An underground vine stem soil cleaning device and a position accurate detection method

Abstract

The application belongs to the technical field of vine trunk soil cleaning, and discloses a soil-under vine trunk soil cleaning device which comprises a portal walking frame, a soil cleaning mechanism, a collecting mechanism and a main server, the portal walking frame is arranged across the soil ridge of the vine trunk, the soil cleaning mechanism is arranged at one end of the portal walking frame, the collecting mechanism comprises an RGBD camera and a high-frequency ground penetrating radar, the RGBD camera is used for acquiring the basic parameters of the grapevine before soil burying and the surrounding environment parameters, the high-frequency ground penetrating radar is used for acquiring the scanning parameters of the grapevine after soil burying, the main server comprises a data processing module, a coordinate system summary module, a deformation fitting module and an execution module, the data processing module is used for outputting a three-dimensional point cloud model and ground penetrating radar three-dimensional data body, the coordinate system summary module executes a unified coordinate model and outputs the same frame parameters, the deformation fitting module executes deformation fitting and outputs the accurate position of the grapevine after soil burying, and the execution module executes the soil cleaning instruction. The problems of strong subjectivity and low positioning accuracy of grapevine soil cleaning are solved.

Description

Technical Field

[0001] This invention belongs to the field of soil clearing technology, specifically relating to an underground soil clearing device and a method for accurate location detection. Background Technology

[0002] In grape cultivation in cold regions, to prevent frost damage to the main trunk, branches, and buds of the grapevines during winter, the vines are usually manually pressed down and covered with soil before winter, and then unearthed in the spring when temperatures rise. After the vines are buried, their main trunk and branches are located inside the soil layer and cannot be directly observed.

[0003] Existing methods for determining the location of buried grapevines mostly rely on manual experience. They are usually roughly estimated based on the location of the grape row, past burying habits, and local soil morphology, and then the soil is cleared by machinery or manual labor.

[0004] However, the above methods for clearing soil from grapevines are highly subjective, have low positioning accuracy, and cannot accurately reflect the actual distribution of grapevines underground. Especially after grapevines are manually pressed down, their spatial posture will bend and redistribute, making it difficult to determine their true extension path and boundary range based solely on experience. Summary of the Invention

[0005] Based on this, this application provides a soil-drying device for grapevines and a method for accurate location detection, in order to solve the problems that soil-drying of grapevines is highly subjective, has low positioning accuracy, and is difficult to accurately reflect the actual distribution of grapevines under the soil. Furthermore, after grapevines are manually pressed down, their spatial posture will be bent and redistributed, making it difficult to determine their true extension path and boundary range based solely on experience.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows: A soil-clearing device for grapevines includes: a portal frame, a soil-clearing mechanism, a data acquisition mechanism, and a main server. The portal frame spans the soil ridges of the grapevines and can move in a preset direction. The soil-clearing mechanism is located at the end of the portal frame away from the direction of travel and can move vertically and horizontally to clear the soil from the grapevines. The data acquisition mechanism includes an RGBD camera and a high-frequency ground-penetrating radar. The RGBD camera is located at the end of the portal frame away from the soil-clearing mechanism and is used to acquire basic parameters of the grapevines and surrounding environmental parameters before burying. The high-frequency ground-penetrating radar is located in the middle of the portal frame, between the RGBD camera and the soil-clearing mechanism, and is used to acquire scanning parameters of the grapevines after burying. The main server includes a data processing unit. The system comprises a data processing module, a coordinate system aggregation module, a deformation fitting module, and an execution module. The data processing module is connected to the RGBD camera and the high-frequency ground-penetrating radar, respectively, and is used to process the basic parameters of the grapevines before burying, the surrounding environment parameters, and the scanning parameters of the grapevines after burying, and output a 3D point cloud model and a 3D ground-penetrating radar data volume. The coordinate system aggregation module is connected to the data processing module, and based on the 3D point cloud model and the 3D ground-penetrating radar data volume, executes a unified coordinate model and outputs frame parameters. The deformation fitting module is connected to the coordinate system aggregation module, and based on the frame parameters, performs deformation fitting to establish a fine underground model and outputs the precise position of the grapevines after burying. The execution module is connected to the deformation fitting module, and based on the precise position of the grapevines after burying, executes a soil-clearing command.

[0007] Preferably, the RGBD camera is an Orbbec Gemini335L, and the high-frequency ground-penetrating radar is an SIR4000 system.

[0008] A method for accurately detecting the location of underground vine trunks, used to execute the underground vine trunk clearing device described in any of the above-mentioned methods, includes the following steps: Step S10: Obtain the basic parameters of the grapevines and the surrounding environment parameters before burying them in the soil; Step S20: Based on the local coordinate system, establish a three-dimensional point cloud model and output the basic center line of the grapevine before burying it in the soil; Step S30: Obtain surrounding environmental parameters and scanning parameters of grapevines after burying in soil to form a regular control grid; Step S40: Based on the rule-based control grid, construct the three-dimensional data volume of the ground penetrating radar and extract the guided target volume; Step S50: Based on the 3D point cloud model, the 3D data volume of ground penetrating radar and the surrounding environmental parameters, establish a unified coordinate model and output the frame parameters. Step S60: Based on the overall deformation fitting module, input the frame parameters to perform deformation fitting, and output the fine underground model; Step S70: Based on the detailed underground model, output the precise location of the grapevines after burying them in the soil.

