ArticlePDF Available

Abstract and Figures

Aroma and other physicochemical parameters are important attributes influencing consumer perception and acceptance of rice. However, current methods using multiple instruments and laboratory analysis make these assessments costly and time-consuming. Therefore, this study aimed to assess rice quality traits of 17 commercial rice using a low-cost electronic nose and portable near-infrared spectrometer coupled with machine learning (ML). Specifically, artificial neural networks (ANN) were used to classify the type of rice and predict rice quality traits (aromas, color, texture and pH of cooked rice) as targets. The ML models developed showed that the chemometrics obtained from both sensor technologies successfully classified the rice (Model 1: 98.7%; Model 2: 98.6%) and predicted the peak area of aromas obtained by gas chromatography-mass spectroscopy found in raw (Model 3: R = 0.95; Model 6: R = 0.95) and cooked rice (Model 4: R = 0.98; Model 7: R = 0.96). Furthermore, a high R=0.98 was obtained for Model 5 to estimate the color, texture, and pH of cooked rice. The proposed method is rapid, low-cost, reliable and may help the rice industry increase high-quality rice production and accelerate the adoption of digital technologies and artificial intelligence to support the rice value chain.
Content may be subject to copyright.


Citation: Aznan, A.; Gonzalez Viejo,
C.; Pang, A.; Fuentes, S. Rapid
Assessment of Rice Quality Traits
Using Low-Cost Digital Technologies.
Foods 2022,11, 1181. https://
doi.org/10.3390/foods11091181
Academic Editor: Ricardo Páscoa
Received: 31 March 2022
Accepted: 18 April 2022
Published: 19 April 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
foods
Article
Rapid Assessment of Rice Quality Traits Using Low-Cost
Digital Technologies
Aimi Aznan 1,2, Claudia Gonzalez Viejo 1, Alexis Pang 1and Sigfredo Fuentes 1,*
1Digital Agriculture, Food and Wine Group (DAFW), School of Agriculture and Food, Faculty of Veterinary
and Agricultural Sciences, University of Melbourne, Parkville, VIC 3010, Australia;
aaznan@student.unimelb.edu.au (A.A.); cgonzalez2@unimelb.edu.au (C.G.V.);
alexis.pang@unimelb.edu.au (A.P.)
2Faculty of Chemical Engineering Technology, University Malaysia Perlis, Arau 02600, Perlis, Malaysia
*Correspondence: sfuentes@unimelb.edu.au; Tel.: +61-42-450-4434
Abstract:
Aroma and other physicochemical parameters are important attributes influencing con-
sumer perception and acceptance of rice. However, current methods using multiple instruments
and laboratory analysis make these assessments costly and time-consuming. Therefore, this study
aimed to assess rice quality traits of 17 commercial rice types using a low-cost electronic nose and
portable near-infrared spectrometer coupled with machine learning (ML). Specifically, artificial neural
networks (ANN) were used to classify the type of rice and predict rice quality traits (aromas, color,
texture, and pH of cooked rice) as targets. The ML models developed showed that the chemometrics
obtained from both sensor technologies successfully classified the rice (Model 1: 98.7%; Model 2:
98.6%) and predicted the peak area of aromas obtained by gas chromatography-mass spectroscopy
found in raw (Model 3: R = 0.95; Model 6: R = 0.95) and cooked rice (Model 4: R = 0.98; Model 7:
R = 0.96
). Furthermore, a high R = 0.98 was obtained for Model 5 to estimate the color, texture, and
pH of cooked rice. The proposed method is rapid, low-cost, reliable, and may help the rice industry
increase high-quality rice production and accelerate the adoption of digital technologies and artificial
intelligence to support the rice value chain.
Keywords: rice aroma; near-infrared; electronic nose; artificial neural networks; machine learning
1. Introduction
Rice is one of the primary energy sources with a high nutritional value from fiber, min-
erals, proteins, vitamins, and antioxidants. It has been consumed as a staple food for more
than half of the world’s population. According to Shahbandeh [
1
], Asian countries are the
major rice consumers, China being the highest (154.9 million metric tons; MMT), followed
by India (103.5 MMT), Vietnam (73.3 MMT), Bangladesh (36.7 MMT), and Indonesia (35.6
MMT). In recent years, it was observed that the growth in socioeconomic status in Asia and
developing countries caused an increasing demand for rice associated with high-quality
traits defined by the consumers [
2
4
]. Therefore, improving rice quality traits is vital to
ensure high acceptability among consumers.
There are diverse types of rice available in the market, mainly produced to suit
consumer preferences and intended cooking dishes. The diversity in rice quality may be
related to the rice cultivar, cultivation practices, postharvest and milling process, storage
condition, and cooking method. Aroma and texture are two important quality traits, as they
may influence the market value and become the driving factors in consumer preferences
for rice [
5
,
6
]. Maleki et al. [
6
] showed that the liking factor among regular rice consumers
for cooked rice using different rice-to-water ratios was divided by their preference for fluffy
or sticky texture. Besides the cooking method, amylose content in rice may also influence
the rice texture, shown by the significant positive correlations between apparent amylose
content and rapid visco-analyzer (RVA) pasting viscosity properties [
7
]. A detailed study
Foods 2022,11, 1181. https://doi.org/10.3390/foods11091181 https://www.mdpi.com/journal/foods
Foods 2022,11, 1181 2 of 16
conducted by Li et al. [
8
] showed that the hardness of cooked rice increased in rice cultivars
that had intermediate to high amylose content. The authors suggested that it was due to
the small size of amylose molecules and higher proportions of amylose branches found in
the cultivars.
The aromatic rice cultivars such as Basmati and Jasmine rice have a higher market
value and preference among rice consumers than nonaromatic rice [
9
,
10
]. More than
300 volatile compounds have been found in aromatic rice cultivars [
11
]. Among them,
2-acetyl-1-pyrroline (2AP) is well-known as one of the key volatile aromatic compounds
associated with popcorn or nutty aroma [
9
,
12
]. Besides 2-AP, Setyaningsih et al. [
13
]
reported pentanal, hexanal, 2-pentyl-furan, 2,4-nonadienal, pyridine, 1-octen-3-ol, and (E)-
2-octenal as the key markers for volatile aromatic compounds that are able to discriminate
between aromatic and nonaromatic rice. Sansenya et al. [
14
] found that although major
volatile aromatic compounds were detected in both aromatic and nonaromatic rice, higher
concentration levels were observed in the aromatic rice. Likewise, the volatiles in raw
and cooked rice with different degrees of milling were significantly different [
15
]. The
latter study showed that the abundance of volatile aromatic compounds found at various
concentrations forms the aroma fingerprinting, which has great potential for developing a
rapid analysis method to assess rice quality traits.
Different techniques have been used to assess rice quality traits. For example, for
the aroma profile of rice, the gas chromatography-mass spectrometry (GC-MS) technique
and descriptive sensory analysis have been used [
12
,
16
18
]. The rice texture profile is
usually analyzed using the RVA and texture analyzer [
7
,
19
]. Additionally, most quality
traits related to the physicochemical properties of rice are measured separately using
different instruments and analysis techniques, as shown by Xia et al. [
20
]. However, these
approaches are costly, time-consuming, and tedious. Therefore, the development of a
simplified method that is robust, portable, rapid, low-cost, and appropriate to analyze
different types of rice simultaneously could be very important for potential application in
the rice industry.
Electric noses (e-noses) are devices consisting of sensor arrays that are sensitive to
volatile compounds, which can mimic the human olfactory system [
21
]. The sensors
are exposed to detect the aroma profile, yielding specific volatile organic compounds
(VOC) [
22
]. Several studies have developed or utilized commercial e-nose sensors in
food and beverages, showing their potential as a promising tool for quality assessment
to distinguish VOC and aroma profiles. For example, the application of e-noses has been
recently reported, such as in rice [
23
,
24
], coffee [
25
], beer and wine [
26
,
27
], broccoli [
28
],
apples [
29
], beef and pork [
30
,
31
], rapeseed [
32
,
33
], wheat bread [
34
], cucumbers [
35
], and
basil [
36
]. On the other hand, near-infrared (NIR) spectroscopy is a rapid screening tool
used to evaluate chemical and physical substances based on the interaction between the
organic material and the lights ranging, for example, between 750 and 2500 nm in the
electromagnetic region [
37
,
38
]. The technique offers feature information on the C-H, N-H,
and O-H vibration bonds that determine the NIR spectra fingerprinting of the organic
substances and VOC [
39
], for which the absorbance values are further correlated with
the concentration of the sample [
40
]. This technique has been used in different studies of
the food, beverage, and grain industry, including rice [
41
], tea [
42
], sugar [
43
], beer [
27
],
wine [
26
], coffee [
25
], and cassava [
44
]. However, a limited number of studies are available
on the application of e-nose and NIR spectroscopy techniques focused on developing rapid
methods using low-cost and portable sensing tools, especially for the rice industry.
Machine learning is a reliable and promising modeling tool that has been used for
the prediction and automation of tasks in many industries, including medical [
45
], oil and
gas [
46
], manufacturing [
47
], automotive [
48
], agriculture [
49
], and food [
50
]. Some specific
applications of machine learning in the agriculture and food industries are on grapes [
51
],
wine [
52
,
53
], beer [
54
,
55
], honey [
56
], and coffee [
25
], among others. According to previ-
ous studies, the application of sensors technology combined with machine learning has
demonstrated that this technique efficiently produces high accuracy and robust prediction
Foods 2022,11, 1181 3 of 16
models [
26
,
57
59
]. Moreover, compared to other supervised machine learning algorithms,
ANNs have been commonly used in agriculture and food studies due to their ability to
predict non-linear relationships between inputs and targets [49,60].
This study aimed to evaluate the use of a handheld NIR spectrometer and a low-cost e-
nose device designed and developed by the Digital Agriculture, Food and Wine group from
The University of Melbourne (DAFW-UoM) [
61
] paired with machine learning modeling
to classify 17 types of commercial rice (Models 1 and 2) and predict the peak area of rice
aromas (Models 3, 4, 6 and 7) and other physicochemical parameters (Model 5). The study
provided new insight into measuring the VOC from cooked and uncooked rice samples
obtained using a low-cost e-nose for simultaneous qualitative and quantitative analysis of
rice quality traits. This approach may benefit the rice industry to perform on-site quality
screening, monitoring, inspection, and authentication using the reliable, non-destructive,
low-cost, and rapid method.
