Content uploaded by Wasswa Shafik
Author content
All content in this area was uploaded by Wasswa Shafik on Jun 21, 2023
Content may be subject to copyright.
Received 5 May 2023, accepted 31 May 2023, date of publication 9 June 2023, date of current version 19 June 2023.
Digital Object Identifier 10.1109/ACCESS.2023.3284760
A Systematic Literature Review on Plant Disease
Detection: Motivations, Classification Techniques,
Datasets, Challenges, and Future Trends
WASSWA SHAFIK 1, (Member, IEEE), ALI TUFAIL 1, (Senior Member, IEEE),
ABDALLAH NAMOUN 2, (Member, IEEE),
LIYANAGE CHANDRATILAK DE SILVA 1, (Senior Member, IEEE),
AND ROSYZIE ANNA AWG HAJI MOHD APONG1
1School of Digital Science, Universiti Brunei Darussalam, Gadong BE1410, Brunei Darussalam
2Faculty of Computer and Information Systems, Islamic University of Madinah, Madinah 42351, Saudi Arabia
Corresponding author: Ali Tufail (ali.tufail@ubd.edu.bn)
This work was supported by the Universiti Brunei Darussalam and the Ministry of Education (MoE) Brunei Darussalam.
ABSTRACT Plant pests and diseases are a significant threat to almost all major types of plants and global
food security. Traditional inspection across different plant fields is time-consuming and impractical for
a wider plantation size, thus reducing crop production. Therefore, many smart agricultural practices are
deployed to control plant diseases and pests. Most of these approaches, for example, use vision-based
artificial intelligence (AI), machine learning (ML), or deep learning (DL) methods and models to provide
disease detection solutions. However, existing open issues must be considered and addressed before AI
methods can be used. In this study, we conduct a systematic literature review (SLR) and present a detailed
survey of the studies employing data collection techniques and publicly available datasets. To begin the
review, 1349 papers were chosen from five major academic databases, namely Springer, IEEE Xplore,
Scopus, Google Scholar, and ACM library. After deploying a comprehensive screening process, the review
considered 176 final studies based on the importance of the method. Several crops, including grapes,
rice, apples, cucumbers, maize, tomatoes, wheat, and potatoes, have tested mainly on the hyperspectral
imagery and vision-centered approaches. Support Vector Machines (SVMs) and Logistic regression (LR)
classifiers demonstrated an increased accuracy in experiments compared to traditional classifiers. Besides
the image taxonomy, disease localization is depicted in these approaches as a bottle neck to disease detection.
Cognitive CNNs with attention mechanisms and transfer learning are showing an increasing trend. There is
no standard model performance assessment though the majority use accuracy, recall, precision, F1 Score, and
confusion matrix. The available 11 datasets are laboratory and in-field based, and 9 are publicly available.
Some laboratory-based datasets are considerably small, making them impractical in experiments. Finally,
there is a need to avail models with fewer parameters, implementable on small devices and large datasets
accommodating several crops and diseases to have robust models.
INDEX TERMS Convolutional neural networks, deep learning, image processing, machine learning, plant
disease detections.
ABBREVIATIONS
ABCK Artificial Bee Colony Method
ACM Association for Computing Machinery
AI Artificial Intelligence
The associate editor coordinating the review of this manuscript and
approving it for publication was Sunil Karamchandani .
ANN Artificial Neural Networks
ASDE Advanced Segmented Dimension Extraction
CART Classification and Regression Tree
CBAM Convolution Block Attention Module
CBAM Carbon Border Adjustment Mechanism
CGAN Conditional Generative Adversarial Network
CNN Convolutional Neural Network
59174
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
DL Deep Learning
DP Disease Prediction
DT Decision Trees
FN False Negatives
FP False Positives
FSL Few-Shot Learning
HIoT Heterogeneous Internet of Things
IEEE Institute of Electrical and Electronics Engineers
IoT Internet of Things
IP Image Processing
KNN K-Nearest Neighbors
MAE Mean Absolute Error
mAP Mean Average Precision
NB Naïve Bayes
NB Naive Bayes classifier
OLPSO Orthogonal learning particle swarm optimization
PDD Plant Disease Detection
PDI Plant Disease Identification
PVD Plant Village Dataset
RAN Residual Attention Network
RF Random Forest
RF Random Forest
RMSE Root Mean Square Error
RPN Risk Priority Number
RPN Region Proposal Network
RSE Relative Standard Error
SCNN Sequential Convolutional Neural Network
SCNN Shallow Convolutional Neural Network
SIFT Scale Invariant Feature Transform
SMO Spider monkey optimization
SPAM Subtractive pixel adjacency model
SSD Single-shot detector
SVM Support Vector Machines
TF Transfer Learning
TN True Negatives
TP True Positives
UAV Unmanned Aerial Vehicle
UN United Nations
WHO World Health Organization
XGB Extreme gradient boosting
I. INTRODUCTION
Plant pathogens and pests cause substantial reduction in plant
production depending on adverse seasonal and environmental
conditions leading to economic and social losses. Contem-
porary pests and pathogen management depend profoundly
on pesticide application, for example, herbicides, fungicides,
and insecticides [1], [2], [3], [4], [5]. According to the United
Nations’ global goals, zero hunger is approximated to be
accomplished by 2030 [6]. Early plant disease detection and
classification significantly increase food security to attain the
target. According to the World Health Organization (WHO),
as of 2021, the total global population is approximately
7.837 billion, indicating that the demand for food increases
along with the global population. Prioritizing food production
is vital to addressing the issue of global hunger since the
UN reported that global hunger numbers rose to as much
as 828 million in 2021 [3], [7], [8], [9], [10]. Farmers have
widely used chemicals, for instance, insecticides and herbi-
cides, to man- age plant diseases and pests and increase crop
yields. In the short run, these chemicals increase plant yields
without decreasing crop quality. However, in the long run,
for instance, in terms of health, chemicals such as herbicides
pollute the ground where they are used and the environment
in general. This is because of their chemical toxicity, which
poses severe health risks and causes nearly three hundred
thousand deaths annually, as demonstrated in [11], [12], [13],
and [14].
Detecting plant diseases early can minimize the use
of hazardous chemicals in plant growth and protection.
For example, IoT methods for detecting and diagnosing
plant health status have been proposed, as have recent
artificial intelligence (AI) methods such as deep and
machine learning and image processing (IP) centered disease
detection [15], [16], [17], [18]. Various methods have been
presented in recent years to appreciate AI’s increased agricul-
tural application, for example, deep learning (DL) approaches
in building advanced AI models for plant disease detec-
tion [17], [18], [19], [20]. Despite advances in AI-based meth-
ods, plant disease detection in natural environments remains
an issue. Drones are being used to monitor plant health to
track diseases. As a result, compared to digital cameras, these
semi- or autonomous aircraft provide robust and dependable
vision methods that can be used on various crops [21].
This study provides a detailed overview of the most
recent disease identification and PDD approaches, focus-
ing on the most used AI (ML and DL) and IP algorithms
for disease identification. Moreover, it systematically eval-
uates the limitations, strengths, and significant traits of these
methods in real-world applications. A search of five major
academic research databases, namely Springer, IEEE Xplore,
Scopus, Google Scholar, and ACM, yielded 1349 papers
for this review. Some keywords that were used during the
search include ‘‘disease classification,’’ ‘‘disease identifica-
tion,’’ ‘‘crop disease detection,’’ ‘‘PDD,’’ ‘‘ML,’’ ‘‘DL,’’ and
‘‘IP.’’ According to the significance of the recent surveys, the
prominence of the approach, performance in PDD, and the
used datasets, 176 studies were selected for the study.
The remainder of this systematic literature review is orga-
nized into six sections: Section II presents a summary of the
study background and the motivations. Section III summa-
rizes the current survey literature in the PDD field and the
rationale for this study. Section IV depicts the essentials of
the survey process, including review questions, followed by
the method of approach. An analysis of the current studies on
PDDs and localization is presented in Section V. Lastly, the
conclusion of the review and the recommendations on future
trends are demonstrated in Section VI.
II. STUDY BACKGROUND
Plant diseases have a global effect on plant production. There-
fore, farmers are supposed to attain expert knowledge and
VOLUME 11, 2023 59175
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
FIGURE 1. Plant disease classification.
thorough training to distinguish early plant pests or viral
symptoms and take appropriate action to prevent disease
continuity. PDD control can help economic development by
reducing hunger and saving the environment through reduced
chemical fertilizer utilization. Environmental factors primar-
ily cause plant diseases, and pathogens (like bacteria, worms,
viruses, fungi, and protozoa) are defined as diseases in plant
pathology. Several plant diseases commonly occur because
of a variety of factors. For instance, depending on its nature,
soil, seed, or air type, it can be caused by a pandemic,
epidemic, or endemic. Other factors include symptoms and
significant causes such as blight, rots, and viruses, as shown
in Fig. 1. Early symptoms are essential in PDD because con-
ventional traits for identifying in-field diseases are based on
several factors, including the type of disease, its color, pattern,
appearance, and location on the plant. These symptoms on
various sections of plants, like the stems, fruits, and leaves,
among others, are primarily utilized in PDD. Farmers and
agricultural experts use traditional surveillance to identify
disease categories. High-tech agricultural systems that use
vision-based learning approaches for PDD can effectively
increase crop yields. There has been an improvement in
efficiency in identifying plant diseases using AI for early
diagnosis and smart inspection automation. Few valid case
studies in third-world countries use automatic approaches
in agriculture. Despite the contributions of the remarkable
endeavors available, a few factors continue to make real-time
PDD complex. The main objective of this review is to present
techniques, available datasets, and challenges in plant disease
detection that need to be addressed to develop comprehen-
sive, intelligent agricultural methods for monitoring and diag-
nosing early plant detection. The following are the main other
contributions to this review:
•Based on the current surveys, this review is the first sys-
tematic study on image-based PDD approaches covering
both localization and disease classification.
•The study shows a complete data collection and prepro-
cessing strategy for PDD used in academia and business.
•The study also looks at three crucial methods: deep
learning, machine learning, and image processing relat-
ing to plant diseases.
•Deep learning methods and their applicability to IoT-
centered smart farming solutions are also investigated.
•All public datasets that can be used with this research
paradigm are examined in the study, and their details are
given.
III. RELATED LITERATURE
Several studies on ML and DL approaches used in agriculture
have been evaluated, leaving a gap to comprehensively exam-
ine image-centered plant disease detection. Most recently,
a review was made of the imperativeness of existing PDD
methods and included segmentation, classification, localiza-
tion, and disease techniques [22]. The study [23] concentrated
on the performance of the CNN method in PDD, primarily on
fruits, vegetables, and various plants. Other studies, like [24],
59176 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
focused on barley, maize, rice, soy- beans, and wheat and
compared the benefits and drawbacks of various IPs, segmen-
tations, extractions, feature selections, and classifications.
The main goal of this survey is to demonstrate the use of
electric impedance spectroscopy, hyperspectral imaging, and
fluorescence spectroscopy, among other things.
This showed that many problems still need to be solved,
such as real-time inspection of a larger farmyard. The authors
compared the performance of various approaches based on
feature extraction, classification, segmentation, and prepro-
cessing using IP. They noticed that the segmentation strat-
egy known as ‘‘k-means’’ was frequently used to identify
plant diseases [25]. A comprehensive review of existing
approaches in the early 2000s was conducted, with implica-
tions for fully automated PDD and identification using digital
approaches [26]. Investigation of the literature was done with
three different methodologies, namely, image processing,
deep and machine learning in one of the early reviews on
PDD, namely, molecular, spectral, and biogenic compound
profiling [27]. The authors revealed that using autonomous
agricultural aircraft, imaging, and spectroscopic methods can
be merged to aid in detection and identification at an early
stage; corresponding to their research, they concluded that the
3D dataset is more efficient in early PDD.
Studies [28] and [29] generally focused on the CNN model
as it is used in the PDD and concluded by showing that there
is still a gap to fill in improving the methods when con-
sidering identical visual environments, respectively. Exist-
ing surveys have covered a few strategies and highlighted
numerous significant problems, but there is a need to assess
the in-situ application potential of various techniques [30].
Since datasets are a vital part of making artificial intelligence
applications, it is not impractical to ignore data collection
techniques and the fact that there are public data sources.
The use of datasets in assessing the development of field
studies is widely appreciated. It begins with data capture
techniques and publicly available datasets. This work de-
livers a thorough assessment of vision-based ML strategies
for PDD. In addition to a brief overview of the devices utilized
for data capture, the study describes the capture conditions,
the regions from which datasets are obtained, and prepro-
cessing procedures. The review also discusses the various
kinds of openly accessible datasets that scientists in the field
use. In addition, this review examines the influence of cur-
rent trends in DL, localization, TF, and attention processes,
together with lightweight approaches in PDD. In addition
to the necessity for additional research on TF strategies, the
impact of lightweight models in real- time applications must
also be investigated. Also, CNNs have surfaced as a potential
approach in recent research years. Their suitability for in-situ
identification of plant diseases must be evaluated. There is no
available study focusing on attention and localization strate-
gies for PDD after an exhaustive search. This review provides
a logical evaluation of existing methodologies with intentions
and a methodical explanation of several classifications and
localization-based PDD methods. An example of existing
TABLE 1. Recent surveys on plant disease classification and detection.
studies with various goals is demonstrated in Table 1, which
shows the used method, current citation, and study scope.
IV. REVIEW PROCEDURE
This study evaluates vision-based PDD and identification
methods and early demonstrated IP, ML, and DL PDD
VOLUME 11, 2023 59177
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
methods. In addition, the study addresses early and modern
disease categorization and localization approaches. This sys-
tematic review follows the study presented in [34] on the
planning, execution, and evaluation of the review process.
Planning establishes the review questions, information on
sources, selection criteria, and evaluation quality procedures.
Research articles were selected using keywords from relevant
academic databases throughout the execution process. The
summary step critically evaluates the existing techniques,
their strengths and weaknesses, and their suitability for actual
in-field implementation.
A. REVIEW QUESTIONS
Besides advancing PDDs and symptom analysis, prema-
ture plant disease identification (PDI) remains challenging.
Researchers have analyzed PDI, visual resemblance, and
crop performance. The necessity for IoT-based cyber physical
agricultural systems raises the following question: ‘‘What
is the present state of classical IP, DL, and ML methods
for PDDs in terms of their application in the field?’’ Other
supplementary questions are also included to assist in the
discovering of appropriate PDD technologies. The basis of
this review process comprises the relevant research questions
listed below.
•RQ1: Which plants and artificial intelligence methods
are the best for disease identification?
•RQ2: How does the practice of localization help farmers
find ways to treat plant diseases in real-time?
•RQ3: What are some more innovative approaches to the
detection of plant diseases?
•RQ5: What kind of impact do lightweight CNNs have
on the process of identifying plant diseases or detecting
them?
•RQ6: How effective have the techniques based on trans-
fer learning been in the identification of disease?
•RQ7: In what ways do conventional image processing
and machine learning approaches add to the complexity
of detecting plant diseases?
•RQ8: Is there an adequate public dataset for machine and
deep learning in the PDD and identification?
•RQ9: Are researchers modeling plant disease detection
techniques that employ data collection and preprocess-
ing approaches?
B. SEARCHING APPROACH
Within this study, the searched research papers were obtained
using different search strings, namely, ‘‘crop disease detec-
tion,’’ ‘‘plant disease identification,’’ ‘‘plant disease classi-
fication,’’ ‘‘ML,’’ ‘‘DL,’’ and ‘‘IP,’’ from five top known
academic databases, including ACM, Springer, Scopus,
IEEE Xplore, and Google Scholar. After retrieving arti-
cles from several search databases, we based our deci-
sion on the research selection criteria to decide which
articles should be included or excluded, as demonstrated in
Table 2.
TABLE 2. Selection criteria.
TABLE 3. Quality assessment measures.
C. SELECTION AND QUALITY EVALUATION
Articles that were not initially written in English or did not
include all relevant facts were not considered. Furthermore,
studies that were discovered to be of high quality because
of the search were removed from the compilation. The qual-
ity chosen for further review was determined based on the
score achieved using the quality assessment benchmarks. The
five questions shown in Table 3were used as the primary
means to sort out the overall quality of the article. The scores
were given to the articles to determine whether they satisfied
inclusion requirements (1 represented ‘‘yes,’’ 0.5 illustrated
‘‘partial,’’ and 0 showed ‘‘nil’’). The final score was found
by adding all the points given for each question about a
specific article. In addition, the articles are sorted according
to the quality score to find the relevant articles to the search.
Based on the scores, 176 studies published between 2012 and
2022 were chosen to be included in this review. The identifi-
cation of plant diseases was approached in these research pub-
lications using several techniques, including IP, DL, and ML.
D. EXTRACTION OF ARTICLES
The abstracts and titles of the publications were initially
reviewed, and then the complete contents of the articles were
59178 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
FIGURE 2. PRISMA systematic literature review flowchart.
FIGURE 3. Utilized in articles with the focus on image processing, deep and machine
learning approaches.
