Intelligent orchard inspection robot and control system and control method thereof

Abstract

The invention discloses an intelligent orchard inspection robot and a control system and method thereof, and relates to the technical field of robots, and the method comprises the steps: customizing an inspection task at an APP client according to an orchard demand, and remotely controlling the operation state and parameter information of the robot; after the robot is started, the robot starts to walk, collects multi-view images and point cloud data of the surrounding environment, and transmits the multi-view images and the point cloud data to the data processing module, the inspection navigation subsystem and the cloud control subsystem; an Alpha Shape algorithm is combined with a grid integration method, and the fruit tree canopy volume is accurately measured and calculated according to the point cloud data; receiving and processing the image and the point cloud data to obtain environment monitoring data and crop growth information, and generating current real-time inspection information, a pest and disease identification result and an equipment laser cloud picture; and processing the point cloud data to obtain a motion track of the robot, and constructing a map to realize autonomous inspection of the robot. According to the invention, accurate path planning can be realized, and autonomous routing inspection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and in particular to an intelligent orchard inspection robot and a control system and a control method thereof. Background Art

[0002] With the rapid development of modern fruit industry, the application of intelligent agricultural equipment is becoming more and more extensive. As an important tool to improve agricultural production efficiency and management level, orchard inspection robots have achieved certain technical results and have been put into practical application in agriculture in various places. Orchard inspection robots can autonomously inspect in orchards, monitor crop growth conditions and environmental changes in real time, and make intelligent judgments on crop growth, pests and diseases, and farming needs based on the built-in database, thereby realizing precision agricultural management. In practical applications, orchard inspection robots have significantly improved the automation and refinement of agricultural management, but their performance in complex agricultural environments still has many shortcomings.

[0003] Existing orchard inspection robots have problems in their applications, such as insufficient positioning accuracy, inaccurate path planning, poor environmental adaptability, and delayed real-time data transmission. In complex orchard environments, due to factors such as light, obstructions, and changing terrain, it is difficult for robots to stably perform simultaneous positioning and mapping (SLAM), which affects the inspection effect. In addition, traditional path planning methods have limited coverage in orchards and cannot effectively cope with the diversity of obstacles, which can easily cause damage to crops. In terms of data transmission, existing systems often lack efficient real-time feedback mechanisms, resulting in slow response speeds for environmental monitoring data and inspection decisions, limiting the effectiveness of intelligent management.

[0004] Therefore, it is urgent to design an orchard inspection robot solution that can achieve accurate path planning and has strong applicability to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent orchard inspection robot and its control system and control method, which can realize flexible and adaptive navigation through high-precision positioning and map construction, and realize accurate path planning, effectively deal with various obstacles, and realize efficient, accurate and intelligent inspection.

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

[0007] An intelligent orchard inspection robot comprises an electric crawler chassis, a box body, a display, and a data acquisition module located on the electric crawler chassis, wherein a data processing module is arranged in the box body;

[0008] The display is arranged in front of the box, and is used to display the robot status, collected environmental information and navigation path in real time; an ultrasonic generator is installed in front of the display, and is used to assist obstacle detection and distance measurement; positioning antennas are symmetrically arranged on both sides of the box, and are used to perform differential positioning of the robot;

[0009] An electric lifting push rod is installed on the box; the data acquisition module includes a binocular camera, a pan-tilt camera, and a laser radar. The binocular camera and the pan-tilt camera are both arranged on the top of the electric lifting push rod, and the laser radar is arranged in front of the electric lifting push rod, and is used to generate high-precision three-dimensional point cloud data through laser scanning to reflect the spatial structure and characteristic information of the orchard environment; the binocular camera is used to capture a stereoscopic visual image of the surrounding environment, and the pan-tilt camera performs all-round shooting through flexible rotation;

[0010] The data processing module is used to receive the images and point cloud data transmitted by the data acquisition module, and realize crop growth status monitoring and terrain perception through analysis.

[0011] The present invention also provides an intelligent orchard inspection robot control system, comprising an inspection robot, an inspection navigation subsystem, a cloud control subsystem, and an APP client; the inspection navigation subsystem is mounted on the intelligent orchard inspection robot;

[0012] The robot includes an electric crawler chassis and a data acquisition module, a data processing module, an ultrasonic generator, a positioning antenna, and a laser radar arranged on the electric crawler chassis; the data acquisition module is used to collect images and point cloud data of the surrounding environment from multiple perspectives, and transmits them to the inspection navigation subsystem and the cloud control subsystem after processing; the data processing module is used to use the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data;

[0013] The cloud control subsystem is used to receive images uploaded by the data acquisition module and process them to obtain environmental monitoring data and crop growth information, thereby generating current real-time patrol inspection information, pest identification results, equipment laser cloud images, and parameter information and operating status of the patrol robot;

[0014] The APP client is respectively connected to the cloud control subsystem and the robot for communication, and is used to customize inspection tasks according to the needs of the orchard and remotely control the operation status of the robot. The inspection tasks include inspection routes, frequencies, inspection areas, and monitoring parameters. The monitoring parameters include the execution order of inspection tasks and working hours.

[0015] The inspection and navigation subsystem is a SLAM system, which is used to pre-process the point cloud data, extract features from the pre-processed point cloud data, and use a graph optimization method for back-end mapping to obtain the robot's motion trajectory, and construct a map based on the motion trajectory to achieve autonomous inspection of the robot.

[0016] Furthermore, the SLAM system includes a visual odometer module, a backend mapping module, a mapping module, and a loop detection module;

[0017] The backend mapping module is used to optimize the cumulative error caused by the visual odometer using a graph optimization method;

[0018] The visual odometry module is used to extract features from point cloud data, estimate the displacement and rotation state of the robot in the environment, and estimate the robot's motion trajectory;

[0019] The loop detection module is based on the closed loop detection LFS process, combined with the coarse search RS and fine search SC double-layer strategy, to locate the loop frame and perform posture registration according to the point cloud data;

[0020] The mapping module is used to construct a refined map in real time according to the motion trajectory.

[0021] The present invention also provides a control method for an intelligent orchard inspection robot, which is applied to the intelligent orchard inspection robot control system and comprises the following steps:

[0022] S1. The user customizes the inspection task according to the needs of the orchard on the APP client, and remotely controls the operation status and parameter information of the robot. The inspection task includes the inspection route, frequency, inspection area and monitoring parameters. The monitoring parameters include the execution order and working time of the inspection task.

