A method and system for identifying early-stage characteristics of settlement and displacement on highway slopes

Abstract

This invention relates to the field of engineering monitoring methods, and particularly to a method and system for identifying early-stage features of settlement and displacement on highway slopes. The method includes: constructing a slope stability confidence map containing several slope monitoring grid units based on the statistical characteristics of the initial slope deformation field; determining the stability region weights of each point cloud point in the current point cloud data based on the slope stability confidence map; performing weighted registration of the current point cloud data and the reference point cloud data based on the stability region weights; updating the current point cloud data according to update transformation parameters; recalculating the slope deformation field based on the updated registered point cloud and the reference point cloud data; and updating the slope stability confidence map based on the recalculated slope deformation field. This invention achieves a closed-loop feedback between the registration process and deformation perception by constructing a slope stability confidence map, ensuring that stable regions dominate the registration while retaining differences in deformed regions, thereby reliably extracting early-stage features.

Description

Technical Field

[0001] This invention relates to the field of engineering monitoring methods, and in particular to a method and system for identifying early-stage features of settlement and displacement on highway slopes. Background Technology

[0002] In highway reconstruction and expansion projects, areas such as the junction of the old and new roadbeds on high embankment slopes and sections where retaining walls have been demolished are high-risk areas for deformation and collapse. In recent years, UAV lidar technology, due to its non-contact, high-precision, and full-coverage characteristics, has been gradually applied to slope deformation monitoring. Its typical workflow involves periodically collecting multi-period point cloud data, registering and aligning the current point cloud with the previous reference point cloud, calculating the elevation difference, and thus identifying deformation areas such as settlement, cracks, and heave. Point cloud registration is a crucial step in achieving multi-period comparison. Currently, the Iterative Closest Point (ICP) algorithm and its improved algorithms are commonly used in engineering. This involves finding the optimal rigid body transformation (rotation matrix R and translation vector T) between two point clouds to align them spatially. The basic assumption of this type of algorithm is that only a rigid body transformation relationship exists between the two point clouds, meaning all points follow the same motion laws.

[0003] However, in actual engineering scenarios of slope monitoring, this assumption deviates from reality. The actual deformation of a slope is non-rigid; some areas (such as stable slopes) should remain aligned, while others (such as cracked zones and slip zones) exhibit real spatial displacement. Existing registration methods treat all areas equally, seeking the globally optimal rigid transformation. This leads to the registration algorithm compromising by using features of deformed areas to calculate transformation parameters when real deformation exists. This results in decreased alignment accuracy in stable areas, and the actual deformation is partially absorbed into the transformation parameters, distorting the difference results. Furthermore, steep slopes often have repetitive textures (such as the regular grid of arched slope protection and vegetation cover), making feature point matching prone to ambiguity. This causes registration to fall into local optima, generating errors. These errors exceed the deformation of early cracks, making early damage unreliable. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for identifying early-stage damage characteristics of highway slope settlement and displacement, aiming to solve the technical problems mentioned in the background art.

[0005] This invention proposes a method for identifying early-stage characteristics of settlement and displacement on highway slopes, including: Acquire baseline point cloud data and current point cloud data for the slope monitoring area; The initial registration point cloud is obtained based on the reference point cloud data and the current point cloud data. The initial slope deformation field is calculated based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. Based on the statistical characteristics of the initial slope deformation field, a slope stability confidence map containing several slope monitoring grid units is constructed. Each slope monitoring grid unit corresponds to a stability confidence value, which is used to characterize the stability region weight that the slope monitoring grid unit should be assigned during the registration process. The stability region weights of each point cloud point in the current point cloud data are determined based on the slope stability confidence map. The current point cloud data and the reference point cloud data are then weighted and registered according to the stability region weights to obtain update transformation parameters. The current point cloud data is then updated according to the update transformation parameters to obtain the updated registered point cloud. The slope deformation field is recalculated based on the updated registration point cloud and the reference point cloud data, and the slope stability confidence map is updated based on the recalculated slope deformation field. Repeat the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence condition to obtain the converged slope deformation field. Early disease characteristics were extracted based on the convergent slope deformation field.

[0006] Preferably, the step of calculating the initial slope deformation field based on the initial registration point cloud and the reference point cloud data includes: The slope monitoring area is projected onto a horizontal plane and divided into regular grids to obtain several slope grid units; Based on the ground points in the benchmark cloud data, a digital elevation model of each slope grid unit is constructed to obtain the three-dimensional coordinates of the center point of each slope grid unit. Obtain the elevation gradient of adjacent slope grid cells, and pre-calculate the normal vector of each slope grid cell based on the elevation gradient; The three-dimensional coordinates of each point in the initial registration point cloud are obtained, and each point in the initial registration point cloud is located to the corresponding slope grid cell. Based on the three-dimensional coordinates of the point in the initial registration point cloud and the three-dimensional coordinates and normal vector of the center point of the corresponding slope grid cell, the slope settlement displacement of each point in the initial registration point cloud to the plane corresponding to the slope grid cell is calculated to obtain the initial slope deformation field.

[0007] Preferably, the step of constructing a slope stability confidence map containing several slope monitoring grid units based on the statistical characteristics of the initial slope deformation field includes: The slope monitoring area is divided into several regular slope monitoring grid units; For each slope monitoring grid cell, the slope settlement displacement of all initial registration point cloud points falling within the slope monitoring grid cell is counted to obtain the set of slope settlement displacement. Calculate the median and median absolute deviation of the set of slope settlement displacements, where the median absolute deviation is the median of the absolute deviations of each slope settlement displacement from the median. Obtain the normal vector of the slope grid cell corresponding to each slope monitoring grid cell, and calculate the slope inclination angle based on the normal vector; The initial stability confidence value of the slope monitoring grid cell is calculated based on the median, the median absolute deviation, and the slope inclination angle: If the median absolute deviation is less than a preset sensor noise threshold, then the initial stability confidence value is calculated based on the absolute value of the median. If the median absolute deviation is greater than or equal to the preset sensor noise threshold, the initial stability confidence value is set to the preset low stability confidence value, indicating that the deformation in the region is inconsistent or there is measurement noise. The initial stability confidence values ​​of each slope monitoring grid unit are combined to obtain the initial slope stability confidence map.

[0008] Preferably, the step of combining the initial stability confidence values ​​of each slope monitoring grid unit to obtain the initial slope stability confidence map includes: Obtain historical slope stability confidence maps, which are slope stability confidence maps after convergence in the previous monitoring period; Determine the current monitoring period number. If the current monitoring period is not the first monitoring period, perform a weighted fusion of the initial slope stability confidence map and the historical slope stability confidence map to obtain the current slope stability confidence map. If the current monitoring period is the first monitoring period, obtain the slope stability confidence map based on conservative initial values.

[0009] Preferably, the step of determining the stable region weight of each point in the current point cloud data based on the slope stability confidence map, and performing weighted registration of the current point cloud data and the reference point cloud data based on the stable region weights to obtain the updated transformation parameters includes: Based on the slope stability confidence map, establish the mapping relationship between each point cloud point in the current point cloud data and the slope monitoring grid unit, and use the stability confidence value of the slope monitoring grid unit to which each point cloud point belongs in the current point cloud data as the stability region weight of that point cloud point. The weighted registration objective function is constructed using the M-estimation loss function; Solve the weighted registration objective function to obtain the updated rotation matrix and the updated translation vector, and use the updated rotation matrix and the updated translation vector as the updated transformation parameters.

[0010] Preferably, the step of repeating the weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence condition to obtain the converged slope deformation field includes: In the first few iterations, a pre-set coarse-scale slope monitoring grid is used to calculate and update the slope stability confidence map. The grid side length of the coarse-scale slope monitoring grid is the first side length. During the iteration process, for slope monitoring grid cells in the preset coarse-scale slope monitoring grid that meet the local refinement conditions, a preset fine-scale slope monitoring grid is used for local densification calculation. The grid side length of the fine-scale slope monitoring grid is the second side length, which is smaller than the first side length. The local refinement conditions include: the slope deformation field gradient within the slope monitoring grid cell exceeds a preset slope deformation gradient threshold, or the stability confidence value of the slope monitoring grid cell is lower than a preset confidence threshold, which correspond to the local deformation abrupt change region and the potential instability region of the slope, respectively. After each iteration, the change between the current slope stability confidence map and the previous slope stability confidence map is calculated; when the change is less than a preset convergence threshold, the slope stability confidence map is determined to meet the convergence condition, and the iteration stops.

