Synthetic and Augmented Data Methods for Deep Learning-Based Plant Disease Detection Models
Open Access DepositedImproving Accuracy of Plant Disease Models Using Data Augmentation andSynthetic Data Generation. Deep learning models for plant disease detection often achieve high accuracy ondatasets collected under laboratory conditions but perform less effectively on real-world variability. This research investigates if increasing training image diversity can improve model generalization. Data augmentation was applied to introduce geometric and photometric variation. Generative models were used to produce synthetic images using GAN, diffusion models or prompt-based tools. Baseline models were compared against models trained on augmented, synthetic or blended datasets. Performance of the models was evaluated on both laboratory and real-world test datasets. Synthetic data varied in realism and diversity of images. GAN-based images were the most realistic and diverse. Diffusion-based methods and online prompt-based tools showed limited realism and diversity. Results indicate that neither augmentation nor synthetic generation produced sufficient generalization. Blended datasets produced models that are more generalizable. Explainability analyses using t-SNE and Grad-CAM confirmed differences in data diversity, class separation, and feature focus. Future work should focus on improving realism and diversity of synthetic data along with deployment strategies that enable continual learning with real-world data and human feedback.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.