A three-degree-of-freedom grape harvesting robot and harvesting method

Abstract

The application provides a three-degree-of-freedom grape picking robot, and relates to the field of intelligent agricultural machinery, which comprises a chassis (10), a walking mechanism (20), a rotating plate (30), a three-coordinate mechanism, an end execution mechanism, a visual assembly, a navigation system and an electric control box, the bottom surface of the chassis (10) is provided with the walking mechanism (20), the end surface of the chassis (10) is rotationally provided with the rotating plate (30), and the end surface of the rotating plate (30) is provided with the three-coordinate mechanism; the three-coordinate mechanism comprises a Y-axis assembly, an X-axis assembly and a Z-axis assembly; the end execution mechanism is arranged at one end of the X-axis assembly away from the Y-axis assembly; the visual assembly is located on the upper side of the end execution mechanism; and the navigation system is located on the side of the chassis. The picking robot adopts a three-degree-of-freedom rectangular coordinate design, simplifies the mechanical structure, avoids environmental interference in the picking process, guarantees the flexibility and accuracy of the picking process, and thus realizes fast, efficient and low-cost grape picking.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural intelligent machinery, in particular to a three-degree-of-freedom grape harvesting robot and a harvesting method. BACKGROUND

[0002] Grape is a woody vine plant of the genus Vitis in the family Vitaceae, which can be divided into fresh food varieties, wine varieties and dried fruit varieties according to its use. The planting methods and harvesting quality requirements of wine varieties and fresh food varieties are different. Wine varieties have a large planting area and grow densely between each cluster of fruits. After ripening, they need to be harvested quickly, efficiently and at low cost. Current grape harvesting mainly relies on manual picking and mechanical picking. Manual picking requires one hand to hold the grape cluster and the other hand to cut the stem, thereby achieving complete and accurate picking of grapes. However, this picking method not only wastes labor resources and increases harvesting costs, but also increases the time and reduces the efficiency of harvesting, making it impossible to achieve the goal of quickly, efficiently and at low cost harvesting wine grapes. Mechanical picking mainly involves vibrating grape plants to make grape fruits fall off, thereby completing the picking and collecting of grapes. This method is prone to damage to grape fruits and is suitable for picking raw materials for dried grapes or grape juice. However, it is not suitable for wine grapes that need to be processed in whole clusters. In addition, with the development of agricultural intelligence, some robots in the prior art use machine vision to locate grapes and achieve automatic picking. However, existing picking robots have difficulties in identifying picking points, low success rates, and low harvesting efficiency. Moreover, most existing picking robots use multi-joint and multi-degree-of-freedom robotic arms, which not only have high manufacturing costs and are difficult to control, but also are prone to contact with grape fruits, vines and other objects during control, resulting in fruit damage, vine entanglement with the robotic arm, and other problems. This increases the rate of fruit damage and even causes the robotic arm to jam, affecting the normal operation of grape harvesting. SUMMARY

[0003] To address the problems in the prior art, the present application provides a three-degree-of-freedom grape harvesting robot. The robot adopts a three-degree-of-freedom Cartesian coordinate design, which simplifies the mechanical structure, avoids disturbing the surrounding fruits or being disturbed by grape vines during harvesting, effectively reduces the complexity and cost of the system, and ensures flexibility and accuracy during harvesting, thereby achieving quick, efficient and low-cost grape harvesting.

[0004] Another object of the present application is to provide a harvesting method for the above-mentioned grape harvesting robot to achieve the harvesting of wine grapes.

[0005] The object of the present application is achieved by the following technical solutions:

[0006] A three-degree-of-freedom grape picking robot comprises a chassis, a walking mechanism, a rotating plate, a three-coordinate mechanism, an end execution mechanism, a vision assembly, a navigation system and an electric control box, the walking mechanism is arranged on the bottom surface of the chassis and adopts a track type walking assembly for controlling the operation of the whole picking robot, the rotating plate is arranged on the end surface of the chassis and the three-coordinate mechanism is arranged on the end surface of the rotating plate; the three-coordinate mechanism comprises a Y-axis assembly, an X-axis assembly and a Z-axis assembly, the Z-axis assembly is arranged on the end surface of the rotating plate, the Y-axis assembly is arranged on the Z-axis assembly, and the X-axis assembly is arranged between the Y-axis assemblies; the end execution mechanism is arranged on one end of the X-axis assembly away from the Y-axis assembly, the vision assembly is arranged on one end of the X-axis assembly corresponding to the end execution mechanism and located on the upper side of the end execution mechanism; the navigation system is located on the side surface of the chassis and the electric control box is arranged on the end surface of the chassis, the electric control box is located away from the three-coordinate mechanism and is electrically connected with the control devices of the walking mechanism and the rotating plate, the three-coordinate mechanism, the end execution mechanism, the navigation system and the vision assembly.

