Interpretable Deep Learning Model for Grape Leaf Disease Classification Based on EfficientNet with Grad-CAM Visualization
Grape leaf diseases pose a significant threat to agricultural productivity, especially in regions with fluctuating climatic conditions that create favorable environments for pathogen growth. Early and accurate disease detection is essential for preventing severe crop losses. Traditional manual inspe...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Pusat Penelitian dan Pengabdian Masyarakat (P3M), Politeknik Negeri Cilacap
2025-06-01
|
| Series: | Journal of Innovation Information Technology and Application |
| Subjects: | |
| Online Access: | https://ejournal.pnc.ac.id/index.php/jinita/article/view/2745 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Grape leaf diseases pose a significant threat to agricultural productivity, especially in regions with fluctuating climatic conditions that create favorable environments for pathogen growth. Early and accurate disease detection is essential for preventing severe crop losses. Traditional manual inspection methods are inefficient and prone to human error, highlighting the need for an automated approach. This study proposes a computer vision-based solution using Convolutional Neural Networks (CNN) improved by EfficientNetB0 to classify grape leaf diseases. The model was trained on a publicly available dataset from Kaggle, which consists of 9,027 images in four classes: ESCA, Leaf Blight, Black Rot, and Healthy. Each image has a resolution of 300 × 300 pixels with a 24-bit color depth, ensuring sufficient detail for analysis. To enhance model performance, data augmentation and hyperparameter tuning were applied. The EfficientNetB0 model was employed due to its strong feature extraction capabilities and computational efficiency. The proposed model achieved 99.36% accuracy, with evaluation metrics including precision (99%), recall (99%), and F1-score (99%), demonstrating its reliability in distinguishing disease categories. Further analysis using a confusion matrix and Grad-CAM visualization provided insights into the model’s decision-making process. The results indicate that this deep learning-based approach is highly effective for grape leaf disease classification. Future research can explore real-time field data collection, attention mechanisms, and self-supervised learning to further improve classification accuracy and model generalization for large-scale agricultural applications. |
|---|---|
| ISSN: | 2716-0858 2715-9248 |