Method for detecting quality of ridging operation in cold field

Abstract

The cold field ridging operation quality detection method of the present application relates to a detection method, obtains the ground point cloud data after ridging, and processes the same to obtain the centralized point cloud, constructs the ground elevation grid by using the centralized point cloud, fills the missing area in the grid, and obtains the continuous ground elevation field; constructs the trend surface based on the continuous ground elevation field, and obtains the residual elevation field; constructs the ridge alignment coordinate system based on the residual elevation field; performs segmented processing on the residual elevation field along the ridge in the ridge alignment coordinate system, and detects the ridge ditch candidate position in each segment; matches and tracks the candidate positions in each segment to obtain the continuous ridge ditch trajectory; divides the single ridge area according to the adjacent continuous ridge ditch trajectory; calculates the ridge body width, height and ridge surface width based on the single ridge area, and detects the results according to the preset threshold. The method of the present application realizes the automatic calculation of the ridge body width, height and ridge surface width, and has high detection efficiency and good spatial continuity.

Description

Technical Field

[0001] This invention relates to a testing method, and more particularly to a method for testing the quality of ridging operations in cold-region fields. Background Technology

[0002] In agricultural areas with cold climates and short frost-free periods, ridge cultivation, by shaping the cultivated land surface into a micro-topography of alternating ridges and furrows, can improve soil temperature and humidity and enhance drainage capacity, making it an important agronomic and field engineering measure for stable crop yields. The benefits of ridge cultivation depend not only on whether ridges are created, but also on whether the ridge geometry meets the requirements of agronomic and mechanized operations. Ridging quality is typically characterized by ridge width, ridge height, and ridge surface width. Deviations in ridge width from the target will reduce the row spacing matching for sowing, fertilization, and cultivation, leading to duplication or omissions. Ridge surface width determines the effective sowing area at the ridge top and the drainage space in the furrows. Insufficient ridge height weakens the warming and drainage effects, while unstable ridge morphology may induce ridge breakage, weaken water retention, and increase the risk of slope runoff and erosion. Therefore, for large-scale ridge cultivation and the construction of high-standard farmland, there is an urgent need for rapid, objective, quantitative detection and spatial evaluation of key ridge geometric indicators at the field scale to provide a basis for operational quality supervision and parameter optimization. However, existing methods for detecting ridging quality mostly rely on manual sampling and measurement, which have problems such as low detection efficiency, insufficient spatial representativeness, strong subjectivity, and difficulty in achieving continuous evaluation at the field scale. These methods are insufficient to meet the needs of rapid monitoring of ridging operation quality in large-scale cold-region fields. Summary of the Invention

[0003] The present invention aims to overcome the shortcomings of the prior art and provide a method for quality inspection of ridging operations in cold regions.

[0004] The method for quality inspection of ridge-making operations in cold regions according to the present invention is implemented through the following steps: (1) Point cloud data preprocessing After ridging, obtain lidar point cloud data of the field surface and perform coordinate centering processing on the point cloud data to obtain a centered point cloud; construct a surface elevation grid based on the centered point cloud and fill in the missing areas in the surface elevation grid with effective support constraints to obtain a continuous surface elevation field. (2) Main ridge direction estimation A trend surface is constructed based on the continuous surface elevation field, and a residual elevation field is obtained from the difference between the continuous surface elevation field and the trend surface; the main ridge orientation of the plot is estimated based on the residual elevation field, and a ridge orientation alignment coordinate system is constructed accordingly. (3) Furrow tracking The residual elevation field is segmented along the ridge direction in the ridge alignment coordinate system, and candidate furrow positions are detected in each segment based on cross-sectional depression features. Cross-segment matching and continuous tracking are performed on the candidate furrow positions in each segment to obtain continuous furrow trajectories. (4) Ridging quality inspection The single-ridge area is divided according to the adjacent continuous furrow trajectory; the ridge width, ridge height and ridge surface width are calculated based on the single-ridge area, and the ridge width, ridge height and ridge surface width are compared according to the preset quality threshold, and the ridge operation quality detection result is output.

[0005] As a further improvement of the present invention, the process of constructing the surface elevation grid in step (1) includes: projecting the centered point cloud onto a two-dimensional plane grid according to a preset grid resolution; and statistically analyzing the point cloud elevation values ​​in each grid cell to generate a median elevation grid. And further generate minimum elevation raster. and maximum elevation grid These serve as auxiliary elevation representations for furrow detection and ridge height calculation, respectively. For the above elevation grid , , When filling in missing regions, a Gaussian normalized convolution method based on an effective mask is used to locally estimate the missing pixels using only effective elevation pixels. For missing pixels with insufficient effective neighborhood support, they are either kept as null values ​​or marked as invalid regions to obtain a continuous surface elevation field. , , .

[0006] As a further improvement of the present invention, the process of constructing the residual elevation field in step (2) is as follows: for the continuous surface elevation field Perform large-scale smoothing to obtain a low-frequency trend surface that characterizes the overall undulation background of the land parcel. ; respectively using continuous surface elevation fields , , Subtract trend surface The residual elevation field is obtained. The residual elevation field is used to characterize the local micro-topographic texture formed by ridge formation; Calculate the residual elevation field Gradient components in two orthogonal directions Construct a global structure tensor based on the gradient components. According to the global structure tensor The principal feature direction determines the principal gradient direction. ; and will be in conjunction with the main gradient direction The direction perpendicular to the main ridge of the plot is determined as the direction of the main ridge. ; Utilizing the main ridge direction of the plot Construct a ridge-direction coordinate system such that the first coordinate axis extends along the ridge direction and the second coordinate axis extends across the ridge direction.