[0011] Preferably, the step of extracting early disease characteristics based on the convergent slope deformation field includes: Calculate the gradient of the convergent slope deformation field to obtain the slope deformation gradient field; Based on the slope deformation gradient field, regions where the slope deformation gradient exceeds a preset gradient threshold are extracted as candidate regions for microcracks. Extract the geometric features of the microcrack candidate region, including the principal orientation angle of the centerline of the microcrack candidate region, the standard deviation of the width along the centerline, and the spatial continuity of the centerline; The microcrack candidate regions are classified according to the geometric features: if the angle between the main direction angle and the slope direction is less than a preset angle threshold and the centerline length is greater than a first length threshold, it is determined to be a slope drainage ditch or a skeleton joint artificial structure; if the standard deviation of the width along the centerline is less than a preset width standard deviation threshold and the centerline length is greater than a second length threshold, it is determined to be an artificial structure; the remaining microcrack candidate regions are determined to be real microcracks. Extract the continuous regions in the convergent slope deformation field where the absolute value of slope settlement displacement exceeds a preset deformation threshold, and use them as candidate regions for micro-uplift or micro-settlement. The transition zone from the low confidence region to the high confidence region in the slope stability confidence map is extracted as the leading edge boundary of the slip body, and the leading edge range of the slip body is extracted based on the area enclosed by the leading edge boundary of the slip body.

[0012] This invention also provides a system for identifying early-stage damage characteristics of highway slope settlement and displacement, comprising: The data acquisition module acquires the baseline point cloud data and the current point cloud data of the slope monitoring area; The initial registration module obtains the initial registration point cloud based on the reference point cloud data and the current point cloud data; The deformation calculation module calculates the initial slope deformation field based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. The confidence building module constructs a slope stability confidence map containing several slope monitoring grid units based on the statistical characteristics of the initial slope deformation field. Each slope monitoring grid unit corresponds to a stability confidence value, which is used to characterize the stability region weight that the slope monitoring grid unit should be assigned during the registration process. The weighted registration module determines the stable region weight of each point in the current point cloud data based on the slope stability confidence map, performs weighted registration between the current point cloud data and the reference point cloud data based on the stable region weight, obtains update transformation parameters, and updates the current point cloud data based on the update transformation parameters to obtain the updated registered point cloud. The update module recalculates the slope deformation field based on the updated registration point cloud and the reference point cloud data, and updates the slope stability confidence map based on the recalculated slope deformation field. The iterative convergence module repeats the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence conditions, thus obtaining the converged slope deformation field. The disease extraction module extracts early disease characteristics based on the convergent slope deformation field. The early disease characteristics include at least one of microcracks, micro-bulges, micro-settlement, and the leading edge of the slip body.

[0013] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for identifying early-stage features of settlement and displacement of highway slopes.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for identifying early-stage features of settlement and displacement of highway slopes.

[0015] The beneficial effects of this invention are as follows: This invention achieves closed-loop feedback between the registration process and deformation perception by constructing a slope stability confidence map, so that the stable region dominates the registration and the deformation region retains the differences, thereby resolving the contradiction between the rigid body registration assumption and the real deformation of non-rigid bodies. On this basis, the M estimation loss function is used to automatically reduce the weight of point pairs with excessive registration residuals, further suppressing the feature point matching ambiguity caused by repeated textures, and realizing the reliable extraction of early defects. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] like Figure 1 As shown, this application provides a method for identifying early-stage features of settlement and displacement on highway slopes, characterized by comprising: S1, acquire the baseline point cloud data and current point cloud data of the slope monitoring area; The baseline point cloud data represents the initial state of the slope or the state after convergence in the previous stable cycle. It is collected before the first monitoring period and serves as the comparison benchmark for all subsequent monitoring periods. The current period's point cloud data is collected according to the preset monitoring cycle, with each monitoring cycle corresponding to one period of slope monitoring data. The setting of the monitoring cycle depends on the construction stage and the slope risk level. In this embodiment, monitoring is conducted daily during high-risk periods (such as after retaining wall demolition or after rain) and every three days during medium-risk periods. S2, perform initial registration based on the reference point cloud data and the current point cloud data to obtain initial transformation parameters, and transform the current point cloud data according to the initial transformation parameters to obtain the initial registered point cloud; The purpose of initial registration is to transform the current point cloud data to the coordinate system of the reference point cloud data, establishing a spatial alignment basis for subsequent deformation calculations. Initial registration employs a random sampling consensus algorithm based on fast point feature histogram features. The initial transformation parameters include a rotation matrix and a translation vector, which are applied to the current point cloud to obtain the initially registered point cloud.

[0021] S3, calculate the initial slope deformation field based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. S4. Construct a slope stability confidence map based on the statistical characteristics of the initial slope deformation field. Each slope monitoring grid unit in the slope stability confidence map corresponds to a stability confidence value. The stability confidence value is used to characterize the stability region weight that the slope monitoring grid unit should be assigned during the registration process. S5. Determine the stable region weight of each point in the current point cloud data according to the slope stability confidence map, and perform weighted registration of the current point cloud data and the reference point cloud data according to the stable region weight to obtain the update transformation parameters, and update the current point cloud data according to the update transformation parameters to obtain the updated registration point cloud. S6. Recalculate the slope deformation field based on the updated registration point cloud and the reference point cloud data, and update the slope stability confidence map based on the recalculated slope deformation field. S7. Repeat the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence condition to obtain the converged slope deformation field. S8. Early disease characteristics are extracted based on the convergent slope deformation field. The early disease characteristics include at least one of microcracks, micro-heavysms, micro-settlement, and the leading edge of the slip body.

[0022] As described in steps S1-S8 above, this invention constructs a feedback mechanism of deformation perception, weighted registration, and iterative optimization, enabling the registration process to distinguish between stable and deformable regions, allowing stable regions to dominate registration while retaining differences in deformable regions. Through multi-scale grid switching and convergence judgment, it achieves a reasonable allocation of computational resources in space and time. Through multi-dimensional geometric feature classification, it achieves automatic identification of various diseases such as cracks, uplift, settlement, and slip bodies. In slope monitoring scenarios, slope deformation is localized and progressive. Stable regions occupy most of the slope area, providing sufficient constraints for weighted registration. The spatial range of deformable regions is limited, and low weighting minimizes their impact on registration. Cracks and artificial structures have significant differences in geometric features, which can be used for differentiation. The spatial abrupt change in the confidence map corresponding to the leading edge of a slip body can be used for boundary extraction. This scheme solves the problems of deformation-induced registration, high false alarm rates of artificial structures, and difficulty in quantitatively extracting slip body boundaries in existing technologies, achieving reliable identification of early-stage diseases.

[0023] In one embodiment of the present invention, the step of calculating the initial slope deformation field based on the initial registration point cloud and the reference point cloud data includes: S31, Project the slope monitoring area onto the horizontal plane, divide it into regular grids, and obtain several slope grid units; S32, construct a digital elevation model for each slope grid unit based on the ground points in the benchmark point cloud data, and obtain the three-dimensional coordinates of the center point of each slope grid unit; S33, obtain the elevation gradient of adjacent slope grid cells, and pre-calculate the normal vector of each slope grid cell based on the elevation gradient; S34, obtain the three-dimensional coordinates of each point in the initial registration point cloud, locate each point in the initial registration point cloud to the corresponding slope grid cell, and calculate the slope settlement displacement from each point in the initial registration point cloud to the plane corresponding to the corresponding slope grid cell based on the three-dimensional coordinates of the point in the initial registration point cloud and the three-dimensional coordinates and normal vector of the center point of the corresponding slope grid cell, to obtain the initial slope deformation field; (for sparse area points that cannot be located to effective slope grid cells, calculate the signed distance between the point and the nearest neighbor point in the reference point cloud data, and use it as the slope settlement displacement of the point).

[0024] Calculation of slope settlement and displacement: ; In the formula, This indicates the slope settlement displacement (in meters; a positive value indicates that the point has risen relative to the reference surface, and a negative value indicates settlement). The three-dimensional coordinates (in meters, obtained from the digital elevation model) of the center point of the slope grid cell represent the slope grid cell center point. This represents the three-dimensional coordinates of the points in the initial registration point cloud (in meters, read directly from the point cloud data). The normal vector of the slope grid cell (dimensionless, calculated from the elevation gradient of adjacent grid cells, and normalized to satisfy) ).

