XAI and Few-shot-based Hybrid Classification Model for Plant Leaf Disease Prognosis

AI can work effectively with limited data: The model achieved over 92% accuracy in identifying crop diseases using few-shot learning, demonstrating that AI systems don't always require massive amounts of labeled data—a significant advantage for farmers in resource-limited regions who may not have extensive disease documentation.

Explainability is crucial for agricultural adoption: The paper uses Grad-CAM visualization to show why the AI makes disease classifications by highlighting specific regions of leaf images. This transparency helps farmers and agricultural experts understand and trust the system's recommendations rather than treating it as a "black box."

Hybrid approaches improve reliability: By combining two different learning techniques (Siamese and Prototypical Networks), the model achieved better performance than using either method alone, suggesting that responsible AI development benefits from integrating multiple approaches rather than relying on single solutions.

Real-world applicability matters: The research specifically addresses "data-constrained" conditions and tests on custom datasets for maize, rice, and wheat—the world's most important food crops—making this work directly relevant to actual farming scenarios rather than theoretical exercises.

Responsible AI supports food security: By enabling accurate, early disease detection with limited computational resources and data requirements, this approach could help smallholder farmers prevent crop losses and maintain productivity, directly supporting global food security goals.