Precision spraying using variable time delays and vision-based velocity estimation

Traditionally, precision farm equipment often relies on real-time kinematics and global positioning systems (RTK-GPS) for accurate position and velocity estimates. This approach proved effective and widely adopted in developed regions where RTK-GPS satellite and base station availability and visibility are not limited. However, RTK-GPS signal can be limited in farm areas due to topographic and economic constraints. Thus, this study developed a precision sprayer that estimated the travel velocity locally by tracking the relative motion of plants using a deep-learning-based machine vision system. Sprayer valves were then controlled by variable time delay (VTD) queuing and dynamic filtering. The proposed velocity estimation approach was tested at different velocities and tracking thresholds. The results showed that the velocity estimates agreed with actual measurements with a mean absolute error of 0.036 m/s. Further, testing the targeting algorithm on rows of artificial crops and weeds at different levels of spraying duration and filter size factor (FSF) showed that short spraying duration and small FSF increase overall spraying accuracy. Finally, testing the sprayer using the optimum settings at 0.87 m/s and 1.03 m/s successfully sprayed all targets. Further, only 2% to 7% of non-targets were sprayed at the low and high test velocities, respectively. With these results, this study suggests that vision-based velocity estimation combined with VTD queuing and dynamic filtering can be an accurate and low-cost solution for targeted spraying without using auxiliary velocity measurement systems.