Path planning method and system for plant protection robots based on UAV high-definition maps

Abstract

The present invention discloses a plant protection robot path planning method and system based on a high-definition map of an unmanned aerial vehicle (UAV), which belongs to the technical field of robot path planning. The method and system specifically include: using a UAV to perform flight mapping of an orchard, obtaining a map image of the orchard, preprocessing the obtained orchard map image, performing regional segmentation on the preprocessed orchard map image, constructing an orchard regional feature model, establishing a kinematic model according to the mechanical structure and motion characteristics of the plant protection robot, determining a set of feasible motion states of the robot in a two-dimensional plane, planning the path of the plant protection robot, obtaining the operation path of the plant protection robot, optimizing and smoothing the operation path of the plant protection robot, and locally adjusting the operation path of the plant protection robot according to the actual motion performance of the plant protection robot. The method can always maintain smooth operation in a complex and changeable orchard environment, will not stagnate or make mistakes due to environmental changes, and improves the level of automated operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot path planning, and in particular to a plant protection robot path planning method and system based on a high-definition map of an unmanned aerial vehicle. Background Art

[0002] In today's agricultural modernization, precision and efficiency in plant protection operations have become key demands. Traditional plant protection methods often rely on manual application of pesticides using backpack sprayers or simple mechanical plant protection equipment. This mode of operation has many drawbacks. On the one hand, manual operation is labor-intensive and inefficient, and prolonged exposure to pesticides poses serious health risks to humans. On the other hand, simple mechanical plant protection equipment lacks intelligent path planning and is typically limited to simple preset routes, unable to flexibly adjust to the actual conditions of the orchard. This can easily lead to uneven pesticide application, wasting pesticide resources and making it difficult to ensure effective pest control.

[0003] With the advancement of technology, plant protection robots have emerged, bringing new solutions to agricultural plant protection. However, existing plant protection robot path planning technology still has significant shortcomings. Some plant protection robots use path planning based on preset coordinate points. However, in actual orchard environments, they may encounter terrain changes, irregular crop layouts, temporary obstacles, new irrigation facilities, and crops that have fallen due to wind and rain. In these situations, they cannot adjust their paths in a timely and effective manner, resulting in operational interruptions or significantly reduced spraying effectiveness. Other plant protection robots rely on simple sensors (such as ultrasonic sensors and infrared sensors) for close-range obstacle detection to plan their paths. However, these sensors have limited detection range and incomplete environmental information, making them difficult to handle in complex and changing large-scale orchard scenarios and unable to fully utilize the advantages of plant protection robots.

[0004] Meanwhile, drone technology has experienced rapid development in recent years, enabling rapid and efficient acquisition of high-definition imagery across large orchards. Combining drones with plant protection robots and leveraging drone-derived HD maps to assist them in path planning has become a highly promising research direction. However, this integration is still in its exploratory stages. Exploiting the rich information contained in drone HD maps to design an accurate, efficient, and adaptable path planning method for plant protection robots in complex orchard environments remains a pressing technical challenge. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes a plant protection robot path planning method and system based on drone high-definition maps, which solves the problems of inaccurate plant protection robot path planning and poor adaptability in the existing technology, and improves the quality and efficiency of plant protection operations.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The path planning method for plant protection robots based on UAV high-definition maps includes:

[0008] Use UAV to conduct flight mapping of the orchard to obtain map images of the orchard, and pre-process the obtained map images of the orchard;

[0009] Perform regional segmentation on the pre-processed orchard map image and construct an orchard regional feature model;

[0010] Based on the mechanical structure and motion characteristics of the plant protection robot, a kinematic model is established to determine the set of feasible motion states of the robot in a two-dimensional plane;

[0011] Plan the path of the plant protection robot to obtain the running path of the plant protection robot;

[0012] The running path of the plant protection robot is optimized and smoothed, and the running path of the plant protection robot is partially adjusted according to the actual motion performance of the plant protection robot.

[0013] Specifically, the region segmentation of the pre-processed orchard map image includes:

[0014] The preprocessed orchard map image is set as I, with a size of H×W, where H represents the height of the preprocessed orchard map image and W represents the width of the preprocessed orchard map image. The improved U-Net neural network is used to establish a region segmentation model, and a loss function is set. The region segmentation model is trained using annotated orchard high-definition map images until the loss function converges and remains unchanged. The training is stopped to obtain a trained region segmentation model.