[0007] As a further improvement of the present invention, the process of detecting candidate furrow locations in step (3) includes: converting the median residual elevation field in the furrow coordinate system into a grid-direction coordinate system. Process the data in segments with a fixed step size along the ridge direction, and extract a one-dimensional cross-sectional profile along the transverse ridge direction within each segment. ; in the one-dimensional cross-sectional section Local minimum points are detected in the middle; and candidate furrow locations are determined by combining the valley bottom platform determination rule or the neighborhood comparison rule; Combining the preset ridge spacing range, candidate position spacing constraints, and candidate position elevation characteristics, the candidate ridge positions are screened to remove false ridge candidates caused by noise, missing measurements, or local disturbances; a matching cost function for candidate ridge positions between adjacent segments is established; the matching relationship is optimized based on position offset continuity, ridge spacing consistency, and trajectory smoothness; the optimized matching results are connected into continuous ridge trajectories; ridge trajectories that are locally missing but meet the continuity constraints are completed; and trajectories with a support length lower than a preset threshold or abnormal continuity are removed to obtain continuous ridge trajectories with stable numbers.

[0008] As a further improvement of the present invention, the process of dividing the single-ridge area in step (4) is as follows: using the trajectories of two adjacent continuous furrows... The left and right boundaries define the corresponding lateral range of a single ridge; and along the ridge direction, corresponding grid areas are extracted according to the left and right boundaries to form a single ridge area; the ridge width is determined by the lateral distance between the trajectories of adjacent continuous furrows within the same single ridge area; the ridge height is determined by the representative value of the ridge top elevation within the single ridge area. Representative value of furrow baseline elevation The difference is determined; the ridge width is determined by the width of the top continuous interval in the cross section of a single ridge that meets the elevation discrimination condition, and the elevation corresponding to the top interval is within 55% to 70% of the height difference between the bottom of the furrow and the top of the ridge.

[0009] Compared with existing technologies, this invention utilizes lidar point clouds to rapidly acquire micro-topography in cold-region fields after ridging. Through continuous surface elevation field construction, main ridge direction estimation, continuous tracking of furrow trajectories, and single ridge area division, it achieves automatic calculation and quality inspection of ridge width, ridge height, and ridge surface width. It has advantages such as high detection efficiency, good spatial continuity, strong objectivity, and suitability for field-scale operation quality supervision. Attached Figure Description

[0010] Figure 1A geometrical diagram illustrating the evaluation indicators for ridge-cultivation operations. Among them, the ridge width is defined as the distance between the centers of adjacent furrows, the ridge height is defined as the elevation difference between the top of the ridge and the straight line connecting the bottom of two adjacent furrows, and the ridge surface width is defined as the horizontal distance between the two edges of the flat top surface of a single ridge. Figure 2 This is a schematic diagram illustrating the acquisition of experimental data; The figure shows the location of the study area in Heilongjiang Province, China, the UAV lidar system used for field data acquisition, the sampling layout used for accuracy verification, and the reconstructed surface point cloud of the experimental plots. Figure 3 A flowchart illustrating the workflow for point cloud preprocessing, residual field construction, and main axis estimation; Its main steps include mean centering, rasterization to generate elevation grids, boundary clipping, trend surface fitting and removal, residual field construction, and global main axis estimation; Figure 4 A flowchart illustrating the workflow for extracting single-ridge quality indicators and mapping field-scale ridging operations based on furrow guidance; The process includes ridge-to-coordinate system transformation, profile-based ridge and furrow detection, cross-segment ridge and furrow tracking, single-ridge strip extraction, index calculation (ridge width, ridge height, ridge surface width), and field-scale spatial mapping. Figure 5 This is a graph showing the relationship between effective pixel coverage and gap fill rate as a function of raster resolution under different point cloud densities. The figure shows how the effective pixel coverage and hole-filling rate of the elevation grid change with the grid resolution under point cloud density conditions of 6.25%, 25%, and 100%. Figure 6 A graph showing the relationship between the ridge angle and raster resolution for different point cloud densities; This figure compares the ridge angle estimation results for different raster resolutions under point cloud density conditions of 6.25%, 25%, and 100%. Figure 7 A graph showing the relationship between the number of detected furrows and raster resolution under different point cloud densities; This figure summarizes the number of detected furrows corresponding to different raster resolutions under point cloud density conditions of 6.25%, 25%, and 100%; Figure 8 A visualization of furrow tracking trajectory; This figure overlays the tracked furrow trajectories onto a background image in a ridge-aligned coordinate system to demonstrate the continuity and spatial consistency of the furrow trajectories throughout the field. Figure 9 A comparison chart showing the manual measurements and model estimates of ridge width, ridge height, and ridge surface width; This figure is a scatter plot of model estimates and manual measurements, and includes a 1:1 reference line and a fitted regression line; Figure 10 A distribution diagram showing the estimation errors for ridge width, ridge height, and ridge surface width; This figure shows the distribution of estimation errors for the three indicators; Figure 11 A spatial distribution map for estimating ridge width, ridge height, and ridge surface width at the field scale; This figure illustrates the spatial variation characteristics of three indicators extracted from UAV lidar point clouds at the field scale; Figure 12 A frequency distribution map of ridge width, ridge height, and ridge width within the field area; This figure shows the frequency distribution of the three indicators across the entire field, and marks the mean and median; Figure 13 This is a pixel-level distribution map of the qualified quality of ridging operations based on agronomic thresholds. The figure shows the pixel-level pass / fail results for ridge width, ridge height, ridge surface width, and overall ridge-making operation quality. Detailed Implementation

