GAN-Sim-GAN: Generative Data Augmentation for Apple Leaf Disease Classification
IEEE Access · 2026 · 14 · 96590-96601
DOI: 10.1109/access.2026.3706432 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Seyedeh Maryam Moosavi und 2 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Babol Noshirvani University of Technology.
Abstract (Originalsprache)
Data augmentation is essential for improving the generalization of machine learning models, particularly in plant disease detection where annotated datasets are limited and costly to obtain. This paper introduces GAN-Sim-GAN, a generative adversarial network (GAN)-based augmentation pipeline designed for apple leaf disease classification. The pipeline first employs an enhanced cycle generative adversarial network (CycleGAN) to transform healthy soybean leaf images into healthy apple leaf counterparts. Simulation-based data augmentation (SDA) is then applied to synthesize diseased leaves by replicating the texture and morphology of apple scab lesions, using statistical moments to guide the generation process. These synthetic samples are further refined using a GAN model, style-consistent image translation (SCIT), which transfers disease features across classes while preserving source style. The augmented dataset by improved CycleGan was evaluated using t-SNE visualization for distribution analysis. Experimental results demonstrate that SegNet achieved an F1 score of 86.75% on simulated diseased data, while DenseNet classification of SDA-SCIT images yielded an 8% accuracy improvement compared to baseline training. The proposed GAN-Sim-GAN framework effectively expands dataset diversity and enhances disease classification performance, offering a scalable solution for agricultural imaging applications.
Förderung
- Babol Noshirvani University of Technology
Zitieren
APA
Moosavi, S. M., Gholipour, M., & Baleghi, Y. (2026). GAN-Sim-GAN: Generative Data Augmentation for Apple Leaf Disease Classification. IEEE Access, 14, 96590-96601. https://doi.org/10.1109/access.2026.3706432
Vancouver
Moosavi SM, Gholipour M, Baleghi Y. GAN-Sim-GAN: Generative Data Augmentation for Apple Leaf Disease Classification. IEEE Access. 2026;14:96590-96601. doi:10.1109/access.2026.3706432
BibTeX
@article{Moosavi2026,
author = {Moosavi, Seyedeh Maryam and Gholipour, Morteza and Baleghi, Yasser},
title = {GAN-Sim-GAN: Generative Data Augmentation for Apple Leaf Disease Classification},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {96590--96601},
doi = {10.1109/access.2026.3706432}
}
EUVIMED