Deep Learning for Classifying Tenera and Dura Oil Palm Using ResNet-50

Abstract

The plantation sector significantly contributes to Indonesia’s economy, with oil palm being a leading commodity in both domestic and international markets. Accurate identification of oil palm fruit varieties, particularly Dura and Tenera, is crucial for maximizing productivity and profitability. However, conventional identification methods are still manual, time-consuming, and prone to human error. This study proposes an automated classification approach using the ResNet-50 deep learning architecture to classify oil palm fruit varieties, and compares the performance of two optimizers: ADAM and SGD. A dataset of 2,000 images representing Dura and Tenera varieties was collected and augmented to enhance training diversity. The images were preprocessed and resized to 224×224 pixels before being input into the model. Experiments were conducted with variations in learning rate, batch size, and dense layers. Model performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix. The results show that the ResNet-50 model trained with the ADAM optimizer achieved the best performance, with 85% testing accuracy and an F1-score of 0.85, outperforming the SGD optimizer. These findings demonstrate that combining ResNet-50 with ADAM results in better generalization and training efficiency. This research supports the development of a more accurate and efficient classification system for oil palm management, with potential for broader application in agricultural automation.