A smart shape-mimicking flower thinning system and method for litchi tree canopy
The intelligent contour-following flower thinning system, which combines 3D point cloud perception and deep neural networks, solves the problem of dynamic adjustment of the flower thinning device for litchi trees under changes in canopy morphology, and achieves efficient and stable flower thinning results.
Patent Information
- Application Number
- CN202610197293.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-11
- Publication Date
- 2026-05-26
AI Technical Summary
In existing technologies, litchi tree thinning devices are difficult to dynamically adjust according to real-time changes in the tree canopy morphology and flower spike distribution, resulting in unstable thinning effects and limited operational adaptability. Two-dimensional image recognition schemes lack accuracy in complex environments and have poor model generalization performance.
An intelligent contour-following flower thinning system combining 3D point cloud perception and deep neural networks is used to achieve precise positioning and dynamic flower thinning control of flower spikes, branches and leaves through 3D point cloud acquisition of litchi tree canopy, point cloud semantic segmentation, 3D canopy envelope surface construction and adaptive contour-following flower thinning mechanism.
It improves the environmental adaptability and operational stability of flower thinning operations, reduces the risk of unintended contact and branch damage, and achieves high-precision and high-efficiency flower thinning results.
Smart Images
Figure CN122074336A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of agricultural engineering technology, specifically relating to an intelligent shape-mimicking flower thinning system and method for litchi tree canopy. Background Technology
[0002] Under natural growing conditions, litchi trees are prone to excessive flowering and dense, uneven distribution of flower spikes. This can lead to nutrient dispersion in the tree, reduced fruit set, decreased fruit quality, and exacerbated alternate bearing (biennial bearing). Therefore, appropriate thinning of flower spikes during peak flowering is an important agronomical measure for achieving stable and high-quality litchi production.
[0003] In the existing technology, the most widely used fruit tree thinning device is based on the principle of mechanical impact or vibration. Although this type of thinning device can replace manual thinning to a certain extent, it still has problems such as insufficient stability of thinning effect and limited adaptability in practical applications. Its main disadvantage is that the thinning intensity and working range mainly rely on pre-set fixed parameters or manual experience for adjustment during the operation, and it is difficult to dynamically adjust according to the real-time changes in the fruit tree canopy morphology and flower spike distribution.
[0004] With the development of computer vision and deep learning technologies, some research has begun to introduce flower or inflorescence recognition methods based on two-dimensional images into the field of fruit tree thinning and refined management. These methods typically utilize RGB cameras to acquire images of fruit trees during their flowering period, and then use convolutional neural networks to detect and locate flowers or inflorescences in the images. The recognition results are then used to assist in thinning decisions or guide the operation of mechanical actuators. However, these methods still have certain limitations in supporting refined thinning operations in complex natural orchard environments. Their main drawbacks are insufficient ability to describe the overall spatial relationships of flower spikes, branches, and the tree canopy, making it difficult to provide reliable spatial location information and operational constraints for the thinning actuators. Two-dimensional images themselves lack depth information and cannot accurately reflect the front-to-back relationships and distance changes between flower spikes and branches. In situations where flower spikes are dense, mutually occluded, or where lighting conditions vary significantly, the recognition results are easily affected. Furthermore, these methods often rely on large-scale labeled data under specific crop varieties and scene conditions for model training; when the tree species, flower spike morphology, or growth stage changes, the model's generalization performance tends to decline. Summary of the Invention
[0005] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and to propose an intelligent shape-mimicking flower thinning system and method for litchi tree canopy.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] An intelligent shape-mimicking flower thinning system for litchi tree canopy includes an information sensing module, a data analysis module, and an operation execution module;
[0008] The information perception module includes a three-dimensional point cloud acquisition unit for the canopy of litchi trees and a point cloud semantic segmentation unit. The three-dimensional point cloud acquisition unit for the canopy of litchi trees is used to acquire three-dimensional point cloud data of the canopy of litchi trees, providing basic data support for subsequent analysis. The point cloud semantic segmentation unit uses a deep neural network model based on PointNet++ to perform semantic segmentation on the acquired point cloud data and identify the regions of flower spikes, branches and leaves.
