Received: 12 August 2023 | Revised: 20 May 2024 | Accepted: 11 August 2024 | Published online: 19 August 2024

Journal of Data Science and Intelligent Systems

RESEARCH ARTICLE 2024, Vol. 00(00) 1–8
DOI: 10.47852/bonviewJDSIS42021530

Early Detection of Banana Leaf Disease Using

Novel Deep Convolutional Neural Network
N. R. Rajalakshmi1, S. Saravanan2,*, J. Arunpandian3, Sandeep Kumar Mathivanan4, Prabhu Jayagopal5,
Saurav Mallik6,7 and Guimin Qin8
1
Department of Computer Science and Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D
Institute of Science and Technology, India
2
Department of Electronics and Communication Engineering, Vel Tech Rangarajan
Dr. Sagunthala R&D Institute of Science and Technology, India
3
Department of Computer Science and Engineering, Vellore Institute of Technology, India
4
School of Computer Science and Engineering, Galgotias University, India
5
School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, India
6
Department of Environmental Health, Harvard T H Chan School of Public Health, USA
7
Department of Pharmacology & Toxicology, The University of Arizona, USA
8
School of Computer Science and Technology, Xidian University, China

Abstract: One of the most widely grown commercial commodities in India is the banana tree, which has important cultural and gastronomic
significance in tropical and subtropical areas where banana leaves are widely used for food delivery and packaging in a variety of cultures.
Regrettably, the incidence of diverse ailments that damage banana leaves present a significant risk to total output, therefore having an instant
effect on the country’s economy. To meet this issue, more efficient monitoring systems must be put in place, and control techniques for early
illness and pest detection must be developed. Using pest indicators makes this proactive strategy easier. With the successful use of these
approaches in a variety of industries, recent advances in agricultural technology have seen the incorporation of deep convolutional neural
networks (DCNN) for disease identification in numerous crops. This study’s main goal is to put into practice a DCNN that is especially
designed to anticipate various illnesses and pest occurrences in banana leaves. Through the use of DCNN, farmers may get vital insights
to apply fertilizers sparingly during the early phases, hence preventing the advent of leaf diseases. Remarkably, the suggested approach,
which uses a convolutional neural network (CNN) for accurate banana leaf disease detection, exhibits an astounding 99% accuracy when
compared to other deep learning techniques. By offering a reliable and precise technique for predicting pest and disease in banana crops,
this study advances agricultural practices. The use of state-of-the-art technologies, like CNN and DCNN, highlights the potential
revolutionary influence on disease control in banana farming, promoting increased yield and sustainable farming methods.
Keywords: deep convolutional neural network, disease prediction, fertilizers

1. Introduction in vitro studies, clinical trials, and animal model testing have
highlighted the therapeutic properties found in different sections
For most developing countries, agriculture is their main source of bananas [1]. These properties make them potential treatments
of income, and the banana sector is one of the most important in the for conditions such as diabetes, high blood pressure, cancer,
world of agribusiness. Three separate genera comprise the ulcers, diarrhea, urolithiasis, Alzheimer’s disease, and infections.
Muscaceae family: Musa, Musella, and Ensete. Bananas are Additionally, bananas find applications in medical fields like
members of the genus Musa, and they have been grown for a very surgical dressing, pain therapy, nanomedicine, pollution control,
long time in many different forms all over the globe. They have apoptosis, and the cell cycle. India is the world’s largest producer
several uses in the food sector and alternative health. Several of bananas, followed by China and Indonesia, according to the
Food and Agriculture Organization of the United Nations [2].
*Corresponding author: S. Saravanan, Department of Electronics and
Communication Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D Agribusiness data indicate that the Philippines, Argentina,
Institute of Science and Technology, India. Email: [email protected] Ecuador, and Indonesia are significant contributors to the world’s

© The Author(s) 2024. Published by BON VIEW PUBLISHING PTE. LTD. This is an open access article under the CC BY License (https://creativecommons.org/
licenses/by/4.0/).