[0015] The pre-processed orchard map image I is input into the trained region segmentation model to obtain the region set R of the orchard map image segmentation, R={r1,r2,...,r n}, where r n Represents the segmented area of ​​the nth orchard map image.

[0016] Specifically, the orchard area characteristic model is constructed, including:

[0017] Extract the geometric features of the segmented regions of the orchard map image, including area, perimeter and shape complexity;

[0018] Extract the texture features of the segmented area of ​​the orchard map image, including: gray level co-occurrence matrix , contrast ,entropy and correlation , d represents the distance offset between pixel pairs, Indicates the direction angle;

[0019] The geometric features and texture features of the segmented area of ​​the orchard map image are extracted, as well as the geographic coordinate information, to construct the farmland area feature model M. , where A i represents the area of ​​the segmented region of the i-th orchard map image, P i represents the perimeter of the segmented region of the i-th orchard map image, C i represents the regional shape complexity of the i-th orchard map image segmentation, Represents the regional geographic coordinate information of the i-th orchard map image segmentation.

[0020] Specifically, the kinematic model is established based on the mechanical structure and motion characteristics of the plant protection robot to determine the set of feasible motion states of the robot in a two-dimensional plane, including:

[0021] According to the mechanical structure of the plant protection robot and the motion characteristics of the omnidirectional wheel, the linear velocity of the plant protection robot in the X direction is set to v x , the linear velocity in the Y direction is v y , the angular velocity is ;

[0022] In discrete time steps Next, a kinematic model is established, and the specific formula is:

[0023] ,

[0024] in, represents the position coordinates of the plant protection robot at time k+1, represents the position coordinates of the plant protection robot at time k, represents the heading angle of the robot at time k+1, represents the heading angle of the robot at time k, sin() represents the sine function, and cos() represents the cosine function;

[0025] Set the motion constraints of the plant protection robot, including: velocity in the X and Y directions, angular velocity, acceleration in the X and Y directions, and wheel slip constraints;

[0026] According to the kinematic model and constraints of the plant protection robot, the set of feasible motion states of the robot in a two-dimensional plane is determined.

[0027] Specifically, planning the path of the plant protection robot to obtain the running path of the plant protection robot includes:

[0028] Set the coordinates of the current node m to (xm ,y m ), the coordinates of the target node G are (x G ,y G ), establish the path function, the specific formula is:

[0029] ,

[0030] in, represents the path function, dl(m) represents the distance term, , cb(m) represents the motion cost of the plant protection robot, , represents the movement cost coefficient, v m represents the speed of the plant protection robot at node m, represents the angular velocity of the plant protection robot at node m, qb(m) represents the characteristic cost item of the orchard area, , 、 and They represent the operating efficiency cost coefficient, penalty cost coefficient and passage cost coefficient, penalty cost coefficient and passage cost coefficient respectively, A crop Indicates the area of ​​the working area, A obs Indicates the area of ​​the obstacle, A road Indicates the area of ​​the road passage area, 、 and represents the weight coefficient;

[0031] Starting from the starting node, add the starting node to the list List, and each time select the node m with the smallest path function from the List to expand. When the current node segment is detected as a surmountable obstacle node, it is marked as a jump node. Based on the motion model of the plant protection robot, the accessibility of the orchard area, and the jump nodes, generate a set of feasible nodes N(m) adjacent to the node m;

[0032] For each adjacent node xl∈N(m), calculate the actual path cost gb(xl) from the starting node to the node xl, and at the same time calculate the path function value h(xl) of the node xl. Evaluate the comprehensive value of the node xl based on the sum of gb(xl) and h(xl). If the node xl is not in the list List, add the node xl to the list List and record the predecessor node of the node xl as m. If the node xl is in the list List and the sum of the newly calculated gb(xl) and h(xl) is less than the old value, update the actual path cost gb(xl) of the node xl and record the predecessor node of the node xl as m.

[0033] Specifically, planning the path of the plant protection robot to obtain the running path of the plant protection robot further includes:

[0034] During the node search process, if the coordinates of one of the nodes are within the preset error range with the coordinates of the target node, it is determined that the target node has been found and the search process ends. Then, starting from the target node, backtrack along the predecessor node pointer of each node, record the passed nodes in turn, and finally obtain the complete path from the starting node to the target node.

