KLASIFIKASI KEMATANGAN BUAH KELAPA SAWIT BERBASIS TRANSFER LEARNING CONVOLUTIONAL NEURAL NETWORK

Nur Nafiiyah (Universitas Islam Lamongan)
Mohamad Habib Havito Ihzami (Universitas Islam Lamongan)
Agus Harjoko (Universitas Gadjah Mada)
Achmad Nizar Hidayanto (University of Indonesia)

Jurnal Sistem Informasi dan Informatika (SIMIKA)
Vol. 9 No. 2 (2026): Jurnal Sistem Informasi dan Informatika (Simika)

The ripeness level of oil palm fruit directly affects the quality and yield of palm oil, making accurate and objective classification methods essential. This study aims to evaluate and compare the performance of several transfer learning Convolutional Neural Network (CNN) architectures for oil palm ripeness classification. The dataset used is a secondary dataset obtained from previous research and consists of four ripeness classes: under-ripe, unripe, ripe, and over-ripe. Data augmentation was applied only to the training data to increase data variability, while the original, non-augmented data were used for testing to ensure an objective evaluation. Seven CNN architectures were evaluated, namely ConvNeXt-Tiny, DenseNet121, InceptionV3, MobileNetV2, NASNetLarge, ResNet50, and Xception, using the same training configuration. The results show that ConvNeXt-Tiny and ResNet50 achieved the best performance, with accuracy and F1-scores of 97.73%. These findings indicate that efficient CNN architectures can provide optimal performance for oil palm ripeness classification.