A Review of Leaf Diseases Detection and Classification by Deep Learning

Abstract

Leaf’s primary function is to produce nutrients through photosynthesis and support the plant’s growth. Leaf diseases caused by bacteria or other pathogens can negatively impact agricultural yields. Immediate and early diagnosis of diseases is vital for plant health. The significant development of deep learning algorithms for leaf diseases classification and detection contributed to a solid tool with a robust and reliable accuracy rate. This study presents a comprehensive review of leaf diseases research in the literature. It also highlights the gaps that need to be filled as well as the obstacles and problems facing research projects. The total number of papers retrieved from five electronic databases is 256. We analyzed and classified them into seven research questions. The results demonstrate that 63% of the papers are journal articles, 35% are conference papers, and 2% are workshop papers.

Full text

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.DOI
A Review of Leaf Diseases Detection and
Classification by Deep Learning
ASSAD S. DOUTOUM1and BULENT TUGRUL1
1Department of Computer Engineering, Ankara University, Ankara, 06830, Turkey
Corresponding author: Bulent Tugrul (e-mail: btugrul@eng.ankara.edu.tr).
ABSTRACT Leaf’s primary function is to produce nutrients through photosynthesis and support the plant’s
growth. Leaf diseases caused by bacteria or other pathogens can negatively impact agricultural yields.
Immediate and early diagnosis of diseases is vital for plant health. The significant development of deep
learning algorithms for leaf diseases classification and detection contributed to a solid tool with a robust and
reliable accuracy rate. This study presents a comprehensive review of leaf diseases research in the literature.
It also highlights the gaps that need to be filled as well as the obstacles and problems facing research projects.
The total number of papers retrieved from five electronic databases is 256. We analyzed and classified them
into seven research questions. The results demonstrate that 63% of the papers are journal articles, 35% are
conference papers, and 2% are workshop papers.
INDEX TERMS Classification, Deep Learning, Leaf Diseases
I. INTRODUCTION
AGRICULTURAL products serve as the main source of
economic output and revenue for most nations. There
are several diseases that affect crops and have a significant
impact on the productivity and income of farmers. Leaf
diseases are the primary issue that reduces agricultural pro-
ductivity [1]. According to the studies, 50% of crop losses
are caused by plant diseases and pets [2]. Managing and
controlling diseases is essential to increasing crop productiv-
ity. Keeping track of crops and diagnosing them at the right
time is essential to eliminating plant diseases. Discovering
diseases in the early stages enables farmers to avoid damage,
lower production costs, and improve profits. Traditional di-
agnosis by the human eye fails to detect diseases in the plant
at an early stage or misdiagnoses them [3]. Machine learning
and deep learning have been widely used in agriculture and
agricultural disease diagnosis and detection in recent years.
Leaf diseases detection and classification at early stages
are essential in agriculture. However, there are different ways
to identify plant diseases. Various types of diseases have
no visible symptoms, which require sophisticated analysis.
Meanwhile, most diseases produce a visible spectrum on the
leaf that a specialist can examine. Achieving accuracy on
plant diseases requires proper monitoring skills to distinguish
feature symptoms [4]. Crops are affected by many diseases
and we can effectively manage their spread. In addition to
minimizing crop losses, it also ensures excellent yields for
economic growth [3]. We review the many types of research
that have been done on plant diseases and plant disease
recognition. The aim is to facilitate the research in this
field that researchers have done previously in detecting and
classifying leaf diseases on images using machine learning
and deep learning architectures [5]. Various machine learn-
ing and deep learning methods have been used to increase
classification and detection accuracy, including the k-means
method, Fuzzy Logic (FL), Random Forest (RF), Artificial
Neural Network (ANN), Support Vector Machine (SVM),
Convolutional Neural Networks (CNNs) [6].
Image processing and cutting-edge deep learning methods
are widely used to diagnose leaf diseases. Different types of
popular CNN architectures have accomplished excellent jobs
in training and testing the image, such as AlexNet [7], LeNet
[8], InceptionV3 [9], VGGNet [2], ResNet [10], GoogLeNet
[11] and DenseNet [12] significantly improves the accuracy
in the detection and classification of leaf diseases [13].
Thus, this literature review aims to classify leaf disease
through state-of-the-art detection and classification using
deep learning and machine learning. The observation exam-
ines considerable research contributions and challenges, and
substantial work has been done to create applications that
improve deep learning architectures [14].
This paper is organized as follows: Section 2 outlines the
background, and Section 3 provides a detailed overview of
the study. We present and discuss our results in Section 4.
VOLUME 4, 2016 1
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
The final section summarizes the conclusions presented in
the paper.
II. BACKGROUND
The detection and classification of diseases on the leaf are
among the hottest topics in computer vision related to agri-
culture and agricultural activities. Farmers use traditional
methods to identify plant diseases. Manual vision leads to
incorrect diagnosis and failed symptom evaluation. After
AlexNet’s debut, other cutting-edge DL models and archi-
tectures for image detection, segmentation, and classification
emerged. The study done to identify and categorize plant
diseases using well-known DL architectures is presented
in this part. Additionally, in other related research papers,
improved/modified DL architectures and novel visualization
techniques were used to produce a better accuracy rate [15].
DL architectures produce more accurate findings than cus-
tomized ML-based techniques, enabling better choices to
be made. Due to the rapid advancement of hardware, DL
frameworks are being extensively researched to find solutions
to difficult issues in a respectably short amount of time. In the
sphere of crops, DL-based approaches demonstrate cutting-
edge precision and generalize well to various jobs. Different
kinds of deep neural networks (DNNs) have surpassed hyper-
spectral evaluation in terms of efficiency [16].
A. DEEP LEARNING
Deep learning (DL) is a subset of machine learning (ML).
Deep learning offers state-of-the-art products in several
computer vision domains. Before deep learning, many ap-
proaches were suggested to identify plant diseases using
image processing and machine learning. These methods are
based on hand-crafted features that lack automation, such as
SIFT, HOG, and SURF [17]. In addition, image labeling must
be manually performed, making data preprocessing very
expensive and time-consuming. Prior studies required a small
dataset for training and testing due to these obstacles, leading
to overfitting. Deep learning has many advantages in plant
disease detection and classification, including an end-to-end
system and the capability to exploit images directly. Deep
learning can train as many images as possible, compared with
traditional machine learning classifiers. The most significant
advantage is that deep learning architecture achieves better
accuracy results [18].
B. CONVOLUTIONAL NEURAL NETWORKS (CNN)
Convolutional Neural Networks (CNNs) are a state-of-the-
art deep learning architecture. CNN was invented several
years ago and has been applied in different domains. CNN
shows excellent success, especially in computer vision-
related tasks. CNN consists of two primary parts: feature
extraction (learning) and classification. Feature extraction
consists of various layers, convolutional ReLU, and pooling
layers. Image classification comprises a fully connected layer
and normalization [19].
CNN is a powerful deep-learning model for image detec-
tion and classification in various computer vision branches.