01
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

agricultural output, making up over 20% of total production. In spite system. Author found that it is important to build intelligent
of this, smallholders control an astounding 85% of farms worldwide systems that can identify plant ailments based on visual symptoms
[3]. These farmers face a variety of biotic and abiotic obstacles. and external appearances similar to human behavior. An artificial
Bananas are the most important fruit and a basic food in the intelligence software for smartphones may be able to provide
Asia-Pacific area. However, a number of pests and illnesses have farmers with early warnings, speed up illness diagnosis, and
been linked to significant productivity losses in producing regions perhaps stop the spread of diseases and pests. As a result, the
[4]. This emphasizes how urgently comprehensive plans and importance of early plant disease identification is highlighted.
solutions are needed to lessen the negative effects of these issues This research constantly places a great deal of focus on deep
on banana farming, protect smallholders’ livelihoods, and learning methodologies. Additionally, this study provides a solid
guarantee the long-term viability of this vital industry. Crop foundation for a great deal of possible future research that will
damage brought on by previously infected plant diseases, bacteria, examine the whole history of the banana. This study’s main goal
fungus, roundworms, and nutritional deficiencies resulting in is to use a deep convolutional neural network (DCNN) to forecast
lower agricultural productivity and worse quality [5]. It also kills various illnesses and pests that harm banana leaves. By offering
the plant, reduces farmer earnings, and raises production expenses early diagnosis, the DCNN acts as a model to assist farmers,
during the control period. However, black leaf stripe diseases, also enabling them to apply the necessary fertilizers as soon as leaf
known as black Sigatoka disease or BLSD, are brought on by the diseases appear. This proactive strategy helps to both avoid and
fungus Pseudocercospora fijiensis and have a significant impact lessen the effects of illnesses on banana crops. This study’s
on banana yield. BLSD is usually regarded as the deadliest of the framework is set up as follows: A variety of cutting-edge
several deadly diseases that threaten bananas worldwide [6]. techniques for banana leaf disease prediction models are covered
After BLSD’s biotrophic phase, a necrotrophic stage with in Section 2. The operation of the suggested model for the
overt symptoms occurs. According to Bhuiyan et al. [7], the identification and categorization of leaf diseases is described in
illness results in a reduction in chlorophyll production in the Section 3. Next, the experimental findings and discussion are
photosynthetic portions of the leaf, which affects the structure of presented in Section 4, and the Conclusion is presented in Section 5.
banana leaves. The early indications of the illnesses are little
black dots on the leaf surface that later develop into thin, brown 2. Related Work
lines 2 to 3 mm long. The adaxial surface of affected leaves also
displays similar lines [8]. The early signs of necrosis are shown Plant disease detection systems that are automatic use photos of
by the stripes eventually becoming black and merging together as afflicted leaves as input to provide an accurate diagnosis of the
the illness progresses. Defoliation and early fruit ripening are illness. The system employs two distinct feature extraction
caused by dead leaves that have dried up. The banana plant’s techniques: first, it takes pictures of diseased leaves and processes
leaves display signs of many illnesses to which it was susceptible them using image processing techniques to extract data from
[9]. Streak Virus, Panama Virus, Black Sigatoka Virus, Yellow them. Using an image processing technique, features including
Sigatoka Virus, and Banana Bunchy Top Virus are the names of texture, shape, color, histogram of gradient (HOG), SURF, and
these diseases. The biotrophic phase, also referred to as the many more are retrieved. The HOG is used to provide information
absence of symptoms, may last for many weeks. According to in the leaf disease diagnostic system developed by Arman et al.
Bhuiyan et al. [7], banana plants may have already sustained [10]. In order to detect plant diseases, author used textural and
severe damage before the illness’s first symptoms appear, which gradient data in conjunction. They examine several different
might result in production losses of up to 85% [10]. Weeks may feature selection algorithms by applying them to a dataset
be used to quantify damage to banana plants. One hundred comprising turmeric crop data. The author claims that feature
percent or more of the world’s banana export and production selection using the Information Gain technique and SVM
might be lost due to illnesses and other climate factors that affect produces superior levels of success in this study including
banana plants. Xanthomonas wilt, bunchy top virus, fusarium wilt principal component analysis (PCA), Information Gain, and
(also called Panama wilt), and black Sigatoka are the four main Relief-f attribute methods. For the research done by Aliff et al.
illnesses that may affect bananas. Thus, the first and most [14] to identify diseases affecting banana leaves, the image
important step in the process is the early detection of diseases and processing method is advised. Their RGB color models are
pests in the field [11]. Agricultural extension workers are involved acquired and then transformed into HSI color models. The HSI
in traditional procedures for identifying pests and leaf diseases, picture is then subjected to preprocessing, threshold-based image
but their efficacy is limited in underdeveloped countries owing to segmentation, and histogram equalization [15]. Three more
a lack of human infrastructure. Asserts that smallholder farmers classifiers are then compared to the classification: PCA, support
are unable to handle the challenges of farming successfully vector machines, and backpropagation neural networks. The
because they do not have access to enough empirical data [12]. author developed a method for identifying leaf disease with an
For farmers, recognizing, evaluating, and managing plant diseases 89% accuracy value by using the SVM classifier and the Gabor
in the fields is an interesting and useful application. To the wavelet transform to extract features. It is suggested in Probojati
maximum degree practical, this procedure should be completed as et al. [16] to use an artificial neural network for the detection and
automatically and effectively as possible, at the lowest possible classification of banana leaf disease. The picture is first acquired
cost. Early detection of pests or agricultural diseases facilitates and pre-processed in the suggested technique, after which the
treatment and reduces adverse effects on food supply chains [13]. color and Histogram of Template feature are retrieved. After
Diseases that impact banana crops may be recognized and training the data set using the artificial neural network, grading is
categorized using an image processing method. It is now feasible done. The query photographs are then assessed based on the total
to automatically analyses and detect early stages of infection and percentage of the affected area. Finally, the picture is categorized
signs of main banana diseases on leaf surfaces by using computer based on the kind of illness. Numerous recent research has shown
vision and machine vision technologies. The methods used for that deep learning techniques may be used to tackle challenges
image processing are essential to the interpretation of data in this related to agricultural disease diagnosis. The LeNet architecture