[0009] Preferably, step S20 further includes establishing a local coordinate system, which includes the following steps: Step S21: Define the grape planting direction as the X-axis; Step S22: Define the direction perpendicular to the grape planting direction as the Y-axis; Step S23: Define the surface normal direction as the Z-axis; Step S24: Define the Earth's surface as a reference surface with Z=0.

[0010] Preferably, in step S40, guiding the target body includes the following steps: Step S41: Perform threshold segmentation based on the three-dimensional data volume of the ground penetrating radar, and determine whether the radar response intensity is not less than the segmentation threshold. If so, mark it to form a set of high-response voxels; Step S42: Based on the high-response voxel set, perform three-dimensional connected component labeling to obtain multiple candidate connected target volumes; Step S43: Determine whether the candidate connected target volume meets the preset threshold: If so, the candidate connected target that meets the preset threshold will be marked as the guiding target corresponding to the grapevine.

[0011] Preferably, the deformation fitting of the overall deformation fitting module in step S60 includes the following steps: Step S61: Using the root ground coordinates as fixed anchor points, establish a regular control mesh outside the 3D point cloud model; Step S62: Using the basic centerline as the driving skeleton and the target centerline as the fitting target, iteratively update the control points of the rule control grid according to the spatial extension direction and path distribution of the target centerline. Step S63: Based on the control points of the updated rule control mesh, recalculate the centerline of the deformed model; Step S64: Determine the maximum spatial deviation between the centerline of the deformed model and the target centerline: If the error exceeds the preset position error threshold, or if the difference between the length of the deformed model's centerline and the target centerline exceeds the preset length error threshold, then continue iteratively updating the control points. If the error is less than or equal to the preset position error threshold, and the difference between the length of the centerline of the deformed model and the target centerline is less than or equal to the preset position error threshold, then the overall model deformation fitting is terminated.

[0012] Preferably, the precise location of the grapevine after burying in step S70 includes: the fixed starting coordinates of the roots on the ground, the location of the center path under the soil, the range of the outer contour of the soil, the range of the burial depth, and the range of the overall spatial distribution.

[0013] Preferably, the construction of the 3D point cloud model in step S20 includes: invalid point removal, outlier removal, voxel downsampling, background separation, multi-frame coarse registration and global fine registration, and point cloud stitching.

[0014] The technical solution adopted in this application can achieve the following beneficial effects: 1. To address the issues that the clearing of grapevines is subjective and has low positioning accuracy, making it difficult to accurately reflect the actual distribution of grapevines underground. Furthermore, after grapevines are manually pressed down, their spatial posture will be bent and redistributed, making it difficult to determine their true extension path and boundary range based solely on experience.

[0015] 2. After the grapevines are buried, relying solely on ground-penetrating radar (GPR) 3D data is insufficient to accurately determine their true location beneath the soil. This is because GPR data is easily affected by soil conditions and stray reflections, leading to blurred spatial boundaries, local distortions, and positional deviations of the underground target. While the pre-burial 3D point cloud model possesses high geometric accuracy, it cannot be directly equated to the post-burial underground morphology, as the grapevines undergo overall bending and posture changes during artificial compaction and covering. Therefore, this application effectively combines the pre-burial 3D point cloud model with the post-burial GPR 3D data volume to achieve precise location detection of the buried grapevines. Simultaneously, it provides the original high-precision geometric morphology of the grapevines, root ground coordinates, and overall length information, offering a reliable foundation for accurate location detection after burial.

[0016] 3. By extracting the target volume and target centerline from the three-dimensional data volume of ground penetrating radar, the approximate extension direction and path distribution of the grapevines under the soil after burying are obtained, thus providing effective underground morphological guidance information for the deformation of the overall model. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall soil-clearing device described in this application. Figure 1 .

[0018] Figure 2 For this application Figure 1 Enlarged view of part A.

[0019] Figure 3 This is a schematic diagram of the overall soil-clearing device described in this application. Figure 2 .

[0020] Figure 4 This is a flowchart of the method for accurately detecting the location of the stem of the *Ipomoea purpurea* described in this application.