VOLUME 11, 2023 59179
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
examined. The PRISMA was used to screen potential articles
and choose those that passed the screening as early as possi-
ble [35]. Out of a total of 1349 papers that were obtained from
several databases, 169, 364, 357, 403, and 44 articles were
obtained from Google Scholar, IEEE Xplore, ACM, Scopus,
and Springer, respectively. Once the duplicates were taken out
of the list, 579 items were chosen. After reading the abstracts
of the papers, the first round of screening resulted in the
elimination of 125 of them since they were irrelevant to the
study. After reading the articles in their entirety for the first
assessment, the second examination focused on determin-
ing whether the remaining 454 articles were relevant to the
topic. Some other articles were removed because they lacked
sufficient information, specifics, and rigor to be considered
credible sources. Fig. 2illustrates the stages of screening and
selecting the research that was included in the study.
V. RESULTS AND DISCUSSION
Within this section, the identified research questions are
reacted to by detailing the contributions made by the
researchers who contributed to each of the 176 research
studies. Also, both quantitative and qualitative assessments of
these studies are provided. Fig. 3below shows the matrices
of the used articles within the study.The significance of these
methodologies for PDDs was considered in the review.
A. RESULTS
1) RQ1: WHICH PLANTS AND ARTIFICIAL INTELLIGENCE
METHODS ARE THE BEST FOR DISEASE IDENTIFICATION?
A wide diversity of plant species is considered when diag-
nosing plant diseases in various parts of the world. It is
evident that most of the research efforts done in the field
of PDD are based mainly in India and China. Plants like
apples, soybeans, avocados, bananas, citrus fruits, maize,
coffee, millet, cucumbers, grapes, oil palms, potatoes, rice,
strawberries, sugar beets, tomatoes, and wheat, among vari-
ous types of crops, are grown worldwide and are considered
for disease detection testing. The results of the PDD tests
conducted on frequently grown crops, such as those listed
in Table 4, include grapes, rice, apples, cucumbers, maize,
tomatoes, wheat, and potatoes. This section also includes a
discussion of the ultimate results obtained for each crop type
and the vision-based approach utilized to obtain those results.
There have been 17 studies on rice that use a variety of IP,
ML, and DL techniques. Some of these studies include [35].
By utilizing the mobile network version 2 model with the SE
block-based attention approach, the authors reached the most
significant accuracy level, which was 99.33%. Furthermore,
11 cucumber studies, like in [36], were examined, employing
the classification and regression trees classifier and attain-
ing 98.7% accuracy. Using a sequential CNN technique that
included squeeze-and-excitation blocks and CBAM in the
shuffle network model, the authors [23] obtained a level of
accuracy of 99.14% out of the 11 research studies on grapes.
Using a feedback network, an area suggestion network, and
the categorization, the study demonstrated the most excel-
lent accuracy among the ten-tomato literature, which was
99.7% [37]. In [38], PDD in corn was discussed by using
a modified Alex Network simulation with multi-scale and
dilated convolutions and obtained an accuracy of 98.62%.
In studies on citrus, the Bagged Tree classifier, which
was used in eight different citrus-related research projects,
attained the most incredible accuracy of 99.9%. Their study
presents a survey of six prior studies on the detection of apple
disease [39].
Using the same dataset, the authors applied the
DenseNet121 model, which obtained an accuracy of
93.71% [40]. The author in [41] is the only one of the
five re- searchers on potato species to attain an accuracy of
99.67% using the support vector machines (SVM) classifier.
The highest level of accuracy was obtained by employing
a six-layer CNN method for feature extraction with a total
accuracy of 97.3%. The data quality impacts the models’ per-
formance; therefore, attention enhances the diagnosis of rice
and grape diseases. We further subdivide these approaches
into two models.
a: HYPERSPECTRAL IMAGING
Researchers have utilized many diverse methods for identify-
ing plant diseases in addition to the more common machine
learning and CNN-based approaches. A concise discussion of
various alternative methods for constructing PDD methods
is presented. Hyperspectral imagery aids in plant disease
diagnosis and plant stress level estimations. Hyperspectral
imaging techniques in the electromagnetic spectrum use a
more comprehensive range of wavelengths, including visible
and infrared light. The author utilised the SVM and spectral
vegetation indices in [20] to predict disease according to
the dataset available in 2010 [42]. Their model attained an
accuracy of 97% when applied to the dataset consisting of
sugar beetroot leaves. The authors utilized KNN and NB
classifiers to identify oil palm leaf diseases. On the vali-
dation DSS, the proposed method attained an accuracy of
92% [43]. To construct a disease classification model utiliz-
ing hyperspectral images, we first performed several image
preprocessing approaches, and then we used a super vector
machine classifier [44]. The authors [45] used aerial photos
obtained via the unmanned aerial vehicle (UAV) with hyper-
spectral sensors to perform tasks. It was demonstrated that
their model has a classification accuracy of 92.50% when
applied to the sugarcane datasets. An ensemble technique
using hyperspectral images was developed to identify plant
diseases and stresses like sand stress [46].
Using the SVM on photos captured in the near-infrared
and shortwave infrared, in terms of potato virus Y detec-
tion, an accuracy of 89.8% was obtained. A nondestructive
remote sensing approach with DT and multilayer perceptron
was used to identify avocado tree deficits. For a variety of
conditions, the model’s accuracy ranged from 82% to 100%.
A nondestructive remote sensing approach with multilayer
perceptron applying DT was used to identify avocado tree
59180 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
deficits. The Monte Carlo approach is combined with support
vector machines (SVM) for disease diagnosis in grapevines.
The model had an efficiency that ranged from 66.67% to
89.93% [22], [47], [48], respectively.
b: VISION-CENTERED SOLUTIONS
RiceTalk is an application that was built after the weather
gathering station AgriTalk served as the inspiration [49].
They concentrated their efforts on gathering meteorological
data to predict rice blasts. To construct a three-layer CNN
model, sensor data was gathered from several different geo-
graphic areas. Based on data collected for 14 days, their
method reached an accuracy of 89.4%. An advanced process
known as airport surface detection equipment utilizing HIT
was developed. Their approach demonstrated remarkable per-
formance, with an astounding accuracy of 98.5% in disease
categorization [50], [48]. Finally, an IoT-built system for the
PDI and pests, in which unmanned aerial vehicle devices
gathered photos of farms following a predetermined and pre-
planned approach, and the recorded images were uploaded to
the cloud for plant disease evaluation [22].
An IoT-aided plant disease diagnosis where sensors were
utilized to gather plant images and a convolutional neural net-
work was employed for disease assessment [51], an accuracy
of 99.2% was obtained using the publicly available PVD.
An intelligent agricultural system consists of three layers:
perception, edge computing, and data analysis. The authors
further compared the various approaches that are accessible,
including the problems as- sociated with their dependability
and security. The estimate of coffee production was the pri-
mary emphasis of their model. XGB model with the highest
performance demonstrated that its root-mean-square error
value was 0.008, its MAE (loss function) value was 0.032,
and its RSE value was 0.585 [52].
IoT-enabled fuzzy network classified the images after
extracting the image features using SIFT, optimizing the
features with the firefly method, and using SIFT for feature
extraction [53]. On a dataset consisting of Alstonia Scholaris
trees, their model reaches a uniqueness of 80.66% and a sen-
sitivity of 80.18%, performing the task of disease detection
using a simulation of IoT systems. They have demonstrated
that their model obtains a maximum accuracy of 91.56%
using data from PlantVillage and the trigonometric classifi-
cation based on neural networks.
An end-to-end approach combines IoTs and DP to diag-
nose rust illness in pearl millet [54]. They classified the
diseases via a 7-layer CNN, utilized the GradCAM model
to visualize the characteristics of the features and reported
an accuracy of 98.78%. Moreover, authors in [55] produced
a prototype for an intelligent agricultural system that uses
a mobile application. They gathered data on the soil sam-
ples and weather to determine the trends in water con-
sumption. They used XGB, NB, RF, and DT on the dataset
that was collected, and they claimed that RF performed
the best out of all available options obtaining 91.59% of
accuracy.
TABLE 4. Current plant disease detection models with globally grown
plants.
2) RQ2: HOW DOES THE PRACTICE OF LOCALIZATION
CONTRIBUTE TO THE DEVELOPMENT OF REAL-TIME
PLANT DISEASE METHODS?
In in-field datasets, image taxonomy is not the only concern;
disease localization also falls into that category. The authors
in [109] have included a CNN equipped with a CBAM as part
of the primary network of the faster RCNN to facilitate the
extraction of visual features. Their model achieved an accu-
racy of classification of 99.95% and a mean average precision
of 77.54% on all 3531 photos included in the strawberry
dataset’s four categories. In [63], we utilized GoogleNet to
extract visual features and the SDD structure in localization.
On the Apple dataset, their model had a mean accuracy
percentage of 78.80%. Reference [73] have used Retinex for
fine-tuning, IP, feature extractions, and RPN for detecting and
localizing northern maize leaf blight in maize leaves. Their
multilevel feature fusion model on the test dataset got an mAP
score of 91.83%.
VOLUME 11, 2023 59181
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
An investigation on using the MobileNetv2-YOLOv3
model in diagnosing the illness known as tomato gray leaf
spot They fed the approach with 2385 tomato leaves taken
under field conditions and used transfer learning to train the
model. Their proposed approach obtained 93.24%, 91.32%,
and 86.98% on F1-score, an average accuracy, and an IoU
score, respectively [77]. Authors [35] have further used the
RPN segment in conjunction with VGG16 and ZFnet. The
trained prototype used a wheat data dataset and attained
a mAP of 88.9% with the VGG16 model. CNN’s model
for localizing northern leaf blight using maize leaf pho-
tos They created an end-to- point method called Cascaded
MRCNN for a segmentation loss function of 2.0437 and
measured 91% accuracy [71]. Image classification methods
were utilised, including translations, YOLO version 2, and
DarkNet-19 approaches. The technique attained 87% canopy
mAP data [111]. Using the apple disease dataset, a CNN
model for the apple disease localization was proposed and
demonstrated a score of 83.12% mAP for the five-class [65].
Using an improved localization SSD method for the plant
disease was proposed, and their model presented 92.2% mAP
using PVD [112].
Faster RCNN, DetectorRS, CenterNet, CascadeRCNN,
FoveaBox, Yolo v4, CenterNet2, and Deformable Detr were
also given some fine-tuning. Using the localization dataset
on citrus disease, the authors were able to get the greatest
mAP possible with CenterNet2, which was 0.914 [68]. There
are not many papers in the field of disease localization since
there are not many DSS, and the expense of labeling large
datasets is costly. Therefore, it is necessary to make more
improvements in the field of plant disease localization studies
to accomplish accurate disease detection at the field level.
3) RQ3: WHAT INFLUENCE DO COGNITIVE CNN MODELS
HAVE ON THE DETECTION OF PLANT DISEASES WHEN
THEY ARE APPLIED?
CNNs with attention mechanisms are becoming increasingly
used in identifying plant diseases. CNNs activate and store
beneficial features via the attention mechanism. This speeds
up CNN convergence and disease detection. Residual CNN
equipped with an attention mechanism for the diagnosis of
tomato crop diseases constructed. After being augmented,
95,999 images of tomato foliage were used and attained an
accuracy of 98% over ten different classes in the PlantVil-
lage dataset [50]. Moreover, the authors proposed a shallow
CNN model that utilized the ShuffleNet model’s SE block to
implement a CBAM method. The model obtained an accuracy
of 99.14% using a dataset of grape leaf images consisting
of 4,062 individual images [87]. In [66], image features
were extracted using a feature segmentation subnetwork, and
feature maps were generated using a spot-aware categoriza-
tion subnetwork. Both subnetworks were utilized in PDI and
PDD. The network was instructed to utilize 404 photos from
the Apple dataset, and the scientists found that it had an
accuracy of 89.4%. Using a straightforward CNN model, self-
attention is represented by residual blocks. Their simulation
achieved an accuracy of 98% on the MK-D2 dataset and
an accuracy of 95.33% on the AES-CD9214 dataset [113].
Convolution, that is, depth-wise separable and spatial CBAM
is available in DenseNet. While the accuracy of the maize in
the PlantVillage DSS was stated to be 98.50%, it was proved
to be 95.86% in some other maize datasets [56]. Denoising
was accomplished by using the binary wavelet transform
with Retinex. In addition, the work of background removal
was carried out using an implementation of KSW that had
been improved using ABCK. RAN was trained to recognize
tomato illness by utilizing preprocessed images as training
data. Their model scored an 89% accuracy rating in the
tests [114]. GAN is equipped with an attention mechanism to
produce images and a classification system for plant diseases.
An accuracy of 97.9% was achieved during the experiment,
which was conducted on a cucumber disease dataset [90].
MobileNet’s CBAM and spatial features were incorpo-
rated, and the program recorded an accuracy of 99.67% on
the PVD. However, on the rice dataset, MobileNet showed
just 98.48% accuracy [69]. In [79], it was suggested to use
an attention approach with advanced learning for PDD in
environments with complex backgrounds. They reached an
accuracy of 98.26% using a combination of the PVD and their
original in-field data. The custom and plant village datasets
yielded 99.78% and 99.33% accuracy, respectively [115].
A modified version of the Mobilenet v2 that incorpo-
rates depth-wise separable convolutions. Within the model,
they incorporated both channel and spatial attention and
attained an accuracy of 99.71% using PVD and 99.13%
on a bespoke (for example, paddy, maize, and cucumber)
dataset. Similarly, [108] utilized a MobileNet multi-head self-
attention rice disease classification model. Bayesian opti-
mization tuned hyperparameters. The model had 94.65%,
92.6%, 89.6%, and 87.4%on accuracy, precision, F1-score,
and recall on 2,370 nutritious rice photos in four classifica-
tions. Inception and residual blocks [116] suggested a CNN
network. The network used a modified CBAM. Their maize,
tomato, and potato dataset models were 99.5% accurate.
4) RQ4: WHAT ARE THE MOST IMPORTANT CRITERIA TO
CONSIDER WHEN EVALUATING THE EFFECTIVENESS OF
DISEASE DETECTION SYSTEMS FOR PLANTS?
Various performance evaluation criteria have been devised to
evaluate the effectiveness and efficiency of these reviewed
methods and classifications in general. For example, we can
often use precision, accuracy, F1-scores, mAP, and recall as
evaluation metrics for classification approaches. The engi-
neering and medical research communities widely use these
statistical parameters in disease localization techniques. The
assessment measures are defined by the terms false negatives
(FN), true negatives (TN), true positives (TP), and false pos-
itives (FP), as defined and expressed below.
a: ACCURACY
The term ‘‘accuracy’’ is the proportion of correct predic-
tions made compared to the total number of data points
59182 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
collected (T). In scientific literature, it is referred to as recog-
nition, correctness, or success rate.
Accuracy =(TN +TP)/T (1)
b: PRECISION
Precision is described as the proportion of actual positive
samples found to the total samples anticipated to be positive.
Precision =TP/(TP +FP) (2)
c: RECALL AND SENSITIVITY
The term ‘‘sensitivity’’ or ‘‘recall’’ refers to the proportion of
correctly anticipated positives to the total number of actual
positive results.
Recall =TP/(FN +TP) (3)
d: F1-SCORE
The F1-score is a definition that refers to the harmonic mean
of both precision and sensitivity (recall).
F1 =2∗(Recall ∗Precision)/(Recall +Precision)(4)
e: SPECIFICITY (SPE)
This is the ratio of precisely anticipated negatives relative to
the total number of observed negatives.
Specificity =TN/(FP +TN)(5)
f: mAP
The area beneath the precision and recall curve can be con-
sidered the standard for precision (AP). A batch of n samples
is used to calculate the mAP, which is the same thing as the
mean of the median precision values APi, i =1, n samples.
Table 5presents the attention mechanism PlantVillage dataset
studies.
1/nX_(i=1)∧nAPi (6)
5) RQ5: WHAT KIND OF IMPACT DO LIGHTWEIGHT CNNS
HAVE ON THE PROCESS OF IDENTIFYING PLANT DISEASES
OR DETECTING THEM?
In recent years, SCNN approaches have become increas-
ingly popular as practical classifications in PDD. There has
been a provision on performance with the most advanced
deep CNNs on several different datasets. A unique CNN
method consisting of pooling layers stochastically and five
convolutional layers exhibited an accuracy of 95.48% after
being applied to 500 photos of rice illness organized into ten
categories [92]. In [123] and [124], they built a CNN model
that consisted of only three convolution layers and claimed
an accuracy of 93.4% on the cucumber dataset. A tiny CNN
model with six layers described a classification precision of
96.46% on 39 different classes after applying various data
augmentation techniques [125]. Image backdrop by putting
them into a ten-layer CNN for soybean crop disease catego-
rization. The accuracy of their model was determined to be
98.14% [126]. Applying a fundamental eight-layered CNN
model, the authors [128] demonstrated a diagnostic accuracy
of 98.4% for tomato diseases using the PVD.
