Urban Agric Regnl Food Sys - 2024 - Bagga - Image‐based detection and classification of plant diseases using deep

Received: 11 August 2023 Accepted: 16 January 2024
DOI: 10.1002/uar2.20053
REVIEW
Special Section: Im proving Livability in Urban Areas: Exam ining Urban and Peri-Urban Soil and Plant
Managem ent
Image-based detection and classification of plant diseases using

deep learning: State-of-the-art review
Manju Bagga1 Sonali Goyal2
1 MMEC, MMICTBM, Maharishi

Markandeshwar (Deemed to be University), Abstract
Mullana, Ambala 133207, Haryana, India Plant diseases are assumed to be one of the primary causes regulating food man-
2 Department of Computer Science & ufacturing and reducing deficits in crop yield, and it is crucial that plant diseases
Engineering, MMEC, Maharishi
Markandeshwar (Deemed To be University),
have rapid spotting and identification. Since the demand for food is rising quickly
Mullana, Ambala 133207, Haryana, India due to population growth, the main issue for any country is plant disease mechaniza-
tion in agricultural science. Recent developments in deep learning techniques have
Correspondence
Manju Bagga, MMICTBM, Maharishi found use in plant disease recognition, offering a robust tool with exceedingly precise
Markandeshwar (Deemed to be University), results. Our main goal is to employ deep learning mechanisms to better understand
Mullana, Ambala, Haryana 133207, India.
diseases and how to identify them quickly. In this regard, we analyzed 94 publica-
Email: [email protected]
tions chosen from the last 7 years (2016–2023) that show how CNN’s philosophy
Assigned to Associate Editor Calogero has evolved over this period with various approaches to treating plant diseases. Fur-
Schillaci.
thermore, a full description of numerous crops, diseases connected to them, various
datasets relating to plant diseases, existing CNN models, and customized CNN archi-
tectures is provided. The results of this state-of-the-art review can be implemented
to comprehend the cutting-edge trends in the application of deep learning (CNNs) to
detect plant diseases as well as pinpoint any research gaps that require the scientific
community’s attention.
1 INTRODUCTION monitoring in addition to prompt and accurate disease diagno-

sis. Phytopathologists advise against these practices the most.
The existence of plant diseases is one of the main factors Accurately recognizing the signs of the main diseases that
influencing food production, which can greatly reduce crops’ harm crops is the biggest issue for agriculture (Hernández
natural or economic profitability and, in certain cases, even & López, 2020). Traditional planting methods that are man-
obstruct this activity. Altieri (2018) argues that proper disease ual and mechanized cannot cover a large area of plantings or
management and control strategies must be implemented in give critical advance data to assist in decision-making proce-
order to reduce output losses and maintain agricultural sus- dures. By using COMPUTER VISION technology to monitor
tainability. He emphasizes the importance of ongoing crop crop growth in real-time rather than manually, it is possible
to detect small changes in crops caused by malnutrition far
Abbreviations: CNN, convolutional neural network; DCNN, deep CNN; sooner than human monitoring and to use this information as
DL, deep learning; ML, machine learning; NN, neural network; RGB, red
a reliable and accurate basis for appropriate management (Das
green blue; ROI, region of interest; SGD, stochastic gradient descent;
TCCNN, three-channel CNN; YOLO, you only look once. et al., 2017).
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original
work is properly cited.
© 2024 The Authors. Urban Agriculture & Regional Food Systems published by Wiley Periodicals LLC on behalf of American Society of Agronomy and Crop Science Society of
America.
Urban Agric Region Food Syst. 2024;9:e20053. wileyonlinelibrary.com/journal/uar2 1 of 21

https://doi.org/10.1002/uar2.20053
25751220, 2024, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/uar2.20053 by Cochrane Colombia, Wiley Online Library on [27/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 of 21 BAGGA AND GOYAL
As a result, practical, dependable, and cost-effective auto-

mated solutions must be created in order to track the health of Core Idea
plants and provide useful data for decision-making (Mahlein,
2016). With major advancements, COMPUTER VISION and ∙ Crop yields can be significantly impacted by plant
artificial intelligence have been creating strategies and meth- diseases, which can result in a smaller harvest and
ods for identifying and categorizing things. In many different financial loss.
fields, deep neural networks (NNs) have lately been success- ∙ Since the demand for food is rising quickly due to
fully used as examples of end-to-end learning. A sequence of population growth, the main issue for any country
stacked layers of nodes in deep NNs simply maps the input is plant diseases.
layer to the output layer. ∙ Recent developments in deep learning techniques
Many convolutional deep NN approaches for detecting have found use in plant disease recognition.
crop diseases already use well-known architectural frame- ∙ The main goal is to employ deep learning mecha-
works such as LeNet (LeCun et al., 1998), AlexNet, VGGNet, nisms to better understand the diseases and how to
GoogLeNet, InceptionV3, ResNet (Türkoğlu & Hanbay, identify them quickly.
2019), and DenseNet (Huang et al., 2017), considerably
enhancing the ability to recognize and identify plant diseases
with a great deal of accuracy. The research by Tian et al.
(2019) emphasizes the use of image data as one of the most microscopy and is neither prompt nor precise. The main issue
cutting-edge techniques to identify and detect plant diseases. with visual assessment is caused by psychological and cogni-
Deep learning has significantly improved this process; how- tive phenomena that might result in mistakes, prejudice, and
ever, there are still several obstacles that make it difficult to optical illusions (Barbedo, 2016).
recognize and classify plant illnesses. The difficult part of
building a deep network is making sure that the network’s 2.1 Various plant-specific diseases
design and structure, functions (nodes), and edges accurately
map the input to the output. Various diseases of plants are categorized based on the kind
When used in actual farming situations, the genomic and of pathogen that causes them. In contrast to abiotic diseases,
phenotype variability among crops, as well as the wide range which are those caused by non-living ecological conditions,
of diseases and pests, has an adverse effect on the results. biotic diseases are those that originate from living crea-
It should also be highlighted that the characteristics of the tures (Pandey et al., 2017). Numerous biological diseases are
databases are restricted to the diagnosis and prediction of crop mostly brought on by fungal organisms, viral and bacterial
diseases brought on by biotic causes. By changing the param- epidemics, and roundworms. The biological pest in this study
eters of the network, deep NNs can be trained to produce is an outbreak of a particular species that affects agricultural
better mapping as training progresses. This computationally development and results in recurrent diseases in varieties of
difficult technique has recently seen significant improve- plants.
ments because of a number of conceptual and engineering
advancements. Using a convolutional neural network (CNN)
to describe the most recent advances for recognizing and iden- 2.2 Disease recognition techniques for
tifying diseased plants, a thorough review of the literature has plants
been conducted.
The remaining portions of the study are structured as fol- The use of images and machine learning (ML) in automated
lows: The background is discussed in Section 2. The review plant disease identification systems, particularly CNNs, has
plan is completely described in Section 3. The findings of significantly improved the precision of a valid examina-
the literature review and outcomes are demonstrated in Sec- tion. CNNs, a branch of artificial intelligence, have gained
tion 4. The discussion is presented in Section 5, and the study popularity as a flexible technique for ingesting copious vol-
is wrapped up in Section 6. umes of diverse data and producing accurate forecasts of
difficult-to-predict occurrences (Heaton et al., 2016).
2 BACKGROUND
2.3 Architectural emergence of CNN
One of the most fundamental and crucial agricultural tasks
is the identification of diseases in plants. The recognition of When threshold logic was proposed in 1943 (McCulloch &
agricultural diseases traditionally relies primarily on arbitrary Pitts, 1943) to create a computer architecture exactly mimick-
human judgment using serological and molecular testing or ing the biological ways of human beings, the deep learning
BAGGA AND GOYAL 3 of 21
(DL) approach was introduced as a subclass of ML. In 1998, 3 SYSTEMATIC PLANNING OF