CNN performs far better than traditional classifiers because
it is less complicated and follows a diverse regularization
approach. It can learn fundamental filters automatically and
sort them accordingly [17]. CNN can train on a vast amount
of data to achieve reliable results. However, this feature does
not apply to traditional approaches. An additional concept
that CNN models can discover with small or large datasets
is transfer learning. Machine learning and deep learning have
been incorporated into CNN developed in recent years. These
architectures have significantly improved optimization, reg-
ularization, and structural reformulation. As a result of its
revised architectural design and structure, CNN has improved
significantly. The newly developed architecture features a
modification to its design. CNN architectures are classi-
fied into seven classes: spatial exploitation, depth, multi-
path, width, feature-map exploitation, channel boosting, and
attention-oriented CNNs. LeNet, AlexNet, VGGNet, and
GoogLeNet are CNN models that rely on spatial exploitation.
A large number of parameters and hyperparameters char-
acterize these models. ResNet and Inception V3 are depth-
based CNN models characterized by increased depth, which
is essential in supervision training. DenseNet is a multi-path-
based CNN model introduced to solve vanishing gradient
problems.
Inception families are width-based CNN models that im-
prove intermediate layer output. SqueezeNet, one of the
feature-map exploitation-based CNN methods, introduces an
innovative block. The block comprises squeeze and excita-
tion procedures to initiate feature-map-wise statistics. The
channel-boosted CNN algorithm boosts the number of input
channels. Image processing relies on image representation
for model performance. Therefore, deep learning classifiers
are concerned with image representation to improve the net-
work’s capacity. Residual attention neural networks (RAN)
and convolutional block attention modules are attention-
based CNN classifiers focusing on image localization and
recognition. The purpose of attention in CNN is to enable
the network to learn objects [20].
C. TYPES OF LEAF DISEASES
There are two main types of leaf diseases: biotic and abiotic
agents. Living organisms are called biotics, and nonliving or-
ganisms are called abiotics. Diseases caused by biotic agents
include insects, bacteria, fungi, and viruses. At the same time,
an abiotic agent includes extremes of temperature, excess
moisture, poor light, poor soil, and insufficient nutrients. Al-
though leaf diseases cause significant crop losses and directly
affect the economy and animal and human health, yield losses
can be reduced and specific toxoids can be adopted to battle
specific pathogens if diseases are appropriately diagnosed
and detected early [21].
2VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
III. LITERATURE REVIEW PLANNING
We revise our literature review plan in this section. According
to Bischoff et al. [22], this methodology aims to facilitate
more reliable and trustworthy discovery by reducing bias
through a precise review approach. This planning section
presents the requirements and the stimulation for conducting
a literature review. In addition, successful outcome protocols
for literature reviews are substantial components [14]. Table
1 presents the description of the research questions.
A. GOAL AND RESEARCH QUESTION
This review aims to contribute to a broad review of the latest
literature and to reveal research gaps, challenges, and barriers
that are valuable to explore from the perspective of leaf
diagnosis, detection, and classification. The main research
question of the literature review that we plan to answer
is: Which machine and deep learning algorithms have been
used to diagnose, classify and detect leaf diseases and which
method gives better accuracy results?
B. RESEARCH STRATEGY
This literature review examines studies with relevant results
and applies them to leaf diseases. Articles are selected ac-
cording to their topic, related to leaf diseases and using
machine learning and deep learning algorithms. Figure 1
shows the various sources used as our search databases.
Those databases provide authenticated articles, conferences,
literature reviews, and workshops. Various papers that detect
leaf diseases with deep learning methods are analyzed. The
primary search keywords examined in this review are leaf
diseases, deep learning, detection, classification, and algo-
rithm. The synonyms belonging to each principle term were
characterized. OR and AND Boolean operators were utilized
in our search criteria. Significantly, search engines can re-
store correlations for leaf disease detection and classification.
C. LITERATURE REVIEW SEARCH SELECTION
CRITERIA
The search selection criteria display how we refine related
articles retrieved from search engines. Our goal is to identify
primary articles related to our topic that answer our research
questions. A total of 256 studies were selected from vari-
ous databases. Articles published between 2006 and 2022
were included in the study analysis (Figure 2). We had to
exclude many irrelevant papers unrelated and duplicated to
the research criteria. A total of seven electronic databases
were used to select research studies for the procedure. Figure
1 shows the seven electronic databases and the number of
papers selected from each database. The papers were catego-
rized into three groups: articles, conferences, and workshops.
Google Scholar has the most articles with the attention of
23.82% (61), and science direct came next with the attention
of 17.18% (44). IEEE Xplore takes the lead for conference
papers with the attention of 30.85% (79) while 26.56 % (68)
of articles and conference papers are distributed between the
other databases. There are four workshop papers, Google
Scholar and IEEE Xplore with the attention of 1.17% (3), and
0.39% (1) respectively. Figure1 below shows the electronic
databases.
FIGURE 1. Research studies on electronic databases
D. PRISMA FLOW DIAGRAM
This literature review uses the PRISMA (Preferred Reporting
Items for Systematic Reviews and Meta-Analyses) diagram,
a research-based practice, to depict the review. The authors
utilize the PRISMA flow diagram to enhance overall paper
selection and improve the quality of the literature review [23].
Refer to the below Figure 2. This literature review covers
only papers written in English that pertain to the detection
and classification of leaf diseases through deep learning,
machine learning, and other methods [24]. For this literature
review, only English papers published in 2006 or later and not
duplicates have been chosen. Any paper that does not meet
these criteria will be excluded [25]. The chosen studies are
identified following these steps:
Identification: Articles were chosen from different
databases, including Google Scholar, IEEE, Springer Link,
Science Direct, Frontiers in Plant Science, Hindawi, and
MDPI. A total of 256 articles that are relevant to the topic
were selected.
Screening: Duplicate articles and papers were taken out.
Following the screening process, ten journal articles were
ineligible for inclusion.
Eligibility: We scanned and analyzed eighty-five research
papers to determine the outcomes. We excluded one hundred
and sixty-one studies for their irrelevance or lack of relevance
to the research questions and discussion.
Included: This systematic review was supported by forty
studies.
E. DATA EXTRACTION
We organized the papers into two categories: articles and
conference papers. The process is to examine each paper
and organize them by publication year. This data extraction
generates a list of research studies and evidence to answer
research questions. This study examines research studies
conducted between 2006 and 2022. In terms of detecting
VOLUME 4, 2016 3
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
RQ1: Which machine/deep learning algorithms
have been used to detect leaf diseases?
Study and evaluate the algorithms utilized in leaf
disease detection
Machine/deep learning
algorithms
RQ2: Which machine/deep learning architectures
have been used to detect leaf diseases?
Recognize the architectures used in leaf disease
detection and classification
Machine/deep learning
architectures
RQ3: Which databases have been used to detect
leaf diseases?
Discover the database where the dataset was col-
lected and the number of images
Name of the databases
RQ4: What type of plant leaf diseases and plants
have been examined?
Determine the disease types and yields that are
most infected
Leaf diseases
RQ5: What types of procedures and frameworks
have been used?
Determine the approaches used to classify and
detect diseases
Techniques and highest
and lowest accuracy
RQ6: Who are the authors and publishers? Locate the website name where the research was
published and the authors’ names
Research studies
RQ7: What are the advantages and disadvantages
of procedures and frameworks?
