Method and device for automatic and precise pesticide application in a dense orchard based on a point cloud map

Abstract

The application provides a point cloud map-based automatic precise pesticide application method and device for a dense orchard, which comprises the following steps: setting a plant protection robot in a target orchard, and constructing an offline point cloud map corresponding to the target orchard; performing semantic segmentation on the offline point cloud map to obtain a pose matching database and a tree crown area database; performing block segmentation on the offline point cloud map based on a fruit tree planting row matrix and a fruit tree planting plant matrix in the target orchard to obtain an orchard topological model; and the plant protection robot travels along a preset route in the target orchard and applies pesticides. According to the scheme, the dynamic objects and the static objects are accurately separated through semantic segmentation, the influence of "tunnel effect" is eliminated by using matching based on topological constraints, the point cloud matching positioning is obtained, and the travel route of the plant protection robot is ensured, so that the tree crown layer is applied with pesticides.

Description

Technical Field

[0001] This application relates to the field of automated agriculture, and in particular to a method and device for automated precision pesticide application in closed orchards based on point cloud maps. Background Technology

[0002] With the rapid development of information technology and robotics, unmanned plant protection robots have become a key development direction for orchard plant protection operations, with autonomous navigation systems and precision spraying systems being their core components. Under current technological systems, autonomous navigation systems typically utilize Global Navigation Satellite Systems (GNSS), lidar, vision technology, or multi-sensor fusion to collect environmental information, thereby enabling the robot's positioning and navigation. Precision spraying systems rely on sensors such as ultrasonic waves, lidar, and vision to acquire data on canopy size and porosity, calculating the amount of pesticide to be applied, thus improving pesticide utilization and reducing environmental pollution.

[0003] However, orchards are typical semi-structured environments, and the complexity of their dynamic interference far exceeds that of outdoor or industrial scenarios, causing existing technologies to reveal many drawbacks. For example, GNSS signals are easily blocked by the canopy of fruit trees, leading to positioning deviations. Moreover, dynamic changes such as wind-induced canopy swaying, branch pruning, and orchard weeding can cause partial inaccuracies in lidar or offline visual maps, severely reducing the positioning accuracy of subsequent matching algorithms. In addition, fruit trees are densely arranged according to specific row and plant spacing, presenting a symmetrical and repetitive "tunnel-like" feature in the lidar field of view. This can easily lead to positioning ambiguity, feature matching errors, and cumulative drift. For example, patent CN116548412A uses GNSS for navigation control, enabling the machine to autonomously drive along the centerline of the passage and using lidar to acquire point clouds of the canopy on one side to calculate pesticide dosage and wind speed. This method relies on GNSS, which is easily affected by canopy occlusion in the orchard environment, impacting positioning accuracy. Other techniques use single-line lidar real-time point cloud (CN120028803A), laser rangefinders (CN119498265A), infrared sensors (CN119867041A), or ultrasound to estimate canopy size. These methods typically extrapolate canopy shape based on a small number of points, resulting in low accuracy. Furthermore, existing autonomous navigation systems and precision spraying systems mostly use independent sensors to acquire the necessary information, which undoubtedly increases the number of sensors used, raising both the overall equipment cost and the complexity of the robot's sensing system. Summary of the Invention

[0004] This application provides an automated and precise pesticide application method and device for closed orchards based on point cloud maps. It achieves precise separation of dynamic and static objects through semantic segmentation, and uses topological constraint-based matching to eliminate the "tunneling effect" and obtain point cloud matching positioning, ensuring the travel route of the plant protection robot, thereby applying pesticides to the canopy layer.

[0005] In a first aspect, embodiments of this application provide an automated and precise pesticide application method for densely canopied orchards based on point cloud maps, the method comprising:

[0006] Plant protection robots are deployed in the target orchard, and an offline point cloud map corresponding to the target orchard is constructed.

[0007] Semantic segmentation is performed on the offline point cloud map to obtain a pose matching database and a canopy area database. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies.

[0008] The orchard topology model is obtained by segmenting the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard.