[0021] In the diagram: gantry frame 100, soil-cleaning wheel 210, drive motor 220, vertical movement component 230, telescopic rod 231, sleeve rod 232, connector 233, drive component 234, horizontal movement component 240, fixing component 241, elastic component 242, Z-shaped plate 243, rotating shaft 244, auxiliary sleeve 245, rotating motor 246, partition plate 247, acquisition mechanism 300, RGBD camera 310, high-frequency ground penetrating radar 320. Detailed Implementation

[0022] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0023] It should be noted that when an element is referred to as being "set on" another element, it can be directly on the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," "top," "bottom," "end," "top," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0025] Please see Figures 1 to 3This application provides a soil-clearing device for underground vine trunks, comprising: a portal frame 100, a soil-clearing mechanism, a data acquisition mechanism 300, and a main server. The portal frame 100 spans the soil ridge of the vine trunk and can move in a preset direction. The soil-clearing mechanism is located at the end of the portal frame 100 away from the direction of travel and can move in both vertical and horizontal directions to clear the soil on the vine trunk. The data acquisition mechanism 300 includes an RGBD camera 310 and a high-frequency ground-penetrating radar 320. The RGBD camera 310 is located at the end of the portal frame 100 away from the soil-clearing mechanism and is used to acquire basic parameters of the grapevine and surrounding environmental parameters before burying. The high-frequency ground-penetrating radar 320 is located in the middle of the portal frame 100, between the RGBD camera 310 and the soil-clearing mechanism, and is used to acquire data on the grapevine after burying. The main server includes a data processing module, a coordinate system aggregation module, a deformation fitting module, and an execution module. The data processing module is connected to the RGBD camera 310 and the high-frequency ground-penetrating radar 320, respectively, and is used to process the basic parameters of the grapevines before burying, the surrounding environment parameters, and the scanning parameters of the grapevines after burying, and output a three-dimensional point cloud model and a three-dimensional data volume of the ground-penetrating radar. The coordinate system aggregation module is connected to the data processing module, and based on the three-dimensional point cloud model and the three-dimensional data volume of the ground-penetrating radar, executes a unified coordinate model and outputs frame parameters. The deformation fitting module is connected to the coordinate system aggregation module, and based on the frame parameters, performs deformation fitting to establish a fine underground model and outputs the precise position of the grapevines after burying. The execution module is connected to the deformation fitting module, and based on the precise position of the grapevines after burying, executes a soil-clearing command.

[0026] Specifically, the gantry frame 100 adopts, but is not limited to, tracked or wheeled movement methods, and spans across the grapevine ridges; the height of the gantry frame 100 is adjusted according to site requirements, and the specific height is adjusted according to the surrounding environment, the height of the cement pillars, etc.; the height adjustment of the gantry frame 100 is achieved by, but is not limited to, telescopic rods 231, hydraulic jacks, sliding rods, etc.; the direction of movement is the same as the direction of the grapevine burying.

[0027] The soil removal mechanism includes a soil removal wheel 210, a vertical moving component 230, and a horizontal moving component 240. Two vertical moving components 230 are provided and symmetrically arranged on both sides of the portal frame 100. The horizontal moving component 240 is rotatably connected to the end of the vertical moving component 230 near the ground, and the vertical moving component 230 can drive the horizontal moving component 240 to move up and down vertically. The soil removal wheel 210 is rotatably located at the end of the horizontal moving component 240 away from the vertical moving component 230, and the horizontal moving component 240 can drive the soil removal wheel 210 to rotate horizontally, and the soil removal wheel 210 can rotate and remove soil.

[0028] The vertical moving component 230 includes a telescopic rod 231, a sleeve rod 232, and a connecting member 233. The connecting member 233 is fixedly connected to the gantry frame 100, and the sleeve rod 232 is rotatably connected to the connecting member 233, with the end of the sleeve rod 232 away from the connecting member 233 fixedly connected to the gantry frame 100. A driving member 234 is provided on the connecting member 233, which is connected to the telescopic rod 231 for driving the telescopic rod 231 to move up and down. The telescopic rod 231 can extend or retract by, but is not limited to, a screw engaging with the driving member 234. The sleeve rod 232 has an internal thread, and the screw is rotated by the driving member 234 to achieve the purpose of vertical extension or retraction. Alternatively, hydraulic or pneumatic jacks can be used directly.

[0029] The horizontal moving component 240 includes a fixing member 241, an elastic member 242, a Z-shaped plate 243, and a rotating shaft 244. The gantry frame 100 has slide rails on both sides. The fixing member 241 has a sliding groove, and the fixing member 241 is slidably connected to the gantry frame 100, with the slide rails and sliding grooves providing a limiting fit. The rotating shaft 244 is rotatably mounted within the fixing member 241, and a rotating motor 246 is connected to one end away from the vertical moving component. The rotating motor 246 drives the rotating shaft 244 to rotate. One end of the Z-shaped plate 243 is fixedly connected to the rotating shaft 244 and rotates horizontally under the drive of the rotating shaft 244. The elastic member 242 is sleeved on the rotating shaft 244 and is located between the top of the Z-shaped plate 243 and the fixing plate. Furthermore, an auxiliary sleeve 245 is fitted onto the rotating shaft 244, with the auxiliary sleeve 245 positioned relative to the rotating shaft 244 and capable of moving up and down along the rotating shaft 244. The rotating shaft 244 can drive the auxiliary sleeve 245 to rotate. A partition plate 247 is provided in the middle of the fixing member 241, with the rotating motor 246 located below the partition plate 247 and the auxiliary sleeve 245 located above the partition plate 247. A Z-shaped plate 243 is fixedly connected to the end of the auxiliary sleeve 245 near the partition plate 247, and an elastic member 242 is fitted onto the auxiliary sleeve 245. A soil-cleaning wheel 210 is rotatably positioned at the end of the Z-shaped plate 243 away from the auxiliary sleeve 245, and a drive motor 220 is provided between the soil-cleaning wheel 210 and the Z-shaped plate 243, driving the soil-cleaning wheel 210 to rotate.

