| Feature recognition method of corn full\|film double\|ridge furrow with “one film used for two years” based on deep learning |
|
View Fulltext View/Add Comment Download reader |
| |
| DOI：10.7606/j.issn.1000-7601.2024.04.30 |
| Key Words: full\|film double\|ridge furrow seed bed characteristics deep learning target detection drone shooting |
| Author Name | Affiliation | | PU Junyu | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | DAI Fei | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | SHI Ruijie | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | WANG Jiuxin | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | SONG Xuefeng | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | LI Lu | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China | | ZHAO Wuyun | College of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou, Gansu 730070, China |
|
| Hits: 581 |
| Download times: 505 |
| Abstract: |
| To more accurately grasp the characteristics of corn full\|film double\|ridge furrow for “one film used for two years” and improve its intelligent level, based on deep learning of target detection network model, feature recognition method of corn full\|film double\|ridge furrow for “one film used for two years” was employed, and the residual membrane, root stubble and overburden zone in the seed bed structure were detected and identified. The results showed that the mean average precision (mAP) of network model of YOLOx was 90.76%, which was better than other network models. In addition, to explore the impact of image acquisition of drone on target detection results, a mathematical model related to mAP was established based on the three factor and three level of BOX Behnken experimental design method, which included three parameters: flight altitude, flight angle and flight speed of drone. The optimal parameter combination of drone for image acquisition was found through experiments and verified. The results showed that, the mAP was maximum when the drone flied at a height of 0.9 m, a flight angle of 100°, and a speed of 2.2 m·s-1. After optimizing the shooting parameters of drone, the mAP of the network model was 92.67%, which was better than other groups. |
|
|
|