Image feature extraction was performed using a shallow
five-layer CNN model by the authors [90], who then moved
on to a 10-layer CNN model. Fold-based cross-validation
support for SVM classification Moreover, in [68], the tomato
varieties were used while removing the background. Their
model has an accuracy of 98.6% across ten different classes.
In [39], enhanced the images using a piecewise log transform,
an improved Retinex method, and wavelet basis function
image enhancement before feeding the images to a CNN
model. Moreover, it employed wavelet-based image enhance-
ment. A PRelu activation function was utilized in their
modified version of the AlexNet model, which had dilated
and multiscale convolutions. The proposed model achieved
98.62% accuracy, and [128] further used the SE inception
module, batch normalization layer, and multiscale feature
extraction method in CNN (FLOPs). In [75], researchers
extracted features from the first two VGG16 blocks to min-
imize the model size. SVM and random forest methods are
classified. Their approach was tested on maize, apple, and
grape plant images from the PlantVillage collection, and a lit-
tle model showed 93% accuracy. A classification model that
was a modified version of DenseNet. The trained corn model
of 12,332 images divided into four groups and achieving an
accuracy of 98.06% by utilizing the Adam optimizer was
presented [62]. A variance of plant symptoms and diseases as
a time function developed a CNN model that could capture
variations. The process of developing their categorization
model consisted of two parts. In the process’s initial step,
a rice dataset was obtained and then partitioned into several
classes because of the intraclass variance. During the second
step, the study investigated both fine-tuning and transfer
learning (TL) as potential solutions, and their model attained
an accuracy of 93.3% [95].
With a total of 841 grapevine infield images across three
different categories, the highest accuracy reported was 94%
after HSV and improved CNN [70] on three distinct CNN
models for the diagnosis of the severity of the Vigna mungo
virus for VirLeafNets. They reported an accuracy of 91.23%,
96.42%, and 97.40% [102] using photos gathered from
unmanned aerial vehicles. Wheat diseases were identified
using a dual wheat disease identification method based on
DTs and CNN models, and it achieved an accurate diag-
nosis rate of 97.3% [42]. Also, the [32] model attained an
accuracy of between 90.6% and 97.9% for estimating the
severity of the condition and classifying it [130], respectively.
A CNN model with seven layers was created by [36] for
classifying and segmenting diseases. On grape photos taken
from the PVD, the model obtained an accuracy of 91.62%;
on 500 leaf images, it produced an accuracy of 93.75%.
Researchers in [89] proposed a basic two-layer CNN for
visual feature extraction using feature selection and a fish
swarm optimizer. The classification was done using ANN,
SVM, and LSTM techniques. On a total of 4,800 photos
VOLUME 11, 2023 59183
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 5. Summary of the studies using attention on CNN for plant disease identification.
of rice leaves, 97.5% was attained. It has been discovered
that CNNs are used with highly restricted datasets or for
only one species at a time. Table 6provides a synopsis of
several lightweight CNN models that were investigated in this
work.
6) RQ6: HOW EFFECTIVE HAVE THE TECHNIQUES BASED
ON TRANSFER LEARNING BEEN IN THE IDENTIFICATION
OF DISEASE?
Initially, the author [15] conducted a groundbreaking study
using GoogleNet and AlexNet to construct a PDD model
59184 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 6. The lightweight CNN overview of crop disease detection.
using their plant disease dataset, which included 54,305
photos divided into 38 categories. In addition to using
TL, a remarkable accuracy of 99.35% was attained by the
GoogleNet model using the TF technique. The number of
photos in the PVD increased to 87,848 over 58 categories.
Therefore, various studies have used this improved dataset
while developing the plant disease classification model. The
study [131] evaluated different neural networks with the
highest accuracy on the enlarged plant village dataset [132].
The quantity of the dataset as well as its variety influenced
the identification of plant diseases. They used TL on the
GoogleNet method using photos that had both the labora-
tory and the backdrop removed [133]. They removed the
background from 1383 photographs in 56 different classes
and reached the greatest accuracy possible of 87%. Authors
in [135] used the VGG16 model to detect diseases in millet
crops and reported a 95%, 90.5%, 94.5%, and 91.75% on
accuracy, precision, recall, and F1-score, respectively, using
the VGG16 model. Fine-tuning was done on Inception v4,
ResNet, and VGG16, among others, using the PVD, which
VOLUME 11, 2023 59185
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 7. Study overview on plant disease detection using transfer learning.
59186 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 7. (Continued.) Study overview on plant disease detection using transfer learning.
has 54,305 images. According to their research findings,
the DenseNet121 model had the best accuracy, coming in
at 99.75% [135]. In a separate piece of research referred to
as [12], the authors employed DenseNet models on multilabel
classification and regression with focal loss to identify plant
diseases and achieved an accuracy of 93.51%, 93.31%, and
93.71%, respectively. These results were based on the clas-
sification of the images using the DenseNet121 model [41].
Another study focused on collecting photographs of cucum-
ber leaves from a greenhouse for disease diagnosis [12].
They analyzed the results obtained from utilizing several
EfficientNet models and compared their effectiveness. Over
38,072 images, the EfficientNet-B4 technique attained 97.5%
accuracy. For disease identification, we gathered greenhouse
cucumber leaf images [87]. EfficientNet scored 97.5% on
38,072 images after augmentation. In another study [136],
GoogleNet scored 44.54% using the IPM dataset, 28.13%
using the Bing dataset, and F1 scores of 99.35% using the
PVD. Also, utilization of the GrabCut technique for image
segmentation was proposed and trained on the 38,072 images
from PVD, and their in-field dataset achieved an accuracy
of more than 80% [137]. With varied preprocessing pro-
cesses in each study, distinct models outperform one another.
TL models are presented in Table 7. It is observed that in most
instances, the VGG16 model proved to be the most effective.
ResNet is the highest-performing model for plant dis-
ease identification out of VGG19, ResNets, LeNet, VGG16,
and AlexNet. It achieved 91.2% accuracy when tested on
VOLUME 11, 2023 59187
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
5941 images of leaf samples from 24 separate classes taken
in the field [138] and employed DCGAN to produce syn-
thetic images to enhance the size of the dataset. They eval-
uated the Inception v3 method’s performance after adding
more data on citrus trees from the PVD and comparing the
results. The model’s accuracy was 92.60 percent, which was
20 percent higher than it would have been without such data
augmentation [117]. The effectiveness of MobileNet V2 and
V3, Xception, and NasNet Mobile on PVD entailing 109,290
tomato leaf pictures for real-time disease diagnosis [64].
Assessment of the accuracy of five diverse neural networks
methods like EfficientNet, Inception v3, AlexNet, ResNet50,
and VGG16 on 61,486 photos from 39 different classes in
the PVD after the dataset was augmented reported a 99.97%
accuracy [139]. The effectiveness of several current models
found that the MobileNet prototype surpassed the competi-
tion with an accuracy of 82.1% for categorising the coconut
crop disease syndrome [140]. Improved accuracy of spotted
wilt virus detection using the Inception v3 standard in peanuts
on the validation dataset; they achieved an AUC of 0.97, with
a sensitivity of 77% and a specificity of 98% [141]. Similarly,
on the same data, the VGG16 approach accomplished the
maximum level of accuracy, which was 97.89% [46]. Utilized
CGAN for data enhancement on the DenseNet 121 model and
a transfer learning technique for classifying plant diseases.
The classification accuracy of their model for tomato diseases
into five categories, seven categories, and ten categories was
99.51 %, 98.65 %, and 97.11 % [63], respectively. Also,
a method known as ‘‘DoubleGAN’’ was devised to produce
high-resolution images using two distinct GAN models. The
first one was WGAN, which was for maintaining a healthy
class image and balance. The next one was a super-resolution
GAN to produce images with high resolution. After using the
DoubleGAN model, they examined the results of VGG16,
ResNet50, and DenseNet121 [142]. Using a subset of the
PVD, they found that DenseNet121 had the highest accuracy
at 99.53%. However, researchers combined DeepLabV3+
and ResNet50 for plant disease segmentation [71]. They use
fuzzy rules to assess disease severity. Their grape leaf model
was 97.5% accurate. According to the investigation, various
studies have been completed utilizing the fine-tuning and TL
of cutting-edge models. They have been applied in the bulk
of works utilizing the PVD, VGG, DenseNet, and ResNet
Inception prototypes.
7) RQ7: IN WHAT WAYS DO CONVENTIONAL IMAGE
PROCESSING AND MACHINE LEARNING APPROACHES
ADD TO THE COMPLEXITY OF DETECTING PLANT DISEASES?
Researchers have analyzed the most essential machine learn-
ing techniques to enhance disease classification and detection
systems. The previous section’s criteria were used to choose
twenty-seven research publications. CNNs are an integral part
of the process by which conventional IP research is advanced.
CNNs have been instrumental in the development of a vari-
ety of approaches to the diagnosis of plant diseases. CNNs
have a wide application range that can be accomplished
through either unsupervised or supervised learning. For pic-
ture clustering and other applications, unsupervised learning
approaches utilize the capability of CNNs to learn feature
representations and extract features for further processing.
For problems involving classification, the CNN architecture
can be employed as a solution that works from beginning to
end. In addition to this, it has also been utilized as a feature
extractor for solving categorization issues. Other classifiers,
including DT, NB, RF, and SVM, are provided with the
characteristics that were extracted from the CNN as presented
in [42], [49], [99], [144], [145], and [146]. There has been
much success in applying the characteristics collected by
CNNs to problems involving the categorization of images,
the localization of plant diseases, and the segmentation of
images. VGG and AlexNet were utilized in addition to par-
allel feature fusion for feature extraction. A genetic algo-
rithm was used to select features, and then the results were
input into an SVM for classification [144]. With a total of
6309 apple leaf and banana leaf photos, the classification
accuracy was 98.60%.
Image feature extraction was accomplished by [49] using a
convolutional autoencoder. The SVM that was used to diag-
nose plant diseases received the retrieved features as input.
According to their model, training on the 6,004 maize and
potato leaf images taken from the PVD achieved 87.01%
accuracy. A convolutional neural network (CNN) incorpo-
rates multiscale features; after being augmented, their model
had a 94.65% success rate when applied to a dataset including
35,000 photos of cucumbers [87]. A method for identify-
ing plant diseases and determining their severity based on
the utilization of residual blocks and shuffle units achieved
98% accuracy, as demonstrated in [146]. The model attained
an accuracy of 91% when estimating the severity of the
condition. Reference [147] used a DCGAN to supplement
the data on a dataset, including information about tea leaf
diseases. They achieved an accuracy rate of 90% when
building the classification model with the help of VGG16
and the tea disease dataset. They have acquired five sepa-
rate in-field crop datasets with 121,955 images. The model
achieved an overall accuracy of 98% when applied to all
five crop datasets, as detailed in [17]. With the model-based
simulator, they got a score of 64.3% in terms of segmen-
tation using the SegNet-based approach [50]. Probabilistic
programming with Bayesian deep learning and fine-tuned
VGG16 model achieved an accuracy of 96% using PVD
54,306 images [60]. On the Kaggle dataset of 15,408 maize
species pictures, the model had 98.2% accuracy [58]. Authors
in [36] merged the VGG-16 pretrained Inception v3 and Ima-
geNet weights with a random weight initialization into a sin-
gle algorithm. Their models were 84.25%, 92%, and 80.38%
accurate on the maize PVD, rice dataset, and bespoke maize
dataset. Authors [98] gathered 5,932 in-field rice images
divided into four categories and evaluated the effectiveness
of eleven CNNs based on transfer learning in addition to
an SVM classifier using the data, achieving an accuracy
of 98.38%.
59188 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
In [94], the AlexNet model extracted features of rice dis-
eased; the model was trained on 60,000 rice leaf photographs
and had an accuracy of 96.8%. Self-supervised tomato plant
disease detection in a location network, including feed-
back, region recommendation, classification, and networks.
Localization networks employed user-defined loss functions.
They trained the model with 16,470 PlantVillage images,
added data, and balanced classes. The dataset was 99.7%
accurate [38]. Asymptotic non-local denoising was followed
by parallel convolution with filters of various sizes. The
CNN model replaced the Softmax layer with an upgraded
PSO extreme learning machine. After augmentation, 102,052
peach images had an accuracy percentage of 88.13% [18].
They have created a hybrid model that combines the ReseNet
50 and Inception v3 datasets. The accuracy of their model was
the greatest possible, coming in at 98.57% for grape disease
categorization. Employing a UAV device, we were able to
acquire a dataset for the categorization of Pinus tree dis-
ease [72]. The AdaBoost classifier was then fed the extracted
features to make its determination. Their model performed
exceptionally well on the dataset, with 78.6%, 95.7%, and
86.3% on precision, recall, and F1-score, accordingly [147].
Studies like [149] focused on developing a method called
FSL to diagnose plant diseases. On the PVD, they utilized
the Inception v3 method on the Siamese network splitting
the dataset into halves, with each section containing either
32 or 6 classes. They were corrected 91.4% of the time out
of 32 classes and 94% of the time in six of the classes.
Citrus disease diagnosis was accomplished by [41] using
FSL in conjunction with zero-shot learning and a conditional
variational autoencoder. It was demonstrated that their model
could achieve 53.4% harmonic mean accuracy with zero-
shot learning. The green channel of visible-range and infrared
pictures has been used by [72] to detect diseases that damage
grapevines with an accuracy of 95.02%.
Furthermore, [66] and [67] implement a degressive tech-
nique for the generator component and dense connectiv-
ity of instances. Grape leaves, totaling 4,062, were used
throughout the training process of the network, which also
resulted in the generation of synthetic data. The highest
possible score that their model could attain was 98.7%. For
example, [149] built a CNN model that diagnosed peach
disease with 98.75% accuracy using the PVD. Reference [69]
introduced DenseNet and reconstructed the residual network.
This approach used leaky ReLU-activated tensor-based con-
volutional layers. The 2018 AI Challenger dataset showed
95% model accuracy. Reference [99] suggested diagnosing
plant diseases with a thick block-based CNN model. They
merged PlantVillage and the iBean Leaf Image Collection,
demonstrating 99.19% accuracy.
Cross-iterative k-means clustering resulted in the develop-
ing a clustering model [150]. Image similarity was assessed
using siamese networks. Three sets of data were used to
evaluate the process. The accuracy of their model varied
between 89.5%, 84.9%, and 52.1% among three distinct
PlantVillage and Citrus databases. CNNs that use VGG16,
Inception v3, and ResNet101 are 5-time trilinear and bilin-
ear. Their model predicted PlantVillage at 99.7% and Plant-
Doc at 75.58%. Reference [83] tested SqueezeNet-MOD2,
AlexNet, GoogLeNet, SqueezeNet-MOD1, SqueezeNet and
ResNet50 on the strawberry dataset. ResNet50 had the high-
est accuracy, at 98.11%. In [145], a comparison was made
between machine and deep learning techniques for citrus
disease detection into five different classes, as early work
showed in [83], [145], [151], and [154]. These models were:
ResNet18, ResNet50, ResNet101, and DenseNet201. On the
Turkey-PlantDataset, which consists of 4447 images divided
into 15 categories, their model achieved an accuracy of
97.56% and 96.83% when employing the ensemble and while
utilizing early fusion.
For wheat disease identification, [152] developed a 24-
layer CNN model. The accuracy of their model was mea-
sured at 97.88% when it was applied to the LWDCD2020
dataset. [153] With the increased amount of in-field back-
drop, the CNN model trained with ResNet50 achieved an
accuracy of 98.50% on the PVD and 72.03% on the collected
field dataset, respectively. VGG16 was used for the multiclass
SVM and extraction of image features [154]. The accuracy
of their model was measured at 91.3% across the eggplant
dataset for all five classes [155]. The model attained an
accuracy of 91.19% when applied to the 87,867 photos that
were included in the PVD.
Cucumber, Corn, and others developed a spatially
pyramid-oriented CNN method for disease diagnosis [156].
Their model had a higher than 90% accuracy rate. Com-
bining DeepLabV3+and U-Net, researchers [82] developed
an innovative CNN model for segmenting plant disease.
The study further used DeepLabV3+to generate segmented
images, while U-Net was used to estimate the disease patch
coverage. Their model has an accuracy of 93.27% for leaf
segmentation, 92.85% for disease severity classification, and
a dice coefficient of 0.6914 for detecting and categorizing
cucumber leaf diseases. Reference [79] have used GAN
for something other than its intended purpose. Using a
GAN developed for image super-resolution brings accuracy
to 92.16 %. CNN-based PDD models take their influence
mostly from industry-standard neural network designs, for
example, VGG, Inception, AlexNet, ResNet, and MobileNet,
according to an assessment of published research. After
undertaking a review of the relevant published research, this
was discovered. In addition, many models have integrated
CNNs with IP-based image preprocessing techniques [159].
While creating their CNN-based models, several researchers
contributed to developing new datasets for the detection of
plant diseases. In the study [78], the use of leaf images as a
diagnostic tool for disease in grapefruit was suggested. Using
the hyperspectral imagery (HSI) color model and an astute
edge detector improved the image’s textural characteristics.
HSI is an abbreviation for the HSI color model. In addition,
the color co-occurrence method was employed in conjunc-
tion with SGDMs to determine the texture characteristics.