Identify the pros and cons of each type of proce-
dure and framework
Advantages and disad-
vantages
TABLE 1. Description of the Research Questions
FIGURE 2. PRISMA flow diagram
and classifying leaf diseases, 2019 was a milestone year.
Figure 3 shows the growth of research studies during this
period. At the termination of the composition activities, Table
9 was intended to be available in the extension, with 40
research studies selected for extraction. Ultimately, the table
consists of the following attributes: article (author’s name),
year (published year), plant (name of the plant), number of
images (number of images), dataset (location of the data),
framework, model or algorithm (the algorithm used in the
experiment), architecture (the architecture utilized), and best
accuracy.
FIGURE 3. Number of publications from the year 2006 to 2022
IV. RESULTS
We discuss 40 studies out of 256 studies in this section that
contributed to the answer to our research questions.
A. RQ1: WHICH MACHINE/DEEP LEARNING
ALGORITHMS HAVE BEEN USED TO DETECT LEAF
DISEASES?
RQ1 is intended to discuss which machine learning and deep
learning algorithms have been utilized to classify and detect
leaf diseases. Based on our findings from the research studies,
we found the most common algorithms to be Convolutional
Neural Networks (CNN), support vector machines (SVM),
artificial neural networks (ANN), and deep neural networks
(DNN). We realized that a Convolutional Neural Network
(CNN) is a widely used algorithm. There is a lot of interest
in deep learning, around 83.2%. There were around 9.5%
of machine learning algorithms in the following categories.
Around 7.4 percent of the research studies were devoted to
machine/deep learning algorithms and traditional algorithms
combined. Additionally, some studies used machine learning
and deep learning algorithms to achieve better accuracy and
compare model performance. For instance, Hasan et al. [26]
used CNN, SVM, KNN, and RF algorithms in Jute plant
diseases. Table 2 shows the prevalent 256 studies.
4VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
Type of Algorithm Number Percentage
Deep Learning 213 83.20%
Machine Learning 24 9.50%
Traditional Algorithms 10 3.90%
Machine and Deep Learning 9 3.50%
TABLE 2. Prevalent approximation of 256 studies
B. RQ2: WHICH MACHINE/DEEP LEARNING
ARCHITECTURES HAVE BEEN USED TO DETECT LEAF
DISEASES?
RQ2 evaluates machine learning and deep learning archi-
tectures for leaf disease detection and classification. Ta-
ble 3 shows the research studies containing different Deep
Learning (DL) architectures used in the selected papers. The
gathered information indicates that deep learning was the
most used with 65% (26), and other architectures received
35(14)%, respectively. Deep learning architectures such as
VGG, AlexNet, LeNet, InceptionV, and others were widely
used in selected research studies. In addition, multiple re-
search studies utilized different deep learning architectures in
training and testing datasets. Table 4 displays classification
and detection examples of different models. Deep learning
has different architectures that have been used in leaf disease
classification and detection. Table 5 shows the comparison of
different deep learning architectures.
Architectures No of articles Percentage
ResNet 5 12.50%
EfficientNet 3 7.50%
VGGNet 2 5.00%
AlexNet 2 5.00%
LeNet 2 5.00%
GoogLeNet 2 5.00%
YoloV 2 5.00%
MobileNet 1 2.50%
Inception 1 2.50%
Other 25 62.50%
TABLE 3. Deep learning architectures
Model Plant Accuracy
ResNet Tomato 97.28%
EfficientNetB7 Apple 99.80%
VGGNet Grape 98.40%
AlexNet Cucumber 94.27%
LeNet Banana 65.93%
GoogLeNet Pomelo 82.70%
YoloV5 Potato 99.75%
MobileNet Multiple 98.65%
InceptionV3 Rice 99.33%
TABLE 4. Classification and detection example
C. RQ3: WHICH DATABASES HAVE BEEN USED TO
DETECT LEAF DISEASES?
RQ3 concerns where the data is collected from and the
databases’ names. Images collected from the field (Self)
Architecture Pros Cons
ResNet Improves the efficiency Complexity
minimizing errors Requires more memory
EfficientNet Improve performance Many computational resources
Fewer parameters perform poorly on hardware
VGGNet Fewer parameters long training time
Smaller kernels large model size
AlexNet Breakthrough performance Complexity
Faster training Overfitting
LeNet Simplicity Overfitting
Easy to use Lack of interpretability
GoogLeNet Faster Large number of parameters
Smaller size Overfitting
TABLE 5. Pros and Cons of different architectures
are the most used in the research studies receiving 75%
(30), followed by PlantVillage at 20% (8). In comparison,
the Kaggle dataset reached 2.50% (1), and open-source and
other databases 2.50% (1), respectively. According to the
investigations of the research studies, most studies utilized
a digital camera and mobile phone to capture images from
the field. Figure 4 shows the database sources.
FIGURE 4. Data sources
D. RQ4: WHAT TYPE OF PLANT LEAF DISEASES AND
PLANTS HAVE BEEN EXAMINED?
In RQ4, the main goal is to discover the most common
leaf type in the literature review. Figure 5 categorizes the
surveyed yields in research studies into multiple kinds of
vegetables, fruits, and flowers.
Our results show that fruits are dominant in the survey.
The total number of crops is 40; multiple pieces are the most
dominant plant, 12.5% (5). Apple and corn are the second,
with a share of about 20 % (8). Cotton and cucumber come
next with a share of about 15% (6). The other crops receive
less attention than the aforementioned plants, 52.5% (21).
VOLUME 4, 2016 5
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
FIGURE 5. Type of plant leaf diseases
E. RQ5: WHAT TYPES OF PROCEDURES AND
FRAMEWORKS HAVE BEEN USED?
Specifically, RQ5 analyzes the contributions of research stud-
ies according to their techniques and frameworks. We dis-
covered multiple frameworks commonly used in traditional
and machine learning by examining research studies. Figure
6 shows the techniques and frameworks used in research
studies. It is clear that Matlab is the most framework used
to classify leaf diseases with a focus of 27.5% (11). The
TensorFlow framework comes in second with 15% (6). In the
third position are the PyTorch and Cafe frameworks together
with 25% (10). TensorFlow with Keras with 7.5% (3). Python
framework attention is 5% (2).
FIGURE 6. Most relevant frameworks used in the research studies
F. RQ6: WHO ARE THE AUTHORS AND PUBLISHERS?
Research studies are classified according to publishers and
types (articles/conferences) in RQ6. The research studies
were selected from 2006 to 2022, with articles related to leaf
diseases. Figures 1 and 3 categorize and classify publications
by year. The year 2019 has the most published papers with
26.29% (66), followed by 2021 with 21.48% (55) papers. The
year 2018 came next with 15.62% (40), followed by 2020
with 12.5% (32). While from 2006 to 2009, the number of
papers decreased by 0.39% (1) respectively.
G. RQ7: WHAT ARE THE ADVANTAGES AND
DISADVANTAGES OF PROCEDURES AND
FRAMEWORKS?