[0011] The present invention provides a method for quality inspection of ridging operations in cold-region fields, which is achieved through the following steps:

[0012] (1) Collection of point cloud data A DJI Matrice 350 RTK drone equipped with a Zenmuse L1 LiDAR was used to acquire surface point cloud data after ridging operations. The test plot was approximately 120 × 60 m in size. To improve geometric redundancy between flight paths and the stability of point cloud reconstruction, the drone's flight altitude was set to 12 meters, the lateral overlap rate to 80%, the main flight path was laid along the ridge direction, and the echo mode was set to 3 echoes. The raw point cloud data was processed using DJI Terra 5.1.0 software, and datasets with three density levels (6.25%, 25%, and 100%) were constructed based on the raw point cloud data and exported in LAS format for subsequent analysis. Within the test area, 100 sampling locations were selected according to the principle of spatial uniformity to measure ridging quality data, and the location information of each point was recorded simultaneously to verify the accuracy of the model.

[0013] (2) Point cloud data preprocessing Using the point cloud data obtained in step (1) as input, the three-dimensional coordinate information of each point is read. Since the numerical magnitude of the UTM projected coordinates is much larger than the spatial scale of the micro-topographic features of the field ridges, excessive coordinate offset will reduce the numerical stability based on matrix operations. Therefore, the point cloud is first subjected to mean centering to obtain a centered point cloud.

[0014] 1) Point cloud mean centering The centered point cloud is transformed into a local coordinate system with the data centroid as the origin to improve the numerical stability in subsequent calculations. The centroid of the point cloud is calculated according to formula (1), and each point is centered using formula (2). This rigid translation transformation preserves the relative distance between points and the local surface geometry. (1) in, This represents the total number of points in the point cloud. The first points coordinate; For all points Sum of coordinates; For all points in The average value along the direction, which is the coordinate of the centroid of the point cloud: (2) in, For the first In the new coordinate system, the points are Direction coordinates. Original The coordinates of the point are After centralization, it becomes .

[0015] 2) Point cloud sampling density assessment The local sampling density of airborne lidar is affected by flight altitude, scanning frequency, and platform attitude, and is usually non-uniformly distributed within the survey area. To quantify the point spacing and guide grid size selection, nearest neighbor distance statistics are calculated. This is applied to each point in the centralized point cloud. Identify its nearest neighbor. And calculate the nearest neighbor distance. Subsequently, the average point spacing was statistically calculated based on this distance. Spacing from median The median distance serves as a robust indicator of central tendency.

[0016] 3) Point cloud rasterization Irregular point sets in the centralized point cloud are processed according to a specified raster resolution. Rasterize to a 2D grid to obtain an elevation raster. For each raster cell, statistics are calculated based on the point cloud elevation values ​​it contains, generating a median elevation raster. Minimum elevation grid With maximum elevation grid Three layers. The median elevation raster is used to represent the default surface elevation, the minimum elevation raster is used to enhance the furrow bottom features, and the maximum elevation raster is used to enhance the ridge top features, thereby providing support for furrow detection and ridge height estimation.

[0017] 4) Pixelless processing After rasterization, an effective mask is applied to empty pixels. The following labels are used: a value of 1 is used when a raster cell contains a data point, and a value of 0 is used otherwise; the elevation value of an empty cell is denoted as NaN. For elevation rasters containing NaN, if convolution is performed directly, it is easy to cause NaN diffusion and boundary artifacts. Therefore, masked Gaussian normalized convolution is used to allow only effective and limited cells to participate in the convolution operation, as shown in formula (3): (3) in, Indicates the location of the original elevation raster. The corresponding effective mask value; A set representing valid raster positions; Indicates the grid position The original validity determination function at the location; Indicates the location of the original elevation raster. The original elevation value at that location.

[0018] To reduce fragmentation in subsequent trend surface fitting, a small-scale masked Gaussian interpolation method is first used to fill in empty pixels, and a support density index is calculated to quantify the degree of neighborhood support. Pixels with insufficient support are retained as NaN to avoid extrapolating results that do not conform to actual terrain patterns in large blank areas. The filling process only applies to empty pixels; the original non-empty pixels retain their original values. After the above processing, a continuous surface elevation field is obtained. , , .

[0019] (3) Main ridge direction estimation 1) Trend surface fitting To mitigate the impact of overall field slope and low-frequency undulations on ridge orientation estimation, a trend surface is first constructed to characterize the slowly changing topographic background at the field scale. A continuous surface elevation field, after gap filling, is then used. As input, a masked Gaussian normalized convolution is used to compute the trend surface. As shown in equation (12): (12) in, The smoothing scale is represented by Two-dimensional Gaussian kernel, This represents a convolution operation. The denominator is passed through an effective mask. Normalization is implemented to ensure that only a limited number of valid grid cells participate in the smoothing calculation; for locations with excessively small denominators, the result is retained as NaN to avoid generating physically meaningless extrapolation results in areas with severely missing data. In this embodiment, to suppress meter-level low-frequency terrain changes while preserving furrow texture, the Gaussian kernel smoothing scale is set to 5.0 m and converted to pixel units according to the raster resolution.

