Abstract: Addressing prevalent issues of low screening efficiency and insufficient recognition accuracy in traditional seed screening equipment, a control system for seed screening device based on visual recognition was designed．Using Raspberry Pi 4B as core control unit, and combining camera and light source components to form an image acquisition module, a visual recognition system was constructed．By configuring system operating environment, completing sample annotation and model training, combining Python programming and trigger mechanism design, and relying on OpenCV image processing library to achieve image analysis tasks, automated operation of seed selection was finally realized, which could effectively improve accuracy of seed selection and stability of system operation．Seed screening results indicated that average screening efficiency was 22 grains/min, with a peak screening efficiency of up to 28 grains/min, which was approximately 1.5 times higher than traditional manual screening efficiency．The device could achieve optimal screening results within 2 minutes and could precisely separate high-quality corn seeds from inferior ones, with an overall recognition accuracy rate of 97.4%．The constructed control system for seed screening device based on visual recognition could effectively enhance efficiency of seed screening, providing a high-precision and low-cost automated solution for seed quality control, and possesses significant industrial application value．