We examine some significant issues that come up when
creating neural network applications. Our goal is to discover
whether the selection of a library can affect the system’s
overall performance, either during training or design and to
derive a set of standards that might be utilized to demonstrate
the benefits and drawbacks of each library under study [27].
Table 6 displays the advantages and disadvantages of proce-
dures and frameworks.
Framework Advantages Disadvantages
Matlab Ease to use Cost
Platform independence Interrupted language
TensorFlow Scalability Process slow
Flexibility Missing Symbolic Loops
Caffe Fast Is not flexible
Open source Limited community
PyTorch Flexibility Low-level API
Easier to debug limited support
Keras Simplicity Inefficient errors
Backend support Low-level API
Python Ease to read Slow speed
Vast libraries support Runtime errors
TABLE 6. Advantages and disadvantages of procedures and frameworks
V. DISCUSSION
Our literature is organized into four main points on leaf
disease using deep learning, machine learning, and other
6VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
architectures. In this section, we will discuss the following
points in detail.
A. TRADITIONAL AND MACHINE LEARNING
APPROACHES
Several architectures have been proposed for the detection
and classification of leaf diseases. Traditional architectures,
such as texture features based on color co-occurrence meth-
ods, are widely used to detect leaf diseases [28]. Fuzzy
feature approaches can significantly enhance leaf disease
diagnosis. This approach uses the mean square error (MSE)
to estimate the performance [29]. Meunkaewjinda et al. [30]
proposed a technique employing a self-organizing feature
map based on a back-propagation neural network to diagnose
grape diseases. SVM, Bayesian, and K-nearest classifiers are
utilized in plant disease diagnosis [31].
B. DEEP LEARNING ALGORITHMS
Since 2012, the state-of-the-art Convolution Neural Network
(CNN) has been used widely in different domains, especially
plant diseases, and pets, to decrease crop loss [32]. Several
algorithms in this research study provide reliable solutions to
leaf diseases.
RQ2 indicates that the ResNet architecture is most com-
monly utilized in the investigated studies, followed by Ef-
ficientNet and AlexNet. Other deep learning architectures,
such as VGG, Faster R-CNN, and Yolo-V3, significantly im-
proved plant disease detection and classification [12]. Gen-
erally, most CNN architectures such as VGGNet, AlexNet,
and ResNet share categories like the number of parameters,
optimizers, hyperparameters, and the number of layers. Ac-
cording to the analysis of the research studies examination,
the use of architecture for leaf diseases varies depending on
the plant type and the architecture used. One architecture
works better for one plant disease and provides better accu-
racy results while not working well for another. Therefore,
several procedures and steps must be taken into consideration
to obtain more accurate results, including overfitting issues,
the number of hyperparameters, and the number of layers.
C. WHICH ARCHITECTURE PRODUCED THE HIGHEST
AND LOWEST ACCURACY?
Table 7 shows the extracted information on the architec-
ture with the highest accuracy for apple plants. The table
shows different studies that utilized various architectures
and obtained different accuracy results for apple plants. The
analysis revealed that the accuracy of the architectures varies.
However, compared to the other architectures, the CNN
architecture has the highest accuracy of 99.2%, while the
SegNet/GANs architecture has the lowest accuracy of 64.3%.
Plant Architecture Ref.# Highest Accu. Lowest Accu.
Apple
VGG16 [33] 90.40% 80.00%
AlexNet [34] 97.62% 86.79%
VGG [35] 85.00% 63.23%
FCNN-LDA [36] 90.00% 89.50%
VGG [37] 84.30% 74.70%
YOLOV3/DenseNet [12] 95.57% 93.84%
SegNet/GANs [38] 64.30% 52.90%
XDNet [39] 98.82% 96.29%
CNN [1] 96.25% 90.00%
ResNet-50 [40] 99.00% 98.74%
EfficientNetB7 [41] 91.46% 63.35%
CNN [42] 99.20% 87.20%
TABLE 7. Highest and Lowest Accuracy for Apple plant diseases
D. WHAT ARE THE MOST USED FEATURE
EXTRACTIONS IN CONVOLUTIONAL NEURAL
NETWORK (CNN) FOR LEAF DISEASE DETECTION AND
CLASSIFICATION?
Feature extraction is crucial for algorithms used in the clas-
sification and detection of leaf diseases. It is a significant
aspect of machine learning and image pattern classifica-
tion. Feature extraction enhances the processing performance
while minimizing redundancy and preserving the relevant
information. Features with the highest accuracy score are
selected for recognition. Some algorithms do not require
feature extraction, as discussed in [24], [43]. Table 8 presents
the feature extractions with the highest accuracy, used in
conjunction with the CNN algorithm. It was determined that
ResNet yielded the highest accuracy compared to other CNN
features.
Algorithm Feature extraction Ref.# Plant Accuracy
Convolutional Neural Network (CNN)
VGG [44] Cucumber 82.30%
InceptionV3 [45] Cassava 93.00%
LeNet [8] Banana 99.72%
VGG16 [33] Apple 90.40%
Faster-RCNN [46] Corn 95.00%
ResNet [47] Wheat 96.00%
SegNet/GANs [38] 64.3% 52.90%
ResNet-101 [48] Tomato 98.80%
ImageNet [49] Apple/Tomato 87.00%
GoogLeNet [11] Cherry 99.60%
ResNet [50] Corn 99.79%
AlexNet [51] Olive 99.11%
DCNN [52] Corn 88.46%
ResNet-101 [53] Maize 91.83%
ResNet-50 [54] Peanut 97.59%
EfficientNet [55] Cassava 87.00%
GoogLeNet [56] Pomelo 82.70%
YoloV5 [57] Sweet cherry 86.10%
EfficientNetB2 [58] Cardamom 98.26%
EfficientNetB7 [59] Grape 98.70%
TABLE 8. Feature extraction for CNN algorithm
E. DATABASES AND DATA-SETS CHARACTERISTICS
Multiple characteristics influence the datasets: the number
of images collected for training and testing, disease types,
pet type, and disease factors. Image preparation and adjust-
ment play an essential role in results accuracy. The reply to
VOLUME 4, 2016 7
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
the question RQ3 categorizes the dataset with the highest
repetition in research studies. We noticed that 75% of the
papers collected their dataset from the field (Self). The plant
village dataset came next with an accuracy of 20%. In Table
9 we categorized the type of dataset into (images collected
from the field (Self), plant village, Kaggle, open-source, and
other databases) that form the datasets. According to the
results, digital cameras, unmanned aerial vehicles (UAV),
and smartphones were the most commonly used devices to
capture field images. Besides, as mentioned earlier, different
other databases offer images for training and testing, such as
AI challenger and other databases.
F. TYPES OF PLANTS AND DISEASES
The research studies were applied to different vegetables
and fruits using different methods and algorithms. RQ4 dis-
cussed a variety of plants. The majority of research studies
focused on vegetables and fruits. Most studies focused on
leaf diseases such as tomato, rice, potato, corn, grape, and
apple plants. Wang et al. [60] discussed tomato disease.
For instance, malformed tomato, puffy tomato, tomato virus
disease, and other tomato diseases were researched in this
paper. Irmak and Saygili [61] researched tomato leaf diseases
such as Septoria leaf spot, yellow leaf curl, and bacterial spot.