[0007] Further optimization based on the above scheme, the Z-axis assembly comprises two groups of horizontal guide rails, two horizontal lead screws, two horizontal sliding blocks and two horizontal sliding seats, the two groups of horizontal guide rails are arranged in parallel on both sides of the end surface of the rotating plate, and one horizontal lead screw is arranged on each side surface of the two groups of horizontal guide rails away from each other, the horizontal lead screws are threadedly sleeved with the horizontal sliding blocks on the outer walls respectively, the horizontal sliding seats are slidably arranged on the horizontal guide rails, and the horizontal sliding seats are fixedly connected with the corresponding horizontal sliding blocks; the Y-axis assembly comprises a connecting cross bar, two groups of vertical guide rails, two vertical lead screws, two vertical sliding blocks and two vertical sliding seats, the bottom surface of the connecting cross bar is fixedly connected with the two horizontal sliding blocks and the end surfaces of the two horizontal sliding seats respectively, and two vertical guide rails are fixedly arranged on the two sides of the end surface of the connecting cross bar in parallel, one vertical lead screw is arranged on each side surface of the two groups of vertical guide rails away from each other, the vertical lead screws are threadedly sleeved with the vertical sliding blocks on the outer walls respectively, the vertical sliding seats are slidably arranged on the outer walls of the vertical guide rails, and the vertical sliding seats are fixedly connected with the corresponding vertical sliding blocks; the X-axis assembly comprises a positioning cross plate, a moving lead screw and a linkage support, the two ends of the positioning cross plate are fixedly connected with the inner side walls of the corresponding vertical sliding seats respectively, and one moving lead screw is arranged on one side surface of the positioning cross plate, the linkage support is perpendicular to the positioning cross plate, and the bottom surface of the linkage support is threadedly connected with the moving lead screw and slidably connected with the positioning cross plate.

[0008] Based on the further optimization of the above scheme, the end effector comprises two servo motors, two adjusting ropes, a support disc, a connecting hose, a universal ball cage, a high-speed motor and a three-edge tool, the end surface of the end of the linkage support away from the positioning horizontal plate is provided with the support disc through the connecting hose, the connecting hose is multiple and uniformly distributed on the outer circle of the end surface of the support disc; the end surface of the end of the linkage support away from the positioning horizontal plate is fixedly provided with the two servo motors, and the output ends of the two servo motors are respectively wound with one adjusting rope, the end of the adjusting rope away from the servo motor passes through the fixed pulley on the end surface of the linkage support, penetrates through the linkage support and is connected with the end surface of the support disc, the adjusting ropes are symmetrically distributed on the two sides of the support disc and are heterotopicly arranged with the connecting hose; the universal ball cage coaxial with the support disc is arranged between the linkage support and the support disc and in the inner circle of the connecting hose; the high-speed motor is fixedly arranged on the bottom surface of the support disc, and the output end of the high-speed motor is fixedly sleeved with the three-edge tool through the shaft coupling, the main body of the three-edge tool is a disc-shaped structure, three main cutting surfaces (i.e. three notches are uniformly arranged on the outer circle of the disc-shaped structure to divide the disc-shaped outer circle into three cutting surfaces) extend from the base, and each cutting surface is provided with a bionic corner groove similar to dinosaur teeth.

[0009] Based on the further optimization of the above scheme, arc-shaped grooves are arranged in the inner circle of the three-edge tool corresponding to the three cutting surfaces.

[0010] Based on the further optimization of the above scheme, the visual assembly comprises a camera support and a visual camera, and the visual camera is fixedly arranged on the end surface of the linkage support through the camera support.

[0011] Based on the further optimization of the above scheme, the navigation system comprises two groups of antenna supports and two groups of Beidou antennas, the two groups of antenna supports are arranged at the front and rear ends of the chassis respectively, and the Beidou antennas are fixedly arranged on the antenna supports.

[0012] A harvesting method of a three-degree-of-freedom grape harvesting robot, comprising:

[0013] Step one, the harvesting robot moves to the area to be harvested, starts the visual camera to capture images and videos, and completes target recognition;

[0014] Step two, the navigation path is extracted in real time through the Beidou antenna, and the harvesting robot is started to move to the target recognized in step one according to the navigation path; during navigation, the start-stop and steering control of the harvesting robot are realized through the recognition of inter-row obstacles;

[0015] Step three, the three-coordinate mechanism and the end effector are started to cut and harvest the target in real time.

[0016] Based on the further optimization of the above scheme, the improved YOLOv10 model is used to realize target recognition in step one, specifically:

[0017] First, an image dataset D={Ii|i=1,2,…,N} containing grape main vines, fruiting branches and fruit stalks is obtained, I 1 ,I 2 ,…,I N}, I i (i=1,2,…,N) represents a single image; and the image dataset is divided into a training set, a validation set and a test set;

[0018] Then, a bounding box and a class label are added to each cluster of fruit using a labeling tool;

[0019] After that, an improved YOLOv10 model is constructed, which includes the deep learning network structures of Backbone, Neck and Head. The network model starts from image input, extracts features through the Backbone part, performs feature fusion through the Neck part, and finally performs regression and classification prediction through the Head part. Therefore, in the improved YOLOv10 model, the FasterBlock module is used to replace the Bottleneck module in C2f of the Backbone part, so as to optimize the calculation efficiency and improve the accuracy of small target detection:

[0020] ;

[0021] In the formula: H, W represents the size of the feature map, k represents convolution and size, C p represents the number of calculated channels;

[0022] In order to meet the detection needs of grape main vines and fruiting branches and other different scale targets, the LSKA attention mechanism is used to replace the SPPF module of the Backbone part. LSKA reduces the computational complexity by decomposing the two-dimensional convolution kernel into horizontal and vertical 1D convolution kernels, and improves the long-distance dependence characteristics and multi-scale adaptability of the model. The output of LSKA in each channel is Z c

