Landslide interface detection method and device

Abstract

The invention relates to a landslide interface detection method and device. The detection method comprises the steps that detection data collected by a muon detector and outer surface data, collected by a topographic surveying and mapping device, of a landslide body are obtained; and based on a compressed sensing iterative reconstruction algorithm, processing the detection data and the outer surface data, and determining a landslide interface. The detection data acquired by the muon detector and the outer surface data, acquired by the topographic surveying and mapping device, of the landslide mass are combined at the same time, so that the adverse effect of the topography on data acquisition is reduced, and the basic accuracy and comprehensiveness of data for constructing the three-dimensional interface of the landslide mass are improved; through a compressed sensing iterative reconstruction algorithm, high-dimensional information is restored by using a small amount of data, the reliability of constructing a three-dimensional interface of a landslide mass is improved, and the accuracy of a detection result is improved.

Description

Technical Field

[0001] This disclosure relates to the field of detection and geological disaster prevention technology, and in particular to a method and apparatus for detecting landslide interfaces. Background Technology

[0002] Landslides are a common and significant geological hazard in mountainous areas, and accurate identification of the interface is crucial for landslide prediction and mitigation. The interface is often located tens to hundreds of meters below the surface, manifesting as a density difference between loose deposits and intact bedrock. Therefore, specific landslide interface detection methods are needed to determine the interface. However, existing methods for detecting landslide interfaces require large sampling coverage to ensure high accuracy and are significantly affected by topography, resulting in inaccurate detection results. Summary of the Invention

[0003] To address the aforementioned technical problems, this disclosure provides a landslide interface detection method and apparatus, aiming to solve the problems of the prior art.

[0004] A first aspect of this disclosure provides a method for detecting landslide interfaces, the method comprising: Acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device; Based on the compressed sensing iterative reconstruction algorithm, the detection data and the outer surface data are processed to determine the landslide interface.

[0005] In some embodiments of this disclosure, the step of processing the probe data and the outer surface data based on the compressed sensing iterative reconstruction algorithm to determine the landslide interface includes: Based on the detection data and the outer surface data, an optimization objective function is constructed; The landslide interface is determined by minimizing the optimization objective function based on the iterative shrinkage threshold algorithm.

[0006] In some embodiments of this disclosure, the step of constructing an optimization objective function based on the probe data and the outer surface data includes: The outer surface data is discretized into a three-dimensional voxel grid based on a preset voxel size to determine the number of voxels; Based on the detection data, the number of paths is determined; Based on the number of paths and the number of voxels, determine the projection matrix and the observation vector; Based on the projection matrix and the observation vector, the voxel density vector is determined; An optimization objective function is constructed based on the voxel density vector, the projection matrix, and the observation vector.

[0007] In some embodiments of this disclosure, the step of minimizing the optimization objective function based on an iterative shrinkage threshold algorithm to determine the landslide interface includes: The density values ​​of each voxel are determined by minimizing the optimization objective function through gradient descent and iterative updates. The landslide interface is determined based on the three-dimensional Gaussian spline generation algorithm and the density values ​​of each voxel.

[0008] In some embodiments of this disclosure, determining the landslide interface based on the three-dimensional Gaussian spline generation algorithm and the density values ​​of each voxel includes: Based on the density values ​​of each voxel, the preliminary three-dimensional density distribution of the landslide body is determined; Based on the aforementioned three-dimensional Gaussian spline generation algorithm, the preliminary three-dimensional density distribution of the landslide body is completed and refined to determine the landslide interface.

[0009] In some embodiments of this disclosure, the step of completing and refining the preliminary three-dimensional density distribution of the landslide body based on the three-dimensional Gaussian spline generation algorithm to determine the landslide interface includes: By placing Gaussian elements into the preliminary three-dimensional density distribution of the landslide body, the density field of the superimposed Gaussian elements is determined. The loss function of the density field is convergent and optimized to determine a continuous Gaussian field. The loss function includes at least a projection consistency term and a regularization term. Based on the density gradient, the continuous Gaussian field is refined or homogenized to determine the landslide interface.

[0010] In some embodiments of this disclosure, placing Gaussian elements in the preliminary three-dimensional density distribution of the landslide body and determining the density field of the superimposed Gaussian elements includes: A first region and a second region are determined for the preliminary three-dimensional density distribution of the landslide body. The first region is configured as a region where the density change exceeds a first threshold, and the second region is a region where the density change is less than a second threshold. The second threshold is less than or equal to the first threshold. A first number of Gaussian elements are placed in the first region, and a second number of Gaussian elements are placed in the second region. The density field of the superimposed Gaussian elements is determined, wherein the first number is greater than the second number.

