Detection (Optical)

Received: 4 September 2020 Revised: 7 June 2021 Accepted: 15 June 2021
DOI: 10.1111/1541-4337.12801
COMPREH ENSIVE REVIEWS IN FOOD SCIENCE AND FOOD SAFETY
Research advancements in optical imaging and

spectroscopic techniques for nondestructive detection of
mold infection and mycotoxins in cereal grains and nuts
Gayatri Mishra1 Brajesh Kumar Panda1 Wilmer Ariza Ramirez1
Hyewon Jung1 Chandra B Singh1,2 Sang-Heon Lee1 Ivan Lee1
1 UniSA STEM, University of South
Australia, Mawson Lakes, South Abstract

Australia, Australia Cereal grains and nuts are represented as the economic backbone of many devel-
2 Centre for Applied Research, Innovation
oped and developing countries. Kernels of cereal grains and nuts are prone to
and Entrepreneurship, Lethbridge
College, Lethbridge, Alberta, Canada mold infection under high relative humidity and suitable temperature condi-
tions in the field as well as storage conditions. Health risks caused by molds and
Correspondence
their molecular metabolite mycotoxins are, therefore, important topics to inves-
Chandra B. Singh, UniSA STEM, Univer-
sity of South Australia, Mawson Lakes, SA tigate. Strict regulations have been developed by international trade regulatory
5095, Australia. bodies for the detection of mold growth and mycotoxin contamination across the
Email: chandra.singh@lethbridgecollege.
food chain starting from the harvest to storage and consumption. Molds and afla-
ca
toxins are not evenly distributed over the bulk of grains, thus appropriate sam-
pling for detection and quantification is crucial. Existing reference methods for
mold and mycotoxin detection are destructive in nature as well as involve skilled
labor and hazardous chemicals. Also, these methods cannot be used for inline
sorting of the infected kernels. Thus, analytical methods have been extensively
researched to develop the one that is more practical to be used in commercial
detection and sorting processes. Among various analytical techniques, optical
imaging and spectroscopic techniques are attracting growers’ attention for their
potential of nondestructive and rapid inline identification and quantification of
molds and mycotoxins in various food products. This review summarizes the
recent application of rapid and nondestructive optical imaging and spectroscopic
techniques, including digital color imaging, X-ray imaging, near-infrared spec-
troscopy, fluorescent, multispectral, and hyperspectral imaging. Advance chemo-
metric techniques to identify very low-level mold growth and mycotoxin con-
tamination are also discussed. Benefits, limitations, and challenges of deploying
these techniques in practice are also presented in this paper.
KEYWORDS
chemometrics, digital colour imaging, hyperspectral imaging, mold infection, mycotoxins,
Spectroscopy
Compr Rev Food Sci Food Saf. 2021;1–40. wileyonlinelibrary.com/journal/crf3 © 2021 Institute of Food Technologists
R 1
2 Nondestructive detection of mold infection and mycotoxins
1 INTRODUCTION through spores then develop visible branched filaments,

called hyphae and the collective mass of hyphae is referred
Food safety and quality are essential parameters from a to as the mycelium (Malloch, 1981). Cereal grains and nuts
consumer standpoint and instrumental in assuring whole- are rich source of nutrients for colonizing mold. Mold
someness in the daily meal. With the advent of urban contamination may occur during crop growth, harvest-
lifestyle, processed food became more popular which not ing, and storage (Magan & Aldred, 2006). The source of
just escalate the horizon of food processing industries but contamination can be a combination of air, dust, water,
imposed a huge responsibility for producing safe food. The soil, insects, birds, or rodents. Also, the infection can be
term “safe food” refers to food that is free from all kinds of spread through contaminated equipment and unsanitary
contaminants, viz. physical, chemical, and biological adul- handling (Liu, Galani Yamdeu, Gong & Orfila, 2020). Mold
terants (Korada et al., 2018). Among the three types of food microflorae of these crops are divided into two groups, viz.
contaminants, biological sources, such as microorganisms, field fungi and storage fungi. The occurrence of various
pose the greatest threat against food safety. Food spoilage fungal microflorae in cereal grains and nuts is given in
caused due to microorganism accounts approximately 42% Table 1. Some types of molds produce wide group of toxic
of the overall foodborne illnesses (Yeni, Acar, Polat, Soyer secondary metabolites during storage, called mycotoxins
& Alpas, 2014). There are several promoting factors for which are low molecular weight compounds (0.3–0.7 kDa)
microbial growth in food materials, but the existing mois- and thermally stable (Majeed, Khaneghah, Kadmi, Khan &
ture in the food available for chemical reactions decides Shariati, 2018). Fungal genera, including Fusarium, Peni-
the shelf-life of food and is commonly expressed in terms cillium, Aspergillus, and Alternaria, produce mycotoxins
of water activity (Labuza, 1980; Lenovich, 2017). Food having diverse potent pharmacological and toxic effects in
spoilage resulting due to microbial growth can be reduced humans and animals (Bailly & Oswald, 2013; Khaneghah,
by limiting the water activity at a lower value; however, a Fakhri, Gahruie, Niakousari & Sant’Ana, 2019). Among the
certain type of microorganism, that is molds, is immune secondary metabolites identified (>300) so far, the occur-
to withstand a water activity value as low as 0.7 (Beuchat, rence of aflatoxins (AFs), deoxynivalenol (DON), ochra-
1983; Troller, 2017). Molds are the principal reason for toxin A, zearalenone, patulin, fumonisin, and citrinin in
spoilage in durable foods, such as cereal grains and nuts cereal grains and nuts reserved notable attention among
(Sun, Su & Shan, 2017). the researchers (Wan, Chen & Rao, 2020). These myco-
Cereal grains are the most produced and consumed food toxins could also pose toxic effects on the kidney, liver,
materials around the globe (Panda & Shrivastava, 2019). and immune system, besides their mutagenic and car-
Larger production of cereal grains as compared to nuts cinogenic effects (Bhat, Rai & Karim, 2010; Heshmati,
is driven by multiple factors, such as agroclimatic condi- Zohrevand, Khaneghah, Nejad & Sant’Ana, 2017). There-
tions, pre and postharvest processing requisites, demand of fore, more than 100 countries established strict regulations
value-added products, nutritional contribution, purchas- on mycotoxins to prevent toxicity in humans and animals
ing capacity of the buyer, and so on. Despite the differ- (Lee & Ryu, 2017).
ence in their global production, both cereal and nuts in Detecting molds and their by-products becomes more
totality contribute significantly toward the economic and challenging for complex food materials analysis. Due to
nutritional wellbeing of the producer as well as the con- subsequent regulatory limits, mycotoxins analysis must
sumer (Awika, 2011). Nuts contain a variety of micronu- be sensitive enough to detect at very low concentra-
trients along with high amounts of proteins and unsatu- tions, down to about a few parts per billion (ppb).
rated fatty acids and hence highly recommended in cases of Over the years, various detection techniques have been
chronic degenerative diseases (Blomhoff, Carlsen, Ander- developed and reported. However, the demand for effi-
sen & Jacobs, 2006). With proven health benefits, nuts are cient and effective detection is significantly increasing for
becoming high in demand as a commercial crop. It is evi- commercial applications. Currently, microscopic examina-
dent that both cereal crops and nuts have their share in tion to plate counts (Gourama & Bullerman, 1995) and
a meal and have a very strong hold in the global market. chemical-based analytical techniques, viz. thin layer chro-
With the production growth, quality assurance became matography (Nicol, 1998), high-performance liquid chro-
more challenging; mold growth in cereal grains and nuts matography (Rahmani, Jinap & Soleimany, 2010), and
is not just a matter of trade barrier that leads to significant liquid chromatography-mass spectrometry (Zhang, Hu,
financial loss, and its associated impact on public health is Zhang & Li, 2016), are reference methods for molds
also an important issue to explore. and mycotoxins analysis. The counting method is dif-
Molds are the group of fungi described as micro- ficult in mold identification as it divides itself into
scopic single-celled organisms that thrive on food moisture tens of individuals during mycelia formation (Levasseur-
(Filtenborg, Frisvad & Thrane, 1996). Mold propagate Garcia, 2012). Sample preparation is the major step in the
TA B L E 1 Occurrence of mold species in cereal grains and nuts and their sources
Grain type Sources of contamination Molds References
Field fungi Storage fungi
Cereal grains Preharvest Birds, insect, and rodent Field fungi: Fusarium, Alternaria, Laca, Mousia, Dıáz, Webb, and
Rain infestation Cladosporium, Pandiella (2006)); Los, Ziuzina,
Pre-existing fungal spores Poor storage practice Helminthosporium, Penicillium and Bourke (2018)); Naresh
Harvesting Storage microflora (mostly Storage fungi: Magan, Hope, Cairns, and Aldred
Nondestructive detection of mold infection and mycotoxins
Contaminated harvesting tool spore forming) A. flavus and Penicillium (2003); Magan and Lacey (1988)
Improper handling
Harvesting after rainfall
Grain damage
Inefficient field drying
Transport
Improper handling
Pre-existing fungal residues in the
transportation vehicle
Nuts Field fungi: Danyluk, Harris, and Sperber
A. flavus, Rhizopus, Penicillium, (2007); Marín and Ramos (2016);
Alternaria, Cladosporium, Tournas, Niazi, and Kohn (2015)
Cunninghamella
Storage fungi:
A. flavus, Penicillium, Rhizopus
3
above-mentioned techniques and estimation of myco- techniques and future scope of its application in cereal
toxin levels is done in representative samples. How- grain and nut industries are discussed.
ever, few kernels of a lot may heavily be infested
and there is a possibility that those kernels may not
be selected for analysis, resulting in ambiguous results 2 FUNDAMENTALS OF OPTICAL
(Chavez, Cheng & Stasiewicz, 2020). The analysis time IMAGING AND SPECTROSCOPIC
of these analytical methods is very high for an individ- TECHNIQUES FOR NONDESTRUCTIVE
ual sample. Rapid diagnostic methods like Enzyme-linked MOLD AND MYCOTOXINS DETECTION
Immunosorbent Assay (ELISA), membrane flow tech-
nique, lateral flow immunoassay techniques with quan- This section provides the brief overview of the principle
tum dots perform quantification close to cutoff level (Krska of operation of the imaging and spectroscopic techniques,
& Molinelli, 2009; Yan et al., 2019). These immunoas- data acquisition techniques, machine learning algorithm,
say kits are for single-use, which can increase the costs and chemometrics. This will be helpful in understanding
of bulk screening and cannot be used for inline detec- the data acquisition and analysis methods used for rapid
tion of mycotoxins. Moreover, the narrow sensitivity of and nondestructive detection of molds and mycotoxins and
the antibodies poses a constraint over the use of the is followed by a separate section for the application of these
ELISA method, resulting in a limited detection range technologies with detailed discussion.
(Turner, Subrahmanyam & Piletsky, 2009). Moreover,
the diversity and abundance of molds and mycotox-
ins demand a higher number of samples for analysis, 2.1 Digital color imaging
which leads to wastage of the food materials. Therefore,
rapid and nondestructive tests are attracting the research Image analysis is an important nondestructive and rapid
attention. method for analyzing food quality. Visual inspection
Over the last two decades, application of computer through human eyes is labor intensive and suffers
vision and nondestructive spectroscopic techniques in from inconsistency; hence, computer vision technique is
food industries for quality and safety analysis has been immerged as a favorite alternative. The fundamental units
thriving due to its potential of rapid inline identifica- of a computer vision system consist of a lighting device,
tion and quantification of molds and mycotoxins. These a camera with optical lens, an image capture interface,
techniques are advantageous over traditional analytical and image processing software (Ma et al., 2019; Saldaña,
techniques as they require minimal sample preparation Siche, Luján & Quevedo, 2013). The quality of the cap-
and no involvement of hazardous chemicals (Kandpal & tured image is greatly influenced by the performance of
Cho, 2014). Digital color imaging, X-ray imaging, near- two components: the camera and the illumination sys-
infrared (NIR) spectroscopy, fluorescence spectroscopy, tem. As good lighting can reduce the effect of shadow,
NIR hyperspectral imaging (HSI) are the main techniques reflection, and noise; and increase the visibility of spe-
for nondestructive detection of microbial contamination in cific features (Vithu & Moses, 2016). Image sensors, such
food products (Liu et al., 2018; Tripathi & Mishra, 2009). as charge-coupled device (CCD) or complementary metal-
Although there are several reviews on the application of oxide-semiconductor (CMOS), applied photovoltaic effect
optical imaging and spectroscopic techniques on quality to transform incoming light sources into an array of dig-
and safety evaluation of food products (Dale et al., 2013; ital signals (Hornberg, 2017). Thus, the captured image
Kaavya et al., 2020; Liu, Pu & Sun, 2017; Ma, Sun, Pu, forms a matrix of numeric values of which each matrix
Cheng & Wei, 2019; Sendin, Williams & Manley, 2018; entry is called a picture value or pixel. The spatial resolu-
Singh & Jayas, 2011; Su, He & Sun, 2017; Wu & Sun, 2013b), tion of a camera is a function of the number of pixels and
there are limited reviews on mold and mycotoxin detec- their size relative to the projected image (Gunasekaran,
tion in cereal grains and nuts. Orina, Manley, and Williams 1996). Once an image has been generated, there are sev-
(2017a) reviewed nondestructive techniques for the detec- eral stages of processing the image for recognition of use-
tion of fungal infection in cereal grains and Wu, Xie, and ful object features, which involves various steps: image
Xu (2018) reviewed imaging and spectroscopic techniques acquisition, preprocessing, segmentation, representation,
for the detection of toxigenic fungi and AFs in nuts and and interpretation. Image preprocessing is required before
dried fruits. Therefore, current review focuses not only extraction of information, is the process of conditioning to
on applications but also recent advances in chemometric enhance the quality of acquired image by noise reduction,
methods in the field of optical and spectroscopic methods gray-level correction and correction of defocusing, and
for detection of toxigenic molds and mycotoxins in cereal so on (Minz, Singh & Sawhney, 2017). The preprocessed
grains and nuts. Also, the challenges associated with these images are then transformed into a matrix of numbers
Nondestructive detection of mold infection and mycotoxins 5
for quantization and feature extraction (da Silva & Men- mycotoxin analysis, including near and mid-infrared
donca, 2005). The next step of image analysis is segmen- (MIR) spectroscopy and HSI. Although some studies
tation for separating the image into different regions by have reported the application of Raman spectroscopy
delineating the area of pixels having a similar background and Fourier transform infrared spectroscopy for detec-
texture. The segmented images are then used for the clas- tion in mold infestation, both methods are destructive.
sification of various objects using a clustering algorithm Therefore, these two methods are not discussed in this
(Tan, 2016). manuscript.
Most of the researchers applied vibrational spectroscopy
for nondestructive detection of mold and mycotoxins in
2.2 X-ray imaging cereal grains and nuts which are mainly focused on NIR
spectroscopy. Spectral region that extends from 750 to 2500
Machine vision with visible light can detect defects on the nm wavelength generally refers to NIR region (Panikut-
surface of the grains and nuts, but many times the sur- tira & O’Donnell, 2018) in which absorption bands sig-
face is intact, and the degradation starts from the interior nify mainly overtones and combinations of fundamental
due to the presence of hidden fungal infestation, which vibrations. NIR instrumentation is an emerging method
cannot be detected by visible cameras. Therefore, X-ray for rapid analyses and flexibility in adapting to different
inspection technique has become increasingly common sample states and is of wide acceptance in many indus-
because of its distinct advantage over other imaging tech- tries. The main parts of NIR spectrometric (NIRS) sys-
niques as it forms an internal image of the object by pen- tem are light source, light analyzer or light isolator, inte-
etrating through it. X-rays are a type of electromagnetic grating sphere, detectors, and sampling devices (Coates,
energy with short wavelengths between 30 petahertz and 1998). NIRS instruments are classified according to light
30 exahertz and high energies in the range of 100 eV–100 separation mechanism, which is crucial to spectrometer
keV (Di Benedetto, Perricone & Corbo, 2010). In earlier design. The NIR instrumentation can be categorized into
times, the most common X-ray detectors used a sheet of diode-based instruments, optical filter instruments, elec-
paper coated with barium platinocyanide. However, inline tronically tunable filter instruments, grating monochro-
inspection requires the use of sensible semiconductors mator instruments, and Fourier transform NIR instru-
(Haff & Toyofuku, 2008). When a food product is exposed ments (Williams & Norris, 1987). Transmittance, diffuse
to X-rays, the energy loses as it penetrates the object. If reflectance, interactance, and/or transflectance are the
an inconsistent area is encountered inside the food, X-ray basic measurements used to detect absorbed NIR radia-
loses further energy than the surrounded area, the hori- tion by the sample, where the suitable acquisition mode
zontal remaining rays are captured by the sensor which is dependent on the sample type and the constituent
converts it into a 2D gray-scale image of the food inte- and/or property being analyzed (Osborne, Fearn & Hin-
rior (Kotwaliwale et al., 2014). This image can be processed dle, 1993). Samples for imaging and spectroscopic tech-
by software to establish if there are internal defects in the niques are usually solids and liquids, which can be per-
food. formed either in laboratory or at a production line. Spe-
cialized sample cups and liquid cells are used for plac-
ing the samples to meet the measurement demand. The
2.3 Vibrational spectroscopy infrared light source penetrates through sample present
in the sampling cell, which is then diffusely reflected and
Complex food material consists of atoms that are in collected by the integrating sphere, and is detected by the
constant motion about their equilibrium positions, called detector (Wang & Paliwal, 2007). NIR instruments used
vibration and have well-defined frequency and energy in commercial applications are integrated with number
(Nawaz et al., 2019). The vibration patterns are character- of interference filters to the incident light for produc-
istics of every material and can be used to measure any tion of particular wavelengths, useful in simple and inex-
changes in chemical alterations. Molecular vibrations can pensive design of spectrophotometers (Gat, 2000). Optical
be of different types, viz. symmetric stretching, asymmet- fiber bundles, also called the dielectric waveguides cou-
ric stretching, bending, twisting, rocking, and wagging pled with microprocessors, enable the NIR spectropho-
(Harris & Bertolucci, 1989). Vibrational spectroscopy is a tometers capture, monitor, and analyze several simultane-
technique used to measure the changes in the molecular ous measurements in real time (Noiseux, Long, Cournoyer
vibration of a material resulting from any form of energy & Vernon, 2004). Besides NIRS, MIR spectroscopy can
absorption (Li-Chan, Ismail, Sedman, & Van de Voort, also be applied for successful detection of mold and myco-
2006). Various vibrational spectroscopic techniques have toxin detection as reported by several researchers (De
been successfully applied for the evaluation of mold and Girolamo et al., 2019).
2.4 Fluorescence spectroscopy 2.6 Image/spectra processing and

