Tomato dynamic growth monitoring method based on robot inspection and digital twinning

Abstract

The invention discloses a tomato dynamic growth monitoring method based on robot inspection and digital twinning. Firstly, test tomato plants are selected and identified according to planting management requirements; then, an inspection map is constructed by utilizing ROS and SLAM algorithms, and an inspection robot inspects according to a set track, accurately identifies sample plants through identification information and parks; secondly, detecting the plant by two depth cameras carried on a liftable rotary bracket to obtain a three-dimensional point cloud model of the tomato, and extracting key phenotype data such as plant height, stem diameter, fruit bunch number, flowering number and the like; and finally, in combination with a digital twinborn technology, constructing a tomato dynamic growth model by using the periodically acquired three-dimensional point cloud model and phenotypic data, and in combination with environmental factors and water and fertilizer records, updating and optimizing a tomato growth management strategy in real time. According to the method, contact-free and damage-free tomato plant automatic phenotype monitoring is realized, the tomato dynamic growth model is constructed, and accurate data support and decision basis are provided for tomato planting management.

Description

Technical Field

[0001] This invention relates to the field of agricultural robots, and in particular to a method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins. Background Technology

[0002] Tomatoes, as an important vegetable crop widely cultivated globally, possess both high nutritional and significant economic value. However, current tomato cultivation and management in large-scale multi-span greenhouses still heavily rely on manual operation for growth monitoring and control, mainly due to three prominent problems: First, manual measurement is time-consuming, labor-intensive, and inefficient, making it difficult to meet the needs of large-scale cultivation; second, manually recording and organizing large amounts of discrete phenotypic data is cumbersome and prone to errors; and third, the collected discrete data is difficult to fully utilize and mine, hindering the construction of intelligent decision-making systems and the intelligent upgrading of the industry.

[0003] In recent years, with the rapid development of agricultural robots and machine vision technology, existing research has largely focused on the identification and detection of specific targets, such as tomato bunch localization, single fruit segmentation, maturity determination, or disease identification. For example, the "Real-time Position Recognition and Tomato Counting Yield Assessment Method for Inspection Robots" proposed in patent CN116681964A generally exhibits a strong single-task orientation, lacking a systematic integration of the overall three-dimensional structure and agronomic traits of the plant, making it difficult to support a comprehensive and dynamic assessment of tomato growth status. In the field of constructing plant growth models, for instance, the "Generator-based Construction of Three-Dimensional Plant Models" method used in patent CN118967971A relies primarily on template scaling and stitching in a pre-set database, with inputs limited to images and environmental variables, failing to reflect the actual three-dimensional structural changes of the plant.

[0004] Based on this, this invention proposes a method for dynamic growth monitoring of tomatoes based on robotic inspection and digital twins. This method uses an inspection robot equipped with two depth cameras to collect data from tomato plants, and combines this data with 3D reconstruction technology to construct a digital twin model of each individual experimental plant, thus achieving dynamic growth monitoring of tomatoes. Furthermore, the system can integrate environmental parameters and water and fertilizer records, supporting on-demand simulation of growth scenarios, providing planting managers with scientific, accurate, and intelligent decision-making support, and effectively promoting the intelligentization of tomato production. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins. Specifically, this method includes:

[0006] S1: Select appropriate experimental plants according to the experimental requirements of tomato growers and identify each sample plant uniquely.

[0007] S2: The inspection map is constructed using the robot operating system ROS and the Simultaneous Localization and Mapping (SLAM) algorithm. The inspection robot inspects along a predetermined track and accurately identifies the sample plants and parks itself by using the identification information.

[0008] S3: By using a liftable and rotating bracket mounted on a greenhouse inspection robot, two depth cameras are used to expand the field of view and simultaneously collect key phenotypic information of the target plant, such as plant height, stem diameter, number of fruit clusters, and number of flowers.

[0009] S4: Integrate the periodically acquired 3D point cloud model of tomatoes with phenotypic data such as plant height, stem diameter, number of fruit clusters, and flowering into a digital twin platform to construct an evolutionary dynamic growth model. This model simultaneously incorporates real-time environmental variables and water and fertilizer records to support growth trend prediction and management strategy optimization.

