A Dual-Stage Deep Learning Framework for Phyllosticta Leaf Disease Severity Estimation in Cardamom Using Lesion Segmentation
Engineering, Technology & Applied Science Research · 11.08.2026 · 16 (4) · 37584-37589
DOI: 10.48084/etasr.19364 ↗Kurz gefasst
Fachartikel in der Zeitschrift Engineering, Technology & Applied Science Research, online erschienen am 11.08.2026. Verfasst von Supreetha Shivaraj und 1 weiteren Autor:innen. Die Arbeit nennt 17 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate estimation of plant disease severity is essential for effective crop health monitoring and precision agriculture. Existing deep learning approaches primarily focus on disease classification while providing limited attention to lesion-level severity quantification and interpretable localization of infected regions. To address this limitation, this study proposes a dual-stage deep learning framework for automated severity estimation of Phyllosticta leaf disease in cardamom using semantic lesion segmentation and Convolutional Neural Network (CNN)-based severity grading. Unlike conventional classification-based approaches, the proposed framework enables continuous severity estimation through lesion-to-leaf area ratio computation combined with categorical severity grading. The framework includes leaf extraction, pixel-level lesion segmentation, quantitative severity estimation using lesion-to-leaf area ratios, and CNN-based severity classification. Experimental evaluation was conducted on a dataset containing 1,444 cardamom leaf images collected under natural field conditions. The proposed framework achieved a lesion segmentation Dice coefficient of 0.914 and a low severity estimation error with a Mean Absolute Error (MAE) of 1.87%. The severity classification network achieved an overall accuracy of 94.2%. Ablation analysis further demonstrates that the proposed dual-stage architecture outperforms single-stage segmentation approaches in lesion localization and severity prediction. The proposed framework provides an interpretable and reliable solution for automated disease severity assessment, supporting early disease detection and precision agriculture applications.
Zitieren
APA
Shivaraj, S., & Haladappa, M. S. (2026). A Dual-Stage Deep Learning Framework for Phyllosticta Leaf Disease Severity Estimation in Cardamom Using Lesion Segmentation. Engineering, Technology & Applied Science Research, 16(4), 37584-37589. https://doi.org/10.48084/etasr.19364
Vancouver
Shivaraj S, Haladappa MS. A Dual-Stage Deep Learning Framework for Phyllosticta Leaf Disease Severity Estimation in Cardamom Using Lesion Segmentation. Engineering, Technology & Applied Science Research. 2026;16(4):37584-37589. doi:10.48084/etasr.19364
BibTeX
@article{Shivaraj2026,
author = {Shivaraj, Supreetha and Haladappa, Manjula Sunkadakatte},
title = {A Dual-Stage Deep Learning Framework for Phyllosticta Leaf Disease Severity Estimation in Cardamom Using Lesion Segmentation},
journal = {Engineering, Technology & Applied Science Research},
year = {2026},
volume = {16},
number = {4},
pages = {37584--37589},
doi = {10.48084/etasr.19364}
}
EUVIMED