A linear moving 3D soil scanner and a three-dimensional soil surface state evaluation method of image scanning

Abstract

The application relates to a linear moving type 3D soil scanner and a three-dimensional soil surface state evaluation method of image scanning, and relates to a soil scanner and a three-dimensional soil surface state evaluation method.The application is used to solve the technical problems of low precision, large error of the existing soil surface measurement method, low automation degree, small scanning width, and the fact that a depth sensor is extremely easy to be disturbed under strong outdoor light of the existing portable soil scanner, and the like.A Z-axis sliding cross beam is slidably arranged on a portal frame, a scanning camera is installed on the Z-axis sliding cross beam through an X-axis linear driving assembly, and a control assembly is used for controlling the scanning camera, the X-axis linear driving assembly and a Z-axis automatic lifting assembly to work.S1, equipment deployment and automatic adjustment of a scanning working distance;S2, automatic scanning and data acquisition;S3, point cloud preprocessing and coordinate calculation;S4, three-dimensional point cloud data registration and data fusion;S5, three-dimensional model reconstruction: calculation of a soil surface flatness index and a soil fragmentation rate index.

Description

Technical Field

[0001] This invention relates to a soil scanner and a three-dimensional soil surface condition assessment method, specifically to a linear moving 3D soil scanner and a three-dimensional soil surface condition assessment method using image scanning. This invention belongs to the field of agricultural soil physical condition detection and analysis technology. Background Technology

[0002] In modern precision agriculture, the three-dimensional micro-topographic features of the soil surface (such as flatness and fragmentation) are key indicators for measuring tillage quality and seedbed preparation effectiveness. Flatness is an important indicator for quality monitoring during soil tillage, and its suitability directly affects crop growth and development. However, according to relevant data analysis, my country loses as much as 40% of its grain yield due to deficiencies in soil guidance. A core reason for this pain point is that existing agricultural machinery cannot effectively guide land cultivation, largely due to a severe shortage of field measurement instruments. This results in a lack of accurate data support for agricultural machinery during operation, hindering precise tillage. For a long time, methods for detecting soil physical conditions have faced two extreme dilemmas. On the one hand, the most commonly used method in grassroots operations is still contact-based manual measurement. However, the traditional crude method of directly measuring with a ruler can no longer meet our requirements for data accuracy and continuity. This method is not only time-consuming and labor-intensive but also highly susceptible to human error. On the other hand, in pursuit of high precision, some scientific research has adopted technologies such as aerial remote sensing, satellite mapping, or airborne lidar. However, considering the cost and inconvenience of using satellite and laser technologies on small plots, these high-precision devices are extremely impractical for routine agricultural production and high-frequency performance testing of agricultural machinery, making it difficult to promote them on a large scale.

[0003] To bridge the gap between traditional manual measurement and expensive surveying equipment, the industry has successively developed several linear 3D soil scanners based on sliding rails. However, in the complex environments of actual fields, these early prototype devices have revealed serious structural defects and technical bottlenecks. First, in terms of mechanical architecture and control logic, existing devices mostly rely on purely mechanical methods such as manually cranking handwheels to adjust the Z-axis height of the camera to achieve focus. This semi-automatic mode, which relies on manual intervention, is not only extremely inefficient, but also prone to micro-vibrations during the cranking process, causing the scanning focal plane to shift and severely weakening the accuracy of subsequent point cloud data reconstruction. Second, limited by considerations of portability and cost, the X-axis guide rail of existing devices is often designed to be too short, resulting in a very narrow lateral width for a single scan. Faced with the vast working surfaces of wide-width heavy tractors or large reversible plows in modern agriculture, this "seeing through a tube" approach with short guide rails requires frequent relocation and resetting of the coordinate system, inevitably introducing huge cumulative stitching errors. Finally, in the most crucial visual perception terminal, early devices generally used first-generation consumer-grade motion-sensing cameras (such as older devices using speckle structured light technology). These sensors, originally designed for indoor environments, would suffer severe interference from sunlight when exposed to the strong outdoor light of open farmland, resulting in large blind spots and "black holes" in the depth image. In addition, the lack of a global shutter made them prone to motion blur during dynamic scanning as they slid along the guide rail, leading to serious distortion of the acquired three-dimensional soil coordinates.

[0004] In conclusion, the current fields of agricultural engineering and soil testing urgently need innovative equipment that can completely break through existing technological barriers. Developing an equipment that overcomes the drawbacks of manual measurement (its roughness and low efficiency) while avoiding the high cost and cumbersome operation of satellite and laser technologies; simultaneously, in its design, achieving fully automatic and precise focusing on the Z-axis and wide-span scanning on the X-axis, and equipped with advanced vision sensors possessing strong outdoor anti-interference capabilities; thereby acquiring large-area three-dimensional physical models of soil at low cost and high efficiency under complex field lighting conditions—this has become a key technical challenge that urgently needs to be overcome to evaluate the effectiveness of agricultural machinery and improve the performance of soil tillage machinery. Summary of the Invention

[0005] In order to solve the technical problems of low accuracy and large error in existing soil surface measurement methods, as well as the low degree of automation, small scanning width, and easy interference of depth sensors under strong outdoor light in existing portable soil scanners, this invention provides a linear mobile 3D soil scanner and a three-dimensional soil surface condition assessment method based on image scanning.

[0006] To address the aforementioned problems, the present invention provides the following technical solution: A linear mobile 3D soil scanner includes a gantry, a scanning camera, an X-axis linear drive assembly, a control assembly, a Z-axis automatic lifting assembly, and a data processing terminal; A Z-axis sliding beam is mounted on the gantry. A scanning camera is mounted on the Z-axis sliding beam via an X-axis linear drive assembly. The scanning camera reciprocates along the X-axis on the Z-axis sliding beam. An automatic Z-axis lifting assembly is mounted on the Z-axis sliding beam and the gantry. The automatic Z-axis lifting assembly drives the Z-axis sliding beam to slide on the gantry. A control assembly is mounted on the gantry and is used to control the operation of the scanning camera, the X-axis linear drive assembly, and the automatic Z-axis lifting assembly. A data processing terminal is communicatively connected to the scanning camera and is used to receive the depth and color images acquired by the scanning camera, and to perform image preprocessing, coordinate transformation, point cloud registration and fusion, and 3D model reconstruction on the depth and color images.

[0007] Furthermore, the X-axis linear drive assembly includes an X-axis stepper motor, a synchronous belt, and a tensioner mechanism; the Z-axis automatic lifting assembly includes two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, two ball screws, and two ball screw bearing housings. The X-axis stepper motor housing and tensioner mechanism are respectively mounted on the Z-axis sliding beam. A synchronous pulley is mounted on the output shaft of the X-axis stepper motor, and a synchronous pulley is mounted on the tensioner mechanism. The synchronous pulleys on the X-axis stepper motor output shaft and the tensioner mechanism are connected by a synchronous belt drive. The scanning camera is fixedly mounted on one side of the synchronous belt, and a horizontal slider is mounted on the scanning camera. The horizontal slider slides on the horizontal slide rail of the Z-axis sliding beam. Two ball screw nuts are installed on the Z-axis sliding beam. Each ball screw nut is internally threaded with a ball screw. The bottom end of the ball screw is rotatably connected to the gantry frame through a ball screw bearing seat. The top end of the ball screw is fixedly connected to the output shaft of a Z-axis stepper motor. Each Z-axis stepper motor is connected to a Z-axis high-microstepping stepper motor driver. The rotation of the Z-axis stepper motor drives the ball screw nuts and the Z-axis sliding beam to slide vertically on the gantry frame. Four anchor bolts are installed at the bottom of the gantry frame to adjust the horizontal angle of the Z-axis sliding beam. A level is installed on the Z-axis sliding beam to observe its horizontal state.

[0008] Furthermore, the control components include a motion controller, an X-axis stepper motor driver, an operation panel, and two photoelectric sensors; The X-axis stepper motor, operation panel, two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, and two photoelectric sensors are all electrically connected to the motion controller. The X-axis stepper motor driver is electrically connected to the X-axis stepper motor and is used to drive the X-axis stepper motor to work. The motion controller controls the two Z-axis high-microstepping stepper motor drivers to drive the two Z-axis stepper motors to work synchronously. Two photoelectric sensors are respectively installed at both ends of the Z-axis sliding beam to detect the movement limit position of the scanning camera in the X-axis direction and feed back the limit signal to the motion controller. The motion controller controls the X-axis stepper motor to stop or rotate in reverse according to the limit signal.

[0009] Furthermore, the data processing terminal is a device with image processing and point cloud computing capabilities; The data processing terminal communicates with the scanning camera via a USB interface, Ethernet interface, or wireless communication module to acquire depth and color images captured by the scanning camera, and performs pixel observation reading, depth image preprocessing, coordinate transformation, point cloud registration and fusion, and 3D model reconstruction on the depth and color images. The data processing terminal is used to process the scanned images and point cloud data, while the control component is used to control the X-axis stepper motor, Z-axis stepper motor, and photoelectric sensor. The data processing terminal and the control component work together to complete the scanning camera's movement and the processing of the scanned images and point cloud data.

[0010] A method for assessing the three-dimensional soil surface condition based on soil scanner images, the method being implemented according to the following steps; S1. Equipment deployment and automatic adjustment of scanning working distance: Place the linear mobile 3D soil scanner on the plot to be tested, adjust the anchor bolts to make the Z-axis sliding beam horizontal, issue a command through the control component to drive the Z-axis stepper motor to drive the ball screw to rotate, automatically adjust the height of the Z-axis sliding beam, and make the scanning camera enter the preset scanning working distance range. S2. Automatic scanning and data acquisition: Start the X-axis stepper motor to drive the horizontal slider to move horizontally at a constant speed on the Z-axis sliding beam. During the reciprocating movement, the scanning camera acquires the depth image and corresponding color image of the soil surface below in real time. S3. Point cloud preprocessing and coordinate calculation: The depth image acquired by the scanning camera is preprocessed with depth image enhancement to obtain the pixel coordinates of each pixel in the depth image and its corresponding depth value; then, based on the camera intrinsic parameter matrix, extrinsic parameter matrix and imaging principle, each pixel in the depth image and its corresponding depth value are converted into three-dimensional point cloud data represented in the world coordinate system. S4. 3D point cloud data registration and data fusion: Through coordinate transformation and iterative nearest point algorithms, the continuous long sequence of 3D point cloud data obtained from multiple scans is registered and stitched together, redundant and overlapping parts are removed, and data fusion is completed. S5. 3D Model Reconstruction: The fused 3D point cloud data is meshed and textured by combining it with color images to finally establish a 3D textured geometric model of the soil physical state without geometric redundancy; the soil surface flatness index and soil fragmentation rate index are calculated based on the soil 3D textured geometric model.