the first CNN architecture, LeNet-5, was introduced. It falls REVIEW
into various categories, including regularization, structural
reformulation, and parameter optimization. It has been rec- The motivation for implementing the systematic review must
ognized, nonetheless, that the essential transformation and be determined during the planning stage. Thus, for this
creation of new blocks are the factors that are mostly responsi- review’s execution to be successful, identifying the objectives
ble for CNN’s performance increase (Shahoveisi et al., 2023). and developing a procedure are crucial. It should be empha-
As a result, according to the innovation ideas published in sized that the effectiveness of the techniques used directly
research by Panwar et al. (2022) and Heaton et al. (2016), this affects the outcomes of the review.
qualifies as an innovative architecture.
The CNN architecture at the cutting edge is set up to handle
classification, segmentation, and object detection problems. 3.1 Research issues
Convolution and subsampling layering are used to encode the
input images in the models to address classification problems, It has been thoroughly acknowledged that CNNs have
and then the classifier technique is utilized to determine the demonstrated superior performance over both conventional
likelihood of each class. To extract features for segmentation computer-aided vision approaches and other ML-based tech-
and detection tasks, it is possible to utilize such architectures niques. Thus, the goal of this review is to pinpoint the factors
with the convolutional layer (Alom et al., 2019). A pre-trained that influence how effectively CNNs are used to recognize
CNN architecture called AlexNet (Rao et al., 2021), Xception, plant diseases. The research questions for this study that we
Densely Connected CNN, Inception residual CNN, and resid- seek to address are as follows:
ual CNN (Kaya et al., 2019)—all of these CNN architectures
with transfer learning methodologies—are utilized to improve Research question 1: What main resources are used to
accuracy from the intended data source. collect data pertaining to plants?
In recent years, a number of semantic segmentation models Research question 2: What automated models with
have been put out (Luo et al., 2023) examining both conven- new architecture are used to identify and categorize
tional segmentation techniques and segmentation techniques diseases of plants?
based on cutting-edge CNN architecture for agricultural Research question 3: How can the performance of
image analysis. Recently, a number of novel models have been automatic recognition systems be enhanced by deep
presented, including SegNet (Badrinarayanan et al., 2017) and learning models?
DeepLab (Chen et al., 2017). Research question 4: What types of cultivated plants are
Since there are numerous ways to characterize plant leaf the subject of the most research employing CNNs?
diseases, the detection challenge is somewhat distinct from Research question 5: What kinds of plant diseases are
categorization and segmentation difficulties. The model’s most thoroughly researched utilizing CNN methods?
objective is to pair together target categories with the appro- Research question 6: Which CNN algorithm dominates
priate locations. The model responds to two queries: What modern methods?
exactly is the problem with classification? Where is the item
or object (regression issue) in the image? An adaptive evolu-
tionary algorithm was suggested by Alshammari et al. (2022) 3.2 Search procedure
for choosing the best parameters. A brand-new DL model
based on the inception layer and residual connection was Any kind of gray publication was not covered in this review
suggested in Hassan and Maji (2022). The most well-known because it concentrated on digital scientific databases. Five
object detection models are you only look once (YOLO) (Ter- indexed digital databases—IEEE Xplore, Web of Science,
ven & Cordova-Esparza, 2023), single shot detector (Ni et al., Scopus Elsevier, PubMed, Google Scholar, and Springer
2023), and Faster region-based CNN (Ren et al., 2015). Nature—were the exploration and selection procedures for
It is crucial to keep in mind that despite CNN’s designs the core study. The academic databases were chosen because
being effective at classifying, segmenting, and detecting they provide the most impactful number of scientific articles,
objects, they require a large amount of training data because which, through CNNs, extensively cover the topic of studies
of the numerous model characteristics and deepened lay- on plant diseases.
ers they use to express generalization (Arivazhagan & Ligi,
2018). Thus, a CNN model for its finetuning requires a
huge dataset of images (Mia et al., 2020), expanding our 3.3 Benchmark for the selection of papers
datasets (Haque et al., 2022), customizing layers (Taye, 2023),
and optimizing parameters of deep NN model (Trang et al., CNNs have generated a lot of interest in the scientific com-
2019). munity over the past years for their potential to identify and
TA B L E 1 Set benchmark to include and exclude papers specified in the literature review.
Serial Benchmark
No. Included Excluded Description
1 ✓ Reviewed articles about neural network approaches currently used
2 ✓ Papers that use CNN as a tool to identify plant diseases
3 ✓ Papers that introduce novel convolutional network models or architectures
4 ✓ Papers outlining techniques for spotting plant diseases using CNNs
5 ✓ Papers with limited text availability
6 ✓ Numerous sources publishing the same thing twice
7 ✓ Papers that do not primarily employ CNN methodology
8 ✓ Include scientific studies and papers that have been referenced in other works
Abbreviation: CNN, convolutional neural network.
recognize plant diseases. So, 7 years’ worth of publications The many data sources that the current literature has used
(from 2016 to 2023) were chosen. This led to the production to assemble its input image datasets have been reviewed and
of a significant number of articles, out of which we chose the identified throughout this phase. Figure 2 shows the two main
most pertinent ones based on the factors listed in Table 1. ways to obtain images, which are real-world data and public
Although various indexed research libraries specify their datasets.
own style for creating search characters, we initially generated By using a fine-tuned CNN from a pre-trained GoogLeNet,
the following general phrase that was then tailored for each 1200 Litchi images divided into 600 healthy and 600 dis-
search engine (“Crop disease detection using Deep Learn- eased images were acquired from smartphones to identify
ing,” “Plant disease detection using CNN Models,” and “Deep Podosphaera pannosa disease (K. Zhang, Zhang, et al.,
Learning in Agriculture”). Using the abovementioned search 2019). CNN is used to help with the diagnosis of Litchi
term, a total of 1895 studies were located, of which 94 arti- fruit disease, and over 9930 augmented images (based on
cles were selected for this state-of-the-art evaluation based 3,310 real images) of both healthy as well as unhealthy leaf
on the criteria given in Table 1 and some qualitative crite- images are collected from various rural places (Haque et al.,
ria (“Includes scientific research and studies that have been 2022). A total of 8438 images of grapes and mangoes were
referenced in other works,” “Are the key findings’ validity, gathered from various sources, including PlantVillage and
reliability, and legitimacy clearly stated?”, “Are the research’ self-acquisition, and were sorted into seven categories, six
limitations spelled out in detail and supported by evidence?”, of which were infected and one of which was healthy (Rao
and “Are the inferences plausible and in line with the stated et al., 2021). A deep CNN (DCNN) was then trained to recog-
results?”). Figure 1 depicts the procedures used to carry out nize diseases in plant leaves. A total of 1200 images are taken
the literature review and provides a summary of the findings. with a digital camera under difficult conditions, and a DCNN
technique is used to automatically identify leaf illnesses in
mango plants (Arivazhagan & Ligi, 2018). A total of 18,222
4 RESULTS maize crop images were acquired from a publicly accessible
repository on Open-Source Framework across three datasets.
This part contains the continuous and discontinuous statistical Lachgar et al. (2022) presented a group of 5571 images of
information that was pulled from a variety of sources to help olive leaves that were laboriously gathered under actual field
with formulating the answers to the six research questions occurrences, and the collected information was then used for
mentioned previously. analysis (DeChant et al., 2017). To test and train the CNN,
an image dataset made up of 3400 olive leaves was collected
using mobile devices (Uğuz & Uysal, 2021). The dataset also
4.1 Discussion on research question 1: includes healthy leaves. DCNN is employed for additional
What main resources are used to collect data analysis of 14,208 symptom images that were taken in the
pertaining to plants? field to help in cucumber leaf disease detection (Ma et al.,
2018). A total of 8927 images were taken in the real field using
By separating studies based on the sources used for data col- diverse camera equipment, and a DCNN framework is sug-
lection, several publications were found for this section. To gested for effective real-time disease and pest recognition on
specify every step of an object, including the training step and tomato plants (Fuentes et al., 2018). Table 2 represents some
the testing step, a suitable and sufficient dataset is required. of the additional literature that was taken into account.
FIGURE 1 Flowchart of selection of studies.
based on three-channel CNN (TCCNN). Instead of using

time-consuming pre-processing, the method allows to auto-
matically extract highly discriminating features by TCCNN
directly from the color image of an infected leaf (S. Zhang,
Huang, et al., 2019). An architecture of MCNN is suggested
for the detection of anthracnose disease, which is one of the
fungal illnesses that may be detected using one of the DL
approaches known as CNN, which has gained popularity in
terms of computation and accuracy (U. P. Singh et al., 2019).
Images cannot be managed by traditional image processing.
CNN architectures are optimized from two perspectives: the
model structure and the training label optimization. Con-
FIGURE 2 Various sources of image data. volution and pooling layers are enhanced for ResNet-50,
InceptionV3, and MobileNetV2 for the detection of plant dis-
eases (Panchal et al., 2023). In contrast to recently created
4.2 Discussion on research question 2: state-of-the-art methods, the suggested DL model achieves
What automated CNN models are used to high accuracy by utilizing an effective customized transfer
identify and categorize diseases of plants? learning mechanism and intelligent data enhancement using
a reasonable quantity of images (Dawei et al., 2019). The
Deep learning is a class of algorithms and topologies, not GoogLeNetwork Inception framework and rainbow concate-
a single strategy, that can be utilized to address a variety nation are introduced in a novel model that is based on
of problems. Following the distribution of these algorithms enhanced CNN for the real-time identification of crop dis-
classified in accordance with these condensed studies is con- eases. A responsive solution in order to detect diseases of
ceivable in Figure 3. Publications were selected for this part leaves at an early stage is provided by the suggested method,
which uses conventional CNN algorithms to identify and which was developed using a substantial collection of 26,377
categorize plant diseases. images (P. Jiang, Chen, et al., 2019). The region of interest
The most effective way to use red green blue (RGB) data (ROI)-based DCNN that has been presented is based on the
to identify vegetable diseases was determined through tests idea that it is helpful to identify diseases in the area where
TA B L E 2 Dataset information of different plants.
References Crop name Data sources Number of images Image type