[0023] ;

[0024] In the formula: W kx1 and W 1xk respectively represent the horizontal and vertical convolution kernels; F c represents the convolution kernel weight generated based on content;

[0025] ​To improve the detection accuracy of small targets such as grape peduncles, the CARAFE module is introduced in the Neck part, and the up-sampling operation is optimized. For a given target position , the corresponding up-sampling feature is :

[0026] ;

[0027] In the formula: , represents the convolution kernel weight generated based on the content, r , represents the neighborhood area radius; , represents the element located on the output feature map of ( i + n, j + m ) after the convolution operation;

[0028] Finally, the improved YOLOv10 model is used to train the labeled training set. After training, the model performance is evaluated using the validation set. The improved YOLOv10 model is deployed to the grape harvesting robot, achieving the purpose of quickly and accurately detecting grape main vines, fruiting branches, and peduncles.

[0029] Based on the further optimization of the above scheme, the specific method for extracting the navigation path in step two is:

[0030] First, the yolov8-seg model is used to identify the grape inter-row road and extract the left and right edge points of the road (x x l ,L l ), (x x r ,L r ), and the left and right grape tree lines are fitted by the least squares method:

[0031] ;

[0032] In the formula: , represents the slope and intercept of the left straight line; , represents the slope and intercept of the right straight line;

[0033] Then, according to the left and right grape tree lines, the reference points of the navigation line are extracted P i : (x x i ,y i ), i = 1, 2, …, n :

[0034] ;

[0035] In the formula: xli ,L li ） denotes the coordinate of any point on the left row line, x ri ,L ri ） denotes the coordinate of the corresponding point on the right row line of the same horizontal line, x li ,L li ） denotes the coordinate of the corresponding point on the right row line of the same horizontal line; β denotes the position parameter of the navigation line;

[0036] Setting the position parameter threshold If , it is considered that the navigation line is in the center of the grape row; if , the navigation line is in the left row, and if , the navigation line is in the right row, and through the control of the position parameter, the harvesting robot moves to the left or right side, thereby realizing the harvesting of the left or right side of the grape.

[0037] Based on the further optimization of the above scheme, the specific method for starting and stopping and steering control of the harvesting robot by identifying the inter-row obstacles in step two is:

[0038] First, the least square method is used to fit the orchard navigation line:

[0039] ;

[0040] Then, m points on the orchard navigation line are selected at equal intervals, that is, from the lower side to the upper side of the image, the orchard navigation line is intercepted by m horizontal lines in turn with a width of H , and the m points are denoted as m : ; );

[0041] After that, the left boundary point slj : P ) and the right boundary point srj : P ) of the safe driving of the harvesting robot are obtained respectively based on the reference: :

[0042] ;

[0043] In the formula: denotes the ratio between the actual width of the road and the actual width of the harvesting robot; denotesRoI The width of the inner lowest end road in the pixel coordinate system;

[0044] Finally, the obstacles on the road during the travel are identified by the yolov8-seg model to obtain the bounding rectangle of the obstacles in the RoI inner; and the leftmost lower vertex coordinates P obl x obl ,y obl and the rightmost lower vertex coordinates P obr x obr ,y obr , if the conditions of or are met, it is determined that the obstacle is in the safe driving area of the harvesting robot, and the robot is started to turn left or right to avoid the obstacle; otherwise, the robot continues to move forward.

[0045] The technical effects of the present application are as follows:

[0046] The present application effectively simplifies the mechanical arm structure through the structural design of the three-coordinate mechanism of the Y-axis assembly, the X-axis assembly and the Z-axis assembly, avoids the entanglement of the multi-degree-of-freedom mechanical arm structure in the picking process or the damage, falling, etc. of the surrounding grape clusters caused by the entanglement of the grapevine, so as to avoid the problems of the whole harvesting robot being stuck, reducing the picking efficiency, etc. The end execution mechanism composed of the servo motor, the adjusting rope, the supporting disc, the connecting hose, the universal ball cage, the high-speed motor and the three-blade cutter not only utilizes the cooperation of the adjusting rope, the connecting hose and the universal ball cage to improve the flexibility and adaptability of the cutter, ensures the efficient operation of the harvesting robot in the complex field environment, reduces the damage of the end execution mechanism to the fruits and the vine, but also further improves the fault tolerance of cutting by cooperating with the setting of the three-blade cutter, ensures accurate and effective cutting when the fruit axis attitude angle or spatial surface error occurs in the cutting process, in addition, it is also helpful to the discharge of the cutting process, avoids the cutting efficiency reduction caused by the cutting process of the three-blade cutter caused by the cutting process of the three-blade cutter, and further improves the harvesting efficiency.

[0047] ​​Meanwhile, the improved YOLOv10 model is used for grape recognition, the detection accuracy of grape main vines, fruiting branches and fruit stalks is improved significantly, the grape picking robot can quickly and accurately identify and locate the grapes, the accuracy and efficiency of harvesting are improved, and the missed picking and mispicking are reduced; through the combination of the visual component and the navigation system, a composite navigation mode is formed, so that the navigation accuracy and stability of the picking robot in the complex field environment are effectively improved, and the picking robot can effectively avoid obstacles and complete accurate picking positioning, and the automation level and operation safety in the grape harvesting process are improved. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 It is a whole structure schematic diagram of the picking robot in the embodiment of the application.