[0025] In slope point cloud data processing, deformation field calculations typically employ two methods: point-to-point distance or point-to-triangular mesh distance. Point-to-point distance calculation yields good results under ideal conditions of uniform point cloud density and no noise. However, in actual slope monitoring, point cloud data is subject to various interference factors such as uneven sampling, vegetation shading, and sensor noise. Point-to-point distance can easily misinterpret sampling differences as deformation, and the calculation results are significantly affected by the selection of nearest neighbor points. While point-to-triangular mesh distance provides smoother results, triangular mesh construction itself consumes substantial computational resources. During multi-round registration processes, rebuilding the triangular mesh in each iteration leads to a sharp increase in computational load, making it difficult to meet the real-time requirements of high-frequency monitoring during construction. Furthermore, the orientation of the triangular mesh normal vectors depends on the order of the triangular facet vertices. In complex terrain areas, inconsistent normal vector orientations can occur, leading to errors in the sign determination of signed distances, misclassifying uplift as settlement or vice versa.

[0026] To address the aforementioned issues, as described in steps S31-S34 above, this invention discretizes the continuous curved surface into a regular grid. By predicting the algorithm vector field and the grid elevation model, the complexity of single-point distance calculation is reduced to a constant level while ensuring the accuracy of deformation field calculation. At the same time, a unified normal vector calculation rule ensures consistent orientation, providing efficiency support for subsequent multi-round iterative registration.

[0027] The slope monitoring area is projected onto a horizontal plane and divided into regular grids to obtain a slope grid cell set. The selection of the grid side length needs to consider two factors: too large a side length will increase the surface discretization error, making it difficult to reflect local slope details; too small a side length will result in insufficient point cloud points within the grid cell, leading to unstable statistical results. In this embodiment, 0.1 meters is chosen, a value comparable to the average point cloud spacing, achieving a balance between preserving terrain details and ensuring statistical stability.

[0028] Based on ground points in the benchmark point cloud data, a digital elevation model (DEM) is constructed for each slope raster cell. For each raster cell, the elevation values ​​of all benchmark point cloud ground points falling within that cell are statistically analyzed, and the median is taken as the elevation value of that raster cell. The median is used instead of the average because, in slope environments, vegetation points may be misclassified as ground points and remain in the data. The median has a natural resistance to such outliers, preventing a single anomaly from significantly affecting the raster elevation value. The three-dimensional center point coordinates of the raster cell are composed of the cell's planar coordinates on the horizontal plane and its elevation value.

[0029] Based on the elevation gradient of adjacent slope grid cells, the normal vector of each slope grid cell is pre-calculated. For the grid cell located at coordinates (i,j), the horizontal component of its normal vector is calculated using the ratio of the elevation difference between adjacent grid cells to the grid spacing. The vertical component of the normal vector is obtained through normalization calculation to ensure that all normal vectors are unit vectors. Since the grid cells are regularly arranged on the horizontal plane, this calculation rule ensures that the normal vectors of all grid cells have the same orientation, pointing outwards from the slope. After the normal vectors of all grid cells are calculated, a slope normal vector field is formed. This normal vector field remains unchanged throughout the entire iteration process and does not need to be recalculated in each iteration.

[0030] For any point in the initial registration point cloud, the corresponding target slope grid cell is located based on the point's planar coordinates. The location process is achieved by dividing the x and y coordinates of the point cloud by the grid side length and rounding, with a time complexity of O(1). After obtaining the three-dimensional center point coordinates and normal vector of the grid cell, the signed distance from the point to the plane corresponding to the grid cell is calculated. The calculation formula is the dot product of the difference vector between the point cloud coordinates and the grid center point coordinates and the normal vector. The physical meaning of this calculation method is that the slope surface represented by the reference point cloud is regarded as a folded surface composed of grid plane pieces, and the signed distance from the current point cloud point to this folded surface is the slope settlement displacement of that point. A positive result indicates that the point is uplifted relative to the reference surface, and a negative result indicates settlement.

[0031] For sparse areas where points cannot be located to valid slope grid cells, a fallback mechanism is used. The causes of sparse areas include: vegetation obstruction preventing ground points from being located within the grid cell, steep cliff areas being difficult for lasers to reach, and void areas lacking contrast points. For these points, the nearest neighbor is searched in the baseline point cloud data, and the signed distance between the point and its nearest neighbor is calculated as the slope settlement displacement for that point. Sparse areas typically constitute a small proportion of slopes, mainly appearing at the edges of steep cliffs or in densely vegetated areas. Using the nearest neighbor distance as an alternative ensures data integrity and avoids interruptions to the entire processing flow due to invalid individual grid cells.

[0032] By utilizing the regular distribution characteristics of point cloud data in slope monitoring scenarios, this invention discretizes continuous curved surfaces into regular grids. Through rasterization, complex geometric calculations in three-dimensional space are transformed into index queries on a two-dimensional plane. In slope monitoring scenarios, the monitoring area is typically a narrow strip with a limited projection range on the horizontal plane, allowing for controllable grid division. Furthermore, slope deformation monitoring focuses on the relative changes of the same spatial location at different times, rather than absolute geometric shapes; therefore, the approximate representation on the grid plane is sufficient to meet accuracy requirements. Additionally, the monitoring process requires multi-period data comparison and multiple rounds of iterative calculations, and the pre-calculated data after rasterization can be reused. This solves the efficiency bottleneck problem in deformation field calculation in existing technologies. In traditional point-to-triangular mesh distance calculations, each iteration requires rebuilding the triangular mesh, and the nearest face search requires traversing the triangular mesh structure. In the registration process that requires multiple iterations, the cumulative time consumption of this calculation method is significant. This step changes the real-time construction of the triangulation mesh to the pre-computation of the raster and the search for the nearest face of the triangulation mesh to the constant-time localization of the raster plane, so that the total amount of computation in the multi-iteration scenario is controllable.

[0033] In one embodiment of the present invention, the step of constructing a slope stability confidence map based on the statistical characteristics of the initial slope deformation field includes: S41 divides the slope monitoring area into several regular slope monitoring grid units; S42, For each slope monitoring grid unit, count the slope settlement displacement of all initial registration point cloud points falling into the slope monitoring grid unit to obtain the slope settlement displacement set; S43, calculate the median and the median absolute deviation of the set of slope settlement displacements, wherein the median absolute deviation is the median absolute deviation of each slope settlement displacement from the median; S44, obtain the normal vector of the slope grid cell corresponding to each slope monitoring grid cell, and calculate the slope inclination angle based on the normal vector; S45, calculate the initial stability confidence value of the slope monitoring grid cell based on the median, the median absolute deviation, and the slope inclination angle: S451, if the median absolute deviation is less than the preset sensor noise threshold, then the initial stability confidence value is calculated based on the absolute value of the median. The initial stability confidence value is negatively correlated with the absolute value of the median, that is, the smaller the slope settlement displacement, the more stable the area. S452, if the median absolute deviation is greater than or equal to the preset sensor noise threshold, then the initial stability confidence value is set to the preset low stability confidence value, indicating that the deformation in the area is inconsistent or there is measurement noise. S46. Combine the initial stability confidence values ​​of each slope monitoring grid unit to obtain the initial slope stability confidence map.

[0034] In slope deformation monitoring, different regions should contribute differently to the registration process: stable regions should dominate the calculation of registration parameters to ensure the accuracy of overall alignment; deformable regions should be downplayed in the registration process to preserve their true displacement information. How to accurately distinguish between stable and deformable regions is a key problem that needs to be solved in this field. Existing methods typically use the absolute value of deformation as the criterion, i.e., regions with small deformation are considered stable regions, and regions with large deformation are considered deformable regions. However, in actual slope scenarios, this method has several drawbacks. First, factors such as sensor noise and vegetation residue can lead to errors in deformation calculation, and relying solely on the absolute value of deformation can easily misjudge noise as deformation. Second, the same deformation represents different risks of instability in steep and gentle slope regions, and the absolute value of deformation alone cannot reflect the inherent stability differences of the slope.

[0035] To address the aforementioned issues, as described in steps S41-S46, this invention divides the monitoring area into regular grid cells. Within each cell, the median of deformation and the median of absolute deviation are statistically analyzed. The median reflects the overall deformation trend, while the median of absolute deviation reflects the consistency of deformation. Simultaneously, the slope angle is introduced as a topographic prior, allowing steep slope areas to achieve lower confidence levels for the same deformation. Through this design, the stability confidence map can more accurately reflect the weights that should be assigned to each region during the registration process.

[0036] The slope monitoring area is divided into regular slope monitoring grid units, resulting in a slope monitoring grid unit set. The selection of grid side length needs to balance statistical stability and spatial resolution: if the side length is too large, multiple deformation features will fall into the same grid unit, making them indistinguishable; if the side length is too small, the number of point cloud points in the grid unit will be insufficient, leading to unstable statistical results. In this embodiment, 0.5 meters is chosen, which matches the spatial scale of slope deformation features. Each grid unit contains an average of about 5 to 12 point cloud points, which meets the statistical requirements.

