332.
brated against color parameters, and the best analytical signal was found Ibraheem, N.A., Hasan, M.M., Khan, R.Z., Mishra, P.K., 2012. Understanding color
models: a review. ARPN J. Sci. Technol. 2 (3), 265–275.
to be M values from CMYK image data. Validation using a set of edible Jiang, S., Xie, Y., Li, M., Guo, Y., Cheng, Y., Qian, H., Yao, W., 2020. Evaluation on the
oils confirmed the good predictive performance of the model for which oxidative stability of edible oil by electron spin resonance spectroscopy. Food Chem.
RMSE and R2 of 3.21 meq O2/kg and 0.97 were obtained. Furthermore, 309, 125714.
Jorge, N. (2009). Química e Tecnologia de Óleos Vegetais. São Paulo: Ed. Cultura
the proposed method reduced the reagent and chemical waste con­ Acadêmica.
sumption by 98.7%, including organic solvents, making the process less Karagiorgos, N., Nenadis, N., Trypidis, D., Siozios, K., Siskos, S., Nikolaidis, S., Tsimidou,
environmentally damaging. It is noteworthy that the same analytical M.Z. (2017). An approach for estimating adulteration of virgin olive oil with soybean
oil using image analysis. In: 6th International Conference on Modern Circuits And
setup and smartphone must be used to analyze standards and samples, as Systems Technologies (p. 1–4). Thessaloniki.
usual for instrumental methods. Finally, the simplicity and low cost of Leão, A.C., Souza, L.A.C., Araújo, A.A., 2007. Gerenciamento de cores – ferramenta
the procedures allow this approach to be used as an alternative to fundamental para a documentação de bens culturais. Rev. Bras. Arqueometria
Restauração Conserv. 1 (4), 215–220.
determining the PV in laboratories and industries, with a minor equip­
Mehta, B.M., Darji, V.B., Aparnathi, K.D., 2015. Comparison of five analytical methods
ment setup required. Moreover, the proposed method eliminates analyst for the determination of peroxide value in oxidized ghee. Food Chem. 185, 449–453.
subjectivity by replacing titration with digital image data analysis. Mitra, I., Saha, A., Roy, K., 2010. Exploring quantitative structure-activity relationship
studies of antioxidant phenolic compounds obtained from traditional Chinese
medicinal plants. Mol. Simul. 36, 1067–1079.
CRediT authorship contribution statement Nelis, J.L.D., Tsagkaris, A.S., Dillon, M.J., Hajslova, J., Elliott, C.T., 2020. Smartphone-
based optical assays in the food safety field. Trends Anal. Chem. 129, 115934.
Nielsen, S.S., 2017. Food Analysis Laboratory Manual. Springer International Publishing,
Amanda Carolina Souza Andrada Anconi: Data curation; Formal
Indiana, USA.
analysis, Methodology, Roles/Writing - original draft; Writing - review Nunes, C.A., Freitas, M.P., Pinheiro, A.C.M., Bastos, S.C., 2012. Chemoface: a novel free
& editing. Nathália Cristina Silva Brito: Methodology, Formal anal­ user-friendly interface for chemometrics. J. Braz. Chem. Soc. 23 (11), 2003–2010.
ysis, Investigation. Cleiton Antônio Nunes: Conceptualization, Formal Peamaroon, N., Jakmunee, J., Moonrungsee, N.A., 2021. Simple colorimetric procedure
for the determination of iodine value of vegetable oils using a smartphone camera.
analysis, Project administration, Supervision. J. Anal. Test. 5, 1.
Peng, B., Chen, G., Li, K., Zhou, M., Shengguo, J., Zhao, Z., 2017. Dispersive liquid-liquid
microextraction coupled with digital image colorimetric analysis for detection of
Declarations of Conflicting Interests
total iron in water and food samples. Food Chem. 230, 667–672.
Peregonchaya, O.V., Korol’kova, N.V., Sokolova, O.V., D’yakonova, S.A. (2020).
