Vegetable transplanting device based on multi-source radar data fusion and real-time path detection

Abstract

A vegetable transplanting device based on multi-source radar data fusion and real-time path recognition, including: - a motion module (1) that serves as a carrier of the device and is designed to move the entire device along a planned path; - planar solid-state laser radars (2) installed on both sides of the motion module (1) and configured to acquire two-dimensional high-density point cloud data of the lateral surrounding areas of the device; - a vegetable transplanting module (3) designed to receive the vegetable seedlings to be transplanted and to carry out the field transplanting; - a vegetable storage module (4) equipped with a vegetable transplanting tray and used to store the seedling containers with the vegetable seedlings to be transplanted; - a path real-time detection module (5) which is connected to the spherical 3D radar (6), the planar solid-state laser radars (2) and the motion module (1) and is configured to fuse the multi-source radar data and plan the transplant path in real time; - a spherical 3D radar (6) configured to acquire three-dimensional point cloud data of the field environment and to obtain information about terrain irregularities, obstacle distribution and spatial arrangement in the vicinity of the device; - a visual module (7) configured to provide visual feedback to the robot arm to adapt the gripping and transplanting pose of the robot arm and thus support highly precise, self-adapting vegetable transplanting operations; - a robot arm module (8) configured to automatically pick up the vegetable seedlings to be transplanted from the seedling containers.

Description

Technical area

[0001] The present utility model relates to the technical field of agricultural automation devices, and in particular to a vegetable transplanting device based on multi-source radar data fusion and real-time path recognition. State of the art

[0002] Transplanting vegetable plants is a key cultivation technique in modern vegetable production. It refers to the process of transplanting seedlings, previously grown in propagation facilities, into open fields or protected growing areas according to a specific planting distance and row spacing. This technique is widely used in the production of many fruit and leafy vegetables. Traditional manual transplanting of vegetable plants is associated with disadvantages such as high labor intensity, low productivity, uneven planting distances, and inconsistent planting depth. It therefore cannot meet the demands of large-scale and intensive cultivation. For this reason, transplanting equipment for vegetable plants is needed to increase transplanting efficiency.

[0003] The autonomous adaptability of current vegetable transplanting devices under complex field conditions is still insufficient. Particularly in unstructured work environments, these devices exhibit drawbacks such as low path recognition accuracy, weak obstacle avoidance, and significant deviations in transplant position. These deficiencies, in turn, negatively impact seedling survival rates and subsequent field management. Existing vegetable transplanting devices lack the application of multi-source radar data fusion technology. Multi-source radar data fusion can effectively overcome the limitations of individual sensors. It improves weather-resistant and highly accurate perception of furrow structures, obstacle profiles, and work paths through complementary information.For this reason, existing vegetable transplanting devices lack intelligent perception based on multi-source radar data fusion and real-time path recognition. They therefore cannot meet the higher demands of future smart agriculture for high reliability, strong environmental adaptability, and fully autonomous operation. Consequently, the present utility model of a vegetable transplanting device based on multi-source radar data fusion and real-time path recognition is proposed. Subject of the utility model

[0004] The purpose of this utility model is to address the problems set out in the background technology described above. To achieve this purpose, the following technical solution is proposed: A vegetable transplanting device based on multi-source radar data fusion and real-time path recognition comprises the following: a motion module that serves as a carrier for the device and is designed to move the entire device along a planned path; Planar solid-state laser radars installed on both sides of the motion module and configured to acquire two-dimensional high-density point cloud data of the lateral surrounding areas of the device; a vegetable transplanting module designed to take over the vegetable seedlings to be transplanted and to carry out the field transplanting; a vegetable storage module equipped with a vegetable transplanting tray and used to store the seedling containers with the vegetable seedlings to be transplanted; a path real-time detection module, each connected to the spherical 3D radar, the planar solid-state laser radars and the motion module, and configured to fuse the multi-source radar data and plan the transplant path in real time; a spherical 3D radar configured to capture three-dimensional point cloud data of the field environment and to obtain information about terrain irregularities, obstacle distribution and spatial arrangement in the vicinity of the device; a visual module configured to provide visual feedback to the robot arm to adjust the gripping and transplanting pose of the robot arm, thus supporting highly precise, self-adapting vegetable transplanting operations; a robot arm module configured to automatically pick up the vegetable seedlings to be transplanted from the seedling containers.