2. Materials and Methods
2.1. Samples Preparation
In this study, 17 types of commercial rice consisting of polished and unpolished rice
were purchased from local markets in Australia (Table 1). The rice was stored in its original
packaging at room temperature until used for experiments. Before cooking, the rice samples
were washed and strained with tap water three times. The amount of water was added
to the rice according to the recommended ratio provided on the packaging with slight
modification after pre-cook testing (Table 1). The rice was cooked using the steam cooking
mode in a combi steamer oven (Convotherm 4 easyDial 6.10 GS, Welbilt Deutschland
GmbH, Eglfing, Germany). After cooking, six cooked rice kernels were drawn from the
samples to test the cooked rice. Two transparent plastic plates were used to press the rice
kernels between the plates, and the rice was considered fully cooked if no white core was
observed. This step is important to ensure the rice samples were thoroughly cooked before
fluffing the rice with a fork and cooling it to room temperature.
Table 1.
Details of commercial rice samples used in the study, including class ID, product category,
type, brandabbreviation, rice-to-water ratio, and cooking time. Abbreviations: v: volume.
Class ID Product
Category Type Brands Abbreviation Rice-to-Water
Ratio (v/v)
Cooking Time
(min)
1
White rice
(Polished rice)
Khoshihikari aSunRice KHO 1:1 1/2 25
2 Sushi rice aSunRice SRS 1:1 1/2 25
3 Bomba aLa Perla BMB 1:2 25
4 Calasparra a
Cooperativa del
Campo CLP 1:2 25
5Arborio bWoolworths ARB 1:2 25
6Calrose bSunRice CLS 1:1 1/2 25
7 Long grain cWoolworths LGW 1:1 1/2 25
8 Jasmine cSunRice JSR 1:1 1/2 25
9 Jasmine-organic cMacro JOM 1:1 1/2 25
10 Basmati cRiviana BSR 1:1 1/3 25
11 Basmati cSunRice BSS 1:1 1/3 25
Foods 2022,11, 1181 4 of 16
Table 1. Cont.
Class ID Product
Category Type Brands Abbreviation Rice-to-Water
Ratio (v/v)
Cooking Time
(min)
12
Whole grain rice
(Unpolished rice)
Biodynamic rice
bHonest to
Goodness BDM 1:1 2/3 55
13 Medium grain bSunRice MGB 1:1 2/3 55
14 Medium
grain—organic bMacro MOB 1:1 2/3 55
15 Doongara cSunRice DGR 1:1 2/3 55
16 Black rice cSunRice BKR 1:1 1/2 60
17 Wild
rice—organic c
Honest to
Goodness WRO 1:1 1/3 60
ashort grain; bmedium grain; clong grain.
2.2. Near-Infrared Spectroscopy and Electronic Nose Measurements
Triplicate samples of raw rice were used to obtain the NIR absorbance value
(1596–2396 nm)
at every 7 to 9 nm interval using the microPHAZIR
RX Analyzer (Thermo Fisher Scientific,
Waltham, MA, USA). Calibration was conducted before the first scanning and repeated
after about 10–15 scans using a standard white reference provided by the manufacturer.
The measurement was obtained using a sample holder attached to the top of the scanning
area to avoid environmental noise, and the samples were poured into the holder. The NIR
measurements were obtained three times per replicate for each sample. The NIR absorbance
values were transformed using the first derivative of the Savitsky–Golay method with
polynomial order two in The Unscrambler X version 10.3 (CAMO Software, Oslo, Norway).
A total of 60 g of raw rice samples was transferred into a 500 mL glass beaker to
obtain the e-nose sensor readings (Supplementary Materials Figure S1) in triplicates. The
e-nose sensor used in the study is portable and was developed by the DAFW-UoM using
low-cost materials [
61
]. It consists of nine sensors (Henan Hanwei Electronics Co., Ltd.,
China) sensitive to different gases (Table 2). This e-nose has holes between each gas sensor
to avoid any oversaturation from volatiles and avoid moisture accumulation. Calibration
procedures were performed between each measurement for about 20 s to let the sensors
reach the baseline reading before taking the new measurement. The sensors were exposed
to the rice samples for one minute to obtain a stable sensor reading from the peak values.
The outputs are given in volts and were analyzed using a supervised code in Matlab 2021a
(Mathworks Inc., Natick, MA, USA) to detect the stable signals from the sensor readings,
followed by ten equidistant subdivisions to obtain ten mean value readings per sensor [
25
].
The same procedure was repeated for cooked rice samples to obtain the e-nose sensor
readings for the 17 rice samples.
Table 2. Gas sensors integrated into the electronic nose and gases they are most sensitive.
Sensor Name Gases
MQ3 Alcohol
MQ4 Methene
MQ7 Carbon monoxide
MQ8 Hydrogen
MQ135 Ammonia, alcohol, benzene
MQ136 Hydrogen sulfide
MQ137 Ammonia
MQ138 Benzene, alcohol, ammonia
MG811 Carbon dioxide
Foods 2022,11, 1181 5 of 16
2.3. Gas Chromatography-Mass Selective Detector Analysis
A gas-chromatography with mass-selective detector 5977B (GC-MSD; Agilent Tech-
nologies, Inc., Santa Clara, CA, USA) equipped with an autosampler system PAL3 (CTC
Analytics AG, Zwingen, Switzerland) was used to analyze the volatile aromatic compounds
of the rice in duplicate samples. Helium was used as the gas carrier at a 1 mL min
1
flow
rate. The raw (4 g) and cooked (3 g) rice samples were weighed and put into 20 mL
screw-top vials with magnetic caps. The headspace method was used for extraction, and
the samples were heated with agitation at 80
C for 30 min to allow the rice to release
the volatile aromatic compounds in the headspace of the vials, followed by the extraction
using divinylbenzene–carboxen–polydimethylsiloxane (DVB–CAR–PDMS) 1.1 mm grey
solid-phase microextraction (SPME) fiber [
62
,
63
] to penetrate the headspace of the vial to
absorb the volatiles for 60 min under 80 C with agitation.
An HP-5MS (Agilent Technologies, Inc.; 30 m
×
0.25 mm
×
0.25
µ
m) column was
used in the analysis; this column was selected considering other studies of rice and rice
bran [
64
66
]. The SPME fiber was desorbed in the injector at 250
C for 5 min using the
splitless mode. The inlet temperature was set at 250
C. The GC oven temperature was
programmed as follows: initial temperature of 40
C for 3 min; then increased to 100
C
at 5
C min
1
and held for 3 min; and ramping to 250
C at 10
C min
1
and held for
4 min
. A blank sample was placed at the start and between raw and cooked rice samples
to prevent the carryover effect. The volatile aromatic compounds were matched with the
National Institute of Standards and Technology (NIST; National Institute of Standards and
Technology, Gaithersburg, MD, USA) library. Only identified compounds greater than
80% certainty are reported in this study. Additionally, compounds with a low relative
abundance identified in less than three rice samples were omitted.
2.4. Colour, pH, and Texture Measurement
The physicochemical quality of the cooked rice was measured to obtain the color,
texture, and pH of samples in triplicates. The Nix Pro color sensor (illuminant: D65,
observer angle: 10
; Nix Sensor Ltd., Hamilton, ON, Canada) was used to determine the L
(lightness), a(red/green), and b(blue/yellow) color components of the cooked rice samples.
Additionally, a pH meter, CT-6021A (Shenzhen Ke Dida Electronics Co., Ltd., Shenzhen,
China), was used to determine the pH of the cooked rice samples. An amount of water
was added to the sample (4 g) and mashed to form a slurry solution before measuring the
pH. Ten grams of cooked rice samples were used to determine the rice texture (T) using a
digital penetrometer (GY-4, China) equipped with an 8 mm diameter cylindrical probe to
obtain the hardness of the rice.
2.5. Statistical Analysis and Machine Learning Modelling
Two matrices were developed using only the significant correlations (p< 0.05) between
(i) the peak area of volatile aromatic compounds using GC-MS and e-nose sensor outputs
for raw rice and (ii) the volatile aromatic compounds using GC-MS and e-nose, color,
texture, and pH of cooked rice.
Seven machine learning models were developed by testing 17 algorithms of artificial
neural network (ANN) pattern recognition in Matlab 2021a using a code developed by
the DAFW-UoM [
67
] to find the algorithm with the best accuracy and performance with
no signs of under- and overfitting, followed by a neuron trimming test (ten, seven, five,
and three neurons). Pattern recognition models of ANN were developed to classify the
17 commercial rice types according to their specific types, as described in Table 1. Figure 1
shows the diagram of the models used in the study to classify the rice using as inputs the
first derivative of raw rice NIR absorbance values (Model 1) and the e-nose outputs of raw
rice (Model 2). The Bayesian regularization (BR) algorithm was selected for these models.
It had the highest accuracy with the lowest mean squared error (MSE) using 70% training
and 30% testing with random data division.
Foods 2022,11, 1181 6 of 16
Foods 2022, 11, x FOR PEER REVIEW 6 of 17
Figure 1. Classification models developed using (a) the first derivative of NIR absorbance values as
inputs (Model 1) and (b) electronic nose sensors data as inputs (Model 2) to classify the 17 types of
commercial rice. A description of the electronic nose sensors can be found in Table 2.
On the other hand, ANN regression models were used to predict the aromas and
other physicochemical parameters (color, texture, and pH) of the rice, as shown in Figure
2. Models 3 and 4 were developed using the first derivative of NIR spectra obtained from
the raw rice as inputs to predict the abundance of volatile aromatic compounds in raw
and cooked rice samples, whereas Model 5 was constructed using the raw rice NIR inputs
to predict the color, texture, and pH of the cooked rice. Based on the optimization, the BR
algorithm was used to develop Models 3 and 4 using random data division with 70% for
training and 30% for testing, whereas Model 5 was developed using a conjugate gradient
with the PowellBeale restarts (CGB) algorithm with a random data division of 60% train-
ing, 20% validation, and 20% testing. The e-nose readings of raw rice were used as inputs
to predict the peak area of volatile aromatic compounds in raw (Model 6) and cooked
(Model 7) rice samples. The BR (random data division: 70% training and 30% testing) and
LevenbergMarquardt (LM; random data division: 70% training, 15% validation, and 15%
testing) algorithms were used to develop Models 6 and 7, respectively.
The time required to train these ANN models depends on the characteristics of the
specific computer used and the number of cores. The development of models from pa-
rameter engineering and testing of different machine learning algorithms, once all of the
data are collected, can be around 24 h, depending on the number of models and the
amount of data used. However, this process is not relevant for the deployment of the
models, which can be in near-real-time. The process of the practical application of the
proposed methods consists of (i) obtaining the sample, (ii) measuring with the e-nose or
NIR device, and (iii) automated analysis with the machine learning model deployment.