[0009] The plant protection robot travels along a preset route and applies pesticides within the target orchard. During its journey, the robot acquires real-time point cloud data and its current block position in the orchard topology model. It also retrieves offline point cloud data of the current block position and adjacent block positions from a pose matching database. The offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volume of the fruit trees on both sides of the robot from a canopy area database, calculates the pesticide application amount based on the canopy volume of the fruit trees on both sides, and applies the pesticide.

[0010] Secondly, embodiments of this application provide an automated precision pesticide application device for densely canopied orchards based on point cloud maps, comprising:

[0011] The setup module is used to set up plant protection robots in the target orchard and build an offline point cloud map corresponding to the target orchard.

[0012] The module constructs a pose matching database and a canopy area database by performing semantic segmentation on the offline point cloud map. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies.

[0013] The segmentation module divides the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard to obtain the orchard topology model.

[0014] In the pesticide application module, the plant protection robot travels along a preset route within the target orchard and applies pesticides. During its movement, the plant protection robot acquires real-time point cloud data and its current block position in the orchard topology model. It also retrieves offline point cloud data of the current block position and adjacent block positions from the pose matching database. The offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volume of the fruit trees on both sides of the plant protection robot from the canopy area database, calculates the pesticide application amount based on the canopy volume of the fruit trees on both sides, and applies the pesticide.

[0015] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute an automated precision pesticide application method for densely canopied orchards based on point cloud maps.

[0016] Fourthly, embodiments of this application provide a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including an automated precision pesticide application method for densely canopied orchards based on point cloud maps.

[0017] The main contributions and innovations of this invention are as follows:

[0018] This application embodiment utilizes an agricultural robot to traverse the orchard in a serpentine or back-and-forth manner. It combines LiDAR data and SLAM algorithms to construct an offline point cloud map and perform semantic segmentation, thereby filtering out dynamic interference and ensuring the stability of point cloud matching. This solves the positioning offset problem caused by GNSS signal obstruction and dynamic environments. Based on the row and plant spacing of the fruit trees, this scheme segments the map into topologically labeled block models, eliminating the "tunnel effect" formed by densely arranged fruit trees and avoiding positioning ambiguity and cumulative drift. During pesticide application, the scheme adjusts the path by matching the topological constraints of real-time point clouds with offline static object point clouds and calculates the pesticide dosage based on the canopy volume, ensuring the stability of the agricultural robot's travel route. Furthermore, precise pesticide application based on canopy volume reduces pesticide waste and environmental pollution, improving the efficiency and accuracy of automated pesticide application in closed orchards.

[0019] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This is a flowchart of an automated and precise pesticide application method for a closed orchard based on a point cloud map, according to an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of an agricultural robot according to an embodiment of this application;

[0023] Figure 3 This is a schematic diagram illustrating the creation of an offline point cloud map according to an embodiment of this application;

[0024] Figure 4 This is a structural block diagram of an automated precision spraying device for a closed orchard based on a point cloud map, according to an embodiment of this application.

[0025] Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0027] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0028] Example 1

[0029] This application provides an automated and precise pesticide application method for closed orchards based on point cloud maps. It achieves accurate separation of dynamic and static objects through semantic segmentation and utilizes topological constraint-based matching to eliminate the "tunneling effect" and obtain point cloud matching positioning, ensuring the travel route of the plant protection robot and thus applying pesticides to the canopy layer. Specifically, refer to... Figure 1 The method includes:

[0030] Plant protection robots are deployed in the target orchard, and an offline point cloud map corresponding to the target orchard is constructed.

[0031] Semantic segmentation is performed on the offline point cloud map to obtain a pose matching database and a canopy area database. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies.

[0032] The orchard topology model is obtained by segmenting the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard.

[0033] The plant protection robot travels along a preset route and applies pesticides within the target orchard. During its journey, the robot acquires real-time point cloud data and its current block position in the orchard topology model. It also retrieves offline point cloud data of the current block position and adjacent block positions from a pose matching database. The offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volume of the fruit trees on both sides of the robot from a canopy area database, calculates the pesticide application amount based on the canopy volume of the fruit trees on both sides, and applies the pesticide.

