Comparison of YOLOv8 Models for Detecting Lemons from Tree

Fruit detection from trees is a critical process in estimating crop productivity using computer vision techniques. This study focuses on detecting lemons from tree images by employing various object detection variants of YOLOv8. Specifically, the YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x variants were applied for fruit detection, and the results were comparatively analyzed. To the best of our knowledge, no prior study has conducted a comparative analysis of YOLOv8 variants for lemon detection. Such comparisons are essential in assessing the suitability of each variant for different application requirements and in identifying the most appropriate model based on specific needs. In this study, a novel dataset of 4,981 lemon images from 127 trees was created, annotated using the bounding box technique, and the models were evaluated using various performance metrics. The comparative analysis of YOLOv8 variants for lemon detection revealed that YOLOv8n is the fastest but has lower accuracy, while YOLOv8l and YOLOv8x exhibit similar performances, with YOLOv8l outperforming in terms of the mAP50 metric.