G. TECHNIQUES AND FRAMEWORKS
Methods and frameworks play a critical role in study ac-
curacy. RQ 5 points out that the Matlab framework is the
most favored, with 27.5%. The TensorFlow framework is
also considered useful and utilized with a concern of 15%,
followed by Caffe and PyTorch frameworks with concerns of
25%. Matlab is primarily used as a framework in traditional
machine-learning studies. Various frameworks were used for
deep learning architectures, including TensorFlow, Keras,
and Caffe. TensorFlow is considered to be the most reliable
framework for detecting and classifying plant leaf diseases in
most research studies.
VI. FUTURE RESEARCH PROSPECT AND POSSIBLE
SOLUTION FOR LIMITATIONS
Processing leaf images in plants is rapidly spreading through-
out the industry, as replicating human visual talents is a
critical first step in the automation of operations. The devel-
opment of a computer vision system for disease diagnosis.
The subsequent next-generation characteristics can be taken
into consideration for additional research according to our
major results from the prior investigations [62].
a) In the future, the current algorithms can be used in natural
environments and integrated with leaf fronts as well as leaf
backs into a single dataset.
b) The automatic assessment of the severity of the identified
problems may also be the focus of future research.
c) By creating sophisticated algorithms, current research can
also be expanded to attain greater speed as well as accuracy.
In the present article, numerous ways of recognizing, and
predicting leaf diseases utilizing image and classification
methods are developed and put into practice. The paper
provides an overview of the technical terms used in the
current approaches that are relevant to the research’s goal.
The classification and detection of leaf diseases is a field
of study that goes beyond the previously discussed potential
future applications. The goal of this work was to list and
explain some of the major obstacles that still need to be
removed before an image-based diagnosis system that is
actually effective is made available [63].
Placing restrictions to restrict the fluctuations in capture
conditions could be one method to get around some of the
constraints that still exist for this kind of technology. The
additional work needed to meet those limits may discourage
many potential users from utilizing the technology, which is
an unfavorable side effect of the method.
The application of more advanced methods taken from the
fields of computer vision and machine learning may be
able to reduce some of the major problems such as graph
theory, mean shift, and other problems that have not yet been
examined.
A lot of issues still need to be resolved in the extremely
difficult scientific field of computer-assisted plant disease de-
tection [32]. There is no doubt that technology will advance
to produce more advanced instruments, but given the com-
plexity of the situation, it is unlikely that plant pathologists
or other plant science experts will be substituted.
VII. CONCLUSION
This study explores distinguishing and demonstrating the
prevalent literature on machine learning for leaf disease de-
tection and classification. We conducted a methodical map-
ping survey to evaluate six research questions. In this study,
we sorted out 256 research studies from seven databases.
A notable advantage of the study is the development and
implementation of a system for classifying and detecting
leaf diseases in an early stages. Classifying and detecting
leaf diseases at early stages will encourage farmers to take
the necessary precautions to reduce yield loss. Our study
reveals that CNN is extensively used in various studies.
Machine learning methods identify problems and difficulties
in plant disease diagnosis and classification. For instance,
data representation, labeling, and collection are the main
challenges facing machine learning development. Overfitting
is one of the existing problems in machine learning and deep
learning. Our study shows considerable progress in using
deep learning architectures for plant disease classification
and detection. Traditional machine-learning algorithms and
the difficulties of capturing image data sets of leaves in the
field affect the models’ performance. Diseases are developing
and rising based on field observation. However, optimization
and customization processes still need to address many prob-
lems and gaps. Lastly, this study considers the background
for more advanced plant diseases detection and classification
research. Farmers can increase productivity by improving
crop disease diagnosis and prediction and avoiding economic
damage.
8VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
Article Year Plant No: of image Dataset Framework Algorithm Architecture Best accuracy
[25] 2006 Citrus 160 Self Matlab Color texture features CCM 95.00%
[29] 2007 Cotton 150 Self N/A FC/FS MSE ———
[30] 2008 Grape 410 Self N/A SVM N/A 86.03%
[31] 2009 Rice 216 Self Visual C++ SVM N/A 97.20%
[64] 2010 Cucumber 336 Self N/A SVM EBF Kernel 91.70%
[65] 2010 Soybean 32 Self N/A RIA/RCI N/A 87.00%
[66] 2011 Cotton 40 Self Matlab SVM and ANN SVM 91.00%
[67] 2011 Multiple 192 Self Matlab NN/K-means N/A 94.00%
[68] 2012 Cotton N/A Self N/A BP NN N/A 89.50%
[69] 2012 Wheat 114 Self N/A MLR and PLSR N/A 90.00%
[70] 2013 Multiple 192 Self Matlab Otus’s method N/A 94.00%
[71] 2013 Multiple 2616 Plant pathology Matlab DWT and PCA MDC/PNN 86.48%
[72] 2014 Multiple 600 Plant Pathology Matlab ANN/Kn-Based c-texture/c-features 94.30%
[72] 2014 Multiple 990 Self Matlab BPNN N-KNN/ANN 91.54%
[73] 2015 Multiple N/A Plant pathology Matlab ANN/N-neighbor GLCM and GLRM 94.85%
[44] 2016 Cucumber 7,520 Self Caffe CNN VGG 82.30%
[45] 2017 Cassava 11,670 Self TensorFlow CNN InceptionV3 93.00%
[8] 2017 Banana 3,700 Self Matlab CNN LeNet 99.72%
[33] 2017 Apple 2,086 Plant village Theano CNN VGG16 90.40%
[74] 2018 Soybeans 25,000 Self PyTorch DCNN PLNet and LeNet 94.13%
[46] 2018 Corn 10,784 Self Caffe CNN Faster R-CNN 95.00%
[75] 2018 Cucumber 14,208 Self Matlab DCNN AlexNet 93.40%
[47] 2018 Wheat 8,178 Self TensorFlow CNN ResNet 96.00%
[48] 2019 Tomato 6,888 Plant village Matlab CNN ResNet-101 98.80%
[49] 2019 Apple and tomato 3663 Plant village TensorFlow CNN ImageNet 87.00%
[76] 2019 Blueberry 800 Self Python Deep learning CNN 84.00%
[11] 2019 Cherry 1,200 Self Python CNN GoogLeNet 99.60%
[12] 2019 Apple 640 Self TensorFlow DNN YOLOV3/DenseNet 95.75%
[50] 2019 Corn 15,240 Self TensorFlow CNN ResNet/crowdsourced 99.79%
[51] 2019 Olive 54,306 Plant village TensorFlow CNN AlexNet 99.11%
[52] 2020 Corn 679 Self TensorFlow/ Keras CNN Deep CNN 88.46%
[77] 2020 Tomato 14,072 Self Caffe/DarkNet53 MobileNetV2 "MobileNetV2/YOL3" 86.98%
[39] 2020 Apple 2,979 Self Keras D-CNN XDNet 98.82%
[53] 2020 Maize 8,152 NLB Caffe CNN ResNet-101 91.83%
[54] 2021 Peanut 6,029 Self PyTorch CNN ResNet50 97.59%
[55] 2021 Cassava 10.839 Kaggle TensorFlow/ Keras CNN EfficientNet 87.00%
[56] 2021 Pomelo 540 Self Caffe CNN GoogLeNet 82.70%
[57] 2021 Sweet cherry 11,676 Self Yolov5 CNN YOLOV5 86.10%
[58] 2022 Cardamom 1,724 Self PyTorch CNN EfficientNetB2 98.26%
[59] 2022 Grape 9,027 Plant village TensorFlow/ Keras CNN EfficientNetB7 98.70%
TABLE 9. Summary of the study characteristics
VOLUME 4, 2016 9
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
REFERENCES
[1] Prakhar Bansal, Rahul Kumar, and Somesh Kumar. Disease detection
in apple leaves using deep convolutional neural network. Agriculture,
11(7):617, 2021.