The time for this process is less than a second per sample.
Figure 1. Classification models developed using (a) the first derivative of NIR absorbance values as
inputs (Model 1) and (
b
) electronic nose sensors’ data as inputs (Model 2) to classify the 17 types of
commercial rice. A description of the electronic nose sensors can be found in Table 2.
On the other hand, ANN regression models were used to predict the aromas and
other physicochemical parameters (color, texture, and pH) of the rice, as shown in Figure 2.
Models 3 and 4 were developed using the first derivative of NIR spectra obtained from
the raw rice as inputs to predict the abundance of volatile aromatic compounds in raw
and cooked rice samples, whereas Model 5 was constructed using the raw rice NIR inputs
to predict the color, texture, and pH of the cooked rice. Based on the optimization, the
BR algorithm was used to develop Models 3 and 4 using random data division with
70% for training and 30% for testing, whereas Model 5 was developed using a conjugate
gradient with the Powell–Beale restarts (CGB) algorithm with a random data division
of 60% training, 20% validation, and 20% testing. The e-nose readings of raw rice were
used as inputs to predict the peak area of volatile aromatic compounds in raw (Model
6) and cooked (Model 7) rice samples. The BR (random data division: 70% training and
30% testing) and Levenberg–Marquardt (LM; random data division: 70% training, 15%
validation, and 15% testing) algorithms were used to develop Models 6 and 7, respectively.
The time required to train these ANN models depends on the characteristics of the
specific computer used and the number of cores. The development of models from parame-
ter engineering and testing of different machine learning algorithms, once all of the data
are collected, can be around 2–4 h, depending on the number of models and the amount of
data used. However, this process is not relevant for the deployment of the models, which
can be in near-real-time. The process of the practical application of the proposed methods
consists of (i) obtaining the sample, (ii) measuring with the e-nose or NIR device, and (iii)
automated analysis with the machine learning model deployment. The time for this process
is less than a second per sample.
Foods 2022,11, 1181 7 of 16
Foods 2022, 11, x FOR PEER REVIEW 7 of 17
Figure 2. Regression models developed using the (a) first derivative of NIR absorbance values to
estimate the aromas (Models 3 and 4) and (b) physicochemical quality (Model 5) of rice, and (c)
electronic nose sensors data as inputs to estimate rice aromas (Models 6 and 7). A description of the
electronic nose sensors can be found in Table 2.
3. Results and Discussion
Figure 3a shows the significant correlations (p < 0.05) between e-nose outputs and
volatile aromatic compounds found in raw rice samples. Among all e-nose sensors, MQ3
had the highest incidence of significant correlations with other variables. There were sig-
nificant positive correlations between MQ3 and MQ7 (r = 0.58), MQ8 (r = 0.56), MQ135 (r
= 0.71), MQ137 (r = 0.88), MQ138 (r = 0.78), MG118 (r = 0.51), and volatile aromatic com-
pounds such as hexanoic acid (C5, r = 0.57), valeric anhydride (C8, r = 0.62), nonanal (C10,
r = 0.49), octanoic acid (C16, r = 0.62), and benzene, 1,4-diethoxy (C23, r = 0.51). Addition-
ally, it can be observed that MQ137 was positively correlated with valeric anhydride (C8,
nonanal (C10, r = 0.49), and octanoic acid (C16, r = 0.62). Both nonanal and octanoic acid
are associated with a waxy and fatty aroma [68]. Interestingly, positive correlations that
have been reported between gas sensors and volatile aromatic compounds related to rice
aging and yellowing processes, such as hexanoic acid, octanoic acid, and nonanal [18,69
71], were identified in the present study. This shows that the outputs obtained from the e-
nose used in this study have great potential as a low-cost alternative tool for rice quality
traits assessment and monitoring applications.
Figure 2.
Regression models developed using the (
a
) first derivative of NIR absorbance values
to estimate the aromas (Models 3 and 4) and (
b
) physicochemical quality (Model 5) of rice, and
(
c
) electronic nose sensors’ data as inputs to estimate rice aromas (Models 6 and 7). A description of
the electronic nose sensors can be found in Table 2.
3. Results and Discussion
Figure 3a shows the significant correlations (p< 0.05) between e-nose outputs and
volatile aromatic compounds found in raw rice samples. Among all e-nose sensors, MQ3
had the highest incidence of significant correlations with other variables. There were
significant positive correlations between MQ3 and MQ7 (r = 0.58), MQ8 (r = 0.56), MQ135
(r = 0.71), MQ137 (r = 0.88), MQ138 (r = 0.78), MG118 (r = 0.51), and volatile aromatic com-
pounds such as hexanoic acid (C5, r = 0.57), valeric anhydride (C8, r = 0.62), nonanal (C10,
r = 0.49), octanoic acid (C16, r = 0.62), and benzene, 1,4-diethoxy (C23, r = 0.51). Addition-
ally, it can be observed that MQ137 was positively correlated with valeric anhydride (C8,
nonanal (C10, r = 0.49), and octanoic acid (C16, r = 0.62). Both nonanal and octanoic acid are
associated with a waxy and fatty aroma [
68
]. Interestingly, positive correlations that have
been reported between gas sensors and volatile aromatic compounds related to rice aging
and yellowing processes, such as hexanoic acid, octanoic acid, and nonanal [
18
,
69
71
],
were identified in the present study. This shows that the outputs obtained from the e-nose
used in this study have great potential as a low-cost alternative tool for rice quality traits
assessment and monitoring applications.
Foods 2022,11, 1181 8 of 16
Foods 2022, 11, x FOR PEER REVIEW 8 of 17
Figure 3. Significant correlations (p < 0.05) between the (a) electronic nose sensors outputs obtained
from raw rice samples with volatile aromatic compounds detected in the raw rice and (b) electronic
nose sensors outputs obtained from cooked rice samples with volatile aromatic compounds, color,
texture, and pH of the cooked rice.
Figure 3.
Significant correlations (p< 0.05) between the (
a
) electronic nose sensors’ outputs obtained
from raw rice samples with volatile aromatic compounds detected in the raw rice and (
b
) electronic
nose sensors’ outputs obtained from cooked rice samples with volatile aromatic compounds, color,
texture, and pH of the cooked rice.
Foods 2022,11, 1181 9 of 16
Figure 3b shows the significant correlations (p< 0.05) between the e-nose outputs
obtained from cooked rice samples with the volatile aromatic compounds, color, texture,
and pH of the cooked rice samples. It was observed that the volatile compounds that
belong to the esters group had a significant positive correlation with some of the e-nose
sensors’ outputs. Dodecanoic acid, ethyl ester (C38) was positively correlated with MQ3
(
r = 0.93
), MQ4 (r = 0.87), MQ7 (r = 0.94), MQ8 (r = 0.74), MQ136 (r = 0.73), and MG811
(
r = 0.87
). Meanwhile, hexadecanoic acid, methyl ester (C39) was positively correlated with
MQ135 (r = 0.83), MQ137 (r = 0.69), and MQ138 (r = 0.65). Dodecanoic acid, ethyl ester, is
associated with sweet, waxy, floral, soapy, and clean aromas, whereas waxy and creamy
are used to describe the aromas of hexadecanoic acid and methyl ester [
68
]. Additionally,
it was observed that the lightness (L) of the cooked rice was negatively correlated with
benzaldehyde (C2, r =
0.66), furan, 2-pentyl- (C4, r =
0.50), and 2,4-decadienal,(E,E) (C37,
r =
0.95). This could be associated with the high abundance of the aforementioned volatile
aromatic compounds detected in pigmented rice samples, such as the wild rice (WRO),
which had lower L values than the other types of rice [
16
,
72
]. Even though some significant
correlations were observed between e-nose outputs of cooked rice and volatile compounds
detected in cooked rice samples, it will be convenient to develop a method to assess cooked
rice quality traits using the raw rice samples to avoid tedious sample preparation and
destructive analysis. Additionally, previous studies showed promising potential through
emerging trends of using sensor technology combined with the chemometrics technique to
indirectly assess cooked rice quality traits using raw rice samples [7375].
Table 3shows the statistical results of the pattern recognition models developed to
classify 17 types of commercial rice using the ANN algorithms based on the first derivative
of NIR absorbances (Model 1) and e-nose readings (Model 2) as inputs. All of the classi-
fication models had high accuracy in all three stages of training, validation, and testing.
The accuracies for Models 1 and 2 were high and similar (98.7% and 98.6%, respectively).
Additionally, the MSE values were low, and the training had lower MSE than the testing
stage, showing that the models had no sign of under- or overfitting. Figure 4shows the
overall receiver operating characteristics (ROC) curves of the classification models. From
the figure, it can be observed that all of the 17 types of rice were close to the true positive
rate on the vertical axis, showing that the models have high sensitivity to classify the rice
samples with respect to the 17 types of rice.
Table 3.
Statistical results of Model 1 and Model 2 developed based on artificial neural network
pattern recognition. Abbreviations: BR: Bayesian regularization; MSE: mean squared error; e-nose:
electronic nose; NIR: near-infrared.
Algorithm Stages Samples Accuracy (%) Error (%) Performance
(MSE)
Model 1 (Inputs: NIR absorbance of raw rice; targets: 17 types of rice)
BR
7 neurons
Training 107 100 0.0 <0.001
Testing 46 95.7 4.3 0.003
Overall 153 98.7 1.3
Model 2 (Inputs: E-nose outputs of raw rice; targets: 17 types of rice)
BR
10 neurons
Training 357 100 0.0 <0.001
Testing 153 95.4 4.6 0.005
Overall 510 98.6 1.4
Foods 2022,11, 1181 10 of 16
Foods 2022, 11, x FOR PEER REVIEW 10 of 17
Figure 4. The receiver operating characteristic (ROC) curves showing the true-positive versus false-
positive rates for (a) Model 1 and (b) Model 2 to classify 17 types of rice samples.
The statistical results of the regression models developed using the ANN algorithm
to predict the aroma and physicochemical qualities of the rice based on the first derivative
of NIR absorbance values (Model 3Model 5) and e-nose outputs (Model 6 and Model 7)
are shown in Table 4. Overall, all models had high accuracy, denoted by the correlation
coefficient (R) values presented in the table. Model 3 and Model 4 developed using the BR
algorithm showed high overall accuracy (Model 3: R = 0.95; Model 4: R = 0.98). Moreover,
the MSE values for both models at the training stage (Model 3: MSE = 1.47 × 108; Model 4:
MSE = 2.22 × 107) were lower than at the testing stage (Model 3: MSE = 1.23 × 1010; Model
4: MSE = 2.54 × 109), showing the best fit models. Model 5, when predicting the color,
texture, and pH of the cooked rice, showed high overall accuracy, at R = 0.96. Lower MSE
values for training (MSE = 0.01) compared with the validation (MSE = 0.07) and testing
(MSE = 0.08) stages and the latter two being similar confirm no signs of under- or overfit-
ting.