analysis
Fluorescent spectroscopy exploits the temporary absorp-
tion of electromagnetic wavelengths from ultraviolet or This section provides an overview of the steps involved in
visible light spectrum by some fluorescent molecules of the processing of images and/or spectra. Figure 1 shows the
the food material and its subsequent emission of higher steps followed using various optical detection techniques,
wavelength and at a lower energy level (Karoui & Blecker, explained in the previous section. Sample preparation is
2011). The emitted light is directed toward a filter and usually a bottleneck for biological materials due to the exis-
onto a detector for measurement and identification of tence of the inherent variation among samples. Sampling
the molecule or changes in the molecule. Some of the should be such that a clear boundary should be established
biomolecules possess intrinsic fluorescence characteris- between the sample groups. Another most important step
tics called fluorophores (Karoui & Blecker, 2011). Fluores- is calibration of the image/spectra capturing device (cam-
cence spectroscopy offers several inherent advantages for era/spectrometer) for maximum reflectance value and res-
the detection of mycotoxins, as it is highly sensitive than olution. Digital color and X-ray techniques involve image
other spectrophotometric techniques, allowing to quantify processing steps after capturing the image, while vibra-
components present in nanomolar ranges (Kulmyrzaev, tional spectroscopy and fluorescence spectroscopy involve
Karoui, De Baerdemaeker & Dufour, 2007). spectral processing steps after scanning of the samples. HSI
technique consists of both image and spectral processing
steps for the development of the calibration model. Details
2.5 Hyperspectral imaging of the image and spectral processing and machine learning
techniques for robust model development are explained in
HSI is the integration of both imaging and spectroscopy the following sections.
(Wu & Sun, 2013a), and can be measured with either vibra-
tional mode (transmittance, diffuse reflectance, interac-
tance, and/or transflectance) or fluorescence mode. Hyper- 2.6.1 Image processing
spectral images are numerous spatial image planes of the
same object at different wavelengths collected by a hyper- The major purpose of image processing is to improve
spectral sensor and create a 3-D data cube called “hyper- the quality of images, to identify desired objects, and
cube,” which represents the composition and appearance automate the management of objects. It is usually per-
of the food material (Feng & Sun, 2012; Huang, Liu & formed following several steps such as (1) preprocessing,
Ngadi, 2014). Each spectral pixel in the hypercube indi- (2) image segmentation, (3) feature extraction based on
cates the spectral signature of the corresponding spatial image representation and description, and (4) classifica-
area, which defines the ability of the food to absorb the tion of objects (Paul et al., 2020). Image preprocessing is
excited light, representing the inherent chemical property used to improve the image quality by suppressing unde-
(Wu & Sun, 2013b). There are three methods for obtaining sired distortions, such as background subtraction, noise
a hyperspectral image that are point scan, line scan, and removal, polygon shape creation, smoothing, histogram
area scan. The line-scan method or pushbroom method operations to enhance the objects, and so on (Paul et al.,
is generally used in food industries as food commodities 2020). Image segmentation is a process of dividing an
use wavelength dispersive system to collect the spectra at image into different regions by grouping them into homo-
a time and gradually construct the hypercube (Qin, Chao, geneous pixels, which is not trivial because of noise in the
Kim, Lu & Burks, 2013). Area scan system uses a mechan- background. Traditional methods have relied on thresh-
ical wavelength filter wheel to acquire entire spatial image olding or morphological operations, which are only effi-
of the sample at one wavelength and combines the images cient in less complicated objects and plain backgrounds.
at different wavelengths to form a hypercube. Due to dif- For this reason, recently, machine learning techniques
ficulties in data processing in HSI in the industrial online have been applied to discriminate the objects, which are
application, a simplified imaging version called multispec- called region of interest. Afterward, various features can
tral imaging system has been developed using area scn be extracted to represent the characteristics of objects,
technique, which is restricted to the limited number of pre- such as color values, shape properties, principal compo-
defined wavelengths. Hyperspectral images are considered nent analysis (PCA), fast Fourier transform, histogram of
to be the fundamental dataset and for a particular applica- wavelength, gray-level co-occurrence matrix, and so on
tion, multispectral imaging can be used to find the opti- (Wan, Lin & Chiou, 2002). Based on the extracted features,
mum waveband. machine learning and deep learning techniques have been
FIGURE 1 Various processing steps involved in optical and spectrometric techniques
used for the classification of objects. Deep learning can broad categories based on their application in fungi and
automatically train deep multiple layers by analyzing red, mycotoxin detection are (1) supervised learning and (2)
green, and blue (RGB) color images or input features. unsupervised learning (Jordan & Mitchell, 2015), as pre-
It has also shown significant advance in food industry sented in Figure 2. Unsupervised learning algorithm cre-
with outstanding performance handling a large amount ates a model that takes an unlabeled feature vector X as
of data (Kamilaris & Prenafeta-Boldú, 2018; Kawano & input and either transforms it into another vector or into a
Yanai, 2014). value that can be used to solve a practical problem. Clus-
tering, k-means, hierarchical cluster analysis (HCA), PCA,
and kernel PCA are the examples of unsupervised learning.
2.6.2 Machine learning Supervised classification strategy involves labeled datasets
for training the model, and subsequently classifies the
Machine learning is concerned with the development of unknown dataset. K-nearest neighbors, linear regression,
techniques and methods that provides systems the ability logistic regression, support vector machine (SVM), deci-
to automatically learn from the raw data that has been fed sion tree, random forests, and artificial neural networks
into it and improve from the experience (Mitchell, 1997). (ANNs) are typical examples of supervised learning mod-
The machine learning techniques are grouped into two els (Alpaydin, 2020).
FIGURE 2 Various machine learning techniques
PCA is a popular data compression (dimesnon reduc- of decision trees at training time and make output as the
tion) technique in which data are orthogonally trans- mean class. The SVM is a popular classification technique,
formed into a new coordinate system in such a way that the which produces high accuracy with moderate computation
highest variance of the data lies in the first few coordinates, power. The basic idea of SVM is to find a hyperplane which
namely, principal components (PCs) (Wold, Esbensen & separates the d-dimensional data perfectly into two classes
Geladi, 1987). The k-means clustering involves dividing (Boswell, 2002). The ANNs employ artificial neurons that
samples into k clusters in such a way that the squared error combine the neuron input with their internal activation
between the empirical mean of a cluster and the points function and an optional threshold to produce a desired
in the cluster is minimized, whereas HCA classifies the output. The training of ANNs is done by the search of the
samples in clusters by calculating distance between each weight and thresholds of the activation function that min-
sample group and form hierarchy by merging most similar imizes the observed error (Hopfield, 1988).
pair of clusters successively (Srivastava, Mishra & Mishra,
2018).
K-nearest-neighbor (KNN) classification is a fundamen- 2.6.3 Chemometrics for spectral analysis
tal supervised classification algorithm, which learns the
categories from the input data and groups the data based Chemometrics can be defined as the application of math-
on a similarity measure (Peterson, 2009). For prediction of ematical and statistical methods to extract and analyze
the dependent variables using a given set of independent the information consisted of a chemical measurement pro-
variables, regression algorithms are used. Linear regres- cess (Workman, Mobley, Kowalski & Bro, 1996). Spectral
sion is used to predict the continuous dependent vari- processing involves the reduction of dimensionality with
able, whereas logistic regression is used to predict the retention of meaningful information and establishment of
categorical dependent variables (Grimm & Yarnold, 1995; the mathematical relationship between the spectra and ref-
Wright, 1995). More complex classification problems are erence data for classification and quantification of impor-
solved by a constructive decision tree algorithm by which tant quality parameters (Amigo, Martí & Gowen, 2013).
an unknown sample is classified into a class using one Optical properties of the food materials, viz. absorbance,
or several decision functions consecutively. Moreover, ran- transmittance, emission at various wavelengths, consti-
dom forest algorithms operate by constructing a multitude tute a spectrum having all the chemical information of
the object. Chemometrics operation involves selection of to find the approximate location of defect characterized
sensitive wavelengths and the weighting vectors or beta- by “valley-like” shape curve in the presence of fungus.
coefficients for each wavelength used for the calibration. He found 74% classification accuracy by using red chan-
Algorithm developers often proceed through three stages nel of the image and considering germ color as white and
of the algorithm when dealing with new data, that are pre- infected region color as dark for deciding on classification.
processing, model construction, and validation. Chemo- To improve the discrimination accuracy, Yorulmaz, Pear-
metrics models correlate the image and/or spectra with the son, and Çetin (2011) extracted cepstral features (inverse
desired properties to detect the anomaly in the food mate- Fourier transform) and applied SVM to classify Penicil-
rials. Chemometrics is a subset of the machine learning lium infected blue-eye-damaged popcorn. Their approach
domain, mainly used to answer linear analytical chemistry reduced the size of the data and emphasized the impor-
problems. tant frequencies by giving higher weight, so that the con-
The common practice of correlating spectrophotomet- tribution of noise to the final decision was reduced. The
ric data to concentration has been carried out for several resultant SVM found has a recognition rate of 93.3% for
algorithms using Beer’s law combined with multiple linear damaged and healthy popcorn kernels. Later, Yorulmaz,
regression, PCs regression, and partial least square regres- Pearson, and Çetin (2012) extended the previous methodol-
sion (PLSR). Quantitative relationship between spectra ogy with the employment of covariance-matrix-based fea-
with the measured properties refers to regression, which tures for detection of blue-eye-damage in corn specially
includes linear and nonlinear regression. PLSR is the cultivated for popcorn and achieved an overall recogni-
basic chemometric technique for establishing a linear rela- tion rate of 96.5%. Singh, Jayas, Paliwal, and White (2012)
tionship between variation in independent variables and used a total of 179 color and textural features of fungal-
response variables by projecting them to a new space for infected and healthy wheat kernels for detection of fun-
solving multicollinearity problem in the dataset (Wold, gal damage due to Penicillium spp. (blue) and Aspergillus
Sjöström & Eriksson, 2001). Response variables are usually spp. A two-way linear and quadratic discriminant analy-
analyzed by standard analytical procedures and the results sis (Linear Discriminant Analysis (LDA) and Quantitative
are considered as ground truth for the model development. Descriptive Analysis (QDA)) and Mahalanobis discrimi-
Several other regression models alone and combined with nant classifiers were used to identify the fungal damage
other spectral-spatial features extracted through machine with 92.7% (7.2 std dev), 90.0% (11.4 std dev), and 85.7% (13.6
learning algorithms are also currently under investigation, std dev) accuracy, respectively.
viz. SVM and least square-SVM (LS-SVM), and deep neural Ropelewska (2018) successfully identified Fusarium
network algorithms (Caporaso, Whitworth & Fisk, 2018). infection in barley kernels with discrimination efficiency
of 95–100% using a discrimination model based on 10 tex-
ture features. It was reported a 100% accuracy in clas-
3 APPLICATION OF DIGITAL COLOR sification of the healthy and infested kernel in case of
IMAGING FOR MOLD AND MYCOTOXINS CIELAB color model with Naive Bayes, LDA, Lazy (IBk),
DETECTION and multiclass classifier. Moreover, the least accuracy was
observed if the XYZ color model and with texture extracted
3.1 Cereal grains by Ranker method are employed to train the multiclass
classifier. Nicolau, Pimentel, Tibola, Fernandes, and Pavan
3.1.1 Detection of mold (2018) applied transfer learning for a pretrained deep neu-
ral network consisting of 120,00 images for detection of
Color cameras became more popular in commercial appli- Fusarium damaged wheat kernels and achieved an overall
cations and gained greater attention of researchers. From accuracy of 94.7%. They used new independent image sets
the captured images, a series of morphological, color, tex- for validation of the classification model and reported 20%
tural, and wavelet features can be extracted. Subsequently, of misclassification, and explained the error resulted due to
classification algorithms can be applied to identify fun- the presence of a small number of Fusarium-damaged ker-
gal damaged kernels from normal kernels. In the same nels and prevalence of normal wheat in the initial training
way, these features can be employed to predict the degree dataset.
of infection (Jayas & Singh, 2012). Several researchers
attempted to detect mold infection in cereal grains in the
last decade and found promising results, which have been 3.1.2 Detection of mycotoxins
summarized in Table 2. Pearson (2009) developed a real-
time detection of blue-eye damage in popcorn kernels DON is a notorious mycotoxin for stored cereal grains, pro-
using CMOS color image sensor and his approach was duced by Fusarium graminearum that need to be detected
10
TA B L E 2 Summary of research works on the application of digital color imaging for detection of mold infection in cereal grains
Grain type Fungi Imaging set up details Reference parameters Chemometrics Accuracy References
Wheat Mildew defect 3-chip CCD color camera 68 Morphological and 78 Stepwise discriminant 99% Luo, Jayas, and Symons
with zoom lens of color features analysis to select best (1999)
10–120 mm focal features and k-NN
length classifier
Wheat Fusarium culmorum CCD camera with 4 Kernels weight per spike Correlation 90% Wiwart, Koczowska, and
and F. avenaceum halogen bulbs of and one thousand Borusiewicz (2001)
35000 K color kernels weight (TKW),
temperature and the values of H S & I
Corn Blue–eye mold Digital color camera and 12 features related to area Probabilistic neural 78% Steenhoek, Misra, Jr, C.,
damage two Ultralume U–tube representing sound network (PNN) and Bern (2001)
(Penicillium) fluorescent lamps germ, hard starch, and
soft starch
Wheat Fusarium and Microscope and a digital DON value Decision tree an NA Perner and Günther
Cladosporium color camera entropy-based (2005)
criterion
Wheat Fusarium Digital reflex camera RGB, HSL color descriptor PCA, HCA 85% Jirsa and Polišenská
and shape descriptor, (2014)
DON content
Wheat Fusarium culmorum Digital color camera with TKW and the shape HCA involving Correlation Suchowilska and Wiwart
24-bit images at a descriptors of the kernel Euclidean coefficient r = (2006)
resolution of 300 dpi image: area, perimeter, distances and Ward’s 0.716
length, and width linkage
Corn A. flavus, Bipolaris Color camera capturing LDA and multilayer 75% Tallada, Wicklow,
zeicola, Diplodia single kernel of 640 × perceptron (MLP) Pearson, and
maydis, Fusarium, 480 neural network Armstrong (2011)
Penicillium, and
Trichoderma
Wheat Fusarium Digital single-lens reflex Visual separation of R2 = 0.8 Maloney et al. (2014)
camera diseased kernels, color
features
Wheat Fusarium Digital color camera with Color features Mask RCNN R2 = 0.920 Qiu et al. (2019)
high field view
Digital color camera with Texture features: histogram, BayesNet, lazy IB1, Classification Zapotoczny et al. (2020)
polarized light gradient, run length multiclass classifier, accuracy for
illumination matrix, co-occurrence meta classification via infested kernel =
matrix, autoregressive regression, tree LMT, 100%
model, and Haar wavelet progressive DA, and
ANN
at their early growth stages to protect the grain from ous types of fungal infections in corn kernels. The result
spoilage. Orlandi, Calvini, Foca, and Ulrici (2018) devel- of the study indicated that the mean X-ray intensity of
oped an automatic quantification method of defective fungal-infected kernels was significantly lower than that
maize kernels which are more prevalent to DON contam- of undamaged kernels as the fungal-infected area absorbed
ination, using digital image analysis of 332 acquired RGB less X-ray energy. They performed stepwise discriminant
images of 83 mixtures consisting of different amounts of analysis using kernel thickness, weight, and length of the
infected kernels. The RGB digital images of the analyzed kernels and found average classification accuracy of 91.5%.
samples in one-dimensional signals (color grams) of the Narvankar, Singh, Jayas, and White (2009) investigated the
images were utilized to develop PLSR models for predic- potential of X-ray imaging for the detection of common
tion of percentage defective kernels with regression coef- storage fungi of wheat grain, namely A. niger, A. glau-
ficient (R2 ) of 0.975 and a root mean square error value of cus, and Penicillium spp., using 34 image features, viz.
2.6%. Though the authors did not identify or quantify the maximum, minimum, mean, median, variance, standard
DON directly by this method, however, the defective ker- deviation, and 28 gray-level co-occurrence matrix feature.
nels were correlated with the presence of DON by ELISA They applied statistical discriminant classifiers (linear,
analysis. It will be helpful for the screening of large maize quadratic, and Mahalanobis) and back-propagation neu-
batches for a preliminary estimation of the degree of DON ral network (BPNN) classifier and found 92.2–98.9% clas-
contamination in maize. Visible color cameras are only sification accuracy from two-class Mahalanobis discrimi-
helpful for the detection of the presence of fungi over the nant classifier. BPNN also identified healthy kernels with
surface of the grains; however, no research results were more than 80% accuracy. No researchers have reported the
reported for direct screening of mycotoxin contamination detection of any mycotoxin by using X-ray imaging, which
in cereal grains. Moreover, it was proclaimed that AF might be due to little contribution of mycotoxin contami-
shows bright greenish-yellow fluorescence (BGYF) under nation on variation in the kernel density.
ultraviolet (UV) light at a wavelength of 365 nm (Wick- The internal structure of Fusarium verticillioides-
low, 1999). Soemantri and Diyono (2019) applied this infor- infected maize was examined by Orina, Manley, and
mation for rapid detection of AF in corn using UV sens- Williams (2017b) using X-ray micro-contrast tomography
ing digital camera with 5-watt UV lamps. They built ANN (CT) to obtain information regarding undergoing physical
architecture considering 10 input parameters, viz. inten- and chemical changes due to fungal infection. Results of
sity of red, green, blue, the color value, area, circumfer- the study indicated loss of kernel integrity and decrease in
ence, length, and hue, saturation, intensity. The above kernel density as the fungal growth proceeded, resulting in
study aimed to classify various AF concentrations rang- low gray value intensities of the images (Figure 3). Thus,
ing between 1-5 ppb based on the area of AF contamina- Fusarium growth in maize was successfully detected by
tion. The detection efficiency of their system obtained from them by evaluating the shifting of gray value histograms
training and validation was found to be 100% and 99%, of control and infested kernel.
respectively. However, it is now clear that fluorescence
(same excitation and emission) is may be due to other fun-
gal metabolites like kojic acid (Kalkan, Güneş, Durmuş & 4.2 Nuts
Kuşçu, 2014). Authors in this study have not mentioned the
source of AF in the corn kernel, whether they are naturally Nuts are known to foster the growth of pathogenic fungus
infected or artificially infected with standard AF solutions. in the field and during storage due to increased humid-
Hence, a more detailed investigation is required to confirm ity in the environment. Mold infection could compro-
the origin of fluorescence signal for detection of AF. mise the quality of nuts and cause serious health con-
cerns for the consumers, which attracted research atten-
tion toward rapid and nondestructive detection of mold
4 APPLICATION OF X-RAY IMAGING infection both from inside and outside of the nuts. Myco-
FOR MOLD AND MYCOTOXINS logical quality of Brazil nuts was evaluated by Franco,
DETECTION Yokaichiya, Kardjilov, and de Oliveira Ferraz (2015) using
X-ray phase-contrast imaging and X-ray micro CT, which
4.1 Cereal grains authors believed to be useful for the separation of infested
nuts to avoid contamination during the automatic shelling
The differences in the X-ray attenuation resulted from vari- process. Results of the study reported that X-ray micro
ation in the atomic composition of the grains can be used to CT technique was able to identify small dots in several
detect fungal infection. Pearson and Wicklow (2006) inves- image slices, which could be attributed to chemical decom-
tigated the applicability of X-ray images to detect 11 vari- position due to fungal infection. They concluded that
F I G U R E 3 Comparison of kernel density changes in 2D X-ray nuts images of (a) control and (b) fungal-infected maize kernels
(approximately same image slice) on day 1 and 14, respectively (Orina et al., 2017a)
both the X-ray phase-contrast imaging and X-ray micro (2012) evaluated the X-ray absorption coefficients of fun-
CT techniques can be applied for automatic sorting for gal infected and healthy nut kernels for possible detec-
Brazil nuts by discriminating morphological features of tion of fungal defects. They noticed a variation in the X-
the nuts. Yanniotis, Proshlyakov, Revithi, Georgiadou, and ray absorption coefficients, indicating change in the speci-
Blahovec (2011) investigated the potential of X-ray imag- men density which might be linked with the compositional
ing for detection of fungal necrotic spot in pistachio by changes as a result ofowing to fungal damage. Moreover,
comparing the X-ray images with the color pictures of Proshlyakov, Yanniotis, and Blahovec (2013) supported the
the same nut after splitting. They captured X-ray images same study indicating that increase in moisture content
of 100 randomly selected intact pistachios and analyzed from 9% to 30% led to increase in the absorption coefficient,
the AF content of the nuts to correlate with the level which can fuzzy the detection parameters. Therefore,
of fungal infection. Results of the study indicated that the researchers suggested keeping the moisture content
pistachios with kernel necrotic spots contained about 60 uniform for potential detection of fungal damage using
times more AF than healthy ones, which can be possi- X-ray images. Mycotoxin contamination can indirectly be
bly screened by X-ray imaging. Further, Georgiadou et al. correlated with the detection of fungal species in cereal
grains and nuts; however, no study was reported regarding the associate wavelength for mold detection. The classifica-
the direct determination of mycotoxin content using X-ray tion accuracy of the partial least square discriminant anal-
imaging or X-ray micro CT. ysis (PLSDA) model developed by them was 98% and it was
improved with the application of second and third deriva-
tives of the spectra. Tao et al. (2019) investigated the poten-
5 APPLICATION OF VIBRATIONAL tial of NIR spectroscopy with two spectral ranges 400–1070
SPECTROSCOPY FOR MOLD AND nm and 1120–2470 nm for detection of A. flavus (strains
MYCOTOXINS DETECTION AF13 and AF 38) contamination on corn kernels using
PLSDA and PLSR chemometric techniques. They utilized
Cereals are subject to infection by mycotoxigenic molds, a total of 1321 and 2701 spectral data variables when estab-
such as Fusarium, which is a major food safety con- lishing the classification models over the spectral range of
cern worldwide. It was estimated that about 25% of the I and II, respectively. They applied standard normal variate
world’s food crops are infected by mycotoxigenic molds, (SNV) preprocessing to the obtained spectra and observed
out of which cereal grains and nuts are affected the most sensitive wavelengths for AF determination are 434, 440,
(Nations, 2015). To detect the molds nondestructively and 764, 890, and 1015 nm in spectral range I and 1175, 1350,
rapidly, many researchers investigated and reported the 1390, 1595, 1605, 2060, 2347, 2365, and 2425 nm in spec-
application of NIR and Fourier transform near infrared tral range II. The wavelength 1175 nm corresponds to the
(FTNIR) spectroscopy for detection of pathogenic molds second overtone region of CH2 and CH3 and 2347 and
and mycotoxins (Jia et al., 2020). The research methodolo- 2365 nm corresponds to the combination band region of
gies and outcomes are summarized in Table 3. Moreover, CH, CH2 , and CH3 . The promising results could be corre-
advance detection techniques have been updated continu- lated with the information that AF is an aromatic ketone
ously as reported in the following sections. and has a cyclic epoxide structure. Based on the above fact
an overall accuracy of 97.78% was achieved through com-
petitive adaptive reweighted sampling (CARS) and PLSDA
5.1 Cereal grains model. Shen et al. (2019) reported the application of vis-
ible/NIR spectroscopy for online detection of Fusarium
5.1.1 Detection of mold and A. flavus and came up with promising results indi-
cating high classification accuracy of 91.7% for classifica-
The fungal infection causes a change in the chemical com- tion of different strains of fungal-infected samples. They
position of the grains resulting in substantial light scat- also presented goodness of fit of PLSR model for pre-
tering effects, which can be detected by vibrational spec- diction of colony counts having R2 values of 0.890 and
troscopy with high accuracy. Falade, Sultanbawa, Fletcher, root mean square error of prediction (RMSEP) values of
and Fox (2017) used NIR spectroscopy in the wave- 0.369 log CFU/g. Later, Shen et al. (2020) investigated the
length range of 800–2600 nm to identify kernels of milk-, feasibility of integrating NIR spectroscopy with computer
dough-, and dent-stage maturities with four doses of vision for more accurate prediction of fungal contamina-
Aspergillus contamination. They observed distinct peak tion in maize grains. They collected the spectroscopic and
differences at 1400 and 1940 nm, which are associated with image information of both contaminated and control sam-
water and carbohydrate-associated changes caused by fun- ples at a speed of 0.15 m/s. PCA indicated that infection-
gal deterioration. Spectral data obtained by them were pre- level detection and fungal strain identification were more
processed using Savitzky–Golay (1d-SG) transformation accurately accomplished by spectroscopic method and
and a regression model was developed to separate the con- computer vision techniques, respectively. Moreover, they
taminated and noncontaminated kernels. The results of developed a PLSR model using fusion features, which pre-
their study showed high accuracy in classification with R2 dicts the colony count well with R2 value of 0.914 and
value of 0.88 and root mean square error of 0.15. Fusarium RMSEP value of 0.617 CFU/g.
is a field fungus common in grains that reduces the yield
of barley and wheat by turning them into empty grains
and the disease is also called as Fusarium head blight. Lim 5.1.2 Detection of mycotoxins
et al. (2017) developed a prediction model for detection of
Fusarium-infected hulled barley, naked barley, and wheat Scab damage in wheat in damp and cool conditions favors
samples using NIR reflectance spectroscopic technique the growth of mold F. graminearum, which can cause ker-
with a wavelength range of 1175–2170 nm. They found max- nels to appear dull, lifeless, or chalky and may produce the
imum variation in the rising peak at 1575 nm using 1236 toxin DON. The presence of scab and DON or ergosterol
number of barley spectra, which might be considered as in wheat adversely affects flour ash, flour color, glutenin
14
TA B L E 3 Summary of research works on the application of vibrational spectroscopy for detection of mold and mycotoxin contamination in cereal grains
Wavelength/ Actual detection
Fungi/ wavenumber No. of Response according to our
Grain mycotoxin range samples variables analysis Chemometrics Accuracy Error References
Fungal infection
Maize Fusarium FTNIR: 650 and – Ergosterol and Fungal constituents PCA and LDA R2 of DON = RMSECV Kos,
4500 cm–1 DON content and change in 0.991 (µg/kg) for Lohninger,
macromolecules R2 of ergosterol DON and
= 0.938 reference: Krska
165.0 and (2003)
ergosterol:
428.9
Maize Fusarium NIR: 400–1100 nm 280 Ergosterol and Fungal constituents Modified PLSR R2 of maize SECV (%) for Berardo
verticillioides fumonisin B1 and change in kernels = 0.75 maize et al.
macromolecules R2 of maize kernels = (2005)
meals = 0.79 7.43; maize
meals =
10.95
Maize Penicillium NIR: 904–1685 nm 864 LDA, MLP ANN LDA: 96% RMSECV Tallada
uninfested (µg/kg) for et al.
versus 74% durum: (2011)
infested 470;
kernels common:
MLPANN: 516;
92% durum+common:
uninfested 555
versus 91%
infested
Corn ears Fusarium NIR: 400–2500 nm 600 Change in grain PCA, soft Healthy grain: – Draganova,
Healthy/diseased chemical independent 99.89% Daskalov,
sample class composition due modeling of Infested grain: and
to fungal infection class analogy 93.7% Tsonev
(SIMCA), (2010)
kNN, and
probabilistic
neural
network
(BPNN)
(Continues)
TA B L E 3 (Continued)
Rice Yellow-green A. NIR: 950–1650 nm 106 Total fungal Fungal constituents PCA and PLSR Correlation SEP = 4.613
flavus infection (%) (chitin, β-glucan) coefficient (r) Sirisomboon,
and change in = 0.437 Put-
macromolecules thang,
of grain and Siri-
somboon
(2013)
Paddy Mold colonies NIR: 918–1045 nm 121 Number of Change in grain Multiple linear Correlation SEP = 33,393 Qiang et al.
rice mold colony chemical regression coefficient: number of (2014)
(colony composition due (MLR) 0.897 colonies
forming unit to fungal infection
[cfu]/ml)
Wheat Fusarium NIR: 1100–1700 nm – 16 mycotoxins Change in grain Pearson’s Positive – Kautzman,
chemical correlation correlation Wick-
composition due between strom,
to fungal infection Fusarium and Scott
damaged (2015)
kernel and
DON = 0.90
Maize A. flavus NIR: 800–2700 nm 180 Low, medium, Water and PCA and R2 = 0.88 Root mean Falade
and high carbohydrate- nonlinear- square et al.
fungal associated PLSR error (2017)
infection changes caused by (RMSE) =
fungal 0.15
deterioration
(Continues)
15
16
2
Hulled Fusarium NIR: 1175–2170 nm 515 Presence of Change in grain PLSR R = 0.944 SEP = 0.108 Lim et al.
barley fusarium chemical (2017)
colony composition due
to fungal infection
Mycotoxin detection
Maize Aflatoxin Transmittance 500 Aflatoxin Changes in kernel PLSDA 95% accuracy for >90% error Pearson,
spectra (500–950 surface color and kernels with for kernels Wick-
nm) and chemical <10 and >100 with low,
reflectance composition due ppb aflatoxin 10–100 ppb Maghi-
spectra (550–1700 to fungal infection levels aflatoxin rang,
nm) levels Xie, and
Dowell
(2001)
Barley Deoxynivalenol NIR: 400–2500 nm 188 DON Change in chemical BPNN R2 = 0.933 SEP = 3.097 Ruan, Li,
(DON) (2 nm interval) properties of grain Lin, and
due to fungal Chen
infection and (2002)
optical property
change due to
DON
Wheat DON FTNIR: 650 and 17 DON Water and PCA, KNN, and Classification RMSEP Abramovic,
4000 cm–1 carbohydrate- PLSR efficiency was (mg/kg) = Jajic,
associated 100%; R2 = 1.639 Abramovic,
changes caused by 0.917 Cosic,
fungal and Juric
deterioration, (2007)
DON chemical
structure
(Continues)
2
Barley Ergosterol NIR: 400–2500 nm 37 Water and Proteins and PSLR R of dataset 1: RMSEP Börjesson,
(dataset ergosterol carbohydrates 0.83 (mg/kg) for Sten-
1) and content change during R2 of dataset dataset 1: berg, and
82 mold growth, 2: 0.59 4.5; dataset Schnürer
(dataset ergosterol 2: 4.7 (2007)
2) chemical structure
Wheat DON NIR: 200–2500 nm Kris: 18; Hardness index Symptom of PLSR R2 of Kris: 0.996 RMSECV Siuda, Bal-
Tur- and DON fusariosis in R2 of Turnia: (ppm) for cerowska,
nia: 32; content UV-VIS region and 0.999 Kris: 1.54; Kupcewicz,
Tonacja: change in grain R2 of Tonacja: Turnia: 2.76 and Lenc
30 macromolecules 0.994 and (2008)

