Garlic shoot harvesting method based on Delta parallel mechanical arm harvesting robot

Abstract

The invention discloses a garlic shoot harvesting method based on a Delta parallel mechanical arm harvesting robot. A cooperative control system controls a harvesting robot moving chassis to walk; a depth camera is used for collecting image information of young garlic shoots, a YOLO11 target detection algorithm is used for identifying the young garlic shoots and accurately positioning coordinates, and a cooperative control system controls a reciprocating needle pricking device to perform needle pricking operation on garlic plants; a Delta mechanical arm of the harvesting device is controlled to pull the young garlic shoots according to the image coordinate positioning result, and then the pulled young garlic shoots are transferred to a collecting device. The method comprises the following steps: acquiring image information of young garlic shoots, accurately identifying and positioning the young garlic shoots on the basis of a YOLO11 target detection algorithm, then controlling a reciprocating needle pricking device to perform needle pricking operation on garlic plants by a cooperative control system, and finally, controlling a Delta mechanical arm of a harvesting device to pull, transfer and collect the young garlic shoots according to an image coordinate positioning result. And automatic harvesting operation is realized.

Description

Technical Field

[0001] The present application relates to the technical field of automated harvesting of agricultural products, and in particular to a method for harvesting garlic stalks based on a Delta parallel robotic arm harvesting robot. Background Art

[0002] Garlic stalks are tender stems that emerge during the growth of garlic. They are in huge demand, have high edible value and economic benefits, and have broad market prospects. Mechanized harvesting of garlic stalks is an inevitable trend in the development of the garlic industry. However, harvesting garlic stalks is a highly seasonal and labor-intensive agricultural activity, which directly affects garlic yield and economic benefits.

[0003] Traditional garlic stalk harvesting mainly relies on manual labor, which is inefficient and has the following problems: First, the labor intensity is high. Garlic stalk harvesting requires farmers to bend over for a long time, which is labor-intensive. Especially in the case of large-scale planting, work efficiency is significantly reduced, affecting production progress; second, the harvesting efficiency is low, and the manual harvesting speed is limited. Usually, each person can only harvest 0.5-1 mu of garlic stalks per day. During the best harvesting period of garlic stalks (usually 7-10 days), manual harvesting is difficult to meet the needs of large-scale farmland; third, the labor cost is high. With the reduction of rural labor and the increase in labor costs, manual harvesting has gradually become the main bottleneck restricting the development of the garlic stalk industry, especially for farmers with large planting areas, the proportion of labor costs has increased significantly; fourth, the quality is unstable. Due to the different technical proficiency of manual harvesting, some garlic stalks may be broken or damaged during the harvesting process, affecting their commodity value. At the same time, the pseudostem of the garlic plant will be damaged during harvesting, which may affect the subsequent garlic yield.

[0004] In the prior art, domestic relevant personnel have conducted research and attempts to solve the above problems, and proposed a variety of solutions, but there are still many problems. In short, realizing automation in garlic scape harvesting is a technical problem to be solved in this field. Summary of the invention

[0005] In order to solve the above technical problems, this application proposes the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for harvesting garlic stalks based on a Delta parallel robotic arm harvesting robot, comprising:

[0007] The Grey Wolf Algorithm is used to optimize and improve the traditional A* algorithm, and the improved A* algorithm is integrated with the DWA algorithm to achieve the global and local planning of the motion chassis of the garlic stalk harvesting robot.

[0008] The image information of garlic stalks is collected using a depth camera, and the garlic stalks are identified and accurately located using the YOLO11 target detection algorithm.

[0009] By integrating the functional modules of the garlic bolt harvesting robot and proposing a cooperative control method, the adaptive adjustment of operation parameters and the optimization of the operation rhythm and transfer mode of the modules are realized;

[0010] According to the image coordinate positioning result, control the reciprocating needle-pricking device of the garlic bolt harvesting robot to perform needle-pricking operations on the garlic plants, and control the Delta robotic arm of the garlic bolt harvesting robot to pull out the garlic bolts to complete the automated harvesting operation.

[0011] In a possible implementation, the traditional A* algorithm is optimized and improved by using the Grey Wolf algorithm, and the improved A* algorithm is fused with the DWA algorithm to realize the global and local planning of the motion chassis of the garlic bolt harvesting robot, including:

[0012] Before path planning, farmland information is collected, including data on plot contour points and obstacle contour points. The plot contour points obtain high-precision position data through the u-blox RTK device and the rtk_gps ROS driver, and data fusion is carried out in combination with the Moncee lidar to support LIO-SAM SLAM mapping. The obstacle contour points are measured by the RTK device and the tape measure tool;

[0013] Generate a geometry_msgs / Polygon model, optimize the obstacle boundary using the pcl_ros package of ROS, and publish it as the global map nav_msgs / OccupancyGrid;

[0014] The path planning is based on the AB line operation mode, which is used to generate a quadrilateral plot, covering the generation of the trapezoidal area for unmanned operation, the division of the turning area and the operation area, the segmentation of the operation strip, the processing of the obstacle strip, the routing of the operation strip, and the generation of the final path.

[0015] In a possible implementation, the path planning is based on the AB line operation mode, which is used to generate a quadrilateral plot, covering the generation of the trapezoidal area for unmanned operation, the division of the turning area and the operation area, the segmentation of the operation strip, the processing of the obstacle strip, the routing of the operation strip, and the generation of the final path, including:

[0016] Generation of the trapezoidal area for unmanned operation: Considering the kinematic model of the garlic bolt harvesting robot and the path planning requirements comprehensively, taking the operation direction as the parallel side of the trapezoidal area, the plot is spatially divided. Within the quadrilateral plot with regular boundaries, a maximized trapezoidal area is constructed through geometric analysis, so that the driving direction of the operation path is aligned with the parallel side of the trapezoidal area, thus ensuring the parallelism of the path;

[0017] Division of the turning area and the operation area: According to the operation width w and turning radius r of the garlic bolt harvesting robot, determine the width W of the turning area turn , and the specific calculation relationship is as follows:

[0018]

[0019] Combined with the operation direction parameter, two parallelograms with a width of W are divided along the non-operation direction boundary of the unmanned operation trapezoidal area for the turning operation of the equipment, where: turn W is the width of the turning area, that is, the space width required for the robot to turn around, w is the working width of the robot, that is, the ground width covered by the robot during operation, r is the turning radius of the robot, that is, the minimum radius at which the robot can complete a smooth turn, W: the total width of the operation area, that is, the width of the entire operation area; turn

