Nobody would contest that physical weed removal methods offer numerous advantages over biochemical alternatives. Within the domain of intelligent mechanical weed control, comprehensive research targeting the entire intelligent weeding machine system remains relatively scarce. To expedite the practical application of intelligent weeding machines, this study explored an enhanced YOLOv5 model with one colour constancy module, which aimed at achieving higher accuracy in crop seedling detection. An innovative "separating and closing" strategy, which allows the machine to precisely avoid crop seedlings while effectively weeding the areas between crop seedlings was employed to facilitate intra-row weeding. By integrating this strategy with a comprehensive design of the mobile platform, inter-row weeding actuators, and harmonious control of these key components, this research successfully developed an intelligent weeding machine capable of simultaneously performing intra-row and inter-row (all-round) weeding. Compared with previous studies, this study put the emphases on complex farm lighting conditions, both inter-row and intra-row weeding functions, and weed regrowth. Field experiments conducted in lettuce (Lactuca sativa var. ramosa Hort.) fields at four different locations on three separate dates demonstrated that this intelligent weeding machine achieved average weeding rates, crop seedling damage rates, and regrowth rates of 96.87%, 1.19%, and 0.34%, respectively. The ability to perform all-round weeding simultaneously is a significant advance in mechanical weeding control. The design and methodology employed in this study have broad implications for advancing the field of precision agriculture and addressing the growing demand for sustainable farming practices.