Page 71
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Tom Leaf Vision: Real-Time Detection of Tomato Leaf Diseases Using
Deep Learning for Early and Late Blight Classification
Urvashi, Saumya Agrawal, Ritu Arya, Riya, Nitin Goyal
IT Professor at RD Engineering College Department of Computer Science & Engineering, RD
Engineering College, Ghaziabad, India
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150400008
Received: 26 March 2026; 01 April 2026; Published: 28 April 2026
ABSTRACT
Tomato cultivation contributes significantly to agricultural production, but it is highly prone to diseases such as
Early Blight and Late Blight, which can severely affect crop yield if not identified at an early stage. These
diseases spread quickly under favorable environmental conditions and can cause major losses to farmers.
Traditional methods of disease identification depend on manual inspection, which is time-consuming, labor-
intensive, and often unreliable, especially during the initial stages of infection. As a result, early symptoms are
frequently overlooked, leading to reduced productivity.
To address this problem, this paper presents TomLeafVision, a deep learning-based system designed for
automated detection of tomato leaf diseases. The proposed approach classifies leaf images into three categories:
Healthy, Early Blight, and Late Blight using a Convolutional Neural Network (CNN). To improve model
performance, input images captured through mobile devices undergo preprocessing steps such as resizing,
normalization, and data augmentation.
Furthermore, transfer learning using the MobileNetV2 architecture is applied to enhance classification accuracy
while reducing training time. The model is trained using the Adam optimizer with categorical cross-entropy as
the loss function. Experimental results indicate that the system performs effectively on unseen data and achieves
high accuracy. The proposed solution is user-friendly, cost-effective, and suitable for real-time deployment in
agricultural environments.
Keywords: Plant disease detection, smartphone-based diagnosis, image-based classification, transfer learning
techniques, deep learning approaches, convolutional neural networks (CNN), tomato leaf disease identification,
early blight, late blight
INTRODUCTION
Agriculture plays an essential role in supporting the economy of many countries, particularly in developing
regions where a significant portion of the population depends on farming as their primary source of income.
Among the various crops cultivated worldwide, tomato is one of the most widely grown and consumed
vegetables due to its nutritional benefits and commercial value.
However, tomato plants are highly susceptible to a range of diseases caused by fungi, bacteria, and viruses.
These diseases can lead to substantial reductions in crop yield and quality if not detected and managed at an
early stage. Among them, Early Blight and Late Blight are considered the most harmful, as they spread rapidly
and can cause severe damage within a short period.
Early Blight is typically characterized by the appearance of dark, circular lesions on older leaves, whereas Late
Blight spreads quickly under humid conditions and can destroy entire crops if left untreated. Therefore, early
and accurate detection of these diseases is critical for effective crop management and minimizing losses.
Page 72
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
In conventional agricultural practices, disease detection is usually performed through visual inspection by
farmers or experts. However, this method is often slow, subjective, and less effective in identifying diseases
during their initial stages. Moreover, manual inspection becomes impractical when dealing with large-scale
farming.
Recent advancements in artificial intelligence and computer vision have opened new opportunities for automated
plant disease detection. Deep learning models, especially Convolutional Neural Networks (CNNs), have shown
remarkable performance in image classification tasks. These models are capable of learning important features
such as color variations, textures, and patterns directly from images, eliminating the need for manual feature
extraction.
Despite their effectiveness, many existing approaches rely on datasets collected under controlled laboratory
conditions. Such datasets often lack variations in lighting, background, and environmental conditions, which
limits their applicability in real-world scenarios.
To overcome these limitations, this work proposes Tom Leaf Vision, a practical and scalable system designed
for real-time tomato leaf disease detection under field conditions. The system allows users to capture leaf images
using a mobile device and receive instant classification results.
Additionally, the system provides basic recommendations to farmers regarding disease prevention and
management. This not only helps in reducing crop losses but also promotes sustainable farming by minimizing
excessive use of pesticides.
The main objectives of the proposed system include:
Developing an automated system for early detection of tomato leaf diseases
Designing a deep learning model that performs reliably in real-world conditions
Providing a simple and accessible interface for farmers
By achieving these goals, the proposed system aims to bridge the gap between advanced technological solutions
and practical agricultural applications.
LITERATURE REVIEW
A number of research studies have explored the use of image processing and deep learning techniques for plant
disease detection. These approaches focus on improving the accuracy and efficiency of identifying diseases in
crops.
Several works have utilized deep learning models that combine different feature extraction strategies to enhance
classification performance. Such approaches aim to capture both global and local features from images, leading
to improved accuracy.
Object detection-based techniques have also been proposed to identify diseased regions in plant leaves. These
methods are capable of providing real-time detection, making them suitable for practical applications.