and chlorophyll in Tonacja:
NIR region, DON 2.24
chemical structure
Maize Aflatoxin B1 FTNIR: 9000–4000 152 Aflatoxin Change in chemical PLSDA R2 of maize = RMSECV Fernández-
and (5–20 ppb) cm–1 (1112–2500 content properties of grain 0.80 value for Ibañez,
barley nm) due to fungal R2 of barley = maize = Soldado,
infection and 0.82 0.211 Martínez-
optical property barley = Fernández,
change due to 0.200 and De
aflatoxin la Roza-
Delgado
(2009)
Durum DON FTNIR: 262 DON content Chemical and PLSR and PCA R2 for durum: De Giro-
and 10,000–4000 optical properties 0.62; lamo,
com- cm–1 change due to the common: 0.63 Lippolis,
mon presence of DON and Nord-
wheat Durum+common: kvist,
0.58 and
Visconti
(2009)
(Continues)
17
18
Maize DON and NIR: 1100–2500 nm 539 DON content Disease severity and PLSR R2 of DON = Square error Bolduan,
Fumonisin change in 0.90 of cross- Montes,
(FUM) chemical R2 of FUM = validation(SECV) Dhillon,
composition of 0.68 (mg/kg) for Mirdita,
grain due to DON = and
infection and 0.50 and Melchinger
mycotoxin SUM = 1.04 (2009)
structure
Wheat DON NIR: 950–1650 nm 862 DON content Changes in PLSR R2 = 0.87 Square error Peiris et al.
carbohydrate, lipid of (2010)
and protein prediction
reserves, and DON (SEP)
levels (ppm) =
60.8
Corn Fumonisin B1, NIR: 650–2500 nm 168 Fuminisin B1 Change in chemical PLSR Multiple RMSECV Gaspardo
meal B2 and B1 composition of correlation (mg/kg) = et al.
grain due to coefficient 0.433 (2012)
infection and (RSQ) = 0.983
fumonisin
chemical structure
Maize Fumonisins NIR: 650–2500 nm 360 Fumonisin Change in chemical PLSR Correlation RMSEP Giacomo
content composition of coefficients (mg/kg) for and
grain due to for first model first model Stefania
infection and (PC = 17): = 0.929 and del
fumonisin 0.908 and second (2013)
chemical structure second model model =
(PC = 21): 0.890
0.909
(Continues)
Paddy Aflatoxin B1 FTNIR: 80 Aflatoxin B1 Change in water, PLSR Correlation SEP = 3.211% Qiang,
rice 10,000–4000 starch, protein, fat, coefficient = Fuguo,
cm–1 (100–2500 and aflatoxin 0.850 Cheng-
nm) infection hai,
Jingkun,
and
Xianzhe
(2014)
Durum DON FTNIR: 500 DON Change in kernel LDA and PLSR Classification False De Giro-
wheat 10,000–4000 composition and accuracy: compliant lamo,
cm–1 pigmentation due 75–90%; rates: 3–7% Cervel-