[0035] Specifically, the preprocessing includes: grayscale and filtering operations. Grayscale is the process of converting the color orchard map image into a grayscale image; and filtering is used to remove noise.

[0036] A plant protection robot path planning system based on a UAV high-definition map is used to implement the plant protection robot path planning method based on a UAV high-definition map, including: a map image acquisition module, a region segmentation modeling module, a motion module, a path planning module, and an optimization and adjustment module;

[0037] The map image acquisition module is used to use a drone to perform flight mapping of the orchard, obtain a map image of the orchard, and pre-process the obtained orchard map image;

[0038] The region segmentation modeling module is used to perform region segmentation on the pre-processed orchard map image and construct an orchard region feature model;

[0039] The motion module is used to establish a kinematic model based on the mechanical structure and motion characteristics of the plant protection robot and determine a set of feasible motion states of the robot in a two-dimensional plane;

[0040] The path planning module is used to plan the path of the plant protection robot to obtain the running path of the plant protection robot;

[0041] The optimization and adjustment module is used to optimize and smooth the running path of the plant protection robot, and to make local adjustments to the running path of the plant protection robot according to the actual motion performance of the plant protection robot.

[0042] Specifically, the region segmentation modeling module includes: a region segmentation unit, a feature extraction unit and a region feature model construction unit;

[0043] The region segmentation unit is used to train a region segmentation model and perform region segmentation on the pre-processed orchard map image;

[0044] The feature extraction unit is used to extract geometric features and texture features of the area segmented from the orchard map image;

[0045] The regional feature model construction unit is used to integrate the geometric features and texture features of the regions segmented from the extracted orchard map image, as well as the geographic coordinate information, to construct a farmland regional feature model.

[0046] Specifically, the path planning module includes: a path function construction unit, a node judgment unit and a path selection unit;

[0047] The path function building unit is used to build a path function according to the current node coordinates and the target node coordinates;

[0048] The node judgment unit is used to generate a set of feasible nodes adjacent to the node and evaluate the comprehensive value of the adjacent nodes;

[0049] The path selection unit is used for node search to obtain a path from a starting node to a target node.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. The present invention proposes a path planning method for a plant protection robot based on a high-definition UAV map. This method can monitor the dynamic changes in the orchard environment in real time. If there are changes in the morphology of trees during growth, or changes in the position of obstacles due to weather or farming activities, the plant protection robot can dynamically adjust the path in a timely manner based on the real-time updated map information to ensure smooth operation in the complex and changing orchard environment. It will not stagnate or make operational errors due to environmental changes, effectively reducing the need for manual intervention and improving the level of automated operation.

[0052] 2. The present invention proposes a path planning method for a plant protection robot based on a high-definition UAV map. Taking into account node jumping and normal walking, a new global optimal path can be formed, making the plant protection robot more reasonable in energy consumption and avoiding energy waste caused by unreasonable paths.

[0053] 3. The present invention proposes a path planning method for a plant protection robot based on a high-definition UAV map. By accurately dividing the operating area and optimizing the path, the amount of pesticide used can be rationally adjusted according to pesticide demand, ensuring the optimal allocation of energy, pesticides and other resources while meeting the plant protection effect, reducing agricultural production costs and improving the economic benefits of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Flowchart of the path planning method for a plant protection robot based on a high-definition UAV map provided by the present invention;

[0055] Figure 2 An example diagram of an orchard provided by the present invention;

[0056] Figure 3This is an architecture diagram of the plant protection robot path planning system based on drone high-definition maps provided by the present invention. DETAILED DESCRIPTION

[0057] The present application is described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but are not intended to limit the present application in any form. It should be noted that those skilled in the art may make several variations and improvements without departing from the scope of the present application. These all fall within the scope of protection of the present application.

[0058] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0059] It should be noted that, if there is no conflict, the various features in the embodiments of the present application can be combined with each other and are all within the scope of protection of the present application. In addition, although the functional modules are divided in the device schematic and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order than the module division in the device or the order in the flow chart. In addition, the words "first", "second", "third", etc. used in this application do not limit the data and execution order, but only distinguish between the same items or similar items with basically the same functions and effects.

[0060] Unless otherwise defined, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which this application belongs. The terms used in this specification and in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the relevant listed items.