[0030] The drive component 234 drives the telescopic rod 231 to move up and down, and the telescopic rod 231 drives the fixed component 241 to slide up and down along the portal frame 100, thereby realizing the vertical adjustment of the cleaning wheel 210; the rotating motor 246 is turned on, driving the rotating shaft 244 and the sleeved and limited auxiliary sleeve 245 to rotate, and the Z-shaped plate 243 on the auxiliary sleeve 245 drives the cleaning wheel 210 to rotate accordingly, thereby realizing the horizontal adjustment of the cleaning wheel 210; at the same time, the auxiliary sleeve 245 can slide in the vertical direction, and the elastic component 242 reduces the buffer, making the operation simpler and more convenient; the drive motor 220 is turned on, driving the cleaning wheel 210 to rotate, thereby achieving the purpose of cleaning soil.

[0031] The RGBD camera 310 in the data acquisition mechanism 300 is an Orbbec Gemini 335L, and the high-frequency ground-penetrating radar 320 is an SIR4000 system. The RGBD camera 310 is handheld and mounted on the gantry frame 100, with a data acquisition range covering the entire row of grapevines. The ground-penetrating radar is mounted above the buried grapevine ridge via a connecting component, located between the soil-clearing mechanism and the RGBD camera 310. The antenna frequency of the ground-penetrating radar is fixed at 1600 MHz, and the scanning method is a regular grid scan.

[0032] The data processing module converts the basic parameters of the grapevines before burial, acquired by the RGBD camera 310, into local point clouds frame by frame, and maps the color image information to corresponding 3D points to form color 3D point cloud data. Subsequently, it sequentially performs invalid depth point removal, outlier removal, voxel downsampling, background separation, coarse registration of adjacent frames, global fine registration, and multi-frame point cloud stitching, thereby obtaining a 3D point cloud model of the entire row of grapevines before burial. The data processing module then segments the scanning parameters of the grapevines after burial acquired by ground-penetrating radar, sequentially performing time zero-point correction, filtering, gain adjustment, and background removal. Finally, according to the scanning grid index, each 2D radar profile is superimposed onto a unified 3D voxel space to construct a 3D radar data volume of the burial area.

[0033] The coordinate system aggregation module first processes the surrounding environmental parameters to establish a local coordinate system. Then, it aggregates the 3D point cloud model output by the data processing module and the 3D data volume of the ground penetrating radar with the local coordinate system to establish a unified coordinate model and outputs the parameters within the same frame. Specifically, the coordinate system aggregation module performs threshold segmentation on the 3D data volume of the ground penetrating radar (the segmentation threshold is set by the operator), forming a high-response voxel set by constructing all spatial points that meet the radar response intensity. The high-response voxel set is then labeled with 3D connected components to obtain multiple candidate connected target bodies. Among all candidate connected target bodies, those that meet preset conditions are selected as the guiding target bodies corresponding to the grapevines. Based on the guiding target bodies, voxel skeleton refinement is performed to extract their 3D skeletons. Finally, spline fitting and smoothing are performed on the 3D skeletons to obtain the target centerline of the grapevines after burying them in the soil.

[0034] The deformation fitting module performs fitting based on the frame parameters and the target centerline. The fitting steps are as follows: (1) Using the root ground coordinates as fixed anchor points, establish a regular control grid outside the three-dimensional point cloud model before burying the soil; (2) Using the base centerline as the driving skeleton and the target centerline as the fitting target; (3) Iteratively update the control grid control points according to the spatial extension direction and path distribution of the target centerline; (4) After each update, recalculate the centerline of the deformed model; (5) Calculate the maximum spatial deviation between the deformed centerline and the target centerline; (6) If it is greater than the preset position error threshold, or the difference between the length of the deformed centerline and the target centerline is greater than the preset length error threshold, then continue iteratively updating the control points; (7) If it is less than or equal to, and the difference between the length of the deformed centerline and the target centerline is less than or equal to, then terminate the overall model deformation fitting. Thus, based on the fitting results, establish a fine underground model of the buried grapevine and output the precise position of the buried grapevine as needed. The execution module is electrically connected to the drive motor 220, the rotation motor 246 and the drive component 234, and is connected to the output end of the deformation fitting module. The execution module adjusts the opening and opening time of the drive motor 220, the rotation motor 246 and the drive component 234 by the precise position of the grapevine after burying the soil, so that the soil cleaning wheel 210 reaches the corresponding position and cleans the soil ridge.

