<![CDATA[Maize is a strategic food commodity in Indonesia; however, its productivity is often reduced by leaf diseases such as Common Rust, Gray Leaf Spot, and Bacterial Leaf Streak. Manual identification is subjective and prone to error due to the visual similarity of symptoms. This study aims to develop an automated maize leaf disease detection system using a Convolutional Neural Network (CNN) based on the MobileNetV2 architecture with a hyperparameter fine-tuning approach. The model was trained and evaluated using the Seasonal Corn Leaf Disease Dataset consisting of 2,943 images across five classes. Data preprocessing included resizing to 224×224 pixels, normalization, and stratified data splitting. The experiment focused on evaluating three augmentation strength levels (0.15, 0.25, and 0.35) using a two-phase training strategy. The results show that the optimal configuration was achieved at augmentation strength 0.25, producing an accuracy of 97.51% and an F1-score of 0.971 on 422 test images. The confusion matrix analysis indicates that most misclassifications occur between visually similar classes, particularly Common Rust and Gray Leaf Spot. The best-performing model was successfully deployed in a Flask-based web application, enabling real-time disease detection with an average inference time of 1.8 seconds per image. These findings demonstrate that appropriate augmentation strength significantly improves model generalization and that the proposed system provides a practical and accessible solution for supporting smart agriculture in Indonesia.]]>