[0010] Preferably, step S1 specifically includes the following steps:

[0011] S11: Select tomato plants that meet the experimental conditions from the greenhouse as samples;

[0012] S12: Affix label 1 with a QR code to the top of each sample, and affix labels 2, 3 and 4 to the middle, lower middle and bottom of the plant, respectively, along the height of the plant. Labels 1, 2, 3 and 4 are identified by the same color, and only label 1 contains a QR code, which is used to uniquely identify the plant.

[0013] S13: After associating the QR code information in label 1 with the plant, store it in the database so that the growth status of each sample plant can be tracked and monitored in the future.

[0014] Preferably, in step S12, the information carried by the QR code on the label 1 includes: the unique ID number of the experimental plant, the variety of the seeds used, and the actual planting time.

[0015] Preferably, step S2 specifically includes the following steps:

[0016] S21: The greenhouse planting environment is scanned using the SLAM mapping method to generate an environmental map file in YAML format for greenhouse tomatoes. This map contains information on the spatial structure and obstacles in various areas of the greenhouse, which is the basis for the navigation of the greenhouse inspection robot.

[0017] S22: Mark the key locations and coordinates within the greenhouse, generate an automatic inspection path, and the greenhouse inspection robot inspects according to the predetermined route;

[0018] S23: After the inspection robot enters the ridge, it activates the overhead camera to simultaneously acquire RGB image streams and depth image streams for detecting the QR code information in tag 1. When the real-time acquired depth value of tag 1 is within the preset threshold range [ɑmin, ɑmax], the inspection robot performs initial parking. At this time, its position coordinates are: .

[0019] Preferably, in step S2, the hardware of the inspection robot navigation section includes a lidar, an inertial measurement unit, an embedded computing unit, and a mobile chassis.

[0020] Preferably, in step S22, the navigation algorithm of the inspection robot adopts a simultaneous localization and mapping algorithm based on LiDAR, specifically including: selecting the gmapping algorithm, remotely controlling the inspection robot to move and scan in the planting environment, enabling the LiDAR to collect environmental point cloud data in real time, record the spatial distribution of obstacles, generate a YAML map file for autonomous navigation, mark several key navigation points on the map, and configure corresponding navigation target files, using the A algorithm for path planning, and using the dynamic window method for local obstacle avoidance, thereby realizing the fully automatic inspection function of the inspection robot in the tomato planting environment.

[0021] Preferably, step S3 specifically includes the following steps:

[0022] S31: The two depth cameras on the greenhouse inspection robot stitch the fields of view vertically upwards. The field of view of the first depth camera covers the upper and middle part of the plant, while the field of view of the second depth camera covers the lower and middle part of the plant and the base of the stem. The fields of view of the two cameras are complementary in the vertical direction and the same in the horizontal direction to ensure that the key organs of the whole tomato plant are within the effective observation range.

[0023] S32: Based on the position of the inspection robot in S23 By combining the color identification information of labels 1 to 4, the position of each label in the image is identified, and the horizontal coordinate of the center of the detection box of label 1 is used as the reference. x-coordinate of the center of the detection box of label 4 Used as a reference for the left and right boundaries of the plant;

[0024] S33: Based on the x-coordinates of the left and right boundaries of the plant obtained from identification. and The geometric constraint benchmark of the scanning trajectory is determined, and the rotating support is controlled to drive the depth camera from the initial observation point. Through the real-time linkage compensation of the horizontal movement of the inspection robot and the rotation angle of the support, a scanning arc trajectory with the spatial point on the side of the plant as the virtual center is constructed. The depth camera moves along the arc trajectory from the rear side of the plant to the front side of the plant, and simultaneously collects the lateral phenotypic structure data and three-dimensional point cloud of the plant, so as to realize the dynamic growth data completion of the occluded area of ​​the plant from the front view.

[0025] S34: Based on the spatial coordinates of labels 1, 2, 3 and 4 in the 3D point cloud model, fit the central axis curve of the main stem of the tomato plant, and obtain the actual growth height of the plant by calculating the length of the curve from the base to the top, which is used as the plant height phenotypic parameter.

[0026] S35: Based on the obtained complete plant point cloud model, extract the target point cloud segment of the main stem of the tomato plant, and perform horizontal cross-sectional slicing sampling along the central axis of the main stem within the segment. Use the least squares method to fit the obtained cross-sectional point cloud to an ellipse, calculate the mean of the major axis and minor axis of the fitted ellipse, and define the mean of the major axis and minor axis as the stem thickness phenotypic parameter at the sampling location.