[0009] The data analysis module consists of a canopy 3D envelope surface construction and inflorescence density analysis unit. Based on the segmented inflorescence point cloud and branch and leaf point cloud, the canopy 3D envelope surface construction and inflorescence density analysis unit constructs the fruit tree canopy 3D envelope surface; on the fruit tree canopy 3D envelope surface, the local inflorescence distribution density and local geometric features are calculated;
[0010] The operation execution module is a three-section arm adaptive contour thinning mechanism, which can accurately adapt to the curved surface of the fruit tree canopy to achieve adaptive contouring. By closely following the three-dimensional envelope surface of the canopy, it can smoothly and efficiently complete the precise flower thinning operation.
[0011] This invention also includes a flower thinning method based on the provided intelligent contour-following flower thinning system, comprising the following steps:
[0012] S1. A depth camera installed at the end of the three-section arm adaptive contour thinning mechanism is used to collect three-dimensional point cloud data of the fruit tree canopy and obtain multimodal point cloud data containing three-dimensional spatial coordinates and color information.
[0013] S2. Use a deep neural network model based on PointNet++ to perform semantic segmentation on the collected point cloud data to identify the flower spike, branch and leaf regions.
[0014] S3. Based on the segmented flower spike point cloud and branch and leaf point cloud, construct a three-dimensional envelope surface of the fruit tree canopy to describe the geometric boundary and spatial morphology of the fruit tree canopy.
[0015] S4. On the three-dimensional envelope surface of the fruit tree canopy, based on the target recognition results of the flower spikes, mark the local density of the flower spikes on the three-dimensional envelope surface of the canopy to obtain a heat map of the spatial distribution of flower spike density.
[0016] S5. Based on the spatial distribution heat map of flower spike density obtained in step S4, dynamically adjust the thinning operation parameters, including the extension length of the three-section arm, the contour angle of the thinning rotor, and the rotor speed, so as to achieve thinning control in different canopy areas.
[0017] S6. Control the three-section arm adaptive contour thinning mechanism to move adaptively and closely along the envelope surface of the fruit tree canopy to perform precise thinning operations.
[0018] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0019] 1. Compared with existing flower thinning devices that mainly rely on fixed mechanical parameters and flower thinning schemes based on two-dimensional vision, this invention introduces three-dimensional visual perception and intelligent decision-making mechanisms, which have stronger environmental adaptability and operational stability under complex litchi canopy conditions. It uses multi-view three-dimensional perception to acquire canopy point cloud information and uses intelligent algorithms to effectively represent the geometric morphology of flower spikes, branches and canopy, so that the flower thinning operation is based on the real spatial structure. This is different from the existing technology that controls the operation based on only empirical parameters or two-dimensional images, and helps to improve the controllability and consistency of the flower thinning operation in scenarios with uneven flower spike distribution and severe shading.
[0020] 2. This invention constructs a closed-loop operation system that combines perception, decision-making, and execution, enabling the thinning execution parameters to be dynamically adjusted according to the canopy geometry and operational status. Combined with a compliant contour-following execution mechanism, it achieves active conformation to the tree canopy outline. Compared to existing thinning equipment with relatively static control methods and limited contour-following capabilities, this technology helps reduce the risk of unintended contact and branch damage, and improves the stability and adaptability of the thinning process, thereby achieving a more balanced and controllable thinning effect in complex orchard environments. Attached Figure Description
[0021] Figure 1 This is an overall schematic diagram of the system of the present invention.
[0022] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation
[0023] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0024] Explanation of relevant terms:
[0025] 3D point cloud: refers to the collection of points in space that represent the surface characteristics of an object, obtained through sensors such as depth cameras.
[0026] Semantic segmentation: a technique in the field of computer vision, refers to classifying each pixel or point in an image or point cloud and assigning it a specific category label.
[0027] PointNet++: A deep neural network model specifically designed for processing 3D point cloud data; it employs a hierarchical feature learning approach to extract local geometric structure features of point clouds at different scales, making it suitable for target recognition and semantic segmentation of complex point cloud data.
[0028] Adaptive Morphing / Profiling refers to the ability of a mechanical actuator to adjust its position and parameters in real time according to changes in the surface contour of the target object, so that the end effector maintains a constant distance or fits the target surface.
[0029] Poisson Surface Reconstruction: An algorithm for extracting continuous, closed mesh surfaces from disordered point clouds.
