In the process of grafting, the vegetable grafting robot needs a gripper to hold the grafted seedlings. Taking the grafting by sticking as an example, the grafting process needs clamping, positioning, carrying and fitting. The gripper in the whole process is related to the success or failure of grafting. In order to better realize the grafting action and improve the intelligence of the machine, this paper proposes the design of grafting seedling clamping system based on machine vision on the basis of the existing grafting robot. Taking the positioning clamping mechanism of the grafting machine as an example, a stable, reliable and high-precision clamping system is studied to improve the grafting efficiency and survival rate.