Disease Prediction in Millet Using Deep Learning: A Custom VGG-19 Based Approach

Abstract

Millet crops suffer from various diseases induced by fungi, bacteria, and viruses, which have a significant impact on crop yields and overall product quality. Thus, precise and early disease recognition is crucial to ensure food production sustainability and security. For this purpose, it gives an idea of developing a deep learning model which will enable automatic disease detection and classification in millets. In particular, a customized 19-layered VGG network architecture with an optimized preprocessing pipeline was selected as the backbone of the proposed model. The latter involves the following steps: image downsizing (resizing to 224×224 pixels), normalization, denoising by applying a Gaussian filter, and advanced data aug-mentation techniques such as image flipping, rotation, zooming, and adjusting brightness. Moreover, the proposed model features a custom-designed VGG19 architecture based on modifications of the classic one, including adding fine-tuned convolutional layers, dropout regularization, and fully connected layers to enhance feature extraction and eliminate the risk of overfitting. The experiments involved training and testing the model on the dataset of different millet leaf images with different diseases. The obtained experimental results prove the reliability and efficiency of the developed algorithm since the test accuracy equals 99.29%, while the test loss was measured at 0.0045. Besides, the model has high classification metrics with a precision rate of 0.98, a recall of 1.00, and an F1-score of 0.99, which suggests that the algorithm has minimal rates of false negatives. In order to ensure the superiority of the algorithm in question, a comparative analysis is performed with other popular deep learning architectures, namely, standard CNN, GoogLeNet, VGG16, VGG19, DenseNet, ResNet50, ResNet101, MobileNetV2, and ConvNeXt. As a result of the analysis, all models show worse performance compared to the custom model.

Introduction

This paper presents a deep learning-based system for millet disease detection and classification using a customized VGG19 convolutional neural network (CNN). The goal is to provide farmers with a fast, accurate, and automated method for identifying diseases in millet crops, improving crop health and food security.

1. Background and Motivation

Millets are highly nutritious crops that are resistant to climate change, making them essential for food security, especially in regions with harsh growing conditions. However, millet crops are vulnerable to diseases such as:

• Leaf blast
• Rust
• Downy mildew
• Smut

Traditional disease detection relies on visual inspection by agricultural experts, which is:

• Time-consuming
• Subjective
• Error-prone
• Often unavailable to farmers in remote areas

With the widespread availability of smartphones and imaging technologies, deep learning offers an effective solution for automated disease diagnosis.

2. Proposed Solution

The study introduces a customized VGG19-based deep learning model designed specifically for millet disease detection.

Key Modifications

The model improves the standard VGG19 architecture by:

• Optimizing convolutional layers
• Adding dropout layers to reduce overfitting
• Enhancing fully connected layers for better classification

Image Preprocessing Pipeline

To improve performance, the following preprocessing techniques are applied:

• Image resizing to 224 × 224 pixels
• Normalization
• Gaussian noise reduction
• Data augmentation:
  • Rotation
  • Zooming
  • Flipping
  • Brightness adjustment

These techniques increase model robustness and reduce class imbalance.

3. Main Contributions

The study provides several innovations:

Customized VGG19 Architecture

A specialized 19-layer VGG19 model tailored for millet disease recognition.

Enhanced Data Processing

A preprocessing and augmentation pipeline that improves generalization and model reliability.

Superior Performance

The proposed model achieves:

The model outperforms several popular deep learning architectures, including:

• VGG16
• Standard VGG19
• ResNet50
• ResNet101
• DenseNet
• MobileNetV2
• GoogLeNet
• ConvNeXt
• Traditional CNN models

4. Literature Review

Recent advances in plant disease detection have increasingly used:

• Convolutional Neural Networks (CNNs)
• Transfer Learning
• Hybrid Deep Learning Models
• UAV-based crop monitoring
• IoT-integrated disease detection systems

Key findings from previous studies include:

• CNNs are highly effective for plant disease classification.
• Ensemble models can achieve accuracies above 98%.
• Transfer learning improves performance on limited datasets.
• Hybrid CNN-LSTM architectures support disease prediction.
• Real-time AI and IoT systems enhance field deployment.

Despite these advances, challenges remain:

• Dataset variability
• High computational requirements
• Real-time deployment limitations
• Lack of specialization for millet diseases

The proposed model aims to address these limitations through a millet-specific optimized architecture.

5. Dataset Description

The study uses the MilletCropHealth Dataset, which contains high-resolution images of millet leaves classified into three categories:

Classes

1. Healthy
   • Normal green leaves without disease symptoms.
2. Blast
   • A fungal disease causing dark lesions with yellow halos.
3. Rust
   • A fungal infection characterized by reddish-brown pustules on leaves.

Dataset Statistics

Class labels:

• Healthy = 0
• Blast = 1
• Rust = 2

The dataset is balanced and includes real-world disease variations, allowing the model to learn meaningful visual patterns and generalize effectively.