[0061] Example 1

[0062] See also Figure 1-Figure 2 The present invention provides an embodiment of a plant protection robot path planning method based on a UAV high-definition map, comprising the following specific steps:

[0063] Step S1: using a UAV to perform flight mapping of the orchard, obtaining a map image of the orchard, and preprocessing the obtained orchard map image;

[0064] Use drones equipped with high-definition cameras to fly over orchards and obtain high-definition image sequences of the orchards. Through image stitching and geolocation technology, these images are synthesized into high-definition maps containing geographic coordinate information.

[0065] The preprocessing includes grayscale conversion and filtering operations. Grayscale conversion is the process of converting a color orchard map image into a grayscale image. Color images usually contain three channels: red, green, and blue, while grayscale images only have one channel, representing brightness information. A common grayscale conversion method is to convert RGB values ​​into grayscale values ​​through weighted averaging. Filtering operations are used to remove noise and unnecessary details, highlighting feature information such as orchard boundaries, obstacles, and crop planting areas.

[0066] Step S2: performing regional segmentation on the pre-processed orchard map image and constructing an orchard regional feature model;

[0067] The specific steps of step S2 are:

[0068] Step S201: Set the pre-processed orchard map image as I, with a size of H×W, where H represents the height of the pre-processed orchard map image and W represents the width of the pre-processed orchard map image, establish a region segmentation model using the improved U-Net neural network, set a loss function, and train the region segmentation model using the annotated orchard high-definition map image until the loss function converges and remains unchanged, then stop training to obtain a trained region segmentation model;

[0069] In this embodiment, the specific formula of the loss function is: , where L represents the loss function, It represents the balance coefficient, which ranges from 0 to 1. Its optimal value is determined by tuning on the validation set. represents the cross entropy loss function, represents the Dice loss function, The formula is: ,in Represents the height of the feature map extracted by the region segmentation model, Represents the width of the feature map extracted by the regional segmentation model. Here, the feature map extracted by the regional segmentation model is the feature map of the annotated orchard high-definition map image. N represents the number of predefined segmentation categories. For example, the farmland area can be divided into 5 categories: fruit tree planting area, road, water source, obstacle and other background areas. represents the true label of the training image, Represents the predicted label of the training image, log() represents the logarithmic function, The formula is: , The loss function can effectively alleviate the problem of category imbalance and has a good optimization effect on the boundary segmentation effect;

[0070] The height and width here are pixel height and pixel width.

[0071] Step S202: Input the pre-processed orchard map image I into the trained region segmentation model to obtain a region set R of the orchard map image segmentation, where R={r1, r2, ..., r n}, where r n represents the segmented area of ​​the nth orchard map image;

[0072] Step S203: extracting geometric features of the region segmented from the orchard map image, including: area, counting the pixels within the region segmented from the orchard map image, and multiplying the actual area represented by each pixel to obtain the area of ​​the region segmented from the orchard map image;

[0073] Perimeter: Use edge detection algorithm to determine the boundary pixels of the segmented area of ​​the orchard map image, connect the boundary pixels in a clockwise or counterclockwise direction, calculate the Euclidean distance between two adjacent points, and add up all the distances to get the perimeter;

[0074] Shape complexity, the ratio of the square of the perimeter to the area of ​​the segmented region of the orchard map image measures the shape complexity of the segmented region of the orchard map image;

[0075] The larger the value of the shape complexity index is, the more irregular the shape of the region is, and vice versa, the closer it is to a regular shape;

[0076] Step S204: Extracting texture features of the segmented area of ​​the orchard map image, including: Gray Level Co-occurrence Matrix , contrast ,entropy and correlation , d represents the distance offset between pixel pairs, Indicates the direction angle, usually 0°, 45°, 90° and 135°;

[0077] Gray-level co-occurrence matrix Elements in Indicates that in the area of ​​the i-th orchard map image segmentation, the gray value is The pixel and gray value are The pixel point is at a distance d and a direction The contrast reflects the clarity and local variation of the texture in the image, and the calculation formula is: , where hdj represents the grayscale level of the region segmented by the i-th orchard map image, Indicates that the gray value in the gray level co-occurrence matrix is The pixel and gray value are The pixel point is at a distance d and a direction The number of times it appears on the image; entropy is used to measure the randomness of the image texture, and the calculation formula is: ; Correlation represents the linear correlation of grayscale values ​​in the image, and the calculation formula is: , and Represent the grayscale mean in the row and column directions respectively, and Represents the grayscale standard deviation in the row and column directions respectively;

[0078] Step S205: Integrate the geometric features and texture features of the segmented regions of the orchard map image, as well as the geographic coordinate information, to construct a farmland region feature model M. , where A i represents the area of ​​the segmented region of the i-th orchard map image, P i represents the perimeter of the segmented region of the i-th orchard map image, C i represents the regional shape complexity of the i-th orchard map image segmentation, Represents the regional geographic coordinate information of the i-th orchard map image segmentation.