[0011] Furthermore, in S1, by independently adjusting the four anchor bolts at the bottom of the gantry and observing the level on the Z-axis sliding beam in real time, the system ensures that the Z-axis sliding beam is level on the uneven ground, establishing a system reference level. The 24V DC control box power supply is then connected, and the motion controller in the control assembly is activated. The motion controller generates a synchronous pulse sequence with trapezoidal acceleration and deceleration curves and sends this sequence to the two Z-axis high-microstepping motor drivers. The two Z-axis high-microstepping motor drivers drive the two Z-axis stepper motors to rotate synchronously, and each stepper motor drives its corresponding ball screw to rotate, thereby causing the Z-axis sliding beam to smoothly rise and fall vertically. The data processing terminal reads the initial depth detection value of the center point returned by the scanning camera in real time. When this depth detection value falls within the preset scanning working distance range of the scanning camera, the motion controller stops sending lifting pulses and uses the two Z-axis high-microstepping motor drivers to maintain the current position of the two Z-axis stepper motors, completing the automatic calibration of the scanning working distance.

[0012] Furthermore, in S2, after the scanning working distance of S1 is calibrated, the translation speed of the X-axis linear drive component is set in the control component. The motion controller sends a control signal to the X-axis stepper motor driver, which drives the X-axis stepper motor to rotate. The X-axis stepper motor drives the scanning camera with a horizontal slider installed to move at a constant speed along the Z-axis sliding beam in the X-axis direction via a synchronous belt. During the movement, the scanning camera synchronously acquires depth images and color images. The scanning camera projects an infrared structured light pattern onto the soil surface below via an active infrared emitter. An infrared camera captures the infrared image formed by the reflection from the soil surface. The visual processing chip built into the scanning camera calculates and generates a depth image based on the infrared image. Simultaneously, the RGB color camera of the scanning camera captures a color image from the same viewpoint as the depth image. The depth image and the color image have corresponding timestamp information. The data processing terminal synchronously acquires the depth image and the color image.

[0013] Furthermore, in S3, to achieve the conversion from depth image to 3D point cloud data, the data processing terminal first acquires the depth image and color image captured by the scanning camera, and then reads the horizontal pixel coordinates of each pixel in the depth image frame by frame. u Vertical pixel coordinates v and their corresponding depth values dThis forms depth pixel observation values; simultaneously, the data processing terminal reads the RGB color values ​​of the positions corresponding to the depth pixel observation values ​​in the color image, which are used for texture mapping of the subsequent 3D model. Subsequently, based on the camera's intrinsic and extrinsic parameter matrices and the imaging principle of a pinhole camera, the data processing terminal converts the depth pixel observations into three-dimensional point cloud data represented in the world coordinate system. Advanced nonlinear lens distortion compensation: First import Brown−Conrady Polynomial distortion model for normalized image coordinates Coupled correction of radial and tangential distortions is performed; whereby normalized image coordinates are used. Pixel coordinates from depth pixel observations 、 The calculation formula, obtained by combining the camera intrinsic parameter matrix, is as follows: (1) (2) in, , These are the equivalent focal lengths of the scanning camera in the horizontal and vertical directions, respectively. 、 The coordinates of the optical principal point of the scanning camera in the image coordinate system; These are the normalized image coordinates to be corrected, calculated based on depth pixel observations.

[0014] Let the corrected distortion-free normalized coordinates be... When performing distortion correction, first let , , as the initial estimate of the distortion-free normalized coordinates; in the first... In this iteration, the squared radial distance is: (3) in, For the first The radial distance corresponding to the distortion-free normalized coordinate estimate at the next iteration. It is the square of the radial distance.

[0015] according to Brown-Conrady Polynomial distortion model, the first The distortion-free normalized coordinates corresponding to the next iteration are calculated as follows: (4) (5) in This is the radial lens distortion coefficient. This is the distortion coefficient of the tangential lens. For the number of iterations, , The first The distortion-free normalized coordinate estimate obtained in the next iteration; , The updated values ​​obtained according to formulas (4) and (5) are respectively the first... The results of the next iteration; the radial lens distortion coefficient and the tangential lens distortion coefficient are obtained by the scanning camera calibration. The data processing terminal repeatedly executes formulas (3) to (5), and stops iterating when the difference between the normalized coordinates obtained from two adjacent iterations is less than the preset convergence threshold, and the final iteration result is displayed. , These are respectively used as the corrected distortion-free normalized coordinates. and .

[0016] 3D spatial projection inverse calculation: After distortion correction, the data processing terminal calculates the depth value. and the corrected distortion-free normalized coordinates Calculate 3D points in camera coordinate system The calculation formula is as follows: (6) (7) (8) (9) Let the rotation matrix from the camera coordinate system to the world coordinate system be... The translation vector is Then the three-dimensional points in the camera coordinate system Convert to 3D points in world coordinate system The relation is: (10) in, , These are the three-dimensional coordinates of the corresponding points in the world coordinate system. and These are the rotation matrix and translation vector from the camera coordinate system to the world coordinate system, respectively. They are obtained through the installation pose calibration between the scanning camera and the Z-axis sliding beam and guide rail, and are updated according to the current scanning position as the scanning camera moves along the X-axis. Thus, the data processing terminal converts the depth pixel observations in each frame of depth image into a discrete 3D point set represented in the world coordinate system, generating initial 3D point cloud data.

[0017] Furthermore, in S4, the data processing terminal performs normal vector estimation, point cloud registration, voxel filtering, noise reduction and fusion, and continuous surface reconstruction on the 3D point cloud data obtained from multiple scans. Normal vector estimation based on principal component analysis: The data processing terminal estimates the normal vector of any point in the 3D point cloud data. Select its neighborhood point set Let the neighborhood point set be... Includes The neighboring points, the th Let the neighboring points be denoted as ,in The geometric center of the neighborhood point set is The calculation formula is as follows: (11) Construct the corresponding local covariance matrix based on the neighborhood point set. The calculation formula is as follows: (12) in, Point to be estimated The corresponding local neighborhood covariance matrix, The number of neighboring points. Point to be estimated The Neighboring points, It is the geometric center of the neighborhood point set; For covariance matrix Eigenvalue decomposition is performed to obtain multiple eigenvalues ​​and their corresponding eigenvectors, and the eigenvector corresponding to the smallest eigenvalue is taken as the value of the point. The local normal vector at point is expressed as: Its expression is: (13) in, Covariance matrix The eigenvector corresponding to the smallest eigenvalue, Used for characterization points The normal direction of the local soil surface; thus, the data processing terminal obtains the local normal vector of each point in the three-dimensional point cloud data; Point-to-plane iterative nearest-point registration: Let the source point cloud be... The target points are clustered as Source points gather The first in The points are Target points are concentrated In and the source point The corresponding target point is Target point The local normal vector at is ,in: (14) in, Indicates the target point The local normal vector at that location, This represents the target point calculated using the principal component analysis method described above. The local normal vector.

[0018] Let the rotation matrix be... The translation vector is The data processing terminal establishes a source point cloud through nearest neighbor search. Gathering with target points The corresponding points between them are identified, and the optimal rotation matrix is ​​solved by minimizing the projection distance of the source point along the normal vector direction of the target point. and optimal translation vector Its core iterative cost function for: (15) Optimal rotation matrix and optimal translation vector The solution is determined by minimizing the core iterative cost function, and its formula is as follows: (16) in To pass KD−Tree The number of corresponding point pairs obtained by nearest neighbor search For robust weighting factors based on feature distance, Represents the source point After rotation and translation, it reaches the target point. The signed distance of the tangent plane; The robust weighting factor The method is determined based on the spatial distance between corresponding point pairs, the angle between normal vectors, or the registration residual. It is used to reduce the impact of weeds, debris, flying insects, local occlusion, and depth measurement anomalies on the soil surface on the point cloud registration results, and to improve the stability of point cloud data registration after multiple scans. After point cloud registration is completed, the data processing terminal uses voxel filtering to denoise and fuse the registered 3D point cloud data; specifically, the registered 3D point cloud space is divided into segments with sides of length 1. The voxel cube grid is divided into sections, where, The side length of the voxel cube grid is used to determine the spatial resolution of the voxel grid and to determine the voxel grid position to which a discrete point belongs.

[0019] Let the first Individual raster contains There are discrete points, denoted as _i_{i} ... ,in, For those falling within the same voxel grid A discrete point, the data processing terminal with its geometric centroid The geometric centroid serves as the output feature point of this voxel grid. The calculation formula is as follows: (17) in, For the first The geometric centroid of an individual pixel grid, For the first The number of discrete points contained within an individual pixel grid. For the first The first individual pixel within the raster A discrete point.

