<![CDATA[Corn leaf diseases pose a significant threat to agricultural productivity, necessitating accurate and efficient detection methods. While extensive deep learning research has been applied to plant disease classification, comparative studies of Hyperparameter Optimization (HPO) frameworks on lightweight architectures such as MobileNetV3 remain scarce. This study aims to compare the performance of Optuna and Hyperopt in optimizing the hyperparameters of MobileNetV3 for corn leaf disease classification. The Kaggle Corn Leaf Disease dataset, comprising 4,000 images across four classes (leaf blight, leaf spot, rust, and healthy), was split 80:10:10 for training, validation, and testing. The hyperparameter search space encompassed learning rate, optimizer type, batch size, and dropout rate, with each framework limited to 15 trials. Results indicate that for MobileNetV3-Small, Optuna achieved an accuracy of 96.75% (F1: 96.66%), marginally outperforming Hyperopt (96.50%; F1: 96.39%). Conversely, for MobileNetV3-Large, Hyperopt demonstrated superior performance with an accuracy of 97.00% (F1: 96.96%) compared to Optuna (96.25%; F1: 96.14%). The Wilcoxon signed-rank test revealed no statistically significant difference between the two frameworks (p > 0.05). These findings suggest that both frameworks are statistically equivalent, and MobileNetV3-Small offers the most favorable balance between classification accuracy and computational efficiency for mobile-based plant disease detection applications.]]>
Copyrights © 2026