02
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

deep learning model is used by Mduma et al. [17] to detect banana To create the feature map, different sized convolutional kernels or
leaf disease. Deep learning models using LeNet architecture perform filters are used to gain different picture attributes. The kernel size
well in a wide range of picture conditions, including ones with step size might be the cause of the information loss. Each neuron in
complicated backgrounds and different sizes and orientations. After the convolutional layer receives just a tiny portion of the outputs
25 epochs, the network stabilizes and achieves a high degree of from the previous layer after they have been convolved using a
accuracy. To identify tomato illness, a potent real-time deep “kernel.” A neuron’s “receptive field” is the range of output values
learning algorithm is used [18]. To identify leaf diseases, Vinta that it is capable of seeing. The second main component is called
et al. [19] used the Alexnet and Googlenet transfer learning models. the “pooling layer.” From each group of the outputs from the
DCNNs were used in Devi et al. [20] to identify maize leaf previous layer, it produces a single neuron. Commonly used
diseases. The deep learning transfer learning and SVM classifier pooling methods include max pooling and average pooling.
approach are used by Ridhovan et al. [21] for early banana disease A unique procedure called pooling is used to minimize the
detection. In this instance, images obtained by hyperspectral remote dimension of the feature map and is useful in the implementation of
sensing are used [22]. The classifiers’ overall performance is downsampling in CNNs. More discriminative and practical
examined, and the average accuracy is determined using spectral characteristics are progressively added to the layers. The layers
and morphological data. About 96% of cases are detected early, reduce the spatial complexity of the employed parameters and solve
90% of cases are detected in the middle, and 92% of cases are the overfitting issue. CNN is composed of eight convolutional
detected late. The distribution of these percentages is as follows: layers and eight pooling layers in this case. The normalization layer
96% accuracy is achieved with early detection [23]. is employed in between to improve training and reduce the
network’s reliance on initialization. Normalization is applied to the
3. Methodology gradient values that traverse the network. An average pooling layer
averages the input data by taking the mean.
Multilayer convolutional neural networks (CNN) are an effective Specific parameters are assigned to each layer in the CNN
way to extract a large number of picture features. The picture architecture that has been discussed. Convolutional layer 1 (Conv
is transformed into a digital image in order to produce a two- 2D1) has 32 filters with a 3 × 3 filter size, a stride value of 2, and
dimensional matrix. Subsequently, the corresponding convolutional no padding. The pooling layer (Pool1) then applies a 2 × 2 filter
and two-dimensional matrix undergo de sampling. In order to with a stride of 1. This pattern is repeated in the second
summarize the presence of features in the input, the CNN will convolutional layer (Conv 2D2), with a stride value of 2 and a
analyze the local components of the input using a series of filters. filter size of 3 × 3. There are 64 filters in total. A 2 × 2 filter with

