Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive
Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive

Full Research Article

Deep Learning-based Classification of Tomato Leaf Diseases

V
Varsha Nemade1,*
V
Vishal Fegade2
D
Deepti Barhate1
S
Suraj Patil1
1Department of Computer Science, Mukesh Patel School of Technology Management and Engineering Shirpur, SVKM's Narsee Monjee Institute of Management Studies, Deemed-To-University, Mumbai-400 056, Maharashtra, India.
2Department of Applied Sciences and Humanities, Mukesh Patel School of Technology Management and Engineering Shirpur, SVKM's Narsee Monjee Institute of Management Studies, Deemed-To-University, Mumbai-400 056, Maharashtra, India.

ABSTRACT

Background: In the Indian economy agriculture plays important role. Many of the crops are damaged due to diseases, therefore plant leaf disease detection at early stage is important. Tomatoes are the second most consumed vegetable in Indian households, with a rank second largest producer and consumption in world. Tomatoes are of economically important crop in India with large scale cultivation. Yet, unfavourable environmental factors tend to cause numerous diseases due to bacteria, fungi and viruses that infect different plant parts. These diseases cause the yield to be lower, leading to heavy economic losses for the farmers. For effective disease management, it is important to detect the disease in early stage using the advanced techniques for maximising the yielding of crop.

Methods: In the proposed approach, tomato leaf images were first enhanced using CLAHE to improve local contrast, after that data augmentation to increase dataset diversity. Four transfer learning models based on pretrained CNN architecture: VGG16, VGG19, ResNet50 and MobileNetV2, were used as fixed feature extractors, where the pre-trained layers were fixed and newly added classification layers were trained for tomato leaf disease classification. Although transfer learning has been widely applied for tomato leaf disease classification, comparative investigations on the effect of CLAHE-enhanced images across different pre-trained architectures remain limited. This study shows performance of these four CNN architectures under identical preprocessing and training conditions to analyze the influence of CLAHE-based contrast enhancement on disease feature representation and classification performance.

Result: The outcomes of experimental results shows strong classification performance on the plant village dataset for detection of tomato leaf disease, achieving best accuracies of 93.75%, 95.75%, 86.38% and 98.54% using the VGG16, VGG19, ResNet50 and MobileNetV2 models respectively.

INTRODUCTION

Detecting plant diseases is of the utmost importance as it paves the way for preventing crop diseases and maintaining the crop quality. Tomato is economically important crop, representing a fair share of dietary needs (Sundararaman et al., 2023). It is among the most widely cultivated horticultural crops. Tomatoes are versatile and hence valuable in many food applications. Tomatoes has rich source of nutrients like vitamin C, vitamin K and antioxidants. Usually, tomatoes are grown throughout the summer season but can be grown throughout the year. However, the diversity of diseases attacking the foliar parts of this plant along with other tissues has threatened both productivity and economic viability in tomato farming (Sundararaman et al., 2023; Attallah, 2023). Generally diseases produced from a different pathogens such as fungi, bacteria and viruses (Kaur et al., 2023) and manifest various symptoms; hence, yield loss and economic hardship among farmers cannot be ignored (Attallah et al., 2023; Pandiyaraju et al., 2024). This investigation considers the most prevalent tomato leaf diseases, as shown in Fig 1 a, b, c, d, e respectively. Disease identification rely heavily on expert visual inspection, which are time-consuming, often subjective and not readily scalable to large-scale agricultural operations (Debnath et al., 2023; Chowdhury et al., 2021). Traditional method depends on the on human observers are inefficient, have poor scalability and are unreliable. To address this challenge, different image processing and computer vision techniques have been employed for the automated analysis of plant diseases (Scharr et al., 2017; Al Farid et al., 2022). In addition, machine learning-based intelligent systems have been developed to enable automatic disease detection that provide timely intervention and improved crop management practices that contribute to enhanced agricultural productivity (Islam et al., 2022).

Fig 1: Tomato leaf diseases.


       
Deep learning (DL) is a subfield of artificial intelligence and this has drastically transformed the character of image analysis and classification procedures (Wagle and Harikrishnan, 2021; Zhang et al., 2018). DL architecture, in general and the convolutional neural network (CNN) in particular, are self-learning in obtaining complex patterns and features from data in images which may lead to developing dependable and efficient strategies in the identification of diseases on the leaves of the tomato Wagle and Harikrishnan, 2021; Zhou et al., 2021). Many studies have proven the success of CNNs in classifying a wide range of plant diseases, including those of tomato leaves (Attallah et al., 2023; Wagle and Harikrishnan, 2021). For leaf image classification deep learning CNN model are always better choice as compared to conventional ML algorithms as it is computationally efficient, providing highest accuracy and faster processing times. CNN are useful to automatically detect important features from the input images without manual feature extraction or human supervision. This allows them to learn hierarchical patterns and complex structures in the data, making them highly effective for leaf classification. Performance of the model can be improve by image enhancement and augmentation technique.
       
For the purpose of giving the accurate and efficient method in detecting and predicting the development of tomato leaf diseases. For image enhancement contrast limited adaptive histogram equalization (CLAHE) is use and    augmentation technique use to expand the dataset. After that CNN architectures such as VGG16, VGG19, ResNet50 and MobileNetV2 are used for the disease classification.
       
Plant village dataset is use in this study for comparing different transfer learning (TL) architectures by using identical preprocessing and training conditions. Although the extensive research on tomato leaf disease classification using transfer learning, most existing studies focuses on evaluating individual CNN architectures or proposing new architectures. Limited research has been given to systematically investigating the impact of CLAHE-based image enhancement across multiple pre-trained architectures under identical experimental settings. This study addresses this gap by providing a comprehensive comparative analysis of VGG16, VGG19, ResNet50 and MobileNetV2, highlighting how contrast enhancement influences feature extraction and classification performance.
       
Major contributions of this work are:
1. A systematic evaluation of CLAHE-based contrast enhancement for tomato leaf disease classification using four different TL architectures.
2. An analysis of how image enhancement and augmentation affect feature learning and classification performance.
A comparative study of lightweight MobileNetV2 and deep architectures: VGG16, VGG19 and ResNet50 under identical training conditions.
3. An investigation into architecture-specific behavior for tomato disease recognition, highlighting the strengths and limitations of each model.

The organization of paper as follows:
       
The literature review section provides prior research in the field of identification and classification plant disease. The proposed methodology section provides the propose methodology for the classification of tomato leaf disease. The results and discussion section provides an analysis of work. Lastly, the conclusion section summarizes the study.
 
Literature survey
 
The traditional methods for detection of plant disease are mainly based on the texture, shape, color and other features of the disease spots for extraction. The technique has low identification efficiency since it relies on a vast amount of expert knowledge in agricultural diseases. Now artificial intelligence (Rangarajan et al., 2018) has proved to be very effective in upgrading the discipline of plant science, ML and DL strategies have emerged as the two most popularly explored methodologies for the detection plant diseases. Most of the existing methods for plant disease analysis are based on disease classification (Mane and Rangarajan, 2017; Patil and More, 2025). Mehta et al., (2025) presented reviews on by using DL frameworks to perform plant disease classification, thereby powering smart farming operations. Nigam and Jain (2020) have evaluated a wide variety of deep learning algorithms for plant disease detection and identification. It was found by them that CV, ML and DL applications greatly increased classification accuracies, which led to more intelligent decision-making in agricultural applications.This section reviews recent and prominent studies focusing on automatic disease identification systems, focusing on novel, design-oriented frameworks and their applications to overcome challenges in this area of study.
       
