An explainable deep learning framework for few shot crop disease detection in rice and sugarcane using CNN based...

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Crop health is critical for global food security, necessitating early disease detection.
• Traditional methods for disease identification are often slow and labor-intensive.
• Advancements in image processing and artificial intelligence (AI) offer potential for automated solutions.

Purpose of the Study:

• To propose an AI-powered framework for early detection and forecasting of rice and sugarcane leaf diseases.
• To develop a cost-effective, precise, and efficient automated system for smart farming.
• To enhance decision-making transparency using Explainable Artificial Intelligence (XAI).

Main Methods:

• Image processing techniques (standardization, resizing, normalization) applied to crop images.
• Convolutional Neural Networks (CNNs) used for feature extraction.
• Few-shot learning (FSL) techniques, including Prototypical Networks and Model-Agnostic Meta-Learning (MAML), implemented for classification.
• Integration with Grad-CAM for explainable AI insights.

Main Results:

• The proposed framework demonstrated high accuracy and specificity in disease identification and prediction.
• Achieved up to 97.6% accuracy for rice leaf diseases with Prototypical Networks and 95.27% with MAML.
• Achieved up to 91.68% accuracy for sugarcane leaf diseases with Prototypical Networks and 90.27% with MAML.
• Outperformed state-of-the-art benchmark algorithms in disease prediction.

Conclusions:

• The developed AI framework effectively detects and forecasts rice and sugarcane leaf diseases.
• Few-shot learning techniques significantly improve disease prediction accuracy in smart farming.
• Explainable AI enhances transparency and trust in AI-driven agricultural decision-making.