Enhanced Tomato Leaf Disease Detection Using DenseNet201 with

Channel and Spatial Attention Mechanisms

Ramana S, Prathiksha V S and Ragul M

Department of Computer Science and Engineering, Kongu Engineering College, Erode, India

Keywords: Tomato Leaf Disease, Dense Net 201, Attention Mechanism, CNN, Disease Detection, Image Classification,

And Data Augmentation.

Abstract: This project presents a comprehensive system for tomato leaf disease detection, utilizing the DenseNet201

architecture enhanced with channel and spatial attention mechanisms. The system is designed to improve the

accuracy and reliability of disease classification, addressing limitations in traditional Convolutional Neural

Networks (CNNs), which previously achieved an accuracy of 95%. By incorporating attention mechanisms,

the proposed approach focuses on critical image features, boosting classification accuracy to 98.07%. The

model was trained on a dataset of 23,896 tomato leaf images across 10 distinct disease classes. The system

architecture also includes data augmentation techniques and robust optimization methods, ensuring the

model's generalization capability and performance. This project represents a significant step toward practical

applications in agriculture, offering an advanced tool for early disease detection, which can aid in more

effective crop management.

1 INTRODUCTION

Tomato leaf diseases significantly affect crop yield

and quality, posing a major challenge for farmers

worldwide. Accurate and early detection of these

diseases is critical for effective disease management

and minimizing economic losses. Traditionally,

disease identification has relied on manual inspection

by experts, which is not only time- consuming but also

prone to human error. In recent years, advancements

in deep learning have enabled the automation of

disease detection, offering a more accurate and

efficient alternative. Our project aims to develop a

deep learning-based system for detecting tomato leaf

diseases using a dataset comprising 23,896 images

across 10 disease categories, including Leaf Mold,

Target Spot, Bacterial Spot, Tomato Yellow Leaf Curl

Virus, and more. The dataset also includes healthy

leaves, ensuring a comprehensive range of

classifications. We began by testing several popular

deep learning models, including VGG19, Efficient

Net, and ResNet, to assess their performance in

identifying the various diseases. After rigorous

testing, DenseNet201 emerged as the top- performing

model, providing superior accuracy and feature

extraction capabilities. However, to further improve

the model’s performance, we integrated channel and

spatial attention mechanisms. These mechanisms help

the model focus on the disease-relevant areas of the

leaf. The introduction of channel attention allowed the

model to emphasize critical features, achieving a

validation accuracy of 97.21%. The spatial attention

mechanism enabled the model to highlight important

regions within the image, resulting in a higher

validation accuracy of 96.07%. When both attention

mechanisms were combined, the model achieved a

validation accuracy of 98.07%. The integration of

these attention mechanisms in our DenseNet201-

based system not only surpassed traditional models

like VGG19 and ResNet but also addressed the key

challenges of detecting subtle visual patterns across

different diseases. This system has real-world

potential in the domain of smart farming, offering

farmers a scalable and efficient solution for

monitoring and managing crop health more

effectively. Additionally, the model’s adaptability

makes it suitable for precision agriculture, where early

disease detection can significantly improve crop

management and reduce losses. Our research

highlights the promise of attention-enhanced deep

learning models in agriculture and opens up

opportunities for future advancements in plant disease

detection using AI-driven approaches.

S, R., V S, P. and M, R.

Enhanced Tomato Leaf Disease Detection Using DenseNet201 with Channel and Spatial Attention Mechanisms.

DOI: 10.5220/0013652000004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 749-756

