<![CDATA[Plants from the palm tree family (Arecaceae), such as coconut, oil palm, and date palm, play an important role in the economy and food security, especially in Indonesia. However, leaf diseases such as leaf spot disease pose a serious threat that can reduce productivity. Manual disease identification is time-consuming and prone to errors, necessitating an image-based automatic classification system. This study aims to apply the ResNet50 Convolutional Neural Network (CNN) architecture for palm tree leaf disease classification and compare two popular optimization algorithms, Adam and Stochastic Gradient Descent (SGD), in terms of model training accuracy and efficiency. The dataset used is public, covering five classes of leaf images: Healthy, White Scale, Brown Spot, Leaf Smut, and Bacterial Leaf Blight. The research process includes data collection and preprocessing (resizing, normalization, and augmentation), dividing the dataset into three parts, namely training, validation, and testing data using the train/validation/test split approach. This approach provides a fairly representative evaluation of model performance while being computationally efficient. Model training was performed using transfer learning with ResNet50, and performance evaluation was performed using a confusion matrix to obtain accuracy, precision, recall, and F1-score values. The results of the two optimizers were compared to determine their effect on model performance. The experimental results show that the ResNet50 model optimized with Adam achieved a higher test accuracy of 87.23% compared to SGD with 85.96%, while SGD demonstrated more consistent performance between validation and testing phases, indicating better training stability.]]>