to fungal prediction and lieri,
infection, DON accuracy: R2 = RMSEP = Visconti,
chemical structure 0.630 1977 µg/kg and
Pascale
(2014)
Maize Fumonisin NIR: 400–2498 nm 117 Fungal count, Change in kernel Decision tree, Classification False Levasseur-
ergosterol, composition and PCA, and accuracy: 82% negative: Garcia,
and pigmentation due QDA 9% Bailly,
fumonisin to fungal infection Kleiber,
content and ergosterol and
content Bailly
(2015)
Maize Aflatoxin FTNIR: 4000–9999 232 Aflatoxin Cellular metabolites KNN, LDA, Classification Regression: Lee, Davis,
cm–1 of fungi and PLS-DA, and accuracy: SEP = 21.7 Her-
inhibition of PLSR highest in µg/kg rman,
macromolecular KNN is 100%; Murray,
synthesis in Regression: r and
maize, aflatoxin = 0.936 Deng
molecular (2015)
structure
(Continues)
19
20
Maize DON FTNIR: 4000–575 110 DON Change in kernel PCA Overall – Kos et al.
cm–1 composition and classification (2016)
pigmentation due accuracy for
to fungal infection 1750 µg/kg
and mycotoxin classifier =
chemical structure 79%; for 500
µg/kg
classifier =
85%
Maize Aflatoxin UV/NIR: 480 Aflatoxin Change in kernel Random forest Overall accuracy – Cheng,
304–1086 nm composition and model = 94.8% Vella,
pigmentation due and
to fungal infection Stasiewicz
and fluorescence (2019)
from the aflatoxin
contamination
Wheat DON FTNIR: 94 DON content Fibers, proteins, PLS-DA and Overall discrimi- No false De Giro-
bran 10,000–4000 starch, and PC-LDA nation rates compliant lamo
cm−1 , moisture for FTNIR: et al.
FTMIR: 4000–350 87–91% and (2019)
cm−1 FTMIR:
86–87%
Maize Aflatoxins NIR: 400–2500 nm 180 Aflatoxin B1 Molecular structure PLSR R2 = 0.91 RMSEP (ppb) Tao et al.
and optical = 284.27 (2020)
properties of
aflatoxin B1
Polished Aflatoxin B1 NIR: 950–1650 nm 105 Aflatoxin B1 Molecular structure PLSR R2 = 0.966 RMSECV Putthang,
rice and optical (µg/kg) = Sirisom-
properties of 0.008 boon,
aflatoxin B1 and Siri-
somboon
(2019)
levels, dough properties, and loaf volume (Dexter, Clear of their study indicated a good discriminant ability of the
& Preston, 1996). Researchers developed techniques for developed spectroscopic technique; however, the authors
measurement of mycotoxin levels, such as DON in sin- have not described whether the spectral variation in the
gle kernels of Fusarium-damaged wheat using NIR spec- contaminated wheat samples is due to the change in the
troscopy (Dowell, Ram & Seitz, 1999; K. Peiris & Dowell, macronutrient composition or the presence of ochratoxin,
2011), and it is proven to be the potential technique for itself. As the samples used for the development of the
detection of different quality defects in wheat. Smeesters, model were naturally contaminated, the variation in the
Meulebroeck, Raeymaekers, and Thienpont (2016) inves- spectra could be due to decrease in the starch and protein
tigated the application of UV-VIS-NIR diffuse reflectance content as a result of fungal invasion. Tao et al. (2020) stud-
spectroscopy for detection of DON in maize kernels in ied the feasibility of VIS-NIR spectroscopy for the detec-
two stages, viz. characterization of optimum illumination tion of AF in corn kernels and observed differences in
as well as the detection wavelength and accurate classifi- absorbance spectra between the fungus-infected and con-
cation of contaminated maize grains. They used 250 mm trol kernels (Figure 4), which can be explained by the scat-
reflection integrating sphere for first-stage optimization tering of light due to metabolic activities of fungus in the
and 30 mm integrating sphere for second-stage classifi- kernel. They further investigated the application of NIR
cation purposes. The results of the study revealed that spectroscopy for the detection of surface contamination
reflectance at 830, 940, and 1220 nm wavelengths and of corn kernels with aflatoxin B1 using PC-LDA and PLS-
the defined criterion successfully classified the contami- DA classification models separately. The result of the study
nated maize batch into high and low contaminated sub- showed that the best discrimination efficiency of 98.6% was
samples. This detection criterion of DON used was vali- achieved for 20–100 ppb threshold when analyzed over a
dated by the authors using measurement of a reference spectral range of 1120–2470 nm (Tao, Yao, Hruska et al.,
DON-contaminated maize powder, indicating robust DON 2019).
identification technique. Qiang et al. (2014) developed a
rapid NIRS method for detection of aflatoxin B1 in paddy
rice by natural and artificial contamination in a moisture 5.2 Nuts
range of 10–22%. A competition-type ELISA method was
followed in reference analysis and the spectral data were Mold infection creates a significant postharvest challenge
modeled by using a PLSR method with a correlation coef- for nut industries and reliable detection at early grow-
ficient of 0.850 and standard error of prediction of 3.211% of ing stages has not yet been accomplished. However, rapid
the AF concentration; the result showed that NIRS can be and nondestructive detection techniques have been inves-
a useful technique for detection of different type of myco- tigated and reported by several researchers. These meth-
toxins in rice; however, NIRS was unfit for the direct detec- ods can be implemented in both static and inline real-
tion of AFB1 in paddy rice with concentrations lower than time detection of mold and mycotoxin in nuts as reviewed
20 µg/kg. Shen, Wu, Shao, and Zhang (2018) worked exten- by Wu et al. (2018). Advance chemometric techniques for
sively on detection of fungal contamination and mycotox- improvement of model robustness, detection limit, and
ins using NIRS and in one of their work, they investigated detection speed have become the major research focus.
the potential quantification of AF in brown rice using NIR Moscetti et al. (2014) studied the feasibility of NIR spec-
and MIR spectroscopy. They analyzed 132 brown rice sam- troscopy having a wavelength range of 1100–2300 nm
ples within the AF concentration range of 0–2435.8 ppb by to classify surface and hidden fungal-infected chestnuts
artificially inoculating A. flavus and A. parasiticus strains using LDA, QDA, and KNN. They found classification
and obtained LDA classification model with an accuracy accuracy as high as 97% for QDA model and error rate
of 96.9% and 90.6% for NIR and MIR spectroscopy, respec- as low as 2.42% for false negative, 2.34% for false posi-
tively. Simultaneously, they found the developed PLSR tive, and 2.38% total error. Shen et al. (2018) investigated
model was also useful in predicting aflatoxins B1, B2, G1, mold contamination levels in peanuts by artificially inoc-
G2 effectively with correlation coefficients in the range of ulating different A. flavus strains in various concentra-
0.936–0.973 and 0.922–0.970 for NIR and MIR, respectively. tions and scanned the peanuts using FTNIR spectroscopy
De Girolamo et al. (2019) screened 229 durum wheat sam- of wavenumber range 4000–120,00 cm–1 . Results of their
ples naturally contaminated with ochratoxin, using both study revealed that 92.11% correct classification rate was
FTNIR and Fourier transform mid-infrared (FTMIR) spec- achieved using LDA with only 4.41% misclassification and
troscopy and reported 94% discrimination efficiency for high colony count prediction accuracy was also achieved
both the techniques for a cutoff limit of 2 µg/kg. Results using PLSR model with R2 value of 0.886.
FIGURE 4 Mean absorbance spectra of control, AF38, and AF13 inoculated corn samples (a) endosperm and (b)germ side (Tao et al.,
2019)
F I G U R E 5 Average smoothed fluorescence spectra for (a) pistachio samples of different level of aflatoxin (b) control, uncontaminated
and contaminated pistachio samples (Wu & Xu, 2020)
6 APPLICATION OF FLUORESCENCE group was always higher than that of the uncontaminated
SPECTROSCOPY FOR MOLD AND from 374 to 800 nm range and reported that the region of
MYCOTOXINS DETECTION 374–609 nm was related to AFB1 and oxidation substances
in samples. They applied LDA for classifying 600 pista-
Fluorescence spectroscopy also showed interesting and chio samples contaminated with AF levels from 5 to 50
promising outcomes in the detection of mycotoxins as µg/kg and found more than 91% classification accuracy
they emit BGYF when subjected to UV/visible light. Some for single variety of pistachio with a detection limit of 5
of the most relevant research works on the application ppb. Further, they were able to predict the level of AF in
of fluorescence spectroscopy for detection of mold and infected pistachios using competitive adaptive reweighted
mycotoxin contamination in cereal grains has been sampling with PLS (CARS-PLS) with less than 6.20 µg/kg
summarized in Table 4. Wu and Xu (2020) developed RMSEP values, irrespective of varieties. Bertani et al.
an online AF detection system for pistachios using the (2020) investigated the application of fluorescent spec-
fluorescence-based spectroscopic method under the troscopy for the detection of AF in grained almonds with
excitation of an optical fiber coupler laser with 360 nm a vision of developing a cheap, rapid, and nondestructive
wavelength and reported fluorescence spectra of various technique using a multispectral camera. They artificially
intensities for a different level of contamination (Figure 5). inoculated 2.7–320.2 ng/g AF B1+B2 over slurried almond
They observed that the intensity for the contaminated and analyzed the fluorescence spectra under 375 nm
TA B L E 4 Summary of research works on the application of fluorescence spectroscopy for detection of mold and mycotoxin contamination in cereal grains
Fungi/ Wavelength/ No. of Response
Grain mycotoxin wavenumber range samples variable Chemometrics Accuracy References
Corn A. flavus Fluorescence: 450–900 nm – Aflatoxin Internal average Peak emission wavelengths Koehler and
kernels relative varied from 500 to 515 nm Wieser
reflectance (2013); Yao,
(IARR) approach Hruska,
Brown, and
Cleveland
(2006)
Corn A. flavus Fluorescence: 400–600 nm 504 Aflatoxin MLR, two spring MLR: R2 = 0.72, Yao et al. (2010)
kernels 1991-class schema classification accuracy:
(TCS) 0.84–0.91 for threshold
20–100 ngg–1
Maize A. flavus Fluorescence: 400–900 nm 90 Aflatoxin ANOVA Hruska et al.
(2013)
Corn ears A. flavus Fluorescence: 400–900 nm 1200 Aflatoxin ANOVA Spectral shift at 501 and 478 Hruska et al.
nm (2013)
Maize A. flavus (AF13 and Fluorescence: 400–700 nm 180 Aflatoxin DA Overall classification Yao et al. (2013)
AF38) accuracy for 100 ppb
threshold = 94.4% on the
germ side
Maize Aflatoxin B1 NIR: 1000–2500 nm 150 Aflatoxin PCA, step-wise Classification accuracy for Wang et al.
factorial validation and (2014)
discriminant verification = 0.94% and
analysis (FDA) 0.88%, respectively
on first 20 PCs
Corn Aflatoxins Fluorescence: 400–700 nm 300 Aflatoxin PCA, kNN, and Best overall prediction Zhu et al. (2016)
kernels LS-SVM accuracy = 95.33% for
LS-SVM model with 100
ppb threshold on the
germ side
23
wavelength excitation. The spectral data demonstrated tions, leading to crucial control of the shelf life of cereal
that there was variation in the peak of contaminated and grains. Zhao, Wang, Chu, Li, and Kimuli (2017) investi-
noncontaminated samples at 410 nm, which could be the gated the application of NIR HSI for the detection of A.
sensitive wavelength for AF detection. They used SVM for parasiticus by acquiring images in the wavelength range
binary classification and found a classification accuracy of 921–2529 nm. The maize kernels were artificially inoc-
of 94% (and false-negative rate 6%) with a threshold set ulated and stored for various incubation times for the
at 6.4 ng/g. Recently, Gu, Chen, Wang, and Wang (2021) development of calibration and validation samples, which
investigated the potential of fluorescence spectroscopy are used for the classification model training and test.
coupled with headspace-gas chromatography-ion mobility The results of the study revealed the classification model
spectrometry (HS-GC-IMS) for detection of aflatoxigenic developed using SVM had the best classification accu-
mold contamination on peanut kernels. They artificially racy of 97.92% and 91.67% for calibration and validation
inoculated the peanut samples with fungal suspension dataset, considering the germ upside images. They selected
for scanning through the spectroscope and observed the 1145, 1408, 1935, 2103, and 2383 nm as the characteris-
largest variation at the peak intensities of 415−440 nm tic wavelength for the discrimination of fungal contam-
region, which were mainly related to the fluorescence ination levels. Sunli et al. (2018) demonstrated the suc-
absorption of aflatoxin B1. The fluorescence signals cessful determination of mold colonies using HSI in the
were further confirmed by HS-GC-IMS fingerprinting wavelength range of 871.607–1766.322 nm by analyzing the
to avoid misclassification due to background fluores- 100 mold-infested rice grains. Characteristic wavelengths
cent proteins of peanuts. Results of the study showed were selected by them using successive projection algo-
accurate characterization (96.7%) of aflatoxigenic and rithm (SPA) and CARS algorithm and used for SVR model
nonaflatoxigenic fungal infections using feature-level development for prediction of mold colonies. Again, a new
data fusion coupled with orthogonal partial least squares swarm intelligent gray wolf optimization (GWO) to opti-
discriminant analysis. They also established regression mize the penalty factor (c) and kernel function parame-
models for predicting colony counts of peanuts infected ter (g) of the prediction model. Authors followed the GWO
with aflatoxigenic fungi and found the best prediction model proposed by Mirjalili and Hashim (2010) in which
performance for A. flavus (R2 = 0.950) and A. parasiticus they simulated the hyperspectral wavelengths as forag-
(R2 = 0.971). ing behavior of the gray wolf population during hunting.
The leader alpha wolves decide hunting sleeping place,
time to wake, and so on. Similarly, the characteristic wave-
7 APPLICATION OF HSI FOR MOLD lengths were optimized to decide the prediction of mold
AND MYCOTOXINS DETECTION infection. The performance of the developed model was
cross-evaluated and found high prediction ability of CARS-
HSI is a promising technique that may be implemented GWO-SVR model with Rc 2 and Rg 2 value of 0.9621, 0.9511,
in the cereal industry for detection of mold and myco- respectively. Fusarium head blight is among the most com-
toxin infection inline, which will help to get rid of sub- mon mold defect that occurs in cereal grains caused due
jectivity, manual classification, and risk of being cross- to F. graminearum which starts at the flowering stage of
contaminated. To reproduce a consistent detection ability, wheat. HSI has emerged as a potential tool for determin-
researchers investigated the potential of HSI and demon- ing Fusarium infection as Delwiche, Rodriguez, Rausch,
strated its accuracy and predictability, as summarized in and Graybosch (2019) used 900–1700 nm wavelength HSI
Table 5. Xing et al. (2019) also reviewed the application of system to detect Fusarium infestation in wheat grown in
HSI for rapid detection of toxigenic mold and mycotoxins Nebraska in 2015. They applied PLSDA to the mean spectra
in various food products, in which applications in cereal of the wheat kernels to select the sensitive wavelengths and
grain are also discussed. Numerous recent applications of using them, LDA models were established. They obtained
this technique in the detection of fungal infection are also more than 95% classification accuracy on the individual
discussed in the following subsections. kernel level by using HSI, resulting reduction in the pro-
cessing time. Zhang et al. (2019) proposed a specific Fusar-
ium head blight classification index (FCI), which is com-
7.1 Cereal grains bination of single wavelength method and waveband com-
bination method to identify the diseased areas and to esti-
7.1.1 Detection of mold mate the degree of infection, respectively. Instability index
(ratio index of the within-class variability and between-
In applications involving HSI for cereal grain safety, detec- class variability) in combination with spectral angle map-
tion of toxigenic molds is one of the most studied applica- per classifier applied to the vis-NIR hyperspectral images
TA B L E 5 Summary of research works on the application of hyperspectral imaging for detection of mold and mycotoxin contamination in cereal grains
Actual detection
No. of according to our
Grain Fungi/mycotoxin Spectral range samples Response variables analysis Chemometrics Remarks References
Mold infection
Wheat Fusarium Vis/NIR: 96 Amount of fungus per Fungal constituents and PLSR Regression coefficient Polder, Van Der
430–900 nm wheat kernel change in (R2 ) = 0.80 Heijden,
and 900–1750 macromoleculess of Waalwijk, and
nm grain Young (2005)
Corn Various strains of Multispectral Infected/noninfected Fungal constituents and Stepwise dis- Classification Pearson and
kernels A. flavus, transmittance: Poineer3394: analogy for change in crimination accuracy for Wicklow
Fusarium, 550–1700 nm 292 and classification macromolecules and analysis infested kernels = (2006)
Diplodia, FS7111: pigments of grain 91.9% and extensive
Penicillium, etc. 148 discoloration =
96.2%
Maize Fusarium MatrixNIR: Two sets: 15 Microbial analysis Changes in moisture, PLSDA Coefficient of Williams,
verticillioides 960–1662 nm and 21 for determination of oil, protein, and determination for Manley, Fox,
and kernels infected/noninfected starch due to fungal matrix NIR = 0.73 and Geladi
sisuChema: class infection and fungal and sisuChema = (2010)
1000–2498 nm cell wall 0.86
Maize Aspergillus spp and Vis/NIR: 12 varieties Infected/noninfected Changes in surface PCA and DA DA on wavelengths: Del Fiore et al.