Tm et al., (2018) proposed three-phase methodology, which is the collection of images of tomato leaf from plant village dataset, data pre-processing with normalization and disease classification using a modified LeNet CNN. The proposed model obtained accuracy of 94-95% that surpassed the AlexNet and GoogleNet models. ResNet (Ahmad et al., 2020) has more layers of convolution than VGG16, which means it can better extract the details of objects. Its layer-skipping structure skips layers and overcomes the gradient vanishing problem due to deep stacking. VGG16 fails to extract detailed features for tomato leaf diseases (Aversano et al., 2020). The deep residual network uses feed-forward neural network residual connections such that layer outputs become inputs to subsequent layers, hence improving efficiency without adding any variables or considerably increasing computation time.
       
Deng et al., (2021) utilized generative adversarial networks (GANs) to augment the dataset and trained different network and evaluating their performance on a test set. Kanda et al., (2022) described the use of four diversity levels: depth size, discriminative learning rates, training-validation split ratios and batch sizes. It reached F1 score:99.5%, surpassing the performance of most existing tomato leaf disease recognition methods. Further testing is conducted on the Flavia leaf image dataset to obtain a 99.23% F1 score, signifying the reliability and efficiency of the proposed approach. Peng et al., (2023) has described a new classification network for tomato leaf disease that combines the dense inception MobileNet-V2 with a parallel convolutional block attention module. The original images of five tomato leaf diseases were expanded to 8190 using data augmentation from 1256. An improved bilateral filtering and threshold function algorithm IBFTF was proposed to effectively filter noise out. While Dense Inception focuses on most intra-class variability but negligible inter-class differences, the PCBAM enhanced MobileNet-V2 minimizes the effects of background complexity. Empirical experiments show that the DIMPCNET achieves an accuracy of 94.44% with an F1-score of 0.9475. Siddiky et al., (2024) applied the deep learning and image processing to enhance the detection of diseases and, therefore, food security. Five models were applied: MobileNet, ResNet50V2, Xception, InceptionV3 and VGG19. The five models were trained on 83,568 images of tomato leaves. The best validation accuracy was achieved by MobileNet at 91%, followed by the others. Chowdhury et al., (2021) applied EfficientNet for tomato leaf disease classification in three scenarios: Two-class (healthy vs. unhealthy), six-class labels (segmented) and ten-class labels. The Adam optimizer achieved 99% accuracy for this task. Tan et al., (2021)  compared traditional machine learning techniques including KNN, SVM and RF with deep learning architectures like AlexNet, VGG16, ResNet34, EfficientNet and MobileNetV2 on the plant village dataset. Classical approach methods have achieved the accuracies between 82.10%-91.00%, whereas the deep learning model achieved the highest accuracy between 91.20% to 99.70% . Sharma et al. (2025) proposed deep learning ensemble model based on MobileNetV2 and ResNet50 is presented for classifying the tomato leaf diseases. By training it on a Kaggle dataset consisting of 11,000 images covering 10 disease classes, the model was able to achieve state-of-the-art performance with 99.91% test accuracy. Adding layers and merging of feature maps improved the ability to extract features and reduce false classifications. This strategy enables the early detection of plant diseases, by reducing crop losses and promoting smart and sustainable agricultural practices. Vinothini et al., (2025) developed a transfer learning-based deep learning model for the classification of tomato leaf diseases.In both the balanced setup and imbalanced setup, 4 models (AlexNet, LeNet-5, DenseNet-121 and InceptionV3) were trained using different sampling methods. The enhanced AlexNet-based model presented the best performance with accuracy, precision, recall and F1-score of 95%. This system provided a dependable means for early disease identification to help enable efficient crop management and sustainable agriculture practice. Mezenner et al. (2025) proposed a novel fuzzy aggregation-based CNN ensemble model for tomato leaf disease classification. First, five deep learning models VGG16, ResNet152V2, MobileNetV2, InceptionV3 and DenseNet201 are pre-trained separately where different feature extractors of each model are used. These models are aggregated via a newly proposed fuzzy Max-Sum integral derived from the fuzzy convex combination operator.

MATERIALS AND METHODS

The experimental work was carried out at the MPSTME Shirpur, Maharashtra. The study utilized the publicly available PlantVillage dataset for tomato leaf disease classification. The experimental work was conducted during the period from May 2025 to December2025. Diseases of tomato leaves considerably reduce agricultural productivity, causing substantial economic problems to the growers. In this context, we highlight the critical necessity of timely and accurate disease diagnosis by using proposed architecture as shown in Fig 2.

The proposed method comprises the following components:

Fig 2: Overall architecture of proposed approach.


 
Dataset and pre-processing
 
This study utilized a dataset comprising images of tomato plants affected by various diseases, serving as a valuable foundation for analysis and research. For this work Plant village dataset (PlantVillage, 2024) is used that is available on Kaggle. We choose tomato leaves images of 6 classes. Class 0: Bacterial spot, Class 1: Early blight, Class 2: Healthy, Class 3: Septoril leaf spot, Class 4: Leaf mold, Class 5: Yellow leaf curl virus, there count is as shown in Fig 3, out of this 7245 images used for training, 1355 images used for validation and 448 images used for testing.

Fig 3: Class wise image count in original dataset.


       
Pre-processing of the images is important step before to applying the actual model, as it helps to further improve the effectiveness of the model. In this work CLAHE is used as described in Algorithm 1. This technique applied on all images in the dataset. It enhances disease features like spots, discoloration, or blight, making them more prominent, reduces the effects of uneven lighting in image and help to improves model performance, especially for models relying on color and texture features. Fig 4 shows tomato leaf images before and after applying CLAHE for each class. This approach was chosen since tomato leaf disease is usually described through various changes in texture, coloring and lesion boundaries, which may not be sufficiently reflected in the image itself. The use of CLAHE will help improve the local contrast of an image without over-amplifying the noise present.

Fig 4: Before and after applying CLAHE.


       
When we are working with limited data size data augment helps to increase the size of the original dataset. This approach helps strengthen the model’s ability to generalize effectively, reduce overfitting and speed up the training. The data augmentation was only done on the training dataset, not on the validation and test datasets, in order to measure the performance of the model precisely. The same technique had been proved repeatedly to be enhancing the performance of deep learning models and their generalization. The augmentations that were done by using height shift, width shift, rotation and zooming. After augmentation training images increases from 7245 to 26658.    
 
Algorithm 1: CLAHE algorithm.
 
Input  
 
I: Input image.   
N: Number of tiles per row and column (N × N).   
Cliplimit: Maximum allowed histogram count before clipping.   
NB: Number of bins for histogram.
 
1. Initialize parameters
 
Divide the image I into (N × N). 
Ti,j= The tile at position (i, j).
 
2. For each tile Ti,j
 
Calculate histogram Hi,j of the tile using NB bins.
 
Clip the histogram 
    
If any bin Hi,j (k) > cliplimit, redistribute excess counts equally to all bins while maintaining the total count.
 