[0034] In some embodiments, the structure of the plant protection robot in this solution is as follows: Figure 2 As shown, the plant protection robot is equipped with hardware such as lidar, IMU, controller, spraying device, and driving chassis. The coordinated operation of these hardware components enables the plant protection robot to move within the target orchard and apply pesticides.

[0035] Specifically, the plant protection robot uses a 16-line lidar to construct an offline point cloud map of the target orchard in the early stages, and to acquire real-time point cloud data during pesticide application. For example, the lidar of the plant protection robot in this solution is a Velodyne VLP-16, which has a maximum range of 100m, an accuracy of ±3cm, a vertical field of view of 30° (±5°), a horizontal field of view of 360°, and a vertical angular resolution of 2°. This solution sets the lidar scanning speed to 10Hz and the horizontal angular resolution to 0.2° to ensure a comprehensive and detailed scan of the fruit trees and surrounding environment in the orchard. The lidar communicates with the main control unit via a 100Mbps Ethernet to ensure efficient and stable data transmission.

[0036] Specifically, the IMU is used to acquire motion state information such as acceleration and angular velocity of the plant protection robot during its movement. For example, the IMU used in this solution is the Wheeltec N200, which has an acceleration measurement range of ±16g, an angular velocity measurement range of ±2000° / s, a data output frequency of 1000Hz, and a zero-bias stability of 2° / h. In this solution, the IMU is rigidly connected to the lidar to acquire the motion state information of the plant protection robot in real time.

[0037] Specifically, the controller runs software modules such as SLAM algorithm, point cloud matching algorithm, and spray control logic, processes data from various sensors, and generates control commands. The spraying device consists of functional components such as a pesticide tank, diaphragm pump, solenoid valves, and nozzles. The nozzles are arranged on the left and right sides, with each nozzle controlled independently by a solenoid valve. During operation, the pesticide solution is transported from the pesticide tank through pipelines to the inlet of the diaphragm pump. Driven by a motor, the diaphragm pump pressurizes the solution and delivers it to the solenoid valves. The solenoid valves receive commands from the controller and adjust their valve core positions to control the output flow rate, thereby achieving variable-rate spraying. A chassis carrying other hardware enables the system to move autonomously within the orchard.

[0038] In some implementations, when constructing an offline point cloud map corresponding to the target orchard, the plant protection robot traverses the orchard along the passageways in a serpentine or back-and-forth manner. During the traversal, the LiDAR scans each object in the target orchard to obtain LiDAR data, and then constructs an offline point cloud map based on the LiDAR data.

[0039] Specifically, when constructing an offline point cloud map, the plant protection robot maintains a relatively stable speed while moving within the target orchard, and the speed cannot be too fast, such as 0.5 to 1 m / s, so as to ensure that the lidar can fully scan every object in the target orchard.

[0040] Furthermore, the motion state information of the plant protection robot as it traverses the orchard is obtained, and the SLAM algorithm is used to obtain an offline point cloud map in real time based on the motion state information and LiDAR data.

[0041] Specifically, SLAM is an information technology in which a robot, while acquiring information about its surroundings through sensors in an unknown environment, calculates its own position and orientation, and constructs an environmental map. In this scheme, the LIO-SAM (Lidar-Inertial Odometry with Scan Matching) algorithm is used as the framework. First, the motion state information is pre-integrated to obtain an initial pose estimate, which provides the initial pose for the LiDAR point cloud. Based on the local map of the target orchard, features are extracted from the LiDAR data to obtain the LiDAR matching factor. Then, the initial pose estimate and the LiDAR matching factor are globally optimized through a sliding window to obtain an offline point cloud map. Figure 3 A schematic diagram for creating an offline point cloud map.

[0042] In some embodiments, a pre-trained segmentation algorithm is used to identify each data point in the offline point cloud map and segment it to obtain a segmentation result. The segmentation result includes dynamic objects and static objects. The point cloud data of all static objects are used to form a pose matching database. All static objects other than the fruit tree canopy point cloud are filtered out and the spatial volume of each fruit tree canopy is calculated to form a canopy area database.