[0027] S36: The detection of the number of fruit clusters and the number of flowers is based on the target detection model. By collecting tomato flower and tomato full growth cycle datasets in advance for pre-training, the trained target detection algorithm is called during the inspection process. At the same time, the target tracking algorithm is combined to eliminate the problem of repeated counting caused by repeated target recognition. Finally, the accurate statistics of the total number of fruit clusters and the number of flowers of a single tomato plant are achieved.

[0028] Preferably, step S4 specifically includes the following steps:

[0029] S41: According to the preset date interval, control the greenhouse inspection robot to periodically inspect the same batch of sample tomato plants, repeat steps S1 to S3, and obtain the three-dimensional point cloud data and corresponding phenotypic parameters at each time point.

[0030] S42: Data collected from the same plant at different time points are matched for identity and time alignment using the QR code in its label 1 to form a time-series dataset based on the plant.

[0031] S43: Based on the time series dataset, perform time series modeling on key phenotypic parameters such as plant height, stem diameter, number of fruit clusters, and number of flowers, and fit their growth change trends;

[0032] S44: The fitting results are fused with the 3D point cloud model to construct a dynamic growth model of the plant in the digital twin platform, enabling visual retrospection and prediction of future growth status.

[0033] S45: Synchronously collect greenhouse environmental factor data and water and fertilizer integrated management records, associate the above environmental factors and water and fertilizer integrated records with the corresponding plant dynamic growth model according to the timestamp, and construct an enhanced digital twin that integrates growth-environment-water and fertilizer information;

[0034] S46: Based on the simulation scenario input by the manager, adjust the virtual water and fertilizer parameters or environmental settings, the model predicts the future growth status of tomato plants, and outputs optimization suggestions.

[0035] Preferably, in step S45, the greenhouse environmental factors include air temperature, relative humidity, carbon dioxide concentration, and light intensity. These environmental factors are collected in real time by IoT sensors deployed in the greenhouse and time-aligned with plant phenotypic data to construct an environmental coupling model.

[0036] Compared with existing technologies, this invention has the following advantages: Existing inspection robots mostly focus on a single target, such as tomato fruit identification, maturity judgment, or yield counting. This invention achieves dynamic monitoring of the entire tomato plant lifecycle and simultaneously collects key phenotypic information such as plant height, stem diameter, number of fruit clusters, and number of flowers, providing data support for precision cultivation and high-throughput breeding. Compared with digital twin plants constructed based on single-view images and generative models, which rely on template scaling and splicing from a preset database, this invention directly constructs a digital twin based on measured 3D point clouds and simultaneously establishes an individualized tomato growth database. This database further integrates environmental variables and water and fertilizer management operations. Plant managers can input different variables and optimize planting strategies in real time based on the inference results of the digital twin model. In summary, this invention overcomes the limitations of existing technologies in terms of single monitoring dimensions and lagging decision support. For the first time, it constructs a closed-loop system integrating "non-destructive perception, high-fidelity reconstruction, multi-source data fusion, and growth prediction decision-making," which significantly improves the intelligence and precision of tomato cultivation and provides a feasible technical path for the high-quality development of facility agriculture. Attached Figure Description

[0037] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0038] Figure 1 A flowchart of an overall method provided by the present invention;

[0039] Figure 2 A flowchart of a visual tag-based inspection robot for precise docking is provided for this invention.

[0040] Figure 3 This is a schematic diagram of three-dimensional reconstruction and phenotypic parameter extraction of a single tomato plant provided by the present invention;

[0041] Figure 4 A schematic diagram of camera rotation scanning based on multi-view fusion provided by the present invention;

[0042] Figure 5 This invention provides a flowchart for constructing a dynamic digital twin model of a tomato plant. Detailed Implementation

[0043] The embodiments of the present invention will be described in detail below. These embodiments are based on the technical solutions of the present invention and provide detailed implementation methods and specific operation processes to further explain the technical solutions of the present invention.

[0044] This embodiment provides a method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins, referencing... Figure 1 As shown, it includes the following steps:

[0045] S1: Select appropriate experimental plants according to the experimental requirements of tomato growers and identify each sample plant uniquely.