[0020] Operation strip division: In the operation area, the basic operation unit is the operation strip, and the center line of the operation strip represents the main path of the operation; the operation area is divided into strips to generate operation straight lines, that is, the center lines of the operation strips; the direction of the operation strip is consistent with the movement direction of the garlic bolt harvesting robot, and its width is the operation width. The division process starts from the starting side of the trapezoidal area, extends along the parallel sides based on the operation direction until the entire operation area is covered, and finally completes the comprehensive division of the operation strips;

[0021] Operation strip routing: With the goal of maximizing the operation distance ratio, a turning mode with a short turning distance is constructed, so that the turning of adjacent sequential operations is in a U-shaped mode, and the distance between adjacent sequential operation straight lines is not less than 2r and should be an integer multiple of the operation width; the block operation mode divides all operation paths into N blocks unit blocks and 1 remaining block, and defines the number of operation strips in each unit block as n block , and the expression is as follows:

[0022]

[0023] Then the number of unit blocks N blocks , and the expression is as follows:

[0024]

[0025] where: n block is the number of operation strips included in each unit block, indicating the number of path strips that the robot can operate within each block, N blocks represents the total number of operation blocks, that is, the number of blocks required to divide the operation area, N operate represents the total number of operation strips, that is, the number of strips of all operation paths within the entire operation area;

[0026] Turning path generation: The turning path is generated between adjacent sequential operation straight lines. According to the turning radius parameter of the garlic bolt harvesting robot and the coordinates of two operation straight lines, a U-shaped turning path in the form of arc-line-arc is generated.

[0027] Final path generation: Insert high-speed driving and operation start instructions at the straight operation path points, and insert low-speed driving and non-operation instructions at the bypassing and turning curve path points. The finally generated path contains instruction information.

[0028] In a possible implementation manner, the use of a depth camera to collect image information of garlic bolts and the use of the YOLO11 object detection algorithm to identify and accurately locate garlic bolts include:

[0029] The YOLO11 object detection model improved based on the Ghost convolutional lightweight network is used for positioning and recognition. It adopts a more efficient convolutional structure and an optimized detection algorithm. Among them, the Ghost module first generates a small number of intrinsic feature maps using ordinary convolution, and then uses linear operations to enhance features and increase channels, achieving the effect of reducing the operation cost and improving the recognition performance. Given the input data X ∈ Rc×h×w, where c is the number of input channels, h and w are the height and width of the input data respectively, the operation formula for any convolutional layer to generate n feature maps is:

[0030] Y = X * f + b

[0031] where * is the convolution operation, b is the bias term, X represents the input data, and f is the convolution kernel;

[0032] To further obtain the required n feature maps, a series of inexpensive linear operations are applied to each intrinsic feature in Y, and s ghost features are generated according to the following function:

[0033]

[0034] where: y ij is a generated ghost feature, which is obtained by operating on the input feature map. Here, the indices i and j represent the i-th original feature and the j-th ghost feature; Φ i,j () is a function or operation that maps the original feature map y′ i to the new ghost feature y ij , y′ i is the original feature map;

[0035] The kernel size is customized through the main convolution in the Ghost module. Pointwise convolution is used to process features across channels, and then depth convolution is used to process spatial information. The theoretical speed-up ratio of upgrading the ordinary convolution by the Ghost module using linear operations of the same size is:

[0036]

[0037] Wherein: n is the number of output feature maps, c is the number of input channels, k is the size of the convolution kernel, s is the multiple for generating Ghost features, and d is the spatial dimension of the Ghost features;

[0038] During the object detection process, an improved YOLO11 model and a depth camera are used to accurately identify and locate the object. First, the image collected by the camera is processed to identify the target object in the image, and the bounding box and confidence information of each target are returned; subsequently, the system uses the depth value of each pixel in the depth image, combines the position of the target object in the image, calculates the three-dimensional coordinates of each target, and transmits them to the host computer for coordinate conversion to obtain the three-dimensional coordinates of the target.

[0039] In a possible implementation, by integrating the functional modules of the garlic bolt harvesting robot and proposing a cooperative control method, the adaptive adjustment of operation parameters and the optimization of the operation rhythm and transfer mode of the modules are realized, including:

[0040] Using a machine vision system to perform real-time detection and positioning of garlic bolts, and identifying the position and size information of garlic bolts through image processing technology and deep learning algorithms;

[0041] According to the visual recognition result, the robot motion chassis adjusts the moving path, and the robotic arm system accurately plucks according to the position information of the target garlic bolt;

[0042] The robot motion chassis dynamically adjusts the moving speed and path according to the garlic bolt distribution density and the harvesting task requirements, so as to avoid excessive time waste in sparse areas. The robot motion chassis is synchronized with the visual system to ensure that the path planning is adjusted according to the real-time visual feedback and is synchronized with the operation speed of the robotic arm system;

[0043] The robotic arm accurately adjusts the action trajectory according to the garlic bolt position and attitude information provided by the visual system, grabs and picks the target garlic bolt. After the picking task is completed, the robotic arm coordinates with the robot motion chassis to ensure that it does not interfere with the walking trajectory during the picking process;

[0044] The robotic arm control module operates relying on the feedback information provided by the visual system to ensure that the grasping action is accurate; at the same time, the motion rhythm of the robotic arm needs to cooperate with the robot motion chassis to avoid conflicts during the picking process;

[0045] Combining the real-time data of garlic bolt harvesting, adaptively adjusting the working parameters of each functional module, including the sensitivity of visual recognition, the speed of the robot motion chassis, and the grasping force of the robotic arm;

[0046] Obtain the working status of each module in real time through the perception layer, coordinate the working rhythms among the modules, avoid conflicts or repeated operations, and optimize the overall harvesting efficiency;

[0047] Adopt a multi-level control structure and dynamically schedule each module according to the task priority and the working status of the module;

[0048] Through a closed-loop control and feedback mechanism, ensure that the working rhythms and actions of each module are coordinated, and avoid resource waste and ineffective operations.

[0049] In a possible implementation manner, controlling the reciprocating needle-pricking device of the garlic bolt harvesting robot to perform needle-pricking operations on the garlic plants according to the image coordinate positioning result, and controlling the Delta robotic arm of the garlic bolt harvesting robot to pull out the garlic bolts to complete the automated harvesting operation, including:

[0050] According to the obtained three-dimensional coordinates of the garlic bolt in the world coordinate system The cooperative control system controls the reciprocating needle-pricking device to perform needle-pricking operations on the garlic plants, and then performs coordinate transformation according to the three-dimensional coordinates of the garlic bolt in the world coordinate system to obtain the relative three-dimensional coordinates in the Delta robotic arm coordinate system. The transformation formula is as follows:

[0051]

[0052] Where: ΔX is the offset in the X direction, ΔY is the offset in the Y direction, ΔZ is the offset in the Z direction, X, Y, Z are the three-dimensional coordinates in the Delta robotic arm coordinate system, X w , Y w , Z w The three-dimensional coordinates of the garlic plant in the world coordinate system;

[0053] After obtaining the three-dimensional coordinates of the garlic bolt in the Delta robotic arm coordinate system, through the inverse kinematics solution equation of the Delta robotic arm, obtain the variable parameters of each joint of the robotic arm;

[0054] Determine the end posture of the robotic arm according to the variable parameters of each joint of the Delta robotic arm;

[0055] Then, according to the detection result, the upper computer automatically controls the robotic arm to accurately pull out the garlic bolt to realize the garlic bolt harvesting operation.