Fusarium spp 400–1000 nm of maize analogy for color, pigmentation, 410, 535, and 945 (2010)
and 3 classification based and water content due nm; classification
moisture on culture to infection accuracy = 99.97%
content: condition
15%, 20%,
and 30%
Red Fusarium Visible/NIR: 400 Severity of damage Changes in surface PCA and LDA Classification Shahin and
spring 400–1000 nm based on mycelia color, pigmentation, accuracy of sound Symons (2011)
wheat growth and water content due and damaged
characterization to infection kernels = 92%
using LDA;
prediction accuracy
of extent of
fusarium damage
= 86%
(Continues)
25
26
Actual detection
Wheat Fusarium Visible/NIR: 512 Infected/noninfected Changes in the kernel LDA Classification Delwiche, Kim,
400–1000 nm based on signs of color, starch, and accuracy of visible and Dong
and 1000–1700 scab and DON protein due to spectrometer = (2011)
nm content infection and the 94% and NIR
presence of chitin and spectrometer = 97%
ergosterol
Wheat Fusarium Vis/NIR: Diseased: Percentage infection Changes in surface PCA and Classification Bauriegel,
400–1000 nm 292 and of blighted spikelets color, reduction of spectral angle accuracy using Giebel, Geyer,
healthy: per ear chlorophyll, water mapper SAM = 87% Schmidt, and
80 content due to (SAM) Herppich
infection (2011)
Wheat A. glaucus FTNIR: 3 control Infected/noninfected Changes in water, PCA and The difference in the Singh, Jayas,
900–4000 and 3 analogy based on starch, proteins, k-means spectra due to Borondics,
cm–1 damaged culture condition lipids, lignin, and clustering reduced lipid (1740 and White
kernels cellulose content of cm–1 ), lignin (1595 (2011)
cut into kernel due to fungal cm–1 ), and cellulose
various infection (1250 cm–1 ) in
sections damaged
endosperm
Wheat Penicillium spp., A. NIR: 700–1100 – Infected/noninfected Changes in water, LDA Significant Singh et al.
glaucus, and A. nm analogy based on starch, proteins, wavelength = 870 (2012)
niger culture condition lipids, lignin, and nm; classification
cellulose content of accuracy =
kernel due to fungal 99–100%
infection
Wheat Ergot bodies NIR: 1100–2400 8 samples Amount of ergot Ergot contaminated PLSDA and R for PLSDA = Vermeulen,
(Claviceps nm bodies artificially pixels and ergot SVM 0.9995; SVM = Pierna,
purpurea) infected to the molecular structure 0.9997 Egmond,
kernels Dardenne,
and Baeten
(2012)
(Continues)
Actual detection
Wheat Fusarium NIR: 400–1000 5221 Extent of fusarium Changes in surface PLSDA Overall classification Shahin and
nm damage based on color, reduction of accuracy = 90% Symons (2012)
mycelial growth chlorophyll, water and false positive of
content due to 9%
infection
Maize Fusarium NIR: 1000–2498 40 kernels Extent of fungal Depletion of food PCA and PLSR Variation in spectra at Williams,
verticillioides nm for 5 growth based on reserves such as 1405 nm (starch); Geladi, Britz,
treat- treatment time starch, protein, and 1900 nm (aromatic and Manley
ments lipids due to fungal structure); 2136 nm (2012)
growth (protein). R2 = 0.98
Durum Fusarium NIR: 1000–1700 3 sets of Vitreous/Fusarium- Changes in chemical PCA and Classification Serranti, Cesare,
wheat nm samples damaged/yellow composition of wheat PLSDA sensitivity = 0.94 and Bonifazi
having berry kernels due to various types of and specificity = (2013)
40 infection 0.99
kernels
each
Wheat Claviceps purpurea NIR: 970–2500 7 samples Quantification of Changes in the chemical PCA and R value for Vermeulen et al.
and rye nm (Burger- 50 g each ergot bodies using composition of PLSDA BurgerMetrics = (2013)
Metrics & stereo-microscope kernels due to fungal 0.94; SisuCHEMA
SisuCHEMA infection = 0.99
Durum Fusarium Vis/MIR: 100 Extent of fusarium Changes in surface PLSDA Classification Menesatti et al.
wheat 450–700 nm infection based on color, reduction of accuracy of 86% (2013)
storage days chlorophyll, water was found after
content due to 8–10 days of
infection infection
(Continues)
27
28
Actual detection
2
Brown A. oryzae Vis/NIR:400– 210 Mold colony forming Formation of fungal PLSR R = 0.97, RMSEP = Siripatrawan
rice 1000 units per gram structures (mycelium 0.39 log CFU/g and Makino
nm and spores), and the (2015)
type and
concentration of
metabolic products
from fungal activities
Wheat Fusarium NIR: 515–1785 800 DON content Fungal constituents and Self-developed Classification Barbedo, Tibola,
nm change in innovative accuracy = 91% and
macromolecules and algorithm Fernandes
pigments of grain and (2015)
changes in grain
chemical composition
due to fungal
infection
Wheat Fusarium NIR 30 Severity of fusarium Changes in the chemical PCA and MLR PCA was helpful for Jaillais, Roumet,
multispectral: head blight (FHB), composition of grain classification of Pinson-
360–950 nm total fusarium DNA due to fungal infested and Gadais, and
content infection healthy kernel and Bertrand
MLR allowed the (2015)
degree of
contamination for
each pixel to be
estimated
Oats Fusarium NIR: 1000–2500 Calibration Presence of fungal Changes in the starch PLSR Infected kernels Tekle, Måge,
nm set = 124 mycelia through and protein structure showed higher Segtnan, and
and vali- microscopy, DON and composition due intensities at 1925, Bj⊘ rnstad
dation content, and kernel to fungal infection 2070, and 2140 nm. (2015)
set = 56 weight R = 0.8
(Continues)
Actual detection
ssss A. glaucus and NIR: 100–1600 3000 Moisture content, Changes in moisture, PCA, pair-wise Classification Senthilkumar,
Barley Penicillium spp nm germination, and starch, and protein two-class and accuracy >94% for Jayas, White,
ochratoxin A content in the grain six-class initial fungal Fields, and
concentrations due to fungal classification infection and 100% Gräfenhan
infection models for infection after 4 (2016)
weeks
Cereal Aspergillus spp. Vis/NIR: 105 Growth stages of the Changes in the color and PCA, SVM, and Classification Lu et al. (2020)
fungi and Penicillium 400–1000 nm fungus at different chmical composition SPA accuracy of
on spp. storage time of growth media SPA-SVM method
culture = 98.89%
medium
Mycotoxins
Milled Fumonisin NIR: 720–940 178 Fumonisin Changes in the chemical Feed-forward Prediction accuracy: Firrao et al.
maize nm composition of grain neural R2 = 0.68 and error: (2010)
due to fungal network MSE = 0.0029 ppb
infection and
fumonisin molecular
structure
Corn Aflatoxin B1 NIR: 1100–1700 195 Aflatoxin B1 Optical properties and PLSDA Classification Kandpal, Lee,
kernels nm molecular structure of accuracy = 96.9% Kim, Bae, and
aflatoxin Cho (2015)
Single Aflatoxin B1 NIR: 1729–2344 120 Aflatoxin B1 Change in chemical PCA and Classification Wang et al.
maize nm composition of grains spectral angle accuracy for three (2015)
kernels due to fungal mapper different varieties
infection and (SAM) = 96.15%, 80%, and
molecular structure of classifier 82.61%
aflatoxin
Maize Aflatoxin B1 Vis/NIR: 150 Aflatoxin B1 Optical properties and PCA, FDA, and Overall classification Wang et al.
kernels 600–1000 nm molecular structure of PCA+FDA accuracy = 98% (2015)
aflatoxin
29
of winter wheat spikelets for single wavelength selection F. gramineraum, associated with maize. The highest inten-
and PLSR was used for waveband set selection. The opti- sity changes were observed at the wavelengths of 1140,
mal single wavelengths found by them for classification 1418, and 1895 nm, which the authors claimed to be acco-
of contaminated and noncontaminated areas were 447, ciated with the presence of carbohydrates and proteins of
539, 668, and 673 nm and the changes in the reflectance fungi cell wall. They applied PLS-DA for classification of
of peak wavelengths mostly caused due to the changes the infected samples from the control and achieved with
in chlorophyll and zeaxanthin content over the spikelet. 100% discrimination accuracy, sensitivity, and specificity
They demonstrated the classification accuracy of devel- for both fungi.
oped FCI to be 89.80%, which is 30% more accurate than
the best-performing spectral vegetation index. Williams,
Bezuidenhout, and Rose (2019) investigated the feasibility 7.1.2 Detection of mycotoxins
of HSI in differentiating maize ear rot pathogens (Fusar-
ium verticillioides, F. graminearums., F. boothiiand, and Awareness about mycotoxins contamination in agri-food
Stenocarpella maydis) within wavelength range of 1000– products has grown enormously as associated health dis-
2500 nm and developed pixel and object-wise classifica- orders in humans and animals are being recognized as
tion algorithms using PCA and PLSDA. They observed a major health and economic problem. AF is one of the
prominent changes in the intensity at 1180, 1430, and 1930 prevalently harmful mycotoxins, which pose a significant
nm, which they demonstrated as the changes in chem- economic burden, causing an estimated 25% or more of the
ical composition due to mold infection and presence of world’s food crops to be infected annually (JECFA, 2018).
carbonyl compounds of F. verticillioides. They correlated HSI is emerging as a promising technology for the detec-
the level of mold infection with the spectral changes and tion of AF kernel-wise to enable growers get an assurance
developed object-wise classification method, which was of properly evaluating a lot of almonds. A brief summary
found to be more effective approach in detecting the molds. of the research conducted by several researchers in the last
Wu et al. (2020) investigated the application of visible and decade is provided in Table 5. Chu, Wang, Yoon, Ni, and
NIR HSI in rice false smut detection and developed a multi- Heitschmidt (2017) studied the feasibility of short-wave
variate regression model using artificially infected kernels infrared (SWIR) HSI in the wavelength range of 1000–2500
with Villosiclava virens fungus. Wavelength optimization nm for detection of aflatoxin B1 in individual maize kernels
was done by comparing three representative variable selec- by artificially inoculating A. flavus. They identified five key
tion methods, viz. SPA. The quantitative analysis of the wavelengths for identification of aflatoxin B1, viz., 1317,
fungal infection was done by PLSDA, SVM, and extreme 1459, 1864, 1934, and 2254 nm, based on large absolute coef-
learning machine (ELM) using the optimal wavelengths. ficients. Using the key wavelengths in SVM method, they
Result of the study demonstrated that the best classifica- achieved 82.50% overall classification accuracy. However,
tion accuracy at 99.33% was achieved by random forest- when the same model was applied on single kernels, a clas-
ELM method using 47 characteristic wavelengths. Chu, sification accuracy of 95.56% was obtained for the lower AT
Wang, Ni, Li, and Li (2020) developed an efficient classi- levels (<20 ppb), 82.35% for intermediate AT levels (20-100
fication model for discrimination of fungal-infected maize ppb), and 75% for the higher at levels (>100 ppb). The pos-
kernels from the healthy one, using NIR HSI within the sible AF distribution map demonstrated by them is given
wavelength range of 935–1720 nm with 3.5 nm spectral res- in Figure 6. Kimuli, Wang, Jiang, Zhao, and Chu (2018)
olution. Their spectral data showed prominent intensity also used SWIR HSI for 600 maize kernels for detection of
changes at 1205, 1252, 1305, 1450, 1581, and 1659 nm, which AF using PCA and PLSDA and found peak differences at
were likely a reflection of differences in chemical compo- 1109, 1205, 1302, 1459, 1865, 1934, and 2235 nm wavelengths.
sitions caused by the fungal metabolism in kernels. They Results of their study indicated that SNV and first deriva-
developed object-wise classification model using the full tive preprocessing combined with PLSDA produced best
spectrum and achieved accuracies of 99.00%, 97.96%, and classification accuracy of 100% and 96% for calibration and
97.87% for the three maize hybrids, respectively. Further, validation sets, respectively. Both the above researchers
SPA was used to select the characteristic wavelengths viz., validated the presence of AF by using a 365 nm UV inci-
1168, 1205, 1344, 1414, and 1673 nm to reduce the informa- dent source and checking BGYF emergence from the ker-
tional redundancy found from full spectra. By using the nels. Therefore, a further confirmational study is needed to
above selected wavelengths SPA-SVM model was devel- identify whether the changes in the reflectance values were
oped, resulting classification accuracy of 100%, 98.98%, and due to the presence of AF or changes in the chemical com-
98.94% for the three maize hybrids. da Conceição et al. position due to mold contamination (Zhongzhi & Limiao,
(2020) implemented NIR-HSI for differentiation of two 2020). DON is also a cereal-based mycotoxin, poses a signif-
mycotoxigenic Fusarium species, viz. F. verticillioides and icant threat to both human and animal health by inhibiting
F I G U R E 6 (a) Maize kernels from BH9740VTTP variety captured by NIR HSI (b) possible distribution maps of AFB1 contents in maize
kernels. The blue indicates low AFB1 content, and red designates a very high AFB1 content. Source: Chu et al. (2017)
protein synthesis and ultimately leads to cease of normal HSI. Liang, Slaughter, Ortega-Beltran, and Michailides
eukaryote cell function. Many countries established lim- (2015) investigated the application of NIR HSI (800–2500
its of DON, ranging from 0.5 mg/kg for processed cereals nm) for detection of A. flavus and A. parasiticus infec-
and 1.25 mg/kg for unprocessed cereals, which necessitates tion in almonds using a canonical classifier and found the
the implementation of rapid and accurate detection meth- algorithm could discriminate the infected almond sam-
ods for low concentration DON. Barbedo, Tibola, and Lima ples from healthy ones with an error rate of 0.26% and
(2017) investigated the detection of DON using HSI in the zero false negatives. They observed that the largest differ-
wavelength range of 585–1785 nm and reported that direct ences in the spectra of infected and healthy kernels were
detection of DON was unfeasible for them, but detection at wavelength bands of 1210, 1720, and 1760 nm, which are
was accurate enough by establishing a correlation between mostly lipid absorbance bands. Kheiralipour et al. (2016)
Fusarium damage and DON content. They developed two- studied the capability of long-wave NIR HSI in rapid and
class and three-class discrimination models and achieved nondestructive detection of different isolates of A. flavus,
an accuracy of 72% and 81%, respectively. Femenias et al. KK11 and R5 in pistachio nuts. They extracted features of
(2020) studied the applicability of NIR HSI (900–1700 nm) healthy and infected kernels from the hypercube to clas-
for detection of DON in naturally contaminated wheat ker- sify the kernels using QDA and found an overall classifi-
nels using the PLSR method and revealed high prediction cation accuracy of 91.7%, which demonstrated the impor-
accuracy with R2 value of 0.76 and root mean square error tance of the technique in the field of mold detection. It
of cross-validation (RMSECV) value of 405 µg/kg. Above is well known that moldy peanut kernels contain poten-
discussed research results are providing evidence of potential carcinogens and efficient identification can prevent
tial application of the HSI technique for the detection of the toxins into the food chain. Jiang et al. (2016) stud-
mycotoxins in cereal grains (Table 4). ied the feasibility of NIR HSI for the detection of moldy
peanut kernels in the wavelength range of 970–2500 nm.
They applied PCA to discriminate the infected peanuts by
7.2 Nuts selecting the sensitive bands at 1100, 1207, 1302, 1509, 1634,
1722, 1847, and 1922 nm and found an overall classification
7.2.1 Detection of mold accuracy of 87.14%. Qiao, Jiang, Qi, Guo, and Yuan (2017)
used short-wave HSI in the wavelength range of 967–2499
Nuts support the growth of toxicogenic molds when stored nm for the detection of fungal-infected peanuts by non-
in a warm and humid climate and produce deadly myco- parametric weighted feature extraction technique. They
toxins. Several researchers have attempted rapid and non- applied SVM for the classification of infected and healthy
destructive technologies for detection of the mold growth peanuts and found pixel-wise overall classification accu-
and mycotoxins and revealed the immense potential of racy of 99.13%, 96.72%, and 99.73% for variety A, B, and
C, respectively. Qi, Jiang, Cui, and Yuan (2019) also inves- cluded that 400–430 nm is the key wavelength range for
tigated the NIR HSI in the same wavelength range as of aflatoxin B1 detection in peanuts. However, the detection
Qiao et al. (2017) by adopting joint sparse representation- accuracy was affected due to the irregular spherical shape
based classification and revealed classification accuracy of the nuts, which was solved by Yuan et al. (2020). They
of 99.2% and 98.8% at pixel scale, which is more accurate noticed the spectral intensity of irregularly shaped peanut
and consistent than SVM classification technique. Further- samples decreases gradually from the central region to the
more, Qi et al. (2020) extracted wavelet spectral features edges. To rectify these irregularities, they applied singular
from hyperspectral images of moldy peanut samples and spectrum analysis-based correction model by assuming the
conducted continuous wavelet transform with separabil- spectral intensity of all pixels of peanuts to be equal. After
ity analysis between healthy and moldy peanuts. They iso- following this approach, the overall classification accuracy
lated five wavelet features and seven optimal bands, viz. was found to be 99.03% and 97.76% for Dabaisha and Huayu
1005, 1208, 1450, 1927, 2078, 2190, and 2251 nm, for moldy varieties, respectively.
peanut identification. The wavelet features combined with
PLS-DA and SVM identified moldy peanut kernels with
96.19% accuracy. 8 CHALLENGES AND FUTURE SCOPE
Optical imaging and spectroscopic techniques have been