[0043] Specifically, the static objects include field ridges and tree trunks. Correspondingly, in addition to static objects, the segmentation algorithm will also segment some dynamic objects, including tree canopies, ground, weeds, etc. The complexity of dynamic disturbances of dynamic objects far exceeds that of outdoor or industrial scenes. Dynamic disturbances such as wind disturbing the canopy, tree pruning, and weeding can cause the offline map to become partially inaccurate, thereby reducing the positioning accuracy of subsequent matching algorithms. Therefore, this solution will filter out all dynamic objects except tree canopies to ensure the accuracy of subsequent point cloud matching.

[0044] Specifically, this solution uses the PointNet++ network model to segment offline point cloud maps. During PointNet++ network training, a large number of orchard point cloud data samples containing information on different categories such as tree canopy, trunk, ground, and field ridges are collected and labeled, marking the category of each point. During training, the initial learning rate is set to 0.001, allowing the model to quickly learn basic features in the early stages of training. As training progresses, a step-wise decay strategy is adopted, multiplying the learning rate by 0.5 every 50 epochs to avoid parameter oscillations and allow the model to converge more accurately to the optimal solution. When the learning rate decays to 1e-5, this value is maintained until the end of training to ensure the model's learning effect on detailed features (such as canopy edge points). The total number of iterations is set to 300-500 epochs. The orchard point cloud dataset typically consists of 1000-5000 samples (each sample is a single fruit tree or a local orchard scene point cloud). During training, model accuracy is monitored using a validation set. Training is terminated early when the validation set accuracy shows no significant improvement (fluctuation less than 0.5%) for 20 consecutive epochs to avoid overfitting. 1024-2048 points are retained for the canopy point cloud sample of each fruit tree.

[0045] In some embodiments, the target orchard is divided into rectangular blocks of planting row spacing × planting plant size to obtain an orchard topology model. In the orchard topology model, the row number and column number are used as the block number of each rectangular block. Furthermore, each rectangular block is divided into multiple sub-voxels, and the row number, column number, and point cloud classification information of the sub-voxel are used as the topology label of the sub-voxel.

[0046] Specifically, since fruit trees in orchards are generally planted in a matrix pattern, with m as the row spacing (distance between two rows of trees) and n as the plant spacing (distance between two trees), this plan divides the target orchard into rectangular blocks of size m×n. Each rectangular block is numbered as follows: , where i is the row number of the corresponding block in the orchard topology model, and j is the column number of the corresponding block in the orchard topology model.

[0047] In this scheme, the size of the sub-voxel is 0.5m×0.5m. The point cloud classification information within the sub-voxel is obtained, and the row number, column number, and point cloud classification information within the sub-voxel are used as the topological label of the sub-voxel. For example, if the sub-voxel belonging to the third row and second column of the rectangular block mainly contains the point cloud category of field ridges, then the topological label of the sub-voxel is "Row 3 Column 2 - Field Ridge Area". If the sub-voxel belonging to the third row and second column of the rectangular block mainly contains the point cloud category of tree trunks, then the topological label of the sub-voxel is "Row 3 Column 2 - Tree Trunk Area".

[0048] Furthermore, an adjacency table is established in the orchard topology model, which stores the adjacency relationships between different blocks. For example, the adjacency table records that "row 3, column 2" and "row 3, column 3" are adjacent.

[0049] In some embodiments, the current position coordinates of the plant protection robot are calculated based on the x-axis acceleration and y-axis acceleration of the plant protection robot at the current moment, and the current block position of the plant protection robot in the orchard topology model is obtained based on the current position coordinates, planting row spacing, and planting plant spacing.

[0050] Specifically, the x-axis acceleration and y-axis acceleration at the current moment are obtained using the motion state information acquired by the IMU in the agricultural machinery. The formula for calculating the position coordinates at the current moment is expressed as follows:

[0051]

[0052] in, Let k be the coordinates of the plant protection robot at the current time. Let x and y be the accelerations at the current time k. For time step.

[0053] Specifically, the formula for obtaining the current block position of the plant protection robot in the orchard topology model is as follows:

[0054]

[0055]

[0056] Where row is the row number of the current block, col is the column number of the current block, m is the planting row spacing, and n is the planting plant spacing. Let x be the x-coordinate of the plant protection robot at the current time k. Let k be the y-coordinate of the plant protection robot at the current time k.

