Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{204664,
        author = {Anurag Anand and RAJ KUMAR SHARMA},
        title = {Automated Disease Prediction in Pomegranate Orchards Using Deep Learning: A Four-Model Comparative Study with YOLOv11, EfficientNetB3, ResNet-50, and VGG-16},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {3625-3635},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=204664},
        abstract = {Pomegranate (Punica granatum L.) is a high-value horticultural crop widely cultivated across the semi-arid belts of Gujarat, Maharashtra, and Rajasthan in India. The crop is threatened by several recurring diseases that, if left undetected, can reduce annual yield by 20–45%. Conventional scouting methods are resource-intensive and inherently subjective, motivating the development of automated, vision-based diagnostic tools. This study presents a rigorous comparative evaluation of four deep learning architectures — YOLOv11, EfficientNetB3, ResNet-50, and VGG-16 — for pomegranate disease classification across five categories: Bacterial Blight (Xanthomonas axonopodis pv. punicae), Anthracnose (Colletotrichum gloeosporioides), Cercospora Leaf Spot (Cercospora punicae), Fruit Borer (Deudorix isocrates), and Alternaria Fruit Rot (Alternaria alternata). A purpose-built dataset of 843 field-collected images was assembled under heterogeneous environmental conditions — spanning five orchards, three geographic locations, four lighting scenarios, and two seasons — and subjected to a detailed augmentation pipeline to maximise data diversity. All models were trained for 100 epochs under controlled, reproducible conditions, and evaluated on a stratified held-out test set of 87 images. YOLOv11 achieved the highest performance across all metrics: accuracy 96.1%, precision 95.9%, recall 95.7%, F1-score 95.8%, and [email protected] 97.4% at an inference latency of only 18 ms. EfficientNetB3, added to extend the comparative scope, attained 94.7% accuracy — the second-best result. ResNet-50 and VGG-16 reached 92.4% and 89.1% respectively. These findings confirm that modern single-stage detection architectures outperform traditional classification CNNs in agricultural disease recognition tasks, even with moderate dataset sizes, provided rigorous augmentation is applied.},
        keywords = {pomegranate disease detection; YOLOv11; EfficientNetB3; ResNet-50; VGG-16; deep learning; transfer learning; [email protected]; agricultural image analysis},
        month = {June},
        }