[0020] 2) Residual field construction After the trend surface is extracted, a residual elevation field is constructed by subtracting the trend surface from the continuous surface elevation field to achieve the separation of ridge-scale micro-topography and low-frequency topographic components, as shown in Equation (13): (13) Using continuous surface elevation fields , , Subtract trend surface The residual elevation field is obtained. The main variation components in the residual elevation field correspond to the ridge-furrow micro-topographic undulations, while the overall slope and low-frequency topographic components are significantly suppressed, thereby reducing the ridge orientation estimation bias caused by gentle surface slopes.

[0021] 3) Main monopoly direction estimation The residual elevation field exhibits significant spatial anisotropy: the variation along the ridge direction is relatively gentle, while the variation perpendicular to the ridge direction is more drastic. To robustly estimate the global main ridge direction, the discrete gradient of the residual elevation field is calculated on a regular grid and aggregated on the effective grid cell set using the global structure tensor. The gradient components in the two orthogonal directions are calculated using central difference, as shown in Equation (15): (15) in, Indicates the elevation field at Gradient components in the direction; This indicates the positions of the two adjacent grid cells to the left and right of the current position; This indicates the positions of the two adjacent grid cells above and below the current position; Indicates the raster resolution.

[0022] Then, a structure tensor per grid cell is constructed based on the gradient components, as shown in Equation (16); and the average is calculated over the effective grid cell set to obtain the global structure tensor, as shown in Equation (17): (16) in, Indicates position The local structure tensor at that location; express The square of the directional gradient component; This represents the product of the gradient components in two directions: (17) in, Represents the global structure tensor; Represents the set of valid locations; Represents a set The number of elements in the grid, i.e., the total number of effective grid cells; This represents the summation of the local structure tensor over all valid grid locations; Represents the global structure tensor in Components in direction; This represents the intersection term in the global structure tensor.

[0023] The eigenvector corresponding to the largest eigenvalue of the global structure tensor is taken as the dominant gradient energy direction, and the gradient principal direction angle is determined accordingly, as shown in equation (18). Since the ridge direction is orthogonal to the dominant gradient direction, the ridge direction angle is calculated based on the gradient principal direction angle: (18) in, This indicates the dominant gradient direction angle; since the ridge direction is orthogonal to the dominant gradient direction, the main ridge direction... according to Calculation. Since the ridge direction has axial equivalence, the ridge angle is folded within a preset angle range to obtain a uniquely represented axial ridge angle. To maintain consistency in the orientation conventions of subsequent ridge direction alignment coordinate systems, when the ridge angle exceeds the preset range, it undergoes an equivalent angle transformation to unify the direction representation.

[0024] (4) Furrow tracking 1) Coordinate system transformation To achieve stable pruning and unified extraction of cross-sectional features in single-ridge areas, the main ridge direction of the plot is utilized. Construct a ridge-oriented coordinate system. Using a two-dimensional rotation matrix, transform each point from the centered planar coordinate system to the ridge-oriented coordinate system, such that the first coordinate axis extends along the ridge direction and the second coordinate axis extends across the ridge direction. The coordinate transformation is defined by equations (19) to (20): (19) (20) in, This represents the axial ridge angle used to construct the ridge-oriented coordinate system; Represents the centered planar coordinates; Represents the axial coordinates after rotation transformation, where The axis extends along the ridge direction. The axis extends across the ridge direction. After rotation, all gridded elevation fields and residual elevation fields are represented in the ridge-direction coordinate system, thus enabling furrow detection and subsequent single-ridge division along a unified direction.

[0025] 2) Furrow detection To enhance the continuity of furrow texture while suppressing pixel-level noise and local data gaps, the median residual elevation field in the furrow-direction coordinate system is used. As input data for furrow detection, the two-dimensional grid is divided into segments with a fixed step size along the ridge direction. Within each segment, a one-dimensional cross-sectional profile is extracted along the direction spanning the ridge to detect candidate furrow locations. Since furrows exhibit a local depression structure on the cross-section of the residual surface, by analyzing each segment... The median of the residual values ​​at each location is used to construct a model of the location. The one-dimensional cross-section is shown in equation (21): (twenty one) in, Indicates the current segment along the ridge direction. Coordinate range This represents a one-dimensional median profile constructed along the transverse direction within the segment.

[0026] To further suppress jagged extrema caused by grid noise, rotational resampling, and locally missing data, a one-dimensional mask Gaussian smoothing is performed on the one-dimensional profile before candidate point detection and geometric determination. Local minima in the smoothed profile are used as candidate furrow points. Considering that flat-bottomed valleys often appear in actual profiles, a platform determination rule is further introduced: if the profile value within a continuous interval is basically constant within the tolerance range, and the profile values ​​on both sides are significantly higher than the platform value, then the center of the platform is considered as a candidate valley point. The tolerance is defined as shown in equation (22): (twenty two) in, This represents the absolute tolerance threshold. This represents the relative tolerance coefficient. This represents the standard deviation of the one-dimensional median profile.