Normalize the histogram:

 
Calculate the cumulative distribution function (CDF):

 
Map the pixel values of Ti,j to the enhanced range:
 
 
Where, 
Pold= The original pixel value.
Pnew= The enhanced value.
 
3. Perform interpolation
 
Use bilinear interpolation:
 
Ienhanced (x, y) = w1. Ti,j + w2. Ti+1 + w3. Ti,j+1 + w4. Ti+1, j+1
 
Where, 
w1, w2, w3, w4= The weights based on the relative distance of (x, y) to the centers of the neighboring tiles.
 
4. Output: Ienhanced: Enhanced image
 
Deep learning architectures
 
We employed four pre-trained CNN architectures-VGG16, VGG19, ResNet50 and MobileNetV2-for tomato leaf disease classification. The pre-trained models were initialized with ImageNet weights using the include_ top=False configuration, which removes the original ImageNet classification layer while retaining the convolutional feature extraction layers. Transfer learning was adopted to leverage the generic visual features learned from the large-scale ImageNet dataset. To preserve the generic visual features learned during pre-training and reduce computational complexity, all layers of the pre-trained base networks were frozen during training. Specifically, the VGG16 base model consisted of 19 layers (1 input layer, 13 convolutional layers and 5 max-pooling layers), while the VGG19 base model consisted of 22 layers (1 input layer, 16 convolutional layers and 5 max-pooling layers), all of which were set as non-trainable. Similarly, the complete pre-trained base networks of ResNet50 (175 layers) and MobileNetV2 (154 layers), comprising convolutional, batch normalization, activation, pooling and residual/inverted residual layers, were entirely frozen. Consequently, only the newly added classification head, consisting of a Flatten layer, a Dropout layer (0.6), a Dense layer with 256 neurons using ReLU activation and L2 regularization (0.002) and a Softmax output layer with six neurons, was trained using the tomato leaf dataset. For each architecture, a custom classification head was appended to the frozen base network. The classification head consisted of a flatten layer, followed by a dropout layer with a dropout rate of 0.6 to reduce overfitting. A fully connected dense layer with 256 neurons, ReLU activation function and L2 regularization  was added to improve feature learning and enhance model generalization. Finally, a Softmax output layer with six neurons was used to classify the tomato leaf images into the six disease categories. For training, Adam optimizer was chosen with the learning rates: 0.001 and 0.0001 to observe the impact of varied optimization techniques. With the learning rate of 0.001, the training process converges quickly; however, the use of learning rate 0.0001 provides slow updates to the weights. Due to the fact that the task is about multi-class classification, the categorical cross-entropy loss function was utilized. Each model was trained for a maximum of 100 epochs with early stopping to prevent overfitting. Model performance was evaluated on the test dataset using accuracy, precision, recall and F1-score and the results of all four architectures were compared to identify the most effective model for tomato leaf disease classification.

RESULTS AND DISCUSSION

This work assesses the performance of different fine-tuned pre-trained models for the classification of tomato leaf diseases, CLAHE for pre-processing and data augmentation methods to improve the dataset. Table 1 and 2 illustrate the performance of four CNN architectures-VGG16, VGG19, ResNet50 and MobileNetV2-on a classification task with varying learning rates LR=0.001 and 0.0001 respectively. The performance is measured using three common metrics: Precision, recall and F1-score, on six classes. Table 3 shows the accuracy comparison of CNN models at different learning rates. All model shows outstanding performance at LR=0.001. VGG 16, VGG19, ResNet50 and MobileNetV2 achieved 93.75 %,95.75%,86.38% and 98.54% accuracy.

Table 1: Performance measure (LR =0.001).



Table 2: Performance measure (LR =0.0001).



Table 3: Accuracy comparison.


       
From the above experiment findings, it is clear that deep learning models that have been pretrained are highly effective in detecting tomato leaf diseases. The application of CLAHE pre-processing led to increased image contrast, feature extraction and overall model performance. Data augmentation was applied to solve the problem of imbalanced classes in the data set, hence increasing the overall robustness of the model. From among the models tested, the MobileNetV2 model proved to be the most effective, recording high accuracy and F1-scores.
       
VGG19 also performed well, especially at high learning rates, showing its capability to work in high precision applications. ResNet50 worked as a feature extractor and could not adapt its deep residual features to the tomato leaf disease dataset, which may have resulted in underfitting and reduced classification performance. ResNet50 functioned as a fixed feature extractor and could not adapt its deep residual features to the tomato leaf disease dataset, which may have resulted in underfitting and reduced classification performance. Learning rate was equally critical in optimizing the model as well. High learning rates were found to give better performance in all cases, showing faster optimization without overfitting, while low learning rates gave poorer performances especially with minority classes.

CONCLUSION

The present work highlights the classification of tomato leaf diseases using DL algorithms which is essential in prevention of agriculture losses.  CLAHE technique used for pre-processing and data augmentation techniques for expanding the dataset. The suggested method achieved impressive levels of classification accuracy. MobileNetV2 algorithm provided the best results by obtaining accuracy level of 98.54% which makes it extremely suitable for low resource settings such as mobile devices.
       
Even though the promising results were obtained in this research, there are certain limitations of this research. Experiments were conducted on the PlantVillage dataset which consists of laboratory acquired images and thus does not entirely represent the variety of images in a real-life agricultural setting. Further research will include conducting experiments on tomato leaf datasets obtained in the field, studying the influence of partial fine-tuning of high layers, using explainable artificial intelligence methods such as Grad-CAM to increase the model interpretability.

CONFLICT OF INTEREST

The authors declare no conflicts of interest of any kind.

REFERENCES


  1. Al Farid, F., Hashim, N., Abdullah, J., Bhuiyan, M.R., Shahida, M.I.W.N., Uddin, J., Haque, M.A. and Husen, M.N. (2022). A structured and methodological review on vision-based hand gesture recognition system. Journal of Imaging. 8(6): 153. https://doi.org/10.3390/jimaging8060153.

  2. Ahmad, I., Hamid, M., Yousaf, S., Shah, S.T. and Ahmad, M.O. (2020). Optimizing pretrained convolutional neural networks for tomato leaf disease detection. Complexity. 2020(1): 8812019. https://doi.org/10.1155/2020/8812019.

  3. Attallah, O. (2023). Tomato leaf disease classification via compact convolutional neural networks with transfer learning and feature selection. Horticulturae. 9(2): 149. https:// doi.org/10.3390/horticulturae9020149.

  4. Aversano, L., Bernardi, M. L., Cimitile, M., Iammarino, M. and Rondinella, S. (2020). Tomato Diseases Classification Based on VGG and Transfer Learning. In 2020 IEEE International Workshop on Metrology for Agriculture and Forestry (MetroAgriFor). IEEE. (pp. 129-133). 10.1109/ MetroAgriFor50201.2020.9277626

  5. Chowdhury, M.E.H., Rahman, T., Khandakar, A., Ayari, M.A., Khan, A.U., Khan, M.S. et al. (2021). Automatic and reliable leaf disease detection using deep learning techniques. Agriengineering. 3(2): 20. https://doi.org/10.3390/ agriengineering3020020.

  6. Debnath, A., Hasan, M. M., Raihan, M., Samrat, N., Alsulami, M.M., Masud, M. and Bairagi, A.K. (2023). A smartphone-based detection system for tomato leaf disease using EfficientNetV2B2 and its explainability with artificial intelligence (AI). Sensors. 23(21): 8685. https://doi.org/10.3390/s23218685.