[0046] The specific steps of S1 include:

[0047] S11: Select tomato plants that meet the experimental conditions from the greenhouse as samples;

[0048] S12: Affix label 1 with a QR code to the top of each sample, and affix labels 2, 3 and 4 to the middle, lower middle and bottom of the plant, respectively, along the height of the plant. Labels 1, 2, 3 and 4 are identified by the same color, and only label 1 contains a QR code, which is used to uniquely identify the plant.

[0049] S13: After associating the QR code information in label 1 with the plant, store it in the database so that the growth status of each sample plant can be tracked and monitored in the future.

[0050] Specifically, in step S12, four visual tags are placed on the main stem of each sample tomato plant: tag 1 is affixed near the top of the main stem, with a unique QR code printed on its surface; tags 2, 3, and 4 are affixed to the middle (approximately 1 / 2 of the plant height), lower middle (approximately 1 / 3 of the plant height), and bottom of the main stem, respectively; tags 1 to 4 are all made of white matte waterproof PET material, presenting the same color (preferably bright yellow or blue) to ensure high contrast with the green tissue of the tomato, facilitating image segmentation and feature extraction; tag 1 is a square mark with a side length of 4cm, with a high-density QR code printed on its surface, used to encode the plant's unique identity ID; tags 2, 3, and 4 are all circular marks with a diameter of 1.5cm, without any encoding information, serving only as auxiliary feature points for geometric registration, used for subsequent multi-view point cloud stitching and plant posture estimation.

[0051] S2: The inspection map is constructed using the robot operating system ROS and the Simultaneous Localization and Mapping (SLAM) algorithm. The inspection robot inspects along a predetermined track and accurately identifies the sample plants by using the marking information and then stops.

[0052] The specific steps of S2 include:

[0053] S21: The greenhouse planting environment is scanned using the SLAM mapping method to generate an environmental map file in YAML format for greenhouse tomatoes. This map contains information on the spatial structure and obstacles in various areas of the greenhouse, which is the basis for the navigation of the greenhouse inspection robot.

[0054] S22: Mark the key locations and coordinates within the greenhouse, generate an automatic inspection path, and the greenhouse inspection robot inspects according to the predetermined route;

[0055] S23: After the inspection robot enters the ridge, it activates the overhead camera to simultaneously acquire RGB image streams and depth image streams for detecting the QR code information in tag 1. When the real-time acquired depth value of tag 1 is within the preset threshold range [ɑmin, ɑmax], the inspection robot performs an initial stop, at which point its position coordinates are: .

[0056] refer to Figure 2As shown, specifically, the inspection robot is equipped with hardware including a LiDAR, an inertial measurement unit (IMU), a Jetson embedded computing platform, etc., and is configured with an Ubuntu operating system. A ROS-based navigation and control software framework is deployed on top of this system to achieve path planning, localization mapping, and autonomous inspection functions. The key points marked in S22 include the starting and ending points of each tomato row, and inflection points between rows. A YAML map constructed based on the SLAM algorithm is loaded with preset navigation points, and an automatic inspection path is generated through a global path planning algorithm (such as A* or Dijkstra). In step S23, the inspection robot patrols along a predetermined trajectory. Upon entering the inter-row area, the top depth camera simultaneously acquires RGB and depth image streams. The unique plant ID in tag 1 is analyzed in real-time using the ZBar QR code decoding library. Simultaneously, the depth value d corresponding to the center pixel of tag 1 is obtained. If d ∈ [ɑmin, ɑmax] (this range is pre-calibrated based on the tag 1 installation height and camera field of view), a navigation interruption signal is triggered, controlling the robot to perform initial docking. Subsequently, the horizontal coordinate of tag 1 in the image is extracted. Proceed to step S3.

[0057] S3: By using a liftable and rotating support mounted on a greenhouse inspection robot, two depth cameras are used to expand the field of view and simultaneously collect key phenotypic information of the target plant, such as plant height, stem diameter, number of fruit clusters, and number of flowers.

[0058] The specific steps of S3 include:

[0059] S31: The two depth cameras on the greenhouse inspection robot stitch the fields of view vertically upwards. The field of view of the first depth camera covers the upper and middle part of the plant, while the field of view of the second depth camera covers the lower and middle part of the plant and the base of the stem. The fields of view of the two cameras are complementary in the vertical direction and basically the same in the horizontal direction to ensure that the key organs of the whole tomato plant are within the effective observation range.

