Precision harvesting for premium tea buds: overcoming recognition and damage-control challenges via multisensor fusion

To address the challenges of complex terrain in Chinese hilly tea garden, high manual picking costs of premium tea, and low mechanization rates, this paper proposes an Intelligent Tea-Picking Robot system based on an improved YOLOv7 algorithm and binocular vision. By integrating HSV color space segmentation and 3D point cloud modeling technologies, the system enhances the accuracy of tea bud recognition and localization. The improved YOLOv7 model enables graded detection of single buds and buds with multiple leaves. The mechanical structure adopts a bionic folding arm design, incorporating an arc blade cutting mechanism and a flexible gripping and collecting module. Ablation experiments and field comparison tests effectively validate the applicability of the lightweight model on embedded devices, providing a practical reference for the mechanized picking of premium tea in hilly tea garden.