A Two-Step CNN-Based Approach for Banana Leaf Disease Classification Using ResNet-50 Optimization with Weight Adjustment and Class Balancing

Abstract

The proposed study introduces a novel two-step convolutional neural network (CNN)-based classification approach using ResNet-50 with class weighting and Synthetic Minority Over-sampling Technique (SMOTE) augmentation to improve banana leaf disease classification accuracy while addressing class imbalance issues. Unlike conventional single-stage models, the proposed hierarchical approach first distinguishes between healthy and diseased leaves, followed by specific disease classification, ensuring more reliable predictions. The binary classification model achieves 98.40% accuracy, while the multi-class classification model reaches 91.92% accuracy, demonstrating the effectiveness of weight adjustment and synthetic data augmentation. The proposed model mitigates bias by leveraging class weighting in binary classification and SMOTE for underrepresented disease classes, enhancing the recognition of rare banana diseases. Furthermore, the model outperforms existing CNN-based plant disease classification methods, providing robust generalization and high precision in disease identification. The experimental results validate that the two-step approach offers a more structured and efficient way of handling imbalanced plant disease datasets, improving detection accuracy. The presented approach outperforms the traditional deep learning models, showing its relevance in the context of precision agriculture in the real world. The results point to the fact that AI-based plant disease recognition systems could be incorporated into mobile or web applications to monitor diseases in real time, guarantee early interventions, reduce losses in yield, and make sustainable agriculture practices possible.