Automated Detection of Banana Leaf Diseases Using ResNet50

Banana leaf diseases pose significant challenges to global agriculture, necessitating early detection and effective management to prevent substantial economic losses. This study presents a robust approach for classifying banana leaf diseases into four categories: cordana, healthy, pestalotiopsis, and sigatoka. Utilizing a dataset of 1,600 images, split into 80% training and 20% testing, this research leverages transfer learning with the ResNet50 architecture. The images were preprocessed to a uniform size of 224 × 224 pixels, and data augmentation was applied to enhance diversity. The ResNet50 model was fine-tuned by freezing pre-trained layers and adding custom classification layers. The model achieved an overall test accuracy of 96.25, with high precision, recall, and F1-scores across all categories. The confusion matrix revealed minimal misclassifications, underscoring the model's robustness in distinguishing similar disease patterns. Compared to previous studies, this approach demonstrates superior performance in banana leaf disease detection and classification, validating the effectiveness of transfer learning. This research highlights the potential of deep learning in agricultural applications, providing a reliable tool for automated disease management.