[0020] The data processing terminal uses the geometric centroid as its reference. As the first The output feature points of the voxel grid are used to replace multiple discrete points within the voxel grid, thereby eliminating redundant overlapping points caused by repeated scanning and reducing high-frequency noise generated during depth image acquisition, and obtaining fused 3D point cloud data. Furthermore, the data processing terminal constructs a normal vector field based on the local normal vectors of each point in the 3D point cloud data. And based on the normal vector field Constructing spatial indicator functions The continuous surface is then solved using the Poisson equation, which is expressed as: (18) in, The normal vector field is composed of the fused 3D point cloud data and its local normal vectors. This is a spatial indicator function used to represent the state of a point in three-dimensional space being located inside or outside a soil entity. For the Laplace operator, Normal vector field The divergence; Subsequently, the data processing terminal performs a spatial indication function. Isosurfaces are extracted to obtain a continuous three-dimensional geometric mesh model corresponding to the soil surface morphology; Through the above-mentioned normal vector estimation, point-to-plane registration, robust error suppression, voxel filtering fusion, and continuous surface reconstruction processing, the three-dimensional point cloud data obtained from multiple scans are unified under the same world coordinate system, forming a continuous, low-redundancy three-dimensional geometric model that can characterize the micro-topographic features of the soil surface.

[0021] Furthermore, in S5, the data processing terminal establishes a three-dimensional soil texture geometric model based on the continuous three-dimensional geometric mesh model obtained in S4 and the color image captured by the scanning camera, and calculates the soil surface flatness index and soil fragmentation rate index based on the three-dimensional soil texture geometric model. Specifically, the data processing terminal calls the color image corresponding to the depth image, and projects the grid vertices or triangular faces in the three-dimensional geometric mesh model to the corresponding coordinate positions of the color image according to the intrinsic and extrinsic parameter matrices of the scanning camera, and reads the RGB color values ​​at the coordinate positions of the color image; then, the RGB color values ​​are mapped to the surface of the three-dimensional geometric mesh model through texture mapping to generate a three-dimensional textured geometric model of soil with real color information and spatial size information. The soil surface smoothness index is calculated as follows: The data processing terminal first uses a random sampling consistency algorithm to fit the effective sampling points in the three-dimensional texture geometry model of the soil to a reference plane, thereby obtaining the soil reference plane. The equation of the soil reference plane is: (19) in, These are the normal vector components in the soil reference plane equation. This is a constant term in the soil reference plane equation. For spatial coordinate variables; Let the number of valid sampling points in the model be... , No. The spatial coordinates of the effective sampling points are: ,in, They represent the first The three-dimensional coordinates of the n valid sampling points in the world coordinate system, then the nth Vertical distance from each valid sampling point to the soil reference plane The calculation formula is as follows: (20) in, Indicates the first The absolute deviation after substituting each valid sampling point into the soil reference plane equation The magnitude of the normal vector to the soil reference plane; Let the average distance from all valid sampling points to the soil reference plane be... Then we have: (twenty one) Soil surface root roughness The calculation formula is as follows: (twenty two) in, As the root mean square roughness of the soil surface, it represents the root mean square value of the vertical distance from all valid sampling points to the soil reference plane, and is used to characterize the overall undulation of the soil surface relative to the soil reference plane. The smaller the value, the smaller the overall deviation of the effective sampling point from the soil reference plane, and the smoother the soil surface.

[0022] Furthermore, to avoid relying entirely on human experience for the height threshold, the data processing terminal uses the root mean square roughness of the soil surface. Adaptive determination of height threshold for raised soil block regions The calculation formula is as follows: (twenty three) in, The height threshold is used to identify areas of raised soil blocks, and to distinguish between ordinary undulating areas and areas of raised soil blocks near the soil reference plane. The average distance from all valid sampling points to the soil reference plane. This is the threshold adjustment coefficient, used to control the sensitivity of identifying raised soil block areas. Based on the height threshold The data processing terminal identifies and segments the raised soil clod areas; specifically, using the soil reference plane as a reference plane, it calculates the height deviation of each grid vertex in the soil three-dimensional texture geometry model relative to the soil reference plane, where the direction pointing upwards on the soil surface is considered positive; and considers areas with height deviations greater than the height threshold. The adjacent grid regions are divided into raised soil block regions, resulting in a set of independent soil block regions; The unit normal vector of the soil reference plane From the soil reference plane equation coefficients in The calculation is as follows: (twenty four) If the unit normal vector of the soil reference plane is calculated according to formula (24) If the direction points below the soil surface, then take its inverse vector to make the unit normal vector of the soil reference plane... It always points upwards towards the soil surface.

[0023] Suppose that the set of independent soil blocks contains There are three triangular facets, among which... This represents the total number of triangular facets in the set of independent soil block regions. The triangular facet number is used to indicate the number of the facet. ;No. The area of ​​each triangular facet is , Used to indicate the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; Each triangular facet relative to the height threshold The average super-high amount is , No. The unit normal vector of each triangular facet is , No. The three vertices of the triangular facet are respectively , and Then the first The formula for calculating the unit normal vector of a triangular facet is as follows: (25) The total volume of the individual soil blocks The calculation formula is as follows: (26) in, Indicates height above the threshold The total volume of the protruding soil clods; Indicates the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; the average superelevation. The directional height deviation of the three vertices of the triangular facet relative to the soil reference plane exceeds a height threshold. The arithmetic mean of a part; Indicates the first The unit normal vector of a triangular facet; The unit normal vector representing the soil reference plane; Indicates the first The absolute value of the cosine of the angle between the normal vector of the triangular facet and the normal vector of the soil reference plane is used to determine the value of the first triangular facet normal vector. The area of ​​each triangular facet is converted into the proportion of its projected area relative to the soil reference plane; The data processing terminal calculates the volume of the region enclosed by the soil surface, the soil reference plane, and the boundary of the scanned area based on the three-dimensional texture geometry model of the soil, and uses this volume as the total volume of disturbed soil within the scanned area. The soil crushing rate index The calculation formula is as follows: (27) in, Indicates height above the threshold The total volume of the protruding soil clods, This indicates the total volume of disturbed soil within the scanned area. This indicates the proportion of the volume of the raised soil clods in the total volume of the disturbed soil. The larger the value, the smaller the proportion of the total volume of the raised soil clods, and the higher the degree of soil fragmentation.

[0024] The beneficial effects of this invention are as follows: A qualitative leap in automation and stability: This invention completely abandons the outdated method of adjusting height using a hand crank, and innovatively introduces a Z-axis automatic motor and ball screw transmission system. Through electronically controlled stepping adjustment, not only is the focusing accuracy improved to the millimeter level, but the frame sway and visual focal plane shift caused by manual cranking are also fundamentally avoided, greatly ensuring the stability of the scanning process and the absolute accuracy of the data.

[0025] The system boasts exceptional resistance to outdoor interference: it has replaced the outdated Kinect motion-sensing camera with the industrial-grade Intel RealSense D435 3D camera. The D435's unique active infrared stereo sensing technology and global shutter perfectly overcome the "black hole" phenomenon in depth data caused by traditional equipment under strong sunlight in outdoor farmland; simultaneously, it eliminates image ghosting during high-speed scanning with the slider, ensuring high-quality point cloud output even under harsh field lighting conditions.

[0026] Large scanning span and high operational efficiency: By adopting an extended aluminum profile crossbeam and an extended synchronous belt design, the bottleneck of the extremely narrow scanning width of the original portable equipment has been completely broken. A single setup can cover a wider operating area of ​​large agricultural machinery, greatly reducing the cumbersome steps of moving equipment in the field, and also reducing the cumulative error caused by frequently stitching together point clouds of different blocks.

[0027] With its high cost-effectiveness and significant guiding value, this device is inexpensive, detachable, and portable, perfectly avoiding the high costs and operational inconveniences associated with satellite remote sensing and large-scale lidar technologies used in small to medium-sized plots. By using this scanner to perform three-dimensional reconstruction and comparative analysis of the soil physical state before and after tillage, it can intuitively and accurately evaluate the actual operational effects of agricultural implements such as tractors and reversible plows, providing invaluable data support for the research and selection of agricultural implements and the improvement of soil tillage machinery performance. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall structure of this application; Figure 2 This is an isometric drawing of the overall structure of this application; Figure 3 This is a front view of the overall structure of this application; Figure 4 This is a side view of the overall structure of this application; Figure 5 This application describes the steps for assessing the three-dimensional soil surface condition using soil scanner images. Figure 6It is the spatial relationship of three-dimensional points in the world coordinate system, which is the transformation of pixel observation values ​​in the depth image into the camera intrinsic and extrinsic parameter matrices. That is, image pixel coordinate system → camera coordinate system → world coordinate system; Figure 7 When the scanning camera moves along the X-axis, the local point clouds obtained at different scanning positions are registered and stitched together in the same world coordinate system to form the spatial relationship of the global point cloud. That is, local point clouds P(i-1), P(i), P(i+1) → global point cloud M through Ropt and Topt registration.

[0029] Figure 8 This is a rendering of the soil 3D texture model generated by scanning and finally reconstructing the model, removing geometric redundancy, according to the present invention. Detailed Implementation

[0030] Combination Figures 1 to 4 This embodiment describes a linear mobile 3D soil scanner, which includes a gantry 1, a scanning camera 2, an X-axis linear drive assembly 3, a control assembly 4, a Z-axis automatic lifting assembly 5, and a data processing terminal 6. A Z-axis sliding beam is slidably mounted on the gantry 1. A scanning camera 2 is mounted on the Z-axis sliding beam via an X-axis linear drive assembly 3. The scanning camera 2 slides back and forth along the X-axis on the Z-axis sliding beam. An automatic Z-axis lifting assembly 5 is mounted on the Z-axis sliding beam and the gantry 1. The automatic Z-axis lifting assembly 5 drives the Z-axis sliding beam to slide on the gantry 1. A control assembly 4 is mounted on the gantry 1. The control assembly 4 is used to control the operation of the scanning camera 2, the X-axis linear drive assembly 3, and the automatic Z-axis lifting assembly 5. A data processing terminal 6 is communicatively connected to the scanning camera 2 and is used to receive the depth and color images acquired by the scanning camera 2. The terminal 6 performs image preprocessing, coordinate transformation, point cloud registration and fusion, and 3D model reconstruction on the depth and color images.