[0035] Furthermore, before the grapevines are buried, the RGBD camera 310 is turned on and driven by the gantry frame 100 to perform the first detection. After the vines are buried, the high-frequency ground-penetrating radar 320 is turned on and driven by the gantry frame 100 to perform the second detection. At the same time, the soil-clearing mechanism is activated, and the soil-clearing work is completed under the control of the main server.

[0036] Please see Figure 4 This application also provides a method for accurately detecting the location of grapevine trunks underground, including the following steps: Step S10, obtaining the basic parameters of the grapevine and the surrounding environment parameters before burying; Step S20, establishing a three-dimensional point cloud model based on a local coordinate system and outputting the basic centerline of the grapevine before burying; Step S30, obtaining the surrounding environment parameters and the scanning parameters of the grapevine after burying, forming a regular control grid; Step S40, constructing a ground-penetrating radar three-dimensional data volume based on the regular control grid and extracting the guiding target volume; Step S50, establishing a unified coordinate model based on the three-dimensional point cloud model, the ground-penetrating radar three-dimensional data volume, and the surrounding environment parameters, and outputting the frame parameters; Step S60, performing deformation fitting based on the overall deformation fitting module, and outputting a fine underground model; Step S70, outputting the accurate location of the grapevine after burying based on the fine underground model. The three-dimensional point cloud model construction in step S20 includes: invalid point removal, outlier removal, voxel downsampling, background separation, multi-frame coarse registration and global fine registration, and point cloud stitching.

[0037] Specifically, the basic parameters include: the size, shape, and location of the grapevine itself, and the area of ​​the grapevine roots above ground; the surrounding environmental parameters include: fixed environmental information such as cement pillars in the field, row direction, and the ground surface; before burying the grapevines, the cement pillar at the beginning of the grape row and the supports set at predetermined intervals along the row are selected as fixed reference objects. Since the spatial position of the fixed reference objects remains unchanged before and after burying, a local coordinate system is established using the fixed reference objects.

[0038] In this application, the direction of the grape row is defined as the X-axis, the direction perpendicular to the grape row is defined as the Y-axis, the direction normal to the ground surface is defined as the Z-axis, and the ground surface is defined as the reference surface with Z=0. The subsequent 3D point cloud model before burial and the 3D data volume of the ground-penetrating radar after burial are both uniformly represented in this local coordinate system.

[0039] The acquisition device uses, but is not limited to, an RGBD camera 310. This application uses the RGBD camera 310 as an example, converting the depth map acquired by the RGBD camera 310 frame by frame into a local point cloud, and mapping the color image information to the corresponding three-dimensional points to form color three-dimensional point cloud data. Subsequently, invalid depth point removal, outlier removal, voxel downsampling, background separation, coarse registration of adjacent frames, fine global registration, and multi-frame point cloud stitching are performed sequentially to obtain a three-dimensional point cloud model of the entire row of grapevines before burying, and the central skeleton line of the vine trunk (i.e., the foundation center line) is extracted. In the three-dimensional point cloud model P0 before burying, the area where the roots of each grapevine grow from the ground is identified, and the coordinates of the roots at the ground are extracted. In this invention, the ground coordinates of the roots are denoted as r0, and r0 serves as a fixed anchor point during the deformation process of the overall model.

[0040] Subsequently, the main path or central skeleton of the grapevine's main point cloud is extracted, and spline fitting and smoothing are performed on the obtained main path or central skeleton to form the basic centerline of the grapevine before burying. In this invention, the basic centerline is denoted as C0(s), where s is the arc length parameter along the centerline, and L0 is the total length of the basic centerline. The total length L0 of the basic centerline is calculated, and the overall length Lp of the entire grapevine is statistically analyzed. In this invention, L0 serves as a constraint for the conservation of the main path length, and Lp serves as a constraint for the overall length to remain essentially unchanged.

[0041] In addition, the overall outer contour range, overall extension direction, and relative spatial relationship with the fixed reference object of the three-dimensional point cloud model before burial are extracted as auxiliary constraint information for subsequent spatial alignment and overall model deformation.

[0042] After the grapevines were manually pressed down and covered with soil along the row direction, a high-frequency ground-penetrating radar 320 was used to perform a regular grid scan of the target area. The scanned area was laid out based on fixed reference objects to form a regular control grid.

[0043] In this invention, the high-frequency ground-penetrating radar 320 is fixed as an SIR4000 system, the antenna frequency is fixed at 1600 MHz, and the scanning method is fixed as regular grid scanning. Since the ground-penetrating radar in this scheme does not record continuous position information, continuous trajectory positioning is not used. Instead, a regular grid layout is used to ensure that each two-dimensional radar profile has a clear row and column index and a relative spatial position.

[0044] The two-dimensional radar profile data acquired after burying the soil were sequentially processed with time zero-point correction, filtering, gain adjustment, and background removal. Then, the two-dimensional radar profiles were superimposed onto a unified three-dimensional voxel space according to the scanning grid index to construct the three-dimensional data volume of the ground-penetrating radar in the buried area.