[0011] In some embodiments of this disclosure, the refinement or homogenization of the continuous Gaussian field based on the density gradient includes: The Gaussian elements are split in the region where the density gradient is greater than a first preset gradient threshold to refine the continuous Gaussian field; The Gaussian elements are averaged in the region where the density gradient is less than a second preset gradient threshold to homogenize the continuous Gaussian field.

[0012] In some embodiments of this disclosure, the convergence optimization of the loss function of the density field to determine the continuous Gaussian field includes: The theory for determining the density field is projected. Based on the aforementioned theory, the projection consistency term in the loss function of the density field is determined through projection and observation of the projection value. Based on the Gaussian elements of the density field, the regularization term in the loss function of the density field is determined.

[0013] In some embodiments of this disclosure, the method further includes: Based on the outer surface data of the landslide and the spatial location of the muon detector, a three-dimensional visible domain model is constructed. Based on the aforementioned three-dimensional visible domain model, the visible voxels are determined; The projection matrix is ​​determined based on the visible voxels.

[0014] In some embodiments of this disclosure, the method further includes: Based on the density field of the superimposed Gaussian elements, density texture features are determined, including the density gradient of each voxel, the directional second derivative, the local density variance, and the layered continuity index. Based on the density texture features, the weak interlayer or slip zone structure of the landslide body is determined; The landslide interface is determined based on the weak interlayers or slip zones of at least a portion of the landslide body.

[0015] In some embodiments of this disclosure, the method further includes: Obtain the credibility score for each voxel; Based on the credibility scores of each voxel, a credible preliminary three-dimensional density distribution is determined.

[0016] A second aspect of this disclosure provides a landslide interface detection device, the landslide interface detection device comprising: The acquisition module is configured to acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device; The processing module is configured to process the detection data and the outer surface data based on the compressed sensing iterative reconstruction algorithm to determine the landslide interface.

[0017] A third aspect of this disclosure provides an electronic device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the landslide interface detection method described above.

[0018] A fourth aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the landslide interface detection method described above.

[0019] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: By simultaneously combining the detection data collected by the muon detector and the external surface data of the landslide body collected by the topographic mapping device, the adverse effects of terrain on data collection are reduced, improving the accuracy and comprehensiveness of the data foundation for constructing the three-dimensional interface of the landslide body. Through the compressed sensing iterative reconstruction algorithm, high-dimensional information can be restored using a small amount of data, improving the reliability of constructing the three-dimensional interface of the landslide body and improving the accuracy of the detection results.

[0020] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this document. Attached Figure Description

[0021] The accompanying drawings, which form part of this document, are used to provide a further understanding of the document. The illustrative embodiments and descriptions herein are used to explain the document and do not constitute an undue limitation thereof. In the drawings: Figure 1 This disclosure illustrates a landslide interface detection method flow according to an exemplary embodiment. Figure 1 ; Figure 2 This disclosure illustrates a landslide interface detection method flow according to an exemplary embodiment. Figure 2 ; Figure 3 This disclosure illustrates a landslide interface detection method flow according to an exemplary embodiment. Figure 3 ; Figure 4 This disclosure illustrates a landslide interface detection method flow according to an exemplary embodiment. Figure 4 ; Figure 5 This disclosure illustrates a landslide interface detection method flow according to an exemplary embodiment. Figure 5 ; Figure 6 This is a block diagram of a landslide interface detection device illustrated in an exemplary embodiment of the present disclosure; Figure 7 This is a block diagram of an electronic device illustrated in an exemplary embodiment of the present disclosure. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this disclosure can be arbitrarily combined with each other.

[0023] Landslides are a common and significant geological hazard in mountainous areas, and accurate identification of the interface is crucial for landslide prediction and mitigation. The interface often lies tens to hundreds of meters below the surface, manifesting as a density difference between loose deposits and intact bedrock. Common landslide interface detection methods include borehole sampling, seismic reflection or resistivity methods, and surface deformation monitoring. While borehole sampling offers high accuracy, it is costly and limited by sampling density, only providing point-like coverage of the interface. Seismic reflection or resistivity methods are significantly affected by external factors such as topography, and the interface shape obtained through reflection or conductivity inversion is unstable, leading to inaccurate results. Surface deformation monitoring only reflects the surface state of the terrain and fails to reveal its internal structure. Therefore, existing interface detection methods are not yet perfect or accurate enough.