[0037] For each slope monitoring grid cell, the slope settlement displacement of all initially registered point cloud points falling within that grid cell is statistically analyzed to obtain a set of slope settlement displacement values. These slope settlement displacement values ​​are derived from the aforementioned projection raster method calculation results and reflect the degree of heave or settlement of each point within that grid cell relative to the baseline state.

[0038] Calculate the median and median absolute deviation of the set of slope settlement displacements. The median represents the overall deformation trend of the grid cell; a larger absolute value indicates more significant deformation in that area. The median absolute deviation reflects the dispersion of deformation at each point relative to the median; a larger value indicates more inconsistent deformation within the grid cell, potentially including crack boundaries, local disturbances, or measurement noise. The formula for calculating the median absolute deviation is: the median of the absolute values ​​of the differences between each deformation and the median. This statistic is more robust to outliers than the standard deviation, preventing individual anomalies from excessively influencing the consistency assessment.

[0039] Obtain the normal vector of the slope grid cell corresponding to each slope monitoring grid cell, and calculate the slope inclination angle based on this normal vector. The slope grid cells are derived from the aforementioned regular grid division, and each slope monitoring grid cell corresponds to multiple slope grid cells. The normal vector is obtained by taking the average of the normal vectors of all slope grid cells within the area covered by the slope monitoring grid cell. The slope inclination angle is calculated as follows: the slope inclination angle is equal to the inverse cosine of the Z component of the normal vector. This inclination angle reflects the steepness of the terrain in the area; a larger value indicates a steeper slope.

[0040] The initial stability confidence value of the slope monitoring grid cell is calculated based on the median, median absolute deviation, and slope inclination angle. The specific judgment rules are as follows: If the median absolute deviation is less than the preset sensor noise threshold, it indicates that the deformation within the grid cell is relatively consistent, eliminating noise interference. In this case, the initial stability confidence value is calculated based on the absolute value of the median and the slope inclination angle. If the median absolute deviation is greater than or equal to the preset sensor noise threshold, it indicates that the deformation within the grid cell is inconsistent, possibly containing crack boundaries or measurement noise. In this case, the initial stability confidence value is set to the preset low stability confidence value, indicating that this area should be weakened during registration.

[0041] The preset sensor noise threshold is set based on the measurement accuracy of the lidar. In this embodiment, the vertical measurement accuracy of the lidar used is 0.01 meters, so the noise threshold is set to 0.01 meters. The preset low stability confidence value is set based on the fact that for regions with inconsistent deformation, even if there is actual deformation, its deformation mode is not suitable as a registration reference and should be given a lower weight. In this embodiment, it is set to 0.3.

[0042] When the absolute value of the median is small, the initial stability confidence value should be close to 1, indicating that the region is stable; when the absolute value of the median increases, the initial stability confidence value should gradually decrease, indicating that the region is deformed. To achieve this mapping relationship, this embodiment uses an exponential decay function: the initial stability confidence value is equal to an exponential function with the natural constant as the base and the product of the negative absolute value of the median and the slope sensitivity coefficient as the exponent, divided by the deformation sensitivity parameter. The slope sensitivity coefficient is equal to 1 plus the slope sensitivity calibration coefficient multiplied by the sine of the slope angle, and its function is to obtain a lower confidence value for steep slope regions under the same amount of deformation. The deformation sensitivity parameter controls the rate at which the confidence value decays with the amount of deformation; in this embodiment, it is taken as 0.01 meters. The slope sensitivity calibration coefficient controls the intensity of the influence of the slope angle on the confidence value; in this embodiment, it is taken as 1.0.

[0043] The initial stable confidence level is calculated as follows: ; In the formula, This represents the initial stability confidence value calculated based on the current slope deformation field (dimensionless, value range [0,1], the larger the value, the more stable the slope monitoring grid unit). This represents the median of the slope settlement displacement set within the slope monitoring grid unit (in meters, reflecting the overall deformation trend of the area). This represents the slope deformation sensitivity parameter (unit: meters, value: 0.01m, controlling the sensitivity of stability confidence to slope settlement displacement, calibrated through engineering tests). Represents the natural constant; ; In the formula, This represents the slope sensitivity coefficient (dimensionless, range of values). gentle slope When the time approaches 1, the slope is steep. As the time approaches 1+λ, the confidence level of a steep slope decays more rapidly. This represents the slope sensitivity calibration coefficient (dimensionless, value range [0.5, 2.0], calibrated through engineering tests to control the intensity of the influence of slope on the stability confidence value). The sine value represents the slope angle (dimensionless, larger for steep slopes, smaller for gentle slopes). ; In the formula, The slope angle (slope angle, in radians, ranging from 0 to π / 2) represents the slope monitoring grid cell and is calculated from the Z component of the grid cell's normal vector. Indicates level, The larger the value, the steeper the slope. Represents the inverse cosine function. The Z component represents the normal vector of the slope grid cell (dimensionless, normalized value range [-1, 1], obtained by pre-calculation of the elevation gradient of adjacent grid cells). The initial stability confidence values ​​of each slope monitoring grid cell are combined to obtain an initial slope stability confidence map. This confidence map corresponds to the spatial distribution of the slope monitoring grid cells. Each grid cell has a confidence value ranging from 0 to 1. A high value indicates that the area is stable and should be given high weight in the registration; a low value indicates that the area may deform and should be weakened in the registration.

[0044] This invention leverages the local spatial correlation of slope deformation, transforming discrete point cloud deformation quantities into continuous regional stability assessments through grid aggregation. It uses the median absolute deviation as a consistency criterion to distinguish between uniform deformation and local disturbances, and employs the slope angle as a topographic prior to ensure the confidence assessment aligns with the physical laws of slope engineering. In slope monitoring scenarios, slope deformation typically manifests as regional settlement or uplift rather than isolated point changes; grid aggregation effectively suppresses the influence of point cloud noise. Abrupt deformation occurs at crack boundaries, with a significantly increased median absolute deviation, serving as an early signal for crack boundary identification. Steep slopes have lower stability reserves and a higher risk of instability for the same deformation; introducing a slope prior makes the confidence assessment more consistent with engineering realities. This solution overcomes the limitations of existing technologies that rely solely on the absolute value of deformation for stability judgment, enabling the confidence assessment to consider both dynamic deformation and inherent topographic stability, achieving a more comprehensive characterization of slope stability. Building upon this foundation, the confidence map constructed in this step serves as the weighting basis for subsequent weighted registration, enabling the registration process to adaptively adjust according to the stability of each region: high confidence in stable regions ensures they dominate the calculation of registration parameters, resulting in more accurate overall alignment; low confidence in deformable regions and crack boundary regions weakens their role in registration, preventing the loss of true deformation due to forced alignment. Simultaneously, the application of the median absolute deviation in this step complements the aforementioned projection raster method. The former focuses on the spatial aggregation of deformation features, while the latter focuses on the rasterization of point cloud data. The combination of these two methods ensures that the data processing from the original point cloud to the confidence map has a complete spatial scale hierarchy.

[0045] In one embodiment of the present invention, the step of combining the initial stability confidence values ​​of each slope monitoring grid unit to obtain an initial slope stability confidence map includes: S47, Obtain the historical slope stability confidence map, which is the slope stability confidence map after convergence in the previous monitoring period; S48, determine the current monitoring period number. If the current monitoring period is not the first monitoring period, perform weighted fusion of the initial slope stability confidence map and the historical slope stability confidence map to obtain the current slope stability confidence map. If the current monitoring period is the first period, obtain the slope stability confidence map based on conservative initialization values, and set the stability confidence values ​​of all slope monitoring grid units to preset conservative values.

[0046] Weighted fusion formula: ; In the formula, This represents the slope stability confidence value for the current period (dimensionless, range [0,1], with a conservative initial value of 0.7 for the first monitoring period, indicating a moderately high stability confidence level, to avoid significant deviations in confidence level due to initial registration errors). This represents the initial stable confidence level. This represents the historical slope stability confidence value (dimensionless, obtained from the slope stability confidence map after convergence in the previous monitoring period). This represents the fusion coefficient (dimensionless, with a value of 0.5 in this implementation, balancing the weights of current monitoring data and historical prior information).