[0060] S32: Based on the position of the inspection robot in S23 By combining the color identification information of labels 1 to 4, the position of each label in the image is identified, and the horizontal coordinate of the center of the detection box of label 1 is used as the reference. x-coordinate of the center of the detection box of label 4 Used as a reference for the left and right boundaries of the plant;

[0061] S33: Based on the x-coordinates of the left and right boundaries of the plant obtained from identification. and The geometric constraint benchmark of the scanning trajectory is determined, and the rotating support is controlled to drive the depth camera from the initial observation point. Through the real-time linkage compensation of the horizontal movement of the inspection robot and the rotation angle of the support, a scanning arc trajectory with the spatial point on the side of the plant as the virtual center is constructed. The depth camera moves along the arc trajectory from the rear side of the plant to the front side of the plant, and simultaneously collects the lateral phenotypic structure data and three-dimensional point cloud of the plant, so as to realize the dynamic growth data completion of the occluded area of ​​the plant from the front view.

[0062] S34: Based on the spatial coordinates of labels 1, 2, 3 and 4 in the 3D point cloud model, fit the central axis curve of the main stem of the tomato plant, and obtain the actual growth height of the plant by calculating the length of the curve from the base to the top, which is used as the plant height phenotypic parameter.

[0063] S35: Based on the obtained complete plant point cloud model, extract the target point cloud segment of the main stem of the tomato plant, and sample the horizontal cross-section slices along the central axis of the main stem within the segment. Use the least squares method to fit the obtained cross-section point cloud to an ellipse, calculate the mean of the major axis and minor axis of the fitted ellipse, and define it as the stem thickness phenotypic parameter at the sampling location.

[0064] S36: The detection of the number of fruit clusters and the number of flowers is based on the target detection model. By collecting tomato flower and tomato full growth cycle datasets in advance for pre-training, the trained target detection algorithm is called during the inspection process. At the same time, the target tracking algorithm is combined to eliminate the problem of repeated counting caused by repeated target recognition. Finally, the accurate statistics of the total number of fruit clusters and the number of flowers of a single tomato plant are achieved.

[0065] refer to Figure 3 As shown, specifically, camera 1 covers the top and middle of the plant, and camera 2 covers the lower middle and bottom of the plant. When labels 1, 2, 3, and 4 are not completely contained within the camera's field of view (F), the x-coordinate of label 1 obtained in step S2 is used. Using the reference point, the inspection robot makes fine adjustments to its position by moving left and right until all tags (tags 1, 2, 3, and 4) are within the camera's field of view (Y), ensuring that the entire tomato plant is completely covered, thereby achieving high-precision 3D reconstruction and phenotypic parameter extraction. Specifically, it identifies four tags and performs identity matching and verification based on their color features and QR code information. Using the horizontal coordinates of tags 1 and 4 as the left and right boundaries, it scans to achieve multi-view 3D information acquisition of the entire plant, obtaining single-shot 3D point cloud data and single-shot phenotypic data.

[0066] refer to Figure 4As shown, specifically, depending on the observation requirements, the camera supports two observation positions: a long-distance coverage mode (away from the plant) and a close-range observation mode (close to the plant). In the long-distance coverage mode, the rotating mechanism of the inspection robot support is centered at a fixed point O1, and its coordinates are (…). Let's define a circular motion with radius R as circle 1. Let the current camera be located at point A on circle 1. Using point A as the tangent point, construct an upper circle that is tangent to the outside of circle 1, denoted as circle 2. Let its center be O2, and its coordinates be ( , y). (y+R), then perform compensation detection on the same circle, let's call it circle 3, with center O3 and coordinates ( +R, y), connecting O2 and point B to form ray O2B, which intersects circle 3 at point C. Simultaneously, circle 1 intersects circle 3 at point D. The inspection robot scans the left side of the plant along arc CD. β is the central angle formed by points C, O2, and D in circle 2, corresponding to the field of view covered by the inspection robot scanning the side of the plant along arc CD. Similarly, a symmetrical structure can be constructed to complete the scanning of the right side of the plant, achieving full-circumference multi-view data acquisition.

[0067] refer to Figure 4 As shown, specifically in close-range observation mode, the rotating mechanism of the inspection robot's support is positioned at point A on circle 2, with fixed point O2 as the center. At this time, the camera rotates along the arc AB path to complete the visual scan of the left side of the plant. Similarly, a symmetrical structure can be constructed to achieve scanning of the right side of the plant.