Some researchers have introduced context-aware models that incorporate additional information such as
environmental conditions to improve prediction accuracy. These approaches highlight the importance of
combining multiple data sources for better performance.
Sensor-based methods have also been explored, where specialized hardware is used to monitor plant health and
detect diseases. Although these techniques can provide accurate results, they often involve high costs and
complex setups.
Page 73
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Convolutional Neural Networks have been widely adopted in many studies due to their strong performance in
image-based tasks. Various CNN-based models have been developed to classify plant diseases with high
accuracy. Some systems also integrate real-time applications for practical usage.
Despite these advancements, certain challenges still remain. Many existing models are trained on limited
datasets, which affects their ability to generalize in real-world conditions. Additionally, some models require
high computational resources, making them unsuitable for mobile deployment.
To address these issues, the proposed work utilizes a combination of CNN and MobileNetV2 architectures. This
approach ensures a balance between accuracy, efficiency, and real-time performance, making it suitable for
practical agricultural applications.
Table I. Comparison of Existing Models for Tomato Leaf Disease Detection
Research Paper Title
Model Used
Performance Level
Limitations
Efficient deep learning-
based tomato leaf disease
detection through global
and local feature fusion
Feature fusion
CNN
High Accuracy
Complex model design
TomatoGuard-YOLO: A
novel efficient tomato
disease detection method
YOLO (Object
Detection)
Very High Accuracy
High computational cost
Context-aware tomato leaf
disease detection using
deep learning
CNN
High Accuracy
Dataset dependency
Spectral sensors-based
device for real-time
detection
Sensors + ML
Moderate Accuracy
Expensive hardware
Tomato leaf disease
detection using CNN
CNN
High Accuracy
Limited Dataset
Real-time CNN-based
tomato leaf disease
classification
CNN
Good Accuracy
Less robust in real
conditions
A study on tomato disease
and pest detection method
Machine
Learning
Moderate Accuracy
Lower accuracy
Early detection and
classification using deep
neural network
Deep Neural
Network
High Accuracy
High training time
PROPOSED METHODOLOGY
The proposed system follows a systematic approach for detecting tomato leaf diseases using deep learning
techniques. The process begins with collecting images of tomato leaves using mobile devices in real-field
environments. The dataset consists of images belonging to three categories: Healthy, Early Blight, and Late
Blight.
During preprocessing, all images are resized to a fixed dimension to maintain consistency. Pixel values are
normalized to improve model stability, and data augmentation techniques such as rotation and flipping are
applied to increase dataset diversity and reduce overfitting.
Page 74
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
The classification process is carried out using a Convolutional Neural Network (CNN), which automatically
extracts relevant features from input images. The model is trained using the Adam optimizer, and categorical
cross-entropy is used as the loss function for multi-class classification.
To enhance performance, transfer learning is implemented using the MobileNetV2 architecture. This model is
chosen due to its lightweight design and efficiency, making it suitable for real-time applications. Fine-tuning the
pre-trained model allows it to adapt effectively to the tomato leaf dataset.
After training, the model is evaluated using unseen data to assess its performance. The results demonstrate that
the system is capable of accurately detecting diseases at an early stage, making it useful for practical deployment.
Dataset Description
The dataset utilized in this research consists of tomato leaf images obtained from real-world conditions,
specifically collected from nearby agricultural fields. It is categorized into three groups: healthy leaves, leaves
affected by early blight, and those impacted by late blight. Efforts were made to ensure a balanced representation
of each category within the dataset.
To enhance the diversity and volume of the data, various augmentation methods were implemented. The
complete dataset was then partitioned into training, validation, and testing subsets using an 80:10:10 split,
allowing for a fair and reliable assessment of the model’s performance.
The dataset also includes representative examples of tomato leaf images, covering healthy samples as well as
those exhibiting symptoms of early blight and late blight.
Page 75
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Figure 1: Sample tomato leaf images from the TomLeafVision Dataset representing respectively
Healthy, Early Blight, and Late Blight.
CNN Model Architecture
The TomLeafVision model is developed using a deep Convolutional Neural Network (CNN) framework that is
capable of automatically identifying and classifying disease patterns in tomato leaf images. In this architecture,
a tomato leaf image is provided as input and processed through several layers of convolution and pooling to
extract meaningful features.
The convolutional layers utilize multiple trainable filters to identify key visual characteristics such as edges,
color differences, textures, and patterns associated with specific diseases on the leaf surface. Each convolution
step is followed by a non-linear activation function, which helps the model capture complex patterns within the
data. Pooling layers are incorporated after these operations to reduce the size of feature maps while preserving
important information, which also helps in lowering computational cost and minimizing overfitting.