[0047] As described in steps S47-S48 above, during long-term slope monitoring, single-period monitoring data may contain random errors due to factors such as weather conditions, flight attitude, and changes in point cloud density. If the stability confidence map is calculated solely based on the current data, these random errors may cause fluctuations in the confidence assessment, thereby affecting the stability of subsequent weighted registration. On the other hand, slope deformation is a gradual process. The confidence map obtained after the convergence of previous monitoring reflects the inherent stability characteristics of each area of ​​the slope, and this historical information has important reference value for current monitoring. However, when monitoring enters the first phase, there is no historical confidence map available for reference. Therefore, this invention obtains the final stability confidence map for the current phase by weighted fusion of the initial confidence map for the current phase and the historical confidence map, thus achieving the smoothing effect of historical information on the current assessment. For the first phase of monitoring, a conservative initialization strategy is adopted, setting the stability confidence value of all grid cells to a preset conservative value to avoid serious deviations in confidence due to initial registration errors.

[0048] Obtain historical slope stability confidence maps. These historical maps are slope stability confidence maps after convergence in the previous monitoring period and are stored in the monitoring system database. During slope construction, the monitoring period is typically daily or every three days. The historical maps reflect the stability status of each area after convergence in the previous monitoring period, encompassing historical knowledge of the inherent stability characteristics of the slope.

[0049] The current monitoring cycle number is determined by the number of monitoring sessions recorded by the system. The first monitoring session refers to the first inspection monitoring conducted after the initial acquisition of the baseline point cloud, at which point there is no historical confidence map available for reference. If the current monitoring cycle is not the first monitoring session, a weighted fusion of the initial slope stability confidence map and the historical slope stability confidence map is performed to obtain the current slope stability confidence map. The weighted fusion calculation formula is: the current stability confidence value equals the fusion coefficient multiplied by the initial stability confidence value plus one minus the fusion coefficient multiplied by the historical stability confidence value. The fusion coefficient ranges from 0 to 1; in this embodiment, it is set to 0.5, indicating that the current data and historical information have equal weight. The design of this fusion coefficient is based on the following considerations: the current initial confidence map is calculated based on the latest acquired point cloud data and can reflect the current state of the slope; the historical confidence map is based on the convergence results of multiple monitoring sessions and has good stability and noise resistance. Each factor carries equal weight, striking a balance between maintaining sensitivity to current changes and preserving the stability of historical understanding. If the current monitoring period is the first monitoring phase, the current slope stability confidence map is conservatively initialized, and the current stability confidence value of all slope monitoring grid units is set to a preset conservative value. The basis for setting the preset conservative value is: in the absence of historical information for reference, all areas should be assigned a moderately high confidence level to avoid some areas being incorrectly identified as deformation or stable areas due to initial registration errors. In this embodiment, the preset conservative value is 0.7. This value is neither equal to 1 (avoiding the forced identification of areas with possible minor deformation as completely stable) nor lower than 0.5 (avoiding the incorrect weakening of stable areas), reserving adjustment space for subsequent iterative optimization.

[0050] This invention leverages the temporal continuity of slope deformation by incorporating historical monitoring results into the current confidence assessment through weighted fusion, thus accumulating information over time. Conservative initialization in the first phase mitigates the risk of misjudgment due to insufficient data in the initial stage. This design aligns with long-term slope monitoring scenarios in the following ways: monitoring frequency is high during slope construction, slope state changes are limited between adjacent monitoring periods, and historical information has significant reference value for current assessment; the registration error between the baseline point cloud and the current point cloud is relatively large during the first monitoring period, and conservative initialization prevents the amplification of initial errors. This design addresses the problem of independent assessment for each monitoring period and neglect of historical information accumulation in existing methods, allowing the confidence assessment to gradually converge to a more accurate state as the monitoring cycle progresses. Regarding the collaborative relationship, this step receives the initial slope stability confidence map constructed in the preceding steps and outputs the current slope stability confidence map after historical fusion. This map provides the final weighting basis for subsequent weighted registration. The conservative initialization in the first phase works in conjunction with the subsequent iterative convergence mechanism: conservative initialization provides a relatively neutral starting point for subsequent iterations, allowing the first few iterations to gradually approach the true state without historical bias; while historical fusion from the previous phase provides better initial values ​​for subsequent iterations, accelerating the iterative convergence process. This step, together with the aforementioned projection raster method and grid aggregation statistics, forms a complete data processing chain: the projection raster method solves the conversion from point cloud data to the deformation field, grid aggregation statistics solves the conversion from the deformation field to the initial confidence level, and historical fusion solves the optimization from the initial confidence level to the current confidence level. Each step progresses sequentially and supports each other.

[0051] In one embodiment of the present invention, the step of determining the stable region weight of each point in the current point cloud data based on the slope stability confidence map, and performing weighted registration of the current point cloud data and the reference point cloud data based on the stable region weight to obtain the updated transformation parameters includes: S51, establish the mapping relationship between each point cloud point in the current point cloud data and the slope monitoring grid unit according to the slope stability confidence map, and use the stability confidence value of the slope monitoring grid unit to which each point cloud point belongs in the current point cloud data as the stability area weight of the point cloud point. S52 uses the M-estimation loss function to construct a weighted registration objective function, so that point cloud points with high weights in stable regions dominate the calculation of registration parameters, while points with low weights in stable regions (potential deformation regions) are weakened in the registration process. S53, solve the weighted registration objective function to obtain the updated rotation matrix and the updated translation vector, and use the updated rotation matrix and the updated translation vector as the updated transformation parameters.

[0052] As described in steps S51-S53 above, the core problem to be solved in this step is how to use the slope stability confidence map to guide the registration process after obtaining the confidence map. The traditional iterative nearest point algorithm treats all point cloud points as equally weighted, with its objective function being to minimize the sum of squared distances between all point pairs. This equal-weighting approach has significant drawbacks: when there is local deformation of the slope, the positions of the point cloud points in the deformed area undergo actual displacement. If the distances between these points and their corresponding points in the reference point cloud are included in the objective function, the registration algorithm will attempt to reduce these distances by adjusting the transformation parameters, resulting in the actual deformation being partially absorbed into the transformation parameters. Simultaneously, the alignment accuracy of the stable region will decrease due to the pull of the deformed region. Therefore, a registration mechanism that can adaptively adjust the weights according to the stability of each region is needed. This invention uses the stability confidence value of each point cloud point as the registration weight, allowing points with high weights in stable regions to dominate the calculation of registration parameters, while points with low weights in deformed regions are weakened during registration. Meanwhile, the Huber loss function is used instead of the traditional squared loss to automatically reduce the weight of point pairs with excessive registration residuals, preventing registration bias caused by incorrect matching or incorrect confidence.

[0053] Based on the slope stability confidence map, a mapping relationship is established between each point in the current point cloud data and the slope monitoring grid unit. The current point cloud data here refers to the point cloud after the initial registration mentioned above, where each point has planar coordinates, allowing it to be located to its corresponding slope monitoring grid unit. The stability confidence value of the slope monitoring grid unit to which the point belongs is used as the stability region weight for that point. The stability confidence value ranges from 0 to 1; a high value indicates that the point is located in a stable region and should be given a high weight in the registration; a low value indicates that the point is located in a potential deformation region or crack boundary and should be weakened in the registration.

[0054] A weighted registration objective function is constructed using the M-estimation loss function. This objective function is in the form of finding the rotation matrix and translation vector that minimizes the weighted point cloud distance error, where the error of each point cloud point is transformed using the Huber loss function. The Huber loss function is a piecewise function: when the registration residual is less than a preset threshold, squared loss is used, in which case the objective function is sensitive to changes in the residual, which is beneficial for accurate alignment; when the registration residual is greater than the preset threshold, linear loss is used, in which case the objective function is insensitive to changes in the residual, which can suppress the influence of outliers. The Huber threshold is set based on the expected registration error; in this embodiment, it is set to 0.02 meters, approximately twice the expected registration error. This loss function design allows the registration process to consider two aspects: for points with high weights in stable regions and small residuals (i.e., truly stable points that need alignment), squared loss ensures accurate alignment; for points with large residuals (which may be deformed regions or mismatched points), linear loss automatically reduces their contribution, preventing these points from affecting the solution of the overall registration parameters.

[0055] The process of solving the weighted registration objective function is as follows: First, calculate the weighted centroid, which is the weighted average of the coordinates of all point cloud points, with the weights being the weights of the stable regions. Then, construct the weighted covariance matrix, which reflects the spatial correlation between the current point cloud and the reference point cloud after weighting. Perform singular value decomposition on the weighted covariance matrix to obtain the left singular vector matrix, the singular value matrix, and the right singular vector matrix. Based on the decomposition results, calculate the rotation matrix as the product of the right singular vector matrix and the transpose of the left singular vector matrix, and calculate the translation vector as the weighted centroid of the reference point cloud minus the rotation matrix multiplied by the weighted centroid of the current point cloud. Use the calculated rotation matrix and translation vector as the update transformation parameters. This solution method is the standard solution to the weighted least squares problem, and the orthogonality of the rotation matrix is ​​guaranteed by singular value decomposition.