[0031] In this embodiment, the scanning camera 2 is an Intel RealSense D435 3D camera. The gantry 1 is an extended gantry support frame, assembled from industrial-grade high-strength 4040 aluminum profiles. The gantry 1 is composed of multiple high-strength aluminum profiles, significantly expanding the lateral width of a single scan. In this embodiment, the scanning working axis of the Intel RealSense D435 3D camera is staggered and parallel to the main load-bearing beam above, ensuring that the top structure of the scanning camera 2 will not mechanically collide or interfere with the beam and transmission components when moving at high speed along the slide, guaranteeing that the camera lens's vertical downward field of view is completely unobstructed by the support. The D435 3D camera employs an advanced stereo vision computing architecture and a more powerful, more complex active infrared projector. Even under the strong direct sunlight of midday in summer, its two infrared CMOS cameras can still clearly capture subtle undulations on the soil surface, completely eliminating the "outdoor light failure" phenomenon of older equipment. It features a built-in industrial-grade global shutter image sensor, meaning that all pixels on the image sensor are exposed simultaneously in a very short instant. Therefore, even when the camera is dragged by an X-axis motor at a relatively high speed (e.g., 0.5 m / s) for lateral reciprocating scanning, the captured depth maps and color images maintain high clarity, reducing motion blur or image distortion, thus providing a reliable data foundation for subsequent high-precision 3D reconstruction. The D435's 3D camera has a wider field of view than traditional devices, approximately 87° horizontally and 58° vertically.

[0032] The Z-axis automatic lifting component 5 drives the scanning camera 2 to automatically adjust the working distance between the scanning camera 2 and the surface of the soil to be measured.

[0033] In this embodiment, the horizontal aluminum profile guide rail at the top of the gantry 1 has been lengthened. This large-span design allows the equipment to completely cover the cross-section of a standard furrow or the single-operation zone of a wide-width agricultural machine with only a single setup, achieving a qualitative leap in scanning coverage. All aluminum profile nodes of the gantry 1 are secured using standard corner slot connectors and high-strength hexagonal bolts. This allows the entire massive frame to be quickly disassembled into several independent long strip components within just 10 minutes, greatly facilitating the vehicle-mounted transfer and transportation of the research team between different test sites, perfectly balancing the seemingly contradictory engineering requirements of "large-span scanning" and "lightweight portability."

[0034] The infrared transmitter, color camera, and infrared CMOS camera in the scanning camera 2 are all aimed at the surface of the soil to be measured, and are used to collect high-precision color images and depth point cloud data of the soil surface in real time under complex outdoor lighting conditions.

[0035] Combination Figures 1 to 4As shown, the X-axis linear drive assembly 3 includes an X-axis stepper motor, a synchronous belt, and a tensioner mechanism; the Z-axis automatic lifting assembly 5 includes two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, two ball screws, and two ball screw bearing housings. The X-axis stepper motor housing and the tensioning pulley mechanism are respectively mounted on the Z-axis sliding beam. A synchronous pulley is mounted on the output shaft of the X-axis stepper motor, and a synchronous pulley is mounted on the tensioning pulley mechanism. The synchronous pulleys on the X-axis stepper motor output shaft and the tensioning pulley mechanism are connected by a synchronous belt drive. The scanning camera 2 is fixedly mounted on one side of the synchronous belt, and a horizontal slider is mounted on the scanning camera 2. The horizontal slider slides on the horizontal slide rail of the Z-axis sliding beam. Two ball screw nuts are installed on the Z-axis sliding beam. Each ball screw nut is internally threaded with a ball screw. The bottom end of the ball screw is rotatably connected to the gantry 1 through a ball screw bearing seat. The top end of the ball screw is fixedly connected to the output shaft of a Z-axis stepper motor. Each Z-axis stepper motor is connected to a Z-axis high-microstepping stepper motor driver. The rotation of the Z-axis stepper motor drives the ball screw nut and the Z-axis sliding beam to slide vertically on the gantry 1. Four anchor bolts are installed at the bottom of the gantry 1. The four anchor bolts are used to adjust the horizontal angle of the Z-axis sliding beam. A level is installed on the Z-axis sliding beam to observe the horizontal state of the Z-axis sliding beam.

[0036] In this embodiment, an X-axis stepper motor is mounted on one end of the Z-axis sliding beam, and a tensioning pulley mechanism is provided on the other end of the Z-axis sliding beam. A synchronous belt connects the synchronous pulley of the X-axis stepper motor and the synchronous pulley of the tensioning pulley mechanism, and fixes the scanning camera 2 to a specific node on the synchronous belt. When the X-axis motor rotates at a constant speed set by the system program, the synchronous belt can drag the slider to make extremely smooth linear X-axis movement on the guide rail.

[0037] This application takes into account the safety protection and reciprocating operation requirements during automatic high-speed scanning. A high-sensitivity photoelectric sensor is installed near the limit positions at both ends of the Z-axis sliding beam. When the high-speed moving transverse slider enters the detection area of ​​either photoelectric sensor, the photoelectric sensor sends a limit signal to the motion controller. The motion controller, based on the limit signal, controls the X-axis stepper motor to stop and rotate in the opposite direction, thereby driving the transverse slider to move in the opposite direction via a synchronous belt. This achieves automatic reciprocating scanning of the scanning camera 2 along the X-axis and reduces the labor intensity of field operators.

[0038] In this embodiment, the Z-axis stepper motor is NEMA 17 or NEMA 23. The output shaft of the Z-axis high-microstepping stepper motor driver is connected to a high-precision ball screw via a high-elasticity coupling. During operation, the control component 4 sends synchronous pulse signals to the Z-axis high-microstepping stepper motor drivers on both sides. The motor rotation drives the ball screw to rotate, efficiently and smoothly converting the rotational motion into the vertical lifting and lowering of the crossbeam. By precisely controlling the number of pulse steps of the motor, the system can achieve an absolute positioning accuracy of up to 0.1 mm or even higher. This mechanism not only achieves extremely smooth lifting and lowering of the camera without any manual intervention, but also ensures that the camera can always be precisely hovered on the optimal optical focusing plane that adapts to different soil undulations, achieving high data acquisition accuracy at the physical level.

[0039] The bottom anchor bolts of the gantry frame 1 are height-adjustable to adapt to uneven farmland terrain. During deployment, operators only need to observe the level on the Z-axis sliding beam and, by rotating the adjustment discs of the anchor bolts at each location, can quickly eliminate the tilt caused by terrain undulations within seconds, bringing the entire extended beam to a perfectly level state. This not only provides a solid and stable foundation for the mechanical structure but is also the primary physical prerequisite for ensuring the accuracy of subsequent optical scanning and point cloud stitching.

[0040] Combination Figures 1 to 4 As shown, control component 4 includes a motion controller, an X-axis stepper motor driver, an operation panel, and two photoelectric sensors; The X-axis stepper motor, operation panel, two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, and two photoelectric sensors are all electrically connected to the motion controller. The X-axis stepper motor driver is electrically connected to the X-axis stepper motor and is used to drive the X-axis stepper motor to work. The motion controller controls the two Z-axis high-microstepping stepper motor drivers to drive the two Z-axis stepper motors to work synchronously. Two photoelectric sensors are respectively installed at both ends of the Z-axis sliding beam to detect the movement limit position of the scanning camera 2 in the X-axis direction and feed back the limit signal to the motion controller. The motion controller controls the X-axis stepper motor to stop or rotate in the reverse direction according to the limit signal.

[0041] In this embodiment, when the transverse slider of the scanning camera 2 moves to the end of the Z-axis sliding beam and enters the detection area of ​​any photoelectric sensor, the photoelectric sensor outputs a limit signal to the motion controller. The motion controller, based on the limit signal, controls the X-axis stepper motor to stop or reverse rotation, and drives the transverse slider to move in the opposite direction via a synchronous belt, thereby realizing automatic reciprocating scanning of the scanning camera 2 along the X-axis. The transverse slider slides in conjunction with the transverse guide rail on the Z-axis sliding beam. A linear bearing is installed inside the transverse slider to reduce frictional resistance during movement and improve the smoothness of the scanning camera 2's movement.

[0042] Combination Figures 1 to 4 As shown, data processing terminal 6 is a device with image processing and point cloud computing capabilities; The data processing terminal 6 can be any one of a computer, industrial control computer, or embedded processing module; The data processing terminal 6 communicates with the scanning camera 2 via a USB interface, Ethernet interface, or wireless communication module to acquire depth images and color images captured by the scanning camera 2, and performs pixel observation value reading, depth image preprocessing, coordinate transformation, point cloud registration and fusion, and three-dimensional model reconstruction processing on the depth images and color images. The data processing terminal 6 is used to process the scanned images and point cloud data, and the control component 4 is used to control the X-axis stepper motor, Z-axis stepper motor and photoelectric sensor. The data processing terminal 6 and the control component 4 cooperate with each other to complete the moving scan of the scanning camera 2 and the processing of the scanned images and point cloud data.

[0043] The buttons on the operation panel of control component 4 are defined as follows: First button (Power On / Power On): Responsible for connecting and initializing the system's main power supply, activating the motor driver and sensors.

[0044] The second button (lift control): actively controls the Z-axis motor to rotate forward, driving the crossbeam and camera to rise vertically.

[0045] The button below the second button (descent control): actively controls the Z-axis motor to reverse, driving the crossbeam to descend smoothly to adjust the focal length.

[0046] The third button (scan start): triggers the X-axis motor to start working, causing the slider to enter the automatic reciprocating scanning mode.

[0047] The fourth button (sensor switch) is used to manually turn the feedback function of the limit sensors at both ends of the guide rail on or off, which is convenient for unidirectional scanning in special terrain.

[0048] The fifth button (emergency stop) has the highest priority. Pressing it will instantly cut off all motor power output, achieving physical locking to protect the equipment.