  7. Deng, H., Luo, D., Chang, Z., Li, H. and Yang, X. (2021). Rahc_gan: A data augmentation method for tomato leaf disease recognition. Symmetry. 13(9): 1597. https://doi.org/ 10.3390/sym13091597.Ā 

  8. Islam, M.N., Ahmed, F., Ahammed, M.T., Rashid, M. and Bari, B.S. (2022). Rice Disease Identification Through Leaf Image and IoT-Based Smart Rice Field Monitoring System. In: Enabling Industry 4.0 through advances in mechatronics. Springer. (pp. 529-539). https://doi.org/10.1007/978-981- 19-2095-0_45.Ā 

  9. Kanda, P.S., Xia, K., Kyslytysna, A. and Owoola, E.O. (2022). Tomato leaf disease recognition on leaf images based on fine-tuned residual neural networks. Plants. 11(21): 2935. https://doi.org/10.3390/plants11212935.Ā 

  10. Kaur, P., Singh, M. P., Mishra, A., Shankar, A., Singh, P., Diwakar, M.Ā et al. (2023). DELM: Deep ensemble learning model for multiclass classification of super-resolution leaf disease images. Turkish Journal of Electrical Engineering and Computer Sciences. 31(3): 3123. https://doi.org/ 10.55730/1300-011X.3123.

  11. Mane, D.T. and Rangarajan, U.V. (2017). A survey on supervised convolutional neural networks and its major applications. International Journal of Rough Sets and Data Analysis. 4(3): 71-82. https://doi.org/10.4018/IJRSDA.2017070105.Ā  Ā Ā 

  12. Mehta, A.R., Kumar, P., Prem, G., Aggarwal, S. and Kumar, R. (2025). Leveraging artificial intelligence for disease diagnosis in agricultural crops: A review. Indian Journal of Agricultural Research. 59(5): 681-690. doi: 10.18805/IJARe.A-6363.

  13. Mezenner, A., Nemmour, H., Chibani, Y., and Hafiane, A. (2025). Fuzzy aggregation-based deep learning ensemble for enhanced tomato leaf disease classification. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems. 33(7): 911-932. https://doi.org/10.1142/ S0218488525400124.Ā 

  14. Nigam, S. and Jain, R. (2020). Plant disease identification using deep learning: A review. The Indian Journal of Agricultural Sciences. 90(2): 249-257. doi: 10.56093/ijas.v90i2.98996.

  15. Pandiyaraju, V., Kumar, A.M.S., Praveen, J.I.R., Venkatraman, S., Kumar, S.P., Aravintakshan, S.A., Abeshek, A. and Kannan, A. (2024). Improved tomato leaf disease classification through adaptive ensemble models with exponential moving average fusion and enhanced weighted gradient optimization. Frontiers in Plant Science. 15: 1382416. https://doi.org/10.3389/fpls.2024. 1382416.

  16. Patil, R.G. and More, A. (2025). A comparative study and optimization of deep learning models for grape leaf disease identification. Indian Journal of Agricultural Research. 59(4): 654- 663. doi: 10.18805/IJARe.A-6242.

  17. Peng, D., Li, W., Zhao, H., Zhou, G. and Cai, C. (2023). Recognition of tomato leaf diseases based on DIMPCNET. Agronomy. 13(7): 1812. https://doi.org/10.3390/agronomy13071812.

  18. PlantVillage Dataset. Accessed: May. 2, 2024. [Online]. Available: https://www.kaggle.com/abdallahalidev/plantvillage- datase.Ā 

  19. Rangarajan, A.K., Purushothaman, R. and Ramesh, A. (2018). Tomato crop disease classification using pre-trained deep learning algorithm. Procedia Computer Science. 133: 1040-1047. https://doi.org/10.1016/j.procs.2018.07.070.

  20. Scharr, H., Pridmore, T.P. and Tsaftaris, S.A. (2017). Computer Vision Problems in Plant Phenotyping, CVPPP 2017- Introduction to the CVPPP 2017 Workshop Papers. In Proceedings of the IEEE International Conference on Computer Vision Workshops. (pp. 2020-2021). https:// doi.org/10.1109/ICCVW.2017.236.Ā 

  21. Siddiky, M.N.A.A., Batra, S., Polok, S.D., Hasan, M.N., Apon, M.O.I., Taki, T.A. and Faisal, F.R. (2024). Performance Assessment of Various Deep Learning-Based Models for Multiclass Classification of Tomato Leaf Diseases. In 2024 International Conference on Communication, Computer Sciences and Engineering (IC3SE). IEEE. (pp. 384-389). https://doi.org/ 10.1109/IC3SE62002.2024.10593577Ā 

  22. Sharma, J., Al-Huqail, A.A., Almogren, A., Doshi, H., Jayaprakash, B., Bharathi, B., Rehman, A.U. and Hussen, S. (2025). Deep learning-based ensemble model for accurate tomato leaf disease classification by leveraging ResNet50 and MobileNetV2 architectures. Scientific Reports. 15(1): 13904. https://doi.org/10.1038/s41598-025-98015-x.Ā 

  23. Sundararaman, B., Jagdev, S. and Khatri, N. (2023). Transformative role of artificial intelligence in advancing sustainable tomato (Solanum lycopersicum) disease management for global food security: A comprehensive review. Sustainability. 15(11): 11681. https://doi.org/10.3390/ su151511681.

  24. Tan, L., Lu, J. and Jiang, H. (2021). Tomato leaf diseases classification based on leaf images: A comparison between classical machine learning and deep learning methods. Agri Engineering. 3(3): 542-558. https://doi.org/10.3390/ agriengineering3030035.

  25. Tm, P., Pranathi, A., SaiAshritha, K., Chittaragi, N.B. and Koolagudi, S.G. (2018). Tomato Leaf Disease Detection Using Convolutional Neural Networks. In: 2018 Eleventh International Conference on Contemporary Computing (IC3). IEEE. (pp. 1-5). https://doi.org/10.1109/IC3.2018. 8530532.Ā 

  26. Vinothini, A. and Aswiga, R.V. (2025). Transfer learning-based deep learning model for classifying tomato plant leaf diseases. Engineering Research Express. 7(2): 025250. https://doi.org/10.1088/2631-8695/add6f5.Ā 

  27. Wagle, S. and Harikrishnan, R. (2021). A deep learning-based approach in classification and validation of tomato leaf disease. Traitement du Signal. 38(3): 629-638. https:// doi.org/10.18280/ts.380317.

  28. Zhang, K., Wu, Q., Liu, A. and Meng, X. (2018). Can deep learning identify tomato leaf disease? Journal of Electrical and Computer Engineering. 2018: 6710865. https://doi.org/ 10.1155/2018/6710865.