[0027] To suppress duplicate detections and nearby false valley points caused by noise, a two-stage geometric constraint method is used to screen candidate valley points. First, in this embodiment, multiple candidate points within a radius of 0.25 m are grouped into the same candidate cluster. For each candidate cluster, a priori information on row spacing is introduced for geometric consistency determination. Let a candidate point within the cluster be... The nearest preserved valley points on its left and right sides are respectively and A geometric consistency cost function is constructed to evaluate the spacing consistency of each candidate point relative to the expected row spacing. Within each candidate cluster, the candidate point with the smallest cost function value is retained; if there are ties for the minimum value, the candidate point with the larger valley depth is preferred to ensure the determinism and physical consistency of the results.

[0028] Secondly, the adjacent spacing of the candidate furrows, sorted in ascending order by the cross-row coordinates, is checked. In this embodiment, if the spacing between two adjacent candidate furrows is less than 0.80 m, the same geometric consistency cost criterion is used to retain only one candidate point, thus obtaining the furrow position sequence within the current segment.

[0029] 3) Furrow span matching Since single-segment extreme value detection may produce missed detections or false furrows near local missing regions and boundaries, a cross-segment tracking method is adopted to further stabilize furrow positions. For each segment, a three-branch dynamic programming matching strategy is used to associate the predicted position of an existing trajectory in the current segment with the furrow candidate sequence obtained from the current segment detection. The three-branch dynamic programming matching strategy allows the following three types of operations: detection matching, observation missing, and skip connections. For unmatched detection results, they are initialized as new trajectories, thereby allowing the number of trajectories to dynamically change between different segments.

[0030] To reduce trajectory fragmentation, missing positions are filled within the observation segment range of each trajectory. Linear interpolation of adjacent segments is used to obtain the... The trajectory in the first Predicted lateral position in segmentation Then, in the section window The inner search identifies valid candidate points for completion. These candidate points must satisfy the following conditions: firstly, they must meet the offset constraint. Secondly, it is a local minimum point or located on a minimum platform; thirdly, it satisfies the minimum spacing constraint with other furrow locations within the same segment. .

[0031] Finally, short trajectories caused by noise are eliminated according to the support length criterion, and each trajectory is sorted and renumbered according to its median lateral position in all segments. Through the above cross-segment matching and completion strategy, the fragmentation and index drift of furrow trajectories can be reduced, providing continuous and stable furrow trajectories for subsequent single-row segmentation and index estimation.

[0032] (5) Ridging quality inspection 1) Ridge width detection After obtaining the continuous furrow trajectories, the furrow positions tracked within each segment are arranged in ascending order along the transverse ridge direction. Adjacent furrow trajectories are considered as the left and right boundaries of a single ridge region, respectively. Within the ridge direction range of the current segment, the segment matrix is ​​trimmed according to the positions of the left and right boundaries in the transverse ridge direction to extract the single ridge strip region, as shown in equation (23): (twenty three) in, Indicates the first A single-ridge strip area, and These represent the boundary positions of two adjacent furrow trajectories in the transverse direction across the furrow.

[0033] The width of the ridge is defined as the distance between two adjacent furrow trajectories in the transverse direction across the ridge, and is calculated using equation (24): (twenty four) in, Indicates the first The width of the ridge in a single ridge area Indicates raster resolution. This represents the lateral cell index corresponding to two adjacent furrow trajectories. Since the width of the furrow is directly determined by the boundaries of the two adjacent furrows, its detection reliability mainly depends on the stability of the furrow trajectory tracking results.

[0034] 2) Ridge height detection Within each segment, extract along the transverse ridge direction. One-dimensional cross-sectional profile The elevations of the ridge top and furrow bottom are constructed separately. Considering that directly using the maximum and minimum values ​​is easily affected by local noise and outliers, this embodiment uses the mean of quantile intervals for robust estimation: the mean of samples within the 93%–98% quantile interval of the maximum value profile distribution is taken as the peak value of the ridge top. The mean of the samples within the 3% to 8% quantile interval of the minimum value profile distribution is taken as the baseline of the trench bottom. The ridge height is defined as the difference between the peak value at the top of the ridge and the baseline at the bottom of the furrow. .

[0035] 3) Ridge width detection The ridge width refers to the lateral span of the ridge top region that is relatively flat and above a preset height threshold. The ridge top determination threshold is defined using the furrow bottom baseline and the ridge height. (25) in, It is used to select candidate elevation zones close to the top of the ridge within the elevation range from the bottom of the furrow to the top of the ridge.

[0036] Within the current single-row strip, a one-dimensional profile of the median residual elevation field is constructed according to equation (26), and the coordinates along the transverse ridge direction are constructed using equation (27). Binary decision mask: (26) in, Indicates the position across the ridge. The one-dimensional median profile value at [location] This represents the median residual elevation field in the axial coordinate system at position. The value at that location, Indicates all The median value is taken for the direction.

[0037] (27) in, Indicates the position across the ridge. Binary decision mask at the location, This indicates that the profile value at this location is higher than or equal to the threshold, and thus belongs to the candidate ridge top region. This indicates that the location is below the threshold and does not belong to the ridge top region. Since scattered high-value fragments may appear in the profile, the longest continuous interval that meets the threshold condition is selected as the effective ridge top range. If the selected continuous interval is... Then the width of the ridge can be calculated according to formula (28): (28) in, Indicates the width of the ridge surface. Indicates raster resolution. This indicates the left and right boundary positions of the effective continuous ridge top region. The method utilizes relative thresholds based on the furrow bottom baseline and ridge height, and constructs a profile from the median residual elevation field to reduce the influence of random extrema, thereby avoiding misclassification of scattered fragments as ridge top regions. Finally, for each segment and ridge, the three indicators are assigned back to the corresponding pixels to generate a field-scale spatial distribution map of ridge quality.