Figure 1
Sample leaves for the banana leaf dataset
healthy sigatoka sigatoka

cordana sigatoka healthy

healthy sigatoka sigatoka

03
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

a stride of 1 is maintained by the equivalent pooling layer (Pool2). Figure 2

With a gradual rise in the number of filters (256) in the final Proposed multi-layered DCNN architecture
convolutional layer, this convolutional and pooling process is
carried out via Conv 2D8 and Pool8. The neural network design
facilitates hierarchical feature extraction and spatial reduction via
the use of stride values of 1 for pooling layers and 2 for
convolutional layers, in addition to alternating padding values of
0 and 1. The layered structure of the DCNN model is seen in
Figure 2. The input picture was sent as a tensor with width,
height, and channel dimensions of 224 × 224 × 3 to the DCNN
model. Because it and the CNN layer share an input and an
output tensor, the sequential layer is chosen. Eight convolutional
layers were followed by the addition of a max pooling layer. The
feature dimensions were compressed using max pooling with a
stride of 1. The most recent max pooling layer provided input to
the global average pooling layer, which averages each feature map
across its spatial dimensions. This reduces the dimension number
of the model and avoids overfitting. Following the global average
pooling layer, the output is linked to an output layer and a fully
connected layer, which use the previously learned features to
conduct classification. The fully connected layer’s neuron count is
based on the number of classes in the classification job, and the
output layer generates the network’s final classification result.
The DCNN model architecture is generally well suited for image
classification tasks, where the goal is to identify the objects
contained in the input picture. A paradigm change in agriculture
is brought about by the use of DCNN for disease detection,
especially in the context of banana leaf diseases. In contrast to
conventional techniques, DCNNs provide unmatched precision
by identifying complex patterns in pictures and guaranteeing
accurate illness detection. Real-time or almost real-time illness
diagnosis is made possible by DCNNs’ automated analytical
capacity, which speeds up the detection process and is
essential for prompt action. DCNNs are skilled at identifying
novel disease strains because to their flexibility and learning
capacities, and the non-intrusive nature of image analysis lowers
the possibility of further contamination during detection. Once
taught, DCNNs’ cost-effectiveness and ability to provide
data-driven insights make them a transformational tool for
farmers, encouraging proactive management, early diagnosis,
and eventually improved crop health.

4. Results and Discussion

This section presents the results of a comprehensive
examination of the importance of the findings on the early
identification of banana leaf disease using the innovative DCNN.
The dataset distribution for different classes is shown in Table 2,
where each row represents a distinct category. The number of
genuine photographs used to train the model is shown in the table
under “Training images,” with Cordana having 86, Healthy
having 134, Pestalotiopsis having 131, and Sigatoka having 422
real training images. The number of augmented photos used for
validation during the training phase is shown in the “Validation”
column; there are 1000 enhanced images for each class. Finally,
the number of photos reserved for assessing the model after
training is specified in the “Testing” column; for each class, 22
genuine and 22 augmented images are assigned. In order to
facilitate a full assessment of the model’s performance, this
structured representation offers a comprehensive overview of the banana leaf disease classes, such as Sigatoka, Cordana, Healthy,
dataset’s composition, comprising actual and augmented photos and Pestalotiopsis. To increase the model’s resilience, the dataset
for training, validation, and testing across multiple leaf disease included 88 enhanced photos in addition to the 768 original
classes. The research made use of a carefully selected dataset of images, as shown in Table 1. Table 2 illustrates the remarkable

