Interpretable Plant Leaf Disease Detection Using Attention-Enhanced CNN

Attention mechanisms improve both accuracy and transparency: The model achieved up to 98.87% accuracy while using attention maps (CBAM) to show where the network focuses when identifying diseases, making the AI's decision-making process visible rather than a "black box."

Multiple interpretability tools provide complementary insights: The researchers used four different explanation methods (CBAM attention maps, Grad-CAM, Grad-CAM++, and Layer-wise Relevance Propagation) to validate their findings, demonstrating that responsible AI development benefits from cross-checking results through multiple transparency approaches.

Generalization across diverse datasets matters for real-world deployment: Testing on five different plant disease datasets showed the model's robustness, which is critical for responsible AI—a system that only works on one dataset could fail farmers in different regions or growing conditions.

Explainability enables human oversight in high-stakes decisions: By visualizing which leaf regions the model uses to make diagnoses, farmers and agricultural experts can verify the AI's reasoning and catch potential errors, maintaining human judgment in food security decisions.

Open-source code supports accountability and reproducibility: Publishing the code publicly allows other researchers to audit the work, verify claims, and build upon it responsibly—a best practice for trustworthy AI development in critical applications like agriculture.