Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Highlights

An improved smart sprayer prototype featuring a 24-nozzle array was developed for precision vegetable weeding.

ABSTRACT. Weed management in vegetable crops faces persistent challenges due to the vulnerability of these cropping systems. Sustainable weed management increasingly focuses on reducing herbicide use through precision control strategies. Advances in machine vision and artificial intelligence (AI) for weed detection and localization have enabled more effective precision weeding. Although blanket spraying systems are commonly used for weed removal, vision-guided smart sprayers offer considerable promise but remain underexplored for vegetable cropping systems. Our group previously developed a smart sprayer, but early field tests showed low weed detection accuracy and spraying hit rates. This study presents an upgraded smart sprayer prototype with several key improvements: a new structure with an array of 24 densely packed nozzles spaced 2.5 cm apart to increase the spatial resolution in weed spraying and target small weeds, reduced nozzle height to mitigate spray drift, YOLOv11 models trained on a multi-season dataset for improved weed detection, and plant tracking to enable precise nozzle scheduling and activation. An optimized multithreading system integrates all modules for efficient real-time operation. In addition, the vision system employs a high-speed GigE camera to achieve video-rate, high-resolution imaging with minimal motion artifacts in dynamic field conditions. Both indoor and field experiments were conducted to assess weed detection and spraying performance, and a comprehensive set of metrics corresponding to image-, video-, and spray-based performance was evaluated in field conditions. In detecting simulated targets in indoor testing, the prototype attained 99.4% mAP@50 for video-based detection and a 100% spraying hit rate. In contrast, in a radish field, the performance declined to a video-based mAP@50 of 69.3% and a spraying hit rate of 74.6%, highlighting the challenges of model generalization under field conditions. This work provides a valuable reference for further research to advance precision weeding technology, offering insights into modular design, system integration, field implementation, and performance evaluation, particularly benefiting small-scale vegetable production.