[0049] Combination Figures 1 to 5 As shown, this embodiment provides a method for assessing soil surface condition using the linear moving 3D soil scanner, the method being implemented according to the following steps: S1. Equipment deployment and automatic adjustment of scanning working distance: The linear mobile 3D soil scanner is placed on the plot to be tested. The anchor bolts are adjusted to make the Z-axis sliding beam horizontal. The control component 4 issues a command to drive the Z-axis high-resolution stepper motor driver to drive the ball screw to rotate, and automatically adjust the height of the Z-axis sliding beam so that the scanning camera 2 enters the preset scanning working distance range. S2. Automatic scanning and data acquisition: Start the X-axis stepper motor to drive the horizontal slider to move horizontally at a constant speed on the Z-axis sliding beam. During the reciprocating movement, the scanning camera 2 acquires the depth image and corresponding color image of the soil surface below in real time. S3. Point cloud preprocessing and coordinate calculation: The depth image acquired by the scanning camera 2 is subjected to depth image enhancement preprocessing to obtain the pixel coordinates of each pixel in the depth image and its corresponding depth value; then, based on the camera intrinsic parameter matrix, extrinsic parameter matrix and imaging principle, each pixel in the depth image and its corresponding depth value are converted into three-dimensional point cloud data represented in the world coordinate system. S4. 3D point cloud data registration and data fusion: Through coordinate transformation and iterative nearest point algorithms, the continuous long sequence of 3D point cloud data obtained from multiple scans is registered and stitched together, redundant and overlapping parts are removed, and data fusion is completed. S5. 3D Model Reconstruction: The fused 3D point cloud data is meshed and textured by combining it with color images to finally establish a 3D textured geometric model of the soil physical state without geometric redundancy; the soil surface flatness index and soil fragmentation rate index are calculated based on the soil 3D textured geometric model.

[0050] This application achieves a fully automated processing flow from real-time acquisition of raw field physical data and multi-dimensional coordinate mapping to the generation of high-precision three-dimensional texture geometric models through the synergistic coupling of electromechanical hardware and software.

[0051] Combination Figures 1 to 5As shown, in S1, by independently adjusting the four anchor bolts at the bottom of the gantry 1 and observing the dual-axis bubble level on the Z-axis sliding beam in real time, the Z-axis sliding beam is ensured to be in a horizontal state on the uneven ground, thus establishing a system reference horizontal plane. The 24V DC control box power supply is turned on and the motion controller in the control component 4 is started. The single-axis controller microprocessor (MCU) of the motion controller generates a synchronous pulse control signal with a trapezoidal acceleration and deceleration curve, and sends the synchronous pulse sequence to the two Z-axis high-microstepping motor drivers respectively. The two Z-axis high-microstepping motor drivers drive the two Z-axis stepper motors to rotate synchronously. The two Z-axis stepper motors drive the corresponding ball screws to rotate, thereby driving the Z-axis sliding beam to rise and fall smoothly in the vertical direction. The data processing terminal 6 reads the initial depth detection value of the center point returned by the scanning camera 2 in real time. When the depth detection value falls within the preset scanning working distance range of the scanning camera 2, the motion controller stops sending lifting control pulses and keeps the two Z-axis stepper motors in their current position through the two Z-axis high-microstepping motor drivers, completing the automatic calibration of the scanning working distance.

[0052] Combination Figures 1 to 5 As shown in S2, after the scanning working distance of S1 is calibrated, the translation speed of the X-axis linear drive component 3 is set in the control component 4. The motion controller sends a control signal to the X-axis stepper motor driver, which drives the X-axis stepper motor to rotate. The X-axis stepper motor drives the scanning camera 2, which is equipped with a horizontal slider, to move at a constant speed along the Z-axis sliding beam in the X-axis direction via a synchronous belt. During the movement, the scanning camera 2 synchronously acquires depth images and color images. The scanning camera 2 projects an infrared structured light pattern onto the soil surface below via an active infrared emitter. The infrared camera captures the infrared image formed by the reflection from the soil surface. The vision processing chip built into the scanning camera 2 calculates and generates a depth image based on the infrared image. At the same time, the RGB color camera of the scanning camera 2 captures a color image from the same viewpoint as the depth image. The depth image and the color image have corresponding timestamp information. The data processing terminal 6 synchronously acquires the depth image and the color image.

[0053] When the horizontal slider enters the detection area of ​​the photoelectric sensor at the end of the guide rail, the photoelectric sensor outputs a limit signal to the motion controller. The motion controller then controls the X-axis stepper motor to stop and rotate in the opposite direction based on the limit signal, causing the horizontal slider to move in the opposite direction. Through this reciprocating scanning method based on limit feedback, the scanning camera 2 can automatically reciprocate along the X-axis, expanding the scanning coverage area of ​​the soil surface.

[0054] Combination Figures 1 to 6As shown in step S3, to convert the depth image into 3D point cloud data, the data processing terminal 6 first acquires the depth image and color image captured by the scanning camera 2, and then reads the horizontal pixel coordinates of each pixel in the depth image frame by frame. u Vertical pixel coordinates v and their corresponding depth values d This forms depth pixel observation values; simultaneously, the data processing terminal 6 reads the RGB color values ​​in the color image corresponding to the depth pixel observation values, which are then used for texture mapping of the subsequent 3D model. Subsequently, the data processing terminal 6 converts the depth pixel observation values ​​into three-dimensional point cloud data represented in the world coordinate system based on the camera intrinsic parameter matrix, extrinsic parameter matrix and pinhole camera imaging principle. Advanced nonlinear lens distortion compensation: Given that light rays at the edges of a wide field of view (FOV) in outdoor environments easily pass through the lens edges, causing severe optical distortion, the following approach is first implemented... Brown-Conrady Polynomial distortion model for normalized image coordinates Coupled correction of radial and tangential distortions is performed; whereby normalized image coordinates are used. Pixel coordinates from depth pixel observations 、 The calculation formula, obtained by combining the camera intrinsic parameter matrix, is as follows: (1) (2) in, , These are the equivalent focal lengths of the scanning camera 2 in the horizontal and vertical directions, respectively. , Let be the coordinates of the optical principal point of scanning camera 2 in the image coordinate system; These are the normalized image coordinates to be corrected, calculated based on depth pixel observations.

[0055] Let the corrected distortion-free normalized coordinates be... When performing distortion correction, first let , , as the initial estimate of the distortion-free normalized coordinates; in the first... In this iteration, the corresponding squared radial distance is: (3) in, For the first The radial distance corresponding to the distortion-free normalized coordinate estimate at the next iteration. It is the square of the radial distance.

[0056] According to the Brown-Conrady polynomial distortion model, the... The distortion-free normalized coordinates corresponding to the next iteration are calculated as follows: (4) (5) in This is the radial lens distortion coefficient. The distortion coefficient of the tangential lens. For the number of iterations, , The first The distortion-free normalized coordinate estimate obtained in the next iteration; , The updated values ​​obtained according to formulas (4) and (5) are respectively the first... The results of the next iteration; the radial lens distortion coefficient and the tangential lens distortion coefficient are obtained by the scanning camera 2. The data processing terminal 6 repeatedly executes formulas (3) to (5), and stops the iteration when the difference between the normalized coordinates obtained from two adjacent iterations is less than the preset convergence threshold, and the final iteration result is displayed. , These are respectively used as the corrected distortion-free normalized coordinates. and .

[0057] 3D spatial projection inverse calculation: After distortion correction, data processing terminal 6 calculates the depth value... and the corrected distortion-free normalized coordinates Calculate 3D points in camera coordinate system The calculation formula is as follows: (6) (7) (8) (9) Let the rotation matrix from the camera coordinate system to the world coordinate system be... The translation vector is Then the three-dimensional points in the camera coordinate system Convert to 3D points in world coordinate system The relation is: (10) in, , These are the three-dimensional coordinates of the corresponding points in the world coordinate system. and These are the rotation matrix and translation vector from the camera coordinate system to the world coordinate system, respectively. They are obtained through the installation pose calibration between the scanning camera 2 and the Z-axis sliding beam and guide rail, and are updated according to the current scanning position as the scanning camera 2 moves along the X-axis. Thus, the data processing terminal 6 converts the depth pixel observations in each frame of the depth image into a discrete 3D point set represented in the world coordinate system, generating initial 3D point cloud data.

[0058] Combination Figures 1 to 7 As shown in S4, the data processing terminal 6 performs normal vector estimation, point cloud registration, voxel filtering, noise reduction and fusion, and continuous surface reconstruction on the three-dimensional point cloud data obtained from multiple scans. Normal vector estimation based on principal component analysis: Data processing terminal 6 pairs any point to be estimated in 3D point cloud data. Select its neighborhood point set Let the neighborhood point set be... Includes The nth neighboring point, the th Let the neighboring points be denoted as ,in The geometric center of the neighborhood point set is The calculation formula is as follows: (11) Construct the corresponding local covariance matrix based on the neighborhood point set. The calculation formula is as follows: (12) in Point to be estimated The corresponding local neighborhood covariance matrix, The number of neighboring points. Point to be estimated The Neighboring points, It is the geometric center of the neighborhood point set; For covariance matrix Eigenvalue decomposition is performed to obtain multiple eigenvalues ​​and their corresponding eigenvectors, and the eigenvector corresponding to the smallest eigenvalue is taken as the value of the point. Local normal vector at point Its expression is: (13) in, Covariance matrix The eigenvector corresponding to the smallest eigenvalue, Used for characterization points The normal direction of the local soil surface; thus, the data processing terminal 6 obtains the local normal vector of each point in the three-dimensional point cloud data; Point-to-plane iterative nearest-point registration: Let the source point cloud be... The target points are clustered as Source points gather The first in The points are Target points are concentrated In and the source point The corresponding target point is Target point The local normal vector at is ,in: (14) in, Indicates the target point The local normal vector at that location, This represents the target point calculated using the principal component analysis method described above. The local normal vector.