The authors declare no conflicts of interest. Possibility of using digital technology in determining the color number of vegetable
oil. In: IOP Conference Series: Earth and Environmental Science (Vol. 422, p.
1755–1315). Tomsk, Russian Federation.
Acknowledgments Pessoa, K.D., Suarez, W.T., Dos Reis, M.F., Franco, M.O.K., Moreira, R.P.L., Dos
Santos, W.B., 2017. A digital image method of spot tests for determination of copper
Authors are thankful to CAPES (Coordenação de Aperfeiçoamento de in sugar cane spirits. Spectrochim. Acta - Part A Mol. Biomol. Spectrosc. 185,
310–316.
Pessoal de Nível Superior – Brasil), CNPq (Conselho Nacional de Rateni, G., Dario, P., Cavallo, F., 2017. Smartphone-based food diagnostic technologies: a
Desenvolvimento Científico e Tecnologico), and FAPEMIG (Fundação de review. Sensors 17 (6).
Amparo à Pesquisa de Minas Gerais) for the financial support of this Rezazadeh, M., Seidi, S., Lid, M., Pedersen-Bjergaard, S., Yamini, Y., 2019. The modern
role of smartphones in analytical chemistry. Trends Anal. Chem. 118, 548–555.
research. This study was financially supported by the Coordination for
the Improvement of Higher Education Personnel - Brazil (CAPES) -

6
A.C.S.A. Anconi et al. Journal of Food Composition and Analysis 113 (2022) 104724

Roda, A., Michelini, E., Zangheri, M., Di Fusco, M., Calabria, D., Simoni, P., 2016. Vucane, S., Sabovics, M., Leitans, L., Cinkmanis, I. (2020). Smartphone-based
Smartphone-based biosensors: a critical review and perspectives. Trends Anal. colorimetric determination of DPPH free radical scavenging activity in vegetable oils
Chem. 79, 317–325. In: 26th Annual International Scientific Conference "Research For Rural
Rohman, A., 2017. The use of infrared spectroscopy in combination with chemometrics Development, (Vol. 35 pp. 106–111) Jelgava, Latvia.
for quality control and authentication of edible fats and oils: a review. Appl. Wsowicz, E., Gramza, A., Hes, M., Jeleñ, H.H., Korczak, J., Matecka, M., Mildner-
Spectrosc. Rev. 52 (7), 1–44. Szkudlarz, S., Rudziñska, M., Samotyja, U., Zawirska-Wojtasiak, R., 2004. Oxidation
Singkhonrat, J., Sriprai, A., Hirunwatthanakasem, S., Angkuratipakorn, T., of lipids in food. Pol. J. Food Nutr. Sci. 13 (54), 87–100.
Preechaburana, P., 2019. Digital image colorimetric analysis for evaluating lipid Xie, Y., Jiang, S., Li, M., Guo, Y., Cheng, Y., Qian, H., Yao, W., 2019. Evaluation on the
oxidation in oils and its emulsion. Food Chem. 286, 703–709. formation of lipid-free radicals in the oxidation process of peanut oil. LWT - Food Sci.
Song, W., Song, Z., Vincent, J., Wang, H., Wang, Z., 2020. Quantification of extra virgin Technol. 104, 24–29.
olive oil adulteration using smartphone videos. Talanta 216, 120920. Xu, L., Yu, X., Liu, L., Zhang, R., 2016. A novel method for qualitative analysis of edible
U.S. Department of Agriculture, Oilseeds: World Markets and Trade (2022). USDA World oil oxidation using an electronic nose. Food Chem. 202, 229–235.
Production, Markets, and Trade Report, Retrieved February 9, 2022. Home Page: Yang, Y., Li, Q., Yu, X., Chen, X., Wang, Y., 2014. A novel method for determining
〈https://www.fas.usda.gov/data/oilseeds-world-markets-and-trade/〉. peroxide value of edible oils using electrical conductivity. Food Control 39,
198–203.