[0056] In a possible implementation manner, after obtaining the three-dimensional coordinates of the garlic bolt in the Delta robotic arm coordinate system, through the inverse kinematics solution equation of the Delta robotic arm, obtaining the variable parameters of each joint of the robotic arm, including:

[0057] Assume that between the moving platform and the static platform is Op = [x y z] T;

[0058] The position vector between the connection of the two rods and the static platform is:

[0059] where

[0060] the spatial position vector of the connecting rod can be obtained as:

[0061] Thus, the kinematic solution equations can be obtained as:

[0062]

[0063] Substituting respectively, we get:

[0064]

[0065] Thus, the variable parameters of each joint of the Delta robot arm are determined, where: The vector between the moving platform and the static platform, Op, represents the position of the moving platform in the coordinate system of the static platform. x, y, and z respectively represent the positions of the moving platform in the X-axis, Y-axis, and Z-axis directions. OBi represents the position vector between the i-th connecting rod or joint connection and the static platform. R is the radius of the static platform, r is the radius of the moving platform, L is the length of the active arm, θi is the joint angle of the i-th rod, controlling the telescopic movement of the rod, the direction angle of the i-th rod, which determines the direction of the connecting rod in the plane. a is a constant, usually representing a known fixed distance between the end effector and a specific reference point of the robot arm.

[0066] In a possible implementation, the garlic bolt harvesting robot includes: a chassis, a dividing device, a Delta parallel robot arm, a flexible clamping and conveying device, a reciprocating needle-pricking device, a camera, a mechanical gripper, a collecting device, a collaborative control system, a lidar, an industrial control computer, an rtk measuring instrument, a DC power supply, and a DC air pump, where: The DC power supply provides the power source for the harvesting robot. The chassis is controlled by a driving motor to make the two rear wheels move, and the two front wheels control the steering. The dividing device is installed at the front end of the flexible clamping and conveying device. The flexible clamping and conveying device is located directly below the Delta parallel robot arm. The camera is installed on the chassis of the harvesting machine, and the shooting end is horizontally oriented towards the flexible clamping and conveying device. The reciprocating needle-pricking device is located directly below the flexible clamping and conveying device. The mechanical gripper is placed at the end of the Delta parallel robot arm. The collecting device is fixed at the end of the chassis. The lidar is installed at the upper end of the frame. The rtk measuring instrument is installed at the rear end of the frame. The industrial control computer is installed on the top of the harvesting machine. The collaborative control system is installed at the side end of the frame, used to realize the orderly operation of various functions of the garlic bolt harvesting robot.

[0067] In a possible implementation, the flexible clamping and conveying device provides power for the clamping belt through the movement of the ground wheels. The movement speed of the clamping belt is the same as the walking speed of the chassis but in the opposite direction, so that the garlic plants can maintain a relatively stable state and prevent the garlic plants from swinging back and forth during subsequent needle insertion.

[0068] In a possible implementation, the reciprocating needle insertion device mainly drives the crank by a servo motor to perform the needle insertion operation on the garlic bolt. The adjacent two cranks move in opposite directions. The garlic plants are conveyed by the flexible clamping and conveying device. When the garlic bolt is recognized by the depth camera, the needle insertion device performs the needle insertion operation on the garlic plants, and the needle insertion operation can effectively reduce the resistance of pulling the bolt.

[0069] In the embodiment of the present application, by acquiring the image information of the garlic bolt, accurate recognition and positioning of the garlic bolt are realized based on the YOLO11 target detection algorithm. Then, the cooperative control system controls the reciprocating needle insertion device to perform the needle insertion operation on the garlic plants. Finally, the Delta robotic arm of the harvesting device is controlled according to the image coordinate positioning result to pull, transfer and collect the garlic bolt, realizing the automated harvesting operation. Description of the Drawings

[0070] Figure 1 It is a schematic flow chart of a garlic bolt harvesting method based on a Delta parallel robotic arm harvesting robot provided by the embodiment of the present application;

[0071] Figure 2 It is a flow chart of the GWO-A* algorithm provided by the embodiment of the present application;

[0072] Figure 3 It is a flow chart of the DWA-A* algorithm provided by the embodiment of the present application;

[0073] Figure 4 It is a schematic diagram of path planning provided by the embodiment of the present application;

[0074] Figure 5 It is a schematic diagram of the camera coordinates provided by the embodiment of the present application;

[0075] Figure 6 It is a schematic diagram of the cooperative control module provided by the embodiment of the present application;

[0076] Figure 7 It is a schematic diagram of the structure of the garlic bolt harvesting robot provided by the embodiment of the present application;

[0077] Figure 8 It is the flexible clamping and conveying device of the garlic bolt harvesting robot provided by the embodiment of the present application;

[0078] Figure 9 It is the reciprocating needle insertion device of the garlic bolt harvesting robot provided by the embodiment of the present application;

[0079] Figure 10 This is the mechanical gripper of the garlic bolt harvesting robot provided by the embodiment of the present application. Specific implementation manners

[0080] The following combines the accompanying drawings and specific implementation manners to elaborate on this solution.

[0081] See Figure 1 , the garlic bolt harvesting method of the Delta parallel manipulator-based harvesting robot provided by this embodiment includes

[0082] S101, using the grey wolf algorithm to optimize and improve the traditional A* algorithm, and fusing the improved A* algorithm with the DWA algorithm to achieve the global and local planning of the motion chassis of the garlic bolt harvesting robot.

[0083] In this embodiment, the grey wolf algorithm is used to optimize and improve the traditional A* algorithm, as Figure 2 shown. The improved A* algorithm is fused with the DWA algorithm (dynamic window method), as Figure 3 shown, so that the motion chassis of the robot obtains better global and local planning capabilities, can cope with various different complex working conditions, and can realize operations such as automatic row following and intelligent obstacle avoidance.