  29. Zhou, C., Zhou, S., Xing, J. and Song, J. (2021). Tomato leaf disease identification by restructured deep residual dense network. IEEE Access. 9: 3058947. https:// doi.org/10.1109/ACCESS.2021.3058947.
Published In
Indian Journal of Agricultural Research
Article Metrics
Metrics Icon
22
Views
0
Citations
Reviewed By
Share Article
Download Article
In this Article
    Contact us
    Follow us

    Full Research Article

    Deep Learning-based Classification of Tomato Leaf Diseases

    V
    Varsha Nemade1,*
    V
    Vishal Fegade2
    D
    Deepti Barhate1
    S
    Suraj Patil1
    1Department of Computer Science, Mukesh Patel School of Technology Management and Engineering Shirpur, SVKM's Narsee Monjee Institute of Management Studies, Deemed-To-University, Mumbai-400 056, Maharashtra, India.
    2Department of Applied Sciences and Humanities, Mukesh Patel School of Technology Management and Engineering Shirpur, SVKM's Narsee Monjee Institute of Management Studies, Deemed-To-University, Mumbai-400 056, Maharashtra, India.

    ABSTRACT

    Background: In the Indian economy agriculture plays important role. Many of the crops are damaged due to diseases, therefore plant leaf disease detection at early stage is important. Tomatoes are the second most consumed vegetable in Indian households, with a rank second largest producer and consumption in world. Tomatoes are of economically important crop in India with large scale cultivation. Yet, unfavourable environmental factors tend to cause numerous diseases due to bacteria, fungi and viruses that infect different plant parts. These diseases cause the yield to be lower, leading to heavy economic losses for the farmers. For effective disease management, it is important to detect the disease in early stage using the advanced techniques for maximising the yielding of crop.

    Methods: In the proposed approach, tomato leaf images were first enhanced using CLAHE to improve local contrast, after that data augmentation to increase dataset diversity. Four transfer learning models based on pretrained CNN architecture: VGG16, VGG19, ResNet50 and MobileNetV2, were used as fixed feature extractors, where the pre-trained layers were fixed and newly added classification layers were trained for tomato leaf disease classification. Although transfer learning has been widely applied for tomato leaf disease classification, comparative investigations on the effect of CLAHE-enhanced images across different pre-trained architectures remain limited. This study shows performance of these four CNN architectures under identical preprocessing and training conditions to analyze the influence of CLAHE-based contrast enhancement on disease feature representation and classification performance.

    Result: The outcomes of experimental results shows strong classification performance on the plant village dataset for detection of tomato leaf disease, achieving best accuracies of 93.75%, 95.75%, 86.38% and 98.54% using the VGG16, VGG19, ResNet50 and MobileNetV2 models respectively.

    INTRODUCTION

    Detecting plant diseases is of the utmost importance as it paves the way for preventing crop diseases and maintaining the crop quality. Tomato is economically important crop, representing a fair share of dietary needs (Sundararaman et al., 2023). It is among the most widely cultivated horticultural crops. Tomatoes are versatile and hence valuable in many food applications. Tomatoes has rich source of nutrients like vitamin C, vitamin K and antioxidants. Usually, tomatoes are grown throughout the summer season but can be grown throughout the year. However, the diversity of diseases attacking the foliar parts of this plant along with other tissues has threatened both productivity and economic viability in tomato farming (Sundararaman et al., 2023; Attallah, 2023). Generally diseases produced from a different pathogens such as fungi, bacteria and viruses (Kaur et al., 2023) and manifest various symptoms; hence, yield loss and economic hardship among farmers cannot be ignored (Attallah et al., 2023; Pandiyaraju et al., 2024). This investigation considers the most prevalent tomato leaf diseases, as shown in Fig 1 a, b, c, d, e respectively. Disease identification rely heavily on expert visual inspection, which are time-consuming, often subjective and not readily scalable to large-scale agricultural operations (Debnath et al., 2023; Chowdhury et al., 2021). Traditional method depends on the on human observers are inefficient, have poor scalability and are unreliable. To address this challenge, different image processing and computer vision techniques have been employed for the automated analysis of plant diseases (Scharr et al., 2017; Al Farid et al., 2022). In addition, machine learning-based intelligent systems have been developed to enable automatic disease detection that provide timely intervention and improved crop management practices that contribute to enhanced agricultural productivity (Islam et al., 2022).

    Fig 1: Tomato leaf diseases.


           
    Deep learning (DL) is a subfield of artificial intelligence and this has drastically transformed the character of image analysis and classification procedures (Wagle and Harikrishnan, 2021; Zhang et al., 2018). DL architecture, in general and the convolutional neural network (CNN) in particular, are self-learning in obtaining complex patterns and features from data in images which may lead to developing dependable and efficient strategies in the identification of diseases on the leaves of the tomato Wagle and Harikrishnan, 2021; Zhou et al., 2021). Many studies have proven the success of CNNs in classifying a wide range of plant diseases, including those of tomato leaves (Attallah et al., 2023; Wagle and Harikrishnan, 2021). For leaf image classification deep learning CNN model are always better choice as compared to conventional ML algorithms as it is computationally efficient, providing highest accuracy and faster processing times. CNN are useful to automatically detect important features from the input images without manual feature extraction or human supervision. This allows them to learn hierarchical patterns and complex structures in the data, making them highly effective for leaf classification. Performance of the model can be improve by image enhancement and augmentation technique.
           
    For the purpose of giving the accurate and efficient method in detecting and predicting the development of tomato leaf diseases. For image enhancement contrast limited adaptive histogram equalization (CLAHE) is use and    augmentation technique use to expand the dataset. After that CNN architectures such as VGG16, VGG19, ResNet50 and MobileNetV2 are used for the disease classification.
           
    Plant village dataset is use in this study for comparing different transfer learning (TL) architectures by using identical preprocessing and training conditions. Although the extensive research on tomato leaf disease classification using transfer learning, most existing studies focuses on evaluating individual CNN architectures or proposing new architectures. Limited research has been given to systematically investigating the impact of CLAHE-based image enhancement across multiple pre-trained architectures under identical experimental settings. This study addresses this gap by providing a comprehensive comparative analysis of VGG16, VGG19, ResNet50 and MobileNetV2, highlighting how contrast enhancement influences feature extraction and classification performance.
           
    Major contributions of this work are:
    1. A systematic evaluation of CLAHE-based contrast enhancement for tomato leaf disease classification using four different TL architectures.
    2. An analysis of how image enhancement and augmentation affect feature learning and classification performance.
    A comparative study of lightweight MobileNetV2 and deep architectures: VGG16, VGG19 and ResNet50 under identical training conditions.
    3. An investigation into architecture-specific behavior for tomato disease recognition, highlighting the strengths and limitations of each model.

    The organization of paper as follows:
           
    The literature review section provides prior research in the field of identification and classification plant disease. The proposed methodology section provides the propose methodology for the classification of tomato leaf disease. The results and discussion section provides an analysis of work. Lastly, the conclusion section summarizes the study.
     
    Literature survey
     
    The traditional methods for detection of plant disease are mainly based on the texture, shape, color and other features of the disease spots for extraction. The technique has low identification efficiency since it relies on a vast amount of expert knowledge in agricultural diseases. Now artificial intelligence (Rangarajan et al., 2018) has proved to be very effective in upgrading the discipline of plant science, ML and DL strategies have emerged as the two most popularly explored methodologies for the detection plant diseases. Most of the existing methods for plant disease analysis are based on disease classification (Mane and Rangarajan, 2017; Patil and More, 2025). Mehta et al., (2025) presented reviews on by using DL frameworks to perform plant disease classification, thereby powering smart farming operations. Nigam and Jain (2020) have evaluated a wide variety of deep learning algorithms for plant disease detection and identification. It was found by them that CV, ML and DL applications greatly increased classification accuracies, which led to more intelligent decision-making in agricultural applications.This section reviews recent and prominent studies focusing on automatic disease identification systems, focusing on novel, design-oriented frameworks and their applications to overcome challenges in this area of study.
           