ISBN: 978-989-758-763-4

749

2 RELATED WORKS

Attallah (Attallah, 2023) proposes a novel approach

for tomato leaf disease classification using image

preprocessing, transfer learning with ResNet-18,

ShuffleNet, and MobileNet, and hybrid feature

selection. Deep features are extracted, refined, and

classified using KNN and SVM, achieving 99.92%

and 99.90% accuracy with 22 and 24 features,

respectively, ensuring efficient and accurate disease

detection for improved agricultural productivity. A

small system containing two CNN models (CNN1

and CNN2) that only require 5.1 MB and 6 MB of

storage is shown by H. Ulutas and V. Aslantas

(Ulutaş, Aslantaş, et al. , 2023). Training periods are

greatly shortened, taking about 1.5 hours. The

ensemble achieves 99.60% accuracy using a dataset

of 18,160 pictures, fine-tuning, and PSO-based

optimization, allowing for the effective and timely

identification of tomato leaf diseases. A methodology

that uses the Otsu segmentation technique for image

processing and the Grey Level Co-Occurrence Matrix

(GLCM) method for feature extraction is put forth by

S. U. Rahman et al. The Support Vector Machine

(SVM) technique is used for classification, and it

achieves impressive accuracy rates. Effective tomato

leaf disease detection and diagnosis are made possible

by this combination of approaches. A. Guerrero-

Ibañez and A. Reyes-Muñoz(Ibañez, and, Muñoz,

2023) describe a CNN-based method for classifying

tomato diseases that outperforms a number of current

models, attaining 99.9% accuracy with good

precision, recall, and F1 scores. In order to improve

accuracy, precision, recall, and F1 scores, K. Roy et

al. [5] suggest a hybrid PCA Deep Net framework

that combines deep neural networks and machine

learning. It outperforms current algorithms in

identifying and categorising tomato leaf diseases,

achieving 99.25% validation accuracy after being

trained on the Plant Village dataset. A modified

InceptionResNet-V2 (MIR-V2) model using RPCA-

enhanced data and adjusted max-pool layers is

presented by P. Kaur et al. (Kaur, Harnal, et al. , 2023)

and achieves 98.92% accuracy. Precision, recall, and

F1-score are used to evaluate the system's overall

performance in plant disease detection, and it

surpasses pre-trained models. Four CNN

architectures (VGG-16, VGG-19, ResNet, and

Inception V3) were assessed by I. Ahmad et al. or the

categorisation of tomato leaf diseases. Using a self-

collected augmented field dataset of 15,216 photos

and a laboratory dataset of 2,364 images, Inception

V3 reached the best accuracy of 93.40% on the

laboratory dataset, however due to real-world

obstacles, its performance was lower on the field

dataset. Performance indicators like as F1-score,

recall, and precision demonstrated Inception V3's

greater performance in both datasets. To solve

misdiagnosis issues, S. G. Paul et al. (Paul, Biswas, et

al. , 2023) offer a method for tomato leaf disease

classification that uses a lightweight proprietary CNN

model and transfer learning-based models (VGG-16,

VGG-19). The algorithm achieved a 95.00%

accuracy and recall rate by applying data

augmentation and classifying eleven types, including

healthy leaves. The agricultural ecosystem benefited

from the implementation of the best-performing

model in web and Android applications, which

offered a comprehensive solution for early disease

identification and treatment options. Using CNN, S.

Z. Khan et al. were able to detect tomato leaf disease

in nine disease classes and one healthy class with an

average accuracy of 91.2%. This demonstrated the

system's efficacy in disease identification and

assisting with timely crop management,

outperforming pre-trained models such as VGG16,

InceptionV3, and Mobile Net. Huang et al. (Huang,

Chen, et al. , 2023) overcame complicated backdrops

by using the FC-SNDPN approach with VGG-16 for

automatic tomato leaf disease identification in

southern China. The solution outperformed

conventional CNN models and supported precision

agriculture with an accuracy of 95.40% using a

custom dataset and segmentation technique. A. Saeed

et al. (Saeed, Aziz, et al. , 2023) used pre-trained

CNNs, Inception V3 and Inception ResNet V2,

trained on a dataset of 5,225 pictures, to create a

tomato leaf disease detection system. With dropout

rates between 5% and 50%, the models' accuracy was

99.22%; Inception V3 and Inception ResNet V2

performed best at 50% and 15%, respectively. This

method has great promise for the identification of

agricultural diseases since it is successful in

differentiating between healthy and sick tomato

leaves. In order to improve feature extraction, T.

Sanida et al. (Sanida, Sideris, et al. , 2023) integrate

the first ten convolutional layers of VGG16 with

inception blocks in their deep learning system for

tomato leaf disease identification. The model pre-

trains on ImageNet and fine-tunes on a tomato leaf

dataset as part of a two-stage transfer learning

process. The training imbalance between classes is

addressed with an enhanced categorical cross-entropy

loss function. The technology outperforms other

cutting-edge methods with an accuracy of 99.23%.

Using 18,160 photos, Ulutaş and Aslantaş (Ulutaş and

Aslantaş, 2023) suggest an ensemble CNN model for

tomato leaf disease detection. They achieved 99.60%

INCOFT 2025 - International Conference on Futuristic Technology

750

accuracy by using grid search, fine-tuning, and

hyperparameter optimisation with particle swarm

optimisation. To stop new infections, the method

offers quick and effective early disease identification.

