Aiming at the risk of end effector damage to the maize root system during mechanical intra-row weeding, a novel intra-row weeding mode based on the protection of the maize root system was proposed and an intelligent robot system for weeding was designed. The robot system consisted of a robot mobile platform, a weeding device, a visual system and a control system. Among them, maize seedlings and weeds were detected by the visual system using advanced YOLO V4 network model; the weeding device was based on the three-dimensional movement of weeding shovels, so that the weeding shovel can complete two operation models, including under and on the ground to avoid maize seedlings. Field experiments showed that when the robot mobile platform advanced at 1.2km/h, the detection rates of maize seedlings and weeds were 96.04% and 92.57%, respectively, and the weeding rate of the two weeding modes were higher than 81%. The average root damage rate under the mode that weeding shovels placed on the ground to avoid maize seedlings was 3.35%, which was 36.40 percentage points lower than that of the mode that weeding shovels were always placed under the ground to avoid maize seedlings. The results proved that the weeding robot system was stable, and the weeding shovel avoiding seedlings and weeding on the soil had better protection performance of the maize root system.