[0079] Step S3: Based on the mechanical structure and motion characteristics of the plant protection robot, a kinematic model is established to determine a set of feasible motion states of the robot in a two-dimensional plane;

[0080] The specific steps of step S3 are:

[0081] Step S301: According to the mechanical structure of the plant protection robot and the motion characteristics of the omnidirectional wheel, the linear velocity of the plant protection robot in the X direction is set to v x , the linear velocity in the Y direction is v y , the angular velocity is ;

[0082] Step S302: In discrete time steps Next, a kinematic model is established, and the specific formula is:

[0083] ,

[0084] in, represents the position coordinates of the plant protection robot at time k+1, represents the position coordinates of the plant protection robot at time k, represents the heading angle of the robot at time k+1, represents the heading angle of the robot at time k, sin() represents the sine function, and cos() represents the cosine function;

[0085] Step S303: setting motion constraints for the plant protection robot, including: velocity in the X and Y directions, angular velocity, acceleration in the X and Y directions, and wheel slip constraints;

[0086] The speed in the X and Y directions must be greater than or equal to the minimum speed and less than or equal to the maximum speed, and the angular velocity must be greater than or equal to the minimum angular velocity and less than or equal to the maximum angular velocity. To prevent the wheels from slipping during movement, the relationship between the linear velocity and angular velocity of the wheels needs to be constrained. Based on the friction coefficient between the wheel and the ground and the acceleration of gravity, the rotational speed of each wheel must be greater than or equal to the product of the negative friction coefficient, the acceleration of gravity, and the radius, and less than or equal to the product of the friction coefficient, the acceleration of gravity, and the radius.

[0087] Step S304: Determine a set of feasible motion states of the robot in a two-dimensional plane according to the kinematic model and constraint conditions of the plant protection robot.

[0088] Step S4: planning the path of the plant protection robot to obtain the running path of the plant protection robot;

[0089] The specific steps of step S4 are:

[0090] Step S401: Set the coordinates of the current node m to (x m ,y m ), the coordinates of the target node G are (x G ,y G ), establish the path function, the specific formula is:

[0091] ,

[0092] in, represents the path function, dl(m) represents the distance term, , cb(m) represents the motion cost of the plant protection robot, , represents the motion cost coefficient, which is determined by experiments and empirical data, including normal walking and jumping costs, and describes the degree of influence of linear velocity and angular velocity on the cost, v m represents the speed of the plant protection robot at node m, represents the angular velocity of the plant protection robot at node m, qb(m) represents the characteristic cost item of the orchard area, and They represent the operating efficiency cost coefficient, penalty cost coefficient and passage cost coefficient respectively, A crop Indicates the area of ​​the working area, A obs Indicates the area of ​​the obstacle, A road Indicates the area of ​​the road passage area, and Represents the weight coefficient; it is used to balance the relative importance of the three items in the path function. A set of more appropriate weight values ​​is determined through multiple experiments and optimizations under different orchard environments and operation tasks.

[0093] Explanation and principle of the above formula: The distance term represents the straight-line distance from the current node to the target node. It is the basic component of the path function and guides the algorithm to search in the target direction. The motion cost term of the plant protection robot is calculated based on the established plant protection robot motion model, taking into account factors such as the robot's energy consumption, time cost, and mechanical wear in different motion states. In the path search process, for each potential expansion node, the corresponding motion cost is calculated based on the speed and angular velocity required for the robot to reach the node from the current node, combined with the motion model, so that the algorithm is more inclined to choose a path with a lower motion cost, thereby optimizing the robot's driving process and improving the accuracy of the path search. High energy utilization efficiency and mechanical life; using the constructed farmland area characteristic model, different cost values ​​are assigned to each area to reflect the quality of the robot's driving in the area. For example, for crop planting areas, a cost coefficient related to the operation efficiency is set according to the growth conditions of the crops (such as height, density, degree of pests and diseases, etc.) and the demand for pesticide spraying. If the crop growth is good and the pests and diseases are serious, a smaller value is taken. On the contrary, if the crops are sparse or the growth condition is poor, a larger value is taken; for obstacle areas, a larger penalty cost coefficient is set to prevent the plant protection robot from entering the area. For roads and other traffic areas, a relatively small traffic cost coefficient is set.