[0049] Figure 2 It is a structure schematic diagram of the Z-axis assembly of the picking robot in the embodiment of the application.

[0050] Figure 3 It is a structure schematic diagram of the Y-axis assembly and the X-axis assembly of the picking robot in the embodiment of the application.

[0051] Figure 4 It is a structure schematic diagram of the X-axis assembly and the end execution mechanism of the picking robot in the embodiment of the application.

[0052] Figure 5 It is a structure schematic diagram of the three-edge blade cutter of the picking robot in the embodiment of the application.

[0053] Figure 6 It is a network structure diagram of the improved YOLOv10 in the embodiment of the application.

[0054] Figure 7 It is a flow chart of extracting a navigation path in the embodiment of the application.

[0055] Figure 8 It is an analysis schematic diagram of the identification of inter-row obstacles in the embodiment of the application.

[0056] Figure 9 It is a fruit stalk low collision ROI schematic diagram in the embodiment of the application.

[0057] Wherein, 10, chassis; 20, walking mechanism; 30, rotating plate; 411, horizontal guide rail; 412, horizontal screw; 413, horizontal slider; 414, horizontal slide; 421, connecting cross bar; 422, vertical guide rail; 423, vertical screw; 424, vertical slider; 425, vertical slide; 431, positioning cross plate; 432, moving screw; 433, linkage bracket; 4330, fixed pulley; 51, servo motor; 52, adjusting rope; 53, supporting disc; 54, connecting hose; 55, universal ball cage; 56, high-speed motor; 57, three-flank cutter; 61, camera bracket; 62, visual camera; 71, antenna bracket; 72, Beidou antenna. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0059] Embodiment 1:

[0060] A three-degree-of-freedom grape harvesting robot, comprising a chassis 10, a walking mechanism 20, a rotating plate 30, a three-coordinate mechanism, an end execution mechanism, a visual assembly, a navigation system and an electric control box, the bottom surface of the chassis 10 is provided with the walking mechanism 20, and the walking mechanism 20 adopts a track type walking assembly (as shown in Figure 1 ), which is used for controlling the operation of the whole harvesting robot, the end surface of the chassis 10 is rotationally provided with the rotating plate 30, and the end surface of the rotating plate 30 is provided with the three-coordinate mechanism (as shown in Figure 1 ); the three-coordinate mechanism comprises a Y-axis assembly, an X-axis assembly and a Z-axis assembly, the Z-axis assembly is arranged on the end surface of the rotating plate, the Y-axis assembly is arranged on the Z-axis assembly, and the X-axis assembly is arranged between the Y-axis assemblies, specifically: the Z-axis assembly comprises two groups of horizontal guide rails 411, two horizontal screws 412, two horizontal sliders 413 and two horizontal slides 414, the two groups of horizontal guide rails 411 are arranged in parallel on both sides of the end surface of the rotating plate 30 (as shown in Figure 2 ), and one rotating horizontal screw 412 (as shown in Figure 2As shown, the horizontal screw rod 412 is arranged on the end surface of the rotating plate 30 on the side of the horizontal guide rail 411 away from the other side through a rotating support, and one end of the horizontal screw rod 412 is provided with a motor for controlling the rotation of the horizontal screw rod 412; the outer wall of the horizontal screw rod 412 is threadedly sleeved with the horizontal sliding block 413; the horizontal guide rail 411 is slidably provided with the horizontal sliding seat 414, and the horizontal sliding seat 414 is fixedly connected with the corresponding horizontal sliding block 413 (so as to drive the horizontal sliding seat 414 to slide on the horizontal guide rail 411 through the horizontal sliding block 413); the Y-axis assembly comprises the connecting cross rod 421, two groups of vertical guide rails 422, two vertical screw rods 423, two vertical sliding blocks 424 and two vertical sliding seats 425; the bottom surface of the connecting cross rod 421 is fixedly connected with the end surfaces of the two horizontal sliding blocks 413 and the two horizontal sliding seats 414, and the end surface of the connecting cross rod 421 is fixedly provided with two vertical guide rails 422 parallel to each other (as shown in Figure 3 As shown, the two groups of vertical guide rails 422 are arranged on the side walls away from each other, and each is provided with a rotating vertical screw rod 423 (as shown in Figure 3 As shown, the two ends of the vertical screw rod 423 are arranged on the side walls away from each other through rotating supports, and the end surface of the connecting cross rod 421 at the bottom of the vertical screw rod 423 is provided with a motor for controlling the rotation of the vertical screw rod 423; the outer wall of the vertical screw rod 523 is threadedly sleeved with the vertical sliding block 424; the outer wall of the vertical guide rail 422 is slidably provided with the vertical sliding seat 425, and the vertical sliding seat 425 is fixedly connected with the corresponding vertical sliding block 424 (so as to drive the vertical sliding seat 425 to ascend and descend on the vertical guide rail 422 through the vertical sliding block 424); the X-axis assembly comprises the positioning cross plate 431, the moving screw rod 432 and the linkage support 433; the two ends of the positioning cross plate 431 are fixedly connected with the inner side walls of the corresponding vertical sliding seats 425 (as shown in Figure 3 As shown, the side of the positioning cross plate 431 is provided with a rotating moving screw rod 432 (as shown in Figure 3 As shown, the two ends of the moving screw rod 432 are arranged on the back surface of the positioning cross plate 431 through rotating supports, and one end of the moving screw rod 432 is provided with a motor for controlling the rotation of the moving screw rod 432; the linkage support 433 is perpendicular to the positioning cross plate 431, and the bottom surface of the linkage support 433 is threadedly connected with the moving screw rod 432 and slidably connected with the positioning cross plate 431.