Figure 2: CNN Model Architecture of the Proposed
Page 76
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
At the final stage, a Softmax activation function is applied to produce probability values for each class, including
healthy, early blight, and late blight. The class with the highest probability is selected as the final prediction.
During the training phase, the predicted output is compared with the actual label to calculate the loss, indicating
the difference between prediction and ground truth.
To enhance the model’s performance, backpropagation is used to adjust the weights of both convolutional and
fully connected layers. Through continuous iterations, the network reduces the loss and improves its prediction
capability. Overall, this CNN-based architecture integrates feature extraction, classification, and optimization
processes to deliver accurate and reliable detection of tomato leaf diseases.
Workflow Description
Figure 3: Workflow of the Proposed TomLeafVision System
The proposed TomLeafVision system begins with the farmer capturing an image of a tomato leaf using a mobile
device. This image is then forwarded to the image acquisition module for further processing. During the
preprocessing stage, the image is resized and normalized to ensure consistency and to improve the overall
performance of the model.
Once preprocessing is complete, the image is passed to the trained CNN model, which analyzes it and classifies
the leaf based on the features it has learned. The model determines whether the leaf is healthy or affected by
early blight or late blight. Finally, the prediction results are displayed to the farmer in a clear and user-friendly
manner, enabling quick and informed decisions for effective disease management.
RESULTSAND ANALYSIS
Model Comparison
The table below presents a comparison between the proposed MobileNetV2 model and several widely used
deep learning architectures, such as ResNet50, VGG16, InceptionV3, and a custom CNN model. The analysis
shows that MobileNetV2 requires significantly fewer parameters while still maintaining a good trade-off
between computational efficiency and predictive performance, making it suitable for real-time applications.
Page 77
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Table II. Comparison of proposed MobileNetV2 Model with Existing Deep Learning Models
Model
Parameters
Accuracy
Speed
Mobile Ready
MobileNetV2 (Ours)
2.5M
94%
Fast
Yes
ResNet50
25.6M
95.1%
Slow
No
VGG16
138M
93.5%
Very Slow
No
InceptionV3
23.9M
94.8%
Medium
Limited
Custom CNN
~1M
88.2%
Fast
Yes
Model Performance Analysis
Table III. Performance Evaluation Metricsof the Proposed TomLeafVision System
Metrics
Value
Training Accuracy
96%
Validation Accuracy
94%
Field Test Accuracy
92%
Precision
93%
Recall
92%
F1-Score
92%
A slight drop in performance during field testing can be attributed to environmental factors such as lighting
conditions, background variations, and image quality. However, the overall results demonstrate that the model
performs reliably under real-world conditions.
System Interface and Prediction Result
The proposed model effectively captured patterns from the training data, achieving a training accuracy of 96%.
It also demonstrated strong performance on unseen samples, with a validation accuracy of 94%, indicating good
generalization and minimal overfitting. The slight decrease in accuracy from training to validation can be
attributed to variations in environmental conditions, such as lighting differences, background complexity, and
image quality. Overall, the results suggest that the CNN-based approach provides a reliable and accurate solution
for detecting diseases in tomato leaves.
Figure 4: Interface for Uploading Images of Tomato Leaves
Page 78
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
The interface allows users to either upload an image or capture a photo of a tomato leaf for disease detection.
Once the image is provided through file upload or camera input, it is forwarded to the trained CNN model for
further analysis and classification..
Prediction Result Page
Figure 5: Results of Real-Time Disease Detection for Various Tomato Leaf Conditions
The proposed system was evaluated using a variety of tomato leaf images, including samples affected by early
blight, late blight, as well as healthy leaves.
In addition to classification, the system presented detailed insights for diseased leaves, such as possible causes,
visible symptoms, preventive strategies, and suggested treatments. For healthy leaves, it offered precautionary
recommendations to help maintain plant health.
These results highlight the effectiveness of the CNN-based approach in accurately distinguishing between
healthy and diseased tomato leaves in real-time conditions.
Page 79
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Furthermore, each convolutional layer is followed by a non-linear activation function, enabling the network to
capture complex patterns within the data. Pooling layers are incorporated after convolution stages to decrease
the spatial size of feature maps while preserving essential information, which enhances computational efficiency
and helps prevent overfitting.
Model Training Performance (Loss Curve)
Figure 6: Training and Validation Loss Curve showing model convergence over epochs
The loss curve illustrates the training and validation losses over 15 epochs. Initially, the training loss is relatively
high, but it steadily decreases as the model learns important features from the dataset. Similarly, the validation
loss also shows a downward trend, indicating improved performance on unseen data. This behavior suggests that
the model is learning effectively with minimal overfitting, demonstrating the efficiency of the CNN approach.