They photographed grapefruit leaves in the laboratory and
VOLUME 11, 2023 59189
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
classified them into oily patches, scabs, and regular citrus
leaves. Each category comprised forty photographs. Accord-
ing to the presented findings, the accuracy of their models
ranged between 95.8% and 100%. Fig. 4depicts a block
diagram of the various processes that are performed by a
typical machine learning-based disease detection system,
including data acquisition, annotation, processing, feature
extraction, and then classification.
Authors in [160] used 117 cotton crop images grown
in laboratory settings; the box-counting approach of fractal
dimension was used for feature assessment, and SVM was
utilized as a classifier. It was demonstrated that a maximum of
94.21% accuracy could be achieved in classification. Using
Fermi energy-based region extraction, [94] identified a rice
pathogen. A rule-based classifier recognized characteristics
chosen using rough set theory (RST). After analyzing 500 rice
pictures in four classes from the lab and the field, the accuracy
of their model was 94.21%. Regional image descriptors
enabled researchers to diagnose a soybean disease with a
CCR of 99.83% [163]. A color histogram and a pairwise
classification technique were employed for illness diagno-
sis [32]. During the segmentation of images, they employed
a GAC technique, using 1335 images shot in the field; their
model attained an accuracy of 58% and a recognition rate
of 91.63 %when it came to identifying cucumber disease.
In [37], we employed the full color feature on 93 images
of cucumber downy mildew captured in the field, and their
approach of image segmentation utilizing morphological
processes obtained 97.29 percent accuracy. Using the color
constancy method and SLIC, [44] aided with the wheat
disease categorization research by finding hotspots on the
plant’s leaves. On a total of 3,637 field images, they achieved
70–85% accuracy. IP stages that have been smoothed,
enhanced, denoised, and aligned. K-means clustering of
L∗A∗b color space images with segmentation FFT and
log-frequency histogram features is combined. PCA and the
sparse representation decision rule were used for feature
reduction and classification [84]. They identified 85.70%
of 420 images of cucumber leaf disease. Using PSO and
association rule mining, [97] developed rule-based
incremental classifications. Decision tree (DT) classifiers,
Bagged Tree, KNN, Boosted Tree, and Cubic SVM classified
gathered features. Their bagged tree-based ensemble model
had 99.9% accuracy on 199 controlled citrus leaf pictures
split into four groups.
PCA reduced features, and SVM classified cucumber dis-
ease using SLIC, EM algorithm, and PHOG. Their model
used in-field photographs to have the highest cucumber
disease identification rate of 65.41% [87]. The image back-
ground was removed from the maximum correlation coeffi-
cient and global thresholding based on OTSU [162]. Their
model predicts 95% okra YVMV disease and 82.67% bitter
gourd disease. Stretching contrast, a top-hat filter, besides the
Gaussian function, enhances images. The high-dimensional
color transforms saliency strategy includes color, texture, and
geometric features. On the 100 features, principal component
analysis and support vector machine classification were
then performed. Their model has a 95.8% accuracy and
0.98 AUC on 580 field and lab images of citrus [163]. The
method demonstrated an accuracy of 89% when tested on
5632 images in four categories from the gathered and PVD.
A region-expanding function was employed to extract image
data, and bacterial foraging was optimized for the training
of CNN. In the hidden layer of the ANN, a radial basis
activation function was utilized [164]. Overall, 270 images
in the PVD achieved a specificity of 85.58%. Retrieved
image features using SPAM and subsequently selected those
features using the exponential SMO. In addition to [165]
and [166], SVM was employed during the classification pro-
cedure. With a total of 1308 photographs obtained in the
field from five distinct species of tea plants, the accuracy of
their model was 98.5%. Pixels are correlated with EM seg-
mentation [52], optimizers of competitive swarms’ selected
characteristics. After tenfold cross-validation on PVD 3,171
apple leaf pictures, their model’s accuracy was 97.20%. Uti-
lizing Pearson’s rank correlations, 15 color characteristics
and color spaces were initiated. Photos were identified pixel-
by-pixel using classification and regression trees (CART).
The proposed detection approach for cucumber disease was
98.7% accurate. GrabCut separates the image foreground
from the background using the Gaussian mixture model and
Orchard–Bouman clustering. Later, LBP and SVM identified
and extracted features 95% of field-grown grapes [123].
Additionally, [168] segmented soybean leaves using
k-means clustering to identify leaf diseases. LBP, SURF,
and SVM were employed to choose the classification pro-
cess’s features. They have supplied evidence indicating that
their algorithm has a 75.8% success rate across 358 images
in 8 distinct categories. Both local tri-directional patterns
(LTridP) and SVM were used to extract picture features [169].
On a total of 1,882 tomato leaf photos across five different
classes from the PVD, their model achieved a 95.7% accuracy
rate. Correspondingly, [74] researchers used fractional-order
Zernike moments and SVM classifier to identify leaf diseases
of grapes and achieved 97.34% accuracy as [79] detected
paddy leaf disease and attained 94.25% accuracy. K-means
clustering was used to partition the dataset [93]; with the
chlorotic and necrotic lesions, an accuracy of 99.67% was
obtained. The segmented image was then translated into the
domain of the wavelet transform. LBP was generated and
then utilized for classification using an ANN classifier on
the segmented image. Their strategy achieved a classifica-
tion accuracy of 95.4% while assessing banana diseases [45]
and [169]. Moreover, in [48], authors developed a model for
identifying groundnut infections. A Harris corner detector,
HOG, and KNN classifier yielded the desired results by
correctly detecting infected groundnut with an accuracy of
97.67%. The photographs were modified to use the L, a,
and b color spaces. LBP feature extraction was performed
on segmented images. ANN, RF, KNN, and SVM classifiers
were applied in the classification procedures. Using an SVM
classifier on a total of 2840 potato images separated into five
59190 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
FIGURE 4. A simplified conventional machine learning approach.
groups, they achieved a 97.4% accuracy rate, 99.7% on the
GrabCut algorithm, and 0.9 IoU on apple leaf images [170]
and [171], respectively, on 6980 images.
Most PDD techniques include color modification. Most
frequently, these approaches employ LBP and histogram fea-
ture extraction algorithms. In extant works, SMO, BoVW,
PSO, and PCA are the prevalent feature selection strategies.
Some academicians use K-means clustering and SLIC to
segment photos pixel-by-pixel. ANN, NB, SVM, and KNN
are prominent algorithms for PDI and PDD. Due to PlantVil-
lage’s extensive range of plants, most authors have created
models utilizing the entire PVD or certain plant species.
8) RQ8: IS THERE AN ADEQUATE PUBLIC DATASET FOR
MACHINE AND DEEP LEARNING IN THE PDD AND
IDENTIFICATION?
Several scientists have contributed by availing themselves
of datasets for a variety of crops. As a result, the public
can now access numerous datasets containing information
on a broader range of plant diseases and pests (detailed in
Fig. 5). Some datasets are generated by capturing images
from agricultural farms, while others are compiled in a lab-
oratory with a fixed or eliminated background. In addition,
specialized databases are compiled for crop-specific research.
These files contain images of a single crop classified into
multiple disease groups. These are referred to as single-crop
datasets. Other types of databases have an extensive range
of crops, and each of these crops has photographs of plants
afflicted by a different set of plant or crop diseases. These
datasets are referred to as ‘‘multi-crop datasets.’’ Finally, this
section contains publicly accessible datasets that researchers
often utilize. These databases are exploited by researchers in
the development of disease detection strategies.
1) PlantVillage: Penn State University introduced this
multi-crop dataset widely used for disease categoriza-
tion. The collection includes 54,305 images of four-
teen distinct plant species. These photographs of plant
leaves were captured in a controlled setting with a
uniform background. The compilation includes data on
38 distinct forms of plant diseases [16].
2) Turkey-PlantDataset: Inonu Universities in Turkey and
the Agricultural Faculty of Bingol used a Nikon
7200d camera, the acquire the data for the dataset,
which consists of 4447 photos organized into fifteen
categories [153].
3) PlantDoc: The collection is compiled from data col-
lected in the field, and it contains 2598 photos orga-
nized into seventeen categories [82].
4) PDD271: A massive dataset consisting of 220,592 pho-
tos divided into 271 categories. The dataset includes
photographs taken in the fields of vegetable, grain,
fruit, and tree plants. A Beijing company called Beijing
PuhuiSannong Technology Company Limited owns the
dataset and has access to a relatively insignificant por-
tion of it [21].
5) LWDCD2020: This dataset has 12,160 images of
wheat diseases organized into nine disease classes and
one healthy class, all taken under actual field condi-
tions [152].
6) AES-CD9214: The data for this collection was gath-
ered in the field under a variety of conditions, including
VOLUME 11, 2023 59191
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
FIGURE 5. Dataset types.
FIGURE 6. Sampled example disease images taken each the datasets.
those about angle, resolutions, and illumination. It has
9,214 images, containing 44 different kinds of plant
leaf images organized into six categories [117].
7) Rice Dataset: China’s Fujian Institute of Subtropical
Botany has also made this dataset available to the
research community. The collection contains a total of
560 different photos of rice diseases organized into five
distinct clusters [119]. It includes photos taken in the
laboratory as well as in the field.
8) Maize Dataset: This modest dataset has been made
available to the public and was compiled by the institute
mentioned above from Xiamen [119]. 481 maize leaf
image samples were taken in the field and categorized
according to one of four disease classifications.
9) Apple Dataset: A disease identification competi-
tion was held on the Kaggle platform before the
CVPR 2020 conference. The competition consisted of
3651 photos of apple leaves infected with the disease
and fell into one of four categories: healthy leaves
despite cedar apple rust, apple scab, and several other
disorders [172].
10) NLB dataset: The dataset includes 18,222 photographs
of maize taken in the field by UAV-mounted cameras,
smartphones, and UAV sensors. There are 105,735
annotations for bounding boxes related to northern leaf
blight disease [61].
11) Rice dataset: This is one of the new rice datasets,
which contains 1426 photos of different rice classes
and images taken in natural field circumstances
and provided by the Bangladesh Rice Research
Institute [96].
12) Strawberry Dataset: A dataset for the localization of
strawberry disease is being compiled, and it has 3,531
photos divided into four types of bounding boxes [114].
13) Embrapa Dataset: The Plant Disease Detection
Database (PDDB) was created by the Embrapa
Agriculture Institute and comprised 2,326 images of
different plant parts that diseases have infected. A sub-
sequent extension of the dataset led to its renaming
as the XDB dataset [133]. The XDB dataset contains
46,513 maize leaf pictures, representing 18 species and
93 disease categories.
14) PlantVillage (extended): The initial dataset from
PlantVillage was expanded to involve 87,848 photos of
25 plant species, each of which was categorized into
58 disease subtypes [134]. Both in-field and laboratory
59192 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 8. Summary of the datasets available (Ref; Reference).
conditions were used to capture the images. However,
these are not available to the public any longer.
15) Bing: To evaluate the efficiency of the approaches, [15]
downloaded 121 photographs from the Bing search
engine. These images had a predetermined background
and in-field area.
16) IPM: By compiling this dataset from various Inter-
net sources, [15] has demonstrated the effectiveness
of their efforts. The dataset consists of 119 test
images, equally distributed between fixed and back-
ground conditions. The datasets are listed in Table 8.
Fig. 6displays some example images taken from the
database.
9) RQ9: ARE RESEARCHERS MODELING PLANT DISEASE
DETECTION TECHNIQUES THAT EMPLOY DATA COLLECTION
AND PREPROCESSING APPROACHES?
Studies have used digital color cameras to determine the
edges of the photos by utilizing a Canny edge detector after
converting the images [79]. For data collection [162], we uti-
lized an HP Scanjet 1200 scanner. Reference [32] collected
information for the Brazilian Agricultural Research Corpo-
ration using mobile cellphones and digital color cameras
(Embrapa). Authors [37] captured the data with a Nikon
Coolpix S3100 camera while working at the Information
Institute of the Tianjin Academy of Agricultural Science.
Moreover, they began by converting the photos into HSV
color space, then moved on to LAB color space, where they
enhanced the colors. To collect what data there was, [44]
utilized an iPad, iPhone 5, iPhone 4, and Dell tablet, among
others.
The Northwest China Agriculture and Forestry Univer-
sity [84] collected data on cucumbers using a smartphone,
a scanner, and a digital camera. For processing, images were
transformed into the L-a-b format [15]. Canon EOS 5D Mark
III (EF 24–70 mm F2.8L II USM) cameras captured images
at a resolution of 5760 by 3840 pixels. Images are then
normalized between 0 and 1 after being scaled to 512 by
512 pixels. Image feature extraction was continued with
additional applications of the PCA and whitening algo-
rithms [23]. They gathered images for their citrus dataset
in the Sargodha area of Pakistan. They used the top-hat
filter in conjunction with the Gaussian filter for preprocess-
ing. In another study conducted by [165], a digital single-
lens reflex camera was utilized to gather tea leaves. The
Anhui Academy of Agricultural Sciences’ Agricultural Eco-
nomics and Information Institute altered the photos so that
the color space was LAB for processing [165]. The Agricul-
tural Scientific Innovation Base of the Information Institute
of the Tianjin Academy of Agricultural Sciences captured
cucumber data using a Nikon Coolpix S3100 digital camera
with 2592 by 1944 pixels. Utilizing segmentation to discern
between the image’s foreground and background [37].
After cutting the images to 240 by 240 pixels, random
cropping, rotating, shifting, resizing, horizontal and vertical
flipping, and intensity transformation-based data augmen-
tation [147], tea leaves were collected with a DJI Phan-
tom 4 Pro UAV at10 meters using a Canon EOS 80D SLR
camera147], tea leaves were collected with a DJI Phantom
4 Pro UAV at 10 meters using a Canon EOS 80D SLR camera.
Smartphones were used to collect multi-crop data in Spain
and Germany [17]. To capture RGB photos, smartphones
and cameras with 1–24 megapixels were utilized. Embrapa,
the Brazilian agricultural research corporation, compiled the
data.
They inverted the vertical and horizontal orientations to
enhance image quality, and rotated, enhanced, and added
noise [133]. Table 9shows the equipment, zones, and pre-
processing techniques used for PDD. It was downloaded and
classified 124 images of mold diseases. They were improved
by zooming, rotating, and flipping [134]. The author in [65]
utilized data augmentation techniques such as center zoom,
random cropping, zooming, and contrast stretching based on
the four classes of tomato species found in the PVD. Only
one dataset exists that was obtained using drones; others
were obtained using smartphones, digital single-lens reflex
cameras, and scanners.
After cutting the images to 240 by 240 pixels, random
cropping, rotating, shifting, resizing, horizontal and vertical
flipping, and intensity transformation-based data augmenta-
tion [147], tea leaves were collected with a DJI Phantom 4 Pro
VOLUME 11, 2023 59193
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 9. Data sites, device(s), and preprocessing methods.
59194 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
TABLE 9. (Continued.) Data sites, device(s), and preprocessing methods.
UAV at 10 meters using a Canon EOS 80D SLR cam-
era147], tea leaves were collected with a DJI Phantom 4 Pro
UAV at 10 meters using a Canon EOS 80D SLR camera.
Smartphones were used to collect multi-crop data in Spain
and Germany [17]. To capture RGB photos, smartphones
and cameras with 1–24 megapixels were utilized. Embrapa,
the Brazilian agricultural research corporation, compiled the
data. They inverted the vertical and horizontal orientations
to enhance image quality, and rotated, enhanced, and added
noise [133]. Table 9shows the equipment, zones, and pre-
processing techniques used for PDD. It was downloaded and
classified 124 images of mold diseases. They were improved
by zooming, rotating, and flipping [134]. The author in [65]
utilized data augmentation techniques such as center zoom,
random cropping, zooming, and contrast stretching based on
the four classes of tomato species found in the PVD. Only
one dataset exists that was obtained using drones; others were
obtained using smartphones, digital single-lens reflex cam-
eras, and scanners. CMOS cameras, spectrometers, and lamps
collected hyperspectral imaging data. In IoT applications,
researchers use sensors to measure temperature, pressure,
and humidity. Digital cameras and other sensors may be
helpful in disease detection since several factors are involved.
Traditional IP and ML plant disease identification systems
preprocess data extensively. These methods use human and
automatic methods. Still, most DL systems scale, normalize
and augment images.
B. DISCUSSION
This discussion section subdivides the discussion into tech-
niques, data availability, and other general findings.
1) TECHNIQUES
According to this review, deep CNN models dominate PDD
compared to localization research. Shallow CNN methods
VOLUME 11, 2023 59195
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
for real-time field applications [173], [174], [175], [176].