[0084] The motion chassis of the robot is equipped with a Beidou / GPS navigation and positioning device, which is connected to the serial server of the industrial computer through the CAN network to transmit motion signals to the chassis controller. To achieve row-by-row harvesting, a path planning method based on the fusion of the improved A* algorithm and the DWA algorithm is designed. Before path planning, farmland information needs to be collected, including data of plot contour points and obstacle contour points. The plot contour points obtain high-precision position data through the u-blox RTK device and the rtk_gps ROS driver, and data fusion is carried out in combination with the Moncee lidar to support LIO-SAM SLAM mapping. The obstacle contour points are measured by the RTK device and the tape measure tool to generate a geometry_msgs / Polygon model. The pcl_ros package of ROS is used to optimize the obstacle boundary and publish it as a global map nav_msgs / OccupancyGrid. The path planning is based on the AB line operation mode, which is applicable to quadrilateral plots and covers modules such as the generation of unmanned operation trapezoidal areas, the division of turning areas and operation areas, the segmentation of operation strips, the processing of obstacle strips, the routing of operation strips, and the generation of the final path.

[0085] As Figure 4As shown in the figure, generation of the trapezoidal area for unmanned operation: In the path planning of the garlic bolt harvesting robot, to ensure the parallelism of the operation path and the operation efficiency, it is necessary to adaptively divide the plot. Considering the kinematic model of the garlic bolt harvesting robot and the requirements of path planning comprehensively, the operation direction is used as the parallel side of the trapezoidal area to divide the plot spatially. Within a quadrilateral plot with regular boundaries, a maximized trapezoidal area is constructed through geometric analysis, making the driving direction of the operation path align with the parallel side of the trapezoidal area, thereby ensuring the parallelism of the path.

[0086] Division of the turning area and the operation area. According to the operation width w and the turning radius r of the garlic bolt harvesting robot, determine the width W of the turning area turn , and the specific calculation relationship is as follows:

[0087]

[0088] On this basis, combined with the operation direction parameter, divide 2 parallelograms with a width of W along the non-operation direction boundary of the trapezoidal area for unmanned operation turn for the turning operation of the equipment. Among them: W turn is the width of the turning area, that is, the spatial width required for the robot to turn around, w is the operation width of the robot, that is, the ground width covered by the robot during operation, r is the turning radius of the robot, that is, the minimum radius at which the robot can complete a smooth turn, and W is the total width of the operation area, that is, the width of the entire operation area.

[0089] Division of operation strips. In the operation area, the basic operation unit is the operation strip, and the center line of the operation strip represents the main path of the operation. First, divide the operation area into strips to generate operation straight lines, that is, the center lines of the operation strips. The direction of the operation strip is consistent with the movement direction of the garlic bolt harvesting robot, and its width is the operation width. The division process starts from the starting side of the trapezoidal area, extends along the parallel side based on the operation direction until it covers the entire range of the operation area, and finally completes the comprehensive division of the operation strips.

[0090] Processing of obstacle strips. The obstacle strip refers to the operation bandwidth area that intersects with the obstacle. Since both line obstacles and point obstacles have been abstracted into polygons represented by vertices, geometric processing methods can be uniformly used. The optimization of the obstacle bypass strategy comprehensively considers the turning radius, the shape of the obstacle, and the comprehensive influence of the operation strip.

[0091] Operation strip routing. To meet the high operation distance ratio requirement of the garlic bolt harvesting robot, the operation strip routing method is used to sort the operation strips. With the maximum operation distance ratio as the goal, a turning mode with a short turning distance is constructed. It is required that the turning between adjacent sequential operations is in a U-shaped mode, and the straight-line distance between adjacent sequential operations is not less than 2r and should be an integer multiple of the operation width. Therefore, the block nested row operation mode is adopted for the design of the operation strip routing method. The block operation mode divides all operation paths into N blocks unit blocks and 1 remaining block. Define the number of operation strips in each unit block as n block , as shown in the following expression:

[0092]

[0093] Then the number of unit blocks N blocks , as shown in the following expression:

[0094]

[0095] Where: n block is the number of operation strips included in each unit block, representing the number of path strips that the robot can operate within each block, and N blocks represents the total number of operation blocks, that is, the number of blocks required to divide the operation area, and N operate represents the total number of operation strips, that is, the number of strips of all operation paths within the entire operation area.

[0096] Turning path generation. The turning path is generated between adjacent sequential operation straight lines. According to the turning radius parameter of the garlic bolt harvesting robot and the coordinates of the two operation straight lines, a U-shaped turning path in the form of "arc-line-arc" can be generated

[0097] Final path generation. To ensure the speed state and implement state of the garlic bolt harvesting robot, two types of basic instructions are designed during the path planning process. Insert high-speed driving and operation start instructions at the straight-line operation path points, and insert low-speed driving and non-operation instructions at the bypassing and turning curve path points. In this way, the finally generated path will contain instruction information, thus forming a sequence of operation path points with instructions that adapts to the operation requirements, ensuring that the garlic bolt harvesting robot can execute tasks at an appropriate speed and state in different operation scenarios.

[0098] S102, Use a depth camera to collect image information of the garlic bolts, and identify and accurately locate the garlic bolts through the YOLO11 object detection algorithm.

[0099] In this embodiment, an Intel RealSense D435i binocular camera is used to obtain depth information. This camera consists of an infrared sensor (IR Stereo), an infrared laser emitter (IR Projector), and a color camera (RGB Camera). The inertial measurement unit (IMU) in the camera is used to obtain the data of the accelerometer and gyroscope, which is used to perceive and understand the motion state of the object. The Jetson Xavier NX edge computing developer kit is used as the computing platform for depth information processing to calculate and process the depth information. The Hand-Eye Calibration package in the ROS robot operating system is used to calibrate the camera on the robot's hand for the robotic arm. Camera calibration is used to obtain the internal parameters of the camera, with the aim of obtaining accurate camera parameters for tasks such as accurate visual measurement, 3D reconstruction, and image correction. Through camera calibration, applications such as eliminating image distortion, determining the true size of an object, achieving accurate positioning and pose estimation can be carried out. And using the camera-on-the-hand method can reduce the interference of environmental factors such as the occlusion of the garlic bolt plant by the robotic arm during grasping, improving the accuracy of positioning and grasping and the stability in a dynamic environment, so as to achieve accurate identification and positioning of the garlic bolt.

[0100] Target positioning is achieved by establishing a correspondence between the information of the target object in the image captured by the D435i camera and the points existing on the actual object in space. Generally, a camera has four coordinate systems, namely the pixel coordinate system, the image coordinate system, the camera coordinate system, and the world coordinate system. Coordinate transformation can be performed between adjacent two coordinate systems through a translation or rotation matrix. This transformation process is actually the transformation of coordinates from the pixel coordinate system to the world coordinate system. Through coordinate transformation, we can obtain the three-dimensional coordinates of the target object in the world coordinate system.

