Research On Visual Servo Control System Of Tomato Collateral Pruning Robot

Tomato collateral pruning robot can improve the efficiency of pruning and provide support for improving agricultural production mode,solving labor shortage,realizing agricultural automation and large-scale.In this paper,aiming at the problems of tomato branch and leaf pruning robot’s difficulty in recognizing tomato side branches and low pruning positioning accuracy,the tomato branch and leaf pruning robot is taken as the research object,and its visual servo control is mainly designed and experimented.The specific research contents are as follows:(1)For the tomato collateral pruning robot,the D-H method was used to construct the connecting rod coordinate system of the robot arm,and the forward kinematics model of the robot arm was established by measuring its parameters.The inverse kinematics model of manipulator was solved by geometric method,and its workspace was solved.Camera model is set up in visual aspects,summarizes the binocular visual positioning principle,and on this basis to solve the image jacobin matrix.(2)In terms of tomato side-branch image processing,RGB threshold segmentation and K-mean clustering segmentation are respectively used to solve the problem.The experiment of background segmentation of tomato plants in different environments shows that k-mean clustering segmentation method is more robust.Secondly,a method to extract the characteristic points of tomato laterals was proposed.By extracting and sorting the horizontal and vertical edges of tomato plants,a set of dual characteristic points of tomato laterals were obtained.(3)In order to solve the problem of camera calibration and image Jacobian singularity in general visual servo system,a visual servo control system based on binocular camera was designed.PSO-BP neural network was used to map the changes of image features and joint angles,and joint angles were predicted by training and testing the network.The simulation results show that the error of the image feature points in the camera module is within 5 pixels when moving to the predicted joint Angle.(4)The end-effector was designed for tomato collateral pruning robot,and its dynamics model and state equation were solved.Sliding mode variable structure control was designed for tomato side branch trimmer robot,and its output model was solved by exponential approaching law and sliding mode surface function.The comparison with PID simulation shows that the sliding mode variable structure control has faster response speed and smaller error.(5)The experimental platform was built and the visual servo control system was established.Two groups of experiments were carried out.In experiment 1,the depth camera was used to locate the target position,and the joint Angle was solved by inverse kinematics to control the rotation of the manipulator.In experiment 2,feature point changes were obtained by binocular camera,and joint angles were predicted in the trained PSO-BP network to control the manipulator.The experimental results show that the manipulator can not perform precise pruning work based on position information,and the average error is 4.98 cm.Based on the image features,the accurate pruning rate reaches 90%,and the average error in the image is 5.6pixes.