    Tm et al., (2018) proposed three-phase methodology, which is the collection of images of tomato leaf from plant village dataset, data pre-processing with normalization and disease classification using a modified LeNet CNN. The proposed model obtained accuracy of 94-95% that surpassed the AlexNet and GoogleNet models. ResNet (Ahmad et al., 2020) has more layers of convolution than VGG16, which means it can better extract the details of objects. Its layer-skipping structure skips layers and overcomes the gradient vanishing problem due to deep stacking. VGG16 fails to extract detailed features for tomato leaf diseases (Aversano et al., 2020). The deep residual network uses feed-forward neural network residual connections such that layer outputs become inputs to subsequent layers, hence improving efficiency without adding any variables or considerably increasing computation time.
           
    Deng et al., (2021) utilized generative adversarial networks (GANs) to augment the dataset and trained different network and evaluating their performance on a test set. Kanda et al., (2022) described the use of four diversity levels: depth size, discriminative learning rates, training-validation split ratios and batch sizes. It reached F1 score:99.5%, surpassing the performance of most existing tomato leaf disease recognition methods. Further testing is conducted on the Flavia leaf image dataset to obtain a 99.23% F1 score, signifying the reliability and efficiency of the proposed approach. Peng et al., (2023) has described a new classification network for tomato leaf disease that combines the dense inception MobileNet-V2 with a parallel convolutional block attention module. The original images of five tomato leaf diseases were expanded to 8190 using data augmentation from 1256. An improved bilateral filtering and threshold function algorithm IBFTF was proposed to effectively filter noise out. While Dense Inception focuses on most intra-class variability but negligible inter-class differences, the PCBAM enhanced MobileNet-V2 minimizes the effects of background complexity. Empirical experiments show that the DIMPCNET achieves an accuracy of 94.44% with an F1-score of 0.9475. Siddiky et al., (2024) applied the deep learning and image processing to enhance the detection of diseases and, therefore, food security. Five models were applied: MobileNet, ResNet50V2, Xception, InceptionV3 and VGG19. The five models were trained on 83,568 images of tomato leaves. The best validation accuracy was achieved by MobileNet at 91%, followed by the others. Chowdhury et al., (2021) applied EfficientNet for tomato leaf disease classification in three scenarios: Two-class (healthy vs. unhealthy), six-class labels (segmented) and ten-class labels. The Adam optimizer achieved 99% accuracy for this task. Tan et al., (2021)  compared traditional machine learning techniques including KNN, SVM and RF with deep learning architectures like AlexNet, VGG16, ResNet34, EfficientNet and MobileNetV2 on the plant village dataset. Classical approach methods have achieved the accuracies between 82.10%-91.00%, whereas the deep learning model achieved the highest accuracy between 91.20% to 99.70% . Sharma et al. (2025) proposed deep learning ensemble model based on MobileNetV2 and ResNet50 is presented for classifying the tomato leaf diseases. By training it on a Kaggle dataset consisting of 11,000 images covering 10 disease classes, the model was able to achieve state-of-the-art performance with 99.91% test accuracy. Adding layers and merging of feature maps improved the ability to extract features and reduce false classifications. This strategy enables the early detection of plant diseases, by reducing crop losses and promoting smart and sustainable agricultural practices. Vinothini et al., (2025) developed a transfer learning-based deep learning model for the classification of tomato leaf diseases.In both the balanced setup and imbalanced setup, 4 models (AlexNet, LeNet-5, DenseNet-121 and InceptionV3) were trained using different sampling methods. The enhanced AlexNet-based model presented the best performance with accuracy, precision, recall and F1-score of 95%. This system provided a dependable means for early disease identification to help enable efficient crop management and sustainable agriculture practice. Mezenner et al. (2025) proposed a novel fuzzy aggregation-based CNN ensemble model for tomato leaf disease classification. First, five deep learning models VGG16, ResNet152V2, MobileNetV2, InceptionV3 and DenseNet201 are pre-trained separately where different feature extractors of each model are used. These models are aggregated via a newly proposed fuzzy Max-Sum integral derived from the fuzzy convex combination operator.

    MATERIALS AND METHODS

    The experimental work was carried out at the MPSTME Shirpur, Maharashtra. The study utilized the publicly available PlantVillage dataset for tomato leaf disease classification. The experimental work was conducted during the period from May 2025 to December2025. Diseases of tomato leaves considerably reduce agricultural productivity, causing substantial economic problems to the growers. In this context, we highlight the critical necessity of timely and accurate disease diagnosis by using proposed architecture as shown in Fig 2.

    The proposed method comprises the following components:

    Fig 2: Overall architecture of proposed approach.


     
    Dataset and pre-processing
     
    This study utilized a dataset comprising images of tomato plants affected by various diseases, serving as a valuable foundation for analysis and research. For this work Plant village dataset (PlantVillage, 2024) is used that is available on Kaggle. We choose tomato leaves images of 6 classes. Class 0: Bacterial spot, Class 1: Early blight, Class 2: Healthy, Class 3: Septoril leaf spot, Class 4: Leaf mold, Class 5: Yellow leaf curl virus, there count is as shown in Fig 3, out of this 7245 images used for training, 1355 images used for validation and 448 images used for testing.

    Fig 3: Class wise image count in original dataset.


           
    Pre-processing of the images is important step before to applying the actual model, as it helps to further improve the effectiveness of the model. In this work CLAHE is used as described in Algorithm 1. This technique applied on all images in the dataset. It enhances disease features like spots, discoloration, or blight, making them more prominent, reduces the effects of uneven lighting in image and help to improves model performance, especially for models relying on color and texture features. Fig 4 shows tomato leaf images before and after applying CLAHE for each class. This approach was chosen since tomato leaf disease is usually described through various changes in texture, coloring and lesion boundaries, which may not be sufficiently reflected in the image itself. The use of CLAHE will help improve the local contrast of an image without over-amplifying the noise present.

    Fig 4: Before and after applying CLAHE.


           
    When we are working with limited data size data augment helps to increase the size of the original dataset. This approach helps strengthen the model’s ability to generalize effectively, reduce overfitting and speed up the training. The data augmentation was only done on the training dataset, not on the validation and test datasets, in order to measure the performance of the model precisely. The same technique had been proved repeatedly to be enhancing the performance of deep learning models and their generalization. The augmentations that were done by using height shift, width shift, rotation and zooming. After augmentation training images increases from 7245 to 26658.    
     
    Algorithm 1: CLAHE algorithm.
     
    Input  
     
    I: Input image.   
    N: Number of tiles per row and column (N × N).   
    Cliplimit: Maximum allowed histogram count before clipping.   
    NB: Number of bins for histogram.
     
    1. Initialize parameters
     
    Divide the image I into (N × N). 
    Ti,j= The tile at position (i, j).
     
    2. For each tile Ti,j
     
    Calculate histogram Hi,j of the tile using NB bins.
     
    Clip the histogram 
        
    If any bin Hi,j (k) > cliplimit, redistribute excess counts equally to all bins while maintaining the total count.
     