[0061] The end effector is arranged on the end of the X-axis assembly away from the Y-axis assembly (specifically, the end of the linkage support 433 away from the positioning cross plate 431, as shown in Figure 4 As shown, the end effector comprises two servo motors 51, two adjusting ropes 52, a support disc 53, a connecting hose 54, a universal ball cage 55, a high-speed motor 56 and a three-blade cutter 57; the end of the linkage support 433 away from the positioning cross plate 431 is provided with the support disc 53 through the connecting hose 54; the connecting hose 54 is a plurality of and uniformly distributed on the outer circle of the end surface of the support disc 53 (the number of the connecting hose 54 is generally 3-8, which is arranged according to the actual situation);Figure 4 As shown, four connecting hoses 54 are used in this embodiment; two servo motors 51 are fixedly arranged at the end face of the linkage bracket 433 away from the positioning horizontal plate 431, and the output ends of the two servo motors 51 are respectively wound with an adjusting rope 52. The end of the adjusting rope 52 away from the servo motor 51 passes through the fixed pulley 4330 arranged at the end face of the linkage bracket 433, penetrates through the linkage bracket 433, and is connected with the end face of the support disc 53. The adjusting ropes 52 are symmetrically arranged on both sides of the support disc 53, and the adjusting ropes 53 are arranged at different positions from the connecting hoses 54. The universal ball cage 55 coaxial with the support disc 53 is arranged between the linkage bracket 433 and the support disc 53 and in the inner circle of the connecting hose 54. The high-speed motor 56 is fixedly arranged at the bottom surface of the support disc 53, and the output end of the high-speed motor 56 is fixedly connected with the three-face blade cutter 57 through a shaft coupling. The three-face blade cutter 57 has a disc-shaped structure (as shown in Figure 5 ), and three main cutting faces are extended from the base (i.e., three notches are uniformly arranged at the outer circle of the disc-shaped structure to divide the disc-shaped outer circle into three cutting faces, as shown in Figure 5 ). Each cutting face is provided with a bionic edge groove similar to dinosaur teeth (as shown in the partial enlarged view of Figure 5 ), and arc-shaped grooves are arranged at the inner circle of the three-face blade cutter 57 corresponding to the three cutting faces. At the same time, in order to reduce the weight of the three-face blade cutter 57 while ensuring its cutting performance, the three-face blade cutter 57 is made of three different materials. Specifically, high-hardness and high-wear-resistance materials such as high-speed steel are used for the cutting edge part, light materials such as alloy steel or aluminum alloy are used for the cutter main body part, and low-density materials such as carbon steel or composite materials are used for the cutter internal part.

[0062] The visual assembly is arranged at one end of the X-axis assembly corresponding to the end effector and located at the upper side of the end effector. The visual assembly includes a camera bracket 61 and a visual camera 62, and the visual camera 62 is fixedly arranged at the end face of the linkage bracket 433 through the camera bracket 61. The navigation system is located at the side of the chassis 10, and the navigation system includes two groups of antenna brackets 71 and two groups of Beidou antennas 72. The two groups of antenna brackets 71 are arranged at the front and rear ends of the chassis 10, and the Beidou antennas 72 are fixedly arranged on the antenna brackets 71 (as shown in Figure 1 ). The electric control box is arranged at the end face of the chassis 10, and the electric control box is arranged at different positions from the three-coordinate mechanism and is electrically connected with the control devices of the walking mechanism 20 and the rotating plate 30, the three-coordinate mechanism, the end effector, the navigation system, and the visual assembly.

[0063] Embodiment 2:

[0064] As a preferred embodiment of the present application, on the basis of the scheme of embodiment 1, in order to realize the protection of the three-edged blade cutter 57 and avoid the damage of the three-edged blade cutter 57 to fruits and vines during movement, a protective shell is arranged at the bottom of the support disc 53 and outside the outer ring of the high-speed motor 56, a cutting shell with an open front end is arranged at the bottom of the protective shell corresponding to the three-edged blade cutter 57, and the three-edged blade cutter 57 is located in the cutting shell and protrudes from the opening of the cutting shell.

[0065] Embodiment 3:

[0066] As a preferred embodiment of the present application, on the basis of the scheme of embodiment 1, a collecting frame (which can be a net collecting device or a hard collecting device, and can be arranged according to actual conditions) is arranged away from the positioning horizontal plate 431 and below the end effector of the linkage support 433, for collecting grape clusters cut by the three-edged blade cutter 57.