Rao et al. (2021) Mango PlantVillage + Field Acquired 8438 RGB
Hernández and López (2020) Multiple PlantVillage 54,306 RGB
Türkoğlu and Hanbay (2019) Multiple Field acquired 1965 RGB
Zeng et al. (2022) Multiple PlantVillage 86,198 RGB
DeChant et al. (2017) Maize Field acquired 1796 RGB
Ma et al. (2018) Cucumber Field acquired 14,208 RGB
Fuentes et al. (2018) Tomato Field acquired 8927 RGB
Johannes et al. (2017) Wheat Field acquired 3500 RGB
Qin et al. (2016) Alfalfa Lab images 899 RGB
Dey et al. (2016) Betel Scanned images 20 RGB
Singh and Misra (2017) Multiple Lab and field acquired 106 RGB
Ramesh and Vydeki (2020) Rice Field images 650 RGB
V. Singh (2019) Sunflower Not mentioned Not mentioned RGB
Jin et al. (2018) Maize Field acquired 10,784 Hyperspectral
Ahmad et al. (2020) Plum Field acquired Not mentioned RGB
Nagasubramanian et al. (2019) Soybean Hyperspectral image dataset 111 Hyperspectral
Wiesner et al. (2018) Maize Cornell University dataset 18,222 RGB
Abbreviation: RGB, red green blue.
F I G U R E 3 The specific quantity of research publications that use convolutional neural network (CNN) models. ANN, artificial neural
network; BPNN, back propagation neural network; CAE, convolutional autoencoder; ROI-CNN, region of interest-convolutional neural network;
YOLO, you only look once.
symptoms of leaf disease first occur (Yu & Son, 2019). An tions, the system has demonstrated its performance (Yadav
ideal DL model has demonstrated astounding efficiency in et al., 2021), and Hang et al. (2019) also focused on the ben-
the identification of diseases. An automated image recogni- efits of using an NN to identify diseased components and
tion system that surpassed ML techniques used an adaptive classify the target locations of the disease. In order to catego-
genetic algorithm to choose the best parameters (Alshammari rize plant diseases, the L2MXception network, an upgraded
et al., 2022). For image segmentation and instinctive recog- version of Xception, is suggested. The architecture used is a
nition of blights (bacteria spot) in plants, gray-level slicing simple and quick method for early detection by using differ-
is used to create an integrated system employing CNN and ent CNN models (Yao et al., 2021). The diseases in grape
an image technique. Even in normal communication situa- and mango fruit were proposed to be examined using a
web-based model called JIT CROPFIX. In order to automat- values of the pixels were also normalized (Haque et al., 2022).
ically extract features and classify them in order to identify The dataset was produced by taking pictures with a digital
infections, if any, a pre-tuned CNN-based architecture called camera in difficult lighting, size, stance, and orientation sit-
AlexNet is designed (Rao et al., 2021). Different CNN mod- uations. The photos are downsized from 4608 × 3456 to 256
els with transfer learning approaches are utilized to recognize × 256 in order to lower the complexity of computation time
and categorize the diseases of the mango fruit by facilitat- (Arivazhagan & Ligi, 2018). In order to improve the dataset
ing the DL principles. Anthracnose fungal disease, crown images, the median noise filtering technique is used (Alsham-
gall, dusty splotches of powdery mildew, and rusts are four mari et al., 2022). All of the images of the dataset were shrunk
distinct illnesses that the model can recognize (Rajbongshi to 224 × 224 in order to suit the required dimensions. In
et al., 2021). The study also conducted experiments using order to obtain a more enhanced image, contrast enhancement
Adam and stochastic gradient descent (SGD) optimization and image normalization were done to adjust the pixel inten-
algorithms to produce better results. The proposed DCNN sity’s value (Ksibi et al., 2022). The acquired images are then
architecture illustrates the effects of data augmentation to enhanced in terms of color, brightness, rotation, and so on,
obtain high accuracy by using transfer learning mechanisms after being resized into the essential pixels for the dataset’s
on Olive leaf (Uğuz & Uysal, 2021). A lightweight and effec- training (Yadav et al., 2021). To cut down on training time,
tive CNN model was developed by concentrating on network the original RGB images were shrunk to 128 × 128 pixels,
depth with fewer parameters in order to achieve the highest and the RGB coefficients were scaled between 0 and 1 (G. B.
classification accuracy results. This model is also appropri- Singh et al., 2021). Scripts of Python built on the Open-Source
ate for mobile devices (Xing et al., 2019). To identify plant CV library automatically computed input images to 224 × 224
leaf diseases, a new automated DCNN architecture based on pixels and 32 × 32 dpi (X. Zhang et al., 2018). Plant images
transfer learning is suggested. Both photographs from pub- are adjusted to 224 pixels per inch in height and width with
licly available databases and images taken in the field can three RGB channels for improved visibility (A. Singh, Son-
be used by the system to identify diseases (DeChant et al., parote, et al., 2020). Rescaling is used to reduce the original
2017; Geetharamani & Pandian, 2019; Kaya et al., 2019). image’s resolution to a more manageable 256 by 256 pixels
Using extremely high-resolution spectral imaging acquired (Pham et al., 2020). Each olive’s hyperspectral imagery was
with unmanned aerial vehicles, a novel DL-based method for saved using a pickle module of Python, which is a quick and
the automatic detection of pathogenic fungi (rust) disease is easy approach to serialize in a Python-oriented pipeline. The
proposed. It had numerous Inception-Resnet multiple layers olives were separated out and scaled to a size of 224 × 224 pix-
for extracting features for a suitable network (X. Zhang, Han, els, which is typical for many CNN architectures (Fazari et al.,
et al., 2019). 2021). In order to reduce the image’s resolution from 3096 ×
3096 to 256 × 256, a scale reduction work is carried out. The
initial action in data pre-processing is to span the data points,
4.3 Discussion on research question 3: How and the second action is to split the data using 80 photos for
can the performance of automatic recognition train set and 20 for test set. Additionally, methods for improve-
systems be enhanced by deep learning models? ment were used to update the distribution of pixels throughout
a wide range of intensities (Lachgar et al., 2022). The noise
Papers were selected for this section by screening them based in leaf images is first removed using a small window median
on the accuracy of automated systems in identifying and cat- filter, and after that the images are reduced to 256 pixels by
egorizing plant diseases. The performance of DCNN with 256 pixels (Cruz et al., 2017). The plant images are reduced
image recognition and categorization has made tremendous in size from 4608 × 3456 to 256 × 256 in order to lower the
advancements in recent years. Working with traditional meth- computational time complexity of the proposed CNN model,
ods is entirely based on specified features. The technique of which consists of one input layer followed by three hidden
pre-processing is employed to improve the image data. The layers (Mia et al., 2020). Before resizing the photos to 224
DL model can increase the algorithm’s recognition accuracy × 224, the images were manually trimmed to the ROI during
and avoid overfitting by being trained on improved images (B. pre-processing. Histogram equalization for contrast enhance-
Jiang, He, et al., 2019). Reducing and transforming the images ment was then applied (Balaji et al., 2023; Nithya et al., 2022;
to ensure they are all identical in both size and type are partic- Rao et al., 2021; Sharma et al., 2022). Image segmentation,
ularly important during the preliminary processing stage. The which further reduces the images to 256 by 256 pixels in size,
most typical type of picture preparation involves shrinking the is a technique used to represent an image in a way that is more
size of the image so that DL models can use it (Kamilaris & meaningful and simpler to analyze. Each cherry leaf image
Prenafeta, 2018). had 100 features that were retrieved using the scale-invariant
Since the NN cannot work with images of different sizes, feature transform technique (K. Zhang, Zhang, et al., 2019).
photos were scaled to a set height and width, and the RGB Image segmentation, which further reduces the images to 256
FIGURE 4 Crops that received the most noticeable consideration in the 94 studies that use convolutional neural networks (CNNs) to find
diseases.
by 256 pixels in size, is a technique utilized to convey an image 4.5 Research question 5: What kinds of
in some way that is more consequential and simpler to analyze plant diseases are most thoroughly researched
(Rao et al., 2021). One of the common techniques is image utilizing CNN methods?
segmentation, which helps the DL process by identifying the
diseased regions (Min et al., 2023; Sa et al., 2016). Finally, it Plant diseases are a threat to global food security, and they can
was realized that using images of a lower size might speed up also have devastating effects on smallholder farmers whose
the training and testing of CNN models. livelihoods depend on producing healthy crops. Plant diseases
are to blame for 35% of India’s annual crop production loss
(D. Singh, Jain, et al., 2020). More than 80% of agricultural
4.4 Research question 4: What types of production in developing countries is generated by small-
cultivated plants are the subject of the most scale cultivators (Ifad, 2013), and claims of yield reductions
research employing CNNs? of over 50% resulting from pests and diseases are common
(Harvey et al., 2014). Smallholder farmers are also dispro-
A summary of current studies on agricultural disease and portionately exposed to disruptions in the food supply caused
pest detection using DL is given in Table 3. It is straightfor- by pathogens because they make up 50% of the world’s
ward to understand that the major part of the table focused hungry people (Sanchez & Swaminathan, 2005). These ele-
on one type of crop. Additionally, “Multiple” categories clas- ments make prompt disease and pest recognition a top priority
sify all research that employ a database with a variety of plant assignment for farmers. Aside from contacting their fellow
phenotypes as well as the disease associated. Each study’s farmers or calling government-funded helplines, farmers do
datasets are briefly described, along with the research’s poten- not have many other options. Public datasets such as PlantVil-
tial future directions. The most often used dataset in this lage, PlantDoc, and Cornell University dataset (Alruwaili
