Automated Guava Disease Detection Using Transfer Learning With ResNet-101

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Manual disease identification in agriculture is inefficient, time-consuming, and error-prone.
• Sustainable agriculture demands automated, accurate disease detection technologies.
• Guava cultivation faces challenges with disease identification impacting yield and quality.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for automated guava disease detection.
• To leverage transfer learning (TL) with ResNet-101 for enhanced classification accuracy.
• To provide an interpretable and transparent AI solution for plant disease identification.

Main Methods:

• Utilized ResNet-101 architecture for direct image analysis without manual feature extraction.
• Applied data augmentation to a dataset of 3784 images, increasing it to 4632 balanced images.
• Preprocessed data using normalization and resizing; split into 80% training, 10% validation, and 10% testing sets.
• Employed Gradient-weighted Class Activation Mapping (Grad-CAM) for visualization and interpretability.

Main Results:

• Achieved an impressive classification accuracy of 98.48% for guava diseases (Anthracnose, Fruit Fly, Healthy).
• Model performance was evaluated using nine key metrics including precision, recall, F1 score, and specificity.
• Grad-CAM visualizations confirmed the model's focus on relevant diseased areas, ensuring transparency.

Conclusions:

• Deep learning with ResNet-101 offers a highly accurate and efficient method for automated guava disease detection.
• The developed AI technology is a viable and scalable solution for large-scale agriculture and sustainable farming practices.
• The interpretable nature of the model enhances trust and applicability in real-world agricultural settings.