Explainable AI for Cotton Leaf Disease Classification: A Metaheuristic-Optimized Deep Learning Approach.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Cotton leaf diseases pose a significant threat to global yield and farmer livelihoods.
• Traditional diagnostic methods are often slow, subjective, and not scalable for agricultural monitoring.

Purpose of the Study:

• To develop an interpretable and efficient deep learning (DL) framework for accurate cotton leaf disease classification.
• To enhance model transparency and trustworthiness using explainable AI (XAI) techniques.

Main Methods:

• A hybrid deep learning architecture combining EfficientNetB3 and InceptionResNetV2 was employed.
• Explainable AI (XAI) techniques, specifically LIME and SHAP, were integrated for model interpretability.

Main Results:

• The framework achieved high performance metrics: 98.0% accuracy, 98.1% precision, 97.9% recall, 98.0% F1-score, and an AUC-ROC of 0.9992.
• The model demonstrated minimal overfitting and high per-class performance, even for visually similar diseases.
• XAI techniques successfully highlighted key visual features, enhancing model transparency.

Conclusions:

• The developed DL framework offers a reliable, interpretable, and efficient solution for cotton leaf disease diagnosis.
• The lightweight and scalable model is suitable for deployment on edge devices for real-time precision agriculture applications.
• Combining transfer learning and XAI shows significant potential for developing trustworthy AI diagnostic tools in agriculture.