.png){kind=icon}
.png){kind=icon}
Return
A dynamic rice seed counting algorithm based on stack elimination
.png){kind=icon}
PRE
AI
DOI：10.1016/j.measurement.2024.114275.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Rice seed counting is an important indicator for evaluating the quality of rice seeds. A dynamic rice seed counting algorithm based on stack elimination was proposed to achieve accurate seed counting. First, a dataset was constructed by capturing videos of falling rice seeds, which was utilized to train the YOLOv7 model and evaluate counting accuracy. Secondly, the YOLOv7 model perception of small rice seeds was improved by integrating the CA module into the backbone network and the P2 small object layer into the detection head. Finally, the motion of the rice seeds during falling was modelled from horizontal and vertical directions. These measures were taken to eliminate the impact of rice seed rotational offset during the falling process. The improved YOLOv7 model achieved a mAP of 99.55%. Furthermore, for rice seed counting, the proposed counting algorithm achieved a MOTA of 100% and a MOTP of 83%.
Keywords:
Rice seed counting
Multi -object tracking
YOLOv7
SORT
Kalman filter