[0057] In some embodiments, a segmentation algorithm is used to segment the real-time point cloud data, and the static objects of the segmentation result are retained as source point cloud data. Offline point cloud data is used as target point cloud data. An initial transformation matrix between the source point cloud data and the target point cloud data is calculated. Then, constraints are set, and the initial transformation matrix is ​​iterated based on the constraints to obtain the optimal transformation matrix. The optimal transformation matrix is ​​used as the matching result. The constraint is: matching the same static objects in the source point cloud data and the target point cloud data.

[0058] In other words, this scheme first performs point cloud matching on the source and target point cloud data using a conventional traversal method to obtain an initial transformation matrix. To ensure the accuracy of point cloud matching, this scheme sets a constraint to update the initial transformation matrix, thereby obtaining a more accurate optimal transformation matrix. Since both the source and target point cloud data in this scheme are point cloud data with dynamic objects filtered out, the constraint is to match the same static objects in the source and target point cloud data. For example, matching tree trunks in the source and target point cloud data, and matching field ridges in the source and target point cloud data, makes the matching results more accurate and obtains the optimal transformation matrix.

[0059] Specifically, this solution obtains point cloud data of different static objects based on the topology labels in the target point cloud data.

[0060] In some embodiments, where row is the row number of the current block position and col is the column number of the current block position, the offline point cloud data of the current block position and adjacent block positions are represented by the following formula:

[0061]

[0062] Where X is the set of markers for the current block position and the positions of adjacent blocks. The block identifier is i, where i is the row number and j is the column number.

[0063] Specifically, during the iteration process of the initial transformation matrix, the SVD algorithm is used to solve for the new transformation matrix and perform iterative optimization. During the iteration process, the error of the corresponding point pair needs to be recalculated in each round. If the error change between two consecutive iterations is less than 0.01 m or the number of iterations reaches 30, the optimization is stopped.

[0064] Specifically, this solution uses an optimal transformation matrix to adjust the position and attitude of the plant protection robot, so that the plant protection robot always travels along the optimal path.

[0065] Specifically, since fruit trees are arranged at a certain row and plant spacing, they have distinct topological structure characteristics. Dense and similar fruit trees form symmetrical and repetitive "tunnel-like" features in the lidar field of view, resulting in high homogeneity of point cloud features in the horizontal direction and limited field of view in the vertical direction. This leads to localization ambiguity, feature matching errors and cumulative drift. This solution achieves accurate separation of dynamic and static features through semantic segmentation and uses topological constraint-based matching to eliminate the influence of the "tunnel effect" and obtain point cloud matching and localization.

[0066] In some specific embodiments, the target orchard is divided into multiple plant protection zones based on the application range of the plant protection robot. The canopy volume of the fruit trees on both sides of the passage in each plant protection zone is obtained based on the canopy area database. The application time is calculated and the application is carried out using the following formula:

[0067] Q=k×V

[0068] Since the plant protection robot applies pesticides through nozzles on both sides, Q is the amount of pesticide applied by a single nozzle, k is the pesticide application coefficient per unit volume (e.g., 0.2 L / m³, which can be adjusted according to the type of pesticide and the variety of fruit tree), and V is the volume of the fruit tree canopy on one side.

[0069] In some specific embodiments, the system determines which plant protection zone the plant protection robot has entered as it moves and calculates the amount of pesticide to apply.

[0070] In some specific embodiments, during the operation of the agricultural robot, the LiDAR continuously collects point clouds of the surrounding environment at a frequency of 10Hz, with each frame containing 20,000-50,000 points. The collected raw real-time point cloud is first subjected to statistical filtering to remove noise points with abnormal distances, and then the point cloud density is downsampled to 5,000-10,000 points / frame using voxel filtering. The IMU outputs the robot's angular velocity and acceleration data at 100Hz, achieving hard synchronization with the LiDAR point cloud through timestamp alignment. The IMU data is preprocessed by Kalman filtering to output the robot's real-time attitude angles and motion velocity, providing initial pose estimation for point cloud matching. The above-mentioned point cloud matching and localization algorithm based on orchard features is used for localization, outputting the robot's position and heading angle in the global coordinate system. The localization process employs a hierarchical search strategy. First, a 10 m × 10 m × 5 m local map around the robot's current location is extracted from the offline map and matched with the real-time point cloud. When the matching error exceeds 0.3 m for three consecutive times, the local map range is automatically expanded to 20 m × 20 m × 5 m and rematched to ensure stable localization even in densely fruit-tree areas.