[0068] S4: The periodically acquired 3D point cloud model of tomatoes is integrated with phenotypic data such as plant height, stem diameter, number of fruit clusters, and number of flowers into the digital twin platform to construct an evolutionary dynamic growth model. This model is combined with real-time environmental variables and water and fertilizer records to support growth trend prediction and management strategy optimization.

[0069] See Figure 5 First, three-dimensional point cloud data of the same plant at different times were acquired through multiple inspections. The point clouds from each period were then identified and timestamped to ensure spatiotemporal consistency. Next, non-rigid registration was performed on the registered point clouds from multiple periods to achieve spatial alignment. Phenotypic parameters such as plant height, crown width, and volume were extracted to form a discrete phenotypic evolution sequence. An interpolation algorithm was then used to complete the growth process and generate a continuous growth animation. Simultaneously, a growth function was used to fit the parameter sequence, establishing a mathematical model of its temporal changes. Based on this, the spatial morphological changes of the point clouds and the temporal evolution of phenotypic parameters were integrated to construct a dynamic growth model encompassing both spatiotemporal dimensions. Finally, environmental factors (such as light and temperature) and water and fertilizer management data were further integrated to construct an enhanced digital twin system with growth prediction capabilities, enabling refined modeling and intelligent management of the entire plant life cycle.

[0070] Specifically, the specific steps of S4 include:

[0071] S41: According to the preset date interval, control the greenhouse inspection robot to periodically inspect the same batch of sample tomato plants, repeat steps S1 to S3, and obtain the three-dimensional point cloud data and corresponding phenotypic parameters at each time point.

[0072] S42: Data collected from the same plant at different time points are matched for identity and time alignment using the QR code in its label 1 to form a time-series dataset based on the plant.

[0073] S43: Based on the time series dataset, perform time series modeling on key phenotypic parameters such as plant height, stem diameter, number of fruit clusters, and number of flowers, and fit their growth change trends;

[0074] S44: The fitting results are fused with the 3D point cloud model to construct a dynamic growth model of the plant in the digital twin platform, enabling visual retrospection and prediction of future growth status.

[0075] S45: Synchronously collect greenhouse environmental factor data and water and fertilizer integrated management records, associate the above environmental factors and water and fertilizer integrated records with the corresponding plant dynamic growth model according to the timestamp, and construct an enhanced digital twin that integrates growth-environment-water and fertilizer information;

[0076] S46: Based on the simulation scenario input by the manager, adjust the virtual water and fertilizer parameters or environmental settings, the model predicts the future growth status of tomato plants, and outputs optimization suggestions.

[0077] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0078] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins, characterized in that, Includes the following steps: S1: Select appropriate experimental plants according to the experimental requirements of tomato growers and identify each sample plant uniquely. S2: The inspection map is constructed using the robot operating system ROS and the Simultaneous Localization and Mapping (SLAM) algorithm. The inspection robot inspects along a predetermined track and accurately identifies the sample plants and parks itself by using the identification information. S3: By using a liftable and rotating bracket mounted on a greenhouse inspection robot, two depth cameras are used to expand the field of view and simultaneously collect key phenotypic information of the target plant, such as plant height, stem diameter, number of fruit clusters, and number of flowers. S4: The periodically acquired 3D point cloud model of tomatoes is integrated with phenotypic data such as plant height, stem diameter, number of fruit clusters, and number of flowers into the digital twin platform to construct an evolutionary dynamic growth model. This model is combined with real-time environmental variables and water and fertilizer records to support growth trend prediction and management strategy optimization.

2. The method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins according to claim 1, characterized in that, S1 specifically includes the following steps: S11: Select tomato plants that meet the experimental conditions from the greenhouse as samples; S12: Affix label 1 with a QR code to the top of each sample, and affix labels 2, 3 and 4 to the middle, lower middle and bottom of the plant, respectively, along the height of the plant. Labels 1, 2, 3 and 4 are identified by the same color, and only label 1 contains a QR code, which is used to uniquely identify the plant. S13: After associating the QR code information in label 1 with the plant, store it in the database so that the growth status of each sample plant can be tracked and monitored in the future.

