Deep learning framework using UAV imagery for multi-disease detection in cereal crops.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Wheat production is vital for global food security but threatened by diseases and environmental factors.
• Traditional disease detection methods are labor-intensive, time-consuming, and subjective.
• Automated, real-time disease monitoring is crucial for modern precision agriculture.

Purpose of the Study:

• To develop an automated, accurate, and real-time Multi-Disease Detection Framework for Wheat Diseases (MDDM-WD) using UAV imagery.
• To integrate deep learning (VGG-16) with machine learning classifiers for enhanced wheat disease identification.
• To address the limitations of conventional disease detection methods in agriculture.

Main Methods:

• A hybrid deep learning approach utilizing the VGG-16 convolutional neural network for feature extraction via transfer learning.
• Classification of extracted features using Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), XGBoost, and Bernoulli Naïve Bayes (BNB).
• Training and evaluation on a custom dataset of wheat diseases including stripe rust, powdery mildew, scab, and yellow dwarf.

Main Results:

• The hybrid MDDM-WD framework achieved high classification performance, with accuracy ranging from 74% to 97%.
• The SVM-based model variant demonstrated superior performance with 96% precision, 95.7% recall, 96% F1-score, and 97% accuracy.
• The system effectively identified multiple wheat diseases, showcasing significant improvements over conventional methods.

Conclusions:

• The proposed two-phase fine-tuned MDDM-WD system is effective and efficient for early detection of multiple wheat diseases.
• This framework offers a resource-efficient and scalable solution for precision agriculture, aiding farmer decision-making.
• The study supports the advancement of sustainable agriculture through automated disease monitoring.