YOLOv5 versus YOLOv3 for Apple Detection

The use of the YOLOv3 and YOLOv5 algorithms for apple detection in fruit-harvesting robots are compared. It is shown that the YOLOv5 algorithm could detect apples in orchards without additional pre- and post-processing with 97.8% Recall (fruit detection rate), and 3.5% False Positive Rate (FPR). It is much better than YOLOv3 that gives 90.8% Recall and 7.8 FPR when combined with special pre- and post-processing procedures, and then 9.1% Recall and 10.0% FPR without pre- and post-processing.