Optimizer Evaluation for Maize Leaf Disease Using Transfer Learning with MobileNetV3-Small

Manual identification of maize leaf disease presents significant challenges, including time- consuming processes, dependence on expert availability, and a high risk of misdiagnosis due to similar symptoms among different diseases. These limitations often lead to delays in disease management, unstable crop yields, and economic losses for farmers. This study aims to address these issues by evaluating the performance of different optimizers in classifying maize leaf disease using transfer learning with the MobileNetV3-Small architecture. A total of 2,850 images of maize leaf disease were used and divided into training, validation, and testing sets. Model evaluation involved systematically comparing the Adam, RMSprop, and SGD optimizers by training each configuration under identical conditions and assessing the resulting model performance. The results show that the RMSprop optimizer provides the best performance with 92.98% accuracy, 93.08% precision, 92.98% recall, and 92.98% F1-score. Based on the evaluation, selecting an appropriate optimizer is essential to improve accuracy and reliability of transfer learning models in maize leaf disease classification. These findings highlight the potential to advance smart agricultural systems by enabling more accurate disease detection, which can reduce crop failure risks and enhance disease management in maize production.