MobileNetV2 Optimization with Enhanced Classification Head and Progressive Fine-Tuning Strategy for Banana Leaf Disease Classification

Banana plants are important horticultural commodities that contribute to food security and economic growth in tropical regions. However, their productivity is often affected by leaf diseases such as sigatoka, cordana, and pestalotiopsis. Manual disease identification has several limitations, including subjectivity, longer processing time, and the risk of misclassification. Therefore, this study proposes an optimized MobileNetV2 model with a fine-tuning strategy for image-based banana leaf disease classification using deep learning. The dataset consists of 1,600 images, divided into 80% training data, 10% validation data, and 10% testing data. Data augmentation techniques, including gaussian blur, horizontal flip, linear contrast, cropping, and rotational shear, were applied to improve model performance. The results show that the proposed model achieved 95.82% accuracy, 95.67% precision, 95.74% recall, and a 95.70% F1-score. These results indicate that the model has good generalization capability and stable performance in classifying each disease category. Compared to previous studies, the proposed approach provides competitive and more optimal results. Therefore, the optimization of MobileNetV2 through a fine-tuning strategy is effective and has strong potential for early detection of banana leaf diseases.