3. The method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins according to claim 1, characterized in that, S2 specifically includes the following steps: S21: The greenhouse planting environment is scanned using the SLAM mapping method to generate an environmental map file in YAML format for greenhouse tomatoes. This map contains information on the spatial structure and obstacles in various areas of the greenhouse, which is the basis for the navigation of the greenhouse inspection robot. S22: Mark the key locations and coordinates within the greenhouse, generate an automatic inspection path, and the greenhouse inspection robot inspects according to the predetermined route; S23: After the inspection robot enters the ridge, it activates the overhead camera to simultaneously acquire RGB image streams and depth image streams for detecting the QR code information in tag 1. When the real-time acquired depth value of tag 1 is within the preset threshold range [ɑmin, ɑmax], the inspection robot performs initial parking. At this time, its position coordinates are: .

4. The method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins according to claim 1, characterized in that, S3 specifically includes the following steps: S31: The two depth cameras on the greenhouse inspection robot stitch the fields of view vertically upwards. The field of view of the first depth camera covers the upper and middle part of the plant, while the field of view of the second depth camera covers the lower and middle part of the plant and the base of the stem. The fields of view of the two cameras are complementary in the vertical direction and the same in the horizontal direction to ensure that the key organs of the whole tomato plant are within the effective observation range. S32: Based on the position of the inspection robot in S23 Combining the color identification information of labels 1 to 4, the position of each label in the image is identified, using the x-coordinate of the center of the detection box of label 1. x-coordinate of the center of the detection box of label 4 Used as a reference for the left and right boundaries of the plant; S33: Based on the x-coordinates of the left and right boundaries of the plant obtained from identification. and The geometric constraint benchmark of the scanning trajectory is determined, and the rotating support is controlled to drive the depth camera from the initial observation point. Through the real-time linkage compensation of the horizontal movement of the inspection robot and the rotation angle of the support, a scanning arc trajectory with the spatial point on the side of the plant as the virtual center is constructed. The depth camera moves along the arc trajectory from the rear side of the plant to the front side of the plant, and simultaneously collects the lateral phenotypic structure data and three-dimensional point cloud of the plant, so as to realize the dynamic growth data completion of the occluded area of ​​the plant from the front view. S34: Based on the spatial coordinates of labels 1, 2, 3 and 4 in the 3D point cloud model, fit the central axis curve of the main stem of the tomato plant, and obtain the actual growth height of the plant by calculating the length of the curve from the base to the top, which is used as the plant height phenotypic parameter. S35: Based on the obtained complete plant point cloud model, extract the target point cloud segment of the main stem of the tomato plant, and perform horizontal cross-sectional slicing sampling along the central axis of the main stem within the segment. Use the least squares method to fit the obtained cross-sectional point cloud to an ellipse, calculate the mean of the major axis and minor axis of the fitted ellipse, and define the mean of the major axis and minor axis as the stem thickness phenotypic parameter at the sampling location. S36: The detection of the number of fruit clusters and the number of flowers is based on the target detection model. By collecting tomato flower and tomato full growth cycle datasets in advance for pre-training, the trained target detection algorithm is called during the inspection process. At the same time, the target tracking algorithm is combined to eliminate the problem of repeated counting caused by repeated target recognition. Finally, the accurate statistics of the total number of fruit clusters and the number of flowers of a single tomato plant are achieved.

5. The method for monitoring the dynamic growth of tomatoes based on robot inspection and digital twins according to claim 1, characterized in that, S4 specifically includes the following steps: S41: According to the preset date interval, control the greenhouse inspection robot to periodically inspect the same batch of sample tomato plants, repeat steps S1 to S3, and obtain the three-dimensional point cloud data and corresponding phenotypic parameters at each time point. S42: Data collected from the same plant at different time points are matched for identity and time alignment using the QR code in its label 1 to form a time-series dataset based on the plant. S43: Based on the time series dataset, perform time series modeling on key phenotypic parameters such as plant height, stem diameter, number of fruit clusters, and number of flowers, and fit their growth change trends; S44: The fitting results are fused with the 3D point cloud model to construct a dynamic growth model of the plant in the digital twin platform, enabling visual retrospection and prediction of future growth status. S45: Synchronously collect greenhouse environmental factor data and water and fertilizer integrated management records, associate the above environmental factors and water and fertilizer integrated records with the corresponding plant dynamic growth model according to the timestamp, and construct an enhanced digital twin that integrates growth-environment-water and fertilizer information; S46: Based on the simulation scenario input by the manager, adjust the virtual water and fertilizer parameters or environmental settings, the model predicts the future growth status of tomato plants, and outputs optimization suggestions.

Images