Deteksi Tahap Perkembangan Putik Sawit Menggunakan YOLO8n

Detection of Palm Oil Pistil Development Stages Using YOLOv8n

Abstract

Oil palm productivity is strongly influenced by the success of the pollination process, particularly in female flowers at the anthesis stage, when they are receptive to fertilization. Identification of this stage is generally performed manually, which is inefficient and prone to detection errors in the field. This study proposes an automatic detection system for oil palm female flowers at the receptive stage using the YOLOv8n model, a lightweight variant of the YOLOv8 family that is well suited for CPU-based computation. The dataset consists of three main classes: female flowers, live male flowers, and dead male flowers. The data are divided into training (82%), validation (12%), and testing (6%) sets. Model training is conducted for 80 epochs with an input resolution of 640×640. Evaluation results show that the model achieves an mAP50 of 0.86, a precision of 0.89, and a recall of 0.85. These results indicate that YOLOv8n is capable of accurately detecting receptive female oil palm flowers even under limited computational resources. This system has the potential to serve as a foundation for the development of computer vision–based technologies for monitoring oil palm phenology and pollination.