Deep learning techniques for early detection and classification of leaf diseases in crops

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Sustainable agriculture requires accurate crop monitoring to mitigate yield losses from plant diseases.
• Automated disease detection is crucial for timely intervention and maintaining food safety.

Purpose of the Study:

• To develop and evaluate a deep learning framework for automated detection and classification of tomato and soybean leaf diseases.
• To enhance model interpretability using explainable AI techniques.

Main Methods:

• Utilized Convolutional Neural Network (CNN) models (DenseNet121, MobileNetV2, InceptionV3) for classification.
• Employed YOLOv12 for object detection and Gradient-Weighted Class Activation Mapping (Grad-CAM) for interpretability.
• Developed a novel Hybrid Attention-Based Stacking Ensemble Model incorporating attention mechanisms (CBAM, spatial attention) and diverse CNN architectures (ResNet152V2, VGG19, EfficientNetB0).

Main Results:

• Individual CNN models achieved high classification accuracies (up to 99.94%).
• YOLOv12 demonstrated strong object detection performance with a mean average precision (mAP) of 99.5%.
• The hybrid ensemble model reached 99.18% accuracy, showcasing improved feature learning and attention integration. Grad-CAM visualizations confirmed accurate localization of disease-affected regions.

Conclusions:

• The proposed deep learning framework demonstrates high accuracy, robustness, and interpretability for plant disease detection.
• Integration of attention mechanisms and explainable AI enhances model reliability for agricultural applications.
• The framework shows significant potential for real-time agricultural monitoring, with future validation needed for diverse crops and field conditions.