<![CDATA[Oil palm is a strategic commodity in Indonesia, and its production quality is greatly influenced by the ripeness of the fruit at harvest. Manual ripeness determination is still subjective and prone to errors due to variations in worker experience and environmental conditions. Advances in computer vision and deep learning technology offer a more objective and consistent automated solution. This study aims to develop and evaluate a model for detecting the ripeness level of palm oil fruit using the YOLOv12m algorithm based on digital images. The dataset used consists of 3,375 images with three ripeness classes (unripe, semi-ripe, ripe), which are divided into training, validation, and testing data with a ratio of 70:20:10. The model was trained for a maximum of 25 epochs with an early stopping mechanism. The evaluation was conducted using precision, recall, mAP@50, and mAP@50–95 metrics. The results showed excellent performance with precision of 0.958, recall of 0.946, mAP@50 of 0.985, and mAP@50–95 of 0.882. Class-by-class analysis shows the best performance in the raw and ripe classes, while the unripe class still poses challenges due to visual similarities between transition phases. Overall, the YOLOv12m model has proven to be effective and has the potential to be applied as a more objective and efficient harvest decision support system.]]>
Copyrights © 2026