Table 4. Statistical results of the artificial neural network regression models developed using the
first derivative of near-infrared absorbances (Model 3Model 5) and electronic nose measurements
(Model 6 and Model 7) as inputs to predict rice quality. Abbreviations: BR: Bayesian regularization;
CGB: conjugate gradient with PowellBeale restarts; LM: LevenbergMarquardt; R: correlation co-
efficient; MSE: mean squared error; e-nose: electronic nose; NIR: near-infrared.
Algorithm
Stages
Observations
(Samples × Targets)
R
Slope
Performance (MSE)
Model 3 (Inputs: raw rice NIR; targets: raw rice aroma)
BR
Training
3531
1.00
1.00
1.47 × 108
10 neurons
Testing
1518
0.85
0.91
1.23 × 1010
Overall
5049
0.95
0.97
Model 4 (Inputs: raw rice NIR; targets: cooked rice aroma)
BR
Training
1070
0.99
1.00
2.22 × 107
7 neurons
Testing
460
0.92
0.98
2.54 × 109
Overall
1530
0.98
0.99
Model 5 (Inputs: raw rice NIR; targets: cooked rice physicochemical quality)
Training
455
0.99
0.98
0.01
CGB
Validation
155
0.93
0.89
0.07
10 neurons
Testing
155
0.90
0.88
0.08
Figure 4.
The receiver operating characteristic (ROC) curves showing the true-positive versus false-
positive rates for (a) Model 1 and (b) Model 2 to classify 17 types of rice samples.
The statistical results of the regression models developed using the ANN algorithm to
predict the aroma and physicochemical qualities of the rice based on the first derivative
of NIR absorbance values (Model 3–Model 5) and e-nose outputs (Model 6 and Model 7)
are shown in Table 4. Overall, all models had high accuracy, denoted by the correlation
coefficient (R) values presented in the table. Model 3 and Model 4 developed using the BR
algorithm showed high overall accuracy (Model 3: R = 0.95; Model 4: R = 0.98). Moreover,
the MSE values for both models at the training stage (Model 3: MSE = 1.47
×
10
8
; Model 4:
MSE = 2.22
×
10
7
) were lower than at the testing stage (Model 3: MSE = 1.23
×
10
10
;
Model 4: MSE = 2.54
×
10
9
), showing the best fit models. Model 5, when predicting the
color, texture, and pH of the cooked rice, showed high overall accuracy, at R = 0.96. Lower
MSE values for training (MSE = 0.01) compared with the validation (MSE = 0.07) and
testing (MSE = 0.08) stages and the latter two being similar confirm no signs of under- or
overfitting.
Table 4.
Statistical results of the artificial neural network regression models developed using the
first derivative of near-infrared absorbances (Model 3–Model 5) and electronic nose measurements
(Model 6 and Model 7) as inputs to predict rice quality. Abbreviations: BR: Bayesian regularization;
CGB: conjugate gradient with Powell–Beale restarts; LM: Levenberg–Marquardt; R: correlation
coefficient; MSE: mean squared error; e-nose: electronic nose; NIR: near-infrared.
Algorithm Stages Samples Observations
(Samples ×Targets) RSlope Performance
(MSE)
Model 3 (Inputs: raw rice NIR; targets: raw rice aroma)
BR Training 107 3531 1.00 1.00 1.47 ×108
10 neurons Testing 46 1518 0.85 0.91 1.23 ×1010
Overall 153 5049 0.95 0.97
Model 4 (Inputs: raw rice NIR; targets: cooked rice aroma)
BR Training 107 1070 0.99 1.00 2.22 ×107
7 neurons Testing 46 460 0.92 0.98 2.54 ×109
Overall 153 1530 0.98 0.99
Model 5 (Inputs: raw rice NIR; targets: cooked rice physicochemical quality)
Training 91 455 0.99 0.98 0.01
CGB Validation 31 155 0.93 0.89 0.07
10 neurons Testing 31 155 0.90 0.88 0.08
Overall 153 765 0.96 0.94
Foods 2022,11, 1181 11 of 16
Table 4. Cont.
Algorithm Stages Samples Observations
(Samples ×Targets) RSlope Performance
(MSE)
Model 6 (Inputs: raw rice e-nose; targets: raw rice aroma)
Training 356 11,748 0.95 0.90 3.37 ×109
LM Validation 77 2541 0.95 0.92 3.47 ×109
10 neurons Testing 77 2541 0.93 0.87 5.15 ×109
Overall 510 16,830 0.95 0.90
Model 7 (Inputs: raw rice e-nose; targets: cooked rice aroma)
BR Training 357 3570 0.98 0.97 5.77 ×108
10 neurons Testing 153 1530 0.92 0.95 3.43 ×109
Overall 510 5100 0.96 0.96
The ANN regression Models 6 and 7 were developed using e-nose measurements
to predict the raw and cooked rice aromas, respectively. Model 6 showed high overall
accuracy (R = 0.95) and showed no overfitting signs, having a lower MSE value for training
(3.37
×
10
9
) compared to validation (MSE = 3.47
×
10
9
) and testing (MSE = 5.15
×
10
9
).
Additionally, Model 7 had high overall accuracy (R = 0.96) with training and testing MSE
values of 5.77
×
10
8
and 3.43
×
10
9
, respectively, showing that the MSE value for training
was lower than the testing stage.
Figure 5shows the overall regression models used to predict the aroma (Models 3 and
4; Models 6 and 7) and color, texture, and pH (Model 5) of the rice. The accuracy for the
regression models developed using the first derivatives of NIR absorbance values (Model
3–5) and e-nose readings (Models 6 and 7) as inputs were high, with the R values of the
models ranging between 0.95 and 0.98. These results are in agreement with previous studies,
which obtained high accuracy predictions using ANN algorithms in food and beverages
research areas, such as in coffee aroma assessment [
25
], mold growth prediction [
76
], and
mulberry fruit grading [
77
]. It should be noted that Model 4, Model 5, and Model 7 are valid
to predict the cooked rice quality traits using the standard method specified in Section 2. It
is because different cooking procedures may affect the cooked rice quality, as suggested in
previous studies by Maleki et al. [6], Fracassetti et al. [78], and Krongworakul et al. [79].
Following the present results, previous studies have shown that the chemometrics of
rice obtained from low-cost sensors have been successfully applied to monitor rice quality.
For example, Arjharn et al. [
80
] used the e-nose developed by the research team from the
Agricultural Engineering Laboratory of Chiang Mai University to detect rancidity and
insect infestation in brown rice during storage. The authors used the e-nose readings
coupled with partial least square-discriminant analysis (PLS-DA) to classify the brown rice
samples according to the normal, rancid, and infested rice groups. However, the study
would have been more complete if it had included different levels of rancidity and severity
of insect infestation over time to assess the sensitivity of the e-nose system. Furthermore,
findings from this study supported previous studies that applied chemical and aroma
fingerprinting for food authentication to detect food fraud [8183].
The application of NIR absorbance values obtained in this study showed superior
accuracy in predicting the physicochemical quality of the rice using the ANN modeling than
those reported by Onmankhong and Sirisomboon [
74
]. The authors used the absorbance
values measured using a Fourier transform NIR (FT-NIR) spectrometer coupled with the
PSLR model to predict the hardness (coefficient of determination, R
2
= 0.70) and toughness
(R
2
= 0.66) of cooked parboiled rice. However, the moderate accuracy of PLSR models
indicated a weak predictive ability to estimate the rice quality adequately [84]. In another
study, Sampaio et al. [
60
] compared the ability of multiple linear regression (MLP) and ANN
to predict the biochemical composition and pasting parameters of rice. The authors found
that the ANN (R
2
= 0.97–0.99) presented a higher determination coefficient than the MLP
(R
2
= 0.27–0.96), showing the ability of ANN to produce a higher accuracy model. Moreover,
one of the advantages of using ANN is that the model can perform multi-target prediction,
Foods 2022,11, 1181 12 of 16
while other machine learning methods are single-target [
85
,
86
]. Since NIR spectra can be
used to predict multiple quality traits, the ANN model may provide a greater advantage to
obtain simultaneous predictions of rice quality traits. The fact that 17 types of rice were
included in this study helps generate a more general machine learning model that can be
applied to classify various types of rice available in the market. This includes the types of
rice from similar cultivars, subspecies (e.g., japonica and indica), and origin.
Foods 2022, 11, x FOR PEER REVIEW 12 of 17
Figure 5. The overall correlation of the regression ANN models developed to predict rice aroma and
physicochemical quality for (a) Model 3; inputs: first derivative of NIR absorbance; targets: raw rice
aroma, (b) Model 4; inputs: first derivative of NIR absorbance; targets: cooked rice aroma, (c) Model
5; inputs: first derivative of NIR absorbance; targets: cooked rice physicochemical quality, (d)
Model 6; inputs: e-nose outputs; targets: raw rice aroma, and (e) Model 7; inputs: e-nose outputs;
targets: cooked rice aroma.
Following the present results, previous studies have shown that the chemometrics of
rice obtained from low-cost sensors have been successfully applied to monitor rice quality.
For example, Arjharn et al. [80] used the e-nose developed by the research team from the
Agricultural Engineering Laboratory of Chiang Mai University to detect rancidity and in-
sect infestation in brown rice during storage. The authors used the e-nose readings cou-
pled with partial least square-discriminant analysis (PLS-DA) to classify the brown rice
samples according to the normal, rancid, and infested rice groups. However, the study
would have been more complete if it had included different levels of rancidity and sever-
ity of insect infestation over time to assess the sensitivity of the e-nose system. Further-
more, findings from this study supported previous studies that applied chemical and
aroma fingerprinting for food authentication to detect food fraud [8183].
The application of NIR absorbance values obtained in this study showed superior
accuracy in predicting the physicochemical quality of the rice using the ANN modeling
than those reported by Onmankhong and Sirisomboon [74]. The authors used the absorb-
ance values measured using a Fourier transform NIR (FT-NIR) spectrometer coupled with
the PSLR model to predict the hardness (coefficient of determination, R2 = 0.70) and tough-
ness (R2 = 0.66) of cooked parboiled rice. However, the moderate accuracy of PLSR models
indicated a weak predictive ability to estimate the rice quality adequately [84]. In another
study, Sampaio et al. [60] compared the ability of multiple linear regression (MLP) and
ANN to predict the biochemical composition and pasting parameters of rice. The authors
found that the ANN (R2 = 0.970.99) presented a higher determination coefficient than the
Figure 5.