[0094] Step S402: Starting from the starting node, add the starting node to the list List. Each time, select the node m with the smallest path function from List for expansion. When the current node segment is detected as a surmountable obstacle node, mark it as a jump node. Based on the motion model of the plant protection robot, the accessibility of the orchard area, and the jump nodes, generate a set of feasible nodes N(m) adjacent to node m.

[0095] In this embodiment, when generating adjacent nodes, not only are the nodes that the plant protection robot can directly reach at its current speed and direction considered, but also the boundary and obstacle information of the farmland area are combined. Based on the plant protection robot's motion performance, it is determined whether the obstacle nodes can be jumped, thereby avoiding the generation of invalid or unreachable nodes. At the same time, the motion costs of node jumping and detour are further analyzed through node jumping. If the jumping cost is higher than the detour cost, the detour is chosen; otherwise, the node is jumped.

[0096] Step S403: For each adjacent node xl∈N(m), calculate the actual path cost gb(xl) from the starting node to the node xl, and at the same time calculate the path function value h(xl) of the node xl, and evaluate the comprehensive value of the node xl according to the sum of gb(xl) and h(xl). If the node xl is not in the list List, add the node xl to the list List and record the predecessor node of the node xl as m. If the node xl is in the list List and the sum of the newly calculated gb(xl) and h(xl) is less than the old value, update the actual path cost gb(xl) of the node xl and record the predecessor node of the node xl as m.

[0097] In this embodiment, the actual path cost gb(xl) from the starting node to the node xl is equal to the distance from the starting node to the current node m plus the movement cost from the node m to the node xl (calculated according to the motion model, for example, considering factors such as linear velocity, angular velocity, and movement distance); if the node xl is in the list List, since when selecting the adjacent node, the node with the smallest path function is selected, that is, for h(xl), the new and old are equal, if the sum of the newly calculated gb(xl) and h(xl) is less than the old value, it indicates that the new gb(xl) is smaller than the old gb(xl). At this time, it is necessary to select a smaller gb(xl), that is, the new gb(xl), so the actual path cost gb(xl) of the node xl needs to be updated and replaced.

[0098] Step S404: During the node search process, if the coordinates of one of the nodes are within the preset error range with the coordinates of the target node, it is determined that the target node is found, and the search process ends. Then, starting from the target node, backtrack along the predecessor node pointer of each node, record the passed nodes in turn, and finally obtain the complete path from the starting node to the target node.

[0099] In this embodiment, path finding can fully consider various practical factors of the plant protection robot in the orchard environment, and effectively plan a path that meets the operational requirements and has high efficiency and safety. Compared with traditional path planning algorithms, it has better adaptability and practicality, and can significantly improve the intelligence level and quality of agricultural plant protection operations.

[0100] Exemplary, reference Figure 2The starting node is the node where the plant protection robot is located, and the destination fruit tree is the target node. The cost of the plant protection robot moving on the road (black squares) is lower than that on the soil (white squares). The paved road is flatter than the soil. When the target node is Q1, it needs to move on the road and avoid obstacles. When approaching Q1, there are 3 paths, which are represented by arrow directions in the figure. The lower arrow goes through 2 black squares and 6 white squares, the upper arrow goes through 6 black squares and 4 white squares, and the middle arrow goes through 4 black squares, 3 white squares and 1 jump square. The path planning method of the existing technology is used to take the path of the lower arrow (the shortest path). According to the path search method of this application, the sum of gb(xl) and h(xl) is calculated and compared with the old value. The jump square is experienced in the middle, and the path is the shortest and the cost is lower than the other two paths, so the path of the middle arrow is selected as the final path.

[0101] Step S5: Optimize and smooth the running path of the plant protection robot, and make partial adjustments to the running path of the plant protection robot according to the actual motion performance of the plant protection robot.