Shallow CNN-based models in implementing more inno-
vative IoT-centered agriculture. The current hi-tech deep
CNN performs well on well-managed datasets considering
early approaches. Due to controlled data collection settings,
datasets like Embrapa and PlantVillage provide low-volatility
data used to build most models. However, field datasets have
not been used much or evaluated during the models’ real-
time performance. Deep CNN models have a more signifi-
cant memory requirement than shallow CNN models, which
could be a hurdle to their adoption in IoT applications. Nev-
ertheless, external networks are susceptible to minor data
limitations. In addition to the required amount of memory,
the number of FLOPs should be considered when deciding
whether a model is acceptable for IoT deployment. Also,
relatively few researchers have examined the models from
FLOP’s perspective. Sized symptom detection: early plant
pathogen is often mild, making identification complex. Thus,
early loss reduction requires technologies that can identify
even the most minor disease indicators. GAN-based image
super-resolution algorithms may potentially help recognize
which body sections are affected by the disease at an early
stage. Consequently, it is possible to improve the efficacy of
illness identification in small populations despite the corre-
sponding increase in the cost of data acquisition.
2) DATA AVAILABILITY
The lack of available in-field datasets is one of the most
significant obstacles to consider when developing PDD
models that can perform adequately in real-time. Scientific
researchers have recently made specific datasets available
to the public, although these databases only cover a select
few types of crops. In recent years, several researchers have
focused their attention, in addition to investigating CNNs’
disease detection capabilities, on the mechanisms that control
attention. The performance of these models was encouraging;
therefore, it is likely that more studies will be conducted
in this area. The following is a synopsis of the most sig-
nificant challenges and potential future areas for research.
Close-range data capture: Most publicly accessible datasets
were assembled using mobile cameras that captured close-
up images. Due to this, the disease-affected portions of the
area are typically the most obvious or cover a significant
portion of the image. Using these datasets to develop disease
detection models for mobile apps or apps with close-range
cameras will yield excellent results. However, these datasets
may not help locate diseases across a more comprehensive
farm or develop UAV-based applications. This is because the
disease may appear very small, and drone cameras may not
be able to get a clear image of their symptoms. Therefore,
not only to gather data from drones but also to make the data
public so that methodologies can be created based on these
datasets. Diverse and complex background: plant disease
detection may be difficult due to lighting, weather, and a
changing background in the field. Moving cameras, such as
those employed in drone-based monitoring, make it harder to
diagnose plants as compared to fixed cameras. This is because
the backdrop conditions are constantly present during the
training.
Lack of accessibility to standard datasets: It is evident
that the original datasets created in the lab were intended for
scientific purposes. In the years that followed, academia had
access to the same databases, including real-world scenarios.
In contrast, practically all infield datasets are created for a
single crop. Another issue with our evaluation was that most
machines and deep learning models created by academics
were based on proprietary datasets or a limited portion of pub-
licly accessible information. Due to this, there was no univer-
sally approved mechanism for comparing the effectiveness of
distinct approaches, even though they were well-designed for
the same plant species. This demonstrates how crucial it is
to have benchmark datasets containing real-world scenarios
that can be employed to construct models.
Deficiency of datasets for actual application development:
Most research uses public datasets, and most are laboratory-
generated. In contrast, works that utilize private datasets fea-
ture a variety of visual backgrounds. Therefore, it is essential
to utilize datasets with field image samples. Although some
enormous datasets are available to the public for developing
real-world applications, other large datasets are available to
the public. As a result, CNN models will benefit from being
trained on a greater variety of samples. Adversarial scenarios
would be helpful to. The deployment of very complex models
in intelligent agriculture solutions involves the fine-tuning of
models using field data. Even though highly potent mod-
els are readily available. As most in-field datasets are not
currently accessible, it is impossible to design reliable plant
pathogen and disease classification and detection methods for
real-time applications. The creation of increased labeled plant
information will improve the development of applications
that can capture a broader range of plant species in different
parts of the globe.
3) CHALLENGES AND FUTURE TRENDS
According to these reviewed studies, plant pests and
diseases have demonstrated huge social, post-harvest, and
economic losses in several countries’ global agricultural pro-
duction industries. This is especially true due to climate
change’s effects over the past several decades. Many prac-
tical approaches for detecting, monitoring, and evaluating
plant diseases have been continuously researched. These
approaches are employed in the fight against plant diseases.
In recent years, there has been a shift toward a more sig-
nificant emphasis on non-invasive technologies. Most imple-
mented models are simulated on small datasets with restricted
image backgrounds; therefore, it is vital to use raw images
considering the natural setup.
Most of these models use pre-trained CNNs, but their
ensemble can improve accuracy in identifying and classifying
different plant diseases. The SVM classifier is noted as being
among the machine learning classifiers that are mainly used;
therefore, testing the suggested models on other classifiers
59196 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
or combining their parameters is a potential trend. Using
traditional approaches has been demonstrated that is not
ineffective in plant pests and disease classification because
of several plant species and disease classification; therefore,
using transfer learning will improve model complexity and
performance and reduce computation resources. In recent
years, one of these technologies, hyperspectral technology,
has received considerable interest. Forthcoming systematic or
general reviews should scrutinize the works and considerable
limitations of innovative agricultural applications, especially
on plant pests and detection. Considering sustainable plant
pests and disease management incorporates plant disease
epidemic insight, healthy agroecosystems will improve and
boost plant health and crop quality produce with proper
natural resource conservation. Different plant diseases and
pests globally affect plant species depending on seasonal
and environmental factors making the centralized approach
impractical in some countries. Thus, an efficient treatment
may be delivered, and large-scale economic losses can be
prevented. It is necessary, for example, to establish how to
quickly compute the effective area of disease and analyze
the severity of infection and insect pests in a region. This is
necessary so that adequate treatment may be delivered. These
issues remain significant obstacles in plant pest control; other
challenges are listed below that must be tackled as soon as
practicable.
•Variations in lighting conditions and image quality
•Disease progression stages and inter-class similarity
•Limited availability of diverse and annotated datasets
•Computational challenges in processing large-scale
datasets
•Real-time disease detection in the field
•Impact of environmental factors on disease detection
accuracy
•Integration of multi-modal data sources
•Development of robust and interpretable models
addressing ethical and social considerations
•Climate change effects on plant diseases
•Integration of IoT devices, drones, and mobile
applications
•Advancements in data augmentation, transfer learning,
and domain adaptation
•Engaging citizen scientists and leveraging crowdsourc-
ing platforms
4) IMPLICATIONS OF THE STUDY
In this section, we briefly demonstrate the implication of plant
disease and pests in various ways, as discussed below:
a: IMPACT ON HUMAN
Plant pests and diseases inhibit water and nutrient absorption,
photosynthesis, fruit production, plant healthy growth, and
cell division. Depending on pathogen aggressivity, host resis-
tance, environment, duration of infection, and other variables,
these diseases may cause moderate symptoms or plant death.
Pathogens and contaminated portions can result in leaf spots,
fruit spots, root and fruit rots, leaf blights, wilting, and death.
For instance, the mild mottle virus interacts with human
immune systems and produces clinical signs. Several plant
diseases reduce food availability or pollute it with hazardous
compounds, causing harm to humans. In addition, humans
may be harmed by plant-disease-fighting microorganisms
applied to soils like herbicides and insecticides, among
others.
b: ENVIRONMENTAL PERSPECTIVES
Since different plants tend to have several pests and diseases,
the traditional farmer opts to apply insecticides and herbicides
that lead to land pollution in the long run. Even though most
plant pathogens cannot infect people, it is vital to avoid eating
moldy or rotten fruits and vegetables from farms where herbi-
cides are used and food tainted with fungi that produce toxins.
This is because both food types represent a significant health
risk. It is likely that by eliminating infected fruit sections,
the amount of pathogen inoculum and decaying fruit can
be reduced. However, this does not necessarily suggest that
all potential sources of contamination have been eliminated,
as certain fungi and toxins can multiply and migrate to areas
of natural plant fruit on the farm.
c: ACADEMIA
Without a doubt, various plant pests and diseases increase
different research bases that increase plant disease early
identification at the pixel level. This has been tested and
found to reduce the chances of using chemicals that raise
environmental (see subsection 6.2) and human concerns (see
subsection 6.1). As always, there is an urgent need for more
study on the direct consequences of plant infections and
diseases on humans; however, considerably large plant dis-
ease datasets are not available, making the current models
practically impossible on small agricultural smart gadgets.
Furthermore, mycotoxin-producing fungi and the presence of
such fungi in human-ingestible foods necessitate heightened
attention and caution. Prioritize food diversification and the
development of effective plant disease control systems to
avert epidemics of plant diseases like the late blight that
afflicted Irish potatoes. This plant disease can be transmitted
from one plant to another.
d: GLOBAL FOOD SECURITY
As explained earlier, plant pests and diseases affect plants
directly, but in the long run, they go beyond traditional
food shortages to a governmental level, thereby affecting the
global food supply caused by pathogens. This is because it
varies in genotype, space, and time, making plant admin-
istration difficult. Reducing pathogen inoculum, decreasing
pathogenicity, and improving crop genetic variety are all
methods for combating a disease that requires governments’
involvement in research to cater to plant pests and diseases
early enough. Plant pests and diseases compromise our food
supply; thus, legislators must support their management at
all costs.
VOLUME 11, 2023 59197
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
5) RESEARCH DIRECTIONS
Plant diseases pose significant threats to crop yields and food
security worldwide. Detecting and diagnosing plant diseases
promptly is crucial to mitigate their impact. In this section,
some identified future research directions are presented:
a: INTEGRATION OF MULTI-MODAL DATA SOURCES
Integrating multi-modal data sources is a promising direction
for future research in plant disease detection. This research
focuses on several attributes and objectives to effectively
leverage the combination of different data modalities. First,
diverse data sources like images, environmental data, and
genomic information could be identified and gathered to
represent plant diseases comprehensively. Robust fusion
techniques must be developed to effectively combine infor-
mation from multiple modalities, considering early, late, or
hybrid fusion approaches. Additionally, handling missing
or noisy data in multi-modal datasets is crucial for reli-
able fusion. Assessing the added value of integrating multi-
modal data sources compared to using individual modalities
is essential in evaluating performance improvements in accu-
racy, speed, and robustness. Finally, interpreting fused data
and understanding the relationships between modalities are
important for gaining insights into disease detection.
b: ROBUST AND INTERPRETABLE MODELS
Developing robust and interpretable models is a crucial direc-
tion for plant disease detection and classification. Robust
models should exhibit resilience to variations in data quality,
lighting conditions, and disease progression stages, while
interpretable models provide transparent explanations for
their predictions. Achieving robustness involves addressing
challenges such as inter-class similarity and variations in
environmental factors. Meanwhile, interpretability ensures
that farmers and stakeholders understand the reasoning
behind disease diagnoses. By focusing on developing robust
and interpretable models, researchers can enhance the trust,
adoption, and practicality of automated plant disease detec-
tion systems.
c: IMPACT OF CLIMATE CHANGE ON PLANT DISEASES
One aspect of climate change is the alteration of temperature
patterns. Rising temperatures can affect the distribution and
prevalence of plant pathogens, influencing their survival,
reproduction, and virulence. Additionally, changes in rainfall
patterns can impact disease spread, as some pathogens thrive
in moist conditions. Extreme weather events like droughts or
floods can further disrupt plant health and make crops more
vulnerable to diseases. Furthermore, shifts in climate can
affect the lifecycle of vectors, such as insects, that transmit
plant diseases. Climate change can alter the geographic range,
intensity, and timing of plant diseases, posing challenges for
agriculture and emphasizing the need for adaptive disease
management strategies. Understanding these impacts and
developing resilient agricultural practices are crucial for
mitigating the detrimental effects of climate change on plant
health.
d: CITIZEN SCIENCE AND CROWDSOURCING
Citizen science and crowdsourcing are valuable approaches
to plant disease detection. Citizen science involves public par-
ticipation in data collection while crowdsourcing leverages
collective intelligence. These methods offer benefits such as
large-scale data collection, collaboration between scientists
and communities, and empowering farmers. In addition, they
bridge knowledge gaps, provide resources, and contribute to
comprehensive disease databases. Citizen science and crowd-
sourcing revolutionize plant disease detection, monitoring,
and management by involving individuals and facilitating
timely interventions to safeguard agricultural productivity.
e: SOCIAL AND ETHICAL CONSIDERATIONS
Social and ethical considerations are vital in developing
and implementing plant disease detection technologies. Pri-
vacy protection, transparency, and informed consent are
essential for data management, and research related to its
impact is vital. Balancing intellectual property rights ensures
fair access and benefits. Equitable access to resources and
technologies should be prioritized for all farmers. Finally,
assessing the social and economic impact, particularly on
marginalized communities and local economies, is necessary.
By addressing these considerations, researchers and policy-
makers can ensure responsible, inclusive, and sustainable
plant disease detection technologies deployment.
VI. CONCLUSION
In this systematic review, we surveyed studies that present
plant disease and pest detection containing IP, ML, DL, and
others. It is clearly demonstrated that plant pests and dis-
eases harm the global agricultural industry. Despite the rapid
expansion of AI-based solutions, there are still many barriers
to overcome before high-performance real-time PDD solu-
tions can be produced, according to a comprehensive assess-
ment of PDD research employing imaging applications. This
system review presented a comprehensive overview of cur-
rent plant pest and disease detection studies. ML, IP, and
DL-based plant disease models and monitoring technolo-
gies have shown promising results. The study considered
176 articles published between 2012 and 2022. These studies
were selected after applying rigorous inclusion criteria from
five academic databases, including ACM, Springer, Scopus,
IEEE Xplorer, and Google Scholar. Our analysis presented
significant relevant findings from the considered studies pro-
viding adequate responses to the research review questions.
Most studies centered extensively around CNN-based disease
detection systems for numerous crops, notably citrus, have
been studied. Lightweight and TL algorithms, CNNs, GANs,
attention mechanisms, and autoencoders have been investi-
gated for high-functioning model construction, and a more
comprehensive range of modifications can still be done in this
paradigm to reduce the computation complexity. However,
the present training paradigm for DL models requires large
59198 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
data sets, making finding remedies for many plant diseases
difficult. There are a limited number of publicly accessible
datasets on this topic. In addition, the bulk of DL mod-
els is created using data collected under laboratory circum-
stances, which may hinder their performance in real-time
utilization. This study further resolves that industry and
academia have many computation complexities and an excel-
lent opportunity to avail models practically visible for real-
field implementation.
Finally, forthcoming reviews should scrutinize the works
and considerable limitations of innovative agricultural appli-
cations, especially on plant pests and detection, which is
another crucial research area. This survey covered a more
comprehensive range of plant pests and diseases methods,
challenges, and dataset checks and demonstrated that future
research sparks new ideas and the concepts of relevant theo-
ries, methods, and practices in industries and academia.
REFERENCES
[1] Q. Dongyu, T. Ghebreyesus, and R. Azevedo, ‘‘Mitigating impacts of
COVID-19 on food trade and markets,’’ in Food and Agriculture Organi-
zation of the United Nations. Italy: Press Release, Sep. 2020.
[2] The Impact of Disasters and Crises on Agriculture and Food Security,
FAO, Rome, Italy, 2018.
[3] New Standards to Curb the Global Spread of Plant Pests and Diseases,
Web Page of the Food and Agriculture Organization of the United
Nations, FAO, Rome, Italy, 2019.
[4] K.-H. Kim, E. Kabir, and S. A. Jahan, ‘‘Exposure to pesticides and
the associated human health effects,’’ Sci. Total Environ., vol. 575,
pp. 525–535, Jan. 2017.
[5] D. Laborde, W. Martin, J. Swinnen, and R. Vos, ‘‘COVID-19 risks to
global food security,’’ Science, vol. 369, no. 6503, pp. 500–502, Jul. 2020.
[6] L. Keeling, H. Tunón, G. O. Antillón, C. Berg, M. Jones, L. Stuardo,
J. Swanson, A. Wallenbeck, C. Winckler, and H. Blokhuis, ‘‘Animal
welfare and the United Nations sustainable development goals,’’ Frontiers
Veterinary Sci., vol. 6, p. 336, Jan. 2019.
[7] M. Khan, ‘‘Integration of selected novel pesticides with trichogramma
chilonis (hymenoptera: Trichogrammatidae) for management of pests in
cotton,’’ J. Agricult. Sci. Technol., vol. 21, no. 4, pp. 873–882, 2019.
[8] S. Savary, A. Ficke, J.-N. Aubertot, and C. Hollier, ‘‘Crop losses due to
diseases and their implications for global food production losses and food
security,’’ Food Secur., vol. 4, no. 4, pp. 519–537, Dec. 2012.
[9] F. Ren, W. Liu, and G. Wu, ‘‘Feature reuse residual networks for insect
pest recognition,’’ IEEE Access, vol. 7, pp. 122758–122768, 2019.
[10] E. A. Heinrichs and R. Muniappan, ‘‘Integrated pest management for
tropical crops: Soyabeans,’’ CABI Rev., vol. 2018, pp. 1–44, Jan. 2018.
[11] M. Margni, D. Rossier, P. Crettaz, and O. Jolliet, ‘‘Life cycle impact
assessment of pesticides on human health and ecosystems,’’ Agricult.,
Ecosyst. Environ., vol. 93, nos. 1–3, pp. 379–392, Dec. 2002.
