Toward Reliable Tea Leaf Disease Diagnosis Using Deep Learning Model: Enhancing Robustness With Explainable AI and Adversarial Training

High accuracy achieved for disease detection: The study demonstrated that deep learning models can effectively identify tea leaf diseases, with EfficientNetB3 reaching 93% accuracy in classification across seven categories (six diseases plus healthy leaves), suggesting automated diagnosis could reduce manual detection errors.

Robustness through adversarial training improves real-world reliability: The researchers enhanced model reliability by using adversarial training, which prepares the system to handle noisy or disturbed inputs—a critical consideration for deploying AI systems in real agricultural settings where image quality may vary.

Explainability makes AI decisions trustworthy: The use of Grad-CAM visualization allowed researchers to identify which specific regions of leaf images influenced the model's predictions, addressing a key responsible AI concern by making the "black box" decision-making process transparent and verifiable.

Practical economic impact for developing economies: By automating tea leaf disease detection in Bangladesh, this technology could help farmers quickly identify crop problems, potentially reducing production losses and supporting the country's economy without requiring specialized expertise.

Dataset size and diversity matter for responsible deployment: The study used 5,278 high-resolution images across multiple disease categories, demonstrating the importance of adequate, representative training data for building AI systems that work reliably across different real-world scenarios.