Conclusion

This analysis presents a robust and efficient framework in deep learning which is based on a modified VGG19 architecture for automated millet disease classification. The proposed sys-tem is capable of categorizing millet leaf images into three distinct classes: Healthy, Blast, and Rust. By incorporat-ing a comprehensive preprocessing pipeline, including im-age resizing, normalization, Gaussian-based noise reduction, and extensive data augmentation, the framework effectively enhances the representation of features and boosts generalization performance. The demonstration of Experimental evaluation of the proposed model achieves a high testing accuracy of 99.29% along with a low testing loss of 0.0045, indicating strong stability and minimal prediction error. In addition, the model attains excellent classification performance with the recall of 1.00, an F1-score of 0.99 and precision of 0.98. Comparative analysis against established a deep learning models, for example CNN, VGG16, ResNet variants, DenseNet, MobileNetV2, and ConvNeXt, confirms the superior performance of the modified VGG19 framework. Moreover, statistical validation through p-value analysis supports the significance and reliability of the obtained results. The high recall score ensures that disease in-stances are accurately identified, minimizing the risk of missed detections, which is critical in agricultural applications for preventing yield loss. The proposed model also demonstrates computational efficiency, making it suitable for deployment in real-world precision agriculture environments. In summary, this study emphasizes the efficacy of deep learning methods in improving the diagnosis of crop diseases. Prospective research avenues involve broadening the framework to accommodate multi-crop disease classification, facilitating real-time implementation on mobile or edge devices, and integrating sophisticated mechanisms like attention modules and hybrid architectures to enhance scalability and performance as well.

References

[1] W. Shafik, A. Tufail, A. Namoun, L. C. De Silva, and R. A. A. H. M. Apong, “A systematic literature review on plant disease detection: motivations, classification techniques, datasets, challenges, and future trends,” IEEE Access, vol. 11, pp. 59,174–59,203, 2023. [2] W. B. Demilie, “Plant disease detection and classification techniques: a comparative study of the performances,” Journal of Big Data, vol. 11, no. 1, p. 5, 2024. [3] I. Johnson, X. A. Mary, A. P. Winifred Raj, J. Chalmers, and M. Karthikeyan, “Deep-millet: a deep learning model for pearl millet disease identification to envisage precision agriculture,” Environmental Research Communications, vol. 6, no. 10, p. 105031, 2024. [4] M. Xu, J.-E. Park, J. Lee, J. Yang, and S. Yoon, “Plant disease recognition datasets in the age of deep learning: challenges and opportunities,” Frontiers in Plant Science, vol. 15, p. 1452551, 2024. [5] A. Divya, D. Sungeetha, and S. Ramesh, “Efficient protection of golden nutri cereal implementing enhanced disease identification injecting hybrid split-attention mechanism through novel sandnes mechanism,” Multimedia Tools and Applications, vol. 83, no. 38, pp. 85,499–85,521, 2024. [6] S. A. A. Qadri, N.-F. Huang, T. M. Wani, and S. A. Bhat, “Advances and challenges in computer vision for image-based plant disease detection: a comprehensive survey of machine and deep learning approaches,” IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 2639–2670, 2024. [7] R. Manoharan, S. Asthana, C. Somanathan Nair, T. Gokhale, D. Nishanth, A. Jaleel, and N. Sood, “Exploring the hidden treasure in arid regions: pseudocereals as sustainable, climate-resilient crops for food security,” Frontiers in Plant Science, vol. 16, p. 1662267, 2025. [8] B. Prajwal, S. Mondal, M. Raghu, and N. Shashank, “Disease detection in finger millet plant using fine-tuned transfer learning models,” in 2025 International Conference on Computing for Sustainability and Intelligent Future (COMP-SIF). IEEE, 2025, pp. 1–6. [9] R. Nalwanga, L. Spreeuwers, E. Talavera, and G. Owomugisha, “Multi-crop disease detection in computer vision for resource-constrained farms—A review,” IEEE Access, vol. 13, pp. 217,265–217,284, 2025. [10] S. Wang, D. Xu, H. Liang, Y. Bai, X. Li, J. Zhou, C. Su, and W. Wei, “Advances in deep learning applications for plant disease and pest detection: A review,” Remote Sensing, vol. 17, no. 4, p. 698, 2025. [11] M. I. Hossen, M. Awrangjeb, S. Pan, and A. A. Mamun, “Transfer learning in agriculture: A review,” Artificial Intelligence Review, vol. 58, no. 4, 2025. [12] S. M. Devi, S. Anitha, L. R. Pillai, and J. Kane-Potaka, “Insight on millet: A potential candidate for climate-resilient food system development,” in Millets: A Sustainable and Nutritious Crop. Springer, 2026, pp. 547–567. [13] B. BK, S. Dahal, M. Koirala, R. Poudel, and B. P. Kandel, “Role of millets for food security under climate change,” Plant-Environment Interactions, vol. 7, no. 1, p. e70128, 2026. [14] S. Naveena and K. Kavitha, “EnsembleNet model: A hybrid approach for classifying Gossypium herbaceum leaf diseases using image feature analysis,” Engineering Applications of Artificial Intelligence, vol. 164, p. 113225, 2026. [15] E. Musazade, I. I. Mrisho, and X. Feng, “Deep learning-driven smart and sustainable agriculture,” in Deep Learning in Plant Science and Agricultural Biotechnology. CABI GB, 2026, pp. 315–355.

Copyright

Copyright © 2026 Mr. Rama Krushna Padhy, Dr. K. Siva Krishna. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.