[0023] S2: After the robot is turned on, it starts to walk and collects multi-view images and point cloud data of the surrounding environment through the data acquisition module, and transmits these data to the data processing module, the inspection navigation subsystem and the cloud control subsystem;

[0024] S3, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data;

[0025] S4, the cloud control subsystem receives the images and point cloud data uploaded by the data acquisition module and processes them to obtain environmental monitoring data and crop growth information, and uses them to generate current real-time inspection information, pest identification results, and equipment laser cloud maps;

[0026] S5. The SLAM system carried by the robot preprocesses the point cloud data, extracts features from the preprocessed point cloud data, and uses a graph optimization method to perform back-end mapping to obtain the motion trajectory of the robot, constructs a map based on the motion trajectory, and realizes autonomous inspection of the robot.

[0027] Furthermore, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data, specifically including:

[0028] The three-dimensional point cloud is projected onto a two-dimensional plane, and the cross product method is used to improve the Alpha Shape algorithm to obtain the contour extraction line; the plane is divided into multiple regular grids, and the vertical distance between the highest point and the lowest point of the point cloud data in each grid is counted as the height difference. The height is multiplied by the grid area to obtain the approximate volume of a single grid. Finally, by accumulating the volumes of all grids, the volume of the object represented by the entire point cloud is obtained.

[0029] Furthermore, the three-dimensional point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is improved by a cross product method to obtain a contour extraction line, which specifically includes:

[0030] S3.1. Project the 3D point cloud onto a 2D plane to obtain 2D views of different planes and related feature information:

[0031] Unified representation method for three-dimensional plane:

[0032] ;

[0033] In the formula, a, b, c are the components of the normal vector of the plane, indicating the direction of the plane. The normal vector (a, b, c) is a vector perpendicular to the plane. x, y, z are the coordinates of any point in space that satisfies the plane equation. d is a constant term in the plane equation that is related to the position of the plane. Let , For two points on the plane, substitute into the above formula:

[0034] ;

[0035] is a vector on the plane. According to the above formula, is perpendicular to this vector, that is is the plane normal vector;

[0036] Assumptions For any point in space, the projection coordinates on the plane are , the vector formed by these two points. That is, the normal vector of the plane, then this vector should be the same as the normal vector In parallel, we get:

[0037] ;

[0038] Parametric expression of the equation of a line:

[0039] ;

[0040] Where (x0, y0, z0) are the coordinates of any point in space, and t is a scalar quantity in the direction represented by the line parameter;

[0041] Substituting the above results into the plane equation formula, we can get:

[0042] ;

[0043] Substitute the plane equation formula back into the parametric expression of the line equation to obtain the projected coordinates and read the projected point cloud;

[0044] S3.2, based on the Alpha Shape contour extraction algorithm, the calculated value of the contour boundary length is obtained, and the cross product method is used to fit the measured value and the calculated value of the contour boundary length to obtain the true contour boundary length l t , the fitting equation is as follows:

[0045] ;

[0046] In the formula, l s The contour length with large error obtained by Alpha Shape contour extraction algorithm, l t is the true contour boundary length;

[0047] Among them, the actual contour boundary length is obtained through CAD software analysis, and the analysis formula is as follows:

[0048] ;

[0049] In the formula, l scad is the length manually marked in CAD by the Alpha Shape contour extraction algorithm, l tcad is the measured value obtained by manual marking in CAD, l t is the true contour length.

[0050] Furthermore, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data, and further includes: using the SOR technology to remove outliers in the point cloud data obtained by the laser radar, and the specific steps are:

[0051] 1) Calculate the average distance from each point to all K neighboring points;

[0052] 2) Calculate the average distance and standard deviation of the entire point set;

[0053] 3) Compare the distance of each point with the preset distance threshold, and mark the points exceeding the threshold as outliers and remove them.

[0054] Furthermore, the front-end odometer data is processed, specifically including:

[0055] First, feature extraction is performed on the point cloud data, and its curvature is calculated. Points with larger curvature are identified as edge points, and points with smaller curvature are identified as plane points.

[0056] Then, by matching the edge points and plane points in the current frame with the previous frame, the displacement and rotation state of the robot in the environment are estimated, thereby realizing the construction of a three-dimensional map of the environment; this process converts the posture information in the local coordinate system into the global coordinate system to ensure the consistency of the map.

[0057] Furthermore, the method further comprises:

[0058] The RS method is used for rough search to screen out potential loop frame candidates: multiple key frame point cloud data at the current time are input into the 3D KD-Tree model, and then a certain dimension is selected for segmentation, and a tree structure is recursively constructed to find candidate frames close to the current frame;

[0059] The SC method is applied to perform a fine search on the candidate frames and locate the cyclic frames.

[0060] Furthermore, the preprocessing includes ground separation, downsampling and noise reduction on the point cloud data.

[0061] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the intelligent orchard inspection robot and its control system and control method provided by the present invention can realize high-precision positioning and map construction in a complex orchard environment by comprehensively applying multiple advanced technologies such as closed-loop detection, fuzzy control navigation, laser radar measurement and cloud data analysis, and provide a flexible and adaptive navigation control system. The virtual inspection robot of the present invention collects environmental data and crop growth information in real time, accurately measures the canopy volume of fruit trees based on laser radar technology, and uses the cloud system to perform multi-dimensional in-depth analysis and visual display of data, which greatly improves the intelligent level of orchard management. With the help of the APP client, users can achieve real-time interaction and remote control with the robot, further optimize the inspection path planning and operation strategy, improve the efficiency and accuracy of the inspection, and reduce manual intervention. Through the integration and application of these technologies, the present invention significantly improves the automation, precision and intelligence level of orchard inspection, and provides an efficient and reliable solution for the management of agricultural sites such as orchards. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0063] Figure 1 This is a schematic diagram of the structure of the intelligent orchard inspection robot of the present invention;

[0064] Figure 2 A specific workflow diagram of the improved LFS method according to an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of a rough RS search according to an embodiment of the present invention;

[0066] Figure 4 This is a flow chart of SC fine search according to an embodiment of the present invention;

[0067] Figure 5 This is a diagram of an orchard SLAM system based on improved LD according to an embodiment of the present invention;

[0068] Figure 6 This is a workflow diagram of the front-end odometer according to an embodiment of the present invention;

[0069] Figure 7 An optimization method for backend mapping according to an embodiment of the present invention;

[0070] Figure 8 A three-dimensional view of the fuzzy control rules of an embodiment of the present invention;

[0071] Fig. 9 This is a plane projection diagram of a three-dimensional point cloud according to an embodiment of the present invention, wherein a is a projection in the X-axis direction; b is a projection in the Y-axis direction; and c is a projection in the Z-axis direction;

[0072] Fig.10 A schematic diagram of an improved Alpha Shape contour extraction algorithm according to an embodiment of the present invention;

[0073] Fig.11 This is a schematic diagram of volume calculation by grid integration according to an embodiment of the present invention;

[0074] Explanation of the reference numerals: 1. Electric crawler chassis; 2. Display; 3. Ultrasonic generator; 4. Positioning antenna; 5. Laser radar; 6. Electric lifting push rod; 7. Binocular camera; 8. PTZ camera; 9. Box. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] The purpose of the present invention is to provide an intelligent orchard inspection robot and its control system and control method, which can realize flexible and adaptive navigation through high-precision positioning and map construction, and realize accurate path planning, effectively deal with various obstacles, and realize efficient, accurate and intelligent inspection.