    Normalize the histogram:

     
    Calculate the cumulative distribution function (CDF):

     
    Map the pixel values of Ti,j to the enhanced range:
     
     
    Where, 
    Pold= The original pixel value.
    Pnew= The enhanced value.
     
    3. Perform interpolation
     
    Use bilinear interpolation:
     
    Ienhanced (x, y) = w1. Ti,j + w2. Ti+1 + w3. Ti,j+1 + w4. Ti+1, j+1
     
    Where, 
    w1, w2, w3, w4= The weights based on the relative distance of (x, y) to the centers of the neighboring tiles.
     
    4. Output: Ienhanced: Enhanced image
     
    Deep learning architectures
     
    We employed four pre-trained CNN architectures-VGG16, VGG19, ResNet50 and MobileNetV2-for tomato leaf disease classification. The pre-trained models were initialized with ImageNet weights using the include_ top=False configuration, which removes the original ImageNet classification layer while retaining the convolutional feature extraction layers. Transfer learning was adopted to leverage the generic visual features learned from the large-scale ImageNet dataset. To preserve the generic visual features learned during pre-training and reduce computational complexity, all layers of the pre-trained base networks were frozen during training. Specifically, the VGG16 base model consisted of 19 layers (1 input layer, 13 convolutional layers and 5 max-pooling layers), while the VGG19 base model consisted of 22 layers (1 input layer, 16 convolutional layers and 5 max-pooling layers), all of which were set as non-trainable. Similarly, the complete pre-trained base networks of ResNet50 (175 layers) and MobileNetV2 (154 layers), comprising convolutional, batch normalization, activation, pooling and residual/inverted residual layers, were entirely frozen. Consequently, only the newly added classification head, consisting of a Flatten layer, a Dropout layer (0.6), a Dense layer with 256 neurons using ReLU activation and L2 regularization (0.002) and a Softmax output layer with six neurons, was trained using the tomato leaf dataset. For each architecture, a custom classification head was appended to the frozen base network. The classification head consisted of a flatten layer, followed by a dropout layer with a dropout rate of 0.6 to reduce overfitting. A fully connected dense layer with 256 neurons, ReLU activation function and L2 regularization  was added to improve feature learning and enhance model generalization. Finally, a Softmax output layer with six neurons was used to classify the tomato leaf images into the six disease categories. For training, Adam optimizer was chosen with the learning rates: 0.001 and 0.0001 to observe the impact of varied optimization techniques. With the learning rate of 0.001, the training process converges quickly; however, the use of learning rate 0.0001 provides slow updates to the weights. Due to the fact that the task is about multi-class classification, the categorical cross-entropy loss function was utilized. Each model was trained for a maximum of 100 epochs with early stopping to prevent overfitting. Model performance was evaluated on the test dataset using accuracy, precision, recall and F1-score and the results of all four architectures were compared to identify the most effective model for tomato leaf disease classification.

    RESULTS AND DISCUSSION

    This work assesses the performance of different fine-tuned pre-trained models for the classification of tomato leaf diseases, CLAHE for pre-processing and data augmentation methods to improve the dataset. Table 1 and 2 illustrate the performance of four CNN architectures-VGG16, VGG19, ResNet50 and MobileNetV2-on a classification task with varying learning rates LR=0.001 and 0.0001 respectively. The performance is measured using three common metrics: Precision, recall and F1-score, on six classes. Table 3 shows the accuracy comparison of CNN models at different learning rates. All model shows outstanding performance at LR=0.001. VGG 16, VGG19, ResNet50 and MobileNetV2 achieved 93.75 %,95.75%,86.38% and 98.54% accuracy.

    Table 1: Performance measure (LR =0.001).



    Table 2: Performance measure (LR =0.0001).



    Table 3: Accuracy comparison.


           
    From the above experiment findings, it is clear that deep learning models that have been pretrained are highly effective in detecting tomato leaf diseases. The application of CLAHE pre-processing led to increased image contrast, feature extraction and overall model performance. Data augmentation was applied to solve the problem of imbalanced classes in the data set, hence increasing the overall robustness of the model. From among the models tested, the MobileNetV2 model proved to be the most effective, recording high accuracy and F1-scores.
           
    VGG19 also performed well, especially at high learning rates, showing its capability to work in high precision applications. ResNet50 worked as a feature extractor and could not adapt its deep residual features to the tomato leaf disease dataset, which may have resulted in underfitting and reduced classification performance. ResNet50 functioned as a fixed feature extractor and could not adapt its deep residual features to the tomato leaf disease dataset, which may have resulted in underfitting and reduced classification performance. Learning rate was equally critical in optimizing the model as well. High learning rates were found to give better performance in all cases, showing faster optimization without overfitting, while low learning rates gave poorer performances especially with minority classes.

    CONCLUSION

    The present work highlights the classification of tomato leaf diseases using DL algorithms which is essential in prevention of agriculture losses.  CLAHE technique used for pre-processing and data augmentation techniques for expanding the dataset. The suggested method achieved impressive levels of classification accuracy. MobileNetV2 algorithm provided the best results by obtaining accuracy level of 98.54% which makes it extremely suitable for low resource settings such as mobile devices.
           
    Even though the promising results were obtained in this research, there are certain limitations of this research. Experiments were conducted on the PlantVillage dataset which consists of laboratory acquired images and thus does not entirely represent the variety of images in a real-life agricultural setting. Further research will include conducting experiments on tomato leaf datasets obtained in the field, studying the influence of partial fine-tuning of high layers, using explainable artificial intelligence methods such as Grad-CAM to increase the model interpretability.

    CONFLICT OF INTEREST

    The authors declare no conflicts of interest of any kind.

    REFERENCES


    1. Al Farid, F., Hashim, N., Abdullah, J., Bhuiyan, M.R., Shahida, M.I.W.N., Uddin, J., Haque, M.A. and Husen, M.N. (2022). A structured and methodological review on vision-based hand gesture recognition system. Journal of Imaging. 8(6): 153. https://doi.org/10.3390/jimaging8060153.

    2. Ahmad, I., Hamid, M., Yousaf, S., Shah, S.T. and Ahmad, M.O. (2020). Optimizing pretrained convolutional neural networks for tomato leaf disease detection. Complexity. 2020(1): 8812019. https://doi.org/10.1155/2020/8812019.

    3. Attallah, O. (2023). Tomato leaf disease classification via compact convolutional neural networks with transfer learning and feature selection. Horticulturae. 9(2): 149. https:// doi.org/10.3390/horticulturae9020149.

    4. Aversano, L., Bernardi, M. L., Cimitile, M., Iammarino, M. and Rondinella, S. (2020). Tomato Diseases Classification Based on VGG and Transfer Learning. In 2020 IEEE International Workshop on Metrology for Agriculture and Forestry (MetroAgriFor). IEEE. (pp. 129-133). 10.1109/ MetroAgriFor50201.2020.9277626

    5. Chowdhury, M.E.H., Rahman, T., Khandakar, A., Ayari, M.A., Khan, A.U., Khan, M.S. et al. (2021). Automatic and reliable leaf disease detection using deep learning techniques. Agriengineering. 3(2): 20. https://doi.org/10.3390/ agriengineering3020020.

