An Approach Towards Identifying Bangladeshi Leaf Diseases through Transfer Learning and XAI

AI can democratize agricultural expertise: The system achieved 98.90% accuracy in identifying 21 distinct leaf diseases across six plants, potentially providing farmers in Bangladesh with disease detection capabilities comparable to expert agronomists, especially valuable in remote areas where expert knowledge is limited.

Explainability is essential for farmer trust: The researchers used five different explainable AI (XAI) techniques—GradCAM, GradCAM++, LayerCAM, ScoreCAM, and FasterScoreCAM—to show farmers which parts of the leaf the model focused on when making predictions, ensuring transparency rather than treating the AI as a "black box."

Transfer learning makes AI more accessible: By using pre-trained models like VGG19 and Xception rather than building systems from scratch, the researchers created a more efficient solution that could be more feasible to implement and maintain in resource-constrained agricultural settings.

AI solutions must address real-world problems: This work directly targets food security in Bangladesh by reducing plant loss and protecting farmer livelihoods, demonstrating how responsible AI development should begin with understanding specific community needs rather than applying generic solutions.

Transparency enables informed decision-making: By highlighting which leaf regions indicated disease, the system empowers farmers to understand why the AI made its recommendation, allowing them to verify predictions and take appropriate action rather than blindly following algorithmic suggestions.