[0101] As Figure 5 shown, the pixel coordinate system UV is the most basic coordinate system, which belongs to the image coordinate system XY. Generally, the origin of the pixel coordinate system is at the upper left corner of the image. Therefore, there is a translational transformation relationship between the two coordinate systems. If the coordinates of the detected target in the pixel coordinate system are (u 0 , v 0 ), then the pixel coordinate system and the image coordinate system satisfy the following relational formula:

[0102]

[0103] Given the coordinates in the image coordinate system, the next step is to calculate the coordinates in the camera coordinate system using the camera internal parameters. During the calculation process, the inverse matrix of the camera internal parameters is used to normalize the three-dimensional coordinates in the camera coordinate system to make them homogeneous coordinates. This normalization process uses the inverse matrix of the camera internal parameters. The transformation formula is:

[0104]

[0105] Among them, f x and f y and c x and c y are some parameters of the camera internal parameter matrix. X c and Y c and Z c are coordinates in the normalized camera coordinate system. Among them, Z c is the depth information, which can be obtained by using the functions of the camera library in the program. Finally, the rotation and translation matrix obtained by the previous joint calibration of the RGB camera and the binocular camera can be used to calculate the three-dimensional coordinates in the world coordinate system. The conversion formula is as follows:

[0106]

[0107] Among them, X w and Y w and Z w are coordinates in the world coordinate system. R is a 3×3 rotation matrix, and t is a 3×1 translation matrix. The two together constitute the external parameters of the camera. Through calculation, the three-dimensional coordinates of the garlic bolt in the world coordinate system are obtained.

[0108] The visual recognition system performs localization and recognition based on a YOLO11 object detection model improved by a Ghost convolutional lightweight network. This model adopts a more efficient convolutional structure and an optimized detection algorithm, and can provide a faster inference speed while ensuring high accuracy. Among them, the Ghost module first generates a small number of intrinsic feature maps using ordinary convolutions, and then uses inexpensive linear operations to enhance features and increase channels, achieving the effect of reducing the computational cost and improving the recognition performance. Given the input data X∈Rc×h×w, where c is the number of input channels, h and w are the height and width of the input data respectively, the operation formula for any convolutional layer to generate n feature maps is:

[0109] Y = X * f + b

[0110] where * is the convolution operation, b is the bias term, X represents the input data, and f is the convolution kernel. To further obtain the required n feature maps, a series of inexpensive linear operations are applied to each intrinsic feature in Y, and s ghost features are generated according to the following function:

[0111]

[0112] where: y ij is a generated ghost feature, which is obtained by performing some operations on the input feature map. Here, the indices i and j represent the i-th original feature and the j-th ghost feature; Φi,j () is a function or operation that maps the original feature map y′ i to the new ghost feature y ij where y′ i is the original feature map.

[0113] The kernel size is customized through the main convolution in the Ghost module. Pointwise convolution is used to process features across channels, and then depthwise convolution is used to process spatial information. The theoretical speedup ratio of upgrading ordinary convolution by the Ghost module using linear operations of the same size is:

[0114]

[0115] where: n is the number of output feature maps, c is the number of input channels, k is the size of the convolution kernel, s is the multiple of generating ghost features, and d is the spatial dimension of the ghost features.

[0116] During the object detection process, an improved YOLO11 model and a depth camera are used to accurately identify and locate objects. First, the images collected by the camera are processed to identify the target objects in the images, and the bounding box and confidence information of each target are returned. Subsequently, the system uses the depth value of each pixel in the depth image, combined with the position of the target object in the image, to calculate the three-dimensional coordinates of each target and transmit them to the host computer for coordinate transformation to obtain the three-dimensional coordinates of the target. Through various data fusion methods, the system can achieve high-precision object positioning, especially under the conditions of complex outdoor environments, and still maintain a high recognition accuracy.

[0117] S103, by integrating the functional modules of the garlic bolt harvesting robot and proposing a cooperative control method, realizes the adaptive adjustment of operation parameters and the optimization of the module operation rhythm and transfer mode.

[0118] Such as Figure 6As shown, a machine vision system is used to perform real-time detection and positioning of garlic bolts. Through image processing technology and deep learning algorithms, the system can identify information such as the position and size of garlic bolts. According to the visual recognition results, the robot motion chassis adjusts the moving path, and the robotic arm system precisely plucks the target garlic bolt based on the position information of the target garlic bolt. The robot motion chassis dynamically adjusts the motion speed and path according to the garlic bolt distribution density and the requirements of the harvesting task to avoid excessive time waste in sparse areas. The robot motion chassis needs to be synchronized with the vision system to ensure that the path planning is adjusted according to the real-time visual feedback and is synchronized with the operating speed of the robotic arm system. The robotic arm precisely adjusts the motion trajectory according to the garlic bolt position and attitude information provided by the vision system, grasps and picks the target garlic bolt. After the picking task is completed, the robotic arm needs to coordinate with the robot motion chassis to ensure that it does not interfere with the walking trajectory during the picking process. The robotic arm control module operates relying on the feedback information provided by the vision system to ensure that the grasping action is accurate; at the same time, the motion rhythm of the robotic arm needs to cooperate with the robot motion chassis to avoid conflicts during the picking process. Combining the real-time data of garlic bolt harvesting (such as the operating environment, operating progress, maturity of garlic bolts, etc.), the system can adaptively adjust the working parameters of each functional module, including the sensitivity of visual recognition, the speed of the robot motion chassis, the grasping force of the robotic arm, etc. By obtaining the working status of each module in real time through the sensing layer, coordinating the working rhythms between modules, avoiding conflicts or repeated operations, and optimizing the overall harvesting efficiency. A multi-level control structure is adopted to dynamically schedule each module according to the task priority and the working status of the module. For example, if the garlic bolt picking task in a certain area is completed, the robot motion chassis can adjust the forward speed, and the robotic arm can also prepare for the next picking task. Through closed-loop control and feedback mechanisms, ensure that the working rhythms and actions of each module are coordinated and consistent, avoiding resource waste and ineffective operations. The action timing of the robotic arm needs to be synchronized with the running rhythm of the robot motion chassis to avoid the robotic arm making operations at inappropriate times.

[0119] S104. According to the image coordinate positioning result, control the reciprocating needle-pricking device of the garlic bolt harvesting robot to perform needle-pricking operations on the garlic plants, and control the Delta robotic arm of the garlic bolt harvesting robot to pluck the garlic bolts to complete the automated harvesting operation.