04
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

Figure 3
Training performance and validation performance

Figure 4
Performance of proposed DCNN

precision of the DCNN model in recognizing disease classes and 94.134%, 95.112%, 91.267%, 95.865%, and 93.171%,
differentiating between healthy leaves. respectively. These results highlight the superior accuracy of the
With an average accuracy of 0.99, Table 3’s performance proposed DCNN, which achieved an accuracy of 98.92%.
metrics for the DCNN model demonstrate excellent accuracy Moreover, the proposed DCNN excels not only in accuracy but
across all classes. Robust precision and recall metrics indicate also in precision, recall, and F1-Score. It achieved a precision of
accurate classification of specific illness groups. F1-scores 99.13%, indicating a high level of confidence in the positive
between 0.991 and 1 demonstrated the model’s reliable performance. predictions made by the model. In comparison, other models
The proposed DCNN achieved a high accuracy of 98.92%, reported by the authors had precision rates of 91.045% by
outperforming other models in the study. Specifically, the Thiagarajan et al. [2] and 92.094% by Bhuiyan et al. [7].
proposed DCNN achieved an impressive accuracy of 98.92%, Additionally, the high recall rate of 99.43% for the proposed
which surpasses the accuracies of other models reported by DCNN suggests that the model is highly effective in correctly
various authors. For instance, the model presented by Thiagarajan identifying all relevant instances in the dataset. This is notably
et al. [2] reached an accuracy of 93.045%, while the models from higher than the recall rates reported by models in Thiagarajan
Bhuiyan et al. [7], Wan and Yao [11], Chowdhury et al. [19], et al. [2] with 89.142% and in Bhuiyan et al. [7] with 91.110%.
Chinpanthana [22], and Demilie [23] achieved accuracies of The F1-Score, which combines precision (the ability to make

05
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

Figure 5
Comparison of proposed DCNN performance with existing techniques

Table 1 Table 4
Deep convolutional neural network (DCNN) Class-wise performance of proposed 8C-DCNN
Conv Conv Conv Mode Acc Pr Recall F1-score
Parameters 2D1 Pool1 2D2 Pool2 : : : 2D8 Pool8 Leaf rust 0.9886 1 0.9821428 0.9909
Size of filter 3×3 2×2 3×3 2×2 3×3 2×2 Healthy 0.9886 0.9814814 1 0.9909
Filters used 32 64 256 Leaf spot 0.9886 1 0.9830508 0.99145
Stride_value 2 1 2 1 2 1 Leaf blight 0.9886 0.9827586 1 0.99130
Padding 0 1 0 1 0 1

Table 5
Table 2
Comparison of proposed DCNN performance
Data description of classes
with existing techniques
Training images
Accuracy Precision Recall F1-score
ID Name Real Augmented Validation Testing Authors (%) (%) (%) (%)
1 Cordana 86 1000 22 22 [2] 93.045 91.045 89.142 90.121
2 Healthy 134 1000 22 22 [7] 94.134 92.094 91.110 91.123
3 Pestalotiopsis 131 1000 22 22 [11] 95.112 91.078 92.067 89.98
4 Sigatoka 422 1000 22 22 [19] 91.267 89.248 88 88.180
[22] 95.865 94.232 94.112 94.152
[23] 93.171 91.142 88.984 88.997
Table 3 Proposed 98.92 99.13 99.43 99.113
Test result of proposed DCNN layered architecture DCNN

TP TN FP FN
Leaf rust 54 31 0 1
Healthy 52 34 1 0
Leaf spot 57 30 0 1 positive predictions (precision) and finding all relevant instances
Leaf blight 56 36 1 0 (recall). In contrast, the other models exhibited lower F1-Scores,
such as 90.121% by Thiagarajan et al. [2] and 91.123% by
Bhuiyan et al. [7], underscoring the superior capability of the
proposed DCNN in providing reliable and robust predictions.
accurate positive predictions) and recall (the ability to find all Overall, these results demonstrate that the proposed DCNN
relevant instances), further emphasizes the efficacy of the outperforms existing models in accuracy, precision, recall, and
proposed DCNN, with a score of 99.113%. This score indicates F1-Score, making it a promising solution for tasks requiring high
that the model has a good balance between making accurate performance and reliability.