area is the PlantVillage dataset. Other datasets that have been et al., 2019), as well as self-acquired data from various field
gathered and utilized for certain studies have not been relied instances, enabled for advancement in the field of disease and
upon by other researchers. For the training of CNNs, huge pest identification.
databases with millions of images are needed. Unfortunately, It can be seen from Table 3 that the majority of researchers
such extensive and varied datasets are still being assembled trained and tested their DL models using the same dataset.
and made available for usage by academics in the field of plant The number of studies on the detection of plant diseases with
disease detection. A compilation of the plants/crops that have various diseases on an individual leaf is likewise evidently
been taken into account in studies looking at disease detection few. The chart also includes future work and testing accuracy
utilizing CNN is shown in Figure 4. for extending the research.
TA B L E 3 Image-based disease detection and identification in diverse crops (year-wise from 2016 to 2023).
Training and
testing data
are from the Accuracy on
same testing field
Reference Disease type Plant Dataset size database? images (%) Future prospects
Geetharamani Multiple Multiple Very Large Yes 96.46 Objective from plant leaves to
and Pandian other parts of the plants, such as
(2019) flowers, fruits, and stems.
Zeng et al. (2022) Powdery mildew and Rubber tree Small Yes 86.5 and 86.8 To enhance the network’s capacity
anthracnose of the rubber tree disease
identification model, the
model’s network architecture
will be further optimized.
Yan et al. (2020) Scab, frogeye spot, Apple Large Yes 99.01 Collect more apple disease images
and cedar rust to train better models and train
other DL models for real-time
detection.
Ksibi et al. (2022) Olive leaf spots, olive Olive Small Yes 97.08 Mobi-Res-Network model might
leaf scab, peacock be put in a processor with
spot Field-Programmable Gate Array
to integrate with drone
monitoring system.
Divyanth et al. Gray leaf spot, Corn Small Yes 92 More robust disease management
(2023) northern leaf blight, system can be developed with
northern leaf spot more classes of disease.
K. Zhang, Zhang, Powdery mildew Cherry Small Yes 99.6 This work’s practical investigation
et al. (2019) can readily be applied to various
crop’s disease identification
difficulties.
Haque et al. Leaf blight, leaf mite, Litchi Small Yes 95.29 By increasing the number of
(2022) and red rust images and classes, accuracy
can be improved.
Li et al. (2022) Downy blight Litchi Small Yes 90.7 This work can be referred for its
application in practical
production.
Rao et al. (2021) Multiple Grapes, Small Yes 99, 89 To help the farmer address that
mango disease, a recommendation
system will be developed that
may provide a sequentially
action plan.
Arivazhagan and Leaf blight, early Mango Small Yes 96.67 Adding more images will help
Ligi (2018) blight spots improve the classification
Sphaeropsis gall, accuracy even more.
leaf webber, leaf
blotches
Rajbongshi et al. Anthracnose, gall Mango Small Yes 98.00 Can get better accuracy for those
(2021) machi, powdery targeted diseases in future.
mildew, red rust
Pham et al. (2020) Anthracnose, gall Mango Small Yes 89.41 The method for choosing features
midge, powdery for a whole disease monitoring
mildew system needs to be fine-tuned.
Sharma et al. (2022) Anthracnose Mango Small Yes 96.16 IOT applications that can be used
to monitor mango leaves can
employ the proposed paradigm.
(Continues)
TA B L E 3 (Continued)
Training and
testing data
same testing field
Trang et al. (2019) Anthracnose, gall Mango Small Yes 88.46 Accuracy can be further enhanced
midge, powdery by augmenting the dataset and
mildew optimizing parameters of deep
NN.
Fazari et al. (2021) Anthracnose Olive fruit Small Yes 91.8 If employed in postharvest
processes, it might ensure a
decent and exact control of olive
anthracnose while retaining the
integrity and financial worth of
olive and oil from olives.
Lachgar et al. Anthracnose, Olive Small Yes 92.59 Create a drone-based, intelligent
(2022) cyclonium, method for diagnosing olive tree
fumigina, diseases.
verticilliose, knot,
Saissetia oleae
Uğuz and Uysal Aculus olearius and Olive Small Yes 95 Web-based application can be
(2021) olive peacock spot developed for further studies.
Bedi Gole Bacterial spot Peach Small Yes 98.38 Can extend work on more diseases.
(2021)
Yadav et al. (2021) Bacteriosis Peach Small Yes 98.75 To function in actual farming
circumstances, this strategy
might be further connected with
the unmanned aerial vehicle.
K. Zhang, Xu, et al. Xanthomonas Peach Small Yes 99.39 This model may be expanded to
(2019) campestris handle more peach leaf disease.
Ahmad et al. (2020) Brown rot, shot hole Plum Small Yes 92 To increase the dataset’s size and
develop a highly effective object
detection model.
Hassan and Maji NLB lesions Maize Large Yes Not specified This restricts the data’s capacity to
(2022) be generalized because signs of
one particular disease in various
areas may manifest or progress
differently.
G. B. Singh Bacterial spot, early Tomato Large Yes 99.1 Need to work with images having
et al. (2021) blight, leaf curl, complex backgrounds.
Septoria blight,
mosaic virus,
tomato spider mite,
leaf mold, tomato
blight
Liu et al. (2023) Multiple Red fuji apple Large Yes 96.2 Need to gather more dataset
images in order to enhance the
model’s recognition capabilities
and enable it to recognize apple
leaves in a range of settings.
D. Zhang et al. Multiple Apple Small Yes 93.765 Try different models and
(2021) optimization techniques in
future work to enhance the
algorithm’s performance.
(Continues)
Training and
testing data
same testing field
Ma et al. (2018) Powdery and downy Cucumber Large Yes 93.4 Prediction accuracy can be
mildew, leaf spots, improved.
anthracnose
Nagasubramanian Charcoal rot Soybean Small Yes 95.73 Can extend work of 3D CNN
et al. (2019) model on more diseases.
Ramesh and Vydeki Bacterial blight, Rice Small Yes 90.57 To improve the detection and
(2020) anthracnose, sheath categorization of plant diseases,
blight and rotten any improved approach can be
neck used to achieve the greatest
efficacy by reducing the error in
classification.
Dey et al. (2016) Leaf rot Betel Small Yes Not mentioned Can expand dataset.
Johannes et al. Rust, septoria, and Wheat Small Yes Not mentioned The next phases will measure how
(2017) yellow or tan leaf this initial phase identification
spot can assist the user in reacting
quickly and planning for various
precautionary operations, such
as crop protection spray.
Qin et al. (2016) Common leaf spot, Alfalfa Small Yes 94.74 Further investigation is required to
rust, assess the possibility of false
leptosphaerulina positives using the proposed
leaf spot, method with other leaf diseases.
Cercospora leaf spot
Zhu et al. (2017) TMV-infected Tobacco Small Yes 98 Data fusion is required and
encouraged in upcoming studies
since it can be done by utilizing
the spatial and spectral
information of hyperspectral
imagery.
Kerkech et al. ESCA Grapevine Not Not mentioned 95.86 The UAV multispectral picture
(2018) mentioned database has to be expanded
with vineyard plots and new
samples of vine diseases in order
to improve the system.
Amara et al. (2017) Banana sigatoka and Banana Small Yes 98.6 Determining the disease’s
banana speckle seveverity automatically is
important because it can help
farmers decide what steps to
take to stop the disease from
spreading.
Ramcharan et al. Brown leaf spot, red Cassava Large Yes 98 As background variations had little
(2017) spider and green impact on the classification
mites, brown streak, accuracy of the model, using
cassava mosaic many images to forecast the
virus disease in the future will
increase diagnostic precision.
Note: Images ≥50,000 equal very large database, where images between 10,000 and 49,999 indicate a large dataset, while images under 10,000 indicate a small database.
Abbreviations: DL, deep learning; IOT, internet of things; NN, neural network; TMV, tobacco mosaic virus; UAV, unmanned aerial vehicle.
FIGURE 5 Most prevalent Frameworks in convolutional neural networks (CNNs).
4.6 Research question 6: Which CNN 5.1 Not enough substantial datasets
algorithm dominates modern methods?
The fundamental challenge in training CNN models for plant
Instead of starting from scratch, researchers might start creat- disease recognition is the lack of big, highly variable image
ing their models using a variety of effective and well-liked datasets with well-annotated labels (Ahmad et al., 2020; Alom
designs. These include Inception-ResNet, CaffeNet, VGG, et al., 2019; Arivazhagan & Ligi, 2018; Dey et al., 2016;
GoogleNet, AlexNet, and many others. As demonstrated in Haque et al., 2022; Ren et al., 2015; Trang et al., 2019; K.
Table 4, each design has distinct benefits and situations that Zhang, Zhang, et al., 2019). The only sizable openly acces-
is more suitable to be employed. It is also crucial to note sible datasets at this moment are the PlantVillage dataset
that nearly all of the NNs in the list have pre-trained weights, (Geetharamani & Pandian, 2019; Hernández & López, 2020;
which means their network has already been trained using a Rao et al., 2021) and PDDB database (Barbedo et al., 2018).
dataset and is capable of precise categorization for a specific In fact, Arivazhagan and Ligi (2018) and Trang et al. (2019)
domain. Additionally, there are numerous platforms and tools suggest that DL approaches are well adapted to plant disease
available that allow researchers to explore DL. TensorFlow, prediction difficulties as long as there is enough data avail-
Theano, Keras (which is an application programming inter- able. However, collecting field data is time-consuming and
face for application developers built on top of TensorFlow and expensive and demands subject-matter specialists for precise
Theano), Caffe, TFLearn, Pylearn2, PyTorch, and DL Matlab annotation. It has been suggested that one way to solve this
are the most widely used ones. Figure 5 shows how the stud- issue is to create standard databases in both lab and field
ies were distributed and organized according to the framework settings for accurate and efficient crop disease identification.
that was utilized to build the strategy. Popular designs such as
the ones listed above (AlexNet, VGG, GoogleNet) are incor-
porated into some of these tools (e.g., Theano, Caffe), either 5.2 Dominant strategies and automated
as libraries or classes. systems
From LeNet-5’s debut in 1998 until the present, numerous