[0005] As a preferred technical solution of the present utility model, the path real-time detection module comprises a data preprocessing unit, a multi-source radar fusion unit, and a path planning unit.

[0006] As a preferred technical solution of the present utility model, the data preprocessing unit is configured to perform filtering, noise suppression and transformation of the coordinate systems into a uniform reference system for the three-dimensional point cloud data acquired by the spherical 3D radar and the two-dimensional point cloud data acquired by the planar solid-state laser radars.

[0007] As the preferred technical solution of the present utility model, the multi-source radar fusion unit is configured to perform feature matching and spatial alignment between the pre-processed three-dimensional and two-dimensional point cloud data in order to generate environmental perception data that integrate three-dimensional spatial information and planar detail information.

[0008] Advantages of the present utility model compared to the prior art: The present utility model fuses data from spherical 3D radar and planar solid-state laser radars. It acquires precise information about terrain irregularities across the entire field area, the distribution of distant obstacles, and the overall spatial arrangement of the field using the spherical 3D radar. It also captures high-density two-dimensional point cloud data on both sides of the replanting path using the planar solid-state laser radars to precisely detect nearby low obstacles and fine features of the furrow boundaries. After data preprocessing, feature matching, and spatial alignment by the multi-source radar fusion unit, environmental perception data is generated that integrates both comprehensive environmental information and local details.This eliminates the disadvantages of individual radars such as blind spots in the field of view and insufficient detail perception, achieves a comprehensive and highly precise perception of the transplanting work environment, and effectively avoids problems such as device collisions and deviations in transplanting positions due to undetected obstacles.

[0009] Based on fused, high-precision environmental perception data, the path planning unit can dynamically plan an optimal transplanting path, taking into account preset transplanting rules such as planting distance and row spacing. It then adjusts the movement path of the motion module in real time based on obstacles and terrain changes encountered in the field. The device is equipped with a six-degree-of-freedom articulated robot arm whose range of motion comprehensively covers the entire work area, from loading and unloading seedlings to field transplanting. Combined with the precise positional information from the visual module and the real-time path detection module, it can adjust the gripping and transplanting pose of the end effector in real time, ensuring that the roots of the vegetable seedlings are not damaged during unloading and that positional deviation during planting is kept to a minimum.

[0010] The motion module, the perception module, the path planning module, and the robot arm module of this utility model achieve complete cooperative coupling across the entire chain. The motion module automatically moves along the planned path, and the robot arm dynamically adjusts its working position to the transplanting rhythm without requiring repeated manual assistance with positioning and device adjustment.

[0011] The spherical 3D radar can precisely detect terrain irregularities in the field and transmit the data to the path planning unit. The path planning unit can dynamically adjust the driving speed and route based on the terrain. Together with the drive system of the drive wheels, it adapts stably to various complex terrain conditions such as sloping fields and soft ground, thus overcoming the limitations of traditional transplanting devices, which are only suitable for flat fields, and expanding the device's range of applications. Description of the drawing Fig. Figure 1 shows a structural view of the present utility model; list of reference symbols in the drawings:

[0012] 1. Motion module; 2. Planar solid-state laser radars; 3. Vegetable transplanting module; 4. Vegetable placement module; 5. Real-time path detection module; 51. Data preprocessing unit; 52. Multi-source radar fusion unit; 53. Path planning unit; 6. Spherical 3D radar; 7. Visual module; 8. Robot arm module. Examples of implementation

[0013] In order to clarify the purposes, technical solutions and advantages of the embodiments of the present utility model, the technical solutions in the embodiments of the present utility model are described clearly and completely below with reference to the drawings.