A. Performance Evaluation using F1-Score, Recall, Precision
Figure 7: Performance Metrics Bar Chart of the Proposed CNN Model
0
0.5
1
1.5
2
2.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Loss
Epochs
Training and Validation Loss Curve for Tomato Leaf Disease
Detection Model
Training Loss
Validation Loss
0.93
0.92 0.92 0.92
0
0.2
0.4
0.6
0.8
1
Precision Recall F1-Score Accuracy
Score Value
Evaluation Metrics
Performance Metrics of the Proposed CNN Model
Page 80
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
The bar chart presents the performance metrics of the proposed CNN model for detecting diseases in tomato
leaves. The model achieves high values in terms of precision, recall, F1-score, and overall accuracy, indicating
strong classification capability. The balanced nature of these metrics suggests that the model effectively
distinguishes between healthy and diseased leaves while keeping misclassification to a minimum.
CONCLUSION
This paper presents a deep learning-based system for early detection of tomato leaf diseases, focusing on Early
Blight and Late Blight. The proposed approach combines CNN with MobileNetV2 to achieve high accuracy
while maintaining computational efficiency.
Experimental results show that the model performs effectively in terms of accuracy, precision, recall, and F1-
score. The system demonstrates good generalization ability and is capable of handling real-world variations.
The mobile-based implementation makes the system accessible and easy to use for farmers, enabling timely
decision-making and reducing crop losses. Overall, the proposed work contributes to the advancement of smart
agriculture by integrating modern technology with practical farming needs.
Future work may include expanding the dataset, improving robustness under varying environmental conditions,
and extending the system to detect additional plant diseases.
REFERENCES
1. Hao Sun, Rui Fu, “Efficient deep Learning-based tomato leaf disease detection through global and
local feature fusion,” BMC Plant Biology, ISSN: 1471-2229, 2025.
2. X. Wang, J. Liu, “ToamtoGuard-YOLO: a novel efficient tomato disease detection method,” Frontiers
in plant Science, ISSN:1664-462X, 2025.
3. D. Karimanizira, “Context-aware tomato leaf disease detection using deep learning in an operational
framework,” Electronics, ISSN: 2079-9292, 2025.
4. K. Kumari, R. Parray, Y.B. Basavaraj, “Spectral sensor-based device for real-time detection and
severity estimation of groundnut bud necrosis virus in tomato,” Journal of Field Robotics, ISSN:1556-
4959, 2025.
5. D.L. Shanhi, K. Vinutha, N. Ashwini, “Tomato Leaf Disease Detection Using CNN,” Procedia
Computer Science (Elsevier), ISSN:1877-0509, 2024.
6. S.G. Paul, A.A. Biswas, A. Saha, “A real-time application-based convolutional neural network
approach for tomato leaf disease classification,” Array (Elsevier Journal), ISSN: 2590-0056, 2023.
7. W. Hu, W. Hong, H. Wang, “A study on tomato disease and pest detection method,” Applied Sciences,
2076-3417, 2023.
8. N.K. Trivedi, V. Gautam, A. Anand, “Early Detection and Classification of Tomato Leaf Disease Using
High-Performance Deep Neural Network,” Sensors, ISSN:1424-8220, 2021.
9. X. Wang, J. Liu, Z. Zhu, “Early real-time detection algorithm of tomato diseases and pests in the natural
environment,” Plant Methods, ISSN:1746-4811, 2021.
10. M. Agarwal, S.K. Gupta, K.K. Biswas, “Development of Efficient CNN Model for Tomato Crop
Disease Identification,” Sustainable Computing: Informatics and Systems (Elsevier Journal), ISSN:
2210-5379, 2020.
11. M. Agarwal, A. Singh, S. Arjaria, “Tomato Leaf Disease Detection using convolution neural network,”
Procedia Computer Science, ISSN:1877-0509, 2020.
12. J. Abdulridha, Y. Ampatzidis, “Detection of target spot and bacterial spot diseases in tomato
usingUAV-based and benchtop-based hyperspectral,” Agriculture, 1385-2256, 2020.
13. A. Kumar, M. Vani, “Image Based Tomato Leaf Disease Detection,” International Conference on
Computing, Communication and Networking Technologies, ISSN: 2150-6864, 2019.
14. Q. Wang, F.Qi, M. Sun, “Identification of tomato disease types and detection of infected areas based
on deep convolutional neural networks and object detection technologies,” Computational
Intelligence,ISSN: 0824-7935, 2019.
Page 81
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
15. Robert G. de Luna, Elmer P. Dadios, Argel A. Bandala, “Automated image capturing system for deep
learning-based tomato plant leaf disease detection and recognition,” IEEE Region 10 Annual
International Conference, ISSN: 2018.