[0045] In both the 3D point cloud model before burial and the 3D data volume of the ground-penetrating radar after burial, a unified local coordinate system is established using cement pillars, support pillars, and the ground coordinates at the base, so that the foundation centerline C0(s) and the target centerline Cg(s) are located in the same coordinate frame, thereby realizing the spatial correspondence of shape guidance and overall model deformation.

[0046] In this invention, the three-dimensional point cloud model P0 before burial is used as the basic geometric model, and the target centerline Cg(s) extracted by ground-penetrating radar after burial is used as the morphological change guidance information. Under the guidance of ground-penetrating radar, the three-dimensional point cloud model before burial is subjected to overall model deformation fitting to obtain a detailed underground model of the grapevine after burial.

[0047] Furthermore, basic parameters such as the morphology, size, and location of the grapevines before burying were collected, along with environmental parameters such as field supports, the ground surface, and planting row direction. This provides raw data support for subsequent 3D modeling and coordinate positioning. A local coordinate system was established, with the grape planting direction defined as the X-axis, the direction perpendicular to the grape planting direction defined as the Y-axis, and the direction normal to the ground surface defined as the Z-axis. Data was collected using an RGBD 310 camera to construct a high-precision 3D point cloud model of the grapevines before burying, and the center line of the vine trunk was extracted to form a high-precision geometric foundation template, ensuring the accuracy of the original morphology and size of the vine trunk.

[0048] Based on the field environment after burying and the scanning parameters of the grapevines after equipment burial, a detection grid was established to ensure that the radar scanning positions were standardized and the spatial coordinates were mappable, providing a unified spatial layout for the construction of the 3D data volume. By acquiring underground information through ground-penetrating radar, 2D profile data was overlaid to construct a 3D data volume, and guiding target bodies matching the grapevines were selected to obtain the approximate extension path and spatial range of the underground vine trunks, achieving coarse positioning of underground targets.

[0049] Multi-source data spatial alignment unifies pre-burying point cloud data and post-burying radar data into the same coordinate system, eliminating spatial deviations and providing directly fused frame parameters for subsequent model deformation fitting. Based on the high-precision pre-burying point cloud as the foundation model, and guided by the target centerline extracted by radar, overall deformation fitting is performed under constraints, outputting a detailed 3D model of the vine trunk beneath the soil. This model retains the geometric accuracy of the point cloud while accurately reflecting the actual posture underground, overcoming the technical shortcomings of low accuracy from standalone radar data and the inability of standalone point clouds to adapt to post-burying deformation. Based on the detailed model, the coordinates of the vine trunk roots, center path, outer contour, burial depth, and overall spatial distribution are determined and output, providing directly usable high-precision location data for mechanized grape harvesting and intelligent operations.

[0050] In the preferred embodiment of this application, step S40, guiding the target body includes the following steps: Step S41, performing threshold segmentation based on the three-dimensional data volume of ground penetrating radar, and determining whether the radar response intensity is not less than the segmentation threshold; if so, marking it to form a high-response voxel set; Step S42, performing three-dimensional connected component marking based on the high-response voxel set to obtain multiple candidate connected target bodies; Step S43, determining whether the candidate connected target bodies meet the preset threshold; if so, marking the candidate connected target bodies that meet the preset threshold as the guiding target bodies corresponding to the grapevine.

[0051] The three-dimensional data volume of the ground-penetrating radar is denoted as V(x, y, z); where V(x, y, z) represents the radar response intensity at the three-dimensional spatial location (x, y, z). Threshold segmentation is performed on the three-dimensional data volume of the ground-penetrating radar V(x, y, z). Let the segmentation threshold be T. Then, all spatial points that satisfy the radar response intensity being greater than or equal to the threshold T constitute a set of high-response voxels.

[0052] Three-dimensional connected component labeling was performed on the high-response voxel set to obtain multiple candidate connected target bodies. Among all candidate connected target bodies, the connected target body that meets the following conditions is selected as the guiding target body corresponding to the grapevine: (1) The starting position of the target body corresponds to the ground coordinate r0 of the root; (2) The overall extension direction of the target body is consistent with the direction of burying the grapevine; (3) The axial length of the target body is basically consistent with the overall length Lp of the grapevine in the three-dimensional point cloud model before burying; (4) The connected target body that meets the above conditions is denoted as Ωg.

[0053] The target body Ωg is refined using voxel skeleton technology to extract its three-dimensional skeleton. Then, spline fitting and smoothing are performed on the three-dimensional skeleton to obtain the target centerline of the buried grapevine. This target centerline is denoted as Cg(s), where s is the arc length parameter along the target centerline, and Lg is the total length of the target centerline. The target centerline Cg(s) is used to characterize the approximate extension direction and path of the buried grapevine underground.

[0054] To further describe the curvature of the target centerline, the amount of curvature change can be calculated based on the directional change of the target centerline at each sampling position, and this amount of curvature change can be used as auxiliary guiding information to participate in the subsequent overall model deformation fitting. However, the core positioning process of this invention does not rely on complex curvature formulas, but rather uses the target centerline itself as the main guiding object.