[0056] This invention utilizes the spatial weight information provided by the stability confidence map to distinguish between stable and deformable regions during the registration process, allowing stable regions to dominate the calculation of transformation parameters. By leveraging the robustness of M-estimation to the residual distribution, the registration process can automatically suppress the influence of mismatches and confidence misjudgments. In slope monitoring scenarios, slope deformation is localized, with stable regions typically occupying most of the slope area. These regions provide sufficient registration constraints, and assigning high weights allows stable regions to dominate the registration direction. While deformable regions exhibit displacement, their spatial extent is limited, and assigning low weights minimizes their impact on the overall registration. Vegetated areas and steep cliff edges, which are prone to mismatches, typically have larger registration residuals. The Huber loss function automatically reduces the weights of these points to prevent them from interfering with the registration results. This design addresses the problems of deformation regions pulling on registration and stable regions reducing alignment accuracy in existing methods, achieving robust optimization of registration parameter calculation.

[0057] Furthermore, this step forms a tight closed-loop feedback with the preceding steps. This step uses the current slope stability confidence map as the weighting basis, which integrates current deformation statistics and historical information. The updated transformation parameters output from this step are used to update the current point cloud data. The updated point cloud returns to the step of recalculating the deformation field, thereby updating the confidence map and forming an iterative optimization loop. In this step, the application of M-estimation and the confidence weights form a dual protection mechanism: the confidence weights distinguish stable and deformable regions spatially, while the M-estimation suppresses the influence of abnormal matching points based on the residual distribution. These two complement each other, making the registration process robust to various interference factors such as vegetation shading, point cloud noise, and initial registration errors. When the confidence weights are correct, high weights dominate the registration, and the M-estimation mainly suppresses residual noise. When there are errors in the confidence weights (such as misclassifying a deformable region as a stable region), although points in that region are assigned high weights, their registration residuals will be large. The linear loss of the M-estimation can automatically reduce its contribution, preventing the true deformation from being forcibly registered out. This dual protection mechanism ensures that reliable registration results can still be obtained even if there are some errors in the confidence map.

[0058] In one embodiment of the present invention, the step of repeating the weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets a preset convergence condition to obtain a converged slope deformation field includes: S71, in the preset first few iterations, a preset coarse-scale slope monitoring grid is used to calculate and update the slope stability confidence map, wherein the grid side length of the coarse-scale slope monitoring grid is the first side length; S72, During the iteration process, for slope monitoring grid units in the preset coarse-scale slope monitoring grid that meet the local refinement conditions, a preset fine-scale slope monitoring grid is used for local refinement calculation. The grid side length of the fine-scale slope monitoring grid is the second side length, which is less than the first side length. The local refinement conditions include: the slope deformation field gradient within the slope monitoring grid unit exceeds a preset slope deformation gradient threshold, or the stability confidence value of the slope monitoring grid unit is lower than a preset confidence threshold, which respectively correspond to the local deformation abrupt change region and the potential instability region of the slope. S73. After each iteration, calculate the change between the current slope stability confidence map and the previous slope stability confidence map; when the change is less than the preset convergence threshold, determine that the slope stability confidence map meets the convergence condition and stop the iteration.

[0059] During the iterative registration process, the slope stability confidence map needs to be repeatedly calculated and updated. Using a single-scale grid throughout the entire iteration presents a dilemma: a larger-scale grid (e.g., 0.5 meters) ensures sufficient point cloud points for statistical analysis within each grid cell, resulting in stable statistical results and strong noise resistance. However, a larger-scale grid struggles to capture local deformation details such as crack boundaries and the leading edge of slip bodies. These details typically range from 0.1 to 0.3 meters in width, and a larger-scale grid smooths them out. A smaller-scale grid (e.g., 0.1 meters) preserves local deformation details, but the number of point cloud points per grid cell is smaller, making the statistical results more susceptible to noise interference, and significantly increasing computational load. Furthermore, in the early stages of iteration, the confidence map is inaccurate, with large registration errors. At this point, pursuing local details is less meaningful; ensuring the correctness of the overall convergence direction is more important. In the later stages of iteration, the confidence map stabilizes, and the registration error decreases. At this stage, capturing local deformation details is necessary to support subsequent disease identification. Therefore, a mechanism is needed that can adaptively adjust the mesh scale based on the iteration process and local features.

[0060] To address the aforementioned issues, as described in steps S71-S73 above, this invention employs a larger-scale grid for confidence map calculation in the initial stage of iteration to ensure the stability of the overall convergence direction. During the iteration process, for grid cells that meet the local refinement conditions, the invention automatically switches to a smaller-scale grid for local calculation to capture local deformation details. The change in the confidence map is used to determine whether the iteration has converged.

[0061] A multi-scale slope monitoring grid is used to calculate and update the slope stability confidence map. In the initial iteration phase, a first-scale slope monitoring grid is used to ensure the correctness of the overall convergence direction. The initial iteration phase refers to the first few iterations; in this embodiment, the first three iterations use the first-scale grid. The side length of the first-scale grid needs to ensure that each grid cell contains enough point cloud points for statistical analysis, while also reflecting the overall deformation trend of the slope. In this embodiment, the side length of the first-scale grid is 0.5 meters, which matches the spatial scale of the slope deformation characteristics. Each grid cell contains an average of approximately 5–12 point cloud points, meeting the statistical requirements.

[0062] During the iteration process, for slope monitoring grid cells that meet the local refinement conditions, the system automatically switches to a second-scale slope monitoring grid for local calculation. The side length of the second-scale grid is shorter than that of the first-scale grid, and the second scale is used to capture local deformation details. The local refinement conditions are set based on the following considerations: areas where the slope deformation field gradient exceeds a preset gradient threshold are often abrupt deformation areas such as crack boundaries and the leading edge of slip bodies, requiring finer grids to characterize deformation details; areas where the stability confidence value is lower than a preset confidence threshold are often potential instability areas, and their boundary locations are crucial for disease identification, also requiring finer grids. In this embodiment, the side length of the second-scale grid is 0.1 meters, consistent with the grid side length of the aforementioned projection raster method, facilitating data reuse. The preset gradient threshold for the local refinement conditions is 0.01 meters per meter, which is approximately half the typical gradient value of crack boundaries, effectively identifying crack boundary areas; the preset confidence threshold is 0.4, and areas below this value are considered potential instability or deformation areas. After each iteration, the change between the current slope stability confidence map and the previous slope stability confidence map is calculated. This change is defined as the maximum change in confidence values ​​of all grid cells, i.e., the maximum absolute difference. This statistic reflects the overall convergence of the confidence map. When the confidence map no longer changes significantly, the registration process is considered stable. When this change is less than a preset convergence threshold, the slope stability confidence map is deemed to meet the convergence condition, and the iteration stops. The preset convergence threshold is set based on the minimum meaningful change in confidence values; in this embodiment, it is set to 0.01. This value is less than the change in confidence values ​​from 0.94 to 0.93, ensuring the sensitivity of the convergence judgment.

[0063] This invention leverages the multi-scale spatial characteristics of slope deformation. In the early stages of iteration, a large-scale grid is used to ensure statistical stability and overall convergence direction. In the later stages, a smaller-scale grid is used to capture details in local feature regions. The change in confidence level is used as a convergence criterion to automatically terminate the iteration process. In slope monitoring scenarios, the overall slope deformation trend is usually dominated by stable regions, and a large-scale grid can effectively reflect this trend. Local deformation features such as cracks and the leading edge of slip bodies have smaller spatial scales and need to be characterized by a fine-scale grid in the later stages of iteration. As iteration progresses, the confidence level gradually stabilizes, and its change can serve as a quantitative indicator of convergence. This scheme solves the problem of balancing statistical stability and detail preservation with a single-scale grid, achieving a rational spatial allocation of computational resources.