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0078] Example 1

[0079] like Figure 1 As shown, the intelligent orchard inspection robot provided by the embodiment of the present invention adopts an electric crawler chassis 1 as a basic platform, and integrates multiple sensing devices, navigation systems, visual devices, voice interaction systems, face recognition and object detection devices thereon to realize intelligent inspection and monitoring functions.

[0080] In this embodiment, it includes an electric crawler chassis 1, a box 9 located on the electric crawler chassis 1, a display 2, and a data acquisition module, wherein a data processing module is disposed in the box 9;

[0081] The display 2 is arranged in front of the box body, and is used to display the robot status, collected environmental information, navigation path and other data in real time, so as to facilitate the operator to monitor and debug; an ultrasonic generator 3 is installed in front of the display 2, which is mainly used to assist obstacle detection and distance measurement, and ensure that the robot can operate safely in a complex environment; positioning antennas 4 are symmetrically arranged on both sides of the box body 9, which are used for differential positioning of the robot, providing navigation support through accurate positioning signals, and ensuring the stability and accuracy of the robot's inspection path;

[0082] An electric lifting push rod 6 is installed on the box body to adjust the height to adapt to different environmental requirements; the data acquisition module includes a binocular camera 7, a pan-tilt camera 8, and a laser radar 5. The binocular camera 7 and the pan-tilt camera 8 are both arranged on the top of the electric lifting push rod 6, and the laser radar 5 is arranged in front of the electric lifting push rod 6, and is used to generate high-precision three-dimensional point cloud data through laser scanning to reflect the spatial structure and characteristic information of the orchard environment. Specifically, the laser radar is used to continuously rotate and scan within a certain angle range through a laser beam during the driving process of the inspection robot in the orchard, and calculate the precise distance between the laser beam and the target by emitting laser pulses and receiving reflected signals, and generate a high-resolution three-dimensional point cloud model in combination with the scanning angle data to fully reflect the spatial structure and characteristic information of the orchard environment; the binocular camera 7 is used to capture stereoscopic visual images of the surrounding environment for crop growth status monitoring and terrain perception; the pan-tilt camera 8 performs all-round shooting through flexible rotation to achieve all-round environmental monitoring and support functions such as pest and disease identification;

[0083] The data processing module is used to receive the images and point cloud data transmitted by the data acquisition module, and realize crop growth status monitoring and terrain perception through analysis.

[0084] The overall design integrates a variety of sensor and visual devices to form an integrated intelligent detection system, which effectively improves the robot's adaptability and operating efficiency in complex orchard environments. This structural design effectively integrates various sensor and visual devices, enhancing the inspection robot's adaptability and working efficiency in complex orchard environments.

[0085] Example 2

[0086] like Figure 1-Figure 11 As shown, based on the intelligent orchard inspection robot described in Example 1, the intelligent orchard inspection robot control system provided by the embodiment of the present invention includes the inspection robot, the inspection navigation subsystem, the cloud control subsystem, and the APP client described in Example 1; the inspection navigation subsystem is mounted on the intelligent orchard inspection robot;

[0087] The robot includes an electric crawler chassis and a data acquisition module, a data processing module, an ultrasonic generator 3, a positioning antenna 4, and a laser radar 5 arranged on the electric crawler chassis; the data acquisition module is used to collect images and point cloud data of the surrounding environment from multiple perspectives, and transmit them to the inspection navigation subsystem and the cloud control subsystem after processing; the data processing module is used to use the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data;

[0088] The cloud control subsystem is used to receive images uploaded by the data acquisition module and process them to obtain environmental monitoring data and crop growth information, thereby generating current real-time inspection information, pest identification results, equipment laser cloud images, and parameter information and operating status of the inspection robot;

[0089] The APP client is respectively connected to the cloud control subsystem and the robot for communication, and is used to customize inspection tasks according to the needs of the orchard and remotely control the operation status of the robot. The inspection tasks include inspection routes, frequencies, inspection areas, and monitoring parameters. The monitoring parameters include the execution order of inspection tasks and working hours.

[0090] The inspection and navigation subsystem is a SLAM system, which is used to pre-process the point cloud data, extract features from the pre-processed point cloud data, and use a graph optimization method for back-end mapping to obtain the robot's motion trajectory, and construct a map based on the motion trajectory to achieve autonomous inspection of the robot.

[0091] In this embodiment, the robot further includes an ultrasonic generator 3, a positioning antenna 4, a laser radar 5, and an electric lifting push rod 6; the ultrasonic generator 3 is installed on the front side of the display 2 to assist obstacle detection and distance measurement; the data acquisition module is installed on the electric lifting push rod 6 to collect images and perceive the environment through multiple perspectives of the surrounding environment; the positioning antenna 4 is arranged on both sides of the display 2 to provide accurate positioning and navigation support; the laser radar 5 is arranged on the front side of the electric lifting push rod 6 to accurately scan the surrounding environment and generate three-dimensional point cloud data;

[0092] The data acquisition module includes a binocular camera 7 and a pan-tilt camera 8.

[0093] In this embodiment, the SLAM system includes a visual odometer module, a backend mapping module, a mapping module, and a loop detection module;

[0094] The backend mapping module is used to optimize the cumulative error caused by the visual odometer using a graph optimization method;

[0095] The visual odometry module is used to extract features from point cloud data, estimate the displacement and rotation state of the robot in the environment, and estimate the robot's motion trajectory;

[0096] The loop detection module is based on the closed loop detection LFS process, combined with the coarse search RS and fine search SC double-layer strategy, to locate the loop frame and perform posture registration according to the point cloud data;

[0097] The mapping module is used to construct a refined map in real time according to the motion trajectory.