[0067] Embodiment 4:

[0068] A harvesting method of a three-degree-of-freedom grape harvesting robot, which adopts any one of the harvesting robots in embodiments 1-3, comprises the following steps:

[0069] Step 1: The harvesting robot moves to the area to be harvested, starts the vision camera to capture images and videos, and completes target recognition; the improved YOLOv10 model is used to realize target recognition, specifically as follows:

[0070] Firstly, an image data set D={Ii|i=1,2,…,N} containing grape main vines, fruiting branches and fruit stalks is obtained, and the image data set is divided into a training set, a verification set and a test set (in this embodiment, the proportions of the training set, the verification set and the test set are 8:1:1); I 1 ,I 2 ,…,I N}, I i (i=1,2,…,N) represents a single image; and the image data set is divided into a training set, a verification set and a test set (in this embodiment, the proportions of the training set, the verification set and the test set are 8:1:1);

[0071] Then, a bounding box and a class label are added to each cluster of fruits by using a labeling tool;

[0072] After that, an improved YOLOv10 model is constructed, which refers to Figure 6As shown: the YOLOv10 network model includes the deep learning network structure of Backbone, Neck and Head; the network model starts from image input, extracts features through the Backbone part, performs feature fusion through the Neck part, and finally performs regression and classification prediction through the Head part; therefore, in the improved YOLOv10 model, the FasterBlock module is used to replace the Bottleneck module in the C2f of the Backbone part, so as to optimize the calculation efficiency and improve the accuracy of small target detection:

[0073] ;

[0074] In the formula: H, W represents the size of the feature map, k represents convolution and size, C p represents the number of channels calculated;

[0075] In order to meet the detection needs of different scale targets such as grape main vines and fruiting branches, the LSKA attention mechanism is used to replace the SPPF module of the Backbone part, and the LSKA reduces the calculation complexity by decomposing the two-dimensional convolution kernel into horizontal and vertical 1D convolution kernels, and improves the long-distance dependence characteristics and multi-scale adaptability of the model. The output of LSKA in each channel is Z c :

[0076] ;

[0077] In the formula: W kx1 and W 1xk respectively represent the horizontal and vertical convolution kernels; F c represents the convolution kernel weight generated based on content;

[0078] In order to improve the detection accuracy of small targets such as grape peduncles, the CARAFE module is introduced in the Neck part to optimize the upsampling operation. For a given target position , the corresponding up-sampling feature is :

[0079] ;

[0080] In the formula: represents the convolution kernel weight generated based on content, r represents the radius of the neighborhood region; represents the element located on the output feature map of i + n, j + m after convolution operation.

[0081] Finally, the improved YOLOv10 model is trained on the labeled training set, and after training, the validation set is used to evaluate the performance of the model. The improved YOLOv10 model is deployed to the grape harvesting robot, so as to quickly and accurately detect the main vine, fruiting branch and fruit stem of the grape.

[0082] Step two, real-time extraction of navigation path through Beidou antenna, and start the harvesting robot to move towards the target identified in step one according to the navigation path, as shown in Figure 7 The specific method for extracting the navigation path is as follows:

[0083] First, the yolov8-seg model is used to identify the grape inter-row road, and the left and right edge points of the road are extracted x l ,L l x r ,L r The left and right side grape tree lines are fitted by the least square method:

[0084] ;

[0085] In the formula: represents the slope and intercept of the left straight line; represents the slope and intercept of the right straight line;

[0086] Then, according to the left and right side grape tree lines, the reference points of the navigation line are extracted P i x i ,y i i = 1, 2, …, n :

[0087] ;

[0088] In the formula: x li ,L li ） represents the coordinates of any point on the left side line, and x ri ,L ri ） represents the coordinates of any point on the right side line x li ,L li ） ​​​The coordinates of the corresponding point on the right side of the same horizontal line; β Indicates the position parameters of the navigation line;

[0089] Set position parameter threshold (Location parameter threshold) The settings are configured according to the actual situation; in this embodiment... ),like If , then the navigation line is considered to be located in the center of the grape row; if If the navigation line is on the left row, then... If the navigation line is on the right side, the harvesting robot can move to the left or right by controlling the position parameters, thereby achieving grape harvesting on the left or right side.

[0090] During navigation, the start, stop, and steering of the harvesting robot are controlled by identifying obstacles between rows. The specific method is as follows:

[0091] First, the orchard navigation line is fitted using the least squares method:

[0092] ;

[0093] Then, select m points at equal intervals along the orchard navigation line, that is, from the bottom of the image to the top, with... H The width is determined by sequentially cutting through m horizontal lines along the selected orchard navigation line. m Each point is recorded as :( );

[0094] Then, refer to Figure 8 As shown, with Based on this, the left boundary points for the safe operation of the harvesting robot are obtained respectively. P slj :( ) and the right boundary point P srj :( ):

[0095] ;

[0096] In the formula: This represents the ratio between the actual width of the road and the actual width of the harvesting robot. express RoI The width of the lowest inner road in pixel coordinates;

[0097] Finally, the YOLOv8-SEG model was used to identify obstacles on the road during the journey, and the location of the obstacles was obtained. RoI The inner bounding rectangle (e.g.) Figure 8 As shown, the distance from the bottom edge of the circumscribed rectangle to the bottom edge of the image is...H s ) and the left-bottom vertex coordinate of the circumscribed rectangle P obl x obl ,y obl ) and the right-bottom vertex coordinate of the circumscribed rectangle P obr x obr ,y obr ), if the following condition is satisfied or , it is determined that the obstacle is in the safe driving area of the harvesting robot, and the robot is started to turn left or right to avoid the obstacle; otherwise, the robot continues to move forward.