    6. Debnath, A., Hasan, M. M., Raihan, M., Samrat, N., Alsulami, M.M., Masud, M. and Bairagi, A.K. (2023). A smartphone-based detection system for tomato leaf disease using EfficientNetV2B2 and its explainability with artificial intelligence (AI). Sensors. 23(21): 8685. https://doi.org/10.3390/s23218685.

    7. Deng, H., Luo, D., Chang, Z., Li, H. and Yang, X. (2021). Rahc_gan: A data augmentation method for tomato leaf disease recognition. Symmetry. 13(9): 1597. https://doi.org/ 10.3390/sym13091597.Ā 

    8. Islam, M.N., Ahmed, F., Ahammed, M.T., Rashid, M. and Bari, B.S. (2022). Rice Disease Identification Through Leaf Image and IoT-Based Smart Rice Field Monitoring System. In: Enabling Industry 4.0 through advances in mechatronics. Springer. (pp. 529-539). https://doi.org/10.1007/978-981- 19-2095-0_45.Ā 

    9. Kanda, P.S., Xia, K., Kyslytysna, A. and Owoola, E.O. (2022). Tomato leaf disease recognition on leaf images based on fine-tuned residual neural networks. Plants. 11(21): 2935. https://doi.org/10.3390/plants11212935.Ā 

    10. Kaur, P., Singh, M. P., Mishra, A., Shankar, A., Singh, P., Diwakar, M.Ā et al. (2023). DELM: Deep ensemble learning model for multiclass classification of super-resolution leaf disease images. Turkish Journal of Electrical Engineering and Computer Sciences. 31(3): 3123. https://doi.org/ 10.55730/1300-011X.3123.

    11. Mane, D.T. and Rangarajan, U.V. (2017). A survey on supervised convolutional neural networks and its major applications. International Journal of Rough Sets and Data Analysis. 4(3): 71-82. https://doi.org/10.4018/IJRSDA.2017070105.Ā  Ā Ā 

    12. Mehta, A.R., Kumar, P., Prem, G., Aggarwal, S. and Kumar, R. (2025). Leveraging artificial intelligence for disease diagnosis in agricultural crops: A review. Indian Journal of Agricultural Research. 59(5): 681-690. doi: 10.18805/IJARe.A-6363.

    13. Mezenner, A., Nemmour, H., Chibani, Y., and Hafiane, A. (2025). Fuzzy aggregation-based deep learning ensemble for enhanced tomato leaf disease classification. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems. 33(7): 911-932. https://doi.org/10.1142/ S0218488525400124.Ā 

    14. Nigam, S. and Jain, R. (2020). Plant disease identification using deep learning: A review. The Indian Journal of Agricultural Sciences. 90(2): 249-257. doi: 10.56093/ijas.v90i2.98996.

    15. Pandiyaraju, V., Kumar, A.M.S., Praveen, J.I.R., Venkatraman, S., Kumar, S.P., Aravintakshan, S.A., Abeshek, A. and Kannan, A. (2024). Improved tomato leaf disease classification through adaptive ensemble models with exponential moving average fusion and enhanced weighted gradient optimization. Frontiers in Plant Science. 15: 1382416. https://doi.org/10.3389/fpls.2024. 1382416.

    16. Patil, R.G. and More, A. (2025). A comparative study and optimization of deep learning models for grape leaf disease identification. Indian Journal of Agricultural Research. 59(4): 654- 663. doi: 10.18805/IJARe.A-6242.

    17. Peng, D., Li, W., Zhao, H., Zhou, G. and Cai, C. (2023). Recognition of tomato leaf diseases based on DIMPCNET. Agronomy. 13(7): 1812. https://doi.org/10.3390/agronomy13071812.

    18. PlantVillage Dataset. Accessed: May. 2, 2024. [Online]. Available: https://www.kaggle.com/abdallahalidev/plantvillage- datase.Ā 

    19. Rangarajan, A.K., Purushothaman, R. and Ramesh, A. (2018). Tomato crop disease classification using pre-trained deep learning algorithm. Procedia Computer Science. 133: 1040-1047. https://doi.org/10.1016/j.procs.2018.07.070.

    20. Scharr, H., Pridmore, T.P. and Tsaftaris, S.A. (2017). Computer Vision Problems in Plant Phenotyping, CVPPP 2017- Introduction to the CVPPP 2017 Workshop Papers. In Proceedings of the IEEE International Conference on Computer Vision Workshops. (pp. 2020-2021). https:// doi.org/10.1109/ICCVW.2017.236.Ā 

    21. Siddiky, M.N.A.A., Batra, S., Polok, S.D., Hasan, M.N., Apon, M.O.I., Taki, T.A. and Faisal, F.R. (2024). Performance Assessment of Various Deep Learning-Based Models for Multiclass Classification of Tomato Leaf Diseases. In 2024 International Conference on Communication, Computer Sciences and Engineering (IC3SE). IEEE. (pp. 384-389). https://doi.org/ 10.1109/IC3SE62002.2024.10593577Ā 

    22. Sharma, J., Al-Huqail, A.A., Almogren, A., Doshi, H., Jayaprakash, B., Bharathi, B., Rehman, A.U. and Hussen, S. (2025). Deep learning-based ensemble model for accurate tomato leaf disease classification by leveraging ResNet50 and MobileNetV2 architectures. Scientific Reports. 15(1): 13904. https://doi.org/10.1038/s41598-025-98015-x.Ā 

    23. Sundararaman, B., Jagdev, S. and Khatri, N. (2023). Transformative role of artificial intelligence in advancing sustainable tomato (Solanum lycopersicum) disease management for global food security: A comprehensive review. Sustainability. 15(11): 11681. https://doi.org/10.3390/ su151511681.

    24. Tan, L., Lu, J. and Jiang, H. (2021). Tomato leaf diseases classification based on leaf images: A comparison between classical machine learning and deep learning methods. Agri Engineering. 3(3): 542-558. https://doi.org/10.3390/ agriengineering3030035.

    25. Tm, P., Pranathi, A., SaiAshritha, K., Chittaragi, N.B. and Koolagudi, S.G. (2018). Tomato Leaf Disease Detection Using Convolutional Neural Networks. In: 2018 Eleventh International Conference on Contemporary Computing (IC3). IEEE. (pp. 1-5). https://doi.org/10.1109/IC3.2018. 8530532.Ā 

    26. Vinothini, A. and Aswiga, R.V. (2025). Transfer learning-based deep learning model for classifying tomato plant leaf diseases. Engineering Research Express. 7(2): 025250. https://doi.org/10.1088/2631-8695/add6f5.Ā 

    27. Wagle, S. and Harikrishnan, R. (2021). A deep learning-based approach in classification and validation of tomato leaf disease. Traitement du Signal. 38(3): 629-638. https:// doi.org/10.18280/ts.380317.

    28. Zhang, K., Wu, Q., Liu, A. and Meng, X. (2018). Can deep learning identify tomato leaf disease? Journal of Electrical and Computer Engineering. 2018: 6710865. https://doi.org/ 10.1155/2018/6710865.

    29. Zhou, C., Zhou, S., Xing, J. and Song, J. (2021). Tomato leaf disease identification by restructured deep residual dense network. IEEE Access. 9: 3058947. https:// doi.org/10.1109/ACCESS.2021.3058947.
    In this Article
      Contact us
      Follow us
      Published In
      Indian Journal of Agricultural Research

      Editorial Board

      View all (0) Ā