Abstract: Broccoli (Brassica oleracea L.var.Italica Plenck) is one of the most important vegetables in recent years. However, manual batch harvesting cannot fully meet the high requirement of selective harvesting, due to the strong seasonal labor demand, high labor intensity, and high cost. In this study, a selective harvesting platform was designed to automatically identify and then cut the broccoli, according to the agronomic properties and morphologies. Firstly, the selective harvesting platform primarily consisted of a walking module, identification in cutting device, triggering collection device, and control system. The control system was acquired for the visual information of broccoli through the upper computer. The lower computer was used to control the overall functioning of the harvesting platform, including overall walking, plant localization, flower ball positioning, stalk cutting, and flower ball grasping. Secondly, an "identify-harvesting" integrated operation mode was designed to autonomously identify and cut operations in the broccoli selective harvesting platform. This operational mode was used to mitigate the interference from independent movements of the executing mechanisms during identification. The precise movement distances of executing mechanisms were calculated to enhance the operational accuracy. The key components of the harvesting platform were selected for the visual identity system and centering cutting mechanism for broccoli. The original image of broccoli was collected by the visual system and then processed through grayscale conversion, Gaussian filtering, and threshold segmentation. Thirdly, the dimension and centroid position of the broccoli head were calculated using pixel area and moment. A better performance was achieved in the image processing time of 2.5 s, the average contour deviation of 2.2%, and the measurement centroid position error of 4.2%. According to the interaction between the broccoli stalks and the cutting knife, a logarithmic spiral was utilized as the cutting curve, in order to design a constant slip angle cutting knife. The key cutting parameters were determined, such as a cutting knife sliding angle of 40º, a cutting radius of 135 mm, and a cutting knife length of 260 mm. Additionally, the explicit dynamic simulation of stalk cutting was conducted using ANSYS Workbench/LS-DYNA software, according to the material property parameters of broccoli stalks. Taking the cutting knife edge angle and rotational speed as the control factors, and the maximum cutting force as the experimental indicator, the optimal parameter combination of the stalk cutting was determined by the orthogonal experimental optimization. An optimal combination was obtained in the cutting knife edge angle of 20° and rotational speed of 1 rad/s. The maximum cutting force was 725.82 N, indicating better cutting quality. Finally, the cutting performance test showed that the centering cutting mechanism rapidly and smoothly cut into and cut off broccoli stalks, where the cutting surface was flat and smooth, with a cutting time of approximately 0.6 s per unit, and the 100% success rate of cutting. The performance tests on the harvesting platform showed that the visual system effectively recognized the mature broccoli plants in natural environments with better detection. The overall leakage rate of the harvesting platform was less than 10%, the detection accuracy was 90%, and the qualified rate of cutting stalks was 88.9%, fully meeting the operational requirements of selective harvesting of broccoli. This finding can provide the theoretical and practical reference for the design and development of selective harvesting equipment for broccoli.