Randomized trial demonstrates enhanced disease classification accuracy in guava using deep learning, suggesting AI's potential for agricultural diagnostics.
Key Points
The aim is to develop an explainable deep learning framework for classifying guava fruit and leaf diseases to improve early detection.
Utilized a dataset of 527 annotated images across five disease classes.
Developed six hybrid deep learning models integrating transfer learning and custom CNN classifiers.
Employed Gradient-weighted Class Activation Mapping for model interpretability.
VGG16 + MobileNetV2 hybrid model achieved 96% accuracy and an F1-score of 0.96.
Confirmed superior performance through confusion matrices and ROC-AUC curves.
The integration of Grad-CAM improved model transparency and trust.