[0098] Step three, start the three-coordinate mechanism and the end effector to cut and harvest the target in real time;

[0099] The picking point acquisition method is specifically as follows:

[0100] For the target grape stem and fruiting branch, a low-collision ROI of the grape stem is constructed (as shown in Figure 9 ); wherein the rectangular frame width is used to limit the horizontal range of the low-collision ROI of the grape stem; the fruiting branch takes the lowest point in the current horizontal range as the upper limit of the vertical direction of the ROI, and the highest point of the grape cluster rectangular frame as the lower limit of the vertical direction of the ROI (as shown in Figure 9 ); the grape stem ROI region is specifically as follows:

[0101]

[0102] In the formula: Wid_roi represents the horizontal range limit of the ROI region; Hei_roi represents the horizontal range limit of the ROI region; P_Clutopl.x represents the x value of the left-top vertex of the grape cluster detection frame, P_Clutopl.y represents the y value of the left-top vertex of the grape cluster detection frame, P_Clutopr.x represents the x value of the right-top vertex of the grape cluster detection frame, P_Stembot.y represents the y value of the lowest point of the current horizontal range by the fruiting branch;

[0103] If the current grape cluster has an ROI, the circumscribed rectangle of the retained grape stem part of the grape stem ROI is obtained, and the centroid of the circumscribed rectangle is taken as the picking point of the three-blade cutter; if the current grape ROI does not exist, it means that the current grape cannot be picked.​​​

Claims

1. A three-degree-of-freedom grape harvesting robot, characterized in that: The system includes a chassis, a walking mechanism, a rotating platform, a coordinate measuring machine (CMM), an end effector, a vision component, a navigation system, and an electrical control box. The walking mechanism, which is a tracked walking mechanism, is mounted on the bottom surface of the chassis. A rotating platform is mounted on the end face of the chassis, and a CMM is mounted on the end face of the rotating platform. The CMM includes a Y-axis assembly, an X-axis assembly, and a Z-axis assembly. The Z-axis assembly is located on the end face of the rotating platform, the Y-axis assembly is mounted on the Z-axis assembly, and the X-axis assembly is located between the Y-axis assemblies. The end effector is located at the end of the X-axis assembly away from the Y-axis assembly. The vision component is located at the end of the X-axis assembly corresponding to the end effector and is positioned above the end effector. The navigation system is located on the side of the chassis, and the electrical control box is located on the end face of the chassis. The electrical control box is offset from the CMM and is electrically connected to the control devices of the walking mechanism, the rotating platform, the CMM, the end effector, the navigation system, and the vision component. The end effector includes two servo motors, two adjusting ropes, a support plate, a connecting hose, a universal ball cage, a high-speed motor, and a three-sided cutting tool. The support plate is mounted on the bottom surface of the end of the linkage bracket away from the positioning plate via a connecting hose. Multiple connecting hoses are evenly distributed around the outer ring of the support plate's end face. Two servo motors are fixedly mounted on the end face of the linkage bracket away from the positioning plate, and each servo motor's output end is wound with an adjusting rope. The end of the adjusting rope away from the servo motor passes through a fixed pulley on the linkage bracket's end face, penetrates the linkage bracket, and connects to the support plate's end face. The adjusting ropes are symmetrically distributed on both sides of the support plate, and are positioned opposite to the connecting hoses. A universal ball cage, coaxial with the support plate, is located between the linkage bracket and the support plate, within the inner ring of the connecting hose. A high-speed motor is fixedly mounted on the bottom surface of the support plate, and the output end of the high-speed motor is fixedly sleeved with a three-sided cutting tool via a coupling. The three-sided cutting tool has a disc-shaped structure, extending from its base with three main cutting surfaces, each with biomimetic angular grooves resembling dinosaur teeth.

2. The three-degree-of-freedom grape harvesting robot according to claim 1, characterized in that: The Z-axis assembly includes two sets of horizontal guide rails, two horizontal lead screws, two horizontal sliders, and two horizontal slide blocks. The two sets of horizontal guide rails are arranged parallel to each other on both sides of the rotating plate end face, with a rotating horizontal lead screw mounted on each side of the two sets of horizontal guide rails that are close to each other. Horizontal sliders are threaded onto the outer walls of the horizontal lead screws. Horizontal slide blocks are slidably mounted on the horizontal guide rails, and the horizontal slide blocks are fixedly connected to their corresponding horizontal sliders. The Y-axis assembly includes a connecting crossbar, two sets of vertical guide rails, two vertical lead screws, two vertical sliders, and two vertical slide blocks. The bottom surface of the connecting crossbar is fixedly connected to the end faces of the two horizontal sliders and the two horizontal slide blocks, respectively. Two parallel vertical guide rails are fixedly installed on both sides of the rod end face. A rotating vertical screw is installed on the side of the two sets of vertical guide rails that are far apart from each other, and a vertical slider is threaded onto the outer wall of the vertical screw. A vertical slide block is slidably installed on the outer wall of the vertical guide rail, and the vertical slide block is fixedly connected to the corresponding vertical slider. The X-axis assembly includes a positioning horizontal plate, a moving screw, and a linkage bracket. The two ends of the positioning horizontal plate are fixedly connected to the inner side wall of the corresponding vertical slide block, and a rotating moving screw is installed on one side of the positioning horizontal plate. The linkage bracket is perpendicular to the positioning horizontal plate, and the bottom surface of the linkage bracket is threadedly connected to the moving screw and slidably connected to the positioning horizontal plate.

3. The three-degree-of-freedom grape harvesting robot according to claim 2, characterized in that: The vision component includes a camera bracket and a vision camera, with the vision camera fixedly mounted on the end face of the linkage bracket via the camera bracket.

