A quick counting method for winter wheat at the seedling stage in fields based on an improved Yolov4 model | [Journal of Animal and Plant sciences • 2022]

Author(s):

1. H. Ma: Collaborative Innovation Center of Machinery Equipment Advanced Manufacturing of Henan Province,Luoyang 471003,China ; College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

2. W. Zhao: College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

3. J. Ji: Collaborative Innovation Center of Machinery Equipment Advanced Manufacturing of Henan Province,Luoyang 471003,China; College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

4. X. Jin: Collaborative Innovation Center of Machinery Equipment Advanced Manufacturing of Henan Province,Luoyang 471003,China ; College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

5. Y. Shi: College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

6. F. Zheng: College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

7. N. Li: College of Agricultural Equipment Engineering, Henan University of science and technology,Luoyang 471003,China

Abstract:

To realize the fast and accurate counting of winter wheat at the seedling stage in fields, we propose a recognition method based on an improved YOLOv4 model. Firstly, we employed a simplified MobileNetv3 neural network instead of the standard CSPDarknet53 network structure. Besides, we added an adaptive image scaling layer in front of the MobileNetv3 network. Finally, we utilized the coyote optimization algorithm (COA) to optimize the learning rate and convolution kernel size. Results showed that the average precision (AP) values of the improved network model for 2-leaf and 3-leaf winter wheat were 96.46% and 93.87%, respectively. The mean average accuracy (mAP) was 95.15% and the average recognition speed was 0.07 s. These indicators were the best, compared with the YOLOv4, YOLOv3, a nd Faster-RCNN models. Also, the mAP was 12.28% higher than the standard YOLOv4 model, and the average recognition speed was 1.49 times faster. Therefore, this method can achieve the fast and accurate counting of winter wheat during the seedling stage in the field.

Page(s): 1666-1681

DOI: 10.36899/JAPS.2022.6.0575

Published: Journal: Journal of Animal and Plant sciences, Volume: 32, Issue: 6, Year: 2022

Keywords:

YOLOv4 , Coyote optimization algorithm , Winter wheat recognition at the seedling stage , Adaptive image scaling , MobileNetv3

References:

References are not available for this document.

Citations

Citations are not available for this document.

Citations

Downloads

Views