[0030] Examples; such as Figure 1 As shown, an intelligent shape-mimicking flower thinning system for litchi tree canopy includes an information sensing module, a data analysis module, and a task execution module.
[0031] The information perception module includes a three-dimensional point cloud acquisition unit for the canopy of litchi trees and a point cloud semantic segmentation unit. The three-dimensional point cloud acquisition unit for the canopy of litchi trees is used to acquire three-dimensional point cloud data of the canopy of litchi trees, providing basic data support for subsequent analysis. The point cloud semantic segmentation unit uses a deep neural network model based on PointNet++ to perform semantic segmentation on the acquired point cloud data and identify the regions of flower spikes, branches and leaves.
[0032] The data analysis module consists of a canopy 3D envelope surface construction and inflorescence density analysis unit. Based on the segmented inflorescence point cloud and branch and leaf point cloud, the canopy 3D envelope surface construction and inflorescence density analysis unit constructs the fruit tree canopy 3D envelope surface; on the fruit tree canopy 3D envelope surface, the local inflorescence distribution density and local geometric features are calculated;
[0033] The operation execution module is a three-section arm adaptive contour thinning mechanism, which can accurately adapt to the curved surface of the fruit tree canopy to achieve adaptive contouring. By closely following the three-dimensional envelope surface of the canopy, it can smoothly and efficiently complete the precise flower thinning operation.
[0034] In this embodiment, the three-dimensional point cloud acquisition unit for the litchi tree canopy specifically employs a stereo vision array composed of depth cameras. The acquired point cloud data is N×6 dimensional data (N is the number of points in the point cloud), including:
[0035] Three-dimensional spatial coordinates (X, Y, Z) are used to describe the spatial relationship between various points in the fruit tree canopy;
[0036] Color information (R, G, B) is used to provide color characteristics of the fruit tree canopy.
[0037] This method can obtain high-precision, dense three-dimensional multimodal point clouds of the fruit tree canopy, providing basic data support for subsequent flower spike identification and thinning strategies.
[0038] In this embodiment, to achieve accurate identification of fruit tree flower spikes, the point cloud semantic segmentation unit uses a deep neural network model based on PointNet++ to perform semantic segmentation on the collected point cloud data, including the following steps:
[0039] Model pre-training: The PointNet++ model is pre-trained on an existing 3D point cloud dataset of fruit trees with similar structures and morphologies to learn the general macroscopic features of fruit tree structure and morphology.
[0040] Freeze the macro feature extraction layer; after pre-training, freeze the network layer in the PointNet++ model used to extract macro features in order to retain its ability to perceive and analyze the overall structure of the fruit tree.
[0041] Fine-tuning the micro-feature extraction layer; only fine-tuning the parameters of the small-scale feature extraction layer related to the inflorescence spike is performed to enable the PointNet++ model to adapt to the inflorescence spike characteristics of different fruit tree varieties and canopy structures.
[0042] Output semantic classification results; use the trained PointNet++ model to output the semantic classification results of point cloud data, including flower spikes, branches and leaves.
[0043] The above method enables high-precision point cloud recognition of flower spikes under small sample conditions, providing accurate target information for subsequent three-segment adaptive contour thinning, thereby improving the efficiency and accuracy of thinning.
[0044] In this embodiment, the canopy three-dimensional envelope surface construction and flower spike density analysis unit first constructs the three-dimensional envelope surface of the fruit tree canopy, and then calculates the local flower spike distribution density and local geometric features, specifically:
[0045] After completing the semantic segmentation of the point cloud, the outer contour point cloud of the fruit tree canopy is extracted and a continuous, closed three-dimensional envelope surface of the fruit tree canopy is constructed using the Poisson surface reconstruction method.
[0046] The three-dimensional envelope surface of the fruit tree canopy is smoothed and its local curvature characteristics are calculated. At the same time, the distribution relationship of the flower spike point cloud in the surface space is combined to realize the accurate calculation of the number of flower spikes, spatial spacing and distribution density, thus providing accurate and stable canopy geometry and spatial distribution basis for flower thinning operations.
[0047] The constructed three-dimensional envelope surface can not only fully describe the overall morphology of the fruit tree canopy, but also serve as a geometric reference basis for the three-section arm to perform contour-following motion along the canopy, thereby realizing a close-fitting flower thinning operation and improving the accuracy and efficiency of flower thinning.
