Potato Leaf Disease Classification Using MobileNetV3 Architecture With Adam and Stochastic Gradient Descent Optimizers

Abstrak

Potato leaf diseases such as Early Blight and Late Blight reduced productivity and could cause crop failure if they were not detected early. This study analyzed the comparative performance of the Adam and Stochastic Gradient Descent (SGD) optimizers using the MobileNetV3-Large architecture for potato leaf disease classification. The dataset consisted of three categories: healthy leaves, Early Blight, and Late Blight, with a total of 4,072 images. All images were processed through preprocessing stages, including resizing to 224 × 224 pixels and pixel value normalization. The data were divided into training, validation, and testing sets with a ratio of 70:20:10. Random undersampling and data augmentation techniques were applied to the training data to address class imbalance and improve the model’s generalization capability. The model training process was conducted using a transfer learning approach with the MobileNetV3-Large architecture through two stages: feature extraction and fine-tuning. Model performance evaluation was based on accuracy, precision, recall, and F1-score metrics. The results showed that the Adam optimizer achieved a test accuracy of 98.75% with an F1-score of 0.9875, while the SGD optimizer achieved a test accuracy of 96.56% with an F1-score of 0.9635. The Adam optimizer also demonstrated faster and more stable convergence during the training process. This study was expected to serve as a reference for determining an appropriate optimizer for deep learning applications in image classification, particularly in plant disease detection.