[0038] The effects of this invention will be further explained below in conjunction with field trials:

[0039] 1. Raster resolution comparison test

[0040] 1.1 Point Cloud Density Analysis Since the selection of raster resolution is affected by the sampling density of the underlying point cloud, the raster resolution should be analyzed in conjunction with the statistical results of the pixel spacing. In practical applications, a raster resolution several times that of the feature point spacing is usually used to achieve a balance between preserving the ridge and groove texture and reducing the number of non-point pixels. For three reconstruction density settings, the raster resolution in the range of 0.01m to 0.10m was tested at 0.01m intervals. The statistical results of the number of point clouds and the nearest neighbor spacing under the three reconstruction density conditions are shown in Table 1.

[0041] Table 1 Statistical characteristics of point cloud density and nearest neighbor distance

[0042] As shown in Table 1, as the point cloud density increases, both the average point spacing and the median point spacing decrease, thus providing the conditions for preserving furrow details using a smaller grid resolution.

[0043] 1.2 Pixel Coverage Analysis To quantify the impact of different raster resolutions on raster data availability and gap filling requirements, a comparative analysis was conducted on three indicators: original coverage, standardized original coverage, and fill rate. Figure 5 The results show that the original coverage increases rapidly with increasing raster resolution, and then gradually plateaus. Under the minimum raster resolution condition, all three density settings produce a large number of empty pixels, with the 6.25% density dataset showing the most significant increase. Once the raster resolution enters a medium range, the increase in original coverage decreases significantly, indicating that further increasing the raster resolution has limited effect on improving coverage.

[0044] The normalized raw coverage is more sensitive to changes in raster resolution. The 6.25% density dataset has significantly lower normalized coverage at lower raster resolutions, but its improvement is more pronounced as the raster resolution increases. In contrast, the 25% and 100% density datasets show higher coverage even at lower raster resolutions and stabilize earlier, indicating that higher point cloud density is more conducive to supporting finer raster resolutions.

[0045] The fill rate is higher at lower raster resolutions, decreases significantly as raster resolution increases, and then gradually stabilizes. This indicates that when the raster resolution is too low, more empty pixels are generated, thus increasing the need for fill. At higher raster resolutions, the remaining missing pixels are mainly concentrated in boundary areas or areas with weak data support, and the effect of fill processing is relatively weakened.

[0046] 1.3 Stability Analysis of Ridge Angle Raster resolution not only affects the effective coverage area but also the continuity of residual texture, thus influencing the statistical results of the global structure tensor. Therefore, the changes in the ridge angle under different raster resolutions were compared. Figure 6 The results show that the estimated ridge orientation angle remains stable overall under different raster resolutions and point cloud densities. This indicates that, under conditions of relatively consistent ridge and furrow orientation within the plot, the global structure tensor aggregation method has good robustness to local missing cells and pixel-level noise. Therefore, raster resolution mainly affects the reliability of ridge orientation estimation, while having a relatively small impact on the main ridge orientation result itself.

[0047] 1.4 Analysis of the number of furrows Under three point cloud density settings, when the raster resolution varied from 0.01 m to 0.10 m, the total number of detected furrows remained within a small range of 58 to 59. Figure 7 The results indicate that, within the tested grid resolution range, both furrow detection and cross-segment tracking remain stable overall, without any phenomenon of a continuous increase in the number of missed detections or a significant increase in duplicate detections due to changes in grid resolution.

[0048] At a few raster resolutions, the number of furrows may fluctuate slightly by about ±1. At lower raster resolutions, broken textures near the furrow bottom may form multiple adjacent local minima, causing a single actual furrow to be split into multiple detected targets, resulting in a slight increase in the total number of furrows counted. At higher raster resolutions, pixelation and local smoothing typically make the furrow bottom position more stable. Even so, the visible range of the furrow bottom may still change locally near boundary regions or areas with missing data, causing occasional connectivity or local merging of boundary furrows. Overall, the fluctuation amplitude is small and does not show a systematic trend with raster resolution.

[0049] Considering factors such as near-saturation of coverage, significantly reduced reliance on gap filling, stable ridge angle, and consistent furrow counts across different point cloud densities, this embodiment selects 0.05 m as the raster resolution for subsequent experiments and model construction. Under this setting, the 6.25% density dataset achieves 96.5% coverage (its maximum), a fill rate of 13.8%, a ridge angle of 77.79°, and detects 58 furrows.

[0050] 2. Furrow detection and tracking test In the ridge alignment coordinate system, the ridge and furrow tracking trajectory across the entire field is as follows: Figure 8As shown, the trajectories are spatially distributed approximately as horizontal straight lines, indicating good consistency in furrow location along the ridge direction. Furthermore, the trajectory numbers between adjacent segments remain stable, demonstrating that the correspondence between furrows remains consistent within the main effective area. In boundary areas where the effective coverage decreases or residual texture support is insufficient, some trajectories may shorten. This phenomenon is normal within boundary areas and does not affect the continuity and numbering stability of furrow trajectories within the main effective area.