[0102] Example 2

[0103] See also Figure 3 , another embodiment provided by the present invention: a plant protection robot path planning system based on a UAV high-definition map, comprising: a map image acquisition module, a region segmentation modeling module, a motion module, a path planning module and an optimization adjustment module;

[0104] The map image acquisition module is used to use a drone to perform flight mapping of the orchard, obtain a map image of the orchard, and pre-process the obtained orchard map image;

[0105] The region segmentation modeling module is used to perform region segmentation on the pre-processed orchard map image and construct an orchard region feature model;

[0106] The motion module is used to establish a kinematic model based on the mechanical structure and motion characteristics of the plant protection robot and determine a set of feasible motion states of the robot in a two-dimensional plane;

[0107] The path planning module is used to plan the path of the plant protection robot to obtain the running path of the plant protection robot;

[0108] The optimization and adjustment module is used to optimize and smooth the running path of the plant protection robot, and to make local adjustments to the running path of the plant protection robot according to the actual motion performance of the plant protection robot.

[0109] The region segmentation modeling module includes: a region segmentation unit, a feature extraction unit and a region feature model construction unit;

[0110] The region segmentation unit is used to train a region segmentation model and perform region segmentation on the pre-processed orchard map image;

[0111] The feature extraction unit is used to extract geometric features and texture features of the area segmented from the orchard map image;

[0112] The regional feature model construction unit is used to integrate the geometric features and texture features of the regions segmented from the extracted orchard map image, as well as the geographic coordinate information, to construct a farmland regional feature model.

[0113] The path planning module includes: a path function construction unit, a node judgment unit and a path selection unit;

[0114] The path function building unit is used to build a path function according to the current node coordinates and the target node coordinates;

[0115] The node judgment unit is used to generate a set of feasible nodes adjacent to the node and evaluate the comprehensive value of the adjacent nodes;

[0116] The path selection unit is used for node search to obtain a path from a starting node to a target node.

[0117] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0118] The above-described specific embodiments further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A path planning method for a plant protection robot based on a UAV high-definition map, characterized in that: include: Use UAVs to conduct flight mapping of orchards, obtain map images of orchards, and pre-process the obtained orchard map images; Perform regional segmentation on the pre-processed orchard map image and construct an orchard regional feature model; Based on the mechanical structure and motion characteristics of the plant protection robot, a kinematic model is established to determine the set of feasible motion states of the robot in a two-dimensional plane; Plan the path of the plant protection robot to obtain the running path of the plant protection robot; Optimize and smooth the running path of the plant protection robot, and make partial adjustments to the running path of the plant protection robot based on its actual motion performance; The path planning of the plant protection robot to obtain the running path of the plant protection robot includes: Set the coordinates of the current node m to (x m ,y m ), the coordinates of the target node G are (x G ,y G ), establish a path function based on the distance term, the plant protection robot motion cost term and the orchard area characteristic cost term; Starting from the starting node, add the starting node to the list List, and each time select the node m with the smallest path function from the List to expand. When the current node segment is detected as a surmountable obstacle node, it is marked as a jump node. Based on the motion model of the plant protection robot, the accessibility of the orchard area, and the jump nodes, generate a set of feasible nodes N(m) adjacent to the node m; For each adjacent node xl∈N(m), calculate the actual path cost gb(xl) from the starting node to the node xl, and calculate the path function value h(xl) of the node xl. Evaluate the comprehensive value of the node xl based on the sum of gb(xl) and h(xl). If the node xl is not in the list List, add the node xl to the list List and record the predecessor node of the node xl as m. If the node xl is in the list List and the sum of the newly calculated gb(xl) and h(xl) is less than the old value, update the actual path cost gb(xl) of the node xl and record the predecessor node of the node xl as m. During the node search process, if the coordinates of one of the nodes are within the preset error range with the coordinates of the target node, it is determined that the target node has been found and the search process ends. Then, starting from the target node, backtrack along the predecessor node pointer of each node, record the passed nodes in turn, and finally obtain the complete path from the starting node to the target node.

2. The plant protection robot path planning method based on a drone high-definition map according to claim 1, characterized in that: The process of performing region segmentation on the pre-processed orchard map image comprises: The preprocessed orchard map image is set as I, with a size of H×W, where H represents the height of the preprocessed orchard map image and W represents the width of the preprocessed orchard map image. The improved U-Net neural network is used to establish a region segmentation model, and a loss function is set. The region segmentation model is trained using annotated orchard high-definition map images until the loss function converges and remains unchanged. The training is stopped to obtain a trained region segmentation model. The pre-processed orchard map image I is input into the trained region segmentation model to obtain the region set R of the orchard map image segmentation, R={r1,r2,...,r n }, where r n Represents the segmented area of ​​the nth orchard map image.