[0048] In this embodiment, the three-section adaptive contouring openwork mechanism includes, from top to bottom, a first section arm, a second section arm, and a third section arm;
[0049] The main support rod is vertically fixed to the mobile operating platform or tracked chassis; the first, second, and third arm sections are connected to different height positions of the main support rod via rotating joints; each arm section is equipped with a flower thinning execution component at its far end, which can be used independently for different canopy areas of the fruit tree; each rotating joint adopts a fixed support structure at the upper and lower ends to enhance impact resistance and prevent axial movement or structural loosening during high-frequency flower thinning; a controllable drive unit is set at the joint connection point to adapt to the shape of the fruit tree canopy by adjusting the rotation angle of the arm.
[0050] The three-section arm adaptive contouring flower thinning mechanism in this embodiment achieves adaptive contouring of the fruit tree canopy surface in the following way:
[0051] Controllable drive joints; each rotating joint between the arms is equipped with a controllable drive unit to adjust the joint angle and adapt to different canopy morphologies;
[0052] Independent posture adjustment; each arm section can be independently adjusted in posture according to the local curvature changes of the canopy during operation, so that the flower thinning execution component can be accurately aligned with the target area;
[0053] Multi-arm coordinated operation; the coordinated movement of each arm enables the thinning execution component to continuously conform to the three-dimensional outer contour of the fruit tree canopy, achieving full-coverage thinning operation;
[0054] The distribution of thinning components: Each arm is equipped with an independent thinning execution component to ensure that different areas can be thinned simultaneously in complex canopy structures.
[0055] Through the above structural design, the three-section arm adaptive contour thinning mechanism of this embodiment can effectively adapt to the spatial characteristics of large changes in the curvature of the litchi tree crown, obvious local concavity and convexity, and complex interlacing of branches. It solves the problem of insufficient contour thinning ability of traditional single-arm or double-arm structures and realizes high-precision and high-efficiency automated flower thinning operation.
[0056] In this embodiment, the three-section arm adaptive contour-following flower-opening mechanism performs precise flower-opening operations including:
[0057] Based on the perceived geometric features and phenotypic analysis results, the motion path, working angle, and flower thinning speed of the three-section adaptive contour thinning mechanism are dynamically calculated. Then, the three-section adaptive contour thinning mechanism is driven, and each section adjusts its posture independently under the guidance of the controllable drive unit. Together with the end flower thinning execution component, it realizes continuous conforming operation along the three-dimensional envelope surface of the tree canopy.
[0058] In this embodiment, the depth camera used for 3D point cloud acquisition can be replaced with:
[0059] LiDAR; it uses the time-of-flight (ToF) or phase difference ranging principle to directly acquire high-precision geometric point cloud data; the active emission of light source by LiDAR makes it more robust to perception and more accurate in orchard environments with strong light, shadow or extreme lighting changes than depth cameras.
[0060] Multi-view vision system: It uses two or more conventional RGB cameras to form a stereo vision array, and calculates the disparity map through stereo matching algorithms such as feature point matching or semi-global matching (SGM), and then reconstructs the three-dimensional point cloud of the fruit tree canopy; this scheme can improve the accuracy of long-distance perception by increasing the baseline length.
[0061] In this embodiment, the PointNet++-based deep neural network model can be replaced with:
[0062] Multidimensional feature point cloud processing networks; such as DGCNN (Dynamic Graph Convolutional Neural Network), PointCNN, RandLA-Net or KPConv, which are point cloud deep learning networks with hierarchical feature extraction capabilities; such networks can extract local geometric structure features of point clouds through different neighborhood aggregation strategies (such as dynamic graph construction or kernel convolution), thereby achieving semantic segmentation functions of flower spikes, branches and leaves similar to those in this embodiment.
[0063] A voxel-based neural network based on dimensionality transformation can first project or transform the original unordered point cloud data into a regular three-dimensional voxel grid, and then use a 3D CNN (three-dimensional convolutional neural network) for feature learning and classification. This scheme uses spatial discretization to transform unstructured point clouds into structured tensors, and uses mature convolution operators to extract spatial features of fruit tree canopies.