[0055] In the above-described scheme of this application, the deformation fitting of the overall deformation fitting module in step S60 includes the following steps: Step S61, using the root ground coordinates as fixed anchor points, establish a regular control grid outside the three-dimensional point cloud model; Step S62, using the base centerline as the driving skeleton and the target centerline as the fitting target, iteratively update the control points of the regular control grid according to the spatial extension direction and path distribution of the target centerline; Step S63, based on the updated control points of the regular control grid, recalculate the centerline of the deformed model; Step S64, determine the maximum spatial deviation between the calculated centerline of the deformed model and the target centerline: if it is greater than a preset position error threshold, or the difference between the length of the centerline of the deformed model and the target centerline is greater than a preset length error threshold, then continue iteratively updating the control points; if it is less than or equal to the preset position error threshold, and the difference between the length of the centerline of the deformed model and the target centerline is less than or equal to the preset position error threshold, then terminate the overall model deformation fitting.

[0056] The precise location of the grapevine after burying it in step S70 includes: the fixed starting coordinates of the roots on the ground, the location of the center path under the soil, the range of the outer contour of the soil, the range of the burial depth, and the range of the overall spatial distribution.

[0057] The overall model deformation is achieved using the free deformation method. Specifically, a regular control grid is constructed outside the three-dimensional point cloud model before burial. By adjusting the position of the control points of the regular control grid, the entire point cloud undergoes continuous spatial deformation. During the overall model deformation process, the following constraints are applied simultaneously: (1) Root coordinate fixed constraint: The root ground coordinates of the deformed model are consistent with the root ground coordinates r0 before burial; (2) Main path length conservation constraint: The centerline length of the deformed model is consistent with the foundation centerline length L0 before burial; (3) Overall length basically unchanged constraint: The difference between the overall length of the whole grapevine after deformation and the overall length Lp before burial is less than the preset length error threshold εL; (4) Target centerline approximation constraint: The centerline of the deformed model approximates the target centerline extracted by the ground penetrating radar point by point; (5) Spatial envelope consistency constraint: The external spatial envelope of the deformed point cloud is consistent with the underground spatial envelope represented by the guiding target body Ωg.

[0058] To facilitate engineering implementation, in this invention, the foundation centerline C0(s) and the target centerline Cg(s) are resampled with equal arc lengths to obtain a sequence of sampling points with the same number and corresponding order. Then, the maximum spatial distance between the sampled points on the deformed centerline and the sampled points on the target centerline is used as the positional deviation index, denoted as ec. Simultaneously, the difference between the length of the deformed centerline and the length of the foundation centerline before burying is used as the length deviation index, and preset positional error thresholds εc and εL are used as deformation termination conditions.

[0059] The overall model deformation fitting process is as follows: (1) Establish a regular control grid outside the three-dimensional point cloud model before burying the soil, with the root ground coordinate r0 as the fixed anchor point; (2) Use the base center line C0(s) as the driving skeleton and the target center line Cg(s) as the fitting target; (3) Iteratively update the control grid control points according to the spatial extension direction and path distribution of the target center line; (4) After each update, recalculate the center line Cd(s) of the deformed model; (5) Calculate the maximum spatial deviation ec between the deformed center line Cd(s) and the target center line Cg(s); (6) If ec is greater than the preset position error threshold εc, or the difference between the length of the deformed center line and L0 is greater than the preset length error threshold εL, then continue iteratively updating the control points; (7) If ec is less than or equal to εc, and the difference between the length of the deformed center line and L0 is less than or equal to εL, then terminate the overall model deformation fitting.

[0060] By deforming and fitting the overall model as described above, a detailed underground model of the grapevine after burying is obtained. In this invention, the detailed underground model is denoted as Pd; Pd represents the three-dimensional point cloud model of the grapevine after burying after deformation.

[0061] This invention, under the constraints of fixed root coordinates, conserved main path length, and essentially unchanged overall length, performs free deformation fitting on the three-dimensional point cloud model before burial. This ensures that the deformation result reflects both the actual bending shape after burial and maintains the rationality of the original geometric properties of the grapevine. Furthermore, based on the detailed underground model Pd, the precise position of the grapevine after burial is determined. This precise position includes at least: the fixed starting coordinates of the root at the ground surface; and the position of the underground center path Cd(s). The range of the underground outer contour; the range of the burial depth; the range of the overall spatial distribution.

[0062] The underground outer contour envelope is determined by the spatial boundary of the deformed point cloud model Pd; the burial depth range is determined by the distribution range of Pd in ​​the Z direction; and the overall spatial distribution range is determined by the overall envelope of Pd in ​​the local coordinate system. The final output precise location result is a high-precision description of the underground position of the grapevine after burial.

