In this study, a set of visual servo control system of mechatronic picking robotic arm based on machine vision is designed. By establishing the kinematics and dynamics model of the robotic arm and using the D-H parametric method for system modeling, it is verified that the model can accurately describe the motion characteristics of the robotic arm. The target recognition algorithm based on deep learning improves the YOLOv3 network structure, which significantly improves the recognition and localization ability in the complex orchard environment. The improved fuzzy neural network sliding mode control algorithm is developed for the nonlinear and strongly coupled characteristics of the robotic arm, and the experiments show that it is better than the traditional PID and SMC algorithms in terms of dynamic response, steady state accuracy and anti-interference ability. The proposed improved stochastic fast search tree algorithm combined with visual servoing dramatically improves the path planning efficiency through techniques such as super ellipsoidal gravity bias sampling. Experimental data show that the algorithm reduces the number of sampling points by 92.9%, the planning time by 86.1%, and the path cost by 35.2%. In the actual test, the success rate of the robotic arm picking integrated with the algorithm reaches 92.9%, which is 16.6 and 7.2 percentage points higher than that of the PID system and the SMC system, respectively. The research results provide effective technical support for orchard picking automation and promote the practical development of picking robots.