[0064] A multimodal information fusion perception model is proposed, employing a deep learning model that simultaneously processes 2D image features and 3D point cloud features. This model enhances the model's accuracy in recognizing subtle textures, color differences, and spatial morphology of flower spikes by fusing RGB texture information from a depth camera with XYZ spatial geometric information across the feature layer or decision layer. This improves the accuracy of identifying sparsely flowered targets in complex environments.
[0065] In this embodiment, the three-section arm adaptive contouring openwork mechanism can be replaced with:
[0066] Multi-stage linear telescopic mechanism; using sleeve-type or guide rail-type telescopic structure to replace some or all rotating joints; this mechanism achieves radial displacement compensation by changing the telescopic length of each level of arm, and can make real-time telescopic adjustments according to the radial depth changes of the fruit tree canopy, making it particularly suitable for operation scenarios with a large canopy depth span and requiring linear contour compensation.
[0067] Multi-degree-of-freedom parallel fine-tuning mechanism: A small six-degree-of-freedom parallel platform (such as the Stewart platform) is added to the end of the serial robotic arm. Utilizing the high rigidity and high frequency response characteristics of the parallel mechanism, the parallel platform realizes high-frequency and fine attitude fine-tuning of the end flower thinning execution component on the basis of the main arm completing a large range of contouring motion, thereby further improving the operation accuracy of complex interlaced flower spikes.
[0068] Flexible continuum contouring mechanism: This mechanism replaces rigid linkages with a flexible continuum mechanism that simulates an elephant trunk or snake-like structure. It uses internal drive ropes or fluid pressure to change the curvature of the flexible body, achieving jointless continuous bending contouring. Due to its infinite degree of freedom, this mechanism has better obstacle avoidance and conformity when facing the extremely irregular spatial distribution and dense branch environment inside the lychee canopy.
[0069] like Figure 2 As shown, the intelligent contour-following flower thinning system in this embodiment includes the following steps when performing flower thinning operations:
[0070] S1. A depth camera installed at the end of the three-section arm adaptive contour thinning mechanism is used to collect three-dimensional point cloud data of the fruit tree canopy and obtain multimodal point cloud data containing three-dimensional spatial coordinates and color information.
[0071] S2. Using a deep neural network model based on PointNet++, semantic segmentation is performed on the collected point cloud data to identify the regions of flower spikes, branches and leaves, thereby achieving precise localization of the flower spike target.
[0072] S3. Based on the segmented flower spike point cloud and branch and leaf point cloud, construct a three-dimensional envelope surface of the fruit tree canopy to describe the geometric boundary and spatial morphology of the fruit tree canopy, and provide a reference for flower thinning path planning.
[0073] S4. On the three-dimensional envelope surface of the fruit tree canopy, based on the target recognition results of the flower spikes, mark the local density of the flower spikes on the three-dimensional envelope surface of the canopy to obtain a heat map of the spatial distribution of flower spike density.
[0074] S5. Based on step S4, obtain the spatial distribution heat map of flower spike density, and dynamically adjust the thinning operation parameters, including the extension length of the three-section arm, the contour angle of the thinning rotor, and the rotor speed, to achieve thinning control in different canopy areas.
[0075] S6. Control the three-section arm adaptive contour thinning mechanism to move adaptively and closely along the envelope surface of the fruit tree canopy to perform precise thinning operations.
[0076] Through the above steps, this embodiment can achieve automation, intelligence and high precision in fruit tree flower thinning operations. The method makes full use of three-dimensional point cloud data and local geometric features to achieve dynamic adjustment of flower thinning execution parameters, so that the flower thinning operation can maintain the best effect under different fruit tree varieties, different canopy structures and different flower spike densities.
[0077] This invention provides an intelligent contour-following flower thinning system and method for litchi tree canopies. Designed for the complex canopy morphology and uneven spatial distribution of flower spikes in litchi and other fruit trees, the system establishes a collaborative working mechanism of 3D point cloud perception, canopy modeling, and three-section arm contour-following control to achieve adaptive adjustment of thinning parameters and high-precision, stable execution. The invention utilizes 3D point cloud data to accurately perceive the canopy structure and flower spike distribution, and constructs a 3D canopy envelope surface to provide a geometric reference for the three-section arm's contour-following movement along the canopy. Simultaneously, the flower thinning execution component can dynamically adjust working parameters based on local flower spike density and canopy geometric features, ensuring efficient and uniform flower thinning operations in different canopy areas, achieving intelligent and automated flower thinning control.