The overall correlation of the regression ANN models developed to predict rice aroma
and physicochemical quality for (
a
) Model 3; inputs: first derivative of NIR absorbance; targets:
raw rice aroma, (
b
) Model 4; inputs: first derivative of NIR absorbance; targets: cooked rice aroma,
(
c
) Model 5; inputs: first derivative of NIR absorbance; targets: cooked rice physicochemical quality,
(
d
) Model 6; inputs: e-nose outputs; targets: raw rice aroma, and (
e
) Model 7; inputs: e-nose outputs;
targets: cooked rice aroma.
Although previous studies had contributed to applying chemical and aroma finger-
printing in assessing rice quality traits, this study presents the feasibility of using low-cost
sensors to assess rice quality traits based on qualitative and quantitative techniques. These
findings suggested that the chemical fingerprinting of rice obtained using the handheld
NIR spectrometer may be used to classify different types of rice and predict the aroma
and physicochemical parameters of the rice. Moreover, the aroma fingerprints obtained
from the e-nose sensors have great possibilities to be utilized as a reliable, low-cost alterna-
tive for rice aroma profiling and quality monitoring since the sensor can be miniaturized
and installed in rice processing facilities. The method proposed in the study has some
advantages in avoiding tedious sample preparation and saving time for routine inspection
to assess cooked rice quality traits under similar cooking procedures. Furthermore, the
proposed method may also reduce the requirement to run multiple analytical tests to assess
various rice quality traits at every routine inspection (e.g., texture analyzer to determine
the cooked rice texture, GC-MS analysis to determine the relative abundance of rice aroma,
Foods 2022,11, 1181 13 of 16
and other physicochemical tests). In further studies, the use of chemical fingerprinting
obtained from low-cost and portable tools could be a means of a high-throughput and
non-invasive approach for on-site quality inspection. This will also provide a promising
solution to prevent destructive testing while obtaining the sample data. Further research
should include other quality traits such as biochemical (starch, amylose, fat, and protein)
data and sensory quality to assess the association between rice quality traits and consumer
preferences.
4. Conclusions
This study set out to assess rice quality using chemical and aroma fingerprinting
obtained from a handheld NIR spectrometer and low-cost e-nose sensors combined with
machine learning modeling. The study has shown that the high overall accuracy of the
machine learning models is robust in assessing rice quality traits. The findings suggest
that the method presented in the study may allow qualitative and quantitative rice quality
monitoring at a lower cost than the conventional method. The results add to the expanding
application of artificial intelligence among rice producers and the government authority to
monitor and assess rice quality at a reasonable price and rapidly. Further research could
explore the potential application of chemical and aroma fingerprinting to predict consumer
preferences concerning the various qualities of rice available in the market.
Supplementary Materials:
The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/foods11091181/s1, Figure S1: Set up of the electronic nose mea-
suring a cooked rice sample.
Author Contributions:
Conceptualization, A.A., C.G.V. and S.F.; methodology, A.A., A.P., C.G.V. and
S.F.; software, C.G.V. and S.F.; validation, C.G.V., A.P. and S.F.; formal analysis, A.A.; investigation,
A.A., C.G.V. and S.F.; resources, C.G.V., A.P. and S.F.; data curation, A.A., C.G.V. and S.F.; writing—
original draft preparation, A.A.; writing—review and editing, C.G.V., A.P. and S.F.; visualization, A.A.,
C.G.V., A.P. and S.F.; supervision, S.F. and A.P. All authors have read and agreed to the published
version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement:
Data and intellectual property belong to the University of Melbourne;
any sharing needs to be evaluated and approved by the university.
Acknowledgments:
The authors would like to acknowledge the support from the Digital Agricul-
ture, Food, and Wine Group from the Faculty of Veterinary and Agricultural Sciences (FVAS), The
University of Melbourne.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Shahbandeh, M. Rice Consumption Worldwide in 2021/2022, by Country (in 1000 metric tons). Available online: https:
//www.statista.com/statistics/255971/top-countries- based-on-rice-consumption-2012-2013/ (accessed on 23 March 2022).
2.
Mané, I.; Bassama, J.; Ndong, M.; Mestres, C.; Diedhiou, P.M.; Fliedel, G. Deciphering urban consumer requirements for rice
quality gives insights for driving the future acceptability of local rice in Africa: Case study in the city of Saint-Louis in Senegal.
Food Sci. Nutr. 2021,9, 1614–1624. [CrossRef] [PubMed]
3.
Bairagi, S.; Demont, M.; Custodio, M.C.; Ynion, J. What drives consumer demand for rice fragrance? Evidence from South and
Southeast Asia. Br. Food J. 2020,122, 3473–3498. [CrossRef]
4.
Wahyudi, A.; Kuwornu, J.K.M.; Gunawan, E.; Datta, A.; Nguyen, L.T. Factors Influencing the Frequency of Consumers’ Purchases
of Locally-Produced Rice in Indonesia: A Poisson Regression Analysis. Agriculture 2019,9, 117. [CrossRef]
5.
Gondal, T.A.; Keast, R.S.J.; Shellie, R.A.; Jadhav, S.R.; Gamlath, S.; Mohebbi, M.; Liem, D.G. Consumer Acceptance of Brown and
White Rice Varieties. Foods 2021,10, 1950. [CrossRef]
6.
Maleki, C.; Oliver, P.; Lewin, S.; Liem, G.; Keast, R. Preference mapping of different water-to-rice ratios in cooked aromatic white
jasmine rice. J. Food Sci. 2020,85, 1576–1585. [CrossRef]
Foods 2022,11, 1181 14 of 16
7.
Pang, Y.; Ali, J.; Wang, X.; Franje, N.J.; Revilleza, J.E.; Xu, J.; Li, Z. Relationship of Rice Grain Amylose, Gelatinization Temperature
and Pasting Properties for Breeding Better Eating and Cooking Quality of Rice Varieties. PLoS ONE
2016
,11, e0168483. [CrossRef]
8.
Li, H.; Prakash, S.; Nicholson, T.M.; Fitzgerald, M.A.; Gilbert, R.G. The importance of amylose and amylopectin fine structure for
textural properties of cooked rice grains. Food Chem. 2016,196, 702–711. [CrossRef]
9.
Routray, W.; Rayaguru, K. 2-Acetyl-1-pyrroline: A key aroma component of aromatic rice and other food products. Food Rev. Int.
2017,34, 539–565. [CrossRef]
10.
Bairagi, S.; Mohanty, S.; Custodio, M.C. Consumers’ preferences for rice attributes in Cambodia: A choice modeling approach. J.
Agribus. Dev. Emerg. Econ. 2019,9, 94–108. [CrossRef]
11.
Hu, X.; Lu, L.; Guo, Z.; Zhu, Z. Volatile compounds, affecting factors and evaluation methods for rice aroma: A review. Trends
Food Sci. Technol. 2020,97, 136–146. [CrossRef]
12. Wei, X.; Sun, Q.; Methven, L.; Elmore, J.S. Comparison of the sensory properties of fragrant and non-fragrant rice (Oryza sativa),
focusing on the role of the popcorn-like aroma compound 2-acetyl-1-pyrroline. Food Chem.
2021
,339, 128077. [CrossRef] [PubMed]
13.
Setyaningsih, W.; Majchrzak, T.; Dymerski, T.; Namiesnik, J.; Palma, M. Key-Marker Volatile Compounds in Aromatic Rice (Oryza
sativa) Grains: An HS-SPME Extraction Method Combined with GCxGC-TOFMS. Molecules
2019
,24, 4180. [CrossRef] [PubMed]
14.
Sansenya, S.; Hua, Y.; Chumanee, S. The Correlation between 2-Acetyl-1-pyrroline Content, Biological Compounds and Molecular
Characterization to the Aroma Intensities of Thai Local Rice. J. Oleo Sci. 2018,67, 893–904. [CrossRef] [PubMed]
15.
Choi, S.; Seo, H.-S.; Lee, K.R.; Lee, S.; Lee, J. Effect of milling degrees on volatile profiles of raw and cooked black rice (Oryza
sativa L. cv. Sintoheugmi). Appl. Biol. Chem. 2018,61, 91–105. [CrossRef]
16.
Tikapunya, T.; Henry, R.J.; Smyth, H. Evaluating the sensory properties of unpolished Australian wild rice. Food Res. Int.
2018
,
103, 406–414. [CrossRef]
17.
Kasote, D.; Singh, V.K.; Bollinedi, H.; Singh, A.K.; Sreenivasulu, N.; Regina, A. Profiling of 2-Acetyl-1-Pyrroline and Other Volatile
Compounds in Raw and Cooked Rice of Traditional and Improved Varieties of India. Foods 2021,10, 1917. [CrossRef]
18.
Gao, C.; Li, Y.; Pan, Q.; Fan, M.; Wang, L.; Qian, H. Analysis of the key aroma volatile compounds in rice bran during storage and
processing via HS-SPME GC/MS. J. Cereal Sci. 2021,99, 103178. [CrossRef]
19.
Pal, S.; Bagchi, T.B.; Dhali, K.; Kar, A.; Sanghamitra, P.; Sarkar, S.; Samaddar, M.; Majumder, J. Evaluation of sensory, physic-
ochemical properties and Consumer preference of black rice and their products. J. Food Sci. Technol.
2019
,56, 1484–1494.
[CrossRef]
20.
Xia, Y.; Sun, Y.; Yuan, J.; Xing, C. Grain quality evaluation of japonica rice effected by cultivars, environment, and their interactions
based on appearance and processing characteristics. Food Sci. Nutr. 2021,9, 2129–2138. [CrossRef]
21.
Ribeiro, C.M.G.; Strunkis, C.d.M.; Campos, P.V.S.; Salles, M.O. Electronic Nose and Tongue Materials for Sensing. In Reference
Module in Biomedical Sciences; Elsevier: Amsterdam, The Netherlands, 2021. [CrossRef]
22.
Jafari, S.-M.; Tsimidou, M.Z.; Rajabi, H.; Kyriakoudi, A. Chapter 16—Bioactive ingredients of saffron: Extraction, analysis,
applications. In Saffron; Koocheki, A., Khajeh-Hosseini, M., Eds.; Woodhead Publishing: Sawston, UK, 2020. [CrossRef]
23.
Sari, I.M.; Wijaya, D.R.; Hidayat, W.; Kannan, R. An Approach to Classify Rice Quality using Electronic Nose Dataset-based
Naïve Bayes Classifier. In Proceedings of the 2021 International Symposium on Electronics and Smart Devices (ISESD), Bandung,
Indonesia, 29–30 June 2021; pp. 1–5.
