Abstract: In view of low efficiency, high cost and large pollution of pesticide spraying in corn seedling stage, a visual recognition and spraying system that can be mounted on a small pesticide spraying robot was proposed．The system used SSD-MobileNetv1 network as corn seedling recognition algorithm, combining methods of data augmentation and transfer learning．The ARM was used as an embedded device．When camera detected corn seedlings within spraying range, control pin generated PWM signals to start praying system to spray accurately．Results showed that recognition accuracy of system for corn seedlings could reach 84.18%, and estimation of distance basically met requirements of robot spraying．