MATERIALS AND METHODS

The proposed work aims to enhance the detection and

classification of tomato leaf diseases by integrating

both spatial and channel attention mechanisms within

the DenseNet201 architecture. This approach enables

the model to effectively focus on relevant features

while suppressing irrelevant information, which is

crucial for accurate diagnosis. The spatial attention

mechanism helps the model identify and prioritize

important regions within the input images,

concentrating on areas where disease symptoms are

most prominent. In parallel, the channel attention

mechanism evaluates the significance of different

feature maps generated by the convolutional layers,

allowing the model to emphasize informative

channels that carry critical information about disease

characteristics while diminishing less relevant

channels. This dual mechanism not only improves the

robustness of the model’s feature representation but

also enhances its ability to adapt to variations in leaf

appearance and environmental conditions, such as

lighting and background noise. The model will be

trained on a curated dataset consisting of tomato leaf

images, with multiple classes representing various

diseases and a healthy class. The training process will

utilize specific parameters, including 50 epochs, an

initial learning rate of 1e-3, a batch size of 32, and

images resized to 128x128 pixels. By leveraging

these attention mechanisms and utilizing the

DenseNet201 architecture, the proposed work seeks

to achieve higher accuracy in disease classification,

ultimately contributing to improved agricultural

outcomes and assisting farmers in timely disease

management.

3.1 Dataset

A well-organized and diverse dataset is essential for

effectively evaluating tomato leaf disease detection

systems. In this project, we leveraged both

laboratory-based and field- based datasets to enhance

the robustness of our model under varied conditions.

The laboratory dataset consists of 23,896 high-

resolution images of tomato leaves, meticulously

categorized into 10 distinct classes representing

various diseases, including early blight, late blight,

and bacterial spot. This dataset was strategically

divided into two parts: 23,896 images were allocated

for training (79.46%), while 5,724 images were set

aside for validation (20.54%) shown in Fig (1). This

division not only ensures sufficient data for

comprehensive model training but also allows for

accurate evaluation of the model's performance. The

balanced approach to dataset creation and partitioning

helps in preventing overfitting and enhances the

model's ability to generalize to unseen data. A

detailed summary of this dataset, including the

distribution of images across classes, is provided in

Table I, illustrating the thorough preparation and

thoughtful curation that underpin this research. This

robust dataset is expected to contribute significantly

to the development of a reliable tomato leaf disease

detection system, capable of performing well in both

controlled laboratory environments and real-world

agricultural settings.

Table 1: Dataset summary for tomato leaf disease

S.No Types of Tomato

Leaf Disease

Training Validation

)

Bacterial S

ot 2826 643

Earl

bli

ht 2988 512

)

Late bli

ht 2455 792

d) Septoria leaf spot 2754 746

e) Spider mites 2882 435

f) Target spot 1747 457

g) Tomato yellow leaf

curl virus

2036 498

h) Tomato mosaic

virus

2153 584

)

Health

3051 805

j) Powdery mildew 1004 252

Total 23896 5724

Figure 1: Sample images for each class of tomato leaf

4 PROPOSED WORKS

4.1 Efficient Net

Efficient Net is an extremely efficient convolutional

neural network that improves performance through

compound scaling, depth, width, and resolution

Enhanced Tomato Leaf Disease Detection Using DenseNet201 with Channel and Spatial Attention Mechanisms

751

balance. It reduces computational costs without

losing accuracy through the use of depth-wise

separable convolutions and inverted residual blocks.

The properties of this network make it a strong feature

extractor, especially in tasks like tomato leaf disease

detection, which are computationally intensive.

Despite its good performance, Efficient Net lacks

mechanisms to focus on disease-relevant regions,

which makes it not sensitive enough to detect subtle

patterns.

4.2 VGG19 (Visual Geometry Group)

VGG19 is an effective model for tomato leaf disease

detection because of its deep architecture of 19 layers,

which enables it to capture features at multiple levels.

It can identify low-level patterns such as edges,

textures, and leaf shapes, as well as high-level

features that distinguish between healthy and

diseased leaves. The use of small 3×3 convolutional

filters ensure parameter efficiency, allowing the

model to handle large datasets of tomato leaf images

without significant computational overhead. With

pre-trained ImageNet weights, VGG19 adapts

quickly to specific datasets through transfer learning,

streamlining the training process. Its ability to

generalize across diverse datasets and detect subtle

differences in leaf health makes it a valuable tool for

disease classification, although it lacks specialized

mechanisms to focus on disease-relevant regions.