[2] Yuan Yuan, Sisi Fang, and Lei Chen. Crop disease image classification
based on transfer learning with dcnns. In Pattern Recognition and Com-
puter Vision: First Chinese Conference, PRCV 2018, Guangzhou, China,
November 23-26, 2018, Proceedings, Part II 1, pages 457–468, 2018.
[3] RS Latha, GR Sreekanth, RC Suganthe, R Rajadevi, S Karthikeyan,
S Kanivel, and B Inbaraj. Automatic detection of tea leaf diseases using
deep convolution neural network. In 2021 International Conference on
Computer Communication and Informatics (ICCCI), pages 1–6, 2021.
[4] Srdjan Sladojevic, Marko Arsenovic, Andras Anderla, Dubravko Culibrk,
and Darko Stefanovic. Deep neural networks based recognition of plant
diseases by leaf image classification. Computational intelligence and
neuroscience, 2016, 2016.
[5] P Goncharov, G Ososkov, A Nechaevskiy, A Uzhinskiy, and I Nestsiarenia.
Disease detection on the plant leaves by deep learning. In International
Conference on Neuroinformatics, pages 151–159, 2018.
[6] Shima Ramesh, Ramachandra Hebbar, M Niveditha, R Pooja, N Shashank,
PV Vinod, et al. Plant disease detection using machine learning. In
2018 International conference on design innovations for 3Cs compute
communicate control (ICDI3C), pages 41–45, 2018.
[7] V Malathi and MP Gopinath. Classification of pest detection in paddy
crop based on transfer learning approach. Acta Agriculturae Scandinavica,
Section B—Soil & Plant Science, 71(7):552–559, 2021.
[8] Jihen Amara, Bassem Bouaziz, and Alsayed Algergawy. A deep learning-
based approach for banana leaf diseases classification. Datenbanksysteme
für Business, Technologie und Web (BTW 2017)-Workshopband, 2017.
[9] Sk Mahmudul Hassan, Arnab Kumar Maji, Michał Jasi´
nski, Zbigniew
Leonowicz, and El˙
zbieta Jasi´
nska. Identification of plant-leaf diseases
using cnn and transfer-learning approach. Electronics, 10(12):1388, 2021.
[10] Keke Zhang, Qiufeng Wu, Anwang Liu, Xiangyan Meng, et al. Can deep
learning identify tomato leaf disease? Advances in multimedia, 2018,
2018.
[11] Keke Zhang, Lei Zhang, and Qiufeng Wu. Identification of cherry leaf
disease infected by podosphaera pannosa via convolutional neural net-
work. International Journal of Agricultural and Environmental Information
Systems (IJAEIS), 10(2):98–110, 2019.
[12] Yunong Tian, Guodong Yang, Zhe Wang, En Li, and Zize Liang. Detection
of apple lesions in orchards based on deep learning methods of cyclegan
and yolov3-dense. Journal of Sensors, 2019, 2019.
[13] Halil Durmu¸s, Ece Olcay Güne¸s, and Mürvet Kırcı. Disease detection
on the leaves of the tomato plants by using deep learning. In 2017 6th
International Conference on Agro-geoinformatics, pages 1–5, 2017.
[14] Andre Abade, Paulo Afonso Ferreira, and Flavio de Barros Vidal. Plant
diseases recognition on images using convolutional neural networks: A
systematic review. Computers and Electronics in Agriculture, 185:106125,
2021.
[15] Muhammad Hammad Saleem, Johan Potgieter, and Khalid Mahmood
Arif. Plant disease detection and classification by deep learning. Plants,
8(11):468, 2019.
[16] Waleed Albattah, Marriam Nawaz, Ali Javed, Momina Masood, and Saleh
Albahli. A novel deep learning method for detection and classification of
plant diseases. Complex & Intelligent Systems, pages 1–18, 2022.
[17] Habibollah Agh Atabay. Deep residual learning for tomato plant leaf
disease identification. Journal of Theoretical & Applied Information
Technology, 95(24), 2017.
[18] Mohammed Brahimi, Kamel Boukhalfa, and Abdelouahab Moussaoui.
Deep learning for tomato diseases: classification and symptoms visualiza-
tion. Applied Artificial Intelligence, 31(4):299–315, 2017.
[19] Stefania Barburiceanu, Serban Meza, Bogdan Orza, Raul Malutan, and
Romulus Terebes. Convolutional neural networks for texture feature ex-
traction. applications to leaf disease classification in precision agriculture.
IEEE Access, 9:160085–160103, 2021.
[20] Asifullah Khan, Anabia Sohail, Umme Zahoora, and Aqsa Saeed Qureshi.
A survey of the recent architectures of deep convolutional neural networks.
Artificial intelligence review, 53(8):5455–5516, 2020.
[21] Zulkifli Bin Husin, Ali Yeon Bin Md Shakaff, Abdul Hallis Bin Abdul
Aziz, and Rohani Binti S Mohamed Farook. Feasibility study on plant
chili disease detection using image processing techniques. In 2012 Third
International Conference on Intelligent Systems Modelling and Simula-
tion, pages 291–296, 2012.
[22] Vinicius Bischoff, Kleinner Farias, Juliano Paulo Menzen, and Gustavo
Pessin. Technological support for detection and prediction of plant
diseases: A systematic mapping study. Computers and Electronics in
Agriculture, 181:105922, 2021.
[23] Oliver Oswin Sumady, Brian Joe Antoni, Randy Nasuta, Edy Irwansyah,
et al. A review of optical text recognition from distorted scene image. In
2022 4th International Conference on Cybernetics and Intelligent System
(ICORIS), pages 1–5, 2022.
[24] Elizabeth Ann Soelistio, Rafael Edwin Hananto Kusumo, Zevira Varies
Martan, and Edy Irwansyah. A review of signature recognition using
machine learning. In 2021 1st International Conference on Computer
Science and Artificial Intelligence (ICCSAI), volume 1, pages 219–223,
2021.
[25] Arifur R. Anik, Kamrul Hasan, Md. Manirul Islam, Md. Mehedi Hasan,
Md. Firoj Ali, and Sajal K. Das. Non-invasive portable technologies
for monitoring breast cancer related lymphedema to facilitate telehealth:
A scoping review. IEEE Journal of Biomedical and Health Informatics,
27(9):4524–4535, 2023.