5 DISCUSSION advancements in CNN architectures have been proposed.
These advancements can be divided into the following cat-
This section provides a summary based on the data and facts egories: parameter optimization, regularization, structural
analyzed from selected articles for this survey. Our study is modification (CONV layer, POOL layer, and ACT or RELU
broken down into four broad categories that summarize the function), cost functions, and quick processing. However, it
state of the art of using CNNs to diagnose plant diseases in has been noted that the design of new blocks and struc-
order to keep the conversation on track with the main research tural reformulation are the factors that are mostly responsible
issues, all of which were addressed in the preceding part. for CNNs performance increase. We discovered that the
TA B L E 4 Summary of implemented convolutional neural network (CNN) architectures based on their strength and weakness.
Transfer
Reference Crop type Architecture learning Framework Strength Weakness Principal strategy
Wang et al. (2017) Apple VGG No Theano Fully automated system that Accuracy still needs to be improved Hyper parameters and
BAGGA AND GOYAL
accurately classifies plant fine-tuning

disease severity
Anim-Ayeko et al. Potato and ResNet-9 model No Pytorch An outstanding testing accuracy of Performance degrades if images It would be advantageous to
(2023) tomato and Saliency 99.25% is achieved even by background is considered implement this model with
maps considering the factors such as an interface that would
farm temperature, soil allow farmers to capture
properties, leaf area properties immediate images of the
crops that are impacted
Geetharamani and Multiple Proposed No scikit-learn, Consistent and reliable model Using some fine-tuning approaches, the Customizing layers
Pandian (2019) DCNN Keras, pillow, increased dataset will enhance the
and opencv model’s efficiency and accuracy
libraries
Zeng et al. (2022) Rubber tree Improved No Pytorch 1.8 Cost effective solution with The detection precision still needs Customizing layers
YOLOv5 reduced parameters development
Yan et al. (2020) Apple Improved Yes Keras Achieved high accuracy with Training time is longer Customizing layers
VGG16 reduced parameters
Ksibi et al. (2022) Olive MobiResNet No TensorFlow and High accuracy Higher training and testing run time, Customizing layers
Keras high-performance h/w is required
Divyanth et al. Corn UNet- No NA Reliable solution for disease Complex model structure Semantic segmentation,
(2023) DeepLabV3 prediction customizing layers
+ model
K. Zhang, Zhang, Cherry GoogLeNet Yes NA High accuracy Limited to detect single disease only Customizing layers
et al. (2019)
Haque et al. (2022) Litchi Proposed CNN No NA Simple and easy to use Recognition rate is not good for all Customizing layers
diseases
Rao et al. (2021) Mango AlexNet Yes TensorFlow Model is appropriate and effective Due to challenging situations, Customizing layers
in systems with minimal misclassified the images into other
hardware classes
Arivazhagan and Mango Proposed No NA Faster convergence rate and good Reliable for smaller dataset only Customizing layers
Ligi (2018) DCNN training performance
Rajbongshi et al. Mango DenseNet201 Yes NA Simple and cost-effective Lower classification accuracy; an Customizing layers
(2021) over-fitting possibility
Sharma et al. (2022) Mango Proposed CNN No TensorFlow Works only for single disease Performance degraded with larger Customizing layers
dataset
(Continues)
13 of 21
14 of 21
Transfer
Trang et al. (2019) Mango Proposed Yes NA Simple and cost-effective Lower classification accuracy; an Contrast enhancement and
DCNN over-fitting possibility customizing layers
Fazari et al. (2021) Olive ResNet101 Yes TensorFlow Pre-processing increased efficacy Adapted to process only two classes of Customizing layers
61-band hyperspectral images
Lachgar et al. Olive MobileNet Yes TensorFlow Use of optimizer algorithms Accuracy and reliability are still a Fine-tuning and customizing
(2022) (Rmsprop) for effective and challenge layers
efficient results
Uğuz and Uysal Olive CNN based on Yes TensorFlow and Superior results by using Adam Compatible with RGB images only Customizing layers
(2021) VGG16, Keras and SGD optimization
VGG19 algorithms
Bedi and Gole Peach CAE + CNN No Keras Required lesser number of training Limited to only single disease detection Dimensionality reduction
(2021) parameters and customizing layers
Yadav et al. (2021) Peach Proposed CNN Yes TensorFlow These models are validated using Limited to only single disease detection Customizing layers
an unseen test dataset
Ahmad et al. (2020) Plum Optimized Yes TensorFlow, Improved performance of the Resource-constrained Fine-tuning and customizing
Inception-v3 MatLab inception network even with layers
background noise
G. B. Singh et al. Tomato Proposed CNN No TensorFlow and Cost-effective and efficient DL Erroneously divided the images into Customizing layers
(2021) inspired by Keras model several classifications as a result of
Alexnet stressful circumstances
Liu et al. (2023) Apple Improved No TensorFlow and Much fewer model parameters, Complex network structure Feature reuse
MobileNetV2 Keras model’s emphasis on infected
Network areas
D. Zhang, Yang Apple Restnet34 Yes Fastai No need to increase the training Challenging for larger datasets Customizing layers
et al. (2021) error percentage
Ma et al. (2018) Cucumber DCNN and Yes TensorFlow To detect the disease at its earliest Testing accuracy is still a challenge Image segmenting and
AlexNet stages, use infrared thermal and customizing layers
hyperspectral imagery
S. Zhang, Huang, Cucumber Three channel No Matlab Avoided the stage of intricate Dusty and noisy background may affect Novel approach with
et al. (2019) and Alexnet preprocessing such lesion the classification accuracy customized layers
tomato segmentation and feature
extraction
(Continues)
BAGGA AND GOYAL
Transfer
BAGGA AND GOYAL
U. P. Singh et al. Mango Mutilayer CNN No TensorFlow/ Computationally simple and Consistency issues that came up while Cutting edge design with
(2019) Alexnet Keras effective, with a classification using a real-time dataset customized layers
accuracy of 97.13%
Panchal et al. (2023) Multiple Improved CNN Yes pytorch0.4.1 Categorizes the extent of the Performance degraded for non-transfer Customized layers
ResNet-B disease into general and severe learning methods
so that appropriate preventive
measures can be chosen
Alruwaili et al. Olive Novel Yes TensorFlow Performs efficiently in contexts Model training takes up a lot of time Customized layers
(2019) DL-based that are more complicated, with
Alexnet a larger and richer dataset
B. Jiang, He, et al. Apple VGGNet No Caffe Achieved good accuracy even with Complex network structure Customized layers
(2019) Inception most complex diseased classes
Shoaib et al. (2023) Multiple Proposed CNN No TensorFlow/ Reduced parameters Limited to single class, accuracy needs Customized layers
Keras to be improved
Yu and Son (2019) Apple ROI-based Yes TensorFlow ROI-based analysis of images; Classification precision needs to be Customized layers
DCNN accuracy has significantly increased, complex structure
improved
Geetharamani and Multiple Nine-layer No scikit-learn, Good classification accuracy with While working with larger datasets, the Customized layers
Pandian (2019) DCNN Keras, pillow, smaller datasets model’s performance and accuracy
and opencv must be enhanced utilizing some
fine-tuning strategies
Hang et al. (2019) Corn, Inception and No Caffe A smaller number of model Testing accuracy still needs to be raised Customized layers
cherry, SE module- parameters result in a significant
apple based reduction in training time and
improved less memory requirements of
CNN parameters
X. Zhang, Han, Wheat Inception- No TensorFlow/ Efficient and feasible solution for Higher training and testing run time; Customized layers
et al. (2019) Resnet-based Keras hyper spectral images high-performance h/w is required
DCNN
Xing et al. (2019) Citrus Weakly No Keras with This model may be fitted to Network overhead due to feature reuse Customized layers
fruits DenseNet TensorFlow various datasets by network method
structure optimization
(Continues)
15 of 21
16 of 21
Transfer
Kaya et al. (2019) Multiple VGG 16, Yes TensorFlow/ When deep features and Limited to plants classification only Fine-tuning and customizing
AlexNet Keras fine-tuning are combined, the layers
model performs better
DeChant et al. Maize CNN Yes Theano Reliable automated system, good Limited to single disease prediction Customizing layers and
(2017) testing accuracy segmenting images
Zhu et al. (2017) Tobacco BPNN, ELM, No NA Infected leaves may be quickly and Accuracy still needs to be improved Customizing layers
LS-SVM painlessly identified using
hyperspectral imaging in a short
amount of time
Ramcharan et al. Cassava Improved Yes TensorFlow This approach bypasses the Performance degraded when leaflet is Customizing layers
(2017) Inception V3 difficult and time-consuming used instead of whole leaf
process of feature extraction
from images for model training,
and the model may be quickly
trained on a desktop and
deployed on a mobile device
Qin et al. (2016) Alfa-alfa K median No Matlab The four main types of leaf Accuracy still needs to be improved for Lesion segmentation
clustering diseases can be identified and other diseases
algorithm recognized with high testing
and linear accuracy
discriminant
analysis
Abbreviations: DL, deep learning; NA, not available; RGB, red green blue; ROI, region of interest.
BAGGA AND GOYAL
explorations that advanced practices for the detection and cat- cumulative table for research question 3 and 4. It is given that
egorization of diseases of plants using CNNs explored various both the training and testing images came from the identical
innovative methods to supply a dependable solution to the dataset, and diseases were initially detected with an accuracy
problem examined when we tried to address research ques- of 99.01%. But when the model was given test images from
tions 2 and 6 of Section 4. Regarding the development of a different dataset, the accuracy fell down poorly (DeChant
new architectures, three trends came to our attention. LeNet, et al., 2017). The present DL-based plant disease identifica-
AlexNet, and ResNet are a few examples of structurally tion methods need each user to build a DL model in their area
reformulated classic designs. Another development includes of expertise under predetermined or regulated image acquisi-
blocks from different architectures are being combined, as tion settings, necessitating a shift in strategy. Most research
seen in the (Tian et al., 2019) (Yolo Densely connected Net) evaluated so far effectively recognized diseases; however, the
and the (X. Zhang, Han, et al., 2019) (Inception-ResNet). trained models did not generalize well to images from other
The most recent trend includes recommendations for imple- datasets and circumstances. As a result, present techniques
menting models specifically created for the recognition and will be just as ineffectual as farmers’ field scouting. The main
categorization of crop diseases, with a focus on hyperspec- takeaway from Table 3 is to develop a reliable model for pre-
tral image-based models. The inherent connections among the cisely classifying diseases. Current DL-based systems also
three categories that have been identified (customizing lay- have trouble distinguishing between different types of dis-
ers, new structure, fine-tuning, and model’s hyperparameters) eases. The majority of research on the identification of plant
are another important aspect of the dominating techniques. diseases concentrated on training DL models to recognize a
The research under investigation falls into categories that face particular crop or a small number of diseases. It is important
similar difficulties, such as lowering the number of param- to ensure that a trained model is capable of global disease
eters in model, shrinking the model size, the necessity of identification from all datasets and conditions.
optimizing hyperparameters, and an increase in prediction
