A variable-rate spraying system for vineyards based on RGB-depth and instance segmentation

Sustainable vineyard management necessitates the precise application of plant protection products (PPP). This study proposes a variable-rate spraying system for vineyards, integrating an RGB-depth (RGB-D) camera with an NVIDIA Jetson computing platform. The system employs a YOLOv8 model for instance segmentation of grapevine canopies, utilizes structured point clouds to generate real-time 3D meshes of canopies, calculates canopy volume with calibration using trellis posts in vineyards, and manages flow rates through pulse-width modulation signals. Field experiments demonstrated a considerable reduction in PPP usage compared to uniform-rate spraying, highlighting its effectiveness and environmental benefits.