[0014] Exemplary embodiment 1: A vegetable transplanting device based on multi-source radar data fusion and real-time path recognition comprises the following: a motion module 1, which serves as a carrier of the device and is designed to move the entire device along a planned path; planar solid-state laser radars 2, installed on both sides of the motion module 1 and configured to acquire two-dimensional high-density point cloud data of the lateral surrounding areas of the device; a vegetable transplanting module 3, which is set up to take over the vegetable seedlings to be transplanted and to carry out the field transplanting; a vegetable storage module 4, which is equipped with a vegetable transplanting tray and serves to store the seedling containers with the vegetable seedlings to be transplanted; a path real-time detection module 5, each connected to the spherical 3D radar 6, the planar solid-state laser radars 2 and the motion module 1, and configured to fuse the multi-source radar data and plan the transplant path in real time; a spherical 3D radar 6 configured to capture three-dimensional point cloud data of the field environment and to obtain information about terrain irregularities, obstacle distribution and spatial arrangement in the vicinity of the device; a visual module 7 configured to provide visual feedback to the robot arm to adapt the gripping and transplanting pose of the robot arm and thus support highly precise, self-adapting vegetable transplanting operations; a robot arm module 8, which is configured to automatically pick up the vegetable seedlings to be transplanted from the seedling containers.

[0015] The components of the device of the invention are listed in Table 1, including its core components. Table 1 module Key components Movement module 1 drive wheels, drive motor Area solid-state laser radars 2 Area laser emission and reception unit Vegetable transplanting module 3 Transplanting execution mechanism Vegetable storage module 4 Vegetable transplanting tray Path real-time detection module 5 Data processing unit 51, multi-source radar fusion unit 52, path planning unit 53 Spherical 3D radar 6 3D scanning radar unit Visual Module 7 Industrial image acquisition unit Robot arm module 8 Six-degree articulated robot arm, end effector

[0016] Operating principle of the vegetable transplanting device based on multi-source radar data fusion and real-time path recognition: The motion module 1 carries the other functional units of the device. Its drive wheels are connected to a drive motor and receive path commands from the path real-time recognition module 5. The drive motor converts electrical energy into mechanical energy and drives the drive wheels into rotation, thereby moving the entire device along the planned path and providing a stable mobile platform base for the transplanting operations.

[0017] The spherical 3D radar 6 continuously transmits and receives 3D detection signals to scan and capture three-dimensional point cloud data of the field environment. By analyzing the spatial coordinates and reflection intensity of the point clouds, information about terrain irregularities, obstacle distribution, and the spatial arrangement of crops in the vicinity of the device is reconstructed, forming the basis for a global 3D spatial perception of the work environment.

[0018] The planar solid-state laser radars 2 are installed on both sides of the motion module 1. They scan the areas on both sides of the device with high-frequency laser beams and generate two-dimensional, high-density point cloud data. Thanks to the advantage of high-density point clouds, they capture precise details such as nearby low obstacles and furrow boundaries on the transplanting path, compensating for the low resolution of the 3D radar in near-field plane perception.

[0019] The path real-time detection module 5 serves as the central control unit of the device. It couples the radar perception units and the mobile execution unit and performs data processing and path planning in three steps: Data preprocessing unit 51: A filter algorithm is executed on the three-dimensional point cloud data input from the spherical 3D radar 6 to remove ambient noise and ineffective reflection points. Simultaneously, the spherical coordinate system of the 3D point clouds is uniformly converted into the device's global coordinate system. Deduplication and smoothing processing are performed on the two-dimensional point cloud data from the planar solid-state laser radars 2 on both sides, and the mapping of planar coordinates to the global coordinate system is completed to output standardized perception data with a uniform format and removed noise.

