Design and Application of an Automated Microinjection System Combining Deep Learning Vision Positioning and Neural Network Sliding Mode Motion Control

Zhihao Deng, Yifan Xu, Shengzheng Kang

Microinjection is one of the most established and effective techniques for introducing foreign substances into cells. However, issues such as cumbersome procedures, low success rates, and poor repeatability in manual cell microinjection have seriously restricted its practical applications in biomedical research and engineering. Responding to such problems, this paper designs an automated microinjection system that combines deep learning visual positioning and adaptive neural network sliding-mode motion control. The machine vision solution based on the deep learning YOLOv8 target detection algorithm is utilized by the system to provide positional prerequisites for automated microinjection. Then, stable and fast puncture is completed by controlling the end effector (composed of a piezoelectric actuator and a displacement amplification mechanism). Since the piezoelectric actuator has strong nonlinearity, the motion control of the end effector adopts the control strategy combining sliding mode variable structure and adaptive neural networks to meet the requirements of precise displacement output of microinjection. At the same time, a host computer control system is developed to integrate hardware equipment, visual positioning algorithms and motion control algorithms to achieve corresponding automated microinjection tasks. Finally, the effectiveness of the designed automated microinjection system is successfully verified on zebrafish embryos.