[12] A. Sabarwal, K. Kumar, and R. P. Singh, ‘‘Hazardous effects of chemi-
cal pesticides on human health–cancer and other associated disorders,’’
Environ. Toxicol. Pharmacol., vol. 63, pp. 103–114, Oct. 2018.
[13] S. Savary, L. Willocquet, S. J. Pethybridge, P. Esker, N. McRoberts, and
A. Nelson, ‘‘The global burden of pathogens and pests on major food
crops,’’ Nature Ecol. Evol., vol. 3, no. 3, pp. 430–439, Feb. 2019.
[14] P. Udmale, I. Pal, S. Szabo, M. Pramanik, and A. Large, ‘‘Global food
security in the context of COVID-19: A scenario-based exploratory anal-
ysis,’’ Prog. Disaster Sci., vol. 7, Oct. 2020, Art. no. 100120.
[15] S. P. Mohanty, D. P. Hughes, and M. Salathé, ‘‘Using deep learning for
image-based plant disease detection,’’ Frontiers Plant Sci., vol. 7, p. 1419,
Sep. 2016.
[16] D. Shah, V. Trivedi, V. Sheth, A. Shah, and U. Chauhan, ‘‘ResTS:
Residual deep interpretable architecture for plant disease detection,’’ Inf.
Process. Agricult., vol. 9, no. 2, pp. 212–223, Jun. 2022.
[17] A. Picon, M. Seitz, A. Alvarez-Gila, P. Mohnke, A. Ortiz-Barredo,
and J. Echazarra, ‘‘Crop conditional convolutional neural networks for
massive multi-crop plant disease classification over cell phone acquired
images taken on real field conditions,’’ Comput. Electron. Agricult.,
vol. 167, Dec. 2019, Art. no. 105093.
[18] S. Huang, G. Zhou, M. He, A. Chen, W. Zhang, and Y. Hu, ‘‘Detection
of peach disease image based on asymptotic non-local means and PCNN-
IPELM,’’ IEEE Access, vol. 8, pp. 136421–136433, 2020.
[19] J. G. A. Barbedo, ‘‘Plant disease identification from individual lesions
and spots using deep learning,’’ Biosystems Eng., vol. 180, pp. 96–107,
Apr. 2019.
[20] J. Chen, J. Chen, D. Zhang, Y. Sun, and Y. A. Nanehkaran, ‘‘Using deep
transfer learning for image-based plant disease identification,’’ Comput.
Electron. Agricult., vol. 173, Jun. 2020, Art. no. 105393.
[21] X. Liu, W. Min, S. Mei, L. Wang, and S. Jiang, ‘‘Plantdisease recognition:
A large-scale benchmark dataset and a visual region and loss reweighting
approach,’’ IEEE Trans. Image Process., vol. 30, pp. 2003–2015, 2021.
[22] Z. Gao, L. R. Khot, R. A. Naidu, and Q. Zhang, ‘‘Early detection of
grapevine leafroll disease in a red-berried wine grape cultivar using
hyperspectral imaging,’’ Comput. Electron. Agricult., vol. 179, Dec. 2020,
Art. no. 105807.
[23] S. K. Noon, M. Amjad, M. A. Qureshi, and A. Mannan, ‘‘Use of deep
learning techniques for identification of plant leaf stresses: A review,’’
Sustain. Comput., Informat. Syst., vol. 28, Dec. 2020, Art. no. 100443.
[24] R. Manavalan, ‘‘Automatic identification of diseases in grains crops
through computational approaches: A review,’’ Comput. Electron.
Agricult., vol. 178, Nov. 2020, Art. no. 105802.
[25] Z. Iqbal, M. A. Khan, M. Sharif, J. H. Shah, M. H. ur Rehman, and
K. Javed, ‘‘An automated detection and classification of citrus plant dis-
eases using image processing techniques: A review,’’ Comput. Electron.
Agricult., vol. 153, pp. 12–32, Oct. 2018.
[26] J. G. A. Barbedo, ‘‘A review on the main challenges in automatic plant
disease identification based on visible range images,’’ Biosyst. Eng.,
vol. 144, pp. 52–60, Apr. 2016.
[27] Y. Tang, Y. Fu, Q. Guo, H. Huang, Z. Tan, and S. Luo, ‘‘Numerical sim-
ulation of the spatial and temporal distributions of the downwash airflow
and spray field of a co-axial eight-rotor plant protection UAV in hover,’’
Comput. Electron. Agricult., vol. 206, Mar. 2023, Art. no. 107634.
[28] A. Abade, P. A. Ferreira, and F. de Barros Vidal, ‘‘Plant diseases
recognition on images using convolutional neural networks: A sys-
tematic review,’’ Comput. Electron. Agricult., vol. 185, Jun. 2021,
Art. no. 106125.
[29] R. I. Hasan, S. M. Yusuf, and L. Alzubaidi, ‘‘Review of the state of the
art of deep learning for plant diseases: A broad analysis and discussion,’’
Plants, vol. 9, no. 10, p. 1302, Oct. 2020.
[30] M. Nagaraju and P. Chawla, ‘‘Systematic review of deep learning tech-
niques in plant disease detection,’’ Int. J. Syst. Assurance Eng. Manage.,
vol. 11, no. 3, pp. 547–560, Jun. 2020.
[31] V. Bischoff, K. Farias, J. P. Menzen, and G. Pessin, ‘‘Technological sup-
port for detection and prediction of plant diseases: A systematic mapping
study,’’ Comput. Electron. Agricult., vol. 181, Feb.2021, Art. no. 105922.
[32] J. Lu, L. Tan, and H. Jiang, ‘‘Review on convolutional neural network
(CNN) applied to plant leaf disease classification,’’ Agriculture, vol. 11,
no. 8, p. 707, Jul. 2021.
[33] M. M. Ali, N. A. Bachik, N. Muhadi, T. N. T.Yusof, and C. Gomes, ‘‘Non-
destructive techniques of detecting plant diseases: A review,’’ Physiolog.
Mol. Plant Pathol., vol. 108, Dec. 2019, Art. no. 101426.
[34] A. H. Alamoodi, B. B. Zaidan, A. A. Zaidan, S. M. Samuri, A. R. Ismail,
O. Zughoul, M. Faiez, G. A. Alshafeay, and M. A. Chyad, ‘‘A review of
data analysis for early-childhood period: Taxonomy, motivations, chal-
lenges, recommendation, and methodological aspects,’’ IEEE Access,
vol. 7, pp. 51069–51103, 2019.
[35] J. Chen, J. Chen, D. Zhang, Y. A. Nanehkaran, and Y. Sun, ‘‘A cognitive
vision method for the detection of plant disease images,’’ Mach. Vis.
Appl., vol. 32, no. 1, pp. 1–18, Jan. 2021.
[36] J. Ma, K. Du, F. Zheng, L. Zhang, and Z. Sun, ‘‘A segmentation method
for processing greenhouse vegetable foliar disease symptom images,’’ Inf.
Process. Agricult., vol. 6, no. 2, pp. 216–223, Jun. 2019.
[37] G. Yang, G. Chen, Y. He, Z. Yan, Y. Guo, and J. Ding, ‘‘Self-supervised
collaborative multi-network for fine-grained visual categorization of
tomato diseases,’’ IEEE Access, vol. 8, pp. 211912–211923, 2020.
[38] M. Lv, G. Zhou, M. He, A. Chen, W. Zhang, and Y. Hu, ‘‘Maize leaf
disease identification based on feature enhancement and DMS-robust
alexnet,’’ IEEE Access, vol. 8, pp. 57952–57966, 2020.
[39] H. Ali, M. I. Lali, M. Z. Nawaz, M. Sharif, and B. A. Saleem, ‘‘Symptom
based automated detection of citrus diseases using color histogram and
textural descriptors,’’ Comput. Electron. Agricult., vol. 138, pp. 92–104,
Jun. 2017.
VOLUME 11, 2023 59199
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
[40] Y. Zhong and M. Zhao, ‘‘Research on deep learning in apple leaf
disease recognition,’’ Comput. Electron. Agricult., vol. 168, Jan. 2020,
Art. no. 105146.
[41] W. Haider, A.-U. Rehman, N. M. Durrani, and S. U. Rehman,
‘‘A generic approach for wheat disease classification and verification
using expert opinion for knowledge-based decisions,’’ IEEE Access,
vol. 9, pp. 31104–31129, 2021.
[42] T. Rumpf, A.-K. Mahlein, U. Steiner, E.-C. Oerke, H.-W. Dehne, and
L. Plümer, ‘‘Early detection and classification of plant diseases with
support vector machines based on hyperspectral reflectance,’’ Comput.
Electron. Agricult., vol. 74, no. 1, pp. 91–99, Oct. 2010.
[43] S. Liaghat, S. Mansor, R. Ehsani, H. Z. M. Shafri, S. Meon, and
S. Sankaran, ‘‘Mid-infrared spectroscopy for early detection of basal stem
rot disease in oil palm,’’ Comput. Electron. Agricult., vol. 101, pp. 48–54,
Feb. 2014.
[44] R. Zhou, S. Kaneko, F. Tanaka, M. Kayamori, and M. Shimizu, ‘‘Disease
detection of Cercospora Leaf Spot in sugar beet by robust template
matching,’’ Comput. Electron. Agricult., vol. 108, pp. 58–70, Oct. 2014.
[45] É. A. S. Moriya, N. N. Imai, A. M. G. Tommaselli, and G. T. Miyoshi,
‘‘Mapping mosaic virus in sugarcane based on hyperspectral images,’’
IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 10, no. 2,
pp. 740–748, Feb. 2017.
[46] A. AlSuwaidi, B. Grieve, and H. Yin, ‘‘Feature-ensemble-based novelty
detection for analyzing plant hyperspectral datasets,’’ IEEE J. Sel. Top-
ics Appl. Earth Observ. Remote Sens., vol. 11, no. 4, pp. 1041–1055,
Apr. 2018.
[47] L. M. Griffel, D. Delparte, and J. Edwards, ‘‘Using support vector
machines classification to differentiate spectral signatures of potato plants
infected with potato virus y,’’ Comput. Electron. Agricult., vol. 153,
pp. 318–324, Oct. 2018.
[48] S. Pandiyan, M. Ashwin, R. Manikandan, K. M. K. Raghunath, and
G. R. A. Raman, ‘‘Heterogeneous Internet of Things organization pre-
dictive analysis platform for apple leaf diseases recognition,’’ Comput.
Commun., vol. 154, pp. 99–110, Mar. 2020.
[49] W. Chen, Y. Lin, F. Ng, C. Liu, and Y. Lin, ‘‘RiceTalk: Rice blast detection
using Internet of Things and artificial intelligence technologies,’’ IEEE
Internet Things J., vol. 7, no. 2, pp. 1001–1010, Feb. 2020.
[50] R. Karthik, M. Hariharan, S. Anand, P. Mathikshara, A. Johnson, and
R. Menaka, ‘‘Attention embedded residual CNN for disease detection in
tomato leaves,’’ Appl. Soft Comput., vol. 86, Jan. 2020, Art. no. 105933.
[51] V. Udutalapally, S. P. Mohanty, V. Pallagani, and V. Khandelwal, ‘‘SCrop:
A novel device for sustainable automatic disease prediction, crop selec-
tion, and irrigation in Internet-of-Agro-Things for smart agriculture,’’
IEEE Sensors J., vol. 21, no. 16, pp. 17525–17538, Aug. 2021.
[52] J. P. Rodríguez, A. I. Montoya-Munoz, C. Rodriguez-Pabon, J. Hoyos,
and J. C. Corrales, ‘‘IoT-agro: A smart farming system to Colom-
bian coffee farms,’’ Comput. Electron. Agricult., vol. 190, Nov. 2021,
Art. no. 106442.
[53] M. Mishra, P. Choudhury, and B. Pati, ‘‘Modified ride-NN optimizer
for the IoT based plant disease detection,’’ J. Ambient Intell. Humanized
Comput., vol. 12, no. 1, pp. 691–703, Jan. 2021.
[54] N. Kundu, G. Rani, V. S. Dhaka, K. Gupta, S. C. Nayak, S. Verma,
M. F. Ijaz, and M. Woźniak, ‘‘IoT and interpretable machine learning
based framework for disease prediction in pearl millet,’’ Sensors, vol. 21,
no. 16, p. 5386, Aug. 2021.
[55] D. Sarpal, R. Sinha, M. Jha, and P. Tn, ‘‘AgriWealth: IoT based
farming system,’’ Microprocessors Microsyst., vol. 89, Mar. 2022,
Art. no. 104447.
[56] P. Chen, W. Li, S. Yao, C. Ma, J. Zhang, B. Wang, C. Zheng, C. Xie, and
D. Liang, ‘‘Recognition and counting of wheat mites in wheat fields by a
three-step deep learning method,’’ Neurocomputing, vol. 437, pp. 21–30,
May 2021.
[57] A. Johannes, A. Picon, A. Alvarez-Gila, J. Echazarra,
S. Rodriguez-Vaamonde, A. D. Navajas, and A. Ortiz-Barredo,
‘‘Automatic plant disease diagnosis using mobile capture devices,
applied on a wheat use case,’’ Comput. Electron. Agricult., vol. 138,
pp. 200–209, Jun. 2017.
[58] A. M. Abdu, M. M. Mokji, and U. U. Sheikh, ‘‘Automatic vegetable dis-
ease identification approach using individual lesion features,’’ Comput.
Electron. Agricult., vol. 176, Sep. 2020, Art. no. 105660.
[59] K. K. Chakraborty, R. Mukherjee, C. Chakroborty, and K. Bora, ‘‘Auto-
mated recognition of optical image based potato leaf blight diseases
using deep learning,’’ Physiolog. Mol. Plant Pathol., vol. 117, Jan. 2022,
Art. no. 101781.
[60] C. Hou, J. Zhuang, Y. Tang, Y. He, A. Miao, H. Huang, and S. Luo,
‘‘Recognition of early blight and late blight diseases on potato leaves
based on graph cut segmentation,’’ J. Agricult. Food Res., vol. 5,
Sep. 2021, Art. no. 100154.
[61] H. F. Pardede, E. Suryawati, R. Sustika, and V. Zilvan, ‘‘Unsupervised
convolutional autoencoder-based feature learning for automatic detection
of plant diseases,’’ in Proc. Int. Conf. Comput., Control, Informat. Appl.
(IC3INA), Nov. 2018, pp. 158–162.
[62] C. Douarre, C. F. Crispim-Junior, A. Gelibert, L. Tougne, and
D. Rousseau, ‘‘Novel data augmentation strategies to boost supervised
segmentation of plant disease,’’ Comput. Electron. Agricult., vol. 165,
Oct. 2019, Art. no. 104967.
[63] P. Jiang, Y. Chen, B. Liu, D. He, and C. Liang, ‘‘Real-time detection
of apple leaf diseases using deep learning approach based on improved
convolutional neural networks,’’ IEEE Access, vol. 7, pp. 59069–59080,
2019.
[64] M. A. Khan, M. I. U. Lali, M. Sharif, K. Javed, K. Aurangzeb,
S. I. Haider, A. S. Altamrah, and T. Akram, ‘‘An optimized method
for segmentation and classification of apple diseases based on strong
correlation and genetic algorithm based feature selection,’’ IEEE Access,
vol. 7, pp. 46261–46277, 2019.
[65] H. Sun, H. Xu, B. Liu, D. He, J. He, H. Zhang, and N. Geng,
‘‘MEAN-SSD: A novel real-time detector for apple leaf diseases using
improved light-weight convolutional neural networks,’’ Comput. Elec-
tron. Agricult., vol. 189, Oct. 2021, Art. no. 106379.
[66] H.-J. Yu and C.-H. Son, ‘‘Leaf spot attention network for apple leaf
disease identification,’’ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern
Recognit. Workshops (CVPRW), Jun. 2020, pp. 52–53.
[67] U. Barman, R. D. Choudhury, D. Sahu, and G. G. Barman, ‘‘Compar-
ison of convolution neural networks for smartphone image based real
time classification of citrus leaf disease,’’ Comput. Electron. Agricult.,
vol. 177, Oct. 2020, Art. no. 105661.
[68] S. Dananjayan, Y. Tang, J. Zhuang, C. Hou, and S. Luo, ‘‘Assessment
of state-of-the-art deep learning based citrus disease detection tech-
niques using annotated optical leaf images,’’ Comput. Electron. Agricult.,
vol. 193, Feb. 2022, Art. no. 106658.
[69] J. Chen, W. Wang, D. Zhang, A. Zeb, and Y. A. Nanehkaran, ‘‘Attention
embedded lightweight network for maize disease recognition,’’ Plant
Pathol., vol. 70, no. 3, pp. 630–642, Apr. 2021.
[70] A. Darwish, D. Ezzat, and A. E. Hassanien, ‘‘An optimized model based
on convolutional neural networks and orthogonal learning particle swarm
optimization algorithm for plant diseases diagnosis,’’ Swarm Evol. Com-
put., vol. 52, Feb. 2020, Art. no. 100616.