06
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

The result is important for the agricultural industry because it References

allows for prompt disease management strategies, reduced crop
losses, and early disease detection. Although the DCNN model for [1] Bakache, A., Douzals, J. P., Bonicelli, B., Cotteux, E., De
the early detection of banana leaf disease has shown promising Lapeyre de Bellaire, L., & Sinfort, C. (2019). Development
results, it still needs to be expanded, transfer learning methods of a rapid methodology for biological efficacy assessment in
used, and real-time implementation. banana plantations: Application to reduced dosages of
contact fungicide for Black Leaf Streak Disease (BLSD)
5. Conclusion control. Pest Management Science, 75(4), 1081–1090.
[2] Thiagarajan, J. D., Kulkarni, S. V., Jadhav, S. A., Waghe, A. A.,
Due to its many applications, banana leaves are essential in Raja, S. P., Rajagopal, S., : : : , & Subramaniam, S. (2024).
the cuisines of many tropical and subtropical nations, which Analysis of banana plant health using machine learning
emphasizes the banana tree’s significance in India’s commerce. techniques. Scientific Reports, 14(1), 15041.
That being said, there is a significant problem with the [3] Kavitha, T., Deepika, S., Nattaraj, K., Shanthini, P., &
prevalence of banana leaf diseases, which puts the country’s Puranaraja, M. (2022). Smart system for crop and diseases
economic productivity at risk. In order to curb the spread of prediction using random forest and resnet architecture. In
illnesses and pests, it becomes imperative to install pest 2022 International Conference on Sustainable Computing
indicators for efficient monitoring and management of these and Data Communication Systems, 1513–1519.
diseases. The apparent efficacy of DCNN in several domains [4] Fuentes, A., Yoon, S., Kim, S. C., & Park, D. S. (2017). A
highlights their possible relevance in the area of agriculture. robust deep-learning-based detector for real-time tomato
The main goal of this research was to use DCNN technology to plant diseases and pests recognition. Sensors, 17(9), 2022.
anticipate illnesses and pests that affect banana leaves in [5] Narayanan, K. L., Krishnan, R. S., Robinson, Y. H., Julie, E. G.,
advance. To lessen the effect of leaf diseases on agricultural Vimal, S., Saravanan, V., & Kaliappan, M. (2022). Banana plant
output, a proactive strategy is introduced using the suggested disease classification using hybrid convolutional neural network.
DCNN model. One of the main components of this plan is Computational Intelligence and Neuroscience, 2022(1),
providing farmers with the tools they need to administer 9153699.
essential treatments, including the right fertilizers, at the earliest [6] Hassan, S. I., Alam, M. M., Illahi, U., Al Ghamdi, M. A.,
stages of disease development. Interestingly, the study’s results Almotiri, S. H., & Su’ud, M. M. (2021). A systematic
show an astounding 99% accuracy rate, outperforming similar review on monitoring and advanced control strategies in
deep learning techniques. The proposed method, which makes smart agriculture. IEEE Access, 9, 32517–32548.
use of a CNN, expedites the process of identifying illnesses in [7] Bhuiyan, M. A. B., Abdullah, H. M., Arman, S. E., Rahman,
banana leaves and produces encouraging results. This S. S., & Al Mahmud, K. (2023). BananaSqueezeNet: A very
development not only protects banana crops but also advances fast, lightweight convolutional neural network for the
the agricultural industry by using state-of-the-art technologies diagnosis of three prominent banana leaf diseases. Smart
to improve disease control and guarantee sustained yield. Agricultural Technology, 4, 100214.
[8] Luna-Moreno, D., Sánchez-Álvarez, A., Islas-Flores, I., Canto-
Ethical Statement Canche, B., Carrillo-Pech, M., Villarreal-Chiu, J. F., &
Rodríguez-Delgado, M. (2019). Early detection of the fungal
This study does not contain any studies with human or animal banana black sigatoka pathogen Pseudocercospora fijiensis by
subjects performed by any of the authors. an SPR immunosensor method. Sensors, 19(3), 465.
[9] Gandhi, S., Walunjkar, S., Choudhary, V., & Joshi, P. (2022).
Conflicts of Interest Banana leaf disease detection using deep learning approach.
International Journal for Research in Applied Science and
The authors declare that they have no conflicts of interest to
Engineering Technology, 10(12), 119–121.
this work.
[10] Arman, S. E., Bhuiyan, M. A. B., Abdullah, H. M., Islam, S.,
Chowdhury, T. T., & Hossain, M. A. (2023). BananaLSD: A
Data Availability Statement banana leaf images dataset for classification of banana leaf
Data sharing is not applicable to this article as no new data were diseases using machine learning. Data in Brief, 50, 109608.
created or analyzed in this study. [11] Wan, G., & Yao, L. (2023). LMFRNet: A lightweight
convolutional neural network model for image analysis.
Author Contribution Statement Electronics, 13(1), 129.
[12] Abdullah, H. M., Mohana, N. T., Khan, B. M., Ahmed, S. M.,
N. R. Rajalakshmi: Conceptualization, Software, Investigation, Hossain, M., Islam, K. S., : : : , & Ahamed, T. (2023). Present
Data Curation, Writing – Original Draft. S. Saravanan: and future scopes and challenges of plant pest and disease
Conceptualization, Software, Investigation, Data Curation, Writing – (P&D) monitoring: Remote sensing, image processing, and
Original Draft. J. Arunpandian: Validation, Formal analysis. artificial intelligence perspectives. Remote Sensing
Sandeep Kumar Mathivanan: Methodology, Writing – Review & Applications: Society and Environment, 32, 100996.
Editing, Supervision, Project administration. Prabhu Jayagopal: [13] Sangeetha, R., Logeshwaran, J., Rocher, J., & Lloret, J. (2023).
Software, Investigation, Resources. Saurav Mallik: Methodology, An improved agro deep learning model for detection of Panama
Writing – Review & Editing, Supervision, Project administration. wilts disease in banana leaves. AgriEngineering, 5(2), 660–679.
Guimin Qin: Resources, Data Curation, Visualization, Supervision, [14] Aliff, M., Luqman, M., Yusof, M. I., Sani, N. S., Syafiq, M. U.,
Project administration. Sadikan, S. F. N., & Mahmud, H. (2024). Utilizing aerial