4.3 Channel Attention Mechanism

This paper deals with tomato leaf disease detection

using a DenseNet201 model that has been improved

by the inclusion of a Channel Attention Mechanism.

The dense connectivity between layers helps in

feature extraction using DenseNet201, as the model

is able to reuse features from earlier layers and hence

helps extract deep and relevant features from the

images. Through channel attention, the model is able

to improve performance by shifting focus to the most

pertinent feature maps while dampening those with

low significance. Two operations are utilized in this

mechanism: GAP and GMP (Global Max Pooling)

which are used to pool down the channel information

to a summary. GAP takes the average value of each

channel, and GMP takes the maximum value, both of

which help in learning the importance of each

channel. These summaries are passed through dense

layers to generate attention weights, which are

squashed using a Sigmoid function, providing a

mechanism for prioritizing useful channels. The

model uses Adam optimizer with the learning rate at

1e-3 that balances speed and stability in convergence;

thus, efficient training is attained. It has a

ReduceLROnPlateau callback used to dynamically

alter the learning rate, thus averting overshooting the

optimal weights and improving training efficiency.

This was to prevent overfitting, by applying the usual

data augmentation such as rotation, zoom, shifts, and

flips that would allow the model to generalize better

by exposing it to various transformations of the

images. The categorical cross-entropy loss function

was selected because the loss function well addresses

multi-class classification problems, thus the model

can be able to manage the multiple classes with their

distinct labels.

Equation 1 illustrates the Channel Attention

Mechanism, where 𝐴



represents the attention

weight, 𝑊



and 𝑊



are learnable weight matrices,

𝐹



is the output from Global Average Pooling, 𝐹



Figure 2: Channel Attention Mechanism architecture

is the output from Global Max Pooling, and 𝛿 denotes the

ReLU activation function. By using this attention

mechanism, the model can selectively focus on the most

important features, which significantly enhances its ability

to accurately classify tomato leaf diseases, which is shown

in Fig. 2.

𝐴

𝑐

= 𝜎(𝑊

𝛿(

𝑊

𝐹

𝑎𝑣𝑔

) + 𝑊

𝛿

(

𝑊

𝐹

𝑚𝑎𝑥)

)

(1)

INCOFT 2025 - International Conference on Futuristic Technology

752

4.4. Spatial Attention Mechanism

The study focuses on developing a deep learning

model for detecting tomato leaf diseases, utilizing

DenseNet201 as the backbone, enhanced with

attention mechanisms like Channel Attention,

Squeeze-and-Excitation (SE) block, and Spatial

Attention. The model starts by preprocessing the

dataset, resizing images to 128x128 pixels,

normalizing pixel values, and converting labels into a

binarized format for multi-class classification. Data

augmentation techniques like rotation, shifting, shear,

zoom, and flipping are used to increase the robustness

of the model. DenseNet201 is used as a feature

extractor, with Channel Attention focusing on global

feature context, SE block for adaptive recalibration,

and Spatial Attention that highlights critical spatial

regions in the image. The Spatial Attention formula

(2) used in the model is given as:

𝐴



=Con

(𝐹)

(2)

where 𝐴



represents the spatial attention map, and F

is the input feature map. The convolution operation

extracts spatial patterns from the feature map, and the

Figure 3: Spatial Attention Mechanism architecture

Sigmoid activation function squashes the output

between 0 and 1, generating the attention map. The

model architecture is illustrated in Fig. 3, which

visualizes the integration of DenseNet201 with the

attention mechanisms. The model is compiled using

the Adam optimizer, and callbacks like learning rate

adjustment, early stopping, and model checkpointing

are included for efficient training. Evaluation metrics

include accuracy, loss, a classification report, and a

confusion matrix, providing insight into the model's

performance in classifying various tomato leaf

diseases.