[0051] In summary, the method proposed in this invention first extracts candidate furrow points from periodic cross-sectional profiles, then reduces the interference of spurious extrema through smoothing, and combines cross-segment tracking and constraint completion to achieve continuous connection and stable numbering of furrow trajectories, thereby providing a reliable structural reference for the subsequent calculation of indicators such as ridge width, ridge height, and ridge surface width. Since these indicators all rely on paired furrow boundaries and their corresponding ridge regions for calculation, the stability of the furrow index plays a crucial role in the accuracy of single-ridge division and the consistency of field-scale indicator mapping.

[0052] 3. Performance test of ridging quality testing method

[0053] 3.1 Sensitivity Analysis of Ridge Width Threshold Table 2 summarizes the sensitivity of the ridge width estimation results to the ridge top determination threshold parameter. The results show that when the threshold parameter is 0.70, the model error is the lowest, the estimation result is basically unbiased, and the correlation coefficient is the highest. This indicates that when the threshold parameter is 0.70, the ridge top area and the two side slope areas can be distinguished relatively stably in the residual elevation profile, thus obtaining better ridge width estimation results.

[0054] Table 2 Sensitivity of Ridge Width Estimation Accuracy to Judgment Threshold

[0055] threshold MAE (m) RMSE (m) Bias (m) r 0.50 0.139 0.145 0.139 0.544 0.55 0.103 0.109 0.103 0.715 0.60 0.077 0.084 0.075 0.639 0.65 0.045 0.057 0.038 0.617 0.70 0.021 0.025 -0.002 0.853 0.75 0.048 0.067 -0.041 0.423 0.80 0.099 0.116 -0.098 0.339 0.85 0.173 0.190 -0.173 0.252 0.90 0.308 0.329 -0.308 0.054 When the threshold is below 0.70, the ridge width exhibits a systematic overestimation, as reflected in the positive deviation results in Table 2. This is because a lower threshold includes a larger area of ​​the ridge side slope within the ridge top region mask. Conversely, when the threshold is above 0.70, it causes a systematic underestimation of the ridge width, resulting in a rapid decline in model performance and a significant decrease in correlation. This indicates that an excessively high threshold retains only a narrower ridge top region and is more sensitive to local noise and subtle irregularities in the ridge top, thus reducing the measurement stability between different samples. Therefore, this embodiment selects 0.70 as the default threshold for ridge width estimation and qualification determination.

[0056] 3.2 Performance Analysis of Site-Level Qualitative Classification of Ridging Quality Testing Methods Figure 9This section compares model estimates with manually measured values ​​for three indicators: ridge width, ridge height, and ridge surface width. The fitted regression line is generally close to the 1:1 reference line, with a deviation close to zero, indicating a small systematic bias. The ridge surface width shows good consistency, while the ridge width exhibits relatively greater dispersion. This is mainly because the ridge width is determined by the combined ridge and furrow boundaries obtained from two tracking methods. The corresponding error distribution is shown below. Figure 10 As shown, the errors of the three indicators are all concentrated near zero, and the errors of most samples are within the centimeter range.

[0057] Furthermore, agronomic thresholds were used to convert continuous estimates into pass / fail criteria, and the resulting point-level classification performance results are summarized in Table 3. The results show that the precision, recall, and accuracy for ridge width were 0.96, 0.95, and 0.91, respectively; for ridge height, they were 0.95, 0.90, and 0.88; and for ridge width, they were 0.97, 0.99, and 0.96. Overall, all three indicators showed high precision, indicating that the pass / fail criteria obtained based on the method of this invention have good reliability. The relatively low recall for ridge height indicates that, in this embodiment, ridge height is the main limiting factor affecting the pass / fail criteria.

[0058] Table 3. Performance of Point-Level Ridging Operation Quality Acceptance Classification Based on Confusion Matrix

[0059] index TP FP TN FN Precision Recall Accuracy Ridge width 89 4 2 5 0.96 0.95 0.91 Ridge height 71 4 17 8 0.95 0.90 0.88 ridge width 92 3 4 1 0.97 0.99 0.96

[0060] 3.3 Quality Evaluation of Ridging Operations Figure 11 This paper presents spatial distribution maps of ridge width, ridge height, and ridge width at the field scale. All three indicators exhibit a continuous striped distribution characteristic consistent with the ridge direction, indicating that the estimation and back-projection methods along the ridge direction can generate spatially continuous results within the effective coverage area. The corresponding field-scale distribution characteristics are summarized in [the relevant section]. Figure 12 Among these, ridge width is mainly concentrated near the nominal ridge spacing, ridge height is mostly distributed near the minimum height requirement, and ridge width reaches its peak within the agronomic expectation range. The mean and median of each indicator are relatively close, indicating that their distribution at the field scale is relatively stable.

[0061] Based on preset agronomic thresholds, qualified distribution maps for each indicator level were further generated and integrated to form an overall qualified distribution map for ridging operations. Figure 13The results showed that most areas met the requirements for ridge width and ridge surface width, while areas with unqualified ridge height occurred relatively frequently, mainly distributed near the field boundaries. The corresponding pass rates were: ridge width 93.41%, ridge height 84.25%, and ridge surface width 93.73%; when all three indicators were met, the overall pass rate was 76.00%.