4. A three-degree-of-freedom grape harvesting robot according to claim 2, characterized in that: The navigation system includes two sets of antenna brackets and two sets of BeiDou antennas. The two sets of antenna brackets are respectively set at the front and rear ends of the chassis, and the BeiDou antennas are fixedly installed on the antenna brackets.

5. The harvesting method of a three-degree-of-freedom grape harvesting robot according to claim 2, characterized in that: include: Step 1: The harvesting robot moves to the harvesting area, activates the vision camera to capture images and videos, and completes target recognition; Step 2: Extract the navigation path in real time using the Beidou antenna, and start the harvesting robot to move towards the target identified in Step 1 according to the navigation path; during the navigation process, the start, stop and turn control of the harvesting robot are realized by identifying obstacles between rows; Step 3: Activate the coordinate measuring machine and the end effector to cut and harvest the target in real time.

6. The harvesting method of a three-degree-of-freedom grape harvesting robot according to claim 5, characterized in that: In step one, an improved YOLOv10 model is used to achieve target recognition, specifically as follows: First, obtain an image dataset D containing the main grapevine, fruiting branches, and fruit stalks. I 1 ,I 2 ,…,I N }, I i (i=1,2,…,N) represents a single image; and the image dataset is divided into training set, validation set and test set; Then, a labeling tool was used to add bounding boxes and category labels to each bunch of fruit; Next, an improved YOLOv10 model was constructed. The YOLOv10 network model includes a deep learning network structure consisting of a Backbone, Neck, and Head. This network model starts with image input, extracts features through the Backbone, performs feature fusion through the Neck, and finally performs regression and classification prediction through the Head. Therefore, in the improved YOLOv10 model, the FasterBlock module replaces the Bottleneck module in the C2f function of the Backbone to optimize computational efficiency and improve the accuracy of small object detection. ； In the formula: H, W Indicates the size of the feature map. k Indicates convolution and size. C p Indicates the number of channels calculated; The SPPF module in the backbone section is replaced with an LSKA attention mechanism. LSKA is applied to the output of each channel. Z c for: ； In the formula: W kx1 and W 1xk These represent the convolution kernels in the horizontal and vertical directions, respectively. F c This represents the convolutional kernel weights generated based on the content. The CARAFE module is introduced into the Neck section, and the upsampling operation is optimized for a given target position. Its corresponding upsampling feature is : ； In the formula: This represents the convolutional kernel weights generated based on the content. r Indicates the radius of the neighborhood region; This indicates that after the convolution operation, the value located at ( i+n,j+m Elements on the output feature map; Finally, the improved YOLOv10 model was trained on the labeled training set. After training, the model performance was evaluated using the validation set. The trained improved YOLOv10 model was then deployed to the grape harvesting robot to achieve the goal of quickly and accurately detecting the main grape vine, fruiting branches, and fruit stalks.

7. The harvesting method of a three-degree-of-freedom grape harvesting robot according to claim 5, characterized in that: The specific method for extracting the navigation path in step two is as follows: First, the YOLOv8-SEG model is used to identify the paths between grape rows and to extract the left and right edge points of the paths. x l , L l ), ( x r ,L r The left and right grapevine row lines were fitted using the least squares method: ； In the formula: This represents the slope and intercept of the line on the left. This represents the slope and intercept of the line on the right. Then, based on the rows of grapevines on the left and right, extract the reference points for the navigation line. P i :( x i ,y i ), i=1,2,…,n : ； In the formula: ( x li ,L li ） Represents the coordinates of any point on the left-hand line, ( x ri ,L ri ） Indicates and ( x li ,L li ） The coordinates of the corresponding point on the right side of the same horizontal line; β Indicates the position parameters of the navigation line; Set position parameter threshold ,like If , then the navigation line is considered to be located in the center of the grape row; if If the navigation line is on the left row, then... If the navigation line is on the right side, the harvesting robot can move to the left or right by controlling the position parameters, thereby achieving grape harvesting on the left or right side.

8. The harvesting method of a three-degree-of-freedom grape harvesting robot according to claim 7, characterized in that: The specific method for controlling the start, stop, and turn of the harvesting robot by identifying obstacles between rows in step two is as follows: First, the orchard navigation line is fitted using the least squares method: ； Then, select m points at equal intervals along the orchard navigation line, that is, from the bottom of the image to the top, with... H The width is determined by sequentially cutting through m horizontal lines along the selected orchard navigation line. m Each point is recorded as :( ); After that, with Based on this, the left boundary points for the safe operation of the harvesting robot are obtained respectively. P slj :( ) and the right boundary point P srj :( ): ； In the formula: This represents the ratio between the actual width of the road and the actual width of the harvesting robot. express RoI The width of the lowest inner road in pixel coordinates; Finally, the YOLOv8-SEG model was used to identify obstacles on the road during the journey, and the location of the obstacles was obtained. RoI The bounding rectangle is obtained from the inner rectangle; and the coordinates of the leftmost bottom vertex of the bounding rectangle are obtained respectively. P obl :( x obl ,y obl ) and the coordinates of the rightmost bottom vertex P obr :( x obr ,y obr If satisfied or If the obstacle is determined to be within the safe driving area of ​​the harvesting robot, the robot will turn left or right to avoid the obstacle; otherwise, the robot will continue to move forward.

Images