24.
Erlangga, F.; Wijaya, D.R.; Wikusna, W. Electronic Nose Dataset for Classifying Rice Quality using Neural Network. In Proceedings
of the 2021 9th International Conference on Information and Communication Technology (ICoICT), Yogyakarta, Indonesia, 3–5
August 2021; pp. 462–466.
25.
Gonzalez Viejo, C.; Tongson, E.; Fuentes, S. Integrating a Low-Cost Electronic Nose and Machine Learning Modelling to Assess
Coffee Aroma Profile and Intensity. Sensors 2021,21, 2016. [CrossRef]
26.
Summerson, V.; Gonzalez Viejo, C.; Pang, A.; Torrico, D.D.; Fuentes, S. Assessment of Volatile Aromatic Compounds in Smoke
Tainted Cabernet Sauvignon Wines Using a Low-Cost E-Nose and Machine Learning Modelling. Molecules
2021
,26, 5108.
[CrossRef] [PubMed]
27.
Gonzalez Viejo, C.; Fuentes, S.; Hernandez-Brenes, C. Smart Detection of Faults in Beers Using Near-Infrared Spectroscopy, a
Low-Cost Electronic Nose and Artificial Intelligence. Fermentation 2021,7, 117. [CrossRef]
28.
Chen, H.-z.; Zhang, M.; Guo, Z. Discrimination of fresh-cut broccoli freshness by volatiles using electronic nose and gas
chromatography-mass spectrometry. Postharvest Biol. Technol. 2019,148, 168–175. [CrossRef]
29.
Zhu, D.; Ren, X.; Wei, L.; Cao, X.; Ge, Y.; Liu, H.; Li, J. Collaborative analysis on difference of apple fruits flavour using electronic
nose and electronic tongue. Sci. Hortic. 2020,260, 108879. [CrossRef]
30.
Wakhid, S.; Sarno, R.; Sabilla, S.I. The effect of gas concentration on detection and classification of beef and pork mixtures using
E-nose. Comput. Electron. Agric. 2022,195, 106838. [CrossRef]
31.
Huang, C.; Gu, Y. A Machine Learning Method for the Quantitative Detection of Adulterated Meat Using a MOS-Based E-Nose.
Foods 2022,11, 602. [CrossRef]
32.
Rusinek, R.; Siger, A.; Gawrysiak-Witulska, M.; Rokosik, E.; Malaga-Toboła, U.; Gancarz, M. Application of an electronic nose for
determination of pre-pressing treatment of rapeseed based on the analysis of volatile compounds contained in pressed oil. Int. J.
Food Sci. Technol. 2020,55, 2161–2170. [CrossRef]
33.
Rusinek, R.; Jele ´n, H.; Malaga-Toboła, U.; Molenda, M.; Gancarz, M. Influence of changes in the level of volatile compounds
emitted during rapeseed quality degradation on the reaction of MOS type sensor-array. Sensors 2020,20, 3135. [CrossRef]
Foods 2022,11, 1181 15 of 16
34.
Gancarz, M.; Malaga-Toboła, U.; Oniszczuk, A.; Tabor, S.; Oniszczuk, T.; Gawrysiak-Witulska, M.; Rusinek, R. Detection and
measurement of aroma compounds with the electronic nose and a novel method for MOS sensor signal analysis during the wheat
bread making process. Food Bioprod. Process. 2021,127, 90–98. [CrossRef]
35.
Tatli, S.; Mirzaee-Ghaleh, E.; Rabbani, H.; Karami, H.; Wilson, A.D. Rapid detection of urea fertilizer effects on voc emissions
from cucumber fruits using a MOS E-Nose Sensor Array. Agronomy 2021,12, 35. [CrossRef]
36.
Strani, L.; D’Alessandro, A.; Ballestrieri, D.; Durante, C.; Cocchi, M. Fast GC E-Nose and Chemometrics for the Rapid Assessment
of Basil Aroma. Chemosensors 2022,10, 105. [CrossRef]
37.
Cozzolino, D. The Ability of Near Infrared (NIR) Spectroscopy to Predict Functional Properties in Foods: Challenges and
Opportunities. Molecules 2021,26, 6981. [CrossRef] [PubMed]
38.
Be´c, K.B.; Grabska, J.; Huck, C.W. Current and future research directions in computer-aided near-infrared spectroscopy: A
perspective. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 2021,254, 119625. [CrossRef] [PubMed]
39.
Zeng, J.; Guo, Y.; Han, Y.; Li, Z.; Yang, Z.; Chai, Q.; Wang, W.; Zhang, Y.; Fu, C. A Review of the discriminant analysis methods for
food quality based on near-infrared spectroscopy and pattern recognition. Molecules 2021,26, 749. [CrossRef] [PubMed]
40.
Pasquini, C. Near infrared spectroscopy: A mature analytical technique with new perspectives—A review. Anal. Chim. Acta
2018
,
1026, 8–36. [CrossRef]
41.
Teye, E.; Amuah, C.L.Y.; McGrath, T.; Elliott, C. Innovative and rapid analysis for rice authenticity using hand-held NIR
spectrometry and chemometrics. Spectrochim. Acta A Mol. Biomol. Spectrosc. 2019,217, 147–154. [CrossRef]
42.
Li, L.; Wang, Y.; Jin, S.; Li, M.; Chen, Q.; Ning, J.; Zhang, Z. Evaluation of black tea by using smartphone imaging coupled with
micro-near-infrared spectrometer. Spectrochim. Acta A Mol. Biomol. Spectrosc. 2021,246, 118991. [CrossRef]
43.
Henrique da Silva Melo, B.; Figueiredo Sales, R.; da Silva Bastos Filho, L.; Souza Povoas da Silva, J.; Gabrielle Carolino de Almeida
Sousa, A.; Maria Camara Peixoto, D.; Pimentel, M.F. Handheld near infrared spectrometer and machine learning methods applied
to the monitoring of multiple process stages in industrial sugar production. Food Chem. 2021,369, 130919. [CrossRef]
44.
Duarte, E.S.A.; de Almeida, V.E.; da Costa, G.B.; de Araujo, M.C.U.; Veras, G.; Diniz, P.; Fernandes, D.D.S. Feasibility study on
quantification and authentication of the cassava starch content in wheat flour for bread-making using NIR spectroscopy and
digital images. Food Chem. 2021,368, 130843. [CrossRef]
45.
Richens, J.G.; Lee, C.M.; Johri, S. Improving the accuracy of medical diagnosis with causal machine learning. Nat. Commun.
2020
,
11, 3923. [CrossRef]
46.
Sircar, A.; Yadav, K.; Rayavarapu, K.; Bist, N.; Oza, H. Application of machine learning and artificial intelligence in oil and gas
industry. Pet. Res. 2021,6, 379–391. [CrossRef]
47. Dogan, A.; Birant, D. Machine learning and data mining in manufacturing. Expert Syst. Appl. 2021,166, 114060. [CrossRef]
48.
Yao, B.; Feng, T. Machine Learning in Automotive Industry; SAGE Publications: London, UK, 2018; Volume 10, p. 1687814018805787.
49.
Liakos, K.G.; Busato, P.; Moshou, D.; Pearson, S.; Bochtis, D. Machine Learning in Agriculture: A Review. Sensors
2018
,18, 2674.
[CrossRef] [PubMed]
50.
Jiménez-Carvelo, A.M.; González-Casado, A.; Bagur-González, M.G.; Cuadros-Rodríguez, L. Alternative data mining/machine
learning methods for the analytical evaluation of food quality and authenticity—A review. Food Res. Int.
2019
,122, 25–39.
[CrossRef]
51.
Basile, T.; Marsico, A.D.; Perniola, R. Use of Artificial Neural Networks and NIR Spectroscopy for Non-Destructive Grape Texture
Prediction. Foods 2022,11, 281. [CrossRef] [PubMed]
52.
Astray, G.; Mejuto, J.C.; Martínez-Martínez, V.; Nevares, I.; Alamo-Sanza, M.; Simal-Gandara, J. Prediction models to control
aging time in red wine. Molecules 2019,24, 826. [CrossRef] [PubMed]
53.
Fuentes, S.; Summerson, V.; Tongson, E.; Viejo, C.G. Latest developments and potential uses of digital technologies and artificial
intelligence (AI) to assess smoke contamination in grapevines, berries and taint in wines. Wine Vitic. J. 2022,37, 62–66.
54.
Gonzalez Viejo, C.; Torrico, D.; Dunshea, F.; Fuentes, S. Emerging Technologies Based on Artificial Intelligence to Assess the
Quality and Consumer Preference of Beverages. Beverages 2019,5, 62. [CrossRef]
55.
Gonzalez Viejo, C.; Fuentes, S.; Li, G.; Collmann, R.; Condé, B.; Torrico, D. Development of a robotic pourer constructed with
ubiquitous materials, open hardware and sensors to assess beer foam quality using computer vision and pattern recognition
algorithms: RoboBEER. Food Res. Int. 2016,89, 504–513. [CrossRef]
56.
Martinez-Castillo, C.; Astray, G.; Mejuto, J.C.; Simal-Gandara, J. Random forest, artificial neural network, and support vector
machine models for honey classification. eFood 2020,1, 69–76. [CrossRef]
57.
Zhang, X.; Gao, Z.; Yang, Y.; Pan, S.; Yin, J.; Yu, X. Rapid identification of the storage age of dried tangerine peel using a hand-held
near infrared spectrometer and machine learning. J. Near Infrared Spectrosc. 2022,30, 31–39. [CrossRef]
58.
Fuentes, S.; Tongson, E.; Unnithan, R.R.; Gonzalez Viejo, C. Early Detection of Aphid Infestation and Insect-Plant Interaction
Assessment in Wheat Using a Low-Cost Electronic Nose (E-Nose), Near-Infrared Spectroscopy and Machine Learning Modeling.
Sensors 2021,21, 5948. [CrossRef] [PubMed]
59.
Ye, Z.; Liu, Y.; Li, Q. Recent Progress in Smart Electronic Nose Technologies Enabled with Machine Learning Methods. Sensors
2021,21, 7620. [CrossRef] [PubMed]
60.
Sampaio, P.S.; Almeida, A.S.; Brites, C.M. Use of Artificial Neural Network Model for Rice Quality Prediction Based on Grain
Physical Parameters. Foods 2021,10, 3016. [CrossRef]
Foods 2022,11, 1181 16 of 16
61.