[0020] Multi-source radar fusion unit 52: Feature matching is performed on the two pre-processed data types. The terrain features in the 3D point clouds and the path boundary features in the 2D point clouds are extracted. Using a spatial alignment algorithm, the detailed information from the 2D plane is embedded into the 3D spatial framework to generate fused environmental perception data with both global spatial field and local plane details, and to fully reconstruct the fieldwork environment.

[0021] Path Planning Unit 53: Based on the fused environmental perception data and taking into account preset transplanting rules such as row spacing, planting distance, and obstacle avoidance, a dynamic path planning algorithm is applied to generate a real-time transplanting path. When temporary obstacles are detected, a detour route is automatically planned, and simultaneously, adjustment commands are sent to the drive motor of motion module 1 to dynamically correct the speed and direction of the motion module, ensuring that the device always travels along the optimal transplanting path.

[0022] The subunits of the path real-time detection module 5 are listed in Table 2 along with their input sources. Table 2 subunit Input data source Data preprocessing unit 51 3D point cloud data from the spherical 3D radar 6; 2D point cloud data from the planar solid-state laser radars Multi-source radar fusion unit 52 Standardized point cloud data output by data preprocessing unit 51 Path planning unit 53 Fused environmental perception data output by the Multi-Source Radar Fusion Unit 52; preset transplant work rules

[0023] The path real-time detection module 5 serves as the central decision unit of the vegetable transplanting device and performs the functions of multi-source radar data fusion and dynamic path planning for transplanting. Internally, it contains the data preprocessing unit 51, the multi-source radar fusion unit 52, and the path planning unit 53. The individual units cooperate to implement a complete chain from environmental perception to path output. The specific workflow and principle are as follows: Data preprocessing unit 51: Standardization of the multi-source radar data: The data preprocessing unit 51 receives the three-dimensional point cloud data of the field environment, acquired by the spherical 3D radar 6, as well as the two-dimensional planar high-density point cloud data of the areas on both sides of the device, acquired by the planar solid-state laser radars 2.The two raw data types differ in format, coordinate basis, and noise characteristics and cannot be directly merged, so standardization conversion through preprocessing is required.

[0024] The three-dimensional point cloud data from the spherical 3D radar 6 is susceptible to interference from factors such as weeds, insects, and soil dust in the field environment. The data preprocessing unit 51 first applies a statistical filtering algorithm: the average distance of each point to its neighboring points is calculated, and isolated noise points with a distance deviation from the mean of more than two standard deviations are removed to eliminate the interference of random individual spots on terrain and obstacle perception. Simultaneously, voxel grid filtering divides the 3D space into voxel grids of a defined size. For each grid, the point closest to the grid center is retained. Assuming the preservation of terrain irregularities and obstacle profile features, the amount of point cloud data is reduced, and the efficiency of subsequent processing is improved.

[0025] For the two-dimensional point cloud data of the planar solid-state laser radars 2, the main disturbances originate from small soil particles in the transplant furrows and reflection points of the radar installation supports. Unit 51 applies a radius filter algorithm: a neighbor radius and a minimum number of neighbor points are defined, and discrete points with a number of neighbor points below the threshold are removed. This accurately preserves the detailed features of path boundaries and nearby obstacles and prevents soil particle mixing points from being falsely identified as obstacles.

[0026] Due to the different installation positions of the spherical 3D radar 6 and the planar solid-state laser radars 2, their raw point cloud data are output based on their respective local coordinate systems and cannot be directly spatially aligned. Unit 51 performs the transformation of the coordinate systems into a unified reference system using preset external calibration results: The geometric center of the motion module 1 is defined as the origin of the global coordinate system. The 3D point cloud data of the spherical 3D radar 6 are converted into the global coordinate system using a rotation matrix and translation vector. The 2D point cloud data of the planar solid-state laser radars 2 are mapped to the corresponding plane of the global coordinate system by height completion (adjusting for the radar installation height) and coordinate rotation translation.Finally, the alignment of the two point cloud data types is implemented in the same coordinate system.