[0063] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for accurately detecting the location of underground vine trunks, characterized in that, The application of the soil-drying and cleaning device includes the following steps: Step S10: Obtain the basic parameters of the grapevines and the surrounding environment parameters before burying them in the soil; Step S20: Based on the local coordinate system, establish a three-dimensional point cloud model and output the basic center line of the grapevine before burying it in the soil; Step S30: Obtain surrounding environmental parameters and scanning parameters of grapevines after burying in soil to form a regular control grid; Step S40: Based on the rule-based control grid, construct the three-dimensional data volume of the ground penetrating radar and extract the guided target volume; Step S50: Based on the 3D point cloud model, the 3D data volume of ground penetrating radar and the surrounding environmental parameters, establish a unified coordinate model and output the frame parameters. Step S60: Based on the overall deformation fitting module, input the frame parameters to perform deformation fitting, and output the fine underground model; Step S70: Based on the detailed underground model, output the precise location of the grapevines after burying them in the soil; The soil-drying and cleaning device includes: A portal frame, which straddles a vine-covered earthen mound and is capable of moving in a predetermined direction; The soil-clearing mechanism is located at one end of the portal frame away from the direction of travel and can move in both vertical and horizontal directions to clear the soil from the vine trunk. The data acquisition mechanism includes an RGBD camera and a high-frequency ground-penetrating radar. The RGBD camera is located at the end of the gantry frame away from the soil-clearing mechanism and is used to acquire the basic parameters of the grapevines and the surrounding environment parameters before burying the soil. The high-frequency ground-penetrating radar is located in the middle of the gantry frame and between the RGBD camera and the soil-clearing mechanism and is used to acquire the scanning parameters of the grapevines after burying the soil. The main server includes a data processing module, a coordinate system aggregation module, a deformation fitting module, and an execution module. The data processing module is connected to the RGBD camera and the high-frequency ground-penetrating radar, respectively, and is used to process the basic parameters of the grapevines before burying, the surrounding environment parameters, and the scanning parameters of the grapevines after burying, and output a 3D point cloud model and a 3D data volume from the ground-penetrating radar. The coordinate system aggregation module is connected to the data processing module, and based on the 3D point cloud model and the 3D data volume from the ground-penetrating radar, executes a unified coordinate model and outputs frame parameters. The deformation fitting module is connected to the coordinate system aggregation module, and based on the frame parameters, performs deformation fitting to establish a fine underground model and outputs the precise position of the grapevines after burying. The execution module is connected to the deformation fitting module, and based on the precise position of the grapevines after burying, executes a soil-clearing command. The RGBD camera is an Orbbec Gemini335L, and the high-frequency ground-penetrating radar is an SIR4000 system.

2. The method for accurately detecting the location of the underground vine trunk as described in claim 1, characterized in that, Step S20 also includes establishing a local coordinate system, which includes the following steps: Step S21: Define the grape planting direction as the X-axis; Step S22: Define the direction perpendicular to the grape planting direction as the Y-axis; Step S23: Define the surface normal direction as the Z-axis; Step S24: Define the Earth's surface as a reference surface with Z=0.

3. The method for accurately detecting the location of the underground vine trunk as described in claim 1, characterized in that, In step S40, guiding the target body includes the following steps: Step S41: Perform threshold segmentation based on the three-dimensional data volume of the ground penetrating radar, and determine whether the radar response intensity is not less than the segmentation threshold. If so, mark it to form a set of high-response voxels; Step S42: Based on the high-response voxel set, perform three-dimensional connected component labeling to obtain multiple candidate connected target volumes; Step S43: Determine whether the candidate connected target volume meets the preset threshold: If so, the candidate connected target that meets the preset threshold will be marked as the guiding target corresponding to the grapevine.

4. The method for accurately detecting the location of underground vine trunks as described in claim 1, characterized in that, Step S60, the overall deformation fitting module, includes the following steps: Step S61: Using the root ground coordinates as fixed anchor points, establish a regular control mesh outside the 3D point cloud model; Step S62: Using the basic centerline as the driving skeleton and the target centerline as the fitting target, iteratively update the control points of the rule control grid according to the spatial extension direction and path distribution of the target centerline. Step S63: Based on the control points of the updated rule control mesh, recalculate the centerline of the deformed model; Step S64: Determine the maximum spatial deviation between the centerline of the deformed model and the target centerline: If the error exceeds the preset position error threshold, or if the difference between the length of the deformed model's centerline and the target centerline exceeds the preset length error threshold, then continue iteratively updating the control points. If the error is less than or equal to the preset position error threshold, and the difference between the length of the centerline of the deformed model and the target centerline is less than or equal to the preset position error threshold, then the overall model deformation fitting is terminated.

5. The method for accurately detecting the location of the underground vine trunk as described in claim 1, characterized in that, The precise location of the grapevine after burying it in step S70 includes: the fixed starting coordinates of the roots on the ground, the location of the center path under the soil, the range of the outer contour of the soil, the range of the burial depth, and the range of the overall spatial distribution.

6. The method for accurately detecting the location of the underground vine trunk as described in claim 1, characterized in that, Step S20, the construction of the 3D point cloud model, includes: invalid point removal, outlier removal, voxel downsampling, background separation, multi-frame coarse registration and global fine registration, and point cloud stitching.

Images