[0024] Based on this, this disclosure provides a method and apparatus for detecting landslide interfaces. By using muon imaging technology combined with compressed sensing iterative reconstruction algorithm, the geological data of the surface and interior of the landslide body can be determined non-invasively, thereby obtaining the landslide interface. This is beneficial for predicting and managing landslide bodies through a three-dimensional interface model.

[0025] Combination Figure 1 As shown, an exemplary embodiment of this disclosure provides a landslide interface detection method, including: S100: Acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device.

[0026] In this step, muon detectors are placed in boreholes upstream or in the middle of the landslide body. By closely monitoring the sliding surface, they acquire localized high-resolution monitoring results, thus collecting detection data. This eliminates the need for large-scale drilling, utilizing natural muons for detection. Topographic mapping devices, such as flat-panel detectors, are placed at the toe of the landslide body, in underground tunnels, or in existing channels to achieve overall coverage of the landslide's outer surface and obtain surface data. Combining borehole-mounted muon detectors with topographic mapping devices placed at the toe of the landslide body, in underground tunnels, or in existing channels enables both large-scale monitoring and the acquisition of localized details. The outer surface data accurately defines the boundaries of density distribution, and the broader coverage of muon detection data further improves the accuracy of the detection results.

[0027] Muons are incident from the atmosphere, penetrate the landslide body, and are captured by a muon detector to record their incident direction and penetration rate. All effective penetration trajectories of atmospheric muons passing through the landslide body from the sky and being detected by the detector form different paths. By statistically analyzing the penetration rate under different paths, projection data related to the density distribution can be obtained.

[0028] For example, when the muon detector is deployed at the foot of a slope or inside a tunnel, the path is mostly obliquely incident from above at θ≈30°~80°, penetrating the entire landslide body; when the muon detector is deployed inside a borehole, the path is mainly in a near-vertical or obliquely upward direction, traversing a local sliding surface area.

[0029] S200: Based on the compressed sensing iterative reconstruction algorithm, the detection data and outer surface data are processed to determine the landslide interface.

[0030] In this step, the probe data is used to characterize the internal geological data, and the external surface data is used to characterize the surface topography. Combining the two can better reflect the actual geological characteristics.

[0031] The principle of the compressed sensing iterative reconstruction algorithm is based on compressed sensing theory. Using sparse data, such as a small amount of detection data and external surface data, the density value of the landslide body can be determined. Then, by using the observation linear relationship between the muon projection data and the density value, the three-dimensional density distribution of the landslide body can be obtained. The landslide interface can be determined through the three-dimensional density distribution, which greatly improves the computational efficiency.

[0032] The landslide interface detection method provided in this embodiment combines detection data collected by a muon detector and external surface data of the landslide body collected by a topographic mapping device. This reduces the adverse impact of terrain on data collection, improves the accuracy and comprehensiveness of the data foundation for constructing the three-dimensional interface of the landslide body, and uses a compressed sensing iterative reconstruction algorithm to restore high-dimensional information with a small amount of data, thereby improving the reliability of constructing the three-dimensional interface of the landslide body and the accuracy of the detection results.

[0033] In the above embodiments, combined with Figure 2 As shown, the methods for determining the landslide boundary include: S210. Based on the detection data and the outer surface data, construct an optimization objective function.

[0034] In this step, based on the detection data and external surface data, the boundary information of the landslide body is determined. This boundary information includes at least the outer surface boundary and the lower boundary of the sliding surface, thus forming a landslide body model outline with a certain volume area. Then, the landslide body model outline is spatially discretized, and the divided landslide body model outline is placed in a coordinate system, with the center of the muon detector as the origin, the vertical upward as the Z-axis, and the muon detector plane as the XY-axis plane, to determine the position coordinates of each voxel.

[0035] S220. Minimize the objective function based on the iterative shrinkage threshold algorithm to determine the density value of each voxel.

[0036] For example, the objective function is a function containing voxel density vectors, and an iterative shrinkage thresholding algorithm, such as FISTA minimization, is used to optimize the objective function and solve for the optimal density vector x. The objective function can be defined as the density values ​​of each voxel. Optionally, the objective function may include at least an observation consistency optimization term, such as the sum of squared errors between theoretical and measured projections, to ensure that the reconstructed density field can reproduce the true muon penetration pattern. The objective function may also include a sparse prior term, such as the L1 norm of the density vector or its total variational TV variant, to effectively suppress redundant information in homogeneous regions and retain the isodense abrupt change regions of the sliding surface, providing a clear foundation for interface identification. The objective function may also include a regularization term, such as the squared L2 norm of the density vector, to avoid high model complexity, ensure the smoothness of the density field, and prevent model overfitting.

