Optimizing Rice Plant Disease Classification Using Data Augmentation with GANs on Convolutional Neural Networks
Background: Rice disease classification using CNN models faces challenges due to limited data, particularly in minority classes, and inconsistent image quality, which affect model performance. Data augmentation techniques can potentially enhance classification accuracy by improving data diversity a...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universitas Nusantara PGRI Kediri
2025-02-01
|
| Series: | Intensif: Jurnal Ilmiah Penelitian Teknologi dan Penerapan Sistem Informasi |
| Subjects: | |
| Online Access: | https://ojs.unpkediri.ac.id/index.php/intensif/article/view/23834 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background: Rice disease classification using CNN models faces challenges due to limited data, particularly in minority classes, and inconsistent image quality, which affect model performance. Data augmentation techniques can potentially enhance classification accuracy by improving data diversity and quality. Objective: This study evaluates the effectiveness of data augmentation techniques, specifically classical augmentation and Deep Convolutional Generative Adversarial Networks (DCGAN), in improving CNN performance for rice disease classification. Methods: A quantitative study was conducted using four CNN training scenarios: no augmentation, classical augmentation, DCGAN augmentation, and a combination of both. Model accuracy was analyzed to determine the impact of each augmentation technique. Results: The baseline CNN model achieved an accuracy of 91.88%. Classical augmentation improved accuracy by 2.56%, while DCGAN augmentation led to a 5.44% increase. The combination of classical augmentation and DCGAN yielded the highest accuracy of 98.13%. Conclusion: Data augmentation significantly enhances CNN performance in rice disease classification, with the combined approach of classical augmentation and DCGAN proving to be the most effective. These findings highlight the importance of augmentation techniques in addressing data limitations and improving classification accuracy. Future research should explore additional augmentation strategies and test the model across different datasets to further validate its effectiveness.
|
|---|---|
| ISSN: | 2580-409X 2549-6824 |