TY - GEN
T1 - Real-Time Bud Detection Using Yolov4 for Automatic Apple Flower Bud Thinning
AU - Sahu, Rashmi
AU - He, Long
N1 - Publisher Copyright:
© 2023 ASABE Annual International Meeting. All Rights Reserved.
PY - 2023
Y1 - 2023
N2 - Robotic bud thinning is an alternative technique of crop load management that regulates fruit bud density in the tree canopy. Flower bud detection in the natural environment is a key step for developing this robotic system to automatically remove flower buds. This study proposed a bud detection model using the You Only Look Once (YOLO) v4 deep learning algorithm. The detection performance of the YOLOv4 model was compared with those of the YOLOv5 and YOLOv7 models. The results show that under the same conditions, YOLOv4 performed better than YOLOv5 and YOLOv7 for buds'detection. On dataset-1 (stereo image dataset) the mean average precision (mAPs) of bud detection with YOLOv4, YOLOv5, and YOLOv7 were 98.9%, 75.5%, and 72.9% for one class, and 95.3%, 80.9%, and 76% for three classes respectively. The mAP from YOLOv4, YOLOv5, and YOLOv7 on dataset-2 (mobile images) was 94.1%, 79%, and 73% for one class, and 98.37%, 76.5%, and 66.9% for three classes respectively. The YOLOv4 model showed great potential for the real-time rapid detection of the apple bud. This model could lay the foundation for the machine vision unit of the robotic apple flower bud thinning system.
AB - Robotic bud thinning is an alternative technique of crop load management that regulates fruit bud density in the tree canopy. Flower bud detection in the natural environment is a key step for developing this robotic system to automatically remove flower buds. This study proposed a bud detection model using the You Only Look Once (YOLO) v4 deep learning algorithm. The detection performance of the YOLOv4 model was compared with those of the YOLOv5 and YOLOv7 models. The results show that under the same conditions, YOLOv4 performed better than YOLOv5 and YOLOv7 for buds'detection. On dataset-1 (stereo image dataset) the mean average precision (mAPs) of bud detection with YOLOv4, YOLOv5, and YOLOv7 were 98.9%, 75.5%, and 72.9% for one class, and 95.3%, 80.9%, and 76% for three classes respectively. The mAP from YOLOv4, YOLOv5, and YOLOv7 on dataset-2 (mobile images) was 94.1%, 79%, and 73% for one class, and 98.37%, 76.5%, and 66.9% for three classes respectively. The YOLOv4 model showed great potential for the real-time rapid detection of the apple bud. This model could lay the foundation for the machine vision unit of the robotic apple flower bud thinning system.
UR - http://www.scopus.com/inward/record.url?scp=85183579240&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85183579240&partnerID=8YFLogxK
U2 - 10.13031/aim.202300423
DO - 10.13031/aim.202300423
M3 - Conference contribution
AN - SCOPUS:85183579240
T3 - 2023 ASABE Annual International Meeting
BT - 2023 ASABE Annual International Meeting
PB - American Society of Agricultural and Biological Engineers
T2 - 2023 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2023
Y2 - 9 July 2023 through 12 July 2023
ER -