[0037] S230. Based on the density values ​​of each voxel, the landslide interface is determined.

[0038] In this step, all the optimal density vectors converged in the previous step are mapped to a three-dimensional voxel density distribution, so that the landslide interface can be initially determined from the three-dimensional voxel density distribution.

[0039] In this embodiment, by introducing an objective function to optimize the collected data using an iterative shrinkage threshold algorithm, the density values ​​of each voxel are determined, the convergence speed is improved, and the initially determined density field distribution information can be quickly output. This distribution information can support subsequent dynamic monitoring of landslide interfaces and provide a basis for rapid detection of interfaces.

[0040] Combination Figure 3 As shown, in one embodiment, the method for constructing the optimization objective function includes: S211. Discretize the outer surface data into a three-dimensional voxel grid based on the preset voxel size, and determine the number of voxels.

[0041] For example, the preset voxel size is selected based on the resolution and statistical accuracy of the muon detector, such as 1-5 meters. Then, the landslide model outline is spatially discretized using voxels, for example, by dividing the volume region into a regular grid of stacked cubes, with each cube defined as a voxel. Assume each voxel k has a density value ρ. k After discretization, the density distribution of the entire landslide mass can be represented as a three-dimensional vector:

[0042] Therefore, the number of three-dimensional voxel grids, i.e., the number of voxels M, can be determined statistically.

[0043] S212. Determine the number of paths based on the detection data.

[0044] In this step, it is assumed that the path is... This means that all paths can be represented as p1, p2, ..., p n The total number of incident paths, or path number n, can be determined by analyzing the probe data.

[0045] S213. Based on the number of paths and the number of voxels, determine the projection matrix and the observation vector.

[0046] In this step, the dimension of the projection matrix = the number of paths n × the number of voxels M.

[0047] Definition: Projection matrix A p,k For path In voxels The length of the passage within, in meters; x k Let k be the density vector of voxel k, in g / cm³. b p For path The observation vector, each element b p Approximately the cumulative mass thickness of path p, in g / cm²:

[0048] In the formula, Indicates along the path The measured penetration probability is the ratio of the residual flux after the muon penetrates the landslide mass to the original flux. , For along the path The actual number of muons detected. For muon flux counts under unobstructed conditions (obtained through open-sky measurements or Monte Carlo simulations), this ratio reflects the landslide mass along the path. The absorption or scattering effect of matter is a direct observation of density inversion; κ is the muon attenuation coefficient obtained by Monte Carlo simulation or empirical calibration, with units of cm² / g.

[0049] The expression in this step is a quantity calculated from experimental observation data, reflecting the cumulative mass thickness along the path, and is an expression in the observation space.

[0050] Optionally, a three-dimensional visible domain model is constructed based on the outer surface data of the landslide body and the spatial location of the muon detector; the visible voxels are determined based on the three-dimensional visible domain model; and the projection matrix is ​​determined based on the visible voxels.

[0051] For example, by using the outer surface data of the landslide body and the spatial location of the muon detector, the relative position of the muon detector, the slope geometry, and the distribution of terrain occlusion can be determined, i.e., the relationship between the terrain boundary and the observation geometry can be determined. Based on these determined boundaries and relationships, a visible cone region originating from the detector's location can be constructed to simulate the actual penetrable spatial range of muons, thereby determining whether each voxel is visible. A voxel is considered visible when it is traversed by at least one muon path; it is considered invisible when it is completely occluded by the terrain and covered by no muon path. When determining the projection matrix based on voxels, invisible voxels are discarded, or the projection matrix elements corresponding to invisible voxels are set to zero, resulting in a corrected projection matrix. This method, in the inversion process from the muon detector to the terrain, can reduce the distribution of errors in severely occluded and impenetrable voxel regions, thereby increasing the condition number of the inversion and reducing artifacts. This further improves the density reconstruction accuracy in complex terrain landslide scenarios, making the detection of landslide interfaces more accurate and stable.

[0052] S214. Determine the voxel density vector based on the projection matrix and the observation vector.

[0053] Based on the observation equation Ax=b, after substitution, the linear expression of the relationship between the observed transmittance and the density integral along the path for each muon path in the model space is determined by inversion:

[0054] In the formula, For path Passing through voxels The length, in cm; , voxels Density, in g / cm³.