[26] Md Zahid Hasan, Md Sazzadur Ahamed, Aniruddha Rakshit, and
KM Zubair Hasan. Recognition of jute diseases by leaf image classi-
fication using convolutional neural network. In 2019 10th International
Conference on Computing, Communication and Networking Technologies
(ICCCNT), pages 1–5, 2019.
[27] Ovidiu-Constantin Novac, Mihai Cristian Chirodea, Cornelia Mihaela No-
vac, Nicu Bizon, Mihai Oproescu, Ovidiu Petru Stan, and Cornelia Emilia
Gordan. Analysis of the application efficiency of tensorflow and pytorch
in convolutional neural network. Sensors, 22(22):8872, 2022.
[28] R Pydipati, TF Burks, and WS Lee. Identification of citrus disease
using color texture features and discriminant analysis. Computers and
electronics in agriculture, 52(1-2):49–59, 2006.
[29] Yan-Cheng Zhang, Han-Ping Mao, Bo Hu, and Ming-Xi Li. Features
selection of cotton disease leaves image based on fuzzy feature selection
techniques. In 2007 International Conference on Wavelet Analysis and
Pattern Recognition, volume 1, pages 124–129, 2007.
[30] A Meunkaewjinda, P Kumsawat, K Attakitmongcol, and A Srikaew. Grape
leaf disease detection from color imagery using hybrid intelligent system.
In 2008 5th international conference on electrical engineering/electronics,
computer, telecommunications and information technology, volume 1,
pages 513–516, 2008.
[31] Qing Yao, Zexin Guan, Yingfeng Zhou, Jian Tang, Yang Hu, and Baojun
Yang. Application of support vector machine for detecting rice diseases
using shape and color texture features. In 2009 international conference
on engineering computation, pages 79–83, 2009.
[32] Alvaro Fuentes, Sook Yoon, Sang Cheol Kim, and Dong Sun Park. A
robust deep-learning-based detector for real-time tomato plant diseases
and pests recognition. Sensors, 17(9):2022, 2017.
[33] Guan Wang, Yu Sun, Jianxin Wang, et al. Automatic image-based
plant disease severity estimation using deep learning. Computational
intelligence and neuroscience, 2017, 2017.
[34] Bin Liu, Yun Zhang, DongJian He, and Yuxiang Li. Identification of apple
leaf diseases based on deep convolutional neural networks. Symmetry,
10(1):11, 2017.
[35] Keunho Park, Young ki Hong, Gook Hwan Kim, and Joonwhoan Lee.
Classification of apple leaf conditions in hyper-spectral images for diagno-
sis of marssonina blotch using mrmr and deep neural network. Computers
and Electronics in Agriculture, 148:179–187, 2018.
[36] Mohit Agarwal, Rohit Kumar Kaliyar, Gaurav Singal, and Suneet Kr
Gupta. Fcnn-lda: a faster convolution neural network model for leaf
disease identification on apple’s leaf dataset. In 2019 12th International
Conference on Information & Communication Technology and System
(ICTS), pages 246–251, 2019.
[37] Hee-Jin Yu and Chang-Hwan Son. Apple leaf disease identification
through region-of-interest-aware deep convolutional neural network. arXiv
preprint arXiv:1903.10356, 2019.
[38] Clément Douarre, Carlos F Crispim-Junior, Anthony Gelibert, Laure
Tougne, and David Rousseau. Novel data augmentation strategies to boost
supervised segmentation of plant disease. Computers and electronics in
agriculture, 165:104967, 2019.
[39] Xiaofei Chao, Guoying Sun, Hongke Zhao, Min Li, and Dongjian He.
Identification of apple tree leaf diseases based on deep learning models.
Symmetry, 12(7):1065, 2020.
[40] Changjian Zhou and Jinge Xing. Improved deep residual network for apple
leaf disease identification. Journal of Information Processing Systems,
17(6), 2021.
10 VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
[41] Subham Divakar, Abhishek Bhattacharjee, and Rojalina Priyadarshini.
Smote-dl: a deep learning based plant disease detection method. In
2021 6th International Conference for Convergence in Technology (I2CT),
pages 1–6, 2021.
[42] Swati Singh, Isha Gupta, Sheifali Gupta, Deepika Koundal, Sultan Al-
jahdali, Shubham Mahajan, and Amit Kant Pandit. Deep learning-based
automated detection of diseases from apple leaf images. Computers,
Materials & Continua, 71(1), 2022.
[43] Muhammad Sharif, Muhammad Attique Khan, Zahid Iqbal, Muham-
mad Faisal Azam, M Ikram Ullah Lali, and Muhammad Younus Javed.
Detection and classification of citrus diseases in agriculture based on
optimized weighted segmentation and feature selection. Computers and
electronics in agriculture, 150:220–234, 2018.
[44] Erika Fujita, Yusuke Kawasaki, Hiroyuki Uga, Satoshi Kagiwada, and
Hitoshi Iyatomi. Basic investigation on a robust and practical plant
diagnostic system. In 2016 15th IEEE international conference on machine
learning and applications (ICMLA), pages 989–992, 2016.
[45] Amanda Ramcharan, Kelsee Baranowski, Peter McCloskey, Babuali
Ahmed, James Legg, and David P Hughes. Deep learning for image-based
cassava disease detection. Frontiers in plant science, 8:1852, 2017.
[46] Shichao Jin, Yanjun Su, Shang Gao, Fangfang Wu, Tianyu Hu, Jin Liu,
Wenkai Li, Dingchang Wang, Shaojiang Chen, Yuanxi Jiang, et al. Deep
learning: individual maize segmentation from terrestrial lidar data using
faster r-cnn and regional growth algorithms. Frontiers in plant science,
9:866, 2018.
[47] Artzai Picon, Maximiliam Seitz, Aitor Alvarez-Gila, Patrick Mohnke,
Amaia Ortiz-Barredo, and Jone Echazarra. Crop conditional convolutional
neural networks for massive multi-crop plant disease classification over
cell phone acquired images taken on real field conditions. Computers and
Electronics in Agriculture, 167:105093, 2019.
[48] Manpreet Kaur and Rekha Bhatia. Development of an improved tomato
leaf disease detection and classification method. In 2019 IEEE Conference
on Information and Communication Technology, pages 1–5, 2019.
[49] Mercelin Francis and C Deisy. Disease detection and classification in
agricultural plants using convolutional neural networks—a visual under-
standing. In 2019 6th International Conference on Signal Processing and
Integrated Networks (SPIN), pages 1063–1068, 2019.
[50] Tyr Wiesner-Hanks, Harvey Wu, Ethan Stewart, Chad DeChant, Nicholas
Kaczmar, Hod Lipson, Michael A Gore, and Rebecca J Nelson.
Millimeter-level plant disease detection from aerial photographs via deep
learning and crowdsourced data. Frontiers in Plant Science, 10:1550, 2019.
[51] Madallah Alruwaili, Saad Alanazi, Sameh Abd El-Ghany, and Abdulaziz
Shehab. An efficient deep learning model for olive diseases detection.
International Journal of Advanced Computer Science and Applications,
10(8), 2019.