[0071] Example 2

[0072] Based on the same concept, referencing Figure 4 This application also proposes an automated precision pesticide application device for densely canopied orchards based on point cloud maps, comprising:

[0073] The setup module is used to set up plant protection robots in the target orchard and build an offline point cloud map corresponding to the target orchard.

[0074] The module constructs a pose matching database and a canopy area database by performing semantic segmentation on the offline point cloud map. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies.

[0075] The segmentation module divides the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard to obtain the orchard topology model.

[0076] In the pesticide application module, the plant protection robot travels along a preset route within the target orchard and applies pesticides. During its movement, the plant protection robot acquires real-time point cloud data and its current block position in the orchard topology model. It also retrieves offline point cloud data of the current block position and adjacent block positions from the pose matching database. The offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volume of the fruit trees on both sides of the plant protection robot from the canopy area database, calculates the pesticide application amount based on the canopy volume of the fruit trees on both sides, and applies the pesticide.

[0077] Example 3

[0078] This embodiment also provides an electronic device, see reference. Figure 5 It includes a memory 404 and a processor 402, wherein the memory 404 stores a computer program and the processor 402 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0079] Specifically, the processor 402 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0080] Memory 404 may include a mass storage device for data or instructions. For example, and not limitingly, memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 404 may include removable or non-removable (or fixed) media. Where appropriate, memory 404 may be internal or external to a data processing device. In a particular embodiment, memory 404 is non-volatile memory. In a particular embodiment, memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0081] The memory 404 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 402.

[0082] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any of the automated precision pesticide application methods for dense orchards based on point cloud maps in the above embodiments.

[0083] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408, wherein the transmission device 406 is connected to the processor 402, and the input / output device 408 is connected to the processor 402.

[0084] The transmission device 406 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 406 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0085] Input / output device 408 is used to input or output information. In this embodiment, the input information may be real-time point cloud data, orchard topology model, etc., and the output information may be the travel route of the plant protection robot, the amount of pesticide applied, etc.

[0086] Optionally, in this embodiment, the processor 402 can be configured to perform the following steps via a computer program:

[0087] Plant protection robots are deployed in the target orchard, and an offline point cloud map corresponding to the target orchard is constructed.

[0088] Semantic segmentation is performed on the offline point cloud map to obtain a pose matching database and a canopy area database. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies.

[0089] The orchard topology model is obtained by segmenting the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard.

[0090] The plant protection robot travels along a preset route and applies pesticides within the target orchard. During its journey, the robot acquires real-time point cloud data and its current block position in the orchard topology model. It also retrieves offline point cloud data of the current block position and adjacent block positions from a pose matching database. The offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volume of the fruit trees on both sides of the robot from a canopy area database, calculates the pesticide application amount based on the canopy volume of the fruit trees on both sides, and applies the pesticide.

[0091] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0092] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0093] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 5 Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0094] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0095] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for automated and precise pesticide application in densely canopied orchards based on point cloud maps, characterized in that, Includes the following steps: Plant protection robots are deployed in the target orchard, and an offline point cloud map corresponding to the target orchard is constructed. Semantic segmentation is performed on the offline point cloud map to obtain a pose matching database and a canopy area database. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies. The orchard topology model is obtained by segmenting the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard. The plant protection robot travels along a preset route and applies pesticides within the target orchard. During its movement, the robot acquires real-time point cloud data and its current block position within the orchard topology model. It also retrieves offline point cloud data of its current block position and adjacent block positions from a pose matching database. This offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot obtains the canopy volumes of the fruit trees on both sides of its path from a canopy area database, calculates the pesticide application amount based on these volumes, and applies the pesticide. A segmentation algorithm is used to segment the real-time point cloud data, retaining the static objects from the segmentation results as source point cloud data. The offline point cloud data is used as the target point cloud data. An initial transformation matrix between the source and target point cloud data is calculated, and constraints are set. The initial transformation matrix is ​​iteratively applied based on these constraints to obtain the optimal transformation matrix, which is then used as the matching result. The constraint condition is to match identical static objects in the source and target point cloud data.

