Citation

Mahmood, Aqsa and Anwar, Waheed and Sattar, Hina and Hassan, Syed Rizwan and Sheraz, Muhammad and Chuah, Teong Chee (2025) Deep learning framework using UAV imagery for multi-disease detection in cereal crops. Scientific Reports, 16 (1). ISSN 2045-2322

Text s41598-025-33304-z.pdf - Published Version Restricted to Repository staff only Download (5MB)

Abstract

Agriculture is a cornerstone of the economies of many countries, and wheat is a staple cereal crop that sustains nearly half of the worldwide population. However, production of wheat is highly vulnerable to biotic stress such as pathogens and pests, as well as adverse environmental conditions. These factors significantly affect yield and quality, posing critical threats to food security and economic resilience. Conventional disease detection methods often involve intense human labor, prolonged procedures, and are predisposed to subjectivity. Therefore, the development of an automated, accurate, and real-time disease monitoring system is imperative for modern precision agriculture. We propose a hybrid deep learning based Multi-Disease Detection Framework for Wheat Diseases (MDDM-WD) for the identification of multiple wheat diseases using UAV imagery. The framework leverages the pre-trained VGG-16 convolutional neural network for deep feature extraction via a transfer learning approach. These features are subsequently classified using Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), XGBoost, and Bernoulli Naïve Bayes (BNB) algorithms of machine learning. The model is trained and evaluated on a custom-curated dataset, containing wheat diseases: stripe rust, powdery mildew, scab (Fusarium head blight), and yellow dwarf. Evaluation of experiments demonstrates that the classification performance is enhanced significantly through our hybrid approach, with accuracy ranging from 74 to 97%, precision from 73 to 96%, and recall from 73 to 95.7%. The SVM-based variant of the model achieved the highest performance, yielding 96% precision, 95.7% recall, 96% F1-score, and 97% accuracy. The proposed two-phase fine-tuned system demonstrates its effectiveness and efficiency in detecting multiple wheat diseases. The MDDMWD model offers a resource-efficient and scalable approach for early disease detection, supporting informed decision-making for farmers, agronomists, and policymakers in advancing sustainable agriculture.

Item Type:	Article
Uncontrolled Keywords:	Hybrid deep learning, Machine learning
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	10 Feb 2026 02:16
Last Modified:	10 Feb 2026 02:16
URII:	http://shdl.mmu.edu.my/id/eprint/15267

Downloads

Edit (login required)