[0120] See Figure 7, the garlic bolt harvesting robot in this example includes: a chassis 4, a dividing device 1, a Delta parallel manipulator 12, a flexible clamping and conveying device 2, a reciprocating needle-pricking device 3, a camera 14, a mechanical gripper 13, a collection device 5, a collaborative control system 10, a lidar 8, an industrial computer 9, an rtk surveying instrument 6, a DC power supply 7, and a DC air pump 11. Among them: the DC power supply 7 provides the power source for the harvesting robot. The DC power supply 7 and the DC air pump 11 are arranged on the top plane of the harvester. The chassis 4 is controlled by a driving motor to make the two rear wheels move, and the two front wheels control the steering. The dividing device 1 is installed at the front end of the flexible clamping and conveying device 2. The flexible clamping and conveying device 2 is located directly below the Delta parallel manipulator 12. The camera 14 is installed on the chassis 4 of the harvester, and the shooting end is horizontally oriented towards the flexible clamping and conveying device 2. The reciprocating needle-pricking device 3 is located directly below the flexible clamping and conveying device 2. The mechanical gripper 13 is arranged at the end of the Delta parallel manipulator 12. The collection device 5 is fixed at the end of the chassis. The lidar 8 is arranged at the upper end of the frame. The rtk surveying instrument 6 is arranged at the rear end of the frame. The industrial computer 9 is arranged on the top of the harvester. The collaborative control system 10 is arranged at the side end of the frame, which is used to realize the orderly operation of various functions of the garlic bolt harvesting robot.

[0121] As Figure 8 shown, the flexible clamping and conveying device mainly consists of a driving wheel 3, a driven wheel 5, a pressing wheel 1, a connecting frame 4, and a clamping belt 2. The flexible clamping and conveying device provides power for the clamping belt through the movement of the ground wheel. The driving wheel 3 drives the driven wheel 5 by driving the clamping belt 2 to move. The moving speed of the clamping belt is the same as the walking speed of the chassis but in the opposite direction. The distance between the two clamping belts is just the distance of one garlic plant. The garlic plant remains relatively stationary during the clamping and conveying process, preventing the garlic plant from swinging back and forth during subsequent needle pricking, and providing a relatively stable working condition for the needle pricking operation.

[0122] As Figure 9 shown, the reciprocating needle-pricking device mainly consists of a needle 1, a flexible baffle 2, a servo motor 3, a crank-rocker mechanism 4, a connecting rod 5, and a deflector 6. The needle 1 and the flexible baffle 2 are arranged on the connecting rod 5. The connecting rod 5 is fixedly connected to the crank-rocker mechanism 4. The deflector 6 is installed at the front end of the connecting rod 5, which is used to guide the garlic plant to the needle-pricking position. The reciprocating needle-pricking device is driven by the servo motor 3 to drive the crank-rocker mechanism 4 to perform reciprocating motion, realizing the needle-pricking operation on the garlic bolt. The adjacent two cranks move in opposite directions, which can effectively prevent the garlic plant from being broken at the needle-pricking position. The garlic plant is conveyed by the flexible clamping and conveying device. When the garlic bolt is recognized by the depth camera, the needle-pricking device performs the needle-pricking operation on the garlic plant. After needle pricking, air seeps into the false stem that was originally in a vacuum state, greatly reducing the pulling force of the garlic bolt. The needle-pricking operation can effectively reduce the resistance of pulling the bolt.

[0123] As Figure 10As shown in the figure, the mechanical gripper consists of a left gripper 1, a right gripper 3, finger pads 2, a connecting rod 4, and a cylinder 5. The mechanical gripper is connected to the Delta parallel manipulator through a connecting plate. Finger pads are installed on both grippers. The finger pads 2 are made of silicone material that meets food safety standards. First, it can increase the friction when pulling garlic sprouts, preventing the garlic sprouts from slipping during the pulling process. Second, it can reduce the damage to the garlic sprouts when clamping them. A pressure sensor is installed on the mechanical gripper. Based on the data feedback from the pressure sensor, an appropriate air pressure is selected to ensure that the garlic sprouts can be successfully pulled out without being damaged.

[0124] Regarding the harvesting of the manipulator, the control device obtains the relative three-dimensional coordinates in the Delta manipulator coordinate system through coordinate transformation based on the three-dimensional coordinates of the spatial position of the target garlic sprout transmitted by the host computer. The transformation formula is as follows: For the harvesting of the manipulator, the control device obtains the relative three-dimensional coordinates in the Delta manipulator coordinate system through coordinate transformation based on the three-dimensional coordinates of the spatial position of the target garlic sprout transmitted by the host computer. The transformation formula is as follows:

[0125]

[0126] Where: ΔX is the offset in the X direction, ΔY is the offset in the Y direction, ΔZ is the offset in the Z direction, X, Y, Z are the three-dimensional coordinates in the Delta manipulator coordinate system, X w , Y w , Z w are the three-dimensional coordinates of the garlic plant in the world coordinate system.

[0127] Then, the joint variables and the end pose of the manipulator are determined through the relative three-dimensional coordinates in the Delta manipulator coordinate system, and the manipulator is controlled to complete the garlic sprout harvesting operation. The problem of determining the joint variables and the end pose is the problem of forward and inverse kinematics solution of the Delta manipulator, which is usually calculated using the following equations:

[0128] Let the position between the moving platform and the static platform be Op = [x y z] T

[0129] The position vector between the connection of the two rods and the static platform is:

[0130] Where

[0131] The spatial position vector of the connecting rod can be obtained as:

[0132] From this, the kinematic solution equations can be obtained as

[0133]

[0134] Among them, substituting respectively, we get:

[0135]

[0136] Thus, the joint variables and the end pose of the Delta robot arm are determined, where: The vector between the moving platform Op and the static platform represents the position of the moving platform in the coordinate system of the static platform. x, y, and z respectively represent the positions of the moving platform in the X-axis, Y-axis, and Z-axis directions. OBi represents the position vector between the i-th link or joint connection and the static platform. R is the radius of the static platform, r is the radius of the moving platform, L is the length of the active arm, θi is the joint angle of the i-th rod, controlling the telescopic movement of the rod. The direction angle of the i-th rod determines the direction of the link in the plane. a is a constant, usually representing a known fixed distance between the end effector and a specific reference point of the robot arm. Through the automatic control of the upper computer, the harvested garlic bolt is transported to the collection device to realize the automated harvesting operation of the garlic bolt.

[0137] In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent the cases of A existing alone, A and B existing simultaneously, and B existing alone. Where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one of the following" and its similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0138] As described above, this is only the specific implementation manner of the present application. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. The protection scope of the present application shall be subject to the protection scope of the claimed rights.