[0064] In one embodiment of the present invention, the step of extracting early disease characteristics based on the convergent slope deformation field includes: S81, Calculate the gradient of the convergent slope deformation field to obtain the slope deformation gradient field; S82, based on the slope deformation gradient field, extract the region where the slope deformation gradient exceeds a preset gradient threshold, and use it as a candidate region for microcracks. S83, extract the geometric features of the microcrack candidate region, the geometric features include the principal orientation angle of the centerline of the microcrack candidate region, the standard deviation of the width along the centerline, and the spatial continuity of the centerline; S84, classify the microcrack candidate regions according to the geometric features: if the angle between the main direction angle and the slope direction is less than a preset angle threshold and the centerline length is greater than a first length threshold, then it is determined to be an artificial structure such as a slope drainage ditch or a skeleton joint; if the standard deviation of the width along the centerline is less than a preset width standard deviation threshold and the centerline length is greater than a second length threshold, then it is determined to be an artificial structure; the remaining microcrack candidate regions are determined to be real microcracks. S85, extract the continuous region in the convergent slope deformation field where the absolute value of slope settlement displacement exceeds the preset deformation threshold, and use it as a micro-uplift candidate region or a micro-settlement candidate region. S86, extract the transition zone from the low confidence region to the high confidence region in the slope stability confidence map, as the leading edge boundary of the slip body, and extract the leading edge range of the slip body based on the area enclosed by the leading edge boundary of the slip body.

[0065] S87, after the step of extracting early disease characteristics based on the convergent slope deformation field, the method further includes: displaying the identified disease areas; and receiving user feedback on the removal of suspected false alarm areas.

[0066] After completing iterative registration and obtaining the converged slope deformation field, it is necessary to extract early-stage defect features that can be used in engineering from this deformation field. However, directly extracting defect areas based on deformation thresholds has significant false alarm problems: the slope surface contains a large number of artificial structures, such as longitudinal drainage ditches, transverse intercepting ditches, and arched skeleton joints. These structures are geometrically similar to cracks and all exhibit local abrupt changes in deformation gradients, making them difficult to distinguish based solely on deformation magnitude and gradient thresholds. Furthermore, the identification of the leading edge of the slip body requires comprehensive consideration of the spatial distribution characteristics of the deformation field; relying solely on deformation cannot accurately define the boundary of the slip body. Therefore, a defect identification method that can comprehensively utilize the geometric features of the deformation field and the spatial distribution of the confidence map is needed. As described in steps S81-S87 above, this invention locates candidate crack regions by calculating the deformation field gradient, extracts the geometric features of the candidate regions (principal orientation angle, width standard deviation, spatial continuity), and classifies and eliminates cracks based on the differences in geometric features between cracks and artificial structures; identifies the leading edge boundary of the slip body through the spatial distribution of the confidence map; and extracts uplift and settlement areas through deformation thresholds.

[0067] The gradient of the convergent slope deformation field is calculated to obtain the slope deformation gradient field. The gradient calculation adopts the central difference method, that is, the partial derivatives of each grid cell in the x and y directions are obtained by the difference in deformation between adjacent grid cells. The deformation gradient field reflects the rate of change of slope settlement displacement in the horizontal direction, and its amplitude increases significantly at the crack boundary, which can be used as a basis for locating crack candidate areas.

[0068] Regions where the slope deformation gradient exceeds a preset gradient threshold are extracted from the slope deformation gradient field and designated as candidate regions for microcracks. The preset gradient threshold is set based on the typical value of the deformation gradient at the crack boundary; in this embodiment, it is set to 0.1 meters per meter. This value effectively identifies crack boundaries while avoiding misjudging normal terrain undulations as cracks.

[0069] Geometric features of microcrack candidate regions are extracted, including the principal orientation angle of the centerline, the standard deviation of the width along the centerline, and the spatial continuity of the centerline. The centerline is extracted using a morphological skeletonization algorithm, extracting a single-pixel-width centerline from the binary image of the crack candidate region. The principal orientation angle is calculated using principal component analysis, taking the direction of the first principal component of the coordinates of all points on the centerline, reflecting the overall extension direction of the crack. The standard deviation of the width is obtained by sampling the boundary points on both sides of the crack along the normal direction of the centerline, calculating the crack width at each sampling point, and then calculating the standard deviation, reflecting the uniformity of the crack width. Spatial continuity is characterized by whether the crack centerline is continuous and whether there are bifurcations or discontinuities.

[0070] Microcrack candidate regions are classified based on geometric characteristics. The classification rules are based on the following engineering experience: artificial structures such as drainage ditches and skeleton joints usually have a regular orientation, parallel or perpendicular to the slope direction, and are relatively long (usually exceeding 10 meters); the width of artificial structures is relatively uniform, and the standard deviation of the width along the centerline is small (usually less than 2 mm); artificial structures are usually continuous over long distances without obvious branching or discontinuity. Real cracks, on the other hand, have random orientations and no obvious regularity with the slope direction; their width varies greatly, and the standard deviation of the width along the centerline is large; they may be discontinuous or branched in space, and their length is usually short. The specific judgment rules are as follows: if the angle between the main direction angle and the slope direction is less than a preset angle threshold and the centerline length is greater than a first length threshold, it is judged as an artificial structure; if the standard deviation of the width along the centerline is less than a preset width standard deviation threshold and the centerline length is greater than a second length threshold, it is judged as an artificial structure; the remaining microcrack candidate regions are judged as real microcracks. In this embodiment, the preset angle threshold is 10 degrees, the first length threshold is 10 meters, the preset width standard deviation threshold is 0.01 meters, and the second length threshold is 5 meters. These thresholds are set based on the typical dimensions of drainage ditches and skeleton joints in slope engineering.

[0071] Continuous regions in the convergent slope deformation field where the absolute value of slope settlement displacement exceeds a preset deformation threshold are extracted as candidate regions for micro-uplift or micro-settlement. The preset deformation threshold is set to 0.01 meters, a value based on the measurement accuracy of the lidar and the engineering definition of early-stage defects. Deformation smaller than this value is considered measurement noise or elastic deformation and does not constitute defects.

[0072] A transition zone is extracted from the slope stability confidence map, where stability confidence values ​​shift from low to high confidence areas. This transition zone serves as the leading edge boundary of the slip body, and the extent of the slip body's leading edge is extracted based on the area enclosed by this boundary. The transition zone is identified as follows: on the confidence map, areas with confidence values ​​below 0.4 are marked as potentially unstable areas, and areas with confidence values ​​above 0.8 are marked as stable areas. The boundary between these two areas is considered the transition zone. The physical significance of this transition zone is that the leading edge of the slip body is the boundary between potentially unstable and stable areas, manifested as a spatial abrupt change in confidence values ​​on the confidence map. The width of the transition zone is typically one grid cell, and its spatial location is used as the leading edge boundary of the slip body.

[0073] This invention classifies and identifies cracks and artificial structures based on their geometric features. It uses deformation thresholds to extract uplift and settlement areas and the spatial distribution of confidence maps to identify the leading edge boundary of slip bodies. In slope monitoring scenarios, artificial structures such as drainage ditches and framework joints are widely present in slope engineering. These structures are difficult to distinguish from real cracks in terms of deformation gradient, but they exhibit significantly different geometric features in terms of direction regularity, width uniformity, and spatial continuity. These features can be used to effectively differentiate them. The leading edge of a slip body is the boundary between potentially unstable and stable regions. The spatial distribution of the confidence map precisely reflects the stability level of each region, and the spatial abrupt changes in confidence values ​​correspond to the physical location of the leading edge of the slip body. This design solves the problems of high false alarm rates for artificial structures and difficulty in quantitatively extracting slip body boundaries in existing methods, achieving automatic identification of various types of defects.

[0074] like Figure 2 As shown, the present invention also provides a system for identifying early-stage damage characteristics of highway slope settlement and displacement, comprising: The data acquisition module acquires the baseline point cloud data and the current point cloud data of the slope monitoring area; The initial registration module performs initial registration based on the reference point cloud data and the current point cloud data to obtain initial transformation parameters, and transforms the current point cloud data according to the initial transformation parameters to obtain the initial registered point cloud. The deformation calculation module calculates the initial slope deformation field based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. The confidence building module constructs a slope stability confidence map based on the statistical characteristics of the initial slope deformation field. Each slope monitoring grid cell in the slope stability confidence map corresponds to a stability confidence value, which is used to characterize the stability region weight that the slope monitoring grid cell should be assigned during the registration process. The weighted registration module determines the stable region weight of each point in the current point cloud data based on the slope stability confidence map, performs weighted registration between the current point cloud data and the reference point cloud data based on the stable region weight, obtains update transformation parameters, and updates the current point cloud data based on the update transformation parameters to obtain the updated registered point cloud. The update module recalculates the slope deformation field based on the updated registration point cloud and the reference point cloud data, and updates the slope stability confidence map based on the recalculated slope deformation field. The iterative convergence module repeats the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence conditions, thus obtaining the converged slope deformation field. The disease extraction module extracts early disease characteristics based on the convergent slope deformation field. The early disease characteristics include at least one of microcracks, micro-bulges, micro-settlement, and the leading edge of the slip body.

[0075] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for identifying early-stage features of settlement and displacement of highway slopes.

