Detection of ripeness level of oil palm fresh fruit bunches using YOLOv4 model in automated harvesting system: A review

Abstract

This research discusses the development of an automated system for detecting ripe oil palm fruits (fresh fruit bunch, FFB) using computer vision and artificial intelligence (AI) technology. The main objective of this study is to improve efficiency and productivity in the oil palm harvesting process by adopting the latest technology. Several related studies have developed non-destructive methods for classifying the ripeness of FFB. This research utilizes the YOLOv4 deep learning model to detect ripe FFB in real-time. Visual data of FFB is obtained by capturing images using an Intel Realsense D435 camera on oil palm trees. The data is then labeled and divided into training and validation sets. Through evaluation, the YOLOv4 model with a network input size of 512 × 512 was found to be the best model for TBS detection task. The training process was conducted for 2000 iterations, achieving a mean average precision (mAP) of 87.9% in the final iteration. This model successfully detects ripe FFB with high accuracy. The results of this research indicate that the use of computer vision and artificial intelligence technology can help optimize the oil palm harvesting process. With this automated system, the palm oil industry can address labor shortages and improve production efficiency. This study provides an important contribution to the development of the oil palm industry with potential applications in broader fields.