[0059] Let the rotation matrix be... The translation vector is Data processing terminal 6 establishes a source point cloud through nearest neighbor search. Gathering with target points The corresponding points between them are identified, and the optimal rotation matrix is ​​solved by minimizing the projection distance of the source point along the normal vector direction of the target point. and optimal translation vector Its core iterative cost function for: (15) Optimal rotation matrix and optimal translation vector The solution is determined by minimizing the core iterative cost function, and its formula is as follows: (16) in To pass KD-Tree The number of corresponding point pairs obtained by nearest neighbor search For robust weighting factors based on feature distance, Represents the source point After rotation and translation, it reaches the target point. The signed distance of the tangent plane; The robust weighting factor The method is determined based on the spatial distance between corresponding point pairs, the angle between normal vectors, or the registration residual. It is used to reduce the impact of weeds, debris, flying insects, local occlusion, and depth measurement anomalies on the soil surface on the point cloud registration results, and to improve the stability of point cloud data registration after multiple scans. After point cloud registration is completed, data processing terminal 6 uses voxel filtering to perform noise reduction and fusion on the registered 3D point cloud data; specifically, the registered 3D point cloud space is divided into segments with sides of length 1. The voxel cube grid is divided into sections, where, The side length of the voxel cube grid is used to determine the spatial resolution of the voxel grid and to determine the voxel grid position to which a discrete point belongs.

[0060] Let the first Individual raster contains There are discrete points, denoted as _i_{i} ... ,in, For those falling within the same voxel grid A discrete point, with its geometric centroid, data processing terminal 6. The geometric centroid serves as the output feature point of this voxel grid. The calculation formula is as follows: (17) in, For the first The geometric centroid of an individual pixel grid, For the first The number of discrete points contained within an individual pixel grid. For the first The first individual pixel within the raster A discrete point.

[0061] Data processing terminal 6 with the geometric centroid As the first The output feature points of the voxel grid are used to replace multiple discrete points within the voxel grid, thereby eliminating redundant overlapping points caused by repeated scanning and reducing high-frequency noise generated during depth image acquisition, and obtaining fused 3D point cloud data. Furthermore, the data processing terminal 6 constructs a normal vector field based on the local normal vectors of each point in the 3D point cloud data. And based on the normal vector field Constructing spatial indicator functions The continuous surface is then solved using the Poisson equation, which is expressed as: (18) in, The normal vector field is composed of the fused 3D point cloud data and its local normal vectors. This is a spatial indicator function used to represent the state of a point in three-dimensional space, indicating whether it is located inside or outside the soil entity. For the Laplace operator, Normal vector field The divergence; Subsequently, data processing terminal 6 performs spatial indication functions. Isosurfaces are extracted to obtain a continuous three-dimensional geometric mesh model corresponding to the soil surface morphology; Through the above-mentioned normal vector estimation, point-to-plane registration, robust error suppression, voxel filtering fusion, and continuous surface reconstruction processing, the three-dimensional point cloud data obtained from multiple scans are unified under the same world coordinate system, forming a continuous, low-redundancy three-dimensional geometric model that can characterize the micro-topographic features of the soil surface.

[0062] Combination Figures 1 to 8 As shown in S5, the data processing terminal 6 establishes a three-dimensional soil texture geometric model based on the continuous three-dimensional geometric mesh model obtained in S4 and the color image collected by the scanning camera 2, and calculates the soil surface flatness index and soil fragmentation rate index based on the soil three-dimensional texture geometric model. Specifically, the data processing terminal 6 calls the color image corresponding to the depth image, and projects the grid vertices or triangular faces in the three-dimensional geometric mesh model to the corresponding color image coordinate positions according to the intrinsic and extrinsic parameter matrices of the scanning camera 2, and reads the RGB color values ​​at the color image coordinate positions; then, the RGB color values ​​are mapped to the surface of the three-dimensional geometric mesh model through texture mapping to generate a soil three-dimensional texture geometric model with real color information and spatial size information. The soil surface smoothness index is calculated as follows: Data processing terminal 6 first uses a random sampling consensus algorithm to fit the effective sampling points in the three-dimensional texture geometry model of the soil to a reference plane, thereby obtaining the soil reference plane. The equation of the soil reference plane is: (19) in, These are the normal vector components in the soil reference plane equation. This is a constant term in the soil reference plane equation. For spatial coordinate variables; Let the number of valid sampling points in the model be... , No. The spatial coordinates of the effective sampling points are: ,in, They represent the first The three-dimensional coordinates of the n valid sampling points in the world coordinate system, then the nth Vertical distance from each valid sampling point to the soil reference plane The calculation formula is as follows: (20) in, Indicates the first The absolute deviation after substituting each valid sampling point into the soil reference plane equation The magnitude of the normal vector to the soil reference plane; Let the average distance from all valid sampling points to the soil reference plane be... Then we have: (twenty one) Soil surface root roughness The calculation formula is as follows: (twenty two) in, As the root mean square roughness of the soil surface, it represents the root mean square value of the vertical distance from all valid sampling points to the soil reference plane, and is used to characterize the overall undulation of the soil surface relative to the soil reference plane. The smaller the value, the smaller the overall deviation of the effective sampling point from the soil reference plane, and the smoother the soil surface.

[0063] Furthermore, to avoid relying entirely on human experience for the height threshold, the data processing terminal 6 uses the root mean square roughness of the soil surface. Adaptive determination of height threshold for raised soil block regions The calculation formula is as follows: (twenty three) in, The height threshold is used to identify areas of raised soil blocks, and to distinguish between ordinary undulating areas and areas of raised soil blocks near the soil reference plane. The average distance from all valid sampling points to the soil reference plane. This is the threshold adjustment coefficient, used to control the sensitivity of identifying raised soil block areas. Based on the height threshold The data processing terminal 6 identifies and segments the raised soil clod areas; specifically, using the soil reference plane as a reference plane, it calculates the height deviation of each grid vertex in the soil three-dimensional texture geometry model relative to the soil reference plane, wherein the direction pointing upwards on the soil surface is considered positive; and it considers areas with height deviations greater than the height threshold. The adjacent grid regions are divided into raised soil block regions, resulting in a set of independent soil block regions; The unit normal vector of the soil reference plane From the soil reference plane equation coefficients in The calculation is as follows: (twenty four) If the unit normal vector of the soil reference plane is calculated according to formula (24) If the direction points below the soil surface, then take its inverse vector to make the unit normal vector of the soil reference plane... It always points upwards towards the soil surface.

[0064] Suppose that the set of independent soil blocks contains There are three triangular facets, among which... This represents the total number of triangular facets in the set of independent soil block regions. The triangular facet number is used to indicate the number of the facet. ;No. The area of ​​each triangular facet is , Used to indicate the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; Each triangular facet relative to the height threshold The average super-high amount is , No. The unit normal vector of each triangular facet is , No. The three vertices of the triangular facet are respectively , and Then the first The formula for calculating the unit normal vector of a triangular facet is as follows: (25) The total volume of the individual soil blocks The calculation formula is as follows: (26) in, Indicates height above the threshold The total volume of the protruding soil clods; Indicates the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; the average superelevation. The directional height deviation of the three vertices of the triangular facet relative to the soil reference plane exceeds a height threshold. Partial arithmetic mean; Indicates the first The unit normal vector of a triangular facet; The unit normal vector representing the soil reference plane; Indicates the first The absolute value of the cosine of the angle between the normal vector of the triangular facet and the normal vector of the soil reference plane is used to determine the value of the first triangular facet normal vector. The area of ​​each triangular facet is converted into the proportion of its projected area relative to the soil reference plane; Data processing terminal 6 calculates the volume of the region enclosed by the soil surface, the soil reference plane, and the boundary of the scanned area based on the three-dimensional texture geometry model of the soil, and uses this volume as the total volume of disturbed soil within the scanned area. The soil crushing rate index The calculation formula is as follows: (27) in, Indicates height above the threshold The total volume of the protruding soil clods, This indicates the total volume of disturbed soil within the scanned area. This indicates the proportion of the volume of the raised soil clods in the total volume of the disturbed soil. The larger the value, the smaller the proportion of the total volume of the raised soil clods, and the higher the degree of soil fragmentation; Through the above-mentioned texture mapping, reference plane fitting, surface roughness calculation, raised soil block region segmentation and soil block volume calculation, the data processing terminal 6 can simultaneously obtain soil surface morphology, flatness index and soil fragmentation rate index based on the same three-dimensional soil texture geometry model, thereby realizing the quantitative evaluation of soil physical state after agricultural machinery cultivation.

Claims

1. A linear moving 3D soil scanner, characterized in that: It includes a gantry (1), a scanning camera (2), an X-axis linear drive assembly (3), a control assembly (4), a Z-axis automatic lifting assembly (5), and a data processing terminal (6). A Z-axis sliding beam is slidably mounted on the gantry (1). The scanning camera (2) is mounted on the Z-axis sliding beam via the X-axis linear drive assembly (3). The scanning camera (2) slides back and forth along the X-axis on the Z-axis sliding beam. The Z-axis automatic lifting assembly (5) is mounted on the Z-axis sliding beam and the gantry (1). The Z-axis automatic lifting assembly (5) drives the Z-axis sliding beam to slide on the gantry (1). The control assembly (4) is mounted on the gantry (1). The control assembly (4) is used to control the operation of the scanning camera (2), the X-axis linear drive assembly (3), and the Z-axis automatic lifting assembly (5). The data processing terminal (6) is connected to the scanning camera (2) for receiving the depth image and color image acquired by the scanning camera (2) and performing image preprocessing, coordinate transformation, point cloud registration and fusion, and three-dimensional model reconstruction on the depth image and color image.

2. The linear mobile 3D soil scanner according to claim 1, characterized in that: The X-axis linear drive assembly (3) includes an X-axis stepper motor, a synchronous belt and a tensioner mechanism; the Z-axis automatic lifting assembly (5) includes two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, two ball screws and two ball screw bearing seats; The X-axis stepper motor housing and the tensioning wheel mechanism are respectively mounted on the Z-axis sliding beam. A synchronous pulley is mounted on the output shaft of the X-axis stepper motor, and a synchronous pulley is mounted on the tensioning wheel mechanism. The synchronous pulleys on the output shaft of the X-axis stepper motor and the synchronous pulleys on the tensioning wheel mechanism are connected by a synchronous belt drive. The scanning camera (2) is fixedly mounted on one side of the synchronous belt. A horizontal slider is mounted on the scanning camera (2), and the horizontal slider slides on the horizontal slide rail of the Z-axis sliding beam. Two ball screw nuts are installed on the Z-axis sliding beam. Each ball screw nut is internally threaded with a ball screw. The bottom end of the ball screw is rotatably connected to the gantry (1) through a ball screw bearing seat. The top end of the ball screw is fixedly connected to the output shaft of a Z-axis stepper motor. Each Z-axis stepper motor is connected to a Z-axis high-microstepping stepper motor driver. The rotation of the Z-axis stepper motor drives the ball screw nut and the Z-axis sliding beam to slide vertically on the gantry (1). Four anchor bolts are installed at the bottom of the gantry (1). The four anchor bolts are used to adjust the horizontal angle of the Z-axis sliding beam. A level is installed on the Z-axis sliding beam to observe the horizontal state of the Z-axis sliding beam.