[52] Sumita Mishra, Rishabh Sachan, and Diksha Rajpal. Deep convolutional
neural network based detection system for real-time corn plant disease
recognition. Procedia Computer Science, 167:2003–2010, 2020.
[53] Jun Sun, Yu Yang, Xiaofei He, and Xiaohong Wu. Northern maize leaf
blight detection under complex field environment based on deep learning.
IEEE Access, 8:33679–33688, 2020.
[54] Haixia Qi, Yu Liang, Quanchen Ding, and Jun Zou. Automatic identifica-
tion of peanut-leaf diseases based on stack ensemble. Applied Sciences,
11(4):1950, 2021.
[55] Vinayakumar Ravi, Vasundhara Acharya, and Tuan D Pham. Attention
deep learning-based large-scale learning classifier for cassava leaf disease
classification. Expert Systems, 39(2):e12862, 2022.
[56] Sirirat Laosim and Taweesak Samanchuen. Classification of pomelo leaf
diseases using convolution neural network. In 2021 18th International
Conference on Electrical Engineering/Electronics, Computer, Telecom-
munications and Information Technology (ECTI-CON), pages 577–580,
2021.
[57] Christos Chaschatzis, Chrysoula Karaiskou, Efstathios G Mouratidis,
Evangelos Karagiannis, and Panagiotis G Sarigiannidis. Detection and
characterization of stressed sweet cherry tissues using machine learning.
Drones, 6(1):3, 2021.
[58] CK Sunil, CD Jaidhar, and Nagamma Patil. Cardamom plant disease
detection approach using efficientnetv2. IEEE Access, 10:789–804, 2021.
[59] Prabhjot Kaur, Shilpi Harnal, Rajeev Tiwari, Shuchi Upadhyay, Surbhi
Bhatia, Arwa Mashat, and Aliaa M Alabdali. Recognition of leaf disease
using hybrid convolutional neural network by applying feature reduction.
Sensors, 22(2):575, 2022.
[60] Qimei Wang, Feng Qi, Minghe Sun, Jianhua Qu, Jie Xue, et al. Iden-
tification of tomato disease types and detection of infected areas based
on deep convolutional neural networks and object detection techniques.
Computational intelligence and neuroscience, 2019, 2019.
[61] Gizem Irmak and Ahmet Saygili. Tomato leaf disease detection and
classification using convolutional neural networks. In 2020 Innovations
in Intelligent Systems and Applications Conference (ASYU), pages 1–5,
2020.
[62] Gittaly Dhingra, Vinay Kumar, and Hem Dutt Joshi. Study of digital
image processing techniques for leaf disease detection and classification.
Multimedia Tools and Applications, 77:19951–20000, 2018.
[63] Jayme Garcia Arnal Barbedo. A review on the main challenges in
automatic plant disease identification based on visible range images.
Biosystems engineering, 144:52–60, 2016.
[64] Zhang Jian and Zhang Wei. Support vector machine for recognition of
cucumber leaf diseases. In 2010 2nd international conference on advanced
computer control, volume 5, pages 264–266, 2010.
[65] Di Cui, Qin Zhang, Minzan Li, Glen L Hartman, and Youfu Zhao.
Image processing methods for quantitatively detecting soybean rust from
multispectral images. Biosystems engineering, 107(3):186–193, 2010.
[66] Viraj A Gulhane and Ajay A Gurjar. Detection of diseases on cotton leaves
and its possible diagnosis. International Journal of Image Processing
(IJIP), 5(5):590–598, 2011.
[67] Sabine D Bauer, Filip Korˇ
c, and Wolfgang Förstner. The potential of auto-
matic methods of classification to identify leaf diseases from multispectral
images. Precision Agriculture, 12(3):361–377, 2011.
[68] P Revathi and M Hemalatha. Advance computing enrichment evaluation
of cotton leaf spot disease detection using image edge detection. In
2012 Third International Conference on Computing, Communication and
Networking Technologies (ICCCNT’12), pages 1–5, 2012.
[69] Jing-Cheng Zhang, Rui-liang Pu, Ji-hua Wang, Wen-jiang Huang, Lin
Yuan, and Ju-hua Luo. Detecting powdery mildew of winter wheat using
leaf level hyperspectral measurements. Computers and Electronics in
Agriculture, 85:13–23, 2012.
[70] Smita Naikwadi and Niket Amoda. Advances in image processing
for detection of plant diseases. International Journal of Application or
Innovation in Engineering & Management, 2(11), 2013.
[71] Jagadeesh D Pujari, Rajesh Yakkundimath, and Abdulmunaf Syedhusain
Byadgi. Automatic fungal disease detection based on wavelet feature
extraction and pca analysis in commercial crops. International Journal
of Image, Graphics and Signal Processing, 6(1):24–31, 2013.
[72] Jagadeesh D Pujari, Rajesh Yakkundimath, and Abdulmunaf S Byadgi.
Recognition and classification of produce affected by identically looking
powdery mildew disease. Acta Technologica Agriculturae, 17(2):29–34,
2014.
[73] Jagadeesh D Pujari, Rajesh Yakkundimath, and Abdulmunaf S Byadgi.
Image processing based detection of fungal diseases in plants. Procedia
Computer Science, 46:1802–1808, 2015.
[74] Sambuddha Ghosal, David Blystone, Asheesh K Singh, Baskar Ganap-
athysubramanian, Arti Singh, and Soumik Sarkar. An explainable deep
machine vision framework for plant stress phenotyping. Proceedings of
the National Academy of Sciences, 115(18):4613–4618, 2018.
[75] Juncheng Ma, Keming Du, Feixiang Zheng, Lingxian Zhang, Zhihong
Gong, and Zhongfu Sun. A recognition method for cucumber diseases
using leaf symptom images based on deep convolutional neural network.
Computers and electronics in agriculture, 154:18–24, 2018.
[76] Cecilia Sullca, Carlos Molina, Carlos Rodríguez, and Thais Fernández.
Diseases detection in blueberry leaves using computer vision and machine
learning techniques. International Journal of Machine Learning and
Computing, 9(5):656–661, 2019.
[77] Jun Liu and Xuewei Wang. Early recognition of tomato gray leaf spot
disease based on mobilenetv2-yolov3 model. Plant Methods, 16(1):1–16,
2020.
VOLUME 4, 2016 11
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/

Doutoum and Tugrul: A Review of Leaf Diseases Detection and Classification by Deep Learning
ASSAD SOULEYMAN DOUTOUM obtained
his B.S. degree in 2005 from the University of
Roi Faycal in Computer Engineering. In 2014 he
received an M.Sc. degree in Information Systems
Concentration in Computer Security Management
from Strayer University. His research interests
include Big Data Analytics, Computer Security
Management, and Machine Learning.
BULENT TUGRUL received his Ph.D. in Com-
puter Engineering from Anadolu University,
Turkey, in 2014. His Ph.D. thesis was on secure
spatial interpolation methods. He is currently an
assistant professor at Ankara University, Ankara,
Turkey. His current research interests include in-
formation security, geostatistics, data mining, and
big data.
12 VOLUME 4, 2016
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3326721
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/