[0062] The index-level and overall compliance distribution maps provided by this invention can form spatially defined diagnostic layers, which helps to identify the distribution location of non-compliant areas and the main limiting factors within a field. These spatial diagnostic layers are difficult to obtain through traditional manual sampling and verification methods, and are therefore more suitable for field-scale ridge-making quality monitoring and evaluation.

Claims

1. A method for quality inspection of ridging operations in cold regions, characterized in that... It includes the following steps: (1) Point cloud data preprocessing After ridging, obtain lidar point cloud data of the field surface and perform coordinate centering processing on the point cloud data to obtain a centered point cloud; construct a surface elevation grid based on the centered point cloud and fill in the missing areas in the surface elevation grid with effective support constraints to obtain a continuous surface elevation field. (2) Main ridge direction estimation A trend surface is constructed based on the continuous surface elevation field, and a residual elevation field is obtained from the difference between the continuous surface elevation field and the trend surface; the main ridge orientation of the plot is estimated based on the residual elevation field, and a ridge orientation alignment coordinate system is constructed accordingly. (3) Furrow tracking The residual elevation field is segmented along the ridge direction in the ridge alignment coordinate system, and candidate furrow positions are detected in each segment based on cross-sectional depression features. Cross-segment matching and continuous tracking are performed on the candidate furrow positions in each segment to obtain continuous furrow trajectories. (4) Ridging quality inspection The single-ridge area is divided according to the adjacent continuous furrow trajectory; the ridge width, ridge height and ridge surface width are calculated based on the single-ridge area, and the ridge width, ridge height and ridge surface width are compared according to the preset quality threshold, and the ridge operation quality detection result is output.

2. The method for quality inspection of ridge-making operations in cold regions as described in claim 1, characterized in that, The process of constructing the surface elevation grid in step (1) includes: projecting the centered point cloud onto a two-dimensional plane grid according to a preset grid resolution; and statistically analyzing the point cloud elevation values ​​within each grid cell to generate a median elevation grid. And further generate minimum elevation raster. and maximum elevation grid These serve as auxiliary elevation representations for furrow detection and ridge height calculation, respectively. For the above elevation grid , , When filling in missing regions, a Gaussian normalized convolution method based on an effective mask is used to locally estimate the missing pixels using only effective elevation pixels. For missing pixels with insufficient effective neighborhood support, they are either kept as null values ​​or marked as invalid regions to obtain a continuous surface elevation field. , , .

3. The method for quality inspection of field ridging operations in cold regions as described in claim 1, characterized in that, The process of constructing the residual elevation field in step (2) is as follows: for the continuous surface elevation field Perform large-scale smoothing to obtain a low-frequency trend surface that characterizes the overall undulation background of the land parcel. ; Continuous surface elevation field , , Subtract trend surface The residual elevation field is obtained. The residual elevation field is used to characterize the local micro-topographic texture formed by ridge formation; Calculate the residual elevation field Gradient components in two orthogonal directions ; Construct a global structure tensor based on the gradient components. According to the global structure tensor The principal feature direction determines the principal gradient direction. ; and with the main gradient direction The direction perpendicular to the main ridge of the plot is determined as the direction of the main ridge. ; Utilizing the main ridge direction of the plot Construct a ridge-direction coordinate system such that the first coordinate axis extends along the ridge direction and the second coordinate axis extends across the ridge direction.

4. The method for quality inspection of ridge-making operations in cold regions as described in claim 1, characterized in that, The process of detecting candidate furrow locations in step (3) includes: converting the median residual elevation field in the furrow-direction coordinate system into a single point. Process the data in segments with a fixed step size along the ridge direction, and extract a one-dimensional cross-sectional profile along the transverse ridge direction within each segment. ; in the one-dimensional cross-sectional section Local minimum points are detected in the middle; and candidate furrow locations are determined by combining the valley bottom platform determination rule or the neighborhood comparison rule; Combining the preset ridge spacing range, candidate position spacing constraints, and candidate position elevation characteristics, the candidate ridge positions are screened to remove false ridge candidates caused by noise, missing measurements, or local disturbances; a matching cost function for candidate ridge positions between adjacent segments is established; the matching relationship is optimized based on position offset continuity, ridge spacing consistency, and trajectory smoothness; the optimized matching results are connected into continuous ridge trajectories; ridge trajectories that are locally missing but meet the continuity constraints are completed; and trajectories with a support length lower than a preset threshold or abnormal continuity are removed to obtain continuous ridge trajectories with stable numbers.

5. The method for quality inspection of ridge-making operations in cold regions as described in claim 1, characterized in that, The process of dividing a single-ridge area in step (4) is as follows: using the trajectories of two adjacent continuous furrows... The left and right boundaries define the corresponding lateral range of a single ridge; and along the ridge direction, corresponding grid areas are extracted according to the left and right boundaries to form a single ridge area; the ridge width is determined by the lateral distance between the trajectories of adjacent continuous furrows within the same single ridge area; the ridge height is determined by the representative value of the ridge top elevation within the single ridge area. Representative value of furrow baseline elevation The difference is determined; the ridge width is determined by the width of the top continuous interval in the cross section of a single ridge that meets the elevation discrimination condition, and the elevation corresponding to the top interval is within 55% to 70% of the height difference between the bottom of the furrow and the top of the ridge.

Images