Related Experiment Video
Updated: May 24, 2026
04:17
DNA Virus Detection System Based on RPA-CRISPR/Cas12a-SPM and Deep Learning
Published on: May 10, 2024
Optimized CNN-based ensemble deep learning approach for potato leaf disease detection with data augmentation
Achin Jain1, Arun Kumar Dubey1, Sunil K Singh2
1Department of Information Technology, Bharati Vidyapeeth's College of Engineering, New Delhi, India.
Scientific Reports
|May 18, 2026
Summary
Optimized convolutional neural networks (CNNs) effectively classify potato leaf diseases. Ensemble Deep Learning (EDL10) achieved 97.0% accuracy, demonstrating the power of data balancing and ensemble methods for disease detection.
Area of Science:
- Computer Science
- Agricultural Science
- Plant Pathology
Background:
- Potato diseases like early and late blight significantly impact crop yield.
- Accurate and timely disease detection is crucial for effective management strategies.
- Traditional methods for disease identification can be labor-intensive and subjective.
Purpose of the Study:
- To develop and evaluate optimized Convolutional Neural Networks (CNNs) for classifying potato leaf diseases.
- To compare the performance of different optimizers (ADAM, SGD, RMSPROP, ADAMAX) within CNN models.
- To investigate the effectiveness of Ensemble Deep Learning (EDL) and data augmentation for improving disease classification accuracy.
Main Methods:
- Utilized a dataset of healthy, early blight, and late blight potato leaf images from Kaggle's Plant Village.
- Preprocessed images by resizing to 128x128 pixels and splitting into 80% training and 20% testing sets.
- Designed CNN models with convolutional, pooling, and fully connected layers, trained using sparse categorical cross-entropy loss and early stopping. Implemented data augmentation and an ensemble of four CNN models (EDL10).
Main Results:
- The EDL10 model, an ensemble of four CNNs with different optimizers, achieved the highest accuracy of 97.0%.
- Data augmentation was crucial for balancing the dataset, particularly by increasing the number of healthy leaf images.
- Various performance metrics including accuracy, loss curves, confusion matrix, ROC curve, precision-recall curve, classification report, and F1 score were used for evaluation.
Conclusions:
- Optimized CNNs, especially when combined in an ensemble approach (EDL10), are highly effective for potato leaf disease classification.
- Data balancing through augmentation significantly enhances model performance.
- The study underscores the potential of deep learning for automated and accurate plant disease detection in agriculture.