A precise crop row detection algorithm in complex farmland for unmanned agricultural machines.

Crop row detection is an important part of precision agriculture. Therefore, an unmanned agricultural machine crop row detection method based on YOLO-R is proposed. This method can solve the problem that traditional image processing methods are easily affected by weeds, light and other factors when extracting crop feature points. First, the YOLO-R object detection algorithm was used to obtain the crop position information, and then, the number of crop rows in the image and the crop in each crop row were obtained by the DBSCAN clustering algorithm. Finally, the function expression for each crop row was obtained by using the least squares method. The experimental results show that the AP values of YOLO-R are 91.69%, 95.34% and 89.13% on the seven-day, 14-day, and 21-day rice datasets, respectively. When the proposed algorithm's number of parameters was only 12.31% of that of YOLOv4 and the FPS was 17.54 higher than that of YOLOv4, the AP value was only 2.2% lower. The accuracy values of crop row detection algorithm are 93.91%, 95.87% and 89.87% on the seven-day, 14-day, and 21-day rice datasets, respectively, which indicates that the algorithm in this paper can effectively identify crop lines. • Proposing a crop row detection method based on YOLO-R object detection algorithm. • Ghostnet and Focal Loss are used in YOLO-R to perform better and faster. • Use channel attention module to coordinate the importance between each channel. • Incremental ablation study on all network designs. [ABSTRACT FROM AUTHOR]