[0076] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for identifying early-stage features of settlement and displacement of highway slopes.

[0077] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0078] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for identifying early-stage characteristics of settlement and displacement defects on highway slopes, characterized in that, include: Acquire baseline point cloud data and current point cloud data for the slope monitoring area; The initial registration point cloud is obtained based on the reference point cloud data and the current point cloud data. The initial slope deformation field is calculated based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. Based on the statistical characteristics of the initial slope deformation field, a slope stability confidence map containing several slope monitoring grid units is constructed. Each slope monitoring grid unit corresponds to a stability confidence value, which is used to characterize the stability region weight that the slope monitoring grid unit should be assigned during the registration process. The stability region weights of each point cloud point in the current point cloud data are determined based on the slope stability confidence map. The current point cloud data and the reference point cloud data are then weighted and registered according to the stability region weights to obtain update transformation parameters. The current point cloud data is then updated according to the update transformation parameters to obtain the updated registered point cloud. The slope deformation field is recalculated based on the updated registration point cloud and the reference point cloud data, and the slope stability confidence map is updated based on the recalculated slope deformation field. Repeat the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence condition to obtain the converged slope deformation field. Early disease characteristics were extracted based on the convergent slope deformation field.

2. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 1, characterized in that, The step of calculating the initial slope deformation field based on the initial registered point cloud and the reference point cloud data includes: The slope monitoring area is projected onto a horizontal plane and divided into regular grids to obtain several slope grid units; Based on the ground points in the benchmark cloud data, a digital elevation model of each slope grid unit is constructed to obtain the three-dimensional coordinates of the center point of each slope grid unit. Obtain the elevation gradient of adjacent slope grid cells, and pre-calculate the normal vector of each slope grid cell based on the elevation gradient; The three-dimensional coordinates of each point in the initial registration point cloud are obtained, and each point in the initial registration point cloud is located to the corresponding slope grid cell. Based on the three-dimensional coordinates of the point in the initial registration point cloud and the three-dimensional coordinates and normal vector of the center point of the corresponding slope grid cell, the slope settlement displacement of each point in the initial registration point cloud to the plane corresponding to the slope grid cell is calculated to obtain the initial slope deformation field.

3. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 2, characterized in that, The step of constructing a slope stability confidence map containing several slope monitoring grid cells based on the statistical characteristics of the initial slope deformation field includes: The slope monitoring area is divided into several regular slope monitoring grid units; For each slope monitoring grid cell, the slope settlement displacement of all initial registration point cloud points falling within the slope monitoring grid cell is counted to obtain the set of slope settlement displacement. Calculate the median and median absolute deviation of the set of slope settlement displacements, where the median absolute deviation is the median of the absolute deviations of each slope settlement displacement from the median. Obtain the normal vector of the slope grid cell corresponding to each slope monitoring grid cell, and calculate the slope inclination angle based on the normal vector; The initial stability confidence value of the slope monitoring grid cell is calculated based on the median, the median absolute deviation, and the slope inclination angle: If the median absolute deviation is less than a preset sensor noise threshold, then the initial stability confidence value is calculated based on the absolute value of the median. If the median absolute deviation is greater than or equal to the preset sensor noise threshold, the initial stability confidence value is set to the preset low stability confidence value, indicating that the deformation in the region is inconsistent or there is measurement noise. The initial stability confidence values ​​of each slope monitoring grid unit are combined to obtain the initial slope stability confidence map.

4. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 3, characterized in that, The step of combining the initial stability confidence values ​​of each slope monitoring grid unit to obtain the initial slope stability confidence map includes: Obtain historical slope stability confidence maps, which are slope stability confidence maps after convergence in the previous monitoring period; Determine the current monitoring period number. If the current monitoring period is not the first monitoring period, perform a weighted fusion of the initial slope stability confidence map and the historical slope stability confidence map to obtain the current slope stability confidence map. If the current monitoring period is the first monitoring period, obtain the slope stability confidence map based on conservative initial values.

5. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 4, characterized in that, The steps of determining the stable region weights of each point in the current point cloud data based on the slope stability confidence map, and performing weighted registration of the current point cloud data and the reference point cloud data based on the stable region weights to obtain the updated transformation parameters include: Based on the slope stability confidence map, establish the mapping relationship between each point cloud point in the current point cloud data and the slope monitoring grid unit, and use the stability confidence value of the slope monitoring grid unit to which each point cloud point belongs in the current point cloud data as the stability region weight of that point cloud point. The weighted registration objective function is constructed using the M-estimation loss function; Solve the weighted registration objective function to obtain the updated rotation matrix and the updated translation vector, and use the updated rotation matrix and the updated translation vector as the updated transformation parameters.

6. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 5, characterized in that, The steps of repeating the weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence condition to obtain the converged slope deformation field include: In the first few iterations, a pre-set coarse-scale slope monitoring grid is used to calculate and update the slope stability confidence map. The grid side length of the coarse-scale slope monitoring grid is the first side length. During the iteration process, for slope monitoring grid cells in the preset coarse-scale slope monitoring grid that meet the local refinement conditions, a preset fine-scale slope monitoring grid is used for local densification calculation. The grid side length of the fine-scale slope monitoring grid is the second side length, which is smaller than the first side length. The local refinement conditions include: the slope deformation field gradient within the slope monitoring grid cell exceeds a preset slope deformation gradient threshold, or the stability confidence value of the slope monitoring grid cell is lower than a preset confidence threshold, which correspond to the local deformation abrupt change region and the potential instability region of the slope, respectively. After each iteration, the change between the current slope stability confidence map and the previous slope stability confidence map is calculated; when the change is less than a preset convergence threshold, the slope stability confidence map is determined to meet the convergence condition, and the iteration stops.

7. The method for identifying early-stage damage characteristics of highway slope settlement and displacement according to claim 6, characterized in that, The step of extracting early disease characteristics based on the convergent slope deformation field includes: Calculate the gradient of the convergent slope deformation field to obtain the slope deformation gradient field; Based on the slope deformation gradient field, regions where the slope deformation gradient exceeds a preset gradient threshold are extracted as candidate regions for microcracks. Extract the geometric features of the microcrack candidate region, including the principal orientation angle of the centerline of the microcrack candidate region, the standard deviation of the width along the centerline, and the spatial continuity of the centerline; The candidate regions for microcracks are classified based on the geometric features described above. If at least one of the following conditions is met, it is determined to be an artificial structure: if the angle between the main direction angle and the slope direction is less than a preset angle threshold and the centerline length is greater than a first length threshold; if the standard deviation of the width along the centerline is less than a preset width standard deviation threshold and the centerline length is greater than a second length threshold. Otherwise, it is determined to be a real micro-crack; Extract the continuous regions in the convergent slope deformation field where the absolute value of slope settlement displacement exceeds a preset deformation threshold, and use them as candidate regions for micro-uplift or micro-settlement. The transition zone from the low confidence region to the high confidence region in the slope stability confidence map is extracted as the leading edge boundary of the slip body, and the leading edge range of the slip body is extracted based on the area enclosed by the leading edge boundary of the slip body.

8. A system for identifying early-stage damage characteristics of highway slope settlement and displacement, characterized in that, include: The data acquisition module acquires the baseline point cloud data and the current point cloud data of the slope monitoring area; The initial registration module obtains the initial registration point cloud based on the reference point cloud data and the current point cloud data; The deformation calculation module calculates the initial slope deformation field based on the initial registration point cloud and the reference point cloud data. The initial slope deformation field includes the slope settlement displacement at each spatial location within the monitoring area. The confidence building module constructs a slope stability confidence map containing several slope monitoring grid units based on the statistical characteristics of the initial slope deformation field. Each slope monitoring grid unit corresponds to a stability confidence value, which is used to characterize the stability region weight that the slope monitoring grid unit should be assigned during the registration process. The weighted registration module determines the stable region weight of each point in the current point cloud data based on the slope stability confidence map, performs weighted registration between the current point cloud data and the reference point cloud data based on the stable region weight, obtains update transformation parameters, and updates the current point cloud data based on the update transformation parameters to obtain the updated registered point cloud. The update module recalculates the slope deformation field based on the updated registration point cloud and the reference point cloud data, and updates the slope stability confidence map based on the recalculated slope deformation field. The iterative convergence module repeats the steps of weighted registration, recalculating the slope deformation field, and updating the slope stability confidence map until the slope stability confidence map meets the preset convergence conditions, thus obtaining the converged slope deformation field. The disease extraction module extracts early disease characteristics based on the convergent slope deformation field. The early disease characteristics include at least one of microcracks, micro-bulges, micro-settlement, and the leading edge of the slip body.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images