[0078] It should also be noted that, in this specification, terms such as "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An intelligent shape-mimicking flower thinning system for the canopy of litchi trees, characterized in that, It includes an information sensing module, a data analysis module, and a job execution module; The information perception module includes a three-dimensional point cloud acquisition unit for the canopy of litchi trees and a point cloud semantic segmentation unit. The three-dimensional point cloud acquisition unit for the canopy of litchi trees is used to acquire three-dimensional point cloud data of the canopy of litchi trees, providing basic data support for subsequent analysis. The point cloud semantic segmentation unit uses a deep neural network model based on PointNet++ to perform semantic segmentation on the acquired point cloud data and identify the regions of flower spikes, branches and leaves. The data analysis module consists of a canopy 3D envelope surface construction and inflorescence density analysis unit. Based on the segmented inflorescence point cloud and branch and leaf point cloud, the canopy 3D envelope surface construction and inflorescence density analysis unit constructs the fruit tree canopy 3D envelope surface; on the fruit tree canopy 3D envelope surface, the local inflorescence distribution density and local geometric features are calculated; The operation execution module is a three-section arm adaptive contour thinning mechanism, which can accurately adapt to the curved surface of the fruit tree canopy to achieve adaptive contouring. By closely following the three-dimensional envelope surface of the canopy, it can smoothly and efficiently complete the precise flower thinning operation.
2. The intelligent shape-mimicking flower thinning system for litchi tree canopy as described in claim 1, characterized in that, The three-dimensional point cloud acquisition unit for the canopy of litchi trees is specifically a stereo vision array composed of depth cameras. The collected point cloud data is N×6 dimensional multimodal data, including: Three-dimensional spatial coordinates are used to describe the spatial relationship between various points in the fruit tree canopy; Color information is used to provide color characteristics of the fruit tree canopy. Where N is the number of points in the point cloud.
3. The intelligent contour-following flower thinning system for litchi tree canopies according to claim 1, characterized in that, The collected point cloud data is semantically segmented using a deep neural network model based on PointNet++, including the following steps: Model pre-training: The PointNet++ model is pre-trained on an existing 3D point cloud dataset of fruit trees with similar structures and morphologies to learn the general macroscopic features of fruit tree structure and morphology. Freeze the macroscopic feature extraction layer; after pre-training, freeze the network layer in the PointNet++ model used to extract macroscopic geometric features in order to retain its ability to perceive and analyze the overall structure of the fruit tree; Fine-tuning the micro-feature extraction layer; only fine-tuning the parameters of the small-scale feature extraction layer related to the inflorescence spike is performed to enable the PointNet++ model to adapt to the inflorescence spike characteristics of different fruit tree varieties and canopy structures. Output semantic classification results; use the trained PointNet++ model to output the semantic classification results of the point cloud, including flower spikes, branches and leaves.
4. The intelligent contour-following flower thinning system for litchi tree canopy as described in claim 1, characterized in that, The canopy 3D envelope surface construction and inflorescence density analysis unit first constructs the 3D envelope surface of the fruit tree canopy, and then calculates the local inflorescence distribution density and local geometric features, specifically: After completing the semantic segmentation of the point cloud, the point cloud of the outer contour of the fruit tree canopy is extracted and a continuous and closed three-dimensional envelope surface of the fruit tree canopy is constructed by using the Poisson surface reconstruction method. The three-dimensional envelope surface of the fruit tree canopy is smoothed and its local curvature characteristics are calculated. At the same time, the distribution relationship of the flower spike point cloud in the surface space is combined to realize the accurate calculation of the number of flower spikes, spatial spacing and distribution density, thus providing accurate and stable canopy geometry and spatial distribution basis for flower thinning operations.
5. The intelligent shape-mimicking flower thinning system for litchi tree canopy as described in claim 1, characterized in that, The three-section arm adaptive contouring openwork mechanism consists of the first arm, the second arm, and the third arm from top to bottom. The main support rod is vertically fixed to the mobile operating platform or tracked chassis; the first, second, and third arm sections are connected to different height positions of the main support rod via rotating joints; each arm section is equipped with a flower thinning execution component at its far end, which can be used independently for different canopy areas of the fruit tree; each rotating joint adopts an upper and lower fixed support structure to enhance impact resistance and prevent axial movement or structural loosening during high-frequency flower thinning. A controllable drive unit is set at the joint connection point, and the rotation angle of the arm is adjusted to adapt to the shape of the fruit tree canopy.