2. The automated and precise pesticide application method for densely canopied orchards based on point cloud maps according to claim 1, characterized in that, When constructing an offline point cloud map corresponding to the target orchard, the plant protection robot traverses the orchard in a serpentine or back-and-forth manner along the passageways. During the traversal, the LiDAR scans each object in the target orchard to obtain LiDAR data, and then constructs an offline point cloud map based on the LiDAR data.

3. The automated precision pesticide application method for densely canopied orchards based on point cloud maps according to claim 2, characterized in that, The motion status information of the plant protection robot when it traverses the orchard is obtained, and the SLAM algorithm is used to obtain an offline point cloud map in real time based on the motion status information and LiDAR data.

4. The automated precision pesticide application method for densely canopied orchards based on point cloud maps according to claim 1, characterized in that, The pre-trained segmentation algorithm is used to identify each data point in the offline point cloud map and segment it to obtain the segmentation result. The segmentation result includes dynamic objects and static objects. The point cloud data of all static objects are used to form a pose matching database. All static objects except the fruit tree canopy point cloud are filtered out and the spatial volume of each fruit tree canopy is calculated to form a canopy area database.

5. The automated precision pesticide application method for densely canopied orchards based on point cloud maps according to claim 1, characterized in that, The target orchard is divided into rectangular blocks of planting row spacing × planting plant spacing to obtain an orchard topology model. Each rectangular block is marked with row number and column number in the orchard topology model. Each rectangular block is further divided into multiple sub-voxels, and the row number, column number, and point cloud classification information of the sub-voxel are used as the topology label of the sub-voxel.

6. The automated precision pesticide application method for densely canopied orchards based on point cloud maps according to claim 1, characterized in that, The current position coordinates of the plant protection robot are calculated based on its x-axis and y-axis accelerations at the current moment. Based on the current position coordinates, planting row spacing, and planting plant spacing, the current block position of the plant protection robot in the orchard topology model is obtained.

7. An automated precision pesticide application device for densely canopied orchards based on point cloud maps, characterized in that, include: The setup module is used to set up plant protection robots in the target orchard and build an offline point cloud map corresponding to the target orchard. The module constructs a pose matching database and a canopy area database by performing semantic segmentation on the offline point cloud map. The pose matching database includes point cloud data of all static objects in the offline point cloud map, and the canopy area database contains point cloud data of the canopy of each fruit tree in the offline point cloud map and the space volume it occupies. The segmentation module divides the offline point cloud map into blocks based on the row and plant spacing of fruit trees in the target orchard to obtain the orchard topology model. The pesticide application module involves a plant protection robot traveling along a preset route within the target orchard and applying pesticides. During its movement, the robot acquires real-time point cloud data and its current block position within the orchard topology model. It also retrieves offline point cloud data of its current block position and adjacent block positions from a pose matching database. This offline point cloud data is matched with the real-time point cloud data, and the robot's path is adjusted based on the matching results. The robot also retrieves the canopy volumes of the fruit trees on either side of it from a canopy area database, calculates the pesticide application amount based on these volumes, and applies the pesticide. The real-time point cloud data is segmented using a segmentation algorithm, and the segmented static objects are retained as source point cloud data. The offline point cloud data is used as the target point cloud data. An initial transformation matrix between the source and target point cloud data is calculated, and constraints are set. The initial transformation matrix is ​​iteratively applied based on these constraints to obtain an optimal transformation matrix, which is then used as the matching result. The constraint condition is to match identical static objects in the source and target point cloud data.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to execute the automated precision pesticide application method for dense orchards based on point cloud maps as described in any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which includes program code for controlling the process to execute the process, the process including an automated precision pesticide application method for densely canopied orchards based on point cloud maps according to any one of claims 1-6.

Images