Performance simulation of real-time vision-based variable rate precision spraying

Abstract

One of the main methods for weed control in crop fields is herbicide application. With the technology development in sensing and control, variable rate (VR) herbicide application is increasingly being adopted in the industry. In post-emergence VR herbicide applications, machine vision is commonly used for weed detection and identification. The recent development of deep learning and artificial intelligence has greatly improved the accuracy and efficiency for weed detection, making cost-effective VR herbicide applications more feasible.  In this paper, spraying simulation models were developed to simulate real-time machine vision-based variable rate precision spraying. The objective was to examine the influence of different design factors and spraying methods on the performance of VR precision spraying, e.g., the total amount of herbicide saved over the uniform application method. Different sprayer travel speeds and different control zone sizes were incorporated in the models. The simulated spraying methods include on/off intermittent spraying, variable rate spraying, and uniform low-dosage base rate plus variable rate spraying. The results show that travel speed has no influence on herbicide savings. For the two variable rate spraying methods, herbicide saving decreases when control zone size increases.  However, for the on/off intermittent spraying method, there is no difference in herbicide savings across different control zone sizes. Overall, the on/off intermittent spraying method resulted in about 10% of herbicide savings, while the variable rate spraying method achieved up to 45% of herbicide saving compared to the uniform spraying method. The results provide valuable suggestions for determining variable rate application strategies for precision weed management in crop production.