capability tuned at a computational cost that is reasonable.
When examining the classifications with customized layer 5.4 Utilizing methods of pre-processing to
structures and novel architectural designs, it becomes clear change the image’s size and data
that the created models follow a different path than conven-
tional structures when dealing with problems relating to the Preparing an image for use in further operations, such as
diagnosis of plant illnesses. While classic architectures such cropping, histogram equalization, contrast enhancement, and
as AlexNet, VGG16, and ResNet gained popularity for their decreasing the unpredictability of natural illumination, is
ability to generalize, the models under investigation have a known as image pre-processing and segmentation. Images
tendency to become more focused on a single type of plant. are uniformly resized and translated into a single image for-
mat before being stored in the database as demonstrated
in research question 3. De-correlation stretching is used
5.3 Type of crops and diseases to enhance downsized photographs in order to emphasize
the differences between the image’s color bands. In short,
The research revealed that in addition to methods that look the standardization of the described approaches has not yet
at a variety of crops, including apples, cucumbers, cherries, been set and realized because such techniques significantly
grapes, wheat, rice, corn, and tomatoes, mangoes, olive, betel, contribute to boosting accuracy.
cassava, and bananas are the subject of the most probes while
using CNNs investigative tools. The dearth of studies with
methodologies that focused solely on these crops was obvi- 6 CONCLUSION
ous given the assumption that grains and millets are the main
sources of energy consumed by people. In addition, there We have conducted a study of DL-based experiments used in
are plenty of studies on stone fruits such as mango, peach, the agricultural area in this publication. Ninety-four papers
and olive with smaller datasets. However, when we look at that are pertinent have been found, and we have examined
other varieties of stone fruits such as plum, apricot, cherry, the topic and issue they addressed in detail. The automated
litchi, and nectarines, there are relatively few papers. If fruit plant disease detection systems suggest identifying and cat-
wines are made, the increasing production of these fruits can egorizing plant diseases by fusing CNNs with the expertise
be employed profitably, leading to the creation of jobs and of phytopathology specialists and the capacity to extract
greater returns for the orchardists. Apricots, peaches, nec- symptomatic signals. The challenge of semantically cata-
tarines, plums, and cherries are planted on roughly 43,000 loguing the data in sets with enough labeled samples and
acres in India alone, and they produce about 0.25 million representative data is exacerbated by the variety of difficul-
tonnes of fruits in a year. It is shown in Table 3 which is a ties and the particularities of real-world settings. This issue
turns into a pertinent and difficult roadblock for creating DL Altieri, M. A. (2018). Agroecology: The science of sustainable agricul-
methodologies that are more practical. Our findings show ture. CRC Press. https://doi.org/10.1201/9780429495465
the tremendous improvements in CNN plant disease detec- Amara, J., Bouaziz, B., & Algergawy, A. (2017). A deep learning-based
approach for banana leaf diseases classification. BTW, 266, 79–88.
tion methods. Despite the dataset’s complexity made up of
Anim-Ayeko, A. O., Schillaci, C., & Lipani, A. (2023). Automatic
images taken in actual agricultural contexts, it was feasible to
blight disease detection in potato (Solanum tuberosum L.) and tomato
see that classical structures paired with optimization and cus- (Solanum lycopersicum, L. 1753) plants using deep learning. Smart
tomization techniques offered a significant accuracy. There Agricultural Technology, 4, 100178.
is an increasing propensity for several methodologies to sug- Arivazhagan, S., & Ligi, S. V. (2018). Mango leaf diseases identification
gest novel CNN models relied on the method of recognizing using convolutional neural network. International Journal of Pure
diseases of plants. Our study demonstrates that methods that and Applied Mathematics, 120(6), 11067–11079.
combine CNNs with multispectral and hyperspectral images, Badrinarayanan, V., Kendall, A., & Cipolla, R. (2017). Segnet: A deep
convolutional encoder-decoder architecture for image segmentation.
which have more potential for data than RGB photos, are
IEEE Transactions on Pattern Analysis and Machine Intelligence,
becoming increasingly popular. Ultimately, we also wish to 39(12), 2481–2495.
issue a warning regarding grains, millets, and stone fruits hav- Balaji, V., Anushkannan, N. K., Narahari, S. C., Rattan, P., Verma, D.,
ing strong financial and food influence, which were frequently Awasthi, D. K., Pandian, A. A., Veeramanickam, M. R., & Mulat, M.
disregarded by virtually all techniques. Additionally, we draw B. (2023). Deep transfer learning technique for multimodal disease
attention to the fact that in the majority of the experiments we classification in plant images. Contrast Media & Molecular Imaging,
have already analyzed, the trained models did not generalize 2023, 5644727.
effectively to images from varied datasets and situations, and Barbedo, J. G. A. (2016). A review on the main challenges in automatic
plant disease identification based on visible range images. Biosystems
the majority of research on diagnosing plant diseases concen-
Engineering, 144, 52–60.
trated on developing DL models to identify a particular crop’s Barbedo, J. G. A., Koenigkan, L. V., Halfeld, V. B. A., Costa, R. V.,
diseases or a limited set of diseases. Nechet, K. L., Godoy, C. V., Junior, M. L., Patricio, F. R. A., Talamini,
V., Chitarra, L. G., & Oliveira, S. A. S. (2018). Annotated plant
AU T H O R C O N T R I B U T I O N S pathology databases for image-based detection and recognition of
Manju Bagga: Writing—original draft; writing—review and diseases. IEEE Latin America Transactions, 16(6), 1749–1757.
editing. Sonali Goyal: Supervision; visualization. Bedi, P., & Gole, P. (2021). Plant disease detection using hybrid
model based on convolutional autoencoder and convolutional neural
network. Artificial Intelligence in Agriculture, 5, 90–101.
AC K N OW L E D G M E N T S
Chen, L. C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A.
The authors would like to thank reviewers for their valuable L. (2017). Deeplab: Semantic image segmentation with deep convo-
comments. lutional nets, atrous convolution, and fully connected CRFs. IEEE
Transactions on Pattern Analysis and Machine Intelligence, 40(4),
C O N F L I C T O F I N T E R E ST STAT E M E N T 834–848.
The authors declare no conflicts of interest. Cruz, A. C., Luvisi, A., De, B. L., & Ampatzidis, Y. (2017). X-FIDO:
An effective application for detecting olive quick decline syndrome
with deep learning and data fusion. Frontiers in Plant Science, 8,
ORCID
1741.
Manju Bagga https://orcid.org/0000-0002-3260-7953
Das, C. S., Goswami, S., Bashyam, S., Samal, A., & Awada, T. (2017).
Automated stem angle determination for temporal plant phenotyp-
REFERENCES ing analysis. Proceedings of the IEEE international conference on
Ahmad, J., Jan, B., Farman, H., Ahmad, W., & Ullah, A. (2020). Disease computer vision workshops, Venice, Italy. https://doi.org/10.1109/
detection in plum using convolutional neural network under true field ICCVW.2017.237
conditions. Sensors, 20(19), 5569. Dawei, W., Limiao, D., Jiangong, N., Jiyue, G., Hongfei, Z., & Zhongzhi,
Alom, M. Z., Taha, T. M., Yakopcic, C., Westberg, S., Sidike, P., Nasrin, H. (2019). Recognition pest by image-based transfer learning. Journal
M. S., Hasan, M., Van-Essen, B. C., Awwal, A. A., & Asari, V. of the Science of Food and Agriculture, 99(10), 4524–4531.
K. (2019). A state-of-the-art survey on deep learning theory and DeChant, C., Wiesner, H. T., Chen, S., Stewart, E. L., Yosinski, J., Gore,
architectures. Electronics, 8(3), 292. M. A., Nelson, R. J., & Lipson, H. (2017). Automated identification
Alruwaili, M., Alanazi, S., Abd, E. G. S., & Shehab, A. (2019). An effi- of northern leaf blight-infected maize plants from field imagery using
cient deep learning model for olive diseases detection. International deep learning. Phytopathology, 107(11), 1426–1432.
Journal of Advanced Computer Science and Applications, 10(8), Dey, A. K., Sharma, M., & Meshram, M. R. (2016). Image processing-
486–492. based leaf rot disease, detection of betel vine (Piper betle L.).
Alshammari, H., Gasmi, K., Krichen, M., Ammar, L. B., Abdelhadi, M. Procedia Computer Science, 85, 748–754.
O., Boukrara, A., & Mahmood, M. A. (2022). Optimal deep learn- Divyanth, L. G., Ahmad, A., & Saraswat, D. (2023). A two-stage
ing model for olive disease diagnosis based on an adaptive genetic deep-learning based segmentation model for crop disease quantifica-
algorithm. Wireless Communications and Mobile Computing, 2022, tion based on corn field imagery. Smart Agricultural Technology, 3,
8531213. 100108.
Fazari, A., Pellicer, V. O. J., Gómez, S. J., Bernardi, B., Cubero, S., neural network based plant classification models. Computers and
Benalia, S., Zimbalatti, G., & Blasco, J. (2021). Application of Electronics in Agriculture, 158, 20–29.
deep convolutional neural networks for the detection of anthrac- Kerkech, M., Hafiane, A., & Canals, R. (2018). Deep leaning approach
nose in olives using VIS/NIR hyperspectral images. Computers and with colorimetric spaces and vegetation indices for vine diseases
Electronics in Agriculture, 187, 106252. detection in UAV images. Computers and Electronics in Agriculture,
Fuentes, A. F., Yoon, S., Lee, J., & Park, D. S. (2018). High-performance 155, 237–243.
deep neural network-based tomato plant diseases and pests diagno- Ksibi, A., Ayadi, M., Soufiene, B. O., Jamjoom, M. M., & Ullah,
sis system with refinement filter bank. Frontiers in Plant Science, 9, Z. (2022). MobiRes-net: A hybrid deep learning model for detect-
1162. ing and classifying olive leaf diseases. Applied Sciences, 12(20),
Geetharamani, G., & Pandian, A. (2019). Identification of plant leaf 10278.
diseases using a nine-layer deep convolutional neural network. Lachgar, M., Hrimech, H., & Kartit, A. (2022). Optimization techniques
Computers & Electrical Engineering, 76, 323–338. in deep convolutional neuronal networks applied to olive diseases
Hang, J., Zhang, D., Chen, P., Zhang, J., & Wang, B. (2019). Classifi- classification. Artificial Intelligence in Agriculture, 6, 77–89.
cation of plant leaf diseases based on improved convolutional neural LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based
network. Sensors, 19(19), 4161. learning applied to document recognition. Proceedings of the IEEE,
Haque, I., Alim, M., Alam, M., Nawshin, S., Noori, S. R. H., & Habib, M. 86(11), 2278–2324. 10.1109/5.726791
T. (2022). Analysis of recognition performance of plant leaf diseases Li, J., Wu, J., Lin, J., Li, C., Lu, H., & Lin, C. (2022). Nondestruc-
based on machine vision techniques. Journal of Human, Earth, and tive identification of litchi downy blight at different stages based on
Future, 3(1), 129–137. spectroscopy analysis. Agriculture, 12(3), 402.
Harvey, C. A., Rakotobe, Z. L., Rao, N. S., Dave, R., Razafimahatratra, Liu, S., Bai, H., Li, F., Wang, D., Zheng, Y., Jiang, Q., & Sun, F. (2023).
H., Rabarijohn, R. H., Rajaofara, H., & MacKinnon, J. L. (2014). An apple leaf disease identification model for safeguarding apple food
Extreme vulnerability of smallholder farmers to agricultural risks safety. Food Science and Technology, 43, e104322.
and climate change in Madagascar. Philosophical Transactions of the Luo, Z., Yang, W., Yuan, Y., Gou, R., & Li, X. (2023). Semantic seg-
Royal Society B: Biological Sciences, 369(1639), 20130089. mentation of agricultural images: A survey. Information Processing
Hassan, S. M., & Maji, A. K. (2022). Plant disease identification using in Agriculture. https://doi.org/10.1016/j.inpa.2023.02.001
a novel convolutional neural network. IEEE Access, 10, 5390–5401. Ma, J., Du, K., Zheng, F., Zhang, L., Gong, Z., & Sun, Z. (2018).
