Image Augmentation for Passionfruit Disease Classification Using Conditional Generative Adversarial Networks (cGANs)
Al-Nahrain Journal of Science · 16.03.2026 · 29 (1) · 117-127
DOI: 10.22401/anjs.29.1.12 ↗Kurz gefasst
Fachartikel in der Zeitschrift Al-Nahrain Journal of Science, online erschienen am 16.03.2026. Verfasst von Khamael Al-Dulaimi und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Early and accurate detection of crop diseases is vital for safeguarding food security and improving agricultural productivity, particularly in regions where access to the prognosis and diagnostic tools as well as expert knowledge is remote. This paper presents the first targeted application of class-conditional Generative Adversarial Networks (cGANs) to passionfruit disease detection using the publicly available Makerere Passionfruit dataset, which contains 3,001 labelled images across three classes: Healthy, Brownspot, and Woodiness. We propose a domain-specific augmentation pipeline that integrates progressive growing, spectral normalization, hinge loss, and gradient penalty to generate high-quality, class-specific synthetic images for under-represented disease categories. Experimental results demonstrate that cGAN augmentation significantly enhances minority class recognition. The F1-score for Brownspot improved from 0.72 to 0.78 and for Woodiness from 0.60 to 0.71, raising the macro-average F1-score from 0.76 to 0.80. To validate robustness, the framework is evaluated across multiple architectures, including LeNet, ResNet-18, and MobileNetV2. In each case, cGAN-based augmentation consistently improved macro-F1 scores by 3–5% without reducing overall accuracy, confirming the model-agnostic nature of the approach. These findings establish a novel and practical foundation for AI-powered plant/crop disease detection in realistic field conditions. The framework is computationally efficient and suitable for deployment on mobile or edge devices, making it particularly relevant for low-resource and inaccessible agricultural environments.
Zitieren
APA
Al-Dulaimi, K., & Banks, J. (2026). Image Augmentation for Passionfruit Disease Classification Using Conditional Generative Adversarial Networks (cGANs). Al-Nahrain Journal of Science, 29(1), 117-127. https://doi.org/10.22401/anjs.29.1.12
Vancouver
Al-Dulaimi K, Banks J. Image Augmentation for Passionfruit Disease Classification Using Conditional Generative Adversarial Networks (cGANs). Al-Nahrain Journal of Science. 2026;29(1):117-127. doi:10.22401/anjs.29.1.12
BibTeX
@article{AlDulaimi2026,
author = {Al-Dulaimi, Khamael and Banks, Jasmine},
title = {Image Augmentation for Passionfruit Disease Classification Using Conditional Generative Adversarial Networks (cGANs)},
journal = {Al-Nahrain Journal of Science},
year = {2026},
volume = {29},
number = {1},
pages = {117--127},
doi = {10.22401/anjs.29.1.12}
}
EUVIMED