Gonzalez Viejo, C.; Fuentes, S.; Godbole, A.; Widdicombe, B.; Unnithan, R.R. Development of a low-cost e-nose to assess aroma
profiles: An artificial intelligence application to assess beer quality. Sens. Actuators B Chem. 2020,308, 127688. [CrossRef]
62.
Astuti, R.D.; Fibri, D.L.N.; Handoko, D.D.; David, W.; Budijanto, S.; Shirakawa, H. The Volatile Compounds and Aroma
Description in Various Rhizopus oligosporus Solid-State Fermented and Nonfermented Rice Bran. Fermentation
2022
,8, 120.
[CrossRef]
63.
Xu, M.; Jin, Z.; Lan, Y.; Rao, J.; Chen, B. HS-SPME-GC-MS/olfactometry combined with chemometrics to assess the impact of
germination on flavor attributes of chickpea, lentil, and yellow pea flours. Food Chem. 2019,280, 83–95. [CrossRef] [PubMed]
64.
Jarunrattanasri, A.; Theerakulkait, C.; Cadwallader, K.R. Aroma components of acid-hydrolyzed vegetable protein made by
partial hydrolysis of rice bran protein. J. Agric. Food Chem. 2007,55, 3044–3050. [CrossRef]
65. Suttiarporn, P.; Sookwong, P.; Mahatheeranont, S. Fractionation and identification of antioxidant compounds from bran of Thai
black rice cv. Riceberry. Int. J. Chem. Eng. Appl. 2016,7, 109. [CrossRef]
66.
Ajarayasiri, J.; Chaiseri, S. Comparative study on aroma-active compounds in Thai, black and white glutinous rice varieties.
Agric. Nat. Resour. 2008,42, 715–722.
67.
Gonzalez Viejo, C.; Torrico, D.D.; Dunshea, F.R.; Fuentes, S. Development of Artificial Neural Network Models to Assess Beer
Acceptability Based on Sensory Properties Using a Robotic Pourer: A Comparative Model Approach to Achieve an Artificial
Intelligence System. Beverages 2019,5, 33. [CrossRef]
68.
The Good Scents Company. The Good Scents Company Information System. Available online: http://www.thegoodscentscompany.
com/ (accessed on 29 March 2022).
69.
Liu, J.; Liu, Y.; Wang, A.; Dai, Z.; Wang, R.; Sun, H.; Strappe, P.; Zhou, Z. Characteristics of moisture migration and volatile
compounds of rice stored under various storage conditions. J. Cereal Sci. 2021,102, 103323. [CrossRef]
70.
Zhao, Q.; Xue, Y.; Shen, Q. Changes in the major aroma-active compounds and taste components of Jasmine rice during storage.
Food Res. Int. 2020,133, 109160. [CrossRef]
71.
Liu, K.; Zhao, S.; Li, Y.; Chen, F. Analysis of volatiles in brown rice, germinated brown rice, and selenised germinated brown rice
during storage at different vacuum levels. J. Sci. Food Agric. 2018,98, 2295–2301. [CrossRef] [PubMed]
72.
Aznan, A.; Gonzalez Viejo, C.; Pang, A.; Fuentes, S. Computer Vision and Machine Learning Analysis of Commercial Rice Grains:
A Potential Digital Approach for Consumer Perception Studies. Sensors 2021,21, 6354. [CrossRef] [PubMed]
73.
Siriphollakul, P.; Nakano, K.; Kanlayanarat, S.; Ohashi, S.; Sakai, R.; Rittiron, R.; Maniwara, P. Eating quality evaluation of Khao
Dawk Mali 105 rice using near-infrared spectroscopy. LWT—Food Sci. Technol. 2017,79, 70–77. [CrossRef]
74.
Onmankhong, J.; Sirisomboon, P. Texture evaluation of cooked parboiled rice using nondestructive milled whole grain near
infrared spectroscopy. J. Cereal Sci. 2021,97, 103151. [CrossRef]
75.
Summerson, V.; Gonzalez Viejo, C.; Torrico, D.D.; Pang, A.; Fuentes, S. Detection of smoke-derived compounds from bushfires in
Cabernet-Sauvignon grapes, must, and wine using Near-Infrared spectroscopy and machine learning algorithms. OENO One
2020,54, 1105–1119. [CrossRef]
76.
Wawrzyniak, J. Prediction of fungal infestation in stored barley ecosystems using artificial neural networks. LWT
2021
,137,
110367. [CrossRef]
77.
Azarmdel, H.; Jahanbakhshi, A.; Mohtasebi, S.S.; Muñoz, A.R. Evaluation of image processing technique as an expert system
in mulberry fruit grading based on ripeness level using artificial neural networks (ANNs) and support vector machine (SVM).
Postharvest Biol. Technol. 2020,166, 111201. [CrossRef]
78.
Fracassetti, D.; Pozzoli, C.; Vitalini, S.; Tirelli, A.; Iriti, M. Impact of cooking on bioactive compounds and antioxidant activity of
pigmented rice cultivars. Foods 2020,9, 967. [CrossRef] [PubMed]
79.
Krongworakul, N.; Naivikul, O.; Boonsupthip, W.; Wang, Y.J. Effect of conventional and microwave heating on physical and
chemical properties of Jasmine brown rice in various forms. J. Food Process Eng. 2020,43, e13506. [CrossRef]
80.
Arjharn, W.; Neamsorn, N.; Changrue, V.; Chanbang, Y.; Kunasakdakul, K.; Theanjumpol, P.; Kantakaew, P.; Junyusen, P.;
Treeamnuk, T.; Junyusen, T. Electronic nose system for rancidity and insect monitoring of brown rice. E3S Web Conf.
2020
,187,
04015. [CrossRef]
81.
Amirvaresi, A.; Nikounezhad, N.; Amirahmadi, M.; Daraei, B.; Parastar, H. Comparison of near-infrared (NIR) and mid-infrared
(MIR) spectroscopy based on chemometrics for saffron authentication and adulteration detection. Food Chem.
2021
,344, 128647.
[CrossRef]
82.
Li, L.; Jin, S.; Wang, Y.; Liu, Y.; Shen, S.; Li, M.; Ma, Z.; Ning, J.; Zhang, Z. Potential of smartphone-coupled micro NIR spectroscopy
for quality control of green tea. Spectrochim. Acta A Mol. Biomol. Spectrosc. 2021,247, 119096. [CrossRef] [PubMed]
83.
Wilde, A.S.; Haughey, S.A.; Galvin-King, P.; Elliott, C.T. The feasibility of applying NIR and FT-IR fingerprinting to detect
adulteration in black pepper. Food Control 2019,100, 1–7. [CrossRef]
84.
Di Bucchianico, A. Coefficient of Determination (R2). In Encyclopedia of Statistics in Quality and Reliability; Wiley: Hoboken, NJ,
USA, 2008; Volume 1.
85.
Junior, S.B.; Mastelini, S.M.; Barbon, A.P.A.; Barbin, D.F.; Calvini, R.; Lopes, J.F.; Ulrici, A. Multi-target prediction of wheat flour
quality parameters with near infrared spectroscopy. Inf. Process. Agric. 2020,7, 342–354.
86.
Junior, S.B.; Santana, E.J.; Badaró, A.T.; Borrás, N.A.; Barbin, D.F. Advantages of Multi-Target Modelling for Spectral Regression.
In Spectroscopic Techniques & Artificial Intelligence for Food and Beverage Analysis; Springer: Berlin/Heidelberg, Germany, 2020;
pp. 95–121.
... Compared to GC-MS analysis, an e-nose provides rapid detection to obtain results in a few seconds or minutes. Previous studies in rice have shown the reliability of e-nose applications such as analyzing the volatile compounds in rice [11], distinguishing expired and non-expired rice [12], monitoring rancidity and insect infestation in brown rice [13], detecting fungal infection in jasmine brown rice [14], and identifying moldy rice [15]. In rice adulteration, Udomkun et al. [16] assessed the feasibility of a commercial e-nose paired with principal component analysis (PCA) to identify the degree of adulteration in Thai jasmine rice in storage conditions. ...
... On the other hand, MG811 was the sensor sensitive to carbon dioxide and had the highest FL in PC2. These results reflect those of a prior study by Aznan et al. [11], who also found significant positive correlations (p < 0.05) between MQ137 and volatile compounds found in raw rice, such as the valeric anhydride (r = 0.53), nonanal (r = 0.49), and octanoic acid (r = 0.54). Octanoic acid is a short-chain fatty acid found in raw rice that was developed through the oxidation of linoleic acid over the storage period [31], while nonanal is one of the major VOCs that contribute to the rice aroma associated with aldehydic, fatty, waxy, citrus, and floral aromas [32]. ...
... Stages Samples (n) All models showed no overfitting signs described by the lower MSE values obtained at the training stage compared to the validation and/or testing stages. Besides, comparable MSE values were obtained between the validation and testing stages in Models 7,8,9,and 11 and the training and testing stages in Models 10 and 12, showing no overfitting signs. Figure 5 shows the overall models, including the 95% prediction bounds for Models 7 to 12 developed using the NIR absorbance values as inputs. ...
Article
Full-text available
Rice fraud is one of the common threats to the rice industry. Conventional methods to detect rice adulteration are costly, time-consuming, and tedious. This study proposes quantitative prediction of rice adulteration levels measured through the packaging using a handheld near-infrared (NIR) spectrometer and electronic nose (e-nose) sensors measured directly on the samples paired with machine learning (ML) algorithms. For these purposes, the samples were prepared by mixing the rice at different ratios from 0% to 100% with a 10% increment based on the rice's weight, con-sisting of (i) rice from different origins, (ii) premium with regular rice, (iii) aromatic with non-aromatic and (iv) organic with non-organic rice. Multivariate data analysis was used to ex-plore the sample distribution and its relationship with the e-nose sensors for parameter engi-neering before ML modeling. Artificial neural network (ANN) algorithms were used to predict the adulteration levels of the rice samples using the e-nose sensors and NIR absorbances readings as inputs. Results showed that both sensing devices could detect rice adulteration at different mixing ratios with high correlation coefficients through direct (e-nose; R = 0.94 – 0.98) and non-invasive measurement through the packaging (NIR; R = 0.95 – 0.98). The proposed method uses low-cost, rapid, and portable sensing devices coupled with ML that have shown to be reliable and accurate to increase the efficiency of rice fraud detection through the rice production chain.