Heaton, J., Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep A recognition method for cucumber diseases using leaf symptom
learning: The MIT Press, 2016, 800 pp, ISBN: 0262035618. Genetic images based on deep convolutional neural network. Computers and
Programming and Evolvable Machines, 19(1–2), 305–307. Electronics in Agriculture, 154, 18–24.
Hernández, S., & López, J. L. (2020). Uncertainty quantification for Mahlein, A. K. (2016). Plant disease detection by imaging sensors–
plant disease detection using Bayesian deep learning. Applied Soft parallels and specific demands for precision agriculture and plant
Computing, 96, 106597. https://doi.org/10.1016/j.asoc.2020.106597 phenotyping. Plant Disease, 100(2), 241–251. https://doi.org/10.
Huang, G., Liu, Z., Van, D. M. L., & Weinberger, K. Q. (2017). Densely 1094/PDIS-03-15-0340-FE
connected convolutional networks. Proceedings of the IEEE confer- McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the
ence on computer vision and pattern recognition, Honolulu, HI, USA. ideas immanent in nervous activity. The Bulletin of Mathematical
https://doi.org/10.1109/CVPR.2017.243 Biophysics, 5, 115–133.
Ifad, U. (2013). Smallholders, food security and the environment. Rome: Mia, M. R., Roy, S., Das, S. K., & Rahman, M. A. (2020). Mango leaf
International Fund for Agricultural Development. https://wedocs. disease recognition using neural network and support vector machine.
unep.org/20.500.11822/32462 Iran Journal of Computer Science, 3, 185–193.
Jiang, B., He, J., Yang, S., Fu, H., Li, T., Song, H., & He, D. (2019). Min, B., Kim, T., Shin, D., & Shin, D. (2023). Data augmentation method
Fusion of machine vision technology and AlexNet-CNNs deep learn- for plant leaf disease recognition. Applied Sciences, 13(3), 1465.
ing network for the detection of postharvest apple pesticide residues. Nagasubramanian, K., Jones, S., Singh, A. K., Sarkar, S., Singh, A.,
Artificial Intelligence in Agriculture, 1, 1–8. & Ganapathysubramanian, B. (2019). Plant disease identification
Jiang, P., Chen, Y., Liu, B., He, D., & Liang, C. (2019). Real- using explainable 3D deep learning on hyperspectral images. Plant
time detection of apple leaf diseases using deep learning approach Methods, 15, 1–10.
based on improved convolutional neural networks. IEEE Access, 7, Ni, J., Wang, R., & Tang, J. (2023). ADSSD: Improved single-shot
59069–59080. detector with attention mechanism and dilated convolution. Applied
Jin, S., Su, Y., Gao, S., Wu, F., Hu, T., Liu, J., Li, W., Wang, D., Chen, Sciences, 13(6), 4038.
S., Jiang, Y., & Pang, S. (2018). Deep learning: Individual maize seg- Nithya, R., Santhi, B., Manikandan, R., Rahimi, M., & Gandomi, A. H.
mentation from terrestrial lidar data using faster R-CNN and regional (2022). Computer vision system for mango fruit defect detection using
growth algorithms. Frontiers in Plant Science, 9, 866. deep convolutional neural network. Foods, 11(21), 3483.
Johannes, A., Picon, A., Alvarez, G. A., Echazarra, J., Rodriguez, V. Panchal, A. V., Patel, S. C., Bagyalakshmi, K., Kumar, P., Khan,
S., Navajas, A. D., & Ortiz, B. A. (2017). Automatic plant disease I. R., & Soni, M. (2023). Image-based plant diseases detec-
diagnosis using mobile capture devices, applied on a wheat use case. tion using deep learning. Materials Today: Proceedings, 80,
Computers and Electronics in Agriculture, 138, 200–209. 3500–3506.
Kamilaris, A., & Prenafeta, B. F. X. (2018). Deep learning in agriculture: Pandey, P., Irulappan, V., Bagavathiannan, M. V., & Senthil, K. M.
A survey. Computers and Electronics in Agriculture, 147, 70–90. (2017). Impact of combined abiotic and biotic stresses on plant
Kaya, A., Keceli, A. S., Catal, C., Yalic, H. Y., Temucin, H., & growth and avenues for crop improvement by exploiting physio-
Tekinerdogan, B. (2019). Analysis of transfer learning for deep morphological traits. Frontiers in Plant Science, 8, 537.
Panwar, P., Roshan, P., Singh, R., Rai, M., Mishra, A. R., & Chauhan, Singh, V. (2019). Sunflower leaf diseases detection using image segmen-
S. S. (2022). DDNet—A deep learning approach to detect driver tation based on particle swarm optimization. Artificial Intelligence in
distraction and drowsiness. Evergreen, 9, 881–892. Agriculture, 3, 62–68.
Pham, T. N., Van, T. L., & Dao, S. V. T. (2020). Early disease classifi- Singh, V., & Misra, A. K. (2017). Detection of plant leaf diseases using
cation of mango leaves using feed-forward neural network and hybrid image segmentation and soft computing techniques. Information
metaheuristic feature selection. IEEE Access, 8, 189960–189973. Processing in Agriculture, 4(1), 41–49.
Qin, F., Liu, D., Sun, B., Ruan, L., Ma, Z., & Wang, H. (2016). Identi- Taye, M. M. (2023). Theoretical understanding of convolutional neu-
fication of alfalfa leaf diseases using image recognition technology. ral network: Concepts, architectures, applications, future directions.
PLoS One, 11(12), e0168274. Computation, 11(3), 52.
Rajbongshi, A., Khan, T., Pramanik, M. M. R. A., Tanvir, S. M., & Terven, J., & Cordova-Esparza, D. A. (2023). Comprehensive review of
Siddiquee, N. R. (2021). Recognition of mango leaf disease using YOLO: From YOLOv1 and beyond. https://doi.org/10.48550/arXiv.
convolutional neural network models: A transfer learning approach. 2304.00501
Indonesian Journal of Electrical Engineering and Computer Science, Tian, Y., Yang, G., Wang, Z., Li, E., & Liang, Z. (2019). Detection
23(3), 1681–1688. of apple lesions in orchards based on deep learning methods of
Ramcharan, A., Baranowski, K., McCloskey, P., Ahmed, B., Legg, J., & CycleGAN and YOLOV3-dense. Journal of Sensors, 2019, 7630926.
Hughes, D. P. (2017). Deep learning for image-based cassava disease https://doi.org/10.1155/2019/7630926
detection. Frontiers in Plant Science, 8, 1852. Trang, K., Ton, T. L., Thao, N. G. M., & Thi, N. T. T. (2019). Mango
Ramesh, S., & Vydeki, D. (2020). Recognition and classification of diseases identification by a deep residual network with contrast
paddy leaf diseases using Optimized Deep Neural network with Jaya enhancement and transfer learning. 2019 IEEE conference on sus-
algorithm. Information Processing in Agriculture, 7(2), 249–260. tainable utilization and development in engineering and technologies
Rao, U. S., Swathi, R., Sanjana, V., Arpitha, L., Chandrasekhar, K., (CSUDET) (pp. 138–142). IEEE.
& Naik, P. K. (2021). Deep learning precision farming: Grapes and Türkoğlu, M., & Hanbay, D. (2019). Plant disease and pest detection
mango leaf disease detection by transfer learning. Global Transitions using deep learning-based features. Turkish Journal of Electrical
Proceedings, 2(2), 535–544. Engineering and Computer Sciences, 27(3), 1636–1651. https://doi.
Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towards org/10.3906/elk-1809-181
real-time object detection with region proposal networks. Advances Uğuz, S., & Uysal, N. (2021). Classification of olive leaf diseases
in Neural Information Processing Systems, 28, 1137–1149. using deep convolutional neural networks. Neural Computing and
Sa, I., Ge, Z., Dayoub, F., Upcroft, B., Perez, T., & McCool, C. (2016). Applications, 33(9), 4133–4149.
Deepfruits: A fruit detection system using deep neural networks. Wang, G., Sun, Y., & Wang, J. (2017). Automatic image-based plant
Sensors, 16(8), 1222. disease severity estimation using deep learning. Computational
Sanchez, P. A., & Swaminathan, M. S. (2005). Cutting world hunger in Intelligence and Neuroscience, 2017, 2917536.
half. Science, 307(5708), 357–359. Wiesner, H. T., Stewart, E. L., Kaczmar, N., DeChant, C., Wu, H.,
Shahoveisi, F., Taheri-Gorji, H., Shahabi, S., Hosseinirad, S., Markell, Nelson, R. J., Lipson, H., & Gore, M. A. (2018). Image set for deep
S., & Vasefi, F. (2023). Application of image processing and transfer learning: Field images of maize annotated with disease symptoms.
learning for the detection of rust disease. Scientific Reports, 13(1), BMC Research Notes, 11(1), 1–3.
5133. Xing, S., Lee, M., & Lee, K. K. (2019). Citrus pests and diseases recog-
Sharma, A., Bijral, R. K., Manhas, J., & Sharma, V. (2022). Mango nition model using weakly dense connected convolution network.
leaf diseases detection using deep learning. International Journal of Sensors, 19(14), 3195.
Knowledge Based Computer Systems, 10(1). https://doi.org/10.1109/ Yadav, S., Sengar, N., Singh, A., Singh, A., & Dutta, M. K. (2021). Iden-
ASIANCON51346.2021.9544689 tification of disease using deep learning and evaluation of bacteriosis
Shoaib, M., Shah, B., Ei-Sappagh, S., Ali, A., Ullah, A., Alenezi, F., in peach leaf. Ecological Informatics, 61, 101247.
Gechev, T., Hussain, T., & Ali, F. (2023). An advanced deep learning Yan, Q., Yang, B., Wang, W., Wang, B., Chen, P., & Zhang, J. (2020).
models-based plant disease detection: A review of recent research. Apple leaf diseases recognition based on an improved convolutional
Frontiers in Plant Science, 14, 1158933. neural network. Sensors, 20(12), 3535.
Singh, A., Sonparote, D., Talekar, N., & Kale, D. (2020). A comparative Yao, N., Ni, F., Wang, Z., Luo, J., Sung, W. K., Luo, C., & Li, G. (2021).
study on crop disease detection using deep learning techniques. Inter- L2MXception: An improved Xception network for classification of
national Research Journal of Engineering and Technology (IRJET), peach diseases. Plant Methods, 17(1), 1–13.
7, 3240–3245. Yu, H. J., & Son, C. H. (2019). Apple leaf disease identification through
Singh, D., Jain, N., Jain, P., Kayal, P., Kumawat, S., & Batra, N. (2020). region-of-interest-aware deep convolutional neural network. https://
PlantDoc: A dataset for visual plant disease detection. Proceedings doi.org/10.48550/arXiv.1903.10356
of the 7th ACM IKDD CoDS and 25th COMAD (pp. 249–253). Zeng, T., Li, C., Zhang, B., Wang, R., Fu, W., Wang, J., & Zhang, X.
Association for Computing Machinery. (2022). Rubber leaf disease recognition based on improved deep con-
Singh, G. B., Rani, R., Sharma, N., & Kakkar, D. (2021). Identifica- volutional neural networks with a cross-scale attention mechanism.
tion of tomato leaf diseases using deep convolutional neural networks. Frontiers in Plant Science, 13, 829479.
International Journal of Agricultural and Environmental Information Zhang, D., Yang, H., & Cao, J. (2021). Identify apple leaf diseases
Systems (IJAEIS), 12(4), 1–22. using deep learning algorithm. https://doi.org/10.48550/arXiv.2107.
Singh, U. P., Chouhan, S. S., Jain, S., & Jain, S. (2019). Multilayer con- 12598
volution neural network for the classification of mango leaves infected Zhang, K., Xu, Z., Dong, S., Cen, C., & Wu, Q. (2019). Identification
by anthracnose disease. IEEE Access, 7, 43721–43729. of peach leaf disease infected by Xanthomonas campestris with deep
learning. Engineering in Agriculture Environment and Food, 12(4), Zhu, H., Chu, B., Zhang, C., Liu, F., Jiang, L., & He, Y. (2017).
388–396. Hyperspectral imaging for presymptomatic detection of tobacco dis-
Zhang, K., Zhang, C. L., & Wu, Q. (2019). Identification of cherry ease with successive projections algorithm and machine-learning
leaf disease infected by Podosphaera pannosa via convolutional neu- classifiers. Scientific Reports, 7(1), 4125.
ral network. International Journal of Agricultural and Environmental
Information Systems, 10(2), 98–110.
Zhang, S., Huang, W., & Zhang, C. (2019). Three-channel convolutional
neural networks for vegetable leaf disease recognition. Cognitive How to cite this article: Bagga, M., & Goyal, S.
Systems Research, 53, 31–41. (2024). Image-based detection and classification of
Zhang, X., Han, L., Dong, Y., Shi, Y., Huang, W., Han, L., González, plant diseases using deep learning: State-of-the-art
M. P., Ma, H., Ye, H., & Sobeih, T. (2019). A deep learning-based
review. Urban Agriculture & Regional Food Systems,
approach for automated yellow rust disease detection from high-
9, e20053. https://doi.org/10.1002/uar2.20053
resolution hyperspectral UAV images. Remote Sensing, 11(13), 1554.
Zhang, X., Qiao, Y., Meng, F., Fan, C., & Zhang, M. (2018). Identifica-
tion of maize leaf diseases using improved deep convolutional neural
networks. IEEE Access, 6, 30370–30377.

