A YOLOv3-Based Rice Vortex Detecting System Using Dual Collaborative UAVs

Abstract

Highlights Used dual collaborative UAVs to track the rice vortex formed on rice canopy. A trained YOLOv3-tiny model was used to recognize and calculate the rice vortex parameters with high accuracy. Discovered a quantitative relationship between rice vortex parameters and UAV flight parameters. Aligned the time axis of data process to enhance the accuracy of the vortex parameter calculation. Abstract. A rice vortex generated by the downwash airflow of the agricultural UAV has a great connection to pesticide droplet deposition. In this study, an unmanned aircraft system (UAS) was designed to obtain rice vortex parameters including position coordinates, area, and relative distance to the UAV. The system uses the flight control system mounted Beidou RTK positioning system and Manifold onboard computer to localize the spraying UAV and trains a YOLOv3-tiny model to recognize motion images of dynamic rice vortices in real-time, achieving a mean accuracy of 98.15% with 0.9326 IOU. Moreover, to optimize the system, a data time axis alignment method based on the robot operating system (ROS) was proposed to alleviate the transmission delay and image processing time-consumption, which increased the accuracy of the vortex parameter by 3.48%. The rice vortex area acquired by the UAS was highly relevant to UAV flight parameters and the relative distance was just correlated to the UAV height. The instantaneous status of the rice vortex helps to lay a data foundation for adjusting the flight parameters of the plant protection UAV. Keywords: Precision agriculture, Rice vortex, ROS, Spray operation, UAV, YOLOv3-tiny.