[71] K. Garg, S. Bhugra, and B. Lall, ‘‘Automatic quantification of plant
disease from field image data using deep learning,’’ in Proc. IEEE Winter
Conf. Appl. Comput. Vis. (WACV), Jan. 2021, pp. 1964–1971.
[72] S. Hernández and J. L. López, ‘‘Uncertainty quantification for plant
disease detection using Bayesian deep learning,’’ Appl. Soft Comput.,
vol. 96, Nov. 2020, Art. no. 106597.
[73] J. Sun, Y. Yang, X. He, and X. Wu, ‘‘Northern maize leaf blight detection
under complex field environment based on deep learning,’’ IEEE Access,
vol. 8, pp. 33679–33688, 2020.
[74] A. Waheed, M. Goyal, D. Gupta, A. Khanna, A. E. Hassanien, and
H. M. Pandey, ‘‘An optimized dense convolutional neural network model
for disease recognition and classification in corn leaf,’’ Comput. Electron.
Agricult., vol. 175, Aug. 2020, Art. no. 105456.
[75] A. Abbas, S. Jain, M. Gour, and S. Vankudothu, ‘‘Tomato plant disease
detection using transfer learning with C-GAN synthetic images,’’ Com-
put. Electron. Agricult., vol. 187, Aug. 2021, Art. no. 106279.
[76] V. Gonzalez-Huitron, J. A. León-Borges, A. E. Rodriguez-Mata,
L. E. Amabilis-Sosa, B. Ramírez-Pereda, and H. Rodriguez, ‘‘Disease
detection in tomato leaves via CNN with lightweight architectures
implemented in raspberry pi 4,’’ Comput. Electron. Agricult., vol. 181,
Feb. 2021, Art. no. 105951.
[77] J. Liu and X. Wang, ‘‘Early recognition of tomato gray leaf spot disease
based on MobileNetv2-YOLOv3 model,’’ Plant Methods, vol. 16, no. 1,
pp. 1–16, Dec. 2020.
[78] H. Nazki, S. Yoon, A. Fuentes, and D. S. Park, ‘‘Unsupervised image
translation using adversarial networks for improved plant disease recog-
nition,’’ Comput. Electron. Agricult., vol. 168, Jan. 2020, Art. no. 105117.
[79] J. Zhou, J. Li, C. Wang, H. Wu, C. Zhao, and Q. Wang, ‘‘A vegetable
disease recognition model for complex background based on region
proposal and progressive learning,’’Comput. Electron. Agricult., vol. 184,
May 2021, Art. no. 106101.
59200 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
[80] M. Ji, L. Zhang, and Q. Wu, ‘‘Automatic grape leaf diseases identification
via UnitedModel based on multiple convolutional neural networks,’’ Inf.
Process. Agricult., vol. 7, no. 3, pp. 418–426, Sep. 2020.
[81] M. Ji and Z. Wu, ‘‘Automatic detection and severity analysis of grape
black measles disease based on deep learning and fuzzy logic,’’ Comput.
Electron. Agricult., vol. 193, Feb. 2022, Art. no. 106718.
[82] P. Kaur, H. S. Pannu, and A. K. Malhi, ‘‘Plant disease recognition using
fractional-order Zernike moments and SVM classifier,’’ Neural Comput.
Appl., vol. 31, no. 12, pp. 8749–8768, Dec. 2019.
[83] M. Kerkech, A. Hafiane, and R. Canals, ‘‘Vine disease detection in
UAV multispectral images using optimized image registration and deep
learning segmentation approach,’’ Comput. Electron. Agricult., vol. 174,
Jul. 2020, Art. no. 105446.
[84] Y. Li, J. Nie, and X. Chao, ‘‘Do we really need deep CNN for plant dis-
eases identification?’’ Comput. Electron. Agricult., vol. 178, Nov. 2020,
Art. no. 105803.
[85] B. Liu, C. Tan, S. Li, J. He, and H. Wang, ‘‘A data augmentation method
based on generative adversarial networks for grape leaf disease identifi-
cation,’’ IEEE Access, vol. 8, pp. 102188–102198, 2020.
[86] Z. Tang, J. Yang, Z. Li, and F. Qi, ‘‘Grape disease image classification
based on lightweight convolution neural networks and channelwise atten-
tion,’’ Comput. Electron. Agricult., vol. 178, Nov. 2020, Art. no. 105735.
[87] R. Pydipati, T. F. Burks, and W. S. Lee, ‘‘Identification of citrus disease
using color texture features and discriminant analysis,’’ Comput. Electron.
Agricult., vol. 52, nos. 1–2, pp. 49–59, Jun. 2006.
[88] Q. H. Cap, H. Tani, S. Kagiwada, H. Uga, and H. Iyatomi, ‘‘LASSR:
Effective super-resolution method for plant disease diagnosis,’’ Comput.
Electron. Agricult., vol. 187, Aug. 2021, Art. no. 106271.
[89] Q. H. Cap, H. Uga, S. Kagiwada, and H. Iyatomi, ‘‘LeafGAN: An
effective data augmentation method for practical plant disease diagnosis,’’
IEEE Trans. Autom. Sci. Eng., vol. 19, no. 2, pp. 1258–1267, Apr. 2022.
[90] J. Ma, K. Du, L. Zhang, F. Zheng, J. Chu, and Z. Sun, ‘‘A segmentation
method for greenhouse vegetable foliar disease spots images using color
information and region growing,’’ Comput. Electron. Agricult., vol. 142,
pp. 110–117, Nov. 2017.
[91] C. Wang, P. Du, H. Wu, J. Li, C. Zhao, and H. Zhu, ‘‘A cucum-
ber leaf disease severity classification method based on the fusion
of DeepLabV3+ and U-Net,’’ Comput. Electron. Agricult., vol. 189,
Oct. 2021, Art. no. 106373.
[92] S. Zhang and Z. Wang, ‘‘Cucumber disease recognition based on
global-local singular value decomposition,’’ Neurocomputing, vol. 205,
pp. 341–348, Sep. 2016.
[93] S. Zhang, X. Wu, Z. You, and L. Zhang, ‘‘Leaf image based cucumber
disease recognition using sparse representation classification,’’ Comput.
Electron. Agricult., vol. 134, pp. 135–141, Mar. 2017.
[94] S. Zhang, S. Zhang, C. Zhang, X. Wang, and Y. Shi, ‘‘Cucumber leaf
disease identification with global pooling dilated convolutional neural
network,’’ Comput. Electron. Agricult., vol. 162, pp. 422–430, Jul. 2019.
[95] S. Zhang, Y. Zhu, Z. You, and X. Wu, ‘‘Fusion of superpixel, expectation
maximization and PHOG for recognizing cucumber diseases,’’ Comput.
Electron. Agricult., vol. 140, pp. 338–347, Aug. 2017.
[96] P. Zhang, L. Yang, and D. Li, ‘‘EfficientNet-B4-ranger: A novel
method for greenhouse cucumber disease recognition under natural com-
plex environment,’’ Comput. Electron. Agricult., vol. 176, Sep. 2020,
Art. no. 105652.
[97] N. V. R. R. Goluguri, K. S. Devi, and P. Srinivasan, ‘‘Rice-Net: An
efficient artificial fish swarm optimization applied deep convolutional
neural network model for identifying the Oryza sativa diseases,’’ Neural
Comput. Appl., vol. 33, no. 11, pp. 5869–5884, Jun. 2021.
[98] F. Jiang, Y. Lu, Y. Chen, D. Cai, and G. Li, ‘‘Image recognition of four
Rice leaf diseases based on deep learning and support vector machine,’’
Comput. Electron. Agricult., vol. 179, Dec. 2020, Art. no. 105824.
[99] Y. Lu, S. Yi, N. Zeng, Y. Liu, and Y. Zhang, ‘‘Identification of Rice
diseases using deep convolutional neural networks,’’ Neurocomputing,
vol. 267, pp. 378–384, Dec. 2017.
[100] A. Nigam, A. K. Tiwari, and A. Pandey, ‘‘Paddy leaf diseases recognition
and classification using PCA and BFO-DNN algorithm by image process-
ing,’’ Mater. Today, Proc., vol. 33, pp. 4856–4862, 2020.
[101] S. Phadikar, J. Sil, and A. K. Das, ‘‘Rice diseases classification using
feature selection and rule generation techniques,’’ Comput. Electron.
Agricult., vol. 90, pp. 76–85, Jan. 2013.
[102] S. Radhakrishnan, ‘‘An improved machine learning algorithm for pre-
dicting blast disease in paddy crop,’’ Mater. Today, Proc., vol. 33,
pp. 682–686, Jan. 2020.
[103] C. R. Rahman, P. S. Arko, M. E. Ali, M. A. I. Khan, S. H. Apon,
F. Nowrin, and A. Wasif, ‘‘Identification and recognition of Rice diseases
and pests using convolutional neural networks,’’ Biosyst. Eng., vol. 194,
pp. 112–120, Jun. 2020.
[104] S. Ramesh and D. Vydeki, ‘‘Recognition and classification of paddy leaf
diseases using optimized deep neural network with Jaya algorithm,’’ Inf.
Process. Agricult., vol. 7, no. 2, pp. 249–260, Jun. 2020.
[105] S. Sengupta and A. K. Das, ‘‘Particle swarm optimization based incre-
mental classifier design for Rice disease prediction,’’ Comput. Electron.
Agricult., vol. 140, pp. 443–451, Aug. 2017.
[106] P. K. Sethy, N. K. Barpanda, A. K. Rath, and S. K. Behera, ‘‘Deep feature
based Rice leaf disease identification using support vector machine,’’
Comput. Electron. Agricult., vol. 175, Aug. 2020, Art. no. 105527.
[107] V. Tiwari, R. C. Joshi, and M. K. Dutta, ‘‘Dense convolutional neural
networks based multiclass plant disease detection and classification using
leaf images,’’ Ecolog. Informat., vol. 63, Jul. 2021, Art. no. 101289.
[108] Y. Wang, H. Wang, and Z. Peng, ‘‘Rice diseases detection and classifi-
cation using attention based neural network and Bayesian optimization,’’
Expert Syst. Appl., vol. 178, Sep. 2021, Art. no. 114770.
[109] G. Zhou, W. Zhang, A. Chen, M. He, and X. Ma, ‘‘Rapid detection of
Rice disease based on FCM-KM and faster R-CNN fusion,’’ IEEE Access,
vol. 7, pp. 143190–143206, 2019.
[110] J. Chen, D. Zhang, M. Suzauddola, Y. A. Nanehkaran, and Y. Sun, ‘‘Iden-
tification of plant disease images via a squeeze-and-excitation MobileNet
model and twice transfer learning,’’ IET Image Process., vol. 15, no. 5,
pp. 1115–1127, Apr. 2021.
[111] R. Gozzovelli, B. Franchetti, M. Bekmurat, and F. Pirri, ‘‘Tip-burn stress
detection of lettuce canopy grown in plant factories,’’ in Proc. IEEE/CVF
Int. Conf. Comput. Vis. Workshops (ICCVW), Oct. 2021, pp. 1259–1268.
[112] J. Wang, L. Yu, J. Yang, and H. Dong, ‘‘DBA_SSD: A novel end-
to-end object detection algorithm applied to plant disease detection,’’
Information, vol. 12, no. 11, p. 474, Nov. 2021.
[113] W. Zeng and M. Li, ‘‘Crop leaf disease recognition based on self-attention
convolutional neural network,’’ Comput. Electron. Agricult., vol. 172,
May 2020, Art. no. 105341.
[114] X. Chen, G. Zhou, A. Chen, J. Yi, W. Zhang, and Y. Hu, ‘‘Iden-
tification of tomato leaf diseases based on combination of ABCK-
BWTR and B-ARNet,’’ Comput. Electron. Agricult., vol. 178, Nov. 2020,
Art. no. 105730.
[115] J. Chen, D. Zhang, M. Suzauddola, and A. Zeb, ‘‘Identifying crop diseases
using attention embedded MobileNet-V2 model,’’ Appl. Soft Comput.,
vol. 113, Dec. 2021, Art. no. 107901.
[116] Y. Zhao, C. Sun, X. Xu, and J. Chen, ‘‘RIC-Net: A plant disease classifi-
cation model based on the fusion of inception and residual structure and
embedded attention mechanism,’’ Comput. Electron. Agricult., vol. 193,
Feb. 2022, Art. no. 106644.
[117] S. Gutiérrez, I. Hernández, S. Ceballos, I. Barrio, A. M. Díez-Navajas,
and J. Tardaguila, ‘‘Deep learning for the differentiation of downy mildew
and spider mite in grapevine under field conditions,’’ Comput. Electron.
Agricult., vol. 182, Mar. 2021, Art. no. 105991.
[118] J. Ma, K. Du, F. Zheng, L. Zhang, Z. Gong, and Z. Sun, ‘‘A recognition
method for cucumber diseases using leaf symptom images based on deep
convolutional neural network,’’ Comput. Electron. Agricult., vol. 154,
pp. 18–24, Nov. 2018.
[119] K. Kc, Z. Yin, M. Wu, and Z. Wu, ‘‘Depthwise separable convolu-
tion architectures for plant disease classification,’’ Comput. Electron.
Agricult., vol. 165, Oct. 2019, Art. no. 104948.
[120] G. Geetharamani and A. Pandian, ‘‘Identification of plant leaf diseases
using a nine-layer deep convolutional neural network,’’ Comput. Electr.
Eng., vol. 76, pp. 323–338, Jun. 2019.
[121] A. Karlekar and A. Seal, ‘‘SoyNet: Soybean leaf diseases classification,’’
Comput. Electron. Agricult., vol. 172, May 2020, Art. no. 105342.
[122] Z. Li, Y. Yang, Y. Li, R. Guo, J. Yang, and J. Yue, ‘‘A solanaceae disease
recognition model based on SE-inception,’’ Comput. Electron. Agricult.,
vol. 178, Nov. 2020, Art. no. 105792.
[123] P. Sharma, Y. P. S. Berwal, and W. Ghai, ‘‘Performance analysis of deep
learning CNN models for disease detection in plants using image segmen-
tation,’’ Inf. Process. Agricult., vol. 7, no. 4, pp. 566–574, Dec. 2020.
VOLUME 11, 2023 59201
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
[124] S. Xiang, Q. Liang, W. Sun, D. Zhang, and Y. Wang, ‘‘L-CSMS: Novel
lightweight network for plant disease severity recognition,’’ J. Plant
Diseases Protection, vol. 128, no. 2, pp. 557–569, Apr. 2021.
[125] M. Agarwal, S. K. Gupta, and K. K. Biswas, ‘‘Development of efficient
CNN model for tomato crop disease identification,’’ Sustain. Comput.,
Informat. Syst., vol. 28, Dec. 2020, Art. no. 100407.
[126] K. P. Ferentinos, ‘‘Deep learning models for plant disease detection
and diagnosis,’’ Comput. Electron. Agricult., vol. 145, pp. 311–318,
Feb. 2018.
[127] J. G. A. Barbedo, ‘‘Impact of dataset size and variety on the effectiveness
of deep learning and transfer learning for plant disease classification,’’
Comput. Electron. Agricult., vol. 153, pp. 46–53, Oct. 2018.
[128] S. Coulibaly, B. Kamsu-Foguem, D. Kamissoko, and D. Traore, ‘‘Deep
neural networks with transfer learning in millet crop images,’’ Comput.
Ind., vol. 108, pp. 115–120, Jun. 2019.
[129] S. H. Lee, H. Goëau, P. Bonnet, and A. Joly, ‘‘New perspectives on
plant disease characterization based on deep learning,’’ Comput. Electron.
Agricult., vol. 170, Mar. 2020, Art. no. 105220.
[130] Y. Xiong, L. Liang, L. Wang, J. She, and M. Wu, ‘‘Identification of cash
crop diseases using automatic image segmentation algorithm and deep
learning with expanded dataset,’’ Comput. Electron. Agricult., vol. 177,
Oct. 2020, Art. no. 105712.
[131] K. Deeba and B. Amutha, ‘‘WITHDRAWN:ResNet–deep neural network
architecture for leaf disease classification,’’ Microprocessors Microsys-
tems, vol. 78, Oct. 2020, Art. no. 103364.
[132] Ü. Atila, M. Uçar, K. Akyol, and E. Uçar, ‘‘Plant leaf disease classification
using EfficientNet deep learning model,’’ Ecolog. Informat., vol. 61,
Mar. 2021, Art. no. 101182.
[133] P. Singh, A. Verma, and J. S. R. Alex, ‘‘Disease and pest infection
detection in coconut tree through deep learning techniques,’’ Comput.
Electron. Agricult., vol. 182, Mar. 2021, Art. no. 105986.
[134] J. W. Clohessy, S. Sanjel, G. K. O’Brien, R. Barocco, S. Kumar,
S. Adkins, B. Tillman, D. L. Wright, and I. M. Small, ‘‘Development of
a high-throughput plant disease symptom severity assessment tool using
machine learning image analysis and integrated geolocation,’’ Comput.
Electron. Agricult., vol. 184, May 2021, Art. no. 106089.
