| The current global production of kiwifruit(Actinidia deliciosa)stands at 4.3 million metric tons(Tones),led by China with 56% of the total.To solve the various problems that the kiwifruit industry experiences,it is necessary to accelerate the development of mechanized kiwifruit production.Mechanization of the kiwifruit industry must adapt to adjustment in how China's economic structure develops.Kiwifruit production in China focuses on the use of cultivation machinery and equipment in orchards.In addition,farmers use artificial labor,to complete harvesting.At present,the development of mechanization in the kiwifruit industry is hindered by technical difficulties in harvesting.It should be pointed out that current studies are focused on harvesting of the mechanization.In essence,the kiwifruit industry in China should pay close attention to these key technologies and the theoretical studies that underpin them in order to promote standardization,mechanization and industrialization.Based on the growth and physical characteristics of kiwifruit,using a picking robot is highly suited to the scaffolding cultivation of kiwifruit.The harvesting robot is a lack of generality in the study of picking robot multi-target recognition and multi-robot cooperative operation.The method of multi-target recognition in wide area and complex environment and a method of task planning based on multi-manipulator harvesting robot are proposed,and a method of multi-manipulator zoning and harvesting sequence planning is validated.The main research results are as follows:(1)An Im-AlexNet method of full field of view kiwifruit target recognition and location is proposed.The Im-AlexNet of Faster R-CNN was used to recognize the full field of view and occluded fruit image,included the sunny backlight,sunny rembrandt light,cloudy,night with illumination condition.In addition,there is more obstructive among fruit clusters and branches and leaves of fruit trees.By modifying the number of nodes in full connection layer of AlexNet model by Transfer Learning,fine-tuning the number of nodes full connection layer L6,L7 to 768 and 256.Through the recognition of 1 823 multi-cluster kiwifruit images trained by Im-AlexNet,the experimental results indicated that the average precision(AP)of far-view and occluded complex condition images was 96.00%,and the recognition speed reached 1 s/range.By comparing with LeNet/AlexNet/VGG16 models of training the same datasets,the AP of Im-AlexNet is 5.73% higher than others Faster R-CNN network,and the rate of false recognition and missing recognition of kiwifruit was reduced by Im-AlexNet.In addition,kiwifruit cluster was located by using Lidar and Point Cloud Processing module of Computer Vision System Toolbox in Matlab function,point clusters are generated according to depth images.The depth maps are registered with RGB images,and the three-dimensional coordinates of point clouds corresponding to any point coordinates on RGB images are obtained.The average three-dimensional coordinates absolute error of X direction was 4.3 mm,and that of Y direction was 3.9 mm.The average absolute error of Z direction(height)of 30 samples measured by radar rangefinder is 3.5 mm.(2)Based on the recognition results of kiwifruit cluster in full field of view,the method of picking task partition and sequential planning for multi-manipulator cooperative operation was determined.Multi-manipulator picking task planning uses K-means multi-manipulator picking task zoning,and simulates the degradation method(SAA)to determine the picking sequence.The K-means clustering algorithm is used to partition the multi-objective fruit sample image of clustered kiwifruit,which includes three steps: firstly,the location of kiwifruit cluster is determined,and the kiwifruit in the 100 mm range adjacent to kiwifruit is defined as the same cluster.Secondly,the neighboring fruit cluster is clustered to determine a working region.Finally,the fruit cluster in the image is divided into four parts.Simulated annealing method was used to validate the picking order of kiwifruit.Several simulated annealing methods were solved Travel salesman problem(TSP),and the path length of picking order was counted.The SAA plan the picking sequence results of kiwifruit shows that the average length of picking path is 807 mm.It basically meets the requirements of picking robot for planning the picking sequence.(3)The method of integrated picking was studied for harvesting kiwifruit by robot.The end-effector was manufactured and the trajectory of integrated harvesting was verified in experiment.We propose an automated method to pick kiwifruit that consists of separating the fruit from its stem on the tree.This method is experimentally verified by using it to pick clustered kiwifruit in a scaffolding canopy cultivation.The end-effector approaches a fruit from below and then envelops and grabs it with two bionic fingers.The fingers are then bent to separate the fruit from its stem.The grabbing,picking,and unloading processes are integrated,with automated picking and unloading performed using a connecting rod linkage following a trajectory model.The trajectory was analyzed and validated by using a simulation implemented in the software Automatic Dynamic Analysis of Mechanical Systems(ADAMS).In addition,a prototype of an end-effector was constructed,and its bionic fingers were equipped with sensors to detect the best position for grabbing the kiwifruit and ensure that the damage threshold was respected while picking.Tolerances for size and shape were incorporated by following a trajectory groove from grabbing and picking to unloading.The end-effector separates clustered kiwifruit and automatically grabs individual fruits.It takes on average 4–5 s to pick a single fruit,with a successful picking rate of 94.2%.This study shows the grabbing–picking–unloading robotic end-effector has significant potential to facilitate the harvesting of kiwifruit.(4)According to the characteristics of the picking from the bottom by robot,the position and pose control method of the picking manipulator based on ROS is researched.The picking posture and motion path of the manipulator are planned,the inverse kinematics of the picking manipulator is solved,and the movement posture characteristics of the picking manipulator are simulated and analyzed through ROS MoveIt! module.The model of the manipulator is created by Solidworks software.The URDF model file of the manipulator is derived by sw2 urdf plugin conversion tool,and the parameters of the manipulator model are configured.Based on ROS,the motion simulation of a 6-DOF manipulator is carried out.Rapidexploration Random Tree(RRT)algorithm in the Open Motion Planning Library(OMPL)is selected to analyze the motion planning of the picking manipulator and drive the UR5 manipulator.The results show that the pose and trajectory of the picking manipulator are determined by controlling the movement of the picking manipulator,and the maximum radius of the rotation of the end effector and the position of the separation point are determined.(5)The performance of kiwifruit harvesting robot has been measured through trialmanufacture and verification experiment.A kiwifruit harvesting robot was developed and validated to test the effectiveness of wide area target recognition and integrated picking methods for kiwifruit.The kiwifruit picking robot is composed of five parts: vision system,manipulator,control unit,end-effector and mobile platform.The validation experiment with picking manipulator mainly validates the full field of view fruit information perception and coherent picking method,and statistics the recognition accuracy of picking robot in the actual working environment,and the effectiveness of the integrated picking method.On the platform,fruit collecting basket,supplementary light and other devices are assembled.The results of evaluation index analysis showed that the recognition accuracy of the picking robot vision recognition system was 95.31%,and the accuracy of kiwifruit target localization was 79.06%.Among them,1.25% were misidentified and 4.69% were missed,and the recognition time was 1.0 s.The harvesting robot can capture and complete the picking,the picking time of single fruit is about 5.8 s,which verifies the effectiveness of the integrated picking method of kiwifruit proposed in this paper. |
PDF Full Text Request