7.2.2 Detection of mycotoxins successfully applied in many food industries for quality
and safety evaluation. These techniques are rapid, non-
Han and Gao (2019) demonstrated pixel-level AF detect- destructive, and exclude major sample preparation steps
ing algorithm based on deep learning, using 146 UV- and involvement of harmful chemicals. However, these
visible-NIR hyperspectral image cubes (292–865 nm) of techniques are still facing some challenges and research
73 peanut samples. They observed 450 nm to be the is undergoing for a possible solution to the difficulties
sensitive wavelength and the reflectance of contaminated faced by the food industries. Stability, consistency, and
pixels was higher than the noncontaminated pixels at 450– accuracy are the three basic challenges of the optical
550 nm. Fisher-PCA method revealed the key wavelengths detection systems. Moreover, automation of optical-based
as 432 and 514 nm, which are mostly the fluorescent spec- nondestructive technologies in food industries is facing
tral values of aflatoxin B1. As the investigators have added difficulty with the influence of physical and biological
the standard AF over the kernels, it can be concluded that variability, which affects the robustness of the features and
the changes in the intensity at the key wavelengths are algorithms. To reduce the effect of biological variability,
due to the presence of AF only. The results of their study diverse libraries need to be developed by considering
revealed an overall classification accuracy of 95% at pixel- cultivar, maturity level, origin, and harvest season, thus
level and better than other classification models they used, the calibration dataset becomes sufficiently rich in vari-
viz. KNN, random forest, radial basis function-SVM, and ation. Another existing issue with the inline detection
BP-ANN. Xing et al. (2017) applied fluorescence visible technique is the inability in detecting the entire surface of
near-infrared HSI with UV illumination source (365 nm an object. However, this problem can be solved by surface
excitation and a 400 nm emission filter) to verify the spec- reconstruction using multiple images captured by the
tral difference between A. flavus and A. parasiticus infected camera, covering the entire surface of the object rotating
and healthy peanut kernels. They recorded UV excitation using a roller. In the case of smaller objects like grains
of the peanut samples with and without skin and interest- and nuts, the front and back side images can be acquired
ingly, they observed fluorescence intensity of peanuts with- by holding the samples on a glass base and placing two
out skin which was higher than that of peanuts with skin cameras directing toward the two surfaces. While digital
(10 times). Furthermore, they were able to separate the color imaging cannot detect the chemical composition,
toxigenic inoculation group from control group with 100% NIR spectroscopic techniques offer a solution to extract
accuracy. Zhongzhi and Limiao (2020) proposed a machine some chemical signature of the sample. NIRS does not
learning AF detection method for peanut, based on fluo- provide spatial distribution map for chemical composition
rescence HSI in the wavelength range of 400–720 nm. For of a sample. Hyperspectral and multispectral imaging
practical applicability, they also optimized the wavelength contain both spatial and spectral information, although
bands and developed a narrow-band recognition model high-dimensional data and excessive processing time need
using SVM. Fisher weights of each band were compared to be addressed for commercial deployment. Therefore,
and found that 410–430 nm waveband was the peak for feature and wavelength selection can be investigated in
AF detection, resulting in classification accuracy of 93.48%. future research. To increase the consistency and accu-
From the above-published research analysis, it was con- racy of detection, various innovative arrangements of
computer vision systems and advance chemometrics can large and heterogenous samples with reduction of sam-
be developed for inline applications. Therefore, multispec- pling error.
tral imaging with selective sensitive wavebands is a viable The comparative assessment of the existing techniques
option for rapid and nondestructive industrial detection for nondestructive estimation of mycotoxin and mold
of molds and mycotoxins in cereal grains and nuts. Sepa- growth, covered in this review, can be utilized by the
ration of sensitive wavelengths for final image processing researcher, growers and processors concerning the food
can be done by placing specific wavelength filters in safety, and quality assurance demanded in grains and nuts
monochrome cameras. Furthermore, detection accuracy industries. This review could be a good starting point to
can be improved by developing innovative calibration be referred for exploring the possibilities of hybrid detec-
and prediction models and introducing new spatial and tion techniques for more accurate and timely detection
spectral features for mold and mycotoxins analysis. of mycotoxin and mold growth with a nondestructive
approach.
9 CONCLUSIONS AC K N OW L E D G M E N T S
We thank Cooperative Research Centres Projects pro-
The current review emphasized on most relevant optical gram (CRCPSIX000081), Australia and Surenut Pty Ltd. for
and spectroscopic techniques facilitating rapid and non- financial support for the project. We also thank University
destructive detection of toxigenic molds and mycotoxins of South Australia for providing necessary infrastructure
in cereal grains and nuts. Based on the detection princi- facilities throughout the project duration.
ple and hardware components, each technique has its own
capability of detecting molds and mycotoxins. Digital color AU T H O R CO N T R I B U T I O N S
imaging has been successfully applied in the food industry Gayatri Mishra: Writing-original draft (Lead). Brajesh
for surface mold detection but very few applications were Kumar Panda: Writing-original draft (Co-lead). Wilmer
reported for mycotoxin detection due to its limitations in Ariza Ramirez: Writing-original draft (Supporting).
obtaining the details of chemical composition. X-ray imag- Hyewon Jung: Writing-original draft (Supporting).
ing technique is being used for detecting internal and exter- Chandra B. Singh: Conceptualization (Equal), Fund-
nal molds and mycotoxins, but the cost and risk of radia- ing acquisition (Equal), Resources (Lead), Supervision
tion exposure to the human body constrain its popularity (Equal), Writing-review & editing (Lead). Sang-Heon
in practice. Interestingly, fluorescence spectroscopy is one Lee: Funding acquisition (Supporting), Supervision
such technique which can detect both mold infection and (Lead), Writing-review & editing (Supporting). Ivan Lee:
mycotoxins with high sensitivity and specificity. However, Funding acquisition (Supporting), Resources (Support-
the background fluorescent elements of the sample usu- ing), Supervision (Supporting), Writing-review & editing
ally interfere with the desired spectra, resulting in mixed (Supporting).
and shifted peaks.
Type of molds, severity of damage, and the presence CONFLICTS OF INTEREST
of mycotoxins as low as 2–20 µg/kg can be accurately The authors declare no conflicts of interest.
detected by vibrational spectroscopic techniques, such as
NIRS and HSI. NIRS could provide extensive information ORCID
about toxigenic mold growth and mycotoxins by analyzing Gayatri Mishra https://orcid.org/0000-0001-8941-5445
the change in secondary molecular structures of the ker- Brajesh Kumar Panda https://orcid.org/0000-0002-
nel and water content. However, its accuracy may be lim- 2257-1600
ited due to its nature of detection at single average point Wilmer Ariza Ramirez https://orcid.org/0000-0002-
and possible heterogeneous distribution of the contami- 2755-2281
nants on different food materials. Hence, multipoint analy- Hyewon Jung https://orcid.org/0000-0002-0057-0818
sis is required in case of NIRS to achieve satisfactory perfor- Chandra B Singh https://orcid.org/0000-0003-4160-
mance. This limitation of NIRS can be overcome through 4655
HSI technique, which combines the advantages of digi- Sang-Heon Lee https://orcid.org/0000-0002-3655-7981
tal imaging and spectroscopic techniques (in spatillay dis- Ivan Lee https://orcid.org/0000-0002-2826-6367
tributed spectra) to detect uneven distribution of molds
and mycotoxins. Because of the nondestructive nature of REFERENCES
detection and rapid response, NIRS and HSI techniques (JECFA), J. F. W. E. C. o. F. A. (2018). Aflatoxins. Retrieved from
can be used to develop compact and portable instruments https://www.who.int/foodsafety/FSDigest_Aflatoxins_EN.pdf
for online detection, which will be helpful in screening a
Abramovic, B., Jajic, I., Abramovic, B., Cosic, J., & Juric, V. (2007). Caporaso, N., Whitworth, M. B., & Fisk, I. D. (2018). Near-infrared
Detection of deoxynivalenol in wheat by Fourier transform spectroscopy and hyperspectral imaging for non-destructive qual-
infrared spectroscopy. Acta Chimica Slovenica, 54(4), 859. ity assessment of cereal grains. Applied Spectroscopy Reviews, 53(8),
Alpaydin, E. (2020). Introduction to machine learning. MIT Press. 667–687. https://doi.org/10.1080/05704928.2018.1425214
Amigo, J. M., Martí, I., & Gowen, A. (2013). Hyperspectral imaging Chavez, R. A., Cheng, X., & Stasiewicz, M. J. (2020). A review of the
and chemometrics: A perfect combination for the analysis of food methodology of analyzing aflatoxin and fumonisin in single corn
structure, composition and quality. In Federico Marini (Ed.), Data kernels and the potential impacts of these methods on food secu-
handling in science and technology (Vol. 28, pp. 343–370). Elsevier. rity. Foods, 9(3), 297.
Awika, J. M. (2011). Major cereal grains production and use around Cheng, X., Vella, A., & Stasiewicz, M. J. (2019). Classification of
the world. In Joseph M. Awika Vieno Piironen & Scott Bean aflatoxin contaminated single corn kernels by ultraviolet to near
(Eds.), Advances in cereal science: Implications to food processing infrared spectroscopy. Food Control, 98, 253–261.
and health promotion (pp. 1–13). ACS Publications. Chu, X., Wang, W., Yoon, S.-C., Ni, X., & Heitschmidt, G. W. (2017).
Bailly, J.-D., & Oswald, I. (2013). Toxins| Mycotoxins. In Paul Worsfold Detection of aflatoxin B1 (AFB1) in individual maize kernels using
Colin Poole Alan Townshend & Manuel Miró (Eds.), Encyclopedia short wave infrared (SWIR) hyperspectral imaging. Biosystems
of analytical science (Vol. 10, pp. 129–140). Elsevier. Engineering, 157, 13–23.
Barbedo, J. G. A., Tibola, C. S., & Fernandes, J. M. C. (2015). Detecting Chu, X., Wang, W., Ni, X., Li, C., & Li, Y. (2020). Classifying maize
Fusarium head blight in wheat kernels using hyperspectral imag- kernels naturally infected by fungi using near-infrared hyperspec-
ing. Biosystems Engineering, 131, 65–76. https://doi.org/10.1016/j. tral imaging. Infrared Physics & Technology, 105, 103242.
biosystemseng.2015.01.003 Coates, J. (1998). Vibrational spectroscopy: Instrumentation for
Barbedo, J. G. A., Tibola, C. S., & Lima, M. I. P. (2017). Deoxyni- infrared and Raman spectroscopy (Reprinted from Analytical
valenol screening in wheat kernels using hyperspectral imag- Instrumentation Handbook, 1997). Applied Spectroscopy Reviews,
ing. Biosystems Engineering, 155, 24–32. https://doi.org/10.1016/j. 33(4), 267–425. https://doi.org/10.1080/05704929808002060
biosystemseng.2016.12.004 da Conceição, R. R. P., Simeone, M. L. F., Queiroz, V. A. V., de
Bauriegel, E., Giebel, A., Geyer, M., Schmidt, U., & Herppich, W. B. Medeiros, E. P., de Araújo, J. B., Coutinho, W. M. . . . de Resende
(2011). Early detection of Fusarium infection in wheat using hyper- Stoianoff, M. A. (2020). Application of near-infrared hyperspec-
spectral imaging. Computers and Electronics in Agriculture, 75(2), tral (NIR) images combined with multivariate image analysis in
304–312. the differentiation of two mycotoxicogenic Fusarium species asso-
Berardo, N., Pisacane, V., Battilani, P., Scandolara, A., Pietri, A., & ciated with maize. Food Chemistry, 344, 128615.
Marocco, A. (2005). Rapid detection of kernel rots and mycotox- da Silva, E., & Mendonca, G. (2005). Digital image processing.
ins in maize by near-infrared reflectance spectroscopy. Journal of Dale, L. M., Thewis, A., Boudry, C., Rotar, I., Dardenne, P., Baeten,
Agriculture and Food Chemistry, 53(21), 8128–8134. https://doi.org/ V., & Pierna, J. A. F. (2013). Hyperspectral imaging applications
10.1021/jf0512297 in agriculture and agro-food product quality and safety control: A
Bertani, F., Businaro, L., Gambacorta, L., Mencattini, A., Brenda, D., review. Applied Spectroscopy Reviews, 48(2), 142–159. https://doi.
Di Giuseppe, D. . . . Gerardino, A. (2020). Optical detection of afla- org/10.1080/05704928.2012.705800
toxins B in grained almonds using fluorescence spectroscopy and Danyluk, M. D., Harris, L. J., & Sperber, W. H. (2007). Nuts and cere-
machine learning algorithms. Food Control, 112, 107073. als. In Michelle D. Danyluk Linda J. Harris & William H. Sperber
Beuchat, L. R. (1983). Influence of water activity on growth, metabolic (Eds.), Food microbiology: Fundamentals and frontiers (3rd ed., pp.
activities and survival of yeasts and molds. Journal of Food Protec- 171–183). American Society of Microbiology.
tion, 46(2), 135–141. https://doi.org/10.4315/0362-028X-46.2.135 De Girolamo, A., Cervellieri, S., Cortese, M., Porricelli, A. C. R., Pas-
Bhat, R., Rai, R. V., & Karim, A. A. (2010). Mycotoxins in food and cale, M., Longobardi, F. . . . Lippolis, V. (2019). Fourier transform
feed: Present status and future concerns. Comprehensive Reviews near-infrared and mid-infrared spectroscopy as efficient tools for
in Food Science and Food Safety, 9(1), 57–81. https://doi.org/10.1111/ rapid screening of deoxynivalenol contamination in wheat bran.
j.1541-4337.2009.00094.x Journal of the Science of Food and Agriculture, 99(4), 1946–1953.
Blomhoff, R., Carlsen, M. H., Andersen, L. F., & Jacobs, D. R., Jr. De Girolamo, A., Cervellieri, S., Visconti, A., & Pascale, M. (2014).
(2006). Health benefits of nuts: Potential role of antioxidants. Rapid analysis of deoxynivalenol in durum wheat by FT-NIR spec-
British Journal of Nutrition, 96 (S2), S52–S60. https://doi.org/10. troscopy. Toxins (Basel), 6(11), 3129–3143. https://doi.org/10.3390/
1017/bjn20061864 toxins6113129
Bolduan, C., Montes, J. M., Dhillon, B. S., Mirdita, V., & Melchinger, De Girolamo, A., Lippolis, V., Nordkvist, E., & Visconti, A. (2009).
A. E. (2009). Determination of mycotoxin concentration by ELISA Rapid and non-invasive analysis of deoxynivalenol in durum
and near-infrared spectroscopy in Fusarium-inoculated maize. and common wheat by Fourier-transform near infrared (FT-NIR)
Cereal Research Communications, 37(4), 521–529. https://doi.org/ spectroscopy. Food Additives and Contaminants Part A-Chemistry
10.1556/Crc.37.2009.4.5 Analysis Control Exposure & Risk Assessment, 26(6), 907–917.
Börjesson, T., Stenberg, B., & Schnürer, J. (2007). Near-infrared spec- https://doi.org/10.1080/02652030902788946
troscopy for estimation of ergosterol content in barley: A com- De Girolamo, A., von Holst, C., Cortese, M., Cervellieri, S., Pascale,
parison between reflectance and transmittance techniques. Cereal M., Longobardi, F. . . . Lippolis, V. (2019). Rapid screening of ochra-
Chemistry, 84(3), 231–236. toxin A in wheat by infrared spectroscopy. Food Chemistry, 282,
Boswell, D. (2002). Introduction to support vector machines. 95–100. https://doi.org/10.1016/j.foodchem.2019.01.008
Departement of Computer Science and Engineering, University of Del Fiore, A., Reverberi, M., Ricelli, A., Pinzari, F., Serranti, S., Fab-
California San Diego. bri, A. A. . . . Fanelli, C. (2010). Early detection of toxigenic fungi
on maize by hyperspectral imaging analysis. International Jour- Gat, N. (2000). Imaging spectroscopy using tunable filters: A review.
nal of Food Microbiology, 144(1), 64–71. https://doi.org/10.1016/j. In Harold H. Szu Martin Vetterli William J. Campbell & James R.
ijfoodmicro.2010.08.001 Buss (Eds.), Wavelet applications VII (Vol. 4056, pp. 50–64). Inter-
Delwiche, S. R., Kim, M. S., & Dong, Y. (2011). Fusarium damage national Society for Optics and Photonics.
assessment in wheat kernels by Vis/NIR hyperspectral imaging. Georgiadou, M., Proshlyakov, Α., Revithi, Α., Tjamos, S., Stringlis,
Sensing and Instrumentation for Food Quality and Safety, 5(2), I., Tsitsigiannis, D. . . . Yanniotis, S. (2012). Detection of insect
63–71. necrotic internal damage and aflatoxin reduction in stored pista-
Delwiche, S. R., Rodriguez, I. T., Rausch, S. R., & Graybosch, R. chio nuts. Bulletin of IOBC-WPRS, 81, 21–29.
A. (2019). Estimating percentages of fusarium-damaged kernels Giacomo, D. R., & Stefania del, Z. (2013). A multivariate regression
in hard wheat by near-infrared hyperspectral imaging. Journal of model for detection of fumonisins content in maize from near
Cereal Science, 87, 18–24. https://doi.org/10.1016/j.jcs.2019.02.008 infrared spectra. Food Chemistry, 141(4), 4289–4294. https://doi.
Dexter, J., Clear, R., & Preston, K. (1996). Fusarium head blight: Effect org/10.1016/j.foodchem.2013.07.021
on the milling and baking of some Canadian wheats. Cereal Chem- Gourama, H., & Bullerman, L. B. (1995). Detection of molds in foods
istry, 73(6), 695–701. and feeds: Potential rapid and selective methods. Journal of Food
Di Benedetto, N., Perricone, M., & Corbo, M. R. (2010). Alter- Protection, 58(12), 1389–1394.
native non thermal approaches: Microwave, ultrasound, pulsed Grimm, L. G., & Yarnold, P. R. (1995). Reading and understanding
electric fields, irradiation. In Antonio Bevilacqua Maria Rosaria multivariate statistics. American Psychological Association.
Corbo & Milena Sinigaglia (Eds.), Application of alternative food- Gu, S., Chen, W., Wang, Z., & Wang, J. (2021). Rapid determina-
preservation technologies to enhance food safety and stability (pp. tion of potential aflatoxigenic fungi contamination on peanut ker-
143–160). nels during storage by data fusion of HS-GC-IMS and fluorescence
Dowell, F., Ram, M., & Seitz, L. (1999). Predicting scab, vomitoxin, spectroscopy. Postharvest Biology and Technology, 171, 111361.
and ergosterol in single wheat kernels using near-infrared spec- Gunasekaran, S. (1996). Computer vision technology for food quality
troscopy. Cereal Chemistry, 76(4), 573–576. assurance. Trends in Food Science & Technology, 7(8), 245–256.
Draganova, T., Daskalov, P., & Tsonev, R. (2010). An approach for Haff, R. P., & Toyofuku, N. (2008). X-ray detection of defects and con-