[135] Q. Zeng, X. Ma, B. Cheng, E. Zhou, and W. Pang, ‘‘GANs-based data
augmentation for citrus disease severity detection using deep learning,’’
IEEE Access, vol. 8, pp. 172882–172891, 2020.
[136] J. Abdulridha, Y. Ampatzidis, R. Ehsani, and A. I. de Castro, ‘‘Evaluating
the performance of spectral features and multivariate analysis tools to
detect laurel wilt disease and nutritional deficiency in avocado,’’ Comput.
Electron. Agricult., vol. 155, pp. 203–211, Dec. 2018.
[137] E. C. Too, L. Yujian, S. Njuki, and L. Yingchun, ‘‘A comparative study
of fine-tuning deep learning models for plant disease identification,’’
Comput. Electron. Agricult., vol. 161, pp. 272–279, Jun. 2019.
[138] R. Sujatha, J. M. Chatterjee, N. Jhanjhi, and S. N. Brohi, ‘‘Performance
of deep learning vs machine learning in plant leaf disease detection,’’
Microprocessors Microsyst., vol. 80, Feb. 2021, Art. no. 103615.
[139] Q. Liang, S. Xiang, Y. Hu, G. Coppola, D. Zhang, and W. Sun, ‘‘PD2SE-
Net: Computer-assisted plant disease diagnosis and severity estimation
network,’’ Comput. Electron. Agricult., vol. 157, pp. 518–529, Feb. 2019.
[140] G. Hu, H. Wu, Y. Zhang, and M. Wan, ‘‘A low shot learning method for
tea leaf’s disease identification,’’ Comput. Electron. Agricult., vol. 163,
Aug. 2019, Art. no. 104852.
[141] D. Argüeso, A. Picon, U. Irusta, A. Medela, M. G. San-Emeterio,
A. Bereciartua, and A. Alvarez-Gila, ‘‘Few-shot learning approach for
plant disease classification using images taken in the field,’’ Comput.
Electron. Agricult., vol. 175, Aug. 2020, Art. no. 105542.
[142] U. Fang, J. Li, X. Lu, L. Gao, M. Ali, and Y. Xiang, ‘‘Self-supervised
cross-iterative clustering for unlabeled plant disease images,’’ Neurocom-
puting, vol. 456, pp. 36–48, Oct. 2021.
[143] M. Turkoglu, B. Yanikoğlu, and D. Hanbay, ‘‘PlantDiseaseNet: Convo-
lutional neural network ensemble for plant disease and pest detection,’’
Signal, Image Video Process., vol. 16, no. 2, pp. 301–309, Mar. 2022.
[144] L. Goyal, C. M. Sharma, A. Singh, and P. K. Singh, ‘‘Leaf and spike
wheat disease detection & classification using an improved deep con-
volutional architecture,’’ Informat. Med. Unlocked, vol. 25, Jan. 2021,
Art. no. 100642.
[145] R. Alguliyev, Y. Imamverdiyev, L. Sukhostat, and R. Bayramov, ‘‘Plant
disease detection based on a deep model,’’ Soft Comput., vol. 25, no. 21,
pp. 13229–13242, Nov. 2021.
[146] P. Gui, W. Dang, F. Zhu, and Q. Zhao, ‘‘Towards automatic field plant
disease recognition,’’ Comput. Electron. Agricult., vol. 191, Dec. 2021,
Art. no. 106523.
[147] A. K. Rangarajan, R. Purushothaman, and M. Pérez-Ruiz, ‘‘Disease clas-
sification in aubergine with local symptomatic region using deep learning
models,’’ Biosyst. Eng., vol. 209, pp. 139–153, Sep. 2021.
[148] Y. Yuan, Z. Xu, and G. Lu, ‘‘SPEDCCNN: Spatial pyramid-oriented
encoder–decoder cascade convolution neural network for crop disease
leaf segmentation,’’ IEEE Access, vol. 9, pp. 14849–14866, 2021.
[149] J. Shin, Y. K. Chang, B. Heung, T. Nguyen-Quang, G. W. Price, and
A. Al-Mallahi, ‘‘A deep learning approach for RGB image-based pow-
dery mildew disease detection on strawberry leaves,’’ Comput. Electron.
Agricult., vol. 183, Apr. 2021, Art. no. 106042.
[150] R. D. L. Pires, D. N. Gonçalves, J. P. M. Oruê, W. E. S. Kanashiro,
J. F. Rodrigues, B. B. Machado, and W. N. Gonçalves, ‘‘Local descriptors
for soybean disease recognition,’’ Comput. Electron. Agricult., vol. 125,
pp. 48–55, Jul. 2016.
[151] S. S. Chouhan, A. Kaul, U. P. Singh, and S. Jain, ‘‘Bacterial foraging
optimization based radial basis function neural network (BRBFNN) for
identification and classification of plant leaf diseases: An automatic
approach towards plant pathology,’’ IEEE Access, vol. 6, pp. 8852–8863,
2018.
[152] Y. Sun, Z. Jiang, L. Zhang, W. Dong, and Y. Rao, ‘‘SLIC_SVM based
leaf diseases saliency map extraction of tea plant,’’ Comput. Electron.
Agricult., vol. 157, pp. 102–109, Feb. 2019.
[153] J. M. M. Araujo and Z. M. A. Peixoto, ‘‘A new proposal for automatic
identification of multiple soybean diseases,’’ Comput. Electron. Agricult.,
vol. 167, Dec. 2019, Art. no. 105060.
[154] H. Hamdani, A. Septiarini, A. Sunyoto, S. Suyanto, and F. Utaminingrum,
‘‘Detection of oil palm leaf disease based on color histogram and super-
vised classifier,’’ Optik, vol. 245, Nov. 2021, Art. no. 167753.
[155] N. Kaur, ‘‘Plant leaf disease detection using ensemble classification
and feature extraction,’’ Turkish J. Comput. Math. Educ. (TURCOMAT),
vol. 12, no. 11, pp. 2339–2352, 2021.
[156] C. Milhorance, F. Howland, E. Sabourin, and J.-F. L. Coq, ‘‘Tackling
the implementation gap of climate adaptation strategies: Understanding
policy translation in Brazil and Colombia,’’ Climate Policy, vol. 22,
nos. 9–10, pp. 1113–1129, Nov. 2022.
[157] P. S. Thakur, P. Khanna, T. Sheorey, and A. Ojha, ‘‘Trends in vision-based
machine learning techniques for plant disease identification: A systematic
review,’’ Expert Syst. Appl., vol. 208, Dec. 2022, Art. no. 118117.
[158] R. Thapa, K. Zhang, N. Snavely, S. Belongie, and A. Khan, ‘‘The plant
pathology challenge 2020 data set to classify foliar disease of apples,’’
Appl. Plant Sci., vol. 8, no. 9, Sep. 2020, Art. no. e11390.
[159] X. Nie, L. Wang, H. Ding, and M. Xu, ‘‘Strawberry verticillium wilt
detection network based on multi-task learning and attention,’’ IEEE
Access, vol. 7, pp. 170003–170011, 2019.
[160] D. Wang, J. Wang, W. Li, and P. Guan, ‘‘T-CNN: Trilinear convolutional
neural networks model for visual detection of plant diseases,’’ Comput.
Electron. Agricult., vol. 190, Nov. 2021, Art. no. 106468.
[161] D. Mathew, C. S. Kumar, and K. A. Cherian, ‘‘Foliar fungal disease
classification in banana plants using elliptical local binary pattern on mul-
tiresolution dual tree complex wavelet transform domain,’’ Inf. Process.
Agricult., vol. 8, no. 4, pp. 581–592, Dec. 2021.
[162] Y. Yuan, L. Chen, M. Li, N. Wu, L. Wan, and S. Wang, ‘‘Diagnosis of
nitrogen nutrition of Rice based on image processing of visible light,’’
in Proc. IEEE Int. Conf. Functional-Structural Plant Growth Modeling,
Simulation, Visualizat. Appl. (FSPMA), Nov. 2016, pp. 228–232.
[163] M. A. Khan, T. Akram, M. Sharif, M. Awais, K. Javed, H. Ali, and
T. Saba, ‘‘CCDF: Automatic system for segmentation and recognition
of fruit crops diseases based on correlation coefficient and deep CNN
features,’’ Comput. Electron. Agricult., vol. 155, pp. 220–236, Dec. 2018.
[164] S. S. Chouhan, A. Kaul, and U. P. Singh, ‘‘Radial basis function neural
network for the segmentation of plant leaf disease,’’ in Proc. 4th Int. Conf.
Inf. Syst. Comput. Netw. (ISCON), Nov. 2019, pp. 713–716.
[165] D. Mondal, D. K. Kole, and K. Roy, ‘‘Gradation of yellow mosaic virus
disease of okra and bitter gourd based on entropy based binning and
naive Bayes classifier after identification of leaves,’’ Comput. Electron.
Agricult., vol. 142, pp. 485–493, Nov. 2017.
[166] J. G. A. Barbedo, ‘‘Factors influencing the use of deep learning for plant
disease recognition,’’ Biosyst. Eng., vol. 172, pp. 84–91, Aug. 2018.
[167] A. Sinha and R. S. Shekhawat, ‘‘A novel image classification technique
for spot and blight diseases in plant leaves,’’ Imag. Sci. J., vol. 68, no. 4,
pp. 225–239, May 2020.
59202 VOLUME 11, 2023
W. Shafik et al.: Systematic Literature Review on Plant Disease Detection
[168] M. M. Agrawal and S. Agrawal, ‘‘Rice plant diseases detection & clas-
sifiation using deep learning models: A systematic review,’’ J. Crit. Rev.,
vol. 7, no. 11, pp. 4376–4390, 2020.
[169] J. Boulent, S. Foucher, J. Théau, and P.-L. St-Charles, ‘‘Convolutional
neural networks for the automatic identification of plant diseases,’’ Fron-
tiers Plant Sci., vol. 10, p. 941, Jul. 2019.
[170] S. Li, K. Li, Y. Qiao, and L. Zhang, ‘‘A multi-scale cucumber disease
detection method in natural scenes based on YOLOv5,’’ Comput. Elec-
tron. Agricult., vol. 202, Nov. 2022, Art. no. 107363.
[171] A. I. Khan, S. M. K. Quadri, S. Banday, and J. L. Shah, ‘‘Deep diagnosis:
A real-time apple leaf disease detection system based on deep learning,’’
Comput. Electron. Agricult., vol. 198, Jul. 2022, Art. no. 107093.
[172] A. Dhakal and S. Shakya, ‘‘Image-based plant disease detection with
deep learning,’’ Int. J. Comput. Trends Technol., vol. 61, no. 1, pp. 26–29,
Jul. 2018.
[173] P. Mishra, G. Polder, and N. Vilfan, ‘‘Close range spectral imaging for
disease detection in plants using autonomous platforms: A review on
recent studies,’’ Current Robot. Rep., vol. 1, no. 2, pp. 43–48, Jun. 2020.
[174] S. S. Sannakki, V. S. Rajpurohit, V. B. Nargund, and P. Kulkarni, ‘‘Diag-
nosis and classification of grape leaf diseases using neural networks,’’
in Proc. 4th Int. Conf. Comput., Commun. Netw. Technol. (ICCCNT),
Jul. 2013, pp. 1–5.
[175] S. Prasad, S. K. Peddoju, and D. Ghosh, ‘‘Multi-resolution mobile vision
system for plant leaf disease diagnosis,’’ Signal, Image Video Process.,
vol. 10, no. 2, pp. 379–388, Feb. 2016.
[176] W. Albattah, M. Nawaz, A. Javed, M. Masood, and S. Albahli, ‘‘A novel
deep learning method for detection and classification of plant diseases,’’
Complex Intell. Syst., vol. 8, pp. 507–524, Feb. 2022.
WASSWA SHAFIK (Member, IEEE) received the
Bachelor of Science degree in information tech-
nology engineering with a minor in mathemat-
ics from Ndejje University, Kampala, Uganda,
in 2016, and the Master of Engineering degree in
information technology engineering (MIT) from
Yazd University, Yazd, Iran, in 2020. He is cur-
rently pursuing the Ph.D. degree in computer sci-
ence with the School of Digital Science, Universiti
Brunei Darussalam, Brunei Darussalam. He is also
the Founder and a Principal Investigator of the Dig Connectivity Research
Laboratory (DCRLab) after serving as a Research Associate at Network
Interconnectivity Research Laboratory, Yazd University. Prior to this, he
worked as a Community Data Analyst at Population Services International
(PSI-Uganda), Community Data Officer at Programme for Accessible Health
Communication (PACE-Uganda), Research Assistant at the Socio-Economic
Data Centre (SEDC-Uganda), Prime Minister’s Office,Kampala, Uganda, an
Assistant Data Officer at TechnoServe, Kampala, IT Support at Thurayya
Islam Media, Uganda, and Asmaah Charity Organization. He has more
than 60 publications in renowned journals and conferences. His research
interests include plant pathology, computer vision, AI-enabled IoT/IoMTs,
the IoT/IIoT/OT security, cyber security, and privacy.
ALI TUFAIL (Senior Member, IEEE) received the
bachelor’s degree in information technology from
the National University of Sciences and Tech-
nology, Pakistan, in 2005, the Master of Science
degree in advanced computing from the University
of Bristol, U.K., in 2006, and the Ph.D. degree in
information and communication engineering from
Ajou University, South Korea, in 2011. He is cur-
rently a Senior Assistant Professor in distributed
and cyber-physical systems with the School of
Digital Science (SDS), Universiti Brunei Darussalam. He is also serving
as a SDS seminar coordinator and a learning technology advisor. He has
more than ten years of teaching experience in countries, such as Pakistan,
South Korea, Turkey, and Saudi Arabia. He has more than 25 publications in
renowned journals and conferences. His Teaching and Research specializa-
tions are in the areas of Wireless Sensor Networks, the Internet of Things,
and Vehicular Adhoc Networks.
ABDALLAH NAMOUN (Member, IEEE) recei-
ved the bachelor’s degree in computer science,
in 2004, and the Ph.D. degree in informat-
ics from The University of Manchester, U.K.,
in 2009. He is currently an Associate Professor in
intelligent interactive systems and the Head of the
Information Technology Department, Faculty of
Computer and Information Systems, Islamic Uni-
versity of Madinah. He has authored more than
80 publications in research areas spanning intelli-
gent systems, human–computer interaction, software engineering, and tech-
nology acceptance and adoption. He has extensive experience in leading
complex research projects (worth more than 21 million Euros) with several
distinguished SMEs, such as SAP, BT, and ATOS. He has investigated
user needs and interaction with modern interactive technologies, design
of composite software services, and methods for testing the usability and
acceptance of human-interfaces. His current research interests include inte-
grating state of the art artificial intelligence approaches in the design and the
development of interactive systems and smart spaces.
LIYANAGE CHANDRATILAK DE SILVA (Senior
Member, IEEE) received the B.Sc.Eng. degree
(Hons.) from the University of Moratuwa,
Sri Lanka, in 1985, the M.Phil. degree from The
Open University of Sri Lanka, in 1989, and the
M.Eng. and Ph.D. degrees from The University
of Tokyo, Japan, in 1992 and 1995, respectively.
He was with the University of Tokyo, Japan, from
1989 to 1995. From April 1995 to March 1997, he
pursued his postdoctoral research as a Researcher
at ATR (Advanced Telecommunication Research) Laboratories, Kyoto,
Japan. In March 1997, he joined The National University of Singapore as
a Lecturer, where he was an Assistant Professor, until June 2003. He was
with Massey University, New Zealand, from 2003 to 2007. Currently, he is
a Professor in engineering and the Deputy Dean of the Faculty of Integrated
Technologies, University of Brunei Darussalam. He has published over
160 technical papers in these areas in international conferences, journals,
and Japanese national conventions, and holds one Japanese national patent,
which was successfully sold to Sony Corporation Japan for commercial
utilization, and he holds one U.S. and one Brunei patents. His works have
been cited as one of the pioneering works in the bimodal (audio and video
signal based) emotion recognition by many researchers. His papers so far
have been cited by more than 4500 times (according to scholar.google.com)
with an H-index of 27. He received the Best Student Paper Award from
SPIE (The International Society for Optical Engineering) for an outstanding
paper contribution to the International Conference on Visual Communication
and Image Processing (VCIP), in 1995. He also received the National
University of Singapore Award, in 2001, and the Department of ECE
Teaching Commendation Award, in 2002. He is the Interim Chair of the
IEEE Brunei Darussalam Subsection. He was the General Chair of the
4th International Conference Computational Intelligence and Robotics and
Autonomous Systems (CIRAS2007) held in New Zealand.
ROSYZIE ANNA AWG HAJI MOHD APONG
received the B.Sc. degree in computer science
from Strathclyde University, in 2004, the M.Sc.
degree in multimedia and internet computing from
Loughborough University, in 2016, and the Ph.D.
degree in computer science from Manchester Uni-
versity, in 2018. She is currently a Lecturer with
the School of Digital Science, Universiti Brunei
Darussalam. She has published and reviewed
about ten papers. Her research interests include
text mining, the Internet of Things, and information retrieval.
VOLUME 11, 2023 59203