6. The intelligent contour-following flower thinning system for litchi tree canopy as described in claim 5, characterized in that, The three-section arm adaptive contour-following stencil performs precise stenciling operations, including: Based on the perceived geometric features and phenotypic analysis results, the motion path, working angle, and flower thinning speed of the three-section adaptive contour thinning mechanism are dynamically calculated. Then, the three-section adaptive contour thinning mechanism is driven, and each section adjusts its posture independently under the guidance of the controllable drive unit. Together with the end flower thinning execution component, it realizes continuous conforming operation along the three-dimensional envelope surface of the tree canopy.
7. The intelligent contour-following flower thinning system for litchi tree canopy as described in claim 2, characterized in that, Replace the depth camera with: LiDAR; uses the principle of laser time-of-flight or phase difference ranging to directly acquire high-precision geometric point cloud data; Multi-view vision system: It uses two or more conventional RGB cameras to form a stereo vision array, calculates disparity map through feature point matching or semi-global matching algorithm, and then reconstructs the three-dimensional point cloud of fruit tree canopy.
8. The intelligent contour-following flower thinning system for litchi tree canopy as described in claim 1, characterized in that, Replace the PointNet++-based deep neural network with: Multidimensional feature point cloud processing networks, including but not limited to dynamic graph convolutional neural networks, PointCNN, RandLA-Net, and KPConv; Voxelization neural network based on dimension transformation: First, the original disordered point cloud data is projected or transformed into a regular three-dimensional voxel grid, and then a three-dimensional convolutional neural network is used for feature learning and classification. A multimodal information fusion perception model is adopted, which uses a deep learning model that simultaneously processes two-dimensional image features and three-dimensional point cloud features.
9. The intelligent contour-following flower thinning system for litchi tree canopy as described in claim 5, characterized in that, Replace the three-section arm adaptive contouring openwork mechanism with: Multi-stage linear telescopic mechanism; using sleeve-type or guide rail telescopic structure to replace some or all rotating joints; Multi-degree-of-freedom parallel fine-tuning mechanism: A small six-degree-of-freedom parallel platform is added to the end of the serial robotic arm. Utilizing the high rigidity and high frequency response characteristics of the parallel mechanism, the parallel platform realizes high-frequency and fine attitude fine-tuning of the end flower thinning execution component on the basis of the main arm completing a large range of contouring motion, thereby further improving the operation accuracy of complex interlaced flower spikes. Flexible continuum contouring mechanism: A flexible continuum mechanism that simulates an elephant trunk or snake-like structure replaces rigid links. It uses internal drive ropes or fluid pressure to change the curvature of the flexible body, achieving jointless continuous bending contouring.
10. A method for thinning flowers based on the intelligent contour-following thinning system according to any one of claims 1-9, characterized in that, Includes the following steps: S1. A depth camera installed at the end of the three-section arm adaptive contour thinning mechanism is used to collect three-dimensional point cloud data of the fruit tree canopy and obtain multimodal point cloud data containing three-dimensional spatial coordinates and color information. S2. Use a deep neural network model based on PointNet++ to perform semantic segmentation on the collected point cloud data to identify the flower spike, branch and leaf regions. S3. Based on the segmented flower spike point cloud and branch and leaf point cloud, construct a three-dimensional envelope surface of the fruit tree canopy to describe the geometric boundary and spatial morphology of the fruit tree canopy. S4. On the three-dimensional envelope surface of the fruit tree canopy, based on the target recognition results of the flower spikes, mark the local density of the flower spikes on the three-dimensional envelope surface of the canopy to obtain a heat map of the spatial distribution of flower spike density. S5. Based on the spatial distribution heat map of flower spike density obtained in step S4, dynamically adjust the thinning operation parameters, including the extension length of the three-section arm, the contour angle of the thinning rotor, and the rotor speed, to achieve thinning control in different canopy areas. S6. Control the three-section arm adaptive contour thinning mechanism to move adaptively and closely along the envelope surface of the fruit tree canopy to perform precise thinning operations.