[0098] like Figure 5 As shown in the figure of the orchard SLAM system based on the improved LD, the improved LFS method improves the robustness and quality of local feature matching by optimizing the feature extraction and screening strategies and combining multimodal data fusion (image features and lidar point cloud), thereby providing more reliable input data for the LD method. On this basis, the LD method combines the optimized feature matching strategy with a more robust similarity measurement algorithm to effectively identify loop areas and correct pose deviations. The two are significantly complementary in workflow: the LFS method provides high-quality local feature data for the LD method, and the LD method achieves global consistency correction of the map by identifying loop areas, while reversely optimizing the feature distribution extracted by LFS, thus forming a virtuous circle. The overall improvement strategy significantly improves the positioning accuracy and structure accuracy of the SLAM system in complex orchard environments. Figure 1 This has laid an important foundation for the efficient and stable autonomous navigation of the intelligent orchard inspection robot.

[0099] The orchard SLAM system improves data quality by preprocessing, ground separation, downsampling and noise reduction of LiDAR point cloud data. The orchard SLAM system uses feature extraction for front-end range processing to reduce cumulative errors, and improves map accuracy through back-end map optimization. Ultimately, real-time map construction and accurate posture output are achieved, providing reliable support for the autonomous inspection of the intelligent orchard inspection robot.

[0100] In this embodiment, the design of the inspection robot system combines cloud control, intelligent algorithms and flexible interactive communication functions. The specific design includes real-time data upload and cloud control system, and APP client remote interaction and management, which greatly improves the intelligence and automation level of orchard inspection.

[0101] The front-end development of the remote control system uses the Vue 3.2 framework, uses TypeScript to write code, combines Vite as a building tool, and uses Element Plus as a UI component library. Routing management uses Vue Router Next, state management is implemented through Pinia, and Composition API setup syntax sugar is used to improve code readability and reusability. Node.js uses version 16.

[0102] Backend development is based on the Java programming language, using the SpringBoot framework to build applications, SpringSecurity to implement authentication and authorization, and Spring Data JPA for database access. Project construction and dependency management are completed through Maven, API documentation is generated using Swagger, and Kafka is used as the message queue to process asynchronous messages and events.

[0103] Database: MySQL 8.0 is used, HikariCP is used to manage the database connection pool, and Flyway is used for database migration and version control.

[0104] Version control: Use Git for version control and host on GitHub for team collaboration and code management.

[0105] Development tools: Use IntelliJ IDEA 2023.1.1 for back-end development and Visual Studio Code for front-end development.

[0106] The unmanned inspection robot data cloud control can view the detailed parameter information and operating status of the inspection robot in detail. Through online map embedding, users can locate the position of the inspection robot in real time and view its panoramic information and online perception data in real time. The cloud control system supports online viewing of current real-time inspection information, pest identification results and equipment laser cloud map.

[0107] Example 3

[0108] The embodiment of the present invention provides a control method for an intelligent orchard inspection robot, which is applied to the intelligent orchard inspection robot described in Example 1 and the intelligent orchard inspection robot control system described in Example 2, and includes the following steps:

[0109] S1. The user customizes the inspection task according to the needs of the orchard on the APP client, and remotely controls the operation status and parameter information of the robot. The inspection task includes the inspection route, frequency, inspection area and monitoring parameters. The monitoring parameters include the execution order and working time of the inspection task.

[0110] S2: After the robot is turned on, it starts to walk and collects multi-view images and point cloud data of the surrounding environment through the data acquisition module, and transmits these data to the data processing module, the inspection navigation subsystem and the cloud control subsystem;

[0111] S3, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data;

[0112] S4, the cloud control subsystem receives the images and point cloud data uploaded by the data acquisition module and processes them to obtain environmental monitoring data and crop growth information, and generates current real-time inspection information, pest identification results, and equipment laser cloud map;

[0113] The environmental monitoring information includes temperature, humidity, wind speed, light, etc.; the crop growth information includes seedling information (such as flowering period, leaf fall period, fruiting period), tree height, canopy volume, and fruit quantity; the inspection information includes plot location, path trajectory, crop type, temperature and humidity, crop height and canopy volume, etc. The pest and disease identification results include real-time video and the display of pest and disease identification results based on preset models, such as pest and disease category, pest and disease identification rate, disease level, etc.;

[0114] S5. The SLAM system carried by the robot preprocesses the point cloud data, extracts features from the preprocessed point cloud data, and uses a graph optimization method to perform back-end mapping to obtain the motion trajectory of the robot, constructs a map based on the motion trajectory, and realizes autonomous inspection of the robot.

[0115] In this embodiment, the closed loop detection (LFS) method is improved, and the inspection and positioning capability of the SLAM system in a complex orchard environment is effectively improved by combining the coarse search (RS) and fine search (SC) double-layer strategy. By combining multiple sensors such as laser radar, stereo vision camera, inertial measurement unit, ultrasonic sensor and infrared sensor, the inspection robot can achieve real-time acquisition and deep fusion of multimodal information; through the fusion of the above multiple sensors and the combination of multi-view data, the inspection robot can enhance its adaptability in a complex orchard environment while building a refined map in real time, significantly improving the operation efficiency and inspection accuracy, and providing strong support for the development of smart agriculture.

[0116] like Figure 2 As shown in the figure, the specific workflow diagram of the improved LFS search method is used. The improved LFS method of the present invention is started by a multi-level LD (closed loop detection) mechanism. First, a rough search is performed in the RS stage to quickly screen out potential cyclic frame candidates, and then the SC fine search is applied on this basis to accurately locate the cyclic frame. In order to further improve the reliability of the results, the system also performs multimodal verification to ensure that the detected ring frames are credible. After completing the rough search and fine search, the system uses the NDT (normal distribution transformation) algorithm to perform preliminary posture alignment, generate a preliminary posture transformation matrix, and use it as the initial value of the ICP (iterative closest point) fine alignment to further optimize the matching accuracy and finally complete the matching of the cyclic frame.

[0117] In this embodiment, Figure 6The figure shows the workflow of the front-end odometer. During the front-end odometer calculation process, the system first extracts features from the point cloud data. Points with larger curvature are identified as edge points, and points with smaller curvature are plane points. By matching edge points and plane points in the current frame with those in the previous frame, the system estimates the displacement and rotation state of the robot in the environment, and then realizes the construction of a three-dimensional map of the environment. This process converts the posture information in the local coordinate system to the global coordinate system to ensure the consistency of the map.

[0118] like Figure 7 The figure shows the optimization method of the back-end mapping. The back-end optimization uses a graph optimization method, with keyframes as graph nodes, and the edges between nodes represent the constraint relationship between adjacent postures. The goal is to minimize the error between each node to obtain the optimal posture estimation. During the optimization process, the nearest keyframe is quickly found through the KD-Tree structure, and the LM algorithm is used to optimize the feature points in the local posture graph to reduce the error and improve the accuracy of posture estimation. The system incorporates the loop information provided by the LD system into the graph optimization, and achieves the accuracy of posture estimation by minimizing the error function of each posture node. The entire process of graph optimization is achieved by minimizing the error function between each posture node:

[0119] .