3. The linear mobile 3D soil scanner according to claim 1, characterized in that: The control component (4) includes a motion controller, an X-axis stepper motor driver, an operation panel, and two photoelectric sensors; The X-axis stepper motor, operation panel, two Z-axis stepper motors, two Z-axis high-microstepping stepper motor drivers, and two photoelectric sensors are all electrically connected to the motion controller. The X-axis stepper motor driver is electrically connected to the X-axis stepper motor and is used to drive the X-axis stepper motor to work. The motion controller controls the two Z-axis high-microstepping stepper motor drivers to drive the two Z-axis stepper motors to work synchronously. Two photoelectric sensors are respectively installed at both ends of the Z-axis sliding beam to detect the movement limit position of the scanning camera (2) in the X-axis direction and feed the limit signal back to the motion controller. The motion controller controls the X-axis stepper motor to stop or rotate in the reverse direction according to the limit signal.

4. The linear moving 3D soil scanner according to claim 3, characterized in that: The data processing terminal (6) is a device with image processing and point cloud computing capabilities; The data processing terminal (6) communicates with the scanning camera (2) via a USB interface, Ethernet interface or wireless communication module to acquire the depth image and color image collected by the scanning camera (2), and performs pixel observation value reading, depth image preprocessing, coordinate transformation, point cloud registration and fusion and three-dimensional model reconstruction processing on the depth image and color image. The data processing terminal (6) is used to complete the data processing of scanned images and point cloud data, and the control component (4) is used to complete the control of the X-axis stepper motor, Z-axis stepper motor and photoelectric sensor. The data processing terminal (6) and the control component (4) cooperate with each other to complete the moving scan of the scanning camera (2) and the processing of scanned images and point cloud data.

5. A method for assessing the three-dimensional soil surface condition using soil scanner images obtained from any one of claims 1 to 4, characterized in that: The method is implemented according to the following steps; S1. Equipment deployment and automatic adjustment of scanning working distance: Place the linear mobile 3D soil scanner on the plot to be tested, adjust the anchor bolts to make the Z-axis sliding beam horizontal, issue an instruction through the control component (4) to drive the Z-axis stepper motor to drive the ball screw to rotate, automatically adjust the height of the Z-axis sliding beam, and make the scanning camera (2) enter the preset scanning working distance range. S2. Automatic scanning and data acquisition: Start the X-axis stepper motor to drive the horizontal slider to move horizontally at a constant speed on the Z-axis sliding beam. During the reciprocating movement, the scanning camera (2) collects the depth image and corresponding color image of the soil surface below in real time. S3. Point cloud preprocessing and coordinate calculation: The depth image acquired by the scanning camera (2) is preprocessed with depth image enhancement to obtain the pixel coordinates of each pixel in the depth image and its corresponding depth value. Subsequently, based on the camera's intrinsic and extrinsic parameters and imaging principles, each pixel in the depth image and its corresponding depth value are converted into three-dimensional point cloud data represented in the world coordinate system. S4. 3D point cloud data registration and data fusion: Through coordinate transformation and iterative nearest point algorithms, the continuous long sequence of 3D point cloud data obtained from multiple scans is registered and stitched together, redundant and overlapping parts are removed, and data fusion is completed. S5. 3D Model Reconstruction: The fused 3D point cloud data is meshed and textured by combining it with color images to finally establish a 3D textured geometric model of the soil physical state without geometric redundancy; the soil surface flatness index and soil fragmentation rate index are calculated based on the soil 3D textured geometric model.

6. The method for assessing the three-dimensional soil surface condition by scanning soil scanner images according to claim 5, characterized in that: In S1, the four anchor bolts at the bottom of the gantry (1) are adjusted independently, and the level on the Z-axis sliding beam is observed in real time to ensure that the Z-axis sliding beam is in a horizontal state on the uneven ground surface, thus establishing the system reference level. The 24V DC control box power supply is turned on and the motion controller in the control component (4) is started. The motion controller generates a synchronous pulse sequence with a trapezoidal acceleration and deceleration curve and sends the synchronous pulse sequence to the two Z-axis high-resolution stepper motor drivers respectively. The two Z-axis high-resolution stepper motor drivers drive the two Z-axis stepper motors to rotate synchronously. The two Z-axis stepper motors drive the corresponding ball screws to rotate, thereby driving the Z-axis sliding beam to rise and fall smoothly in the vertical direction. The data processing terminal (6) reads the initial depth detection value of the center point returned by the scanning camera (2) in real time. When the depth detection value falls into the preset scanning working distance range of the scanning camera (2), the motion controller stops sending the lifting pulse and keeps the two Z-axis stepper motors in their current position through the two Z-axis high-resolution stepper motor drivers, thus completing the automatic calibration of the scanning working distance.

7. The method for assessing the three-dimensional soil surface condition by scanning soil scanner images according to claim 6, characterized in that: In S2, after the scanning working distance of S1 is calibrated, the translation speed of the X-axis linear drive component (3) is set in the control component (4). The motion controller sends a control signal to the X-axis stepper motor driver. The X-axis stepper motor driver drives the X-axis stepper motor to rotate. The X-axis stepper motor drives the scanning camera (2) with the horizontal slider installed to move at a constant speed along the Z-axis sliding beam in the X-axis direction through the synchronous belt. The scanning camera (2) performs synchronous acquisition of depth image and color image during the movement. The scanning camera (2) projects an infrared structured light pattern onto the soil surface below through an active infrared emitter. The infrared camera captures the infrared image formed by the reflection from the soil surface. The visual processing chip built into the scanning camera (2) calculates and generates a depth image based on the infrared image. At the same time, the RGB color camera of the scanning camera (2) captures a color image with a viewpoint corresponding to the depth image. The depth image and the color image have corresponding timestamp information. The data processing terminal (6) synchronously acquires the depth image and the color image.

8. The method for assessing the three-dimensional soil surface condition by scanning soil scanner images according to claim 5, characterized in that: In S3, in order to realize the conversion of depth image to three-dimensional point cloud data, the data processing terminal (6) first acquires the depth image and color image collected by the scanning camera (2), and reads the horizontal pixel coordinate u, vertical pixel coordinate v and its corresponding depth value d of each pixel in the depth image frame by frame to form depth pixel observation value; at the same time, the data processing terminal (6) reads the RGB color value of the position corresponding to the depth pixel observation value in the color image for subsequent texture mapping of the three-dimensional model; Subsequently, the data processing terminal (6) converts the depth pixel observation values ​​into three-dimensional point cloud data represented in the world coordinate system based on the camera intrinsic parameter matrix, extrinsic parameter matrix and pinhole camera imaging principle; High-order nonlinear lens distortion compensation: First, import the Brown-Conrady polynomial distortion model and normalize the image coordinates. Coupled correction of radial and tangential distortions is performed; whereby normalized image coordinates are used. Pixel coordinates from depth pixel observations , The calculation formula, obtained by combining the camera intrinsic parameter matrix, is as follows: （1） （2） in, , These are the equivalent focal lengths of the scanning camera (2) in the horizontal and vertical directions, respectively. , Let the coordinates of the optical principal point of the scanning camera (2) in the image coordinate system be given. The coordinates of the image to be corrected are calculated based on the depth pixel observations. Let the corrected distortion-free normalized coordinates be... When performing distortion correction, first let , , as the initial estimate of the distortion-free normalized coordinates; in the first... In this iteration, the corresponding squared radial distance is: （3） in, For the first The radial distance corresponding to the distortion-free normalized coordinate estimate at the next iteration. The square of the radial distance. According to the Brown-Conrady polynomial distortion model, the... The distortion-free normalized coordinates corresponding to the next iteration are calculated as follows: （4） （5） in This is the radial lens distortion coefficient. This is the distortion coefficient of the tangential lens. For the number of iterations, , The first The distortion-free normalized coordinate estimate obtained in the next iteration; , The updated values ​​obtained according to formulas (4) and (5) are respectively the first... The results of the next iteration; the radial lens distortion coefficient and the tangential lens distortion coefficient are obtained by the scanning camera (2) and the data processing terminal (6) repeatedly executes formulas (3) to (5). When the difference between the normalized coordinates obtained from two adjacent iterations is less than the preset convergence threshold, the iteration stops and the final iteration result is displayed. , These are respectively used as the corrected distortion-free normalized coordinates. and , Inverse calculation of 3D spatial projection: After completing distortion correction, the data processing terminal (6) calculates the depth value. and the corrected distortion-free normalized coordinates Calculate 3D points in camera coordinate system The calculation formula is as follows: （6） （7） （8） （9） Let the rotation matrix from the camera coordinate system to the world coordinate system be... The translation vector is Then the three-dimensional points in the camera coordinate system Convert to 3D points in world coordinate system The relation is: （10） in, , These are the three-dimensional coordinates of the corresponding points in the world coordinate system. and These are the rotation matrix and translation vector from the camera coordinate system to the world coordinate system, respectively. They are obtained by the installation pose calibration between the scanning camera (2) and the Z-axis sliding beam and guide rail, and are updated according to the current scanning position during the movement of the scanning camera (2) along the X-axis. Thus, the data processing terminal (6) converts the depth pixel observation values ​​in each frame of depth image into discrete three-dimensional point sets represented in the world coordinate system, generating initial three-dimensional point cloud data.

