Novel dual-input stream-based hybrid approach for wheat leaf disease classification using edge-aware features

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Wheat diseases significantly impact global food security by reducing crop yield and quality.
• Accurate and early disease detection is crucial for effective crop management and sustainable agriculture.

Purpose of the Study:

• To propose and evaluate a hybrid deep learning model, EffiXB3, for enhanced wheat crop disease classification.
• To improve the accuracy and robustness of wheat disease identification using advanced AI techniques.

Main Methods:

• Developed a hybrid deep learning (DL) model, EffiXB3, integrating Xception and EfficientNetB3 architectures.
• Employed a dual-input stream architecture processing structural and textural features via Canny edge detection.
• Evaluated model performance on a multi-class classification task with five wheat leaf categories: Blast, Brown Rust, Healthy, Leaf Blight, and Septoria.

Main Results:

• The hybrid EffiXB3 model achieved a classification accuracy of 98.5%, outperforming individual Xception (95%) and EfficientNetB3 (93%) models.
• Integration of edge-aware features significantly enhanced classification performance, especially for visually similar disease patterns.
• The model demonstrated high robustness in differentiating between various wheat leaf conditions.

Conclusions:

• Hybrid DL models, like EffiXB3, incorporating edge-aware features are highly effective for agricultural disease diagnosis.
• EffiXB3 presents a promising tool for improving disease detection in wheat cultivation.
• This research contributes to enhanced crop management strategies and strengthens global food security.