[0120] Where E is the error function, which represents the key frame g i and g j The constraint error between ig represents the key frame g actually observed i and g j The relative pose between them. The predicted value h(gi,gj) represents the relative pose between the current key frame g i and g j The relative position calculated by the posture. T is the transpose symbol, which represents the transpose of the matrix and is used to construct the quadratic error. Ωij is the information matrix, which represents the confidence or accuracy of the measurement value, Z ij -h(gi-gj) is the residual, which represents the difference between the measured value and the predicted value.

[0121] The objective function for optimizing the entire graph can be expressed as the sum of the errors of all adjacent nodes, as shown in the following formula:

[0122] .

[0123] Where F is the objective function, which represents the sum of constraint errors between all key frames in the entire graph, and min means that the goal of the optimization process is to minimize the value of F, that is, to reduce the error between all adjacent nodes in the graph. i} represents the pose of all key frames in the graph, and the edge set ξ in (i,j)∈ξ represents the set of all adjacent node pairs (i,j). E(gi,gj) represents the pose of key frame g i and g j The constraint error between .

[0124] This embodiment proposes a multi-perspective combined display method to enhance the display effect of the SLAM generated map. The system provides a two-dimensional plane perspective and a three-dimensional perspective, which is convenient for users to comprehensively observe the relationship between the robot's motion trajectory and the environment from different angles. The two-dimensional perspective is suitable for quickly understanding the overall environment layout, and the three-dimensional perspective can present the spatial structure details of complex environments, such as tree height and terrain undulations. The multi-perspective display method helps operators to plan and make decisions on paths in complex terrain, further improving the reliability of robot inspection and positioning.

[0125] In this embodiment, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data, specifically including:

[0126] like Fig.11 The figure shows a schematic diagram of volume calculation by grid integration. The three-dimensional point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is improved by the cross product method to obtain the contour extraction line. The plane is divided into multiple regular grids, and the vertical distance between the highest point and the lowest point of the point cloud data in each grid is the height difference. The height is multiplied by the grid area to obtain the approximate volume of a single grid. Finally, the volume of the object represented by the entire point cloud is obtained by accumulating the volumes of all grids.

[0127] In this embodiment, an improved Alpha Shape algorithm is used in combination with a grid integration method to accurately calculate the canopy volume of fruit trees, and a machine learning algorithm is used to build a prediction model to further improve the accuracy of the measurement. In this embodiment, the data processing module uses the Alpha Shape algorithm in combination with a grid integration method to accurately calculate the canopy volume of fruit trees based on the point cloud data. It also includes: using SOR technology to remove outliers in the point cloud data acquired by the lidar: performing a statistical analysis on the neighborhood of each point, calculating the average distance to all neighboring points, assuming that the result is a Gaussian distribution, and the shape of the distribution is determined by the mean and standard deviation, then points outside the standard deviation range can be defined as outliers and removed from the point cloud data. The specific steps are as follows:

[0128] 1) Calculate the average distance from each point to all K neighboring points;

[0129] 2) Calculate the average distance and standard deviation of the entire point set;

[0130] 3) Compare the distance of each point with the preset distance threshold, and mark the points exceeding the threshold as outliers and remove them.

[0131] For a plane point cloud of any shape, if a circle with a radius of a is rolled around it, the points formed by its rolling trajectory are contour points. The alpha shapes algorithm can only detect the contour points of two-dimensional points. Therefore, when extracting the contour point straight line of the point cloud, it is necessary to project the points on a two-dimensional plane, generally the XOY plane. In this embodiment, the three-dimensional point cloud is projected onto a two-dimensional plane, and the cross product method is used to improve the Alpha Shape algorithm to obtain the contour extraction line, which specifically includes:

[0132] S3.1. Project the 3D point cloud onto a 2D plane to obtain 2D views of different planes and related feature information:

[0133] A 3D point cloud is composed of a bunch of disordered points in 3D space. In 3D, it can be found on any 2D plane. Projection is performed in the unified representation method of three-dimensional plane:

[0134] .

[0135] In the formula, a, b, c are the components of the normal vector of the plane, indicating the direction of the plane. The normal vector (a, b, c) is a vector perpendicular to the plane. x, y, z are the coordinates of any point in space that satisfies the plane equation. d is a constant term in the plane equation that is related to the position of the plane. Let , For two points on the plane, substitute into the above formula:

[0136] .

[0137] is a vector on the plane. According to the above formula, is perpendicular to this vector, that is is the plane normal vector;

[0138] Assumptions For any point in space, the projection coordinates on the plane are , the vector formed by these two points; that is, the normal vector of the plane, then this vector should be equal to the normal vector In parallel, we get:

[0139] .

[0140] Parametric expression of the equation of a line:

[0141] .

[0142] Where (x0, y0, z0) are the coordinates of any point in space, and t is a scalar quantity representing the direction of the line parameter;

[0143] Substituting the above results into the plane equation formula, we can get:

[0144] .

[0145] Substitute the plane equation formula back into the parametric expression of the line equation to obtain the projected coordinates and read the projected point cloud;

[0146] S3.2, based on the Alpha Shape contour extraction algorithm, the calculated value of the contour boundary length is obtained, and the cross product method is used to fit the measured value and the calculated value of the contour boundary length to obtain the true contour boundary length l t , the fitting equation is as follows:

[0147] .

[0148] In the formula, l s The contour length with large error obtained by Alpha Shape contour extraction algorithm, l t is the true contour boundary length;

[0149] Among them, the actual contour boundary length is obtained through CAD software analysis, and the analysis formula is as follows:

[0150] .

[0151] In the formula, l scad is the length manually marked in CAD by the Alpha Shape contour extraction algorithm, l tcad is the measured value obtained by manual marking in CAD, l t is the true contour length; l scad and tcad like Fig.10 As shown, the fitting l in SPSS software t and l s .

[0152] In this embodiment, the processing of the front-end odometer data specifically includes:

[0153] First, feature extraction is performed on the point cloud data, and its curvature is calculated. Points with larger curvature are identified as edge points, and points with smaller curvature are identified as plane points.

[0154] Then, by matching the edge points and plane points in the current frame with the previous frame, the displacement and rotation state of the robot in the environment are estimated, thereby realizing the construction of a three-dimensional map of the environment; this process converts the posture information in the local coordinate system into the global coordinate system to ensure the consistency of the map.