9. The method for assessing the three-dimensional soil surface condition by scanning soil scanner images according to claim 5, characterized in that: In S4, the data processing terminal (6) performs normal vector estimation, point cloud registration, voxel filtering noise reduction and fusion, and continuous surface reconstruction on the three-dimensional point cloud data obtained from multiple scans. Normal vector estimation based on principal component analysis: Data processing terminal (6) performs normal vector estimation on any point to be estimated in the 3D point cloud data. Select its neighborhood point set Let the neighborhood point set be... Includes The neighboring points, the th Let the neighboring points be denoted as ,in The geometric center of the neighborhood point set is The calculation formula is as follows: (11) Construct the corresponding local covariance matrix based on the neighborhood point set. The calculation formula is as follows: （12） in, Point to be estimated The corresponding local neighborhood covariance matrix, The number of neighboring points. Point to be estimated The Neighboring points, It is the geometric center of the neighborhood point set; For the covariance matrix C i Eigenvalue decomposition is performed to obtain multiple eigenvalues ​​and their corresponding eigenvectors, and the eigenvector corresponding to the smallest eigenvalue is taken as the value of the point. Local normal vector at point Its expression is: （13） in, Let C be the covariance matrix. i The eigenvector corresponding to the smallest eigenvalue, Used for characterization points The normal direction of the local soil surface; thus, the data processing terminal (6) obtains the local normal vector of each point in the three-dimensional point cloud data; Point-to-plane iterative nearest-point registration: Let the source point cloud be... The target points are clustered as Source points gather The first in The points are Target points are concentrated In and the source point The corresponding target point is Target point The local normal vector at is ,in: （14） in, Indicates the target point The local normal vector at that location, This represents the target point calculated using the principal component analysis method described above. The local normal vector, Let the rotation matrix be... The translation vector is The data processing terminal (6) establishes a source point cloud through nearest neighbor search. Gathering with target points The corresponding points between them are identified, and the optimal rotation matrix is ​​solved by minimizing the projection distance of the source point along the normal vector direction of the target point. and optimal translation vector Its core iterative cost function for: （15） Optimal rotation matrix and optimal translation vector The solution is determined by minimizing the core iterative cost function, and its formula is as follows: （16） in This represents the number of corresponding point pairs obtained through KD-Tree nearest neighbor search. For robust weighting factors based on feature distance, Represents the source point After rotation and translation, it reaches the target point. The signed distance of the tangent plane; The robust weighting factor The method is determined based on the spatial distance between corresponding point pairs, the angle between normal vectors, or the registration residual. It is used to reduce the impact of weeds, debris, flying insects, local occlusion, and depth measurement anomalies on the soil surface on the point cloud registration results, and to improve the stability of point cloud data registration after multiple scans. After point cloud registration is completed, the data processing terminal (6) uses voxel filtering to perform noise reduction and fusion on the registered 3D point cloud data; specifically, the registered 3D point cloud space is divided into segments with sides of length 1. The voxel cube grid is divided into sections, where, The side length of the voxel cube grid is used to determine the spatial resolution of the voxel grid and to determine the voxel grid position to which a discrete point belongs. Let the first Individual raster contains There are discrete points, denoted as _i_{i} ... ,in, For those falling within the same voxel grid A discrete point, the data processing terminal (6) with its geometric centroid The geometric centroid serves as the output feature point of this voxel grid. The calculation formula is as follows: (17) in, For the first The geometric centroid of an individual pixel grid, For the first The number of discrete points contained within an individual pixel grid. For the first The first individual pixel within the raster discrete points, The data processing terminal (6) uses the geometric centroid as its reference. As the first The output feature points of the voxel grid are used to replace multiple discrete points within the voxel grid, thereby eliminating redundant overlapping points caused by repeated scanning and reducing high-frequency noise generated during depth image acquisition, and obtaining fused 3D point cloud data. Furthermore, the data processing terminal (6) constructs a normal vector field based on the local normal vectors of each point in the three-dimensional point cloud data. And based on the normal vector field Constructing spatial indicator functions The continuous surface is then solved using the Poisson equation, which is expressed as: (18) in, The normal vector field is composed of the fused 3D point cloud data and its local normal vectors. This is a spatial indicator function used to represent the state of a point in three-dimensional space, indicating whether it is located inside or outside the soil entity. For the Laplace operator, Normal vector field The divergence; Subsequently, the data processing terminal (6) performs a spatial indication function. Isosurfaces are extracted to obtain a continuous three-dimensional geometric mesh model corresponding to the soil surface morphology; Through the above-mentioned normal vector estimation, point-to-plane registration, robust error suppression, voxel filtering fusion, and continuous surface reconstruction processing, the three-dimensional point cloud data obtained from multiple scans are unified under the same world coordinate system, forming a continuous, low-redundancy three-dimensional geometric model that can characterize the micro-topographic features of the soil surface.

10. The method for assessing the three-dimensional soil surface state by scanning soil scanner images according to claim 5, characterized in that: In S5, the data processing terminal (6) establishes a three-dimensional texture geometric model of soil based on the continuous three-dimensional geometric mesh model obtained in S4 and the color image collected by the scanning camera (2), and calculates the soil surface flatness index and soil fragmentation rate index based on the three-dimensional texture geometric model of soil. Specifically, the data processing terminal (6) calls the color image corresponding to the depth image, and according to the intrinsic and extrinsic parameters of the scanning camera (2), projects the grid vertices or triangular faces in the three-dimensional geometric grid model to the corresponding color image coordinate positions, and reads the RGB color values ​​at the color image coordinate positions. Subsequently, the RGB color values ​​are mapped onto the surface of the three-dimensional geometric mesh model through texture mapping to generate a three-dimensional soil texture geometric model with real color information and spatial size information. The soil surface smoothness index is calculated as follows: The data processing terminal (6) first uses the random sampling consistency algorithm to fit the effective sampling points in the three-dimensional texture geometry model of the soil to the reference plane, and obtains the soil reference plane. The equation of the soil reference plane is: （19） in, These are the normal vector components in the soil reference plane equation. This is a constant term in the soil reference plane equation. For spatial coordinate variables; Let the number of valid sampling points in the model be... , No. The spatial coordinates of the effective sampling points are: ,in, They represent the first The three-dimensional coordinates of the n valid sampling points in the world coordinate system, then the nth Vertical distance from each valid sampling point to the soil reference plane The calculation formula is as follows: （20） in, Indicates the first The absolute deviation after substituting each valid sampling point into the soil reference plane equation The magnitude of the normal vector to the soil reference plane; Let the average distance from all valid sampling points to the soil reference plane be... Then we have: （21） Soil surface root roughness The calculation formula is as follows: （22） in, As the root mean square roughness of the soil surface, it represents the root mean square value of the vertical distance from all valid sampling points to the soil reference plane, and is used to characterize the overall undulation of the soil surface relative to the soil reference plane. The smaller the value, the smaller the overall deviation of the effective sampling point from the soil reference plane, and the smoother the soil surface. Furthermore, to avoid relying entirely on human experience for the height threshold, the data processing terminal (6) calculates the height threshold based on the root mean square roughness of the soil surface. Adaptive determination of height threshold for raised soil block regions The calculation formula is as follows: （23） in, The height threshold is used to identify areas of raised soil blocks, and to distinguish between ordinary undulating areas and areas of raised soil blocks near the soil reference plane. The average distance from all valid sampling points to the soil reference plane. This is the threshold adjustment coefficient, used to control the sensitivity of identifying raised soil block areas. Based on the height threshold The data processing terminal (6) identifies and segments the raised soil block area; specifically, using the soil reference plane as a reference plane, it calculates the height deviation of each grid vertex in the soil three-dimensional texture geometry model relative to the soil reference plane, wherein the direction pointing upwards on the soil surface is the positive direction; and the height deviation is greater than the height threshold. The adjacent grid regions are divided into raised soil block regions, resulting in a set of independent soil block regions; The unit normal vector of the soil reference plane From the soil reference plane equation coefficients in The calculation is as follows: （24） If the unit normal vector of the soil reference plane is calculated according to formula (24) If the direction points below the soil surface, then take its inverse vector to make the unit normal vector of the soil reference plane... Always pointing upwards towards the soil surface. Suppose that the set of independent soil blocks contains There are three triangular facets, among which... This represents the total number of triangular facets in the set of independent soil block regions. The triangular facet number is used to indicate the number of the facet. ;No. The area of ​​each triangular facet is , Used to indicate the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; Each triangular facet relative to the height threshold The average super-high amount is , No. The unit normal vector of each triangular facet is , No. The three vertices of the triangular facet are respectively , and Then the first The formula for calculating the unit normal vector of a triangular facet is as follows: （25） The total volume of the individual soil blocks The calculation formula is as follows: （26） in, Indicates height above the threshold The total volume of the protruding soil clods; Indicates the first The actual surface area of ​​each triangular facet in the three-dimensional geometric mesh model; the average superelevation. The directional height deviation of the three vertices of the triangular facet relative to the soil reference plane exceeds a height threshold. The arithmetic mean of a part; Indicates the first The unit normal vector of a triangular facet; The unit normal vector representing the soil reference plane; Indicates the first The absolute value of the cosine of the angle between the normal vector of the triangular facet and the normal vector of the soil reference plane is used to determine the value of the first triangular facet normal vector. The area of ​​each triangular facet is converted into the proportion of its projected area relative to the soil reference plane; The data processing terminal (6) calculates the volume of the area enclosed by the soil surface, the soil reference plane, and the boundary of the scanned area based on the three-dimensional texture geometry model of the soil, and uses the volume as the total volume of disturbed soil within the scanned area. The soil crushing rate index The calculation formula is as follows: （27） in, Indicates height above the threshold The total volume of the protruding soil clods, This indicates the total volume of disturbed soil within the scanned area. This indicates the proportion of the volume of the raised soil clods in the total volume of the disturbed soil. The larger the value, the smaller the proportion of the total volume of the raised soil clods, and the higher the degree of soil fragmentation.

Images