Classification of Coconut Fruit Ripeness Level Using Convolutional Neural Network (CNN) Method

Muhammad Rizki (Universitas Indo Global Mandiri)
Rudi Heriansyah (Universitas Indo Global Mandiri)
Dwi Verano (Universitas Indo Global Mandiri)

Jurnal Komputer, Informasi dan Teknologi
Vol. 6 No. 1 (2026): June

Manual assessment of coconut ripeness is often subjective and causes post-harvest losses of up to 25% in Indonesia, the world's largest coconut producer. This study aims to develop a CNN VGG-19 model for automatic classification of three ripeness levels (immature, medium, mature) with accuracy >95%. The quantitative experimental method uses supervised learning with a dataset of 900 original images (300/class) from local plantations in South Sumatra, augmented to 3000 images. Instruments include Python/TensorFlow on Google Colab, preprocessing (rembg background removal, resizing 224x224), training 10 epochs of the Adam optimizer. Analysis uses a confusion matrix, accuracy, precision, recall, and F1-score. The results show a progressive accuracy from 14% (40 test data/class) to 98% (200 test data/class). Conclusion: VGG-19 transfer learning with data augmentation is effective for local coconut ripeness classification, potentially integrating into mobile applications for the processing industry.