07
Journal of Data Science and Intelligent Systems Vol. 00 Iss. 00 2024

imagery and deep learning techniques for identifying banana Automatic and reliable leaf disease detection using deep
plants diseases. In ITM Web of Conferences, 60. learning techniques. AgriEngineering, 3(2), 294–312.
[15] Krishnan, V. G., Deepa, J. R. V. P., Rao, P. V., Divya, V., & [20] Devi, S., Dand, H., Waghole, D., & Goriparthi, R. (2024).
Kaviarasan, S. (2022). An automated segmentation and Agriculture system for potato leaf disease detection using deep
classification model for banana leaf disease detection. Journal learning and GENAI. International Journal of Intelligent
of Applied Biology and Biotechnology, 10(1), 213–220. Systems and Applications in Engineering, 12(4), 1808–1816.
[16] Probojati, R. T., Listyorini, D., Sulisetijono, S., & Wahyudi, D. [21] Ridhovan, A., Suharso, A., & Rozikin, C. (2022). Disease
(2021). Phylogeny and estimated genetic divergence times of detection in banana leaf plants using densenet and inception
banana cultivars (Musa spp.) from Java Island by maturase method. Information Systems Engineering Articles, 6(5), 1–6.
K (mat K) genes. Bulletin of the National Research Centre, [22] Chinpanthana, N. (2023). Image classification of tomato leaf
45, 1–13. diseases using convolutional neural network. Journal of
[17] Mduma, N., & Leo, J. (2023). Dataset of banana leaves and Information Science and Technology, 13(2), 1–12.
stem images for object detection, classification and [23] Demilie, W. B. (2024). Plant disease detection and
segmentation: A case of Tanzania. Data in Brief, 49, 109322. classification techniques: A comparative study of the
[18] Vinta, S. R., Koshariya, A. K., & Gottimukkala, A. (2024). performances. Journal of Big Data, 11(1), 5.
Investigation of early symptoms of tomato leaf disorder by
using analysing image and deep learning models. EAI How to Cite: Rajalakshmi, N. R., Saravanan, S., Arunpandian, J., Mathivanan, S.
Endorsed Transactions on Internet of Things, 10. K., Jayagopal, P., Mallik, S., & Qin, G. (2024). Early Detection of Banana Leaf
[19] Chowdhury, M. E., Rahman, T., Khandakar, A., Ayari, M. A., Disease Using Novel Deep Convolutional Neural Network. Journal of Data
Science and Intelligent Systems. https://doi.org/10.47852/bonviewJDSIS42021530
Khan, A. U., Khan, M. S., : : : , & Ali, S. H. M. (2021).

08