Research On Target Recognition And Grasp Of Tomato Picking Binocular Robot

Picking mature tomato fruit is the most time-consuming and laborious link in tomato production.With the development of science and technology and labor shortage,it is necessary for tomato picking robot to replace human labor.In this paper,the mature tomato fruit in greenhouse is taken as the object,the tomato picking robot is designed,and the hardware of tomato picking robot is assembled.The identification of mature tomato fruit,the positioning of target fruit and the kinematic analysis and trajectory planning of manipulator are studied,and the grasping of mature tomato fruit is realized.The main research work is as follows:(1)The background significance and progress of tomato picking robot at home and abroad are introduced,the identification and location of mature tomato fruit and the trajectory planning of manipulator are analyzed,the overall design of tomato picking robot is completed,and the hardware is selected.complete the construction of the hardware of the whole machine.(2)Based on the improved YOLOv4 algorithm,the fruit recognition of mature tomato was studied.In order to solve the problem of limited computing power of target and edge equipment recognized by tomato picking binocular robot,the deep backbone feature extraction network of YOLOv4 algorithm is replaced by lightweight network,and the enhanced feature extraction network is optimized for deep separable convolution.The improved Ghostnet-YOLOv4 algorithm has a good overall effect,the number of network parameters is reduced to 16% of the original,and the size of the trained model is 42.5m.The average accuracy of mature tomato fruit recognition is 94.80%,and the detection speed is 0.012 s.Considering the situation that the fruit is occluded in the real production environment,the mature tomato fruits with different occlusion degrees are divided and identified,and the improved YOLOv4 algorithm is used for recognition.The results show that with the increase of the degree of occlusion,the recognition accuracy tends to decrease.(3)The fruit localization of mature tomato was studied based on binocular stereo vision.The internal and external parameters and distortion coefficient of the binocular camera are calibrated,and the binocular correction is completed.The corrected image is stereo matched by SGBM stereo matching algorithm to get parallax.In order to further locate the recognized mature tomato fruit,binocular stereo vision and YOLOv4 algorithm are combined.The distance between the fruit and the camera is calculated according to the identified coordinate frame information of the target fruit and the parallax obtained by stereo matching,and the effect of location is verified by the ranging experiment.In the ranging experiment of a single fruit,the maximum proportion of ranging error is less than1.5%.(4)The trajectory planning of the manipulator is completed based on the kinematics of the manipulator.The Dmurh parameter method is used to model the manipulator,and the forward kinematics and inverse kinematics of the manipulator are solved.The effects of cubic polynomial and quintic polynomial on the single joint trajectory planning of the manipulator are compared and analyzed.The quintic polynomial is selected for the trajectory planning of the manipulator,and the trajectory planning simulation experiment of quintic polynomial is carried out by using matlab to model the manipulator.The results show that the quintic polynomial can plan the steady motion of the manipulator.(5)The experiment of target fruit recognition and grasping was carried out on the whole machine.First of all,the identification and location experiment of several mature tomato fruits is carried out,and the distance measurement is carried out while identifying the mature tomato fruit,the maximum ranging error is 0.009 m,and the error ratio is less than1.5%,indicating the reliability of the identification and positioning,and then the robot arm is modeled by matlab,and the simulation experiment of multi-axis coordinated motion is completed,which verifies that the quintic polynomial can satisfy the planning of the smooth motion of the manipulator.Finally,the grasping experiment of ripe tomato fruit was completed,and the success rate was 86%,so as to verify the feasibility of the whole grasping operation.