Abstract: Abstract: The potted-flower is largely needed during festivals in China, as the parterre takes a more and more important role for decorating the city because of its timely and beautiful sight. However, the task to transplant the flower seedling from the tray to the pot still relies on a manual operation, which is tiring and costly. In order to improve the mechanical transplanting for the flower seedling, key parts of the automatic transplanter were designed. The machine could autonomously identify the seedlings fitting for transplanting based on the machines' vision technology, and separately plant them into the flower pots from the seedling tray, with the intention to relieve labor intensity. A color camera is used to acquire the image of the seedlings in the tray, and the RGB aberration T was used to distinguish the seedling leaves from the root soil. Then the segmentation threshold for the target was obtained according to the maximum variance of the gray value between the seedling leaves and the background. The seedling quality was evaluated by counting the target pixels in every tray-hole. If the pixels quantity was more than 700, the seedling was identified as good-quality, fit for transplanting, and the position data of the tray hole growing the good seedling also was measured by the vision system. A flexible grasper was applied to hold the seedling root to prevent hurting the seedling. After inserting the fingers into the soil, the ring-shaped spring drove the fingers to close (grasp) to remove the seedling from the tray hole. Considering the grasp force should be strong enough to pick up the seedling, and the fingers should be long enough to prevent hurting the seedling leaves, the parameter of the spring was calculated according to the kinetics equations. A UFSP5-0.5-80 spring produced by MISUMI was selected. In order to test the performance of the grasper, a virtual test was carried out on ADAMS, and the result showed the grasping pressure on the root soil stayed at 2.3N when reaching static equilibrium, which was enough to overcome the resistance to pick the seedling out from the tray hole. In other words, the spring selected was fit for the grasper. Additionally, the swing angle of its fingers was adjustable so as to make the grasper adaptive for different trays. A PC and a series of SMC6480 modules composed the control system of the multi-motor machine, and they were linked through an Ethernet workgroup. Every SMC6480 could separately send the signal from the PC to 3 motors or graspers. The structure was convenient to extend a different number of graspers trough the router. Finally, the test for the key parts showed that when the transplanter completed less than 800 cycles per hour, the grasper holding and releasing the seedlings was perfect, and more than 95% of transplant cycles were finished successfully. 87% of the good-quality seedlings were recognized successfully, and the control system could run well under the transplanting efficiency from 700-900 cycles per hour.