identifying of Fusarium infected maize grains by spectral analysis taminants in the food industry. Sensing and Instrumentation for
in the visible and near infrared region, SIMCA models, paramet- Food Quality and Safety, 2(4), 262–273.
ric and neural classifiers. International Journal Bioautomation, 14, Han, Z., & Gao, J. (2019). Pixel-level aflatoxin detecting based on deep
119–128. learning and hyperspectral imaging. Computers and Electronics in
Falade, T., Sultanbawa, Y., Fletcher, M. T., & Fox, G. (2017). Agriculture, 164, 104888.
Near infrared spectrometry for rapid non-invasive modelling of Harris, D. C., & Bertolucci, M. D. (1989). Symmetry and spectroscopy:
Aspergillus-contaminated maturing kernels of maize (Zea mays An introduction to vibrational and electronic spectroscopy. Courier
L.). Agriculture, 7(9), 77. Corporation.
Feng, Y.-Z., & Sun, D.-W. (2012). Application of hyperspectral imaging Heshmati, A., Zohrevand, T., Khaneghah, A. M., Nejad, A. S. M., &
in food safety inspection and control: A review. Critical Reviews in Sant’Ana, A. S. (2017). Co-occurrence of aflatoxins and ochratoxin
Food Science and Nutrition, 52(11), 1039–1058. A in dried fruits in Iran: Dietary exposure risk assessment. Food
Femenias, A., Gatius, F., Ramos, A. J., Sanchis, V., & Marín, S. (2020). and Chemical Toxicology, 106, 202–208.
Standardisation of near infrared hyperspectral imaging for quan- Hopfield, J. J. (1988). Artificial neural networks. IEEE Circuits and
tification and classification of DON contaminated wheat samples. Devices Magazine, 4(5), 3–10.
Food Control, 111, 107074. Hornberg, A. (2017). Handbook of machine and computer vision: The
Fernández-Ibañez, V., Soldado, A., Martínez-Fernández, A., & De guide for developers and users. John Wiley & Sons.
la Roza-Delgado, B. (2009). Application of near infrared spec- Hruska, Z., Yao, H., Kincaid, R., Darlington, D., Brown, R. L., Bhat-
troscopy for rapid detection of aflatoxin B1 in maize and bar- nagar, D., & Cleveland, T. E. (2013). Fluorescence imaging spec-
ley as analytical quality assessment. Food Chemistry, 113(2), troscopy (FIS) for comparing spectra from corn ears naturally and
629–634. artificially infected with aflatoxin producing fungus. Journal of
Filtenborg, O., Frisvad, J. C., & Thrane, U. (1996). Moulds in food Food Science, 78(8), T1313–T1320.
spoilage. International Journal of Food Microbiology, 33(1), 85–102. Huang, H., Liu, L., & Ngadi, M. O. (2014). Recent developments in
Firrao, G., Torelli, E., Gobbi, E., Raranciuc, S., Bianchi, G., & Locci, hyperspectral imaging for assessment of food quality and safety.
R. (2010). Prediction of milled maize fumonisin contamination by Sensors, 14(4), 7248–7276.
multispectral image analysis. Journal of Cereal Science, 52(2), 327– Jaillais, B., Roumet, P., Pinson-Gadais, L., & Bertrand, D. (2015).
330. Detection of Fusarium head blight contamination in wheat ker-
Franco, M. K. K. D., Yokaichiya, F., Kardjilov, N., & de Oliveira Ferraz, nels by multivariate imaging. Food Control, 54, 250–258.
A. C. (2015). Microfocus X-ray imaging of Brazil nuts for quality Jayas, D. S., & Singh, C. B. (2012). Grain quality evaluation by com-
control. Semina: Ciências Agrárias, 36(4), 2565–2575. puter vision. In Da-Wen Sun (Ed.), Computer vision technology in
Gaspardo, B., Del Zotto, S., Torelli, E., Cividino, S. R., Firrao, G., Della the food and beverage industries (pp. 400–421). Elsevier.
Riccia, G., & Stefanon, B. (2012). A rapid method for detection Jia, B., Wang, W., Ni, X. Z., Chu, X., Yoon, S. C., & Lawrence, K.
of fumonisins B1 and B2 in corn meal using Fourier transform C. (2020). Detection of mycotoxins and toxigenic fungi in cereal
near infrared (FT-NIR) spectroscopy implemented with integrat- grains using vibrational spectroscopic techniques: A review. World
ing sphere. Food Chemistry, 135(3), 1608–1612. Mycotoxin Journal, 13(2), 163–178.
Jiang, J., Qiao, X., & He, R. (2016). Use of Near-Infrared hyperspec- ban (Eds.), Food safety and preservation (pp. 1–12). Academic
tral images to identify moldy peanuts. Journal of Food Engineering, Press.
169, 284–290. Kos, G., Lohninger, H., & Krska, R. (2003). Development of a method
Jirsa, O., & Polišenská, I. (2014). Identification of Fusarium dam- for the determination of Fusarium fungi on corn using mid-
aged wheat kernels using image analysis. Acta Universitatis Agri- infrared spectroscopy with attenuated total reflection and chemo-
culturae et Silviculturae Mendelianae Brunensis, 59(5), 125–130. metrics. Analytical Chemistry, 75(5), 1211–1217.
Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, per- Kos, G., Sieger, M., McMullin, D., Zahradnik, C., Sulyok, M., Oner, T.
spectives, and prospects. Science, 349(6245), 255–260. . . . Krska, R. (2016). A novel chemometric classification for FTIR
Kaavya, R., Pandiselvam, R., Mohammed, M., Dakshayani, R., spectra of mycotoxin-contaminated maize and peanuts at regu-
Kothakota, A., Ramesh, S. V. . . . Ashokkumar, C. (2020). Appli- latory limits. Food Additives and Contaminants Part A-Chemistry
cation of infrared spectroscopy techniques for the assessment Analysis Control Exposure & Risk Assessment, 33(10), 1596–1607.
of quality and safety in spices: A review. Applied Spectroscopy https://doi.org/10.1080/19440049.2016.1217567
Reviews, 55(7), 1–19. Kotwaliwale, N., Singh, K., Kalne, A., Jha, S. N., Seth, N., & Kar,
Kalkan, H., Güneş, A., Durmuş, E., & Kuşçu, A. (2014). Non-invasive A. (2014). X-ray imaging methods for internal quality evaluation
detection of aflatoxin-contaminated figs using fluorescence and of agricultural produce. Journal of Food Science and Technology,
multispectral imaging. Food Additives & Contaminants: Part A, 51(1), 1–15.
31(8), 1414–1421. Krska, R., & Molinelli, A. (2009). Rapid test strips for analysis of
Kamilaris, A., & Prenafeta-Boldú, F. X. (2018). Deep learning in agri- mycotoxins in food and feed. Analytical and Bioanalytical Chem-
culture: A survey. Computers and Electronics in Agriculture, 147, istry, 393(1), 67–71.
70–90. Kulmyrzaev, A., Karoui, R., De Baerdemaeker, J., & Dufour, E. (2007).
Kandpal, L. M., & Cho, B.-K. (2014). A review of the applications of Infrared and fluorescence spectroscopic techniques for the deter-
spectroscopy for the detection of microbial contaminations and mination of nutritional constituents in foods. International Jour-
defects in agro foods. Journal of Biosystems Engineering, 39(3), 215– nal of Food Properties, 10(2), 299–320.
226. Labuza, T. P. (1980). The effect of water activity on reaction kinetics
Kandpal, L. M., Lee, S., Kim, M. S., Bae, H., & Cho, B.-K. (2015). Short of food deterioration. Food Technology, 34(4), 36–41.
wave infrared (SWIR) hyperspectral imaging technique for exam- Laca, A., Mousia, Z., Dıáz, M., Webb, C., & Pandiella, S. S. (2006). Dis-
ination of aflatoxin B1 (AFB1) on corn kernels. Food Control, 51, tribution of microbial contamination within cereal grains. Journal
171–176. of Food Engineering, 72(4), 332–338.
Karoui, R., & Blecker, C. (2011). Fluorescence spectroscopy measure- Lee, H. J., & Ryu, D. (2017). Worldwide occurrence of mycotoxins
ment for quality assessment of food systems—A review. Food and in cereals and cereal-derived food products: Public health per-
Bioprocess Technology, 4(3), 364–386. spectives of their co-occurrence. Journal of Agricultural and Food
Kautzman, M. E., Wickstrom, M. L., & Scott, T. A. (2015). The use of Chemistry, 65(33), 7034–7051.
near infrared transmittance kernel sorting technology to salvage Lee, K.-M., Davis, J., Herrman, T. J., Murray, S. C., & Deng, Y. (2015).
high quality grain from grain downgraded due to Fusarium dam- An empirical evaluation of three vibrational spectroscopic meth-
age. Animal Nutrition, 1(1), 41–46. ods for detection of aflatoxins in maize. Food Chemistry, 173, 629–
Kawano, Y., & Yanai, K. (2014). Food image recognition with deep con- 639. https://doi.org/10.1016/j.foodchem.2014.10.099
volutional features. Paper presented at the Proceedings of the 2014 Lenovich, L. M. (2017). Survival and death of microorganisms as
ACM International Joint Conference on Pervasive and Ubiquitous influenced by water activity. In Louis B. Rockland & Larry R.
Computing Adjunct Publication. Beuchat (Ed.), Water activity (pp. 119–136). Routledge.
Khaneghah, A. M., Fakhri, Y., Gahruie, H. H., Niakousari, M., & Levasseur-Garcia, C. (2012). Infrared spectroscopy applied to identifi-
Sant’Ana, A. S. (2019). Mycotoxins in cereal-based products during cation and detection of microorganisms and their metabolites on
24 years (1983–2017): A global systematic review. Trends in Food cereals (corn, wheat, and barley). In: Aflakpui, G. (Ed.), Agricul-
Science & Technology, 91, 95–105. tural science, (pp.185–196).
Kheiralipour, K., Ahmadi, H., Rajabipour, A., Rafiee, S., Javan- Levasseur-Garcia, C., Bailly, S., Kleiber, D., & Bailly, J.-D. (2015).
Nikkhah, M., Jayas, D. S., & Siliveru, K. (2016). Detection of fun- Assessing risk of fumonisin contamination in maize using near-
gal infection in pistachio kernel by long-wave near-infrared hyper- infrared spectroscopy. Journal of Chemistry, 2015, 1–10.
spectral imaging technique. Quality Assurance and Safety of Crops Li-Chan, E. C. Y., Ismail, A. A., Sedman, J., & Van de Voort, F. R.
& Foods, 8(1), 129–135. (2006). Vibrational spectroscopy of food and food products. In
Kimuli, D., Wang, W., Jiang, H., Zhao, X., & Chu, X. (2018). Applica- Handbook of vibrational spectroscopy.
tion of SWIR hyperspectral imaging and chemometrics for iden- Liang, P.-S., Slaughter, D. C., Ortega-Beltran, A., & Michailides, T. J.
tification of aflatoxin B1 contaminated maize kernels. Infrared (2015). Detection of fungal infection in almond kernels using near-
Physics & Technology, 89, 351–362. infrared reflectance spectroscopy. Biosystems Engineering, 137, 64–
Koehler, P., & Wieser, H. (2013). Chemistry of cereal grains. In 72.
Marco Gobbetti & Michael Gänzle (Eds.), Handbook on sourdough Lim, J., Kim, G., Mo, C., Oh, K., Yoo, H., Ham, H., & Kim, M. S. (2017).
biotechnology (pp. 11–45). Springer. Classification of Fusarium-infected Korean hulled barley using
Korada, S. K., Yarla, N. S., Putta, S., Hanumakonda, A. S., Lakkappa, near-infrared reflectance spectroscopy and partial least squares
D. B., Bishayee, A., . . . Peluso, I. (2018). Chapter 1 - A critical discriminant analysis. Sensors, 17(10), 2258.
appraisal of different food safety and quality management tools Liu, Q., Sun, K., Peng, J., Xing, M., Pan, L., & Tu, K. (2018). Identi-
to accomplish food safety. In A. M. Grumezescu & A. M. Hol- fication of bruise and fungi contamination in strawberries using
hyperspectral imaging technology and multivariate analysis. Food Moscetti, R., Monarca, D., Cecchini, M., Haff, R. P., Contini, M., &
Analytical Methods, 11(5), 1518–1527. Massantini, R. (2014). Detection of mold-damaged chestnuts by
Liu, Y., Galani Yamdeu, J. H., Gong, Y. Y., & Orfila, C. (2020). A near-infrared spectroscopy. Postharvest Biology and Technology, 93,
review of postharvest approaches to reduce fungal and mycotoxin 83–90.
contamination of foods. Comprehensive Reviews in Food Science Narvankar, D. S., Singh, C. B., Jayas, D. S., & White, N. D. G. (2009).
and Food Safety, 19(4), 1521–1560. Assessment of soft X-ray imaging for detection of fungal infection
Liu, Y., Pu, H., & Sun, D.-W. (2017). Hyperspectral imaging technique in wheat. Biosystems Engineering, 103(1), 49–56.
for evaluating food quality and safety during various processes: A Nations, F. a. A. O. o. T. U. (2015). The state of agricultural commodity
review of recent applications. Trends in Food Science & Technology, markets 2015–16. Rome: FAO.
69, 25–35. Nawaz, H., Majeed, M. I., Anwar, S., Ali, A., Khalid, A., Rashid, N., &
Los, A., Ziuzina, D., & Bourke, P. (2018). Current and future technolo- Byrne, H. J. (2019). Vibrational spectroscopy for food analysis. In
gies for microbiological decontamination of cereal grains. Journal Advances in noninvasive food analysis. CRC Press.
of Food Science, 83(6), 1484–1493. Nicol, R. W. (1998). Analysis of Fusarium toxins in maize and wheat
Lu, Y., Wang, W., Huang, M., Ni, X., Chu, X., & Li, C. (2020). Evalua- using thin layer chromatography. Mycopathologia, 142(2), 107–113.
tion and classification of five cereal fungi on culture medium using Nicolau, M., Pimentel, M. B. M., Tibola, C. S., Fernandes, J. M.
visible/near-infrared (Vis/NIR) hyperspectral imaging. Infrared C., & Pavan, W. (2018). Fusarium damaged kernels detection
Physics & Technology, 105, 103206. using transfer learning on deep neural network architecture. arXiv
Luo, X., Jayas, D. S., & Symons, S. J. (1999). Identification of damaged preprint arXiv:1802.00030v1. http://arxiv.org/abs/1802.00030.
kernels in wheat using a colour machine vision system. Journal of Noiseux, I., Long, W., Cournoyer, A., & Vernon, M. (2004). Simple
Cereal Science, 30(1), 49–59. fiber-optic-based sensors for process monitoring: An application
Ma, J., Sun, D.-W., Pu, H., Cheng, J.-H., & Wei, Q. (2019). Advanced in wine quality control monitoring. Applied Spectroscopy, 58(8),
techniques for hyperspectral imaging in the food industry: Princi- 1010–1019.
ples and recent applications. Annual Review of Food Science and Orina, I., Manley, M., & Williams, P. J. (2017a). Non-destructive tech-
Technology, 10, 197–220. niques for the detection of fungal infection in cereal grains. Food
Magan, N., & Aldred, D. (2006). Managing microbial spoilage in Research International, 100, 74–86.
cereal and baking products. In Food spoilage microorganisms. Orina, I., Manley, M., & Williams, P. J. (2017b). Use of high-resolution
Magan, N., Hope, R., Cairns, V., & Aldred, D. (2003). Post-harvest X-ray micro-computed tomography for the analysis of internal
fungal ecology: Impact of fungal growth and mycotoxin accumula- structural changes in maize infected with Fusarium verticillioides.
tion in stored grain. In Xiangming Xu John A. Bailey & B. Michael Food Analytical Methods, 10(9), 2919–2933.
Cooke (Eds.), Epidemiology of mycotoxin producing fungi (pp. 723– Orlandi, G., Calvini, R., Foca, G., & Ulrici, A. (2018). Automated
730). Springer. quantification of defective maize kernels by means of multivari-
Magan, N., & Lacey, J. (1988). Ecological determinants of mould ate image analysis. Food Control, 85, 259–268.
growth in stored grain. International Journal of Food Microbiology, Osborne, B. G., Fearn, T., & Hindle, P. H. (1993). Practical NIR spec-
7(3), 245–256. troscopy with applications in food and beverage analysis. Longman
Majeed, M., Khaneghah, A. M., Kadmi, Y., Khan, M. U., & Shariati, Scientific and Technical.
M. A. (2018). Assessment of ochratoxin A in commercial corn and Panda, B. K., & Shrivastava, S. L. (2019). Microwave assisted rapid
wheat products. Current Nutrition & Food Science, 14(2), 116–120. hydration in starch matrix of paddy (Oryza sativa L.): Process
Malloch, D. (1981). Moulds, their isolation, cultivation, and identifica- development, characterization, and comparison with conven-
tion. University of Toronto Press. tional practice. Food Hydrocolloids, 92, 240–249.
Maloney, P. V., Petersen, S., Navarro, R. A., Marshall, D., McKendry, Panikuttira, B., & O’Donnell, C. P. (2018). Process analytical technol-
A. L., Costa, J. M., & Murphy, J. P. (2014). Digital image analy- ogy for the fruit juice industry. In Gaurav Rajauria & Brijesh K.
sis method for estimation of Fusarium-damaged kernels in wheat. Tiwari (Eds.), Fruit juices (pp. 835–847). Elsevier.
Crop Science, 54(5), 2077–2083. Paul, A., Ghosh, S., Das, A. K., Goswami, S., Das Choudhury, S., &
Marín, S., & Ramos, A. J. (2016). Molds and mycotoxins in nuts. In Sen, S. (2020). A review on agricultural advancement based on com-
Parthena Kotzekidou (Ed.), Food hygiene and toxicology in ready- puter vision and machine learning. Singapore.
to-eat foods (pp. 295–312). Elsevier. Pearson, T. (2009). Hardware-based image processing for high-speed
Menesatti, P., Antonucci, F., Pallottino, F., Giorgi, S., Matere, A., inspection of grains. Computers and Electronics in Agriculture,
Nocente, F. . . . Costa, C. (2013). Laboratory vs. in-field spectral 69(1), 12–18.
proximal sensing for early detection of Fusarium head blight infec- Pearson, T. C., & Wicklow, D. T. (2006). Detection of corn kernels
tion in durum wheat. Biosystems Engineering, 114(3), 289–293. infected by fungi. Transactions of the ASABE, 49(4), 1235–1245.
Minz, P. S., Singh, C., & Sawhney, I. K. (2017). Machine vision tech- Pearson, T. C., Wicklow, D. T., Maghirang, E. B., Xie, F., & Dowell,
nology in food processing industry: Principles and applications— F. E. (2001). Detecting aflatoxin in single corn kernels by trans-
A review. In M. R. Goyal & D. K. Verma (Eds.), Engineering inter- mittance and reflectance spectroscopy. Transactions of the ASAE,
ventions in agricultural processing. (3–31). CRC Press. 44(5), 1247.
Mirjalili, S., & Hashim, S. Z. M. (2010). A new hybrid PSOGSA algo- Peiris, K., & Dowell, F. (2011). Determining weight and moisture
rithm for function optimization. Paper presented at the 2010 Inter- properties of sound and Fusarium-damaged single wheat kernels
national Conference on Computer and Information Application. by near-infrared spectroscopy. Cereal Chemistry, 88(1), 45–50.
Mitchell, T. M. (1997). Machine learning. Burr Ridge, IL: McGraw Peiris, K. H. S., Pumphrey, M. O., Dong, Y., Maghirang, E. B.,
Hill. Berzonsky, W., & Dowell, F. E. (2010). Near-infrared spectroscopic
method for identification of fusarium head blight damage and pre- Sendin, K., Williams, P. J., & Manley, M. (2018). Near infrared hyper-
diction of deoxynivalenol in single wheat kernels. Cereal Chem- spectral imaging in quality and safety evaluation of cereals. Critical
istry, 87(6), 511–517. Reviews in Food Science and Nutrition, 58(4), 575–590.
Perner, P., & Günther, T. (2005). Detection of hygiene-relevant parame- Senthilkumar, T., Jayas, D. S., White, N. D. G., Fields, P. G., & Gräfen-
ters from cereal grains based on intelligent image interpretation and han, T. (2016). Detection of fungal infection and ochratoxin A
data mining. contamination in stored barley using near-infrared hyperspectral
Peterson, L. E. (2009). K-nearest neighbor. Scholarpedia, 4(2), 1883. imaging. Biosystems Engineering, 147, 162–173.
Polder, G., Van Der Heijden, G., Waalwijk, C., & Young, I. T. (2005). Serranti, S., Cesare, D., & Bonifazi, G. (2013). The development of
Detection of Fusarium in single wheat kernels using spectral imag- a hyperspectral imaging method for the detection of Fusarium-
ing. Seed Science and Technology, 33(3), 655–668. damaged, yellow berry and vitreous Italian durum wheat kernels.