Claims

1. A garlic stalk harvesting method based on a Delta parallel mechanical arm harvesting robot, characterized in that: include: The Grey Wolf Algorithm is used to optimize and improve the traditional A* algorithm, and the improved A* algorithm is integrated with the DWA algorithm to achieve the global and local planning of the motion chassis of the garlic stalk harvesting robot. The image information of garlic stalks is collected using a depth camera, and the garlic stalks are identified and accurately located using the YOLO11 target detection algorithm. By integrating the functional modules of the garlic stalk harvesting robot and proposing a collaborative control method, the adaptive adjustment of operating parameters and the optimization of module operation rhythm and switching mode are achieved; According to the image coordinate positioning results, the reciprocating needle-piercing device of the garlic stalk harvesting robot is controlled to perform needle-piercing operations on the garlic plants, and the Delta robotic arm of the garlic stalk harvesting robot is controlled to pull the garlic stalks to complete the automated harvesting operation.

2. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The gray wolf algorithm is used to optimize and improve the traditional A* algorithm, and the improved A* algorithm is integrated with the DWA algorithm to achieve global and local planning of the motion chassis of the garlic stalk harvesting robot, including: Before path planning, farmland information is collected, including land contour point and obstacle contour point data. The land contour points obtain high-precision position data through the u-blox RTK device and rtk_gps ROS driver, and are combined with the Moncee lidar for data fusion to support LIO-SAM SLAM mapping. The obstacle contour points are measured through the RTK device and tape measure tool. Generate geometry_msgs / Polygon model, optimize obstacle boundaries using ROS pcl_ros package, and publish as global map nav_msgs / OccupancyGrid; Path planning is based on the AB line operation mode and is used to generate quadrilateral plots, covering the generation of unmanned operation trapezoidal areas, the division of U-turn areas and operation areas, the segmentation of operation strips, the handling of obstacle strips, the routing of operation strips and the generation of the final path.

3. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The path planning is based on the AB line operation mode and is used to generate quadrilateral plots, covering the generation of unmanned operation trapezoidal areas, the division of U-turn areas and operation areas, the segmentation of operation strips, the processing of obstacle strips, the routing of operation strips and the generation of the final path, including: Unmanned operation trapezoidal area generation: Taking into account the kinematic model and path planning requirements of the garlic stalk harvesting robot, the operation direction is used as the parallel side of the trapezoidal area, and the land is spatially divided. In a quadrilateral land with regular boundaries, a maximized trapezoidal area is constructed through geometric analysis, so that the driving direction of the operation path is aligned with the parallel side of the trapezoidal area, thereby ensuring the parallelism of the path; Division of U-turn area and working area: Determine the width W of the U-turn area according to the working width w and turning radius r of the garlic stalk harvesting robot turn , the specific calculation relationship is as follows: Combined with the working direction parameters, two zones with a width of W are divided along the non-working direction boundary of the unmanned working trapezoidal area. turn The parallelogram of is used for the U-turn operation of the equipment, where: W turn is the width of the U-turn area, i.e., the width of the space required for the robot to make a U-turn; w is the width of the robot's operation, i.e., the width of the ground covered by the robot during operation; r is the turning radius of the robot, i.e., the minimum radius at which the robot can smoothly complete a turn; W: the total width of the operation area, i.e., the width of the entire operation area; Operation strip segmentation: In the operation area, the basic operation unit is the operation strip, where the center line of the operation strip represents the main operation path; the operation area is divided into strips to generate an operation straight line, that is, the center line of the operation strip; the direction of the operation strip is consistent with the movement direction of the garlic stalk harvesting robot, and its width is the operation width. The segmentation process starts from the starting edge of the trapezoidal area, and extends along the parallel edges based on the operation direction until the entire operation area is covered, and finally the comprehensive segmentation of the operation strip is completed; Operation strip routing: With the maximum operation distance ratio as the goal, a U-turn pattern with a short U-turn distance is constructed, so that the U-turn of adjacent sequential operations is a U-shaped pattern, and the distance between adjacent sequential operation lines is not less than 2r, and should be an integer multiple of the operation width; the block operation mode divides all operation paths into N blocks unit blocks and 1 remaining block, and define the number of job stripes for each unit block as n block , the expression is as follows: Then the number of unit blocks N blocks , the expression is as follows: Where: n block The number of operation strips contained in each unit block, indicating the number of path strips that the robot can operate in each block, N blocks Indicates the total number of operating blocks, that is, the number of blocks required to divide the operating area, N operate Indicates the total number of operation stripes, that is, the number of stripes of all operation paths in the entire operation area; U-turn path generation: The U-turn path is generated between adjacent working lines. According to the turning radius parameter of the garlic stalk harvesting robot and the coordinates of the two working lines, a U-turn path in the form of arc-line-arc is generated; Final path generation: high-speed driving and operation start instructions are inserted at the straight operation path points, and low-speed driving and non-operation instructions are inserted at the detour and U-turn curve path points. The final generated path contains the instruction information.

4. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The method of collecting image information of garlic scapes using a depth camera and identifying and accurately locating the garlic scapes using a YOLO11 target detection algorithm includes: Improved YOLO11 target detection model based on Ghost convolution lightweight network This model performs positioning and recognition, adopts a more efficient convolution structure and optimized detection algorithm, in which the Ghost module uses ordinary convolution to generate a small number of intrinsic feature maps, and then uses linear operations to enhance features and increase channels, so as to reduce computational costs and improve recognition performance; given the input data X∈Rc×h×w, where c is the number of input channels, h and w are the height and width of the input data respectively, the operation formula for generating n feature maps by any convolution layer is: Y=X*f+b Where: * is the convolution operation, b is the bias term, X represents the input data, and f is the convolution kernel; To further obtain the required n feature maps, a series of cheap linear operations are applied to each intrinsic feature in Y to generate s ghost features according to the following function: Where: y ij is a generated ghost feature, which is obtained by some operation on the input feature map. Here, the indexes i and j represent the i-th original feature and the j-th ghost feature; Φ i,j () is a function or operation that transforms the original feature map y ′ i Mapped to the new ghost feature y ij Up, y′ i is the original feature map; The kernel size is customized through the main convolution in the Ghost module, point-wise convolution is used to process features across channels, and then deep convolution is used to process spatial information. The theoretical speedup ratio of upgrading the ordinary convolution to the Ghost module is obtained by linear operations of the same size: Among them: n is the number of output feature maps, c is the number of input channels, k is the size of the convolution kernel, s is the multiple of generating Ghost features, and d is the spatial dimension of Ghost features; During the target detection process, the improved YOLO11 model and depth camera are used to accurately identify and locate the target. First, the image captured by the camera is processed to identify the target object in the image, and the bounding box and confidence information of each target are returned; then, the system uses the depth value of each pixel in the depth image, combined with the position of the target object in the image, to calculate the three-dimensional coordinates of each target and transmit it to the host computer for coordinate conversion to obtain the three-dimensional coordinates of the target.

5. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The method integrates the function modules of the garlic stalk harvesting robot and proposes a collaborative control method to realize the adaptive adjustment of the operation parameters and the optimization of the module operation rhythm and the switching mode, including: Use machine vision systems to detect and locate garlic scapes in real time, and use image processing technology and deep learning algorithms to identify the location and size of garlic scapes. Based on the visual recognition results, the robot's motion chassis adjusts its movement path, and the robotic arm system accurately plucks the target garlic stalks based on their location information; The robot motion chassis dynamically adjusts the motion speed and path according to the distribution density of garlic stalks and the requirements of the harvesting task to avoid wasting too much time in sparse areas. The robot motion chassis is synchronized with the visual system to ensure that the path planning is adjusted according to real-time visual feedback and keeps pace with the operating speed of the robotic arm system; The robotic arm accurately adjusts its motion trajectory based on the position and posture information of the garlic stalks provided by the visual system, grabs and picks the target garlic stalks; after the picking task is completed, the robotic arm coordinates with the robot's motion chassis to ensure that it does not interfere with the walking trajectory during the picking process; The robot arm control module relies on the feedback information provided by the visual system to operate, ensuring that the grasping action is accurate and correct; at the same time, the movement rhythm of the robot arm needs to be coordinated with the robot motion chassis to avoid conflicts during the picking process; Combined with the real-time data of garlic stalk harvesting, the working parameters of each functional module are adaptively adjusted, including the sensitivity of visual recognition, the speed of the robot's motion chassis, and the gripping force of the robotic arm; The working status of each module is obtained in real time through the perception layer, and the working rhythm between modules is coordinated to avoid conflicts or duplication of operations and optimize the overall harvesting efficiency; Adopt a multi-level control structure to dynamically schedule each module according to task priority and module working status; Through closed-loop control and feedback mechanisms, the working rhythm and actions of each module are ensured to be coordinated and consistent, avoiding waste of resources and ineffective operations.

6. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The method controls the reciprocating needle piercing device of the garlic stalk harvesting robot to perform needle piercing operation on the garlic plants according to the image coordinate positioning result, controls the Delta mechanical arm of the garlic stalk harvesting robot to extract the garlic stalks, and completes the automated harvesting operation, including: According to the three-dimensional coordinates of the garlic stalk in the world coordinate system The collaborative control system controls the reciprocating needle-piercing device to perform needle-piercing operations on garlic plants, and then performs coordinate conversion based on the three-dimensional coordinates of the garlic stalks in the world coordinate system to obtain the relative three-dimensional coordinates in the Delta robot arm coordinate system. The conversion formula is as follows: Where: ΔX is the offset in the X direction, ΔY is the offset in the Y direction, ΔZ is the offset in the Z direction, X, Y, Z are the three-dimensional coordinates in the Delta robot arm coordinate system, X w , Y w , Z w The three-dimensional coordinates of the garlic plant in the world coordinate system; After obtaining the three-dimensional coordinates of the garlic scape in the Delta robot arm coordinate system, the variable parameters of each joint of the robot arm are obtained by inversely solving the Delta robot arm kinematics equations; Determine the end posture of the robot arm according to the variable parameters of each joint of the Delta robot arm; Then, based on the test results, the upper computer automatically controls the robotic arm to accurately pull out the garlic stalks to realize the garlic stalk harvesting operation.

7. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 6, characterized in that: After obtaining the three-dimensional coordinates of the garlic scape in the Delta robot arm coordinate system, the variable parameters of each joint of the robot arm are obtained by inversely solving the Delta robot arm kinematics equation, including: Assume that Op = [xyz] between the moving platform and the static platform T ; The position vector between the connection of the two rods and the static platform is: in The spatial position vector of the connecting rod can be obtained as: The kinematic solution equation is: In this, Substitute them respectively to get: Thus, the variable parameters of each joint of the Delta robot arm are determined, where: Op is the vector between the moving platform and the static platform, represents the position of the moving platform in the static platform coordinate system, x, y, z represent the position of the moving platform in the X-axis, Y-axis and Z-axis directions respectively, OBi represents the position vector between the ith link or joint connection and the static platform, R is the radius of the static platform, r is the radius of the moving platform, L is the length of the active arm, θi is the joint angle of the ith rod, and controls the telescopic movement of the rod. The orientation angle of the i-th rod determines the orientation of the link in the plane, and a is a constant that usually represents the known fixed distance between the end effector and a specific reference point of the robotic arm.

8. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 1, characterized in that: The garlic stalk harvesting robot comprises: a chassis, a grass-separating device, a Delta parallel mechanical arm, a flexible clamping and conveying device, a reciprocating pin-piercing device, a camera, a mechanical gripper, a collecting device, a collaborative control system, a laser radar, an industrial computer, an RTK measuring instrument, a DC power supply, and a DC air pump, wherein: the DC power supply provides a power source for the harvesting robot, the chassis is controlled by a driving motor to control the walking of two rear wheels, and the steering is controlled by two front wheels, the grass-separating device is installed at the front end of the flexible clamping and conveying device, and the flexible clamping and conveying device is located directly below the Delta parallel mechanical arm, the camera is installed on the harvester chassis, and the shooting end is horizontally facing the flexible clamping and conveying device, the reciprocating pin-piercing device is located directly below the flexible clamping and conveying device, the mechanical gripper is arranged at the end of the Delta parallel mechanical arm, the collecting device is fixed at the end of the chassis, the laser radar is arranged at the upper end of the frame, the RTK measuring instrument is arranged at the rear end of the frame, the industrial computer is arranged at the top of the harvester, and the collaborative control system is arranged at the side end of the frame, so as to realize the orderly operation of various functions of the garlic stalk harvesting robot.

9. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 8, characterized in that: The flexible clamping and conveying device provides power to the clamping belt through the movement of the ground wheel. The movement speed of the clamping belt is the same as the walking speed of the chassis but in the opposite direction, so that the garlic plants can maintain a relatively stable state and prevent the garlic plants from swinging back and forth during subsequent needle piercing.

10. The garlic stalk harvesting method based on the Delta parallel mechanical arm harvesting robot according to claim 9, characterized in that: The reciprocating needle-piercing device mainly uses a servo motor to drive a crank to achieve the needle-piercing operation on the garlic stalk. The two adjacent cranks move towards each other, and the garlic plants are transported by the flexible clamping and conveying device. When the garlic stalk is recognized by the depth camera, the needle-piercing device performs the needle-piercing operation on the garlic plant. The needle-piercing operation can effectively reduce the resistance to pulling out the stalk.

Images