Evaluasi dan Perbandingan Model CNN Dan Transfer Learning Dalam Klasifikasi Kematangan Buah Kelapa Sawit

Abstract

The identification of oil palm fruit ripeness is an important factor in maintaining harvest quality and improving palm oil productivity. Manual identification processes still have limitations, including subjectivity and inconsistency in assessment results. This study aims to evaluate and compare the performance of a Baseline Convolutional Neural Network (CNN) model with several transfer learning architectures, namely EfficientNetB3, ResNet50, and DenseNet121, for oil palm fruit ripeness classification. The dataset consisted of 302 original images of oil palm fruits categorized into three classes: ripe, unripe, and rotten. To prevent data leakage, the dataset was first divided using a stratified split into training, validation, and testing sets before data augmentation was applied exclusively to the training set. The augmentation techniques included rotation, translation, zooming, brightness adjustment, and horizontal flipping to increase data variability and reduce overfitting. All models were trained using an input size of 224 × 224 pixels, the Adam optimizer, and categorical cross-entropy as the loss function. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics to assess the classification capability of each model. In addition, confusion matrix analysis was conducted to identify classification error patterns across the ripeness categories. The results indicate that transfer learning models outperformed the Baseline CNN model. DenseNet121 achieved the best overall performance, followed by EfficientNetB3 and ResNet50. These findings demonstrate that transfer learning is an effective approach for oil palm fruit ripeness classification, particularly when working with limited datasets. Nevertheless, further studies using larger and more diverse datasets are recommended to improve model generalization capabilities.