[0027] Multi-Source Radar Fusion Unit 52: Feature Matching and Spatial Information Fusion: The preprocessed 3D point cloud data can fully reconstruct terrain irregularities and the spatial distribution of obstacles in the field, but exhibit low accuracy in the detailed perception of path boundaries and small obstacles in furrows. The preprocessed 2D point cloud data has extremely high plane resolution and can accurately capture furrow edge profiles and small nearby obstacles on both sides of the transplant path, but lacks elevation information and is unable to detect three-dimensional obstacles such as raised mounds of earth and fallen straws. The Multi-Source Radar Fusion Unit 52 achieves the combined benefits of both through feature matching and spatial alignment.

[0028] The multi-source radar fusion unit 52 first performs feature extraction from both point clouds: For the 3D point clouds, normal features, curvature features, and key points are extracted to identify areas of terrain slope changes and the topside profiles of obstacles. For the 2D point clouds, edge features and vertex features are extracted to mark path boundary lines and the planar positions of small obstacles. A random sample consensus (RANSAC) algorithm is then applied for feature matching: Based on the path boundary lines identified by the 2D point clouds, the elevation data of the corresponding positions in the 3D point clouds are matched, adding elevation attributes to the planar boundary information.At the same time, based on the position data of the three-dimensional obstacles identified by the 3D point clouds, the detail features of the 2D plane of these positions are matched in the 2D point clouds, so that the contour information of the obstacles is completed.

[0029] After feature matching is complete, the multi-source radar fusion unit 52 uses the least-squares method to trigger the spatial transformation matrix of the two point clouds. This further optimizes the accuracy of the coordinate alignment and ensures that the plane projection of the 3D point clouds fully matches the detail features of the 2D point clouds. Finally, environmental perception data is generated, integrating three-dimensional spatial information and planar detail information. This data includes the elevation irregularities of the field terrain and the spatial distribution of three-dimensional obstacles, as well as the path boundaries of the transplanting path and the planar details of small obstacles in the furrows, thus providing a complete environmental perception basis for path planning.

[0030] Path Planning Unit 53: Dynamic transplant path planning and drive control. The input for Path Planning Unit 53 comprises two parts: firstly, the fused environmental perception data output by the multi-source radar fusion unit 52, and secondly, the preset transplanting work rules, including constraint conditions such as the required row spacing in the furrows, the planting distance of the seedlings, and the safety distance for obstacle avoidance. Simultaneously, the working area of ​​the robot arm module 8 is taken into account to ensure that the transplanting work area covered by the planned path lies within the range of motion of the articulated robot arm with six degrees of freedom.

[0031] Path planning unit 53 first converts the fused environmental perception data into a two-dimensional raster map, assigning each raster to one of three categories: the flat and obstacle-free areas are marked as traversable areas; the three-dimensional obstacles and the areas outside the path boundaries are marked as obstacle areas; the preset furrow positions are marked as transplanting work areas.

[0032] Based on the preset transplant furrow layout, the path planning unit 53 generates an initial transplant path and ensures that the path extends linearly along the furrows and meets the planting distance requirements. The initial path is simultaneously transmitted to the motion module 1, and the drive motor powers the drive wheels to travel along the planned path.

[0033] During the transplanting operation, the path planning unit 53 receives real-time updates of the fused environmental perception data. If non-predefined obstacles, such as fallen straw or raised mounds of soil, are detected in the furrows, the path planning unit 53 applies the dynamic window method to perform a local adjustment of the initial path: starting from the current position of motion module 1, several feasible local paths are generated, provided that obstacles are safely avoided. Based on the transplanting work rules, the optimal path is selected, which both meets the obstacle avoidance requirements and is as close as possible to the original furrow. The adjusted path is simultaneously transmitted to the drive motor of motion module 1 to prevent the robot arm module 8 from colliding with obstacles during the operation.