[0155] The inspection robot navigation system described in this embodiment applies an intelligent inspection navigation control method based on fuzzy control. Fuzzy control has good robustness and can effectively respond to environmental changes or external interference. The fuzzy control system is mainly divided into the following three steps: The fuzzy controller design of the inspection robot navigation control method based on fuzzy control mainly includes:

[0156] (1) Fuzzification of input and output quantities

[0157] The input lateral deviation d and heading deviation θ are converted into fuzzy quantities, and the output is the expected turning angle δ. The present invention introduces an adaptive parameter quantization factor and combines multimodal sensor data to adjust the fuzzy values ​​of lateral deviation and heading deviation in real time. The specific configuration is as follows: lateral deviation d / cm

[0158] Horizontal deviation d: basic domain is [−30,30], quantization level is {-6,...,6}, quantization factor K d =0.2;

[0159] Heading deviation θ: The basic domain is [−10,10], the quantization level is {-6,...,6}, and the quantization factor is K θ =0.6;

[0160] Expected rotation angle δ: basic domain is [−6,6], quantization level is {-6,...,6}, scaling factor K δ =1.

[0161] In the navigation system of the inspection robot of the present invention, the direction and symbol of each fuzzy variable are defined to facilitate description and control.

[0162] Lateral deviation d: The deviation on the right side of the navigation path is positive, and the deviation on the left side is negative. That is, when the robot deviates from the navigation path to the right, the lateral deviation d is positive; when it deviates to the left, the lateral deviation d is negative.

[0163] Heading deviation θ: Heading deviation is defined as the angular deviation relative to the navigation forward direction. Clockwise deviation is positive, and counterclockwise deviation is negative. Therefore, when the robot deviates from the forward direction to the right, the heading deviation θ is positive; when it deviates to the left, the heading deviation θ is negative.

[0164] Expected rotation angle δ: The expected rotation angle is defined as the angle at which the robot corrects its heading. The expected rotation angle for turning right is positive, and the expected rotation angle for turning left is negative.

[0165] According to the direction and size relationship of these deviations, fuzzy sets are defined to describe the degree of these deviations and angles. Each set of fuzzy variables includes the following seven fuzzy levels: negative big (NB) - the deviation or angle is large and negative; negative medium (NM) - the deviation or angle is moderately negative; negative small (NS) - the deviation or angle is small and negative; zero (ZO) - the deviation or angle is close to zero; positive small (PS) - the deviation or angle is small and positive; positive medium (PM) - the deviation or angle is moderately positive; positive big (PB) - the deviation or angle is large and positive.

[0166] (2) Establishing fuzzy control rules

[0167] The design of fuzzy control rules is based on human driving experience in order to optimize the robot's navigation decisions under various deviation conditions. When both the lateral deviation d and the heading deviation θ are large, the expected turning angle δ needs to be significantly adjusted for rapid deviation correction, that is, the expected turning angle should also be set to a large positive value (PB) to correct the deviation as quickly as possible. It is expressed as a fuzzy conditional statement: if d=NB andθ=NB, then δ=PB. Through a similar analysis method, all possible deviation combinations are derived, and finally 49 fuzzy control rules are established. These rules cover different deviation combinations to achieve a comprehensive control effect. Each rule specifically defines the adjustment direction and amplitude of the expected turning angle δ under a specific lateral deviation d and heading deviation θ.

[0168] like Figure 8 As shown in the figure, it is a three-dimensional view of the fuzzy control rules obtained according to the above fuzzy control rules. Figure 7 As shown. It shows the trend of the expected turning angle δ with the lateral deviation d and the heading deviation θ. In this three-dimensional view, each control rule corresponds to a specific deviation combination, and the height of the graph reflects the size and direction of the expected turning angle under this combination. This three-dimensional view helps to intuitively display the effect of the control strategy under various deviation conditions. Especially in areas where the lateral deviation and heading deviation are large, the adjustment of the expected turning angle is also more significant to speed up the error correction; when the deviation is small, the adjustment range is relatively gentle, which is conducive to smooth navigation.

[0169] (3) Defuzzification and establishment of fuzzy query table

[0170] The improved defuzzification method is used to combine the traditional area centroid method with the fuzzy neural network. The fuzzy output is fine-tuned through the learning function of the neural network to achieve dynamic calculation of the desired turning angle δ. For example, when the input lateral deviation d=5 and the heading deviation θ=5, the output desired turning angle δ=−5.25, and the actual control is obtained by multiplying the proportional factor.

[0171] In this embodiment, the method further includes:

[0172] The RS method is used for rough search to screen out potential loop frame candidates: multiple key frame point cloud data at the current time are input into the 3D KD-Tree model, and then a certain dimension is selected for segmentation, and a tree structure is recursively constructed to find candidate frames close to the current frame;

[0173] Apply the SC method to perform a fine search on candidate frames and locate the cyclic frames;

[0174] The details are as follows:

[0175] like Figure 3 The figure shows a schematic diagram of RS rough search. This embodiment uses the efficient data structure of multi-dimensional spatial data - k-dimensional tree (KD-Tree) to accelerate the execution of LFS, which can significantly improve the efficiency of point cloud data query. In LFS, KD-Tree recursively divides the space into sub-areas to accelerate the search process. When constructing KD-Tree, multiple key frame point cloud data at the current time are input into the 3D KD-Tree model, and then a certain dimension (such as X, Y or Z) is selected for segmentation, and the tree structure is recursively constructed to quickly find candidate frames close to the current frame. This method improves the search efficiency and ensures the real-time and accuracy of closed-loop detection. In the RS rough search stage, the present invention defines the search range by setting a fixed search radius, and searches for potential cyclic frames within this range. RS rough search calculates the Euclidean distance between the pose of the current frame and the previous frame. When the distance is less than the preset search radius, it is considered that they may form a closed loop, that is, a potential cyclic frame. During the search process, this study calculates the Euclidean distance between the central node of the current frame and the candidate node:

[0176] .

[0177] Where d is the Euclidean distance between the center node and the candidate node in the current coordinate system, (x0, y0, z0) is the coordinate of the center node in the current frame, (x i ,y i , z i ) are the coordinates of the candidate nodes. When the Euclidean distance satisfies the condition of being less than the search radius, the candidate nodes are regarded as potential cycle frames within the specified search radius, and these nodes will be retained and further processed. The efficient query capability of KD-Tree can improve the efficiency of RS coarse search. Compared with directly traversing all point cloud data, KD-Tree can reduce the complexity of the search by recursively partitioning the space. When processing large-scale point cloud data in an orchard, the calculation and search time can be significantly reduced.