3. The plant protection robot path planning method based on the UAV high-definition map according to claim 2 is characterized in that: The orchard area characteristic model is constructed, comprising: Extract the geometric features of the segmented regions of the orchard map image, including area, perimeter and shape complexity; Extract the texture features of the segmented area of ​​the orchard map image, including: gray level co-occurrence matrix , contrast ,entropy and correlation , d represents the distance offset between pixel pairs, Indicates the direction angle; The geometric features and texture features of the segmented area of ​​the orchard map image are extracted, as well as the geographic coordinate information, to construct the farmland area feature model M. , where A i represents the area of ​​the segmented region of the i-th orchard map image, P i represents the perimeter of the segmented region of the i-th orchard map image, C i represents the regional shape complexity of the i-th orchard map image segmentation, Represents the regional geographic coordinate information of the i-th orchard map image segmentation.

4. The plant protection robot path planning method based on a drone high-definition map according to claim 3 is characterized in that: The kinematic model is established based on the mechanical structure and motion characteristics of the plant protection robot to determine the set of feasible motion states of the robot in a two-dimensional plane, including: According to the mechanical structure of the plant protection robot and the motion characteristics of the omnidirectional wheel, the linear velocity of the plant protection robot in the X direction is set to v x , the linear velocity in the Y direction is v y , the angular velocity is ; In discrete time steps Next, a kinematic model is established, and the specific formula is: ， in, represents the position coordinates of the plant protection robot at time k+1, represents the position coordinates of the plant protection robot at time k, represents the heading angle of the robot at time k+1, represents the heading angle of the robot at time k, sin() represents the sine function, and cos() represents the cosine function; Set the motion constraints of the plant protection robot, including: velocity in the X and Y directions, angular velocity, acceleration in the X and Y directions, and wheel slip constraints; According to the kinematic model and constraints of the plant protection robot, the set of feasible motion states of the robot in a two-dimensional plane is determined.

5. The plant protection robot path planning method based on UAV high-definition map according to claim 4 is characterized in that: The preprocessing includes: grayscale and filtering operations. Grayscale is the process of converting a color orchard map image into a grayscale image; and filtering is used to remove noise.

6. A plant protection robot path planning system based on a drone high-definition map, used to implement the plant protection robot path planning method based on a drone high-definition map according to any one of claims 1 to 5, characterized in that: include: Map image acquisition module, region segmentation modeling module, motion module, path planning module and optimization adjustment module; The map image acquisition module is used to use a drone to perform flight mapping of the orchard, obtain a map image of the orchard, and pre-process the obtained orchard map image; The region segmentation modeling module is used to perform region segmentation on the pre-processed orchard map image and construct an orchard region feature model; The motion module is used to establish a kinematic model based on the mechanical structure and motion characteristics of the plant protection robot and determine a set of feasible motion states of the robot in a two-dimensional plane; The path planning module is used to plan the path of the plant protection robot to obtain the running path of the plant protection robot; The optimization and adjustment module is used to optimize and smooth the running path of the plant protection robot, and to make local adjustments to the running path of the plant protection robot according to the actual motion performance of the plant protection robot.

7. The plant protection robot path planning system based on UAV high-definition map according to claim 6, characterized in that: The region segmentation modeling module includes: a region segmentation unit, a feature extraction unit and a region feature model construction unit; The region segmentation unit is used to train a region segmentation model and perform region segmentation on the pre-processed orchard map image; The feature extraction unit is used to extract geometric features and texture features of the area segmented from the orchard map image; The regional feature model construction unit is used to integrate the geometric features and texture features of the regions segmented from the extracted orchard map image, as well as the geographic coordinate information, to construct a farmland regional feature model.

8. The plant protection robot path planning system based on UAV high-definition map according to claim 7, characterized in that: The path planning module includes: a path function construction unit, a node judgment unit and a path selection unit; The path function building unit is used to build a path function according to the current node coordinates and the target node coordinates; The node judgment unit is used to generate a set of feasible nodes adjacent to the node and evaluate the comprehensive value of the adjacent nodes; The path selection unit is used for node search to obtain a path from a starting node to a target node.

Images