Article
Full-text available
The aim of this work is to assess the potentialities of the synergistic combination of an ultra-fast chromatography-based electronic nose as a fingerprinting technique and multivariate data analysis in the context of food quality control and to investigate the influence of some factors, i.e., basil variety, cut, and year of crop, in the final aroma of the samples. A low = level data fusion approach coupled with Principal Component Analysis (PCA) and ANOVA—Simultaneous Component Analysis (ASCA) was used in order to analyze the chromatographic signals acquired with two different columns (MXT-5 and MXT-1701). While the PCA analysis results highlighted the peculiarity of some basil varieties, differing either by a higher concentration of some of the detected chemical compounds or by the presence of different compounds, the ASCA analysis pointed out that variety and year are the most relevant effects, and also confirmed the results of previous investigations.
Article
Full-text available
Rice bran is known to have beneficial nutrients. Current studies suggest that solid-state fermentation affects the rice bran’s volatile profile. The aim of this study is to identify the volatile compounds and aroma description of fermented and nonfermented rice bran (FRB and NFRB) of Ciherang, Inpari30, IR64 and Inpari42. The fermentation was conducted using Rhizopus oligosporus solid-state fermentation. Headspace-solid phase microextraction coupled with GC/MS was performed, and the aroma was translated by 10 trained panelists through quantitative descriptive analysis (QDA). The result showed that 72 and 68 compounds were identified in FRB and NFRB, respectively. They are aldehydes, ketones, alcohols, acids, esters, fatty acid, phenol, benzenes, furan, thiazole, pyrazines, pyridine, lactones, terpenes, and hydrocarbons. The PCA showed that FRB was dominated by alcohols, whereas NFRB was dominated by aldehydes. The QDA described nine aromas, i.e., rancid, smoky, musty, grassy, green, earthy, cereal, and sweet in NFRB. The fermentation process added fermented attributes to the aroma description to FRB and enhanced the rancid, smoky, and musty aromas. These studies indicated that fermented rice bran might increase the volatile compound of rice bran. Thus, it may provide opportunities to develop the production of fermented rice bran as a functional ingredient.
Article
Full-text available
Meat adulteration is a global problem which undermines market fairness and harms people with allergies or certain religious beliefs. In this study, a novel framework in which a one-dimensional convolutional neural network (1DCNN) serves as a backbone and a random forest regressor (RFR) serves as a regressor, named 1DCNN-RFR, is proposed for the quantitative detection of beef adulterated with pork using electronic nose (E-nose) data. The 1DCNN backbone extracted a sufficient number of features from a multichannel input matrix converted from the raw E-nose data. The RFR improved the regression performance due to its strong prediction ability. The effectiveness of the 1DCNN-RFR framework was verified by comparing it with four other models (support vector regression model (SVR), RFR, backpropagation neural network (BPNN), and 1DCNN). The proposed 1DCNN-RFR framework performed best in the quantitative detection of beef adulterated with pork. This study indicated that the proposed 1DCNN-RFR framework could be used as an effective tool for the quantitative detection of meat adulteration.
Article
Full-text available
In this article, a combination of non-destructive NIR spectroscopy and machine learning techniques was applied to predict the texture parameters and the total soluble solids content (TSS) in intact berries. The multivariate models obtained by building artificial neural networks (ANNs) and applying partial least squares (PLS) regressions showed a better prediction ability after the elimination of uninformative spectral ranges. A very good prediction was obtained for TSS and springiness (R2 0.82 and 0.72). Qualitative models were obtained for hardness and chewiness (R2 0.50 and 0.53). No satisfactory calibration model could be established between the NIR spectra and cohesiveness. Textural parameters of grape are strictly related to the berry size. Before any grape textural measurement, a time-consuming berry-sorting step is compulsory. This is the first time a complete textural analysis of intact grape berries has been performed by NIR spectroscopy without any a priori knowledge of the berry density class.
Article
Full-text available
The widespread use of nitrogen chemical fertilizers in modern agricultural practices has raised concerns over hazardous accumulations of nitrogen-based compounds in crop foods and in agricultural soils due to nitrogen overfertilization. Many vegetables accumulate and retain large amounts of nitrites and nitrates due to repeated nitrogen applications or excess use of nitrogen fertilizers. Consequently, the consumption of high-nitrate crop foods may cause health risks to humans. The effects of varying urea-nitrogen fertilizer application rates on VOC emissions from cucumber fruits were investigated using an experimental MOS electronic-nose (e-nose) device based on differences in sensor-array responses to volatile emissions from fruits, recorded following different urea fertilizer treatments. Urea fertilizer was applied to cucumber plants at treatment rates equivalent to 0, 100, 200, 300, and 400 kg/ha. Cucumber fruits were then harvested twice, 4 and 5 months after seed planting, and evaluated for VOC emissions using an e-nose technology to assess differences in smellprint signatures associated with different urea application rates. The electrical signals from the e-nose sensor array data outputs were subjected to four aroma classification methods , including: linear and quadratic discriminant analysis (LDA-QDA), support vector machines (SVM), and artificial neural networks (ANN). The results suggest that combining the MOS e-nose technology with QDA is a promising method for rapidly monitoring urea fertilizer application rates applied to cucumber plants based on changes in VOC emissions from cucumber fruits. This new monitoring tool could be useful in adjusting future urea fertilizer application rates to help prevent nitrogen overfertilization.
Article
Full-text available
The main goal of this study was to test the ability of an artificial neural network (ANN) for rice quality prediction based on grain physical parameters and to conduct a comparison with multiple linear regression (MLR) using 66 samples in duplicate. The parameters used for rice quality prediction are related to biochemical composition (starch, amylose, ash, fat, and protein concentration) and pasting parameters (peak viscosity, trough, breakdown, final viscosity, and setback). These parameters were estimated based on grain appearance (length, width, length/width ratio, total whiteness, vitreous whiteness, and chalkiness), and milling yield (husked, milled, head) data. The MLR models were characterized by very low coefficient determination (R 2 = 0.27-0.96) and a root-mean-square error (RMSE) (0.08-0.56). Meanwhile, the ANN models presented a range for R 2 = 0.97-0.99, being characterized for R 2 = 0.98 (training), R 2 = 0.88 (validation), and R 2 = 0.90 (testing). According to these results, the ANN algorithms could be used to obtain robust models to predict both biochemical and pasting profiles parameters in a fast and accurate form, which makes them suitable for application to simultaneous qualitative and quantitative analysis of rice quality. Moreover, the ANN prediction method represents a promising approach to estimate several targeted biochemical and viscosity parameters with a fast and clean approach that is interesting to industry and consumers, leading to better assessment of rice classification for authenticity purposes.
Article
Full-text available
Near infrared (NIR) spectroscopy is considered one of the main routine analytical methods used by the food industry. This technique is utilised to determine proximate chemical compositions (e.g., protein, dry matter, fat and fibre) of a wide range of food ingredients and products. Novel algorithms and new instrumentation are allowing the development of new applications of NIR spectroscopy in the field of food science and technology. Specifically, several studies have reported the use of NIR spectroscopy to evaluate or measure functional properties in both food ingredients and products in addition to their chemical composition. This mini-review highlights and discussed the applications, challenges and opportunities that NIR spectroscopy offers to target the quantification and measurement of food functionality in dairy and cereals.
Article
Examining the purity of meat is a classical problem in developing countries, especially in Indonesia. The high economic value of beef causes counterfeiting to occur frequently. The forgery process is done through the simple practice of mixing in a certain percentage of pork. Several recent studies have shown that e-noses can examine beef purity through gas detection. This study aimed to determine the effect of gas concentration on the results of detection and classification of beef and pork mixtures by characterizing different meat samples in 3 chambers with a different size. The meat mixture dataset was retrieved by an e-nose device with an array of MQ series sensors that are sensitive to chemical scents. Classification of the meat mixtures was done in several stages: data acquisition from the 3 different sample chambers, statistical feature extraction, classification, ensemble learning, and performance evaluation based on a confusion matrix. The experimental results from this study indicate that the sample chamber with the highest gas concentration yielded the highest accuracy. The best accuracy result, i.e. 95.71%, was obtained with a 50-ml sample chamber using an ensemble method with the statistical parameters of kurtosis and skewness.
Article
Dried tangerine peel is a Chinese medicine with high medicinal value. The storage age is an important indicator of its medicinal value, so it is very significant to accurately identify the storage age of dried tangerine peel. Traditional physical and chemical analysis methods can be used to achieve this goal, but these methods are limited by their operability and convenience. Near infrared (NIR) spectroscopy and machine learning have excellent performance in the rapid detection of food and pharmaceutical samples. This study investigated the novel application of integrating a hand-held NIR spectrometer combined with machine learning to rapidly and accurately identify the storage age of Xinhui dried tangerine peel. Savitzky–Golay convolution smoothing, standard normal variate (SNV), first derivative, and second derivative pretreatments were employed to preprocess spectral data. Principal component analysis (PCA) was used to reduce the spectral data dimensions and obtain the characteristic spectral variables of each sample. Support vector machine (SVM) and k-nearest neighbor were applied to establish the qualitative discriminant models. The SNV-PCA-SVM model discriminant accuracy was 99.60% in the validation set and was 96.50% in the test set, showing excellent generalization performance. The results indicated that the method of using a hand-held NIR spectrometer combined with machine learning could be applied to rapidly identify the storage age of Xinhui dried tangerine peel. This is a promising and economical hand-held NIR spectroscopic method for assuring the dried tangerine peel age on-site.
Article
While starting a new 2021-22 grape-growing season in Australia, potential climatic anomalies such as bushfires are in the consciousness of many grape growers and winemakers. Increasing ambient temperatures has resulted in associated climatic anomalies, such as extreme wildfires in Australia, California, Siberia, Greece and Turkey as forecasted and reviewed by the latest Intergovernmental Panel on Climate Change report (IPPC-AR6). The latest IPCC report also claims a “virtually certain” increase in the frequency and intensity of heatwaves due to greenhouse emissions from burning fossil fuels. Many of these wildfires or bushfires occur in Mediterranean countries that cultivate grapes for winemaking. Recent review papers on smoke taint in wines related to mitigation techniques have assessed the state of the art research efforts related to smoke taint and developed methods for mitigation and incorporating some of the latest digital technologies for its assessment. Both reviews agreed that the best practices to remediate smoke taint in wine are activated carbon fining and reverse osmosis treatments. This article focuses on the latest advances to monitor the levels of smoke contamination in grapevines and grapes and smoke taint in wines using novel digital and non-invasive technologies. Much of these latest efforts are the product of research from the Digital Agriculture, Food and Wine group from The University of Melbourne. The latest research has produced tools, and artificial intelligence (AI) models with high accuracy (Table 1) to be deployed in near-real-time using affordable technologies either in the field between veraison to harvest and in the winery that can be accessible to winegrowers and winemakers.