[0178] like Figure 4As shown in the figure, it is a flowchart of SC fine search. The present invention further uses SC to perform a more detailed search on the candidate loop frames, and realizes the accurate matching of the loop frames by constructing a global descriptor. SC fine search divides the input single-frame point cloud data into radial areas and circumferential areas. The radial division divides the point cloud data into several concentric circular areas, and each radial area is further divided into multiple sectors, thereby forming a two-dimensional matrix structure to represent the entire point cloud data. In order to extract meaningful features in each area, the SC method uses the maximum height value (MHV) as the feature value of each area. The MHV feature value in each intersection area (i.e., each sector) is:

[0179] .

[0180] Where H ij Indicates the maximum height of the area, z(p) is the height of point p, P ij is the specified area point set.

[0181] The MHV values ​​are combined into a matrix, where the rows and columns represent the number of radial and circumferential regions, respectively. This matrix is ​​used as a global feature of the current frame for similarity matching with other candidate frames. In order to compare the similarity between different frames, the SC method uses cosine distance as a metric. The following is used to calculate the cosine distance between each sector vector:

[0182] .

[0183] In the formula, D(G q ,G c ) is region G q and G c The cosine distance, N s is the dimension of the feature vector, For region G q and G c The eigenvalues ​​in the jth dimension, ||cq j|| ||cc j|| are the region G q and G c The length of the eigenvector of .

[0184] The smaller the cosine distance between two point cloud frames, the higher the similarity. In order to solve the problem that the sector features may be offset due to the change of viewing angle of the laser radar, the SC method will traverse all possible sector offsets and select the offset that minimizes the cosine distance as the final matching result. As shown in the following formula:

[0185] .

[0186] Where Δ is the set of all possible offsets, δ is the offset, Indicates the offset corresponding to the minimum cosine distance. The final matching distance is

[0187] .

[0188] In the formula, D(G q ,G c ) is region G q and G c The minimum matching distance, N s is the set of all dimensions in the feature space, D n (G q ,G c ) is the matching distance between two regions in the nth dimension.

[0189] In order to further reduce the computational complexity, the SC method introduces a loop key to perform a two-stage hierarchical search. The specific steps are as follows:

[0190] (1) Ring key selection: First, the candidate frame with the most similar features to the frame to be matched is selected by using the ring key. The role of the ring key is to help quickly screen out the potential most suitable candidate frames, thereby reducing the amount of calculation.

[0191] (2) Screening valid candidate frames: When performing the final screening among candidate frames, the distance between each candidate frame and the frame to be matched is calculated and compared with the set threshold. If the distance of the candidate frame is less than or equal to the threshold, the candidate frame is considered to be a valid ring frame.

[0192] (3) Select the best matching frame: Select the frame with the smallest distance from the screened candidate frames as the final loop frame and as the matching result of the subsequent LFS.

[0193] (4) NDT-ICP combination method: Combining the advantages of both, the NDT algorithm is first used for initial registration to obtain a rough pose transformation matrix. Subsequently, the ICP algorithm refines and optimizes the initial registration result to obtain a more accurate pose transformation matrix. Through this combination, the high efficiency of the initial registration can be ensured, and high-precision matching results can be obtained through ICP fine-tuning, thereby improving the overall accuracy and robustness of loopback frame matching.

[0194] The APP client in this embodiment allows the user to have real-time two-way interaction with the inspection robot. Through the APP on devices such as mobile phones or tablets, the robot's operating status can be controlled, and inspection tasks can be remotely scheduled, paused, started, or the robot's operation can be adjusted. The APP client supports users to plan and set inspection tasks on the interface, and can customize the robot's inspection route, frequency, inspection area, and specific monitoring parameters according to the different needs of the orchard. According to user needs, the APP client supports personalized operation strategy settings, and adjusts the robot's inspection task execution order, path selection, working hours and other parameters through the APP to customize strategies that meet the actual needs of the orchard.

[0195] In this embodiment, the preprocessing includes ground separation, downsampling and noise reduction on the point cloud data.

[0196] The remaining technical features in this embodiment can be flexibly selected by those skilled in the art according to actual conditions to meet different specific practical needs. However, it is obvious to those skilled in the art that it is not necessary to adopt these specific details to implement the present invention. In other examples, in order to avoid confusing the present invention, the well-known components, structures or parts are not specifically described, which are all within the technical protection scope defined by the technical solution claimed for protection in the claims of the present invention.

[0197] Modifications and changes made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the scope of protection of the claims attached to the present invention. In the above description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is obvious to those of ordinary skill in the art that these specific details are not necessary to practice the present invention. In other examples, in order to avoid confusing the present invention, well-known technologies, such as specific construction details, operating conditions and other technical conditions, are not specifically described.

[0198] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An intelligent orchard inspection robot, characterized in that: It includes an electric crawler chassis, a box, a display, and a data acquisition module located on the electric crawler chassis, wherein a data processing module is arranged in the box; The display is arranged in front of the box, and is used to display the robot status, collected environmental information and navigation path in real time; an ultrasonic generator is installed in front of the display, and is used to assist obstacle detection and distance measurement; positioning antennas are symmetrically arranged on both sides of the box, and are used to perform differential positioning of the robot; An electric lifting push rod is installed on the box; the data acquisition module includes a binocular camera, a pan-tilt camera, and a laser radar, the binocular camera and the pan-tilt camera are both arranged on the top of the electric lifting push rod, and the laser radar is arranged in front of the electric lifting push rod, and is used to generate high-precision three-dimensional point cloud data through laser scanning to reflect the spatial structure and characteristic information of the orchard environment; The binocular camera is used to capture stereoscopic images of the surrounding environment, and the pan-tilt camera performs omnidirectional photography by flexibly rotating; The data processing module is used to receive the images and point cloud data transmitted by the data acquisition module, and realize crop growth status monitoring and terrain perception through analysis.

