Performance Evaluation of YOLOv9, YOLOv10, and YOLOv11 for Real-Time Early Detection of Ganoderma Boninense in Oil Palm

Abstract

Early detection of Ganoderma boninense infection is essential to reduce yield losses in oil palm plantations. This study aims to evaluate the performance of three recent YOLO architectures, namely YOLOv9, YOLOv10, and YOLOv11, for real-time detection of early infection symptoms under natural field conditions. A dataset of 2,000 annotated RGB images was used with a 70:20:10 split for training, validation, and testing. Model performance was evaluated using precision, recall, F1-score, mean average precision (mAP50 and mAP50–95), and inference speed. The results show that YOLOv9 achieved the highest detection accuracy with an mAP50 of 0.989 and F1-score of 0.968. Meanwhile, YOLOv11 demonstrated the best computational efficiency with an inference speed of 35 FPS and processing time of 28.5 ms per frame. These findings indicate a trade-off between accuracy and speed, where YOLOv9 is suitable for accuracy-oriented applications, while YOLOv11 is more appropriate for real-time deployment in precision agriculture.