This repository contains source codes for the peformance evaluation of YOLOv8 and YOLOv9 for blueberry dection, counting, and harvest maturity assessment. This study presents the first publicly available dataset of blueberry canopy images with 17,809 “Blue” (ripe) and “Unblue” (unripe) for blueberries, which were captured in diverse orchard conditions. YOLOv8l (large) and YOLOv9-c (compact) with comparable complexity were trained for blueberry detection and whereby fruit counting and “Blue” fruit percentage estimation. Both models performed similarly in terms of detection accuracy and speeds, except that YOLOv9-c was far more time-consuming to train. Trained with the input of high-resolution images of 3520×3520 pixels, YOLOv8l achieved an overall mAP@50 of nearly 93%, and a root-mean-square error of 10.4 in fruit counting and 3.62% in estimating the percentage of ripe fruit of each image. The dataset of blueberry canopy images will be made available very soon.
Please consider cite our work if you find this repo is helpful.
@article{Update soon,
title={Canopy Image-based Blueberry Detection by YOLOv8 and YOLOv9 },
author={Boyang Deng, Yuzhen Lu*},
journal={Update soon},
volume={Update soon},
pages={Update soon},
year={2024},
publisher={update soon}
}
Wokring protocol: BlueberryAnnotationProtocol 03132023.pdf
Dataset URL: {Update soon}