Urban Agric Regnl Food Sys - 2024 - Bagga - Image‐based detection and classification of plant diseases using deep

Uploaded by

Copyright:

Available Formats

Urban Agric Regnl Food Sys - 2024 - Bagga - Image‐based detection and classification of plant diseases using deep

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Urban Agric Regnl Food Sys - 2024 - Bagga - Image‐based detection and classification of plant diseases using deep

Uploaded by

Copyright:

Available Formats

Received: 11 August 2023 Accepted: 16 January 2024

Image-based detection and classification of plant diseases using

Manju Bagga1 Sonali Goyal2

1 MMEC, MMICTBM, Maharishi

1 INTRODUCTION monitoring in addition to prompt and accurate disease diagno-

Urban Agric Region Food Syst. 2024;9:e20053. wileyonlinelibrary.com/journal/uar2 1 of 21

As a result, practical, dependable, and cost-effective auto-

(DL) approach was introduced as a subclass of ML. In 1998, 3 SYSTEMATIC PLANNING OF

Abbreviation: CNN, convolutional neural network.

FIGURE 1 Flowchart of selection of studies.

based on three-channel CNN (TCCNN). Instead of using

TA B L E 2 Dataset information of different plants.

References Crop name Data sources Number of images Image type

Abbreviation: RGB, red green blue.

FIGURE 5 Most prevalent Frameworks in convolutional neural networks (CNNs).

From LeNet-5’s debut in 1998 until the present, numerous

accurately classifies plant fine-tuning

You might also like