2. An intelligent orchard inspection robot control system, characterized in that: The intelligent orchard inspection robot comprises the intelligent orchard inspection robot according to claim 1, and further comprises an inspection navigation subsystem, a cloud control subsystem, and an APP client; the inspection navigation subsystem is mounted on the intelligent orchard inspection robot; The data acquisition module on the intelligent orchard inspection robot is used to collect multi-perspective images and point cloud data of the surrounding environment, and transmit them to the inspection navigation subsystem and the cloud control subsystem after processing; the data processing module is used to use the AlphaShape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data; The cloud control subsystem is used to receive images uploaded by the data acquisition module and process them to obtain environmental monitoring data and crop growth information, thereby generating current real-time inspection information, pest identification results, equipment laser cloud images, and parameter information and operating status of the inspection robot; The APP client is respectively connected to the cloud control subsystem and the robot for communication, and is used to customize inspection tasks according to the needs of the orchard and remotely control the running status of the inspection robot. The inspection tasks include inspection routes, frequencies, inspection areas, and monitoring parameters. The monitoring parameters include the execution order of inspection tasks and working hours. The inspection and navigation subsystem is a SLAM system, which is used to pre-process the point cloud data, extract features from the pre-processed point cloud data, and use a graph optimization method for back-end mapping to obtain the robot's motion trajectory, and construct a map based on the motion trajectory to achieve autonomous inspection of the robot.

3. The intelligent orchard inspection robot control system according to claim 2 is characterized in that: The SLAM system includes a visual odometer module, a backend mapping module, a mapping module, and a loop detection module; The backend mapping module is used to optimize the cumulative error caused by the visual odometer using a graph optimization method; The visual odometry module is used to extract features from point cloud data, estimate the displacement and rotation state of the robot in the environment, and estimate the robot's motion trajectory; The loop detection module is based on the closed loop detection LFS process, combined with the coarse search RS and fine search SC double-layer strategy, to locate the loop frame and perform posture registration according to the point cloud data; The mapping module is used to construct a refined map in real time according to the motion trajectory.

4. A control method for an intelligent orchard inspection robot, applied to the intelligent orchard inspection robot control system according to any one of claims 2-3, characterized in that: The following steps are involved: S1. The user customizes the inspection task according to the needs of the orchard on the APP client, and remotely controls the operation status and parameter information of the robot. The inspection task includes the inspection route, frequency, inspection area and monitoring parameters. The monitoring parameters include the execution order and working time of the inspection task. S2: After the robot is turned on, it starts to walk and collects multi-view images and point cloud data of the surrounding environment through the data acquisition module, and transmits these data to the data processing module, the inspection navigation subsystem and the cloud control subsystem; S3, the data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data; S4, the cloud control subsystem receives the images and point cloud data uploaded by the data acquisition module and processes them to obtain environmental monitoring data and crop growth information, and generates current real-time inspection information, pest identification results, and equipment laser cloud map; S5. The SLAM system carried by the robot preprocesses the point cloud data, extracts features from the preprocessed point cloud data, and uses a graph optimization method to perform back-end mapping to obtain the motion trajectory of the robot, constructs a map based on the motion trajectory, and realizes autonomous inspection of the robot.

5. The intelligent orchard inspection robot control method according to claim 4 is characterized in that: The data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data, specifically including: The three-dimensional point cloud is projected onto a two-dimensional plane, and the cross product method is used to improve the Alpha Shape algorithm to obtain the contour extraction line; the plane is divided into multiple regular grids, and the vertical distance between the highest point and the lowest point of the point cloud data in each grid is counted as the height difference. The height is multiplied by the grid area to obtain the approximate volume of a single grid. Finally, by accumulating the volumes of all grids, the volume of the object represented by the entire point cloud is obtained.

6. The intelligent orchard inspection robot control method according to claim 5, characterized in that: The three-dimensional point cloud is projected onto a two-dimensional plane, and the Alpha Shape algorithm is improved by a cross product method to obtain a contour extraction line, specifically including: S3.

1. Project the 3D point cloud onto a 2D plane to obtain 2D views of different planes and related feature information: Unified representation method for three-dimensional plane: ； Where a, b, and c are the components of the normal vector of the plane, indicating the direction of the plane; (x, y, z) are the coordinates of any point in space that satisfies the plane equation; d is the constant term in the plane equation, which is related to the position of the plane; set up , For two points on the plane, substitute into the above formula: ； is a vector on the plane. According to the above formula, is perpendicular to this vector, that is is the plane normal vector; Assumptions For any point in space, the projection coordinates on the plane are , the vector formed by these two points is the normal vector of the plane, then this vector should be In parallel, we get: ； Parametric expression of the equation of a line: ； Where (x0, y0, z0) are the coordinates of any point in space, and t is a scalar quantity in the direction represented by the line parameter; Substituting the above results into the plane equation formula, we get: ； Substitute the plane equation formula back into the parametric expression of the line equation to obtain the projected coordinates and read the projected point cloud; S3.2, based on the Alpha Shape contour extraction algorithm, the calculated value of the contour boundary length is obtained, and the cross product method is used to fit the measured value and the calculated value of the contour boundary length to obtain the true contour boundary length l t , the fitting equation is as follows: ； In the formula, l s The contour length obtained by the Alpha Shape contour extraction algorithm has a large error, l t is the true contour boundary length; Among them, the actual contour boundary length is obtained through CAD software analysis, and the analysis formula is as follows: ； In the formula, l scad is the length manually marked in CAD by the Alpha Shape contour extraction algorithm, l tcad is the measured value obtained by manual marking in CAD, l t is the true contour length.

7. The intelligent orchard inspection robot control method according to claim 6, characterized in that: The data processing module uses the Alpha Shape algorithm combined with the grid integration method to accurately calculate the canopy volume of the fruit tree according to the point cloud data, and also includes: using the SOR technology to remove outliers in the point cloud data obtained by the laser radar, and the specific steps are: 1) Calculate the average distance from each point to all K neighboring points; 2) Calculate the average distance and standard deviation of the entire point set; 3) Compare the distance of each point with the preset distance threshold, and mark the points exceeding the threshold as outliers and remove them.

8. The intelligent orchard inspection robot control method according to claim 7, characterized in that: Process the front-end odometer data, including: First, feature extraction is performed on the point cloud data, and its curvature is calculated. Points with larger curvature are identified as edge points, and points with smaller curvature are identified as plane points. Then, by matching the edge points and plane points in the current frame with the previous frame, the robot’s displacement and rotation state in the environment are estimated, thereby realizing the construction of a three-dimensional map of the environment; this process converts the posture information in the local coordinate system into the global coordinate system to ensure the consistency of the map.

9. The intelligent orchard inspection robot control method according to claim 8, characterized in that: The method further comprises: The RS method is used for rough search to screen out potential loop frame candidates: multiple key frame point cloud data at the current time are input into the 3D KD-Tree model, and then a certain dimension is selected for segmentation, and a tree structure is recursively constructed to find candidate frames close to the current frame; The SC method is applied to perform a fine search on the candidate frames and locate the cyclic frames.

10. The intelligent orchard inspection robot control method according to claim 9, characterized in that: The preprocessing includes ground separation, downsampling and noise reduction on the point cloud data.

Images