[0034] The path planning unit 53 sends the planting coordinates along the planned path to the robot arm module 8 in real time. Based on the received coordinates, the robot arm adjusts the pose of its end effector gripper and combines the visual position information from the visual module 7 to precisely grasp the vegetable seedlings from the vegetable transplanting trays of the vegetable storage module 4. It then moves to the corresponding transplanting position and performs the planting operation, thus achieving dynamic coordination between the robot arm's path and transplanting movement.

[0035] The foregoing embodiments serve only to illustrate the present utility model and not to limit the described technical solutions. Although the present description of the present utility model has been detailed with reference to the foregoing embodiments, the present utility model is not limited to the foregoing specific embodiments. Consequently, all modifications or equivalent subdivisions of the present utility model, as well as all technical solutions and improvements that do not depart from the spirit and scope of the utility model, fall within the scope of protection of the claims of the present utility model.

Claims

[1] A vegetable transplanting device based on multi-source radar data fusion and real-time path detection, comprising: - a motion module (1) that serves as a carrier of the device and is designed to move the entire device along a planned path; - planar solid-state laser radars (2) installed on both sides of the motion module (1) and configured to acquire two-dimensional high-density point cloud data of the lateral surrounding areas of the device; - a vegetable transplanting module (3) designed to receive the vegetable seedlings to be transplanted and to carry out the field transplanting; - a vegetable storage module (4) equipped with a vegetable transplanting tray and used to store the seedling containers with the vegetable seedlings to be transplanted; - a path real-time detection module (5) which is connected to the spherical 3D radar (6), the planar solid-state laser radars (2) and the motion module (1) and is configured to fuse the multi-source radar data and plan the transplant path in real time; - a spherical 3D radar (6) configured to acquire three-dimensional point cloud data of the field environment and to obtain information about terrain irregularities, obstacle distribution and spatial arrangement in the vicinity of the device; - a visual module (7) configured to provide visual feedback to the robot arm to adapt the gripping and transplanting pose of the robot arm and thus support highly precise, self-adapting vegetable transplanting operations; - a robot arm module (8) configured to automatically pick up the vegetable seedlings to be transplanted from the seedling containers. [2] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 1, characterized by , that the path real-time detection module (5) comprises a data preprocessing unit (51), a multi-source radar fusion unit (52) and a path planning unit (53). [3] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 1, characterized by , that the data preprocessing unit (51) is configured to perform filtering, noise reduction and transformation of the coordinate systems into a unified reference system for the three-dimensional point cloud data acquired by the spherical 3D radar (6) and the two-dimensional point cloud data acquired by the planar solid laser radars (2). [4] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 1, characterized by, that the multi-source radar fusion unit (52) is configured to perform feature matching and spatial alignment between the pre-processed three-dimensional and two-dimensional point cloud data to generate environmental perception data that integrate three-dimensional spatial information and planar detail information. [5] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 1, characterized by , that the path planning unit (53) is configured to plan the transplanting path in real time based on the fused environmental perception data and taking into account preset transplanting work rules and to dynamically adjust the movement path of the movement module (1). [6] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 1, characterized by, that the motion module (1) is equipped with drive wheels which are connected to a drive motor. [7] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 2, characterized by , that the robot arm module (8) uses an articulated robot arm with six degrees of freedom, the range of motion of which covers the vegetable storage module (4), the vegetable transplanting storage plate and the movement path of the motion module (1) corresponding to the transplanting work zone, and which is able to adjust the pose of the end effector according to the transplanting position output by the path real-time detection module (5). [8] The vegetable transplanting device based on multi-source radar data fusion and real-time path recognition according to claim 7, characterized by , that the robot arm module (8) is equipped with the visual module (7).

Images