Proshlyakov, A., Yanniotis, S., & Blahovec, J. (2013). Pistachio dete- Biosystems Engineering, 115(1), 20–30.
rioration detected by X-ray absorption. Czech Journal of Food Sci- Shahin, M. A., & Symons, S. J. (2011). Detection of Fusarium damaged
ences, 31(2), 126–131. kernels in Canada Western Red Spring wheat using visible/near-
Putthang, R., Sirisomboon, P., & Sirisomboon, C. D. (2019). Shortwave infrared hyperspectral imaging and principal component analysis.
near-infrared spectroscopy for rapid detection of aflatoxin B1 con- Computers and Electronics in Agriculture, 75(1), 107–112.
tamination in polished rice. Journal of Food Protection, 82(5), 796– Shahin, M. A., & Symons, S. J. (2012). Detection of fusarium dam-
803. age in Canadian wheat using visible/near-infrared hyperspectral
Qi, X., Jiang, J., Cui, X., & Yuan, D. (2019). Identification of fungi- imaging. Journal of Food Measurement & Characterization, 6(1–4),
contaminated peanuts using hyperspectral imaging technology 3–11.
and joint sparse representation model. Journal of Food Science and Shen, F., Huang, Y., Jiang, X., Fang, Y., Li, P., Liu, Q. . . . Liu, X.
Technology, 56(7), 3195–3204. (2020). On-line prediction of hazardous fungal contamination in
Qi, X., Jiang, J., Cui, X., & Yuan, D. (2020). Moldy peanut kernel iden- stored maize by integrating Vis/NIR spectroscopy and computer
tification using wavelet spectral features extracted from hyper- vision. Spectrochimica Acta Part A: Molecular and Biomolecular
spectral images. Food Analytical Methods, 13(2), 445–456. Spectroscopy, 229, 118012.
Qiang, Z., Cheng-hai, L., Jing-kun, S., Yi-juan, C., Qun, L., Fu-guo, J., Shen, F., Wu, Q., Shao, X., & Zhang, Q. (2018). Non-destructive and
& Xian-zhe, Z. (2014). Rapid non-destructive detection for molds rapid evaluation of aflatoxins in brown rice by using near-infrared
colony of paddy rice based on near infrared spectroscopy. Jour- and mid-infrared spectroscopic techniques. Journal of Food Sci-
nal of Northeast Agricultural University (English Edition), 21(4), ence and Technology, 55(3), 1175–1184.
54–60. Shen, F., Zhao, T., Jiang, X., Liu, X., Fang, Y., Liu, Q. . . . Liu, X. (2019).
Qiang, Z., Fuguo, J., Chenghai, L., Jingkun, S., & Xianzhe, Z. (2014). On-line detection of toxigenic fungal infection in wheat by visi-
Rapid detection of aflatoxin B1 in paddy rice as analytical quality ble/near infrared spectroscopy. LWT, 109, 216–224.
assessment by near infrared spectroscopy. International Journal of Singh, C. B., & Jayas, D. S. (2011). Spectroscopic techniques for fungi
Agricultural and Biological Engineering, 7(4), 127–133. and mycotoxins detection. In Sarah De Saeger (Ed.), Determining
Qiao, X., Jiang, J., Qi, X., Guo, H., & Yuan, D. (2017). Utilization of mycotoxins and mycotoxigenic fungi in food and feed (pp. 401–414).
spectral-spatial characteristics in shortwave infrared hyperspec- Elsevier.
tral images to classify and identify fungi-contaminated peanuts. Singh, C. B., Jayas, D. S., Borondics, F., & White, N. D. G. (2011). Syn-
Food Chemistry, 220, 393–399. chrotron based infrared imaging study of compositional changes
Qin, J., Chao, K., Kim, M. S., Lu, R., & Burks, T. F. (2013). Hyper- in stored wheat due to infection with Aspergillus glaucus. Journal
spectral and multispectral imaging for evaluating food safety and of Stored Products Research, 47(4), 372–377.
quality. Journal of Food Engineering, 118(2), 157–171. Singh, C. B., Jayas, D. S., Paliwal, J., & White, N. D. G. (2012). Fungal
Qiu, R., Yang, C., Moghimi, A., Zhang, M., Steffenson, B. J., & Hirsch, damage detection in wheat using short-wave near-infrared hyper-
C. D. (2019). Detection of Fusarium head blight in wheat using a spectral and digital colour imaging. International Journal of Food
deep neural network and color imaging. Remote Sensing, 11(22), Properties, 15(1), 11–24.
2658. Siripatrawan, U., & Makino, Y. (2015). Monitoring fungal growth
Rahmani, A., Jinap, S., & Soleimany, F. (2010). Validation of the pro- on brown rice grains using rapid and non-destructive hyperspec-
cedure for the simultaneous determination of aflatoxins ochra- tral imaging. International Journal of Food Microbiology, 199, 93–
toxin A and zearalenone in cereals using HPLC-FLD. Food Addi- 100.
tives & Contaminants: Part A, 27(12), 1683–1693. Sirisomboon, C. D., Putthang, R., & Sirisomboon, P. (2013). Applica-
Ropelewska, E. (2018). The application of computer image analysis tion of near infrared spectroscopy to detect aflatoxigenic fungal
based on textural features for the identification of barley kernels contamination in rice. Food Control, 33(1), 207–214.
infected with fungi of the genus Fusarium. Agricultural Engineer- Siuda, R., Balcerowska, G., Kupcewicz, B., & Lenc, L. (2008). A mod-
ing, 22(3), 49–56. ified approach to evaluation of DON content in scab-damaged
Ruan, R., Li, Y., Lin, X., & Chen, P. (2002). Non–destructive deter- ground wheat by use of diffuse reflectance spectroscopy. Food
mination of deoxynivalenol levels in barley using near–infrared Analytical Methods, 1(4), 283.
spectroscopy. Applied Engineering in Agriculture, 18(5), 549. Smeesters, L., Meulebroeck, W., Raeymaekers, S., & Thienpont, H.
Saldaña, E., Siche, R., Luján, M., & Quevedo, R. (2013). Computer (2016). Non-destructive detection of mycotoxins in maize kernels
vision applied to the inspection and quality control of fruits using diffuse reflectance spectroscopy. Food Control, 70, 48–57.
and vegetables. Brazilian Journal of Food Technology, 16(4), 254– Soemantri, A. S., & Diyono, W. (2019). Rapid testing of aflatoxin by
272. using image processing and artificial neural network.
Srivastava, S., Mishra, G., & Mishra, H. N. (2018). Identification and of a method based on near-infrared hyperspectral imaging for the
differentiation of insect infested rice grains varieties with FTNIR detection and quantification of ergot bodies in cereals. Analytical
spectroscopy and hierarchical cluster analysis. Food Chemistry, and Bioanalytical Chemistry, 405(24), 7765–7772.
268, 402–410. Vermeulen, P. H., Pierna, J. A. F., Egmond, H. P. v., Dardenne, P.,
Steenhoek, L. W., Misra, M. K., Hurburgh Jr, C. R., & Bern, C. & Baeten, V. (2012). Online detection and quantification of ergot
J. (2001). Implementing a computer vision system for corn ker- bodies in cereals using near infrared hyperspectral imaging. Food
nel damage evaluation. Applied Engineering in Agriculture, 17(2), Additives & Contaminants: Part A, 29(2), 232–240.
235. Vithu, P., & Moses, J. A. (2016). Machine vision system for food grain
Su, W.-H., He, H.-J., & Sun, D.-W. (2017). Non-destructive and rapid quality evaluation: A review. Trends in Food Science & Technology,
evaluation of staple foods quality by using spectroscopic tech- 56, 13–20.
niques: A review. Critical Reviews in Food Science and Nutrition, Wan, Y. N., Lin, C. M., & Chiou, J. F. (2002). Rice quality classification
57(5), 1039–1051. using an automatic grain quality inspection system. Transactions
Suchowilska, E., & Wiwart, M. (2006). Multivariate analysis of image of the ASAE, 45(2), 379.
descriptors of common wheat (Triticum aestivum) and spelt (T. Wan, J., Chen, B., & Rao, J. (2020). Occurrence and preventive
spelta) grain infected by Fusarium culmorum. International Agro- strategies to control mycotoxins in cereal-based food. Compre-
physics, 20(4), 345. hensive Reviews in Food Science and Food Safety, 19(3), 928–
Sun, X. D., Su, P., & Shan, H. (2017). Mycotoxin contamination of 953.
maize in China. Comprehensive Reviews in Food Science and Food Wang, W., Heitschmidt, G. W., Ni, X., Windham, W. R., Hawkins,
Safety, 16(5), 835–849. S., & Chu, X. (2014). Identification of aflatoxin B1 on maize
Sunli, C., Jun, S., Hanping, M., Xiaohong, W., Pei, W., & Xiaodong, Z. kernel surfaces using hyperspectral imaging. Food Control, 42,
(2018). Non-destructive detection for mold colonies in rice based 78–86.
on hyperspectra and GWO-SVR. Journal of the Science of Food and Wang, W., Heitschmidt, G. W., Windham, W. R., Feldner, P., Ni, X., &
Agriculture, 98(4), 1453–1459. Chu, X. (2015). Feasibility of detecting aflatoxin B1 on inoculated
Tallada, J. G., Wicklow, D. T., Pearson, T. C., & Armstrong, P. R. (2011). maize kernels surface using Vis/NIR hyperspectral imaging. Jour-
Detection of fungus-infected corn kernels using near-infrared nal of Food Science, 80(1), M116–M122.
reflectance spectroscopy and color imaging. Transactions of the Wang, W., Ni, X., Lawrence, K. C., Yoon, S.-C., Heitschmidt, G. W.,
ASABE, 54(3), 1151–1158. & Feldner, P. (2015). Feasibility of detecting aflatoxin B1 in single
Tan, Y. (2016). GPU-based parallel implementation of swarm intelli- maize kernels using hyperspectral imaging. Journal of Food Engi-
gence algorithms. Morgan Kaufmann. neering, 166, 182–192.
Tao, F., Yao, H., Hruska, Z., Liu, Y., Rajasekaran, K., & Bhatnagar, Wang, W., & Paliwal, J. (2007). Near-infrared spectroscopy and imag-
D. (2019). Detection of aflatoxin B1 on corn kernel surfaces using ing in food quality and safety. Sensing and Instrumentation for
visible-near infrared spectroscopy. Journal of Near Infrared Spec- Food Quality and Safety, 1(4), 193–207.
troscopy, 28(2), 59–69. https://doi.org/10.1177/0967033519895686 Wicklow, D. T. (1999). Influence of Aspergillus flavus strains on afla-
Tao, F., Yao, H., Hruska, Z., Liu, Y., Rajasekaran, K., & Bhatnagar, toxin and bright greenish yellow fluorescence of corn kernels.
D. (2020). Detection of aflatoxin B 1 on corn kernel surfaces using Plant Disease, 83(12), 1146–1148.
visible-near infrared spectroscopy. Journal of Near Infrared Spec- Williams, P., Manley, M., Fox, G., & Geladi, P. (2010). Indirect detec-
troscopy, 28(2), 59–69. tion of Fusarium verticillioides in maize (Zea mays L.) kernels
Tao, F., Yao, H., Zhu, F., Hruska, Z., Liu, Y., Rajasekaran, K., & Bhat- by near infrared hyperspectral imaging. Journal of Near Infrared
nagar, D. (2019). A rapid and nondestructive method for simul- Spectroscopy, 18(1), 49–58.
taneous determination of aflatoxigenic fungus and aflatoxin con- Williams, P., & Norris, K. (1987). Near-infrared technology in the
tamination on corn kernels. Journal of Agricultural and Food agricultural and food industries. American Association of Cereal
Chemistry, 67(18), 5230–5239. Chemists, Inc.
Tekle, S., Måge, I., Segtnan, V. H., & Bj⊘ rnstad, Å. (2015). Near- Williams, P. J., Bezuidenhout, C., & Rose, L. J. (2019). Differentiation
infrared hyperspectral imaging of Fusarium-damaged oats (Avena of maize ear rot pathogens, on growth media, with near infrared
sativa L.). Cereal Chemistry, 92(1), 73–80. hyperspectral imaging. Food Analytical Methods, 12(7), 1556–
Tournas, V. H., Niazi, N. S., & Kohn, J. S. (2015). Fungal presence in 1570.
selected tree nuts and dried fruits. Microbiology Insights, 8, 1–10. Williams, P. J., Geladi, P., Britz, T. J., & Manley, M. (2012). Investi-
Tripathi, S., & Mishra, H. N. (2009). A rapid FT-NIR method for esti- gation of fungal development in maize kernels using NIR hyper-
mation of aflatoxin B1 in red chili powder. Food Control, 20(9), spectral imaging and multivariate data analysis. Journal of Cereal
840–846. Science, 55(3), 272–278.
Troller, J. (2017). Adaptation and growth of microorganisms in envi- Wiwart, M., Koczowska, I., & Borusiewicz, A. (2001). Estimation of
ronments with reduced water activity. In Louis B. Rockland & Fusarium head blight of triticale using digital image analysis of
Larry R. Beuchat (Eds.), Water activity and food. (101–117). Else- grain.
vier. Wold, S., Esbensen, K., & Geladi, P. (1987). Principal component
Turner, N. W., Subrahmanyam, S., & Piletsky, S. A. (2009). Analyti- analysis. Chemometrics and Intelligent Laboratory Systems, 2(1–3),
cal methods for determination of mycotoxins: A review. Analytica 37–52.
Chimica Acta, 632(2), 168–180. Wold, S., Sjöström, M., & Eriksson, L. (2001). PLS-regression: A basic
Vermeulen, P., Pierna, J. A. F., Van Egmond, H. P., Zegers, J., Dard- tool of chemometrics. Chemometrics and Intelligent Laboratory
enne, P., & Baeten, V. (2013). Validation and transferability study Systems, 58(2), 109–130.
Workman Jr, J. J., Mobley, P. R., Kowalski, B. R., & Bro, R. (1996). Yorulmaz, O., Pearson, T. C., & Çetin, A. E. (2011). Cepstrum based
Review of chemometrics applied to spectroscopy: 1985–95, Part I. feature extraction method for fungus detection.
Applied Spectroscopy Reviews, 31(1–2), 73–124. Yorulmaz, O., Pearson, T. C., & Çetin, A. E. (2012). Detec-
Wright, R. E. (1995). Logistic regression. In L. G. Grimm & P. R. tion of fungal damaged popcorn using image property covari-
Yarnold (Eds.), Reading and understanding multivariate statis- ance features. Computers and Electronics in Agriculture, 84, 47–
tics (pp. 217–244). American Psychological Association. 52.
Wu, D., & Sun, D.-W. (2013a). Advanced applications of hyperspectral Yuan, D., Jiang, J., Qiao, X., Qi, X., & Wang, W. (2020). An appli-
imaging technology for food quality and safety analysis and assess- cation to analyzing and correcting for the effects of irregular
ment: A review—Part II: Applications. Innovative Food Science & topographies on NIR hyperspectral images to improve identi-
Emerging Technologies, 19, 15–28. fication of moldy peanuts. Journal of Food Engineering, 280,
Wu, D., & Sun, D.-W. (2013b). Hyperspectral imaging technology: 109915.
A nondestructive tool for food quality and safety evaluation and Zapotoczny, P., Reiner, J., Mrzygłód, M., & Lampa, P. (2020). The use
inspection. In Stavros Yanniotis Petros Taoukis Nikolaos G. Sto- of polarized light and image analysis in evaluations of the severity
foros & Vaios T. Karathanos (Eds.), Advances in food process engi- of fungal infection in barley grain. Computers and Electronics in
neering research and applications (pp. 581–606). Springer. Agriculture, 169, 105154.
Wu, N., Jiang, H., Bao, Y., Zhang, C., Zhang, J., Song, W., . . . Liu, Zhang, N., Pan, Y., Feng, H., Zhao, X., Yang, X., Ding, C., & Yang, G.
F. (2020). Practicability investigation of using near-infrared hyper- (2019). Development of Fusarium head blight classification index
spectral imaging to detect rice kernels infected with rice false smut using hyperspectral microscopy images of winter wheat spikelets.
in different conditions. Sensors and Actuators B: Chemical, 308, Biosystems Engineering, 186, 83–99.
127696. Zhang, Z., Hu, X., Zhang, Q., & Li, P. (2016). Determination for multi-
Wu, Q., Xie, L., & Xu, H. (2018). Determination of toxigenic fungi and ple mycotoxins in agricultural products using HPLC–MS/MS via a
aflatoxins in nuts and dried fruits using imaging and spectroscopic multiple antibody immunoaffinity column. Journal of Chromatog-
techniques. Food Chemistry, 252, 228–242. raphy B, 1021, 145–152.
Wu, Q., & Xu, H. (2020). Design and development of an on-line flu- Zhao, X., Wang, W., Chu, X., Li, C., & Kimuli, D. (2017). Early detec-
orescence spectroscopy system for detection of aflatoxin in pista- tion of Aspergillus parasiticus infection in maize kernels using
chio nuts. Postharvest Biology and Technology, 159, 111016. near-infrared hyperspectral imaging and multivariate data anal-
Xing, F., Yao, H., Hruska, Z., Kincaid, R., Zhu, F., Brown, R. L. . . . Liu, ysis. Applied Sciences, 7(1), 90.
Y. (2017). Detecting peanuts inoculated with toxigenic and atoxienic Zhongzhi, H., & Limiao, D. (2020). Aflatoxin contaminated degree
Aspergillus flavus strains with fluorescence hyperspectral imagery. detection by hyperspectral data using band index. Food and Chem-
Xing, F., Yao, H., Liu, Y., Dai, X., Brown, R. L., & Bhatnagar, D. (2019). ical Toxicology, 137, 111159.
Recent developments and applications of hyperspectral imaging Zhu, F., Yao, H., Hruska, Z., Kincaid, R., Brown, R. L., Bhatna-
for rapid detection of mycotoxins and mycotoxigenic fungi in food gar, D., & Cleveland, T. E. (2016). Integration of fluorescence and
products. Critical Reviews in Food Science and Nutrition, 59(1), 173– reflectance visible near-infrared (VNIR) hyperspectral images for
180. detection of aflatoxins in corn kernels. Transactions of the ASABE,
Yan, T., Zhang, Q., Wang, D., Li, P., Tang, X., & Zhang, W. (2019). 59(3), 785–794.
Determination of deoxynivalenol by ELISA and immunochro-
matographic strip assay based on monoclonal antibodies. Toxin
Reviews, 1–7.
Yanniotis, S., Proshlyakov, A., Revithi, A., Georgiadou, M., & Bla- S U P P O RT I N G I N F O R M AT I O N
hovec, J. (2011). X-ray imaging for fungal necrotic spot detection Additional supporting information may be found online
in pistachio nuts. Procedia Food Science, 1, 379–384. in the Supporting Information section at the end of the
Yao, H., Hruska, Z., Brown, R. L., & Cleveland, T. E. (2006). Hyper-
article.
spectral bright greenish-yellow fluorescence (BGYF) imaging of afla-
toxin contaminated corn kernels.
Yao, H., Hruska, Z., Kincaid, R., Brown, R., Cleveland, T., & Bhatna-
gar, D. (2010). Correlation and classification of single kernel fluo-
How to cite this article: Mishra G, Panda BK,
rescence hyperspectral data with aflatoxin concentration in corn
Ramirez WA, Jung H, Singh CB, Lee S.-H., Lee I.
kernels inoculated with Aspergillus flavus spores. Food Additives
and Contaminants, 27(5), 701–709. Research advancements in optical imaging and
Yao, H., Hruska, Z., Kincaid, R., Brown, R. L., Bhatnagar, D., & Cleve- spectroscopic techniques for nondestructive
land, T. E. (2013). Detecting maize inoculated with toxigenic and detection of mold infection and mycotoxins in
atoxigenic fungal strains with fluorescence hyperspectral imagery. cereal grains and nuts. Compr Rev Food Sci Food
Biosystems Engineering, 115(2), 125–135. Saf. 2021;1–40.
Yeni, F., Acar, S., Polat, Ö., Soyer, Y., & Alpas, H. (2014). Rapid and https://doi.org/10.1111/1541-4337.12801
standardized methods for detection of foodborne pathogens and
mycotoxins on fresh produce. Food Control, 40, 359–367.

Detection (Optical)

Uploaded by

Copyright:

Available Formats

Detection (Optical)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Detection (Optical)

Uploaded by

Copyright:

Available Formats

Received: 4 September 2020 Revised: 7 June 2021 Accepted: 15 June 2021

COMPREH ENSIVE REVIEWS IN FOOD SCIENCE AND FOOD SAFETY

Research advancements in optical imaging and

1 UniSA STEM, University of South

Australia, Mawson Lakes, South Abstract

1 INTRODUCTION through spores then develop visible branched filaments,

2.4 Fluorescence spectroscopy 2.6 Image/spectra processing and

FIGURE 1 Various processing steps involved in optical and spectrometric techniques

FIGURE 2 Various machine learning techniques

30 macromolecules 0.994 and (2008)

cm–1 pigmentation due 75–90%; rates: 3–7% Cervel-

Optical imaging and spectroscopic techniques have been

You might also like