Model comparison of object detection algorism YOLO based on the cone detection from the imagery of tree crowns of Abies sachalinensis

Abstract

Differences in the accuracy of cone detection on the tree crown of Abies sachalinensis were compared using multiple models of the object detection algorithm “You Only Look Once” (YOLO). Training and validation were conducted using YOLOv4 and YOLOv5m, which both showed an average precision (AP) of around 0.9. Five models of YOLOv5 (YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x) with different neural network layer sizes were also compared. AP was 0.9 in all models except for YOLOv5n, and the accuracy was similar regardless of the neural network layer size.