4.5 Hybrid Attention Mechanism

The hybrid attention mechanism in your code

significantly enhances the performance of the tomato

leaf disease detection model by integrating channel

and spatial attention, allowing the model to focus on

important features and regions within the images. The

implementation processes images resized to 128x128

pixels and uses a dataset consisting of 23,896 images

across 10 classes. Data augmentation techniques,

such as random flips and rotations, are applied to

increase dataset variability and improve the model's

robustness against overfitting, leading to better

generalization on unseen data. The DenseNet201

architecture is selected for its capability to efficiently

extract features through its deep residual learning

framework, which allows for improved information

flow and addresses the vanishing gradient problem.

In your implementation, the channel attention

mechanism emphasizes important feature channels

using global average and max pooling, while the

spatial attention mechanism highlights critical spatial

regions by concatenating average and max pooled

features and applying a convolutional layer. This dual

approach enables the model to focus on both

significant feature channels and relevant spatial

details, enhancing overall detection accuracy. The

code also includes a visualization function that

displays the original images alongside the attention

maps generated from the model which is shown in Fig

(4), illustrating how the hybrid attention mechanism

directs focus toward crucial areas for detecting

tomato leaf diseases. The formula for Hybrid

Attention can be expressed as (3):

𝐹𝑓𝑖𝑛𝑎𝑙 = 𝐹 ∗ 𝜎(𝐹𝐶(𝐺𝐴𝑃(𝐹)) + 𝐹𝐶(𝐺𝑀𝑃(𝐹)) ∗ 𝜎

(𝐶𝑜𝑛𝑣 (𝐶𝑜𝑛𝑐𝑎𝑡(𝐴𝑣𝑔𝑃𝑜𝑜𝑙(𝐹), 𝑀𝑎𝑥𝑃𝑜𝑜𝑙(𝐹))))

(3)

The hybrid attention mechanism processes the input

feature map F by applying Global Average Pooling

(GAP) and Global Max Pooling (GMP) to capture

global information, followed by a fully connected

layer (FC) for channel attention (2). It uses a

convolution operation (Conv) on concatenated

pooled features to focus on important image regions,

generating spatial attention. Both attention maps are

scaled using a sigmoid activation (σ) and multiplied

Enhanced Tomato Leaf Disease Detection Using DenseNet201 with Channel and Spatial Attention Mechanisms

753

with the input feature map to emphasize key features

and regions. Evaluation metrics, including

experimental results for the five models shown in

Table II, performance comparison shown in Fig (5),

predicted output shown in Fig (6), confusion matrix

shown in Fig (7), accuracy curve shown in Fig (8),

and loss curve show in Fig (9) indicated that the

attention mechanism significantly improved the

model’s ability to identify subtle disease patterns.

Figure 4: Hybrid Attention Mechanism architecture

RESULT AND MODEL

EVALUATION

Table 2: Experimental Results

Figure 5: Performance Metrics Comparison of each model

Figure 6: Predicted output of Hybrid attention mechanism

Figure 7: Confusion matrix of Hybrid Attention

mechanism

Perfor

mance

Metric

VGG19

Efficient

NET

Channel

Attention

Mechanism

Special

Attention

Mechanism

Hybrid

Attention

Mechanism

Accur

88.9

94.2

97.21 96.07 98.07

Precisi

87.1

89.9

97 92 97

Recall 88.3 90.1 98 90 98

score

89.6 88.3 97 94 98

INCOFT 2025 - International Conference on Futuristic Technology

754

Figure 8: Accuracy graph of Hybrid attention mechanism

Figure 9: Loss graph of Hybrid attention mechanism

6 DISCUSSIONS

In this work, we developed a deep learning model for

tomato leaf disease detection by integrating a hybrid

attention mechanism that combines both spatial and

channel attention. After experimenting with various

architectures, including VGG19, Efficient Net, and

DenseNet121, we identified DenseNet201 as the most

effective model for our task due to its superior

accuracy and robustness. The DenseNet201

architecture excels in feature extraction through its

densely connected layers, which promote feature

reuse and mitigate the vanishing gradient problem.

By employing a hybrid attention mechanism, we

enhanced the model's ability to focus on relevant

features within the images, effectively improving the

detection of subtle disease symptoms. The spatial

attention mechanism enables the model to

concentrate on important regions in the image, while

the channel attention mechanism emphasizes the

most informative feature maps, resulting in a more

comprehensive understanding of the data. This dual

attention approach allows the model to capture both

fine- grained details and overall structural patterns in

the tomato leaves, leading to better classification

performance. Our experimental results demonstrate

that the model achieved an impressive testing

accuracy of 98.23%, outperforming other state-of-

the-art methods. The hybrid attention mechanism

contributed significantly to this achievement by

enhancing the model's sensitivity to critical features

associated with various leaf diseases. The model's

architecture, coupled with well- tuned

hyperparameters, played a crucial role in its ability to

learn and generalize effectively from the training

data. The performance metrics, including precision,

recall, F1 score, and accuracy, underscore the

effectiveness of our model. With precision ensuring

that a high proportion of the predicted positive cases

are true positives, recall measuring the model's ability

to identify all relevant instances, and the F1 score

providing a balance between precision and recall, our

model exhibited a well-rounded performance across

all categories. The overall accuracy of 98.23%

indicates the model's robust capability in

distinguishing between healthy and diseased leaves.

In terms of efficiency, our proposed model

demonstrated significant improvements in training

and inference times compared to other architectures.

The training time per epoch was approximately 2.73

minutes, with an inference time of just 0.008 seconds.

This efficient performance makes our model suitable

for real-time applications, providing timely and

accurate disease detection to support farmers in

managing crop health effectively. Overall, the

integration of a hybrid attention mechanism with

DenseNet201 has proven to be a promising approach

for tomato leaf disease detection, achieving a balance

between high accuracy and computational efficiency.

This combination positions our model as a valuable

tool for automated disease identification, offering

significant advantages in agricultural practices and

contributing to advancements in precision

agriculture.

REFERENCES

Attallah, O., 2023. Tomato leaf disease classification via

compact convolutional neural networks with transfer

learning and feature selection. Horticulturae, 9(2),

p.149.

Ulutaş, H. and Aslantaş, V., 2023. Design of efficient

methods for the detection of tomato leaf disease

utilizing proposed ensemble CNN model. Electronics,

12(4), p.827.

Guerrero-Ibañez, A. and Reyes-Muñoz, A., 2023.

Monitoring tomato leaf disease through convolutional

neural networks. Electronics, 12(1), p.229.

Enhanced Tomato Leaf Disease Detection Using DenseNet201 with Channel and Spatial Attention Mechanisms

755

Huang, X., Chen, A., Zhou, G., Zhang, X., Wang, J., Peng,

N., Yan, N. and Jiang, C., 2023. Tomato leaf disease

detection system based on FC- SNDPN. Multimedia

tools and applications, 82(2), pp.2121-2144.

Kaur, P., Harnal, S., Gautam, V., Singh, M.P. and Singh,

S.P., 2023. A novel transfer deep learning method for

detection and classification of plant leaf disease.

Journal of Ambient Intelligence and Humanized

Computing, 14(9), pp.12407-12424.

Paul, S.G., Biswas, A.A., Saha, A., Zulfiker, M.S., Ritu,

N.A., Zahan, I., Rahman, M. and Islam, M.A., 2023. A

real-time application-based convolutional neural

network approach for tomato leaf disease classification.

Array, 19, p.100313.

M.S., Ritu, N.A., Zahan, I., Rahman, M. and Islam, M.A.,

2023. A real-time application-based convolutional

neural network approach for tomato leaf disease

classification. Array, 19, p.100313.

Rahman, S.U., Alam, F., Ahmad, N. and Arshad, S., 2023.

Image processing-based system for the detection,

identification and treatment of tomato leaf diseases.

Multimedia Tools and Applications, 82(6), pp.9431-

9445.

Roy, K., Chaudhuri, S.S., Frnda, J., Bandopadhyay, S., Ray,

I.J., Banerjee, S. and Nedoma, J., 2023. Detection of

tomato leaf diseases for agro-based industries using

novel PCA DeepNet. Ieee Access, 11, pp.14983-15001.

Saeed, A., Abdel-Aziz, A.A., Mossad, A., Abdelhamid,

M.A., Alkhaled, A.Y. and Mayhoub, M., 2023. Smart

detection of tomato leaf diseases using transfer

learning-based convolutional neural networks.

Agriculture, 13(1), p.139.

Sanida, T., Sideris, A., Sanida, M.V. and Dasygenis, M.,

2023. Tomato leaf disease identification via two–stage

transfer learning approach. Smart Agricultural

Technology, 5, p.100275.

Ulutaş, H. and Aslantaş, V., 2023. Design of efficient

methods for the detection of tomato leaf disease

utilizing proposed ensemble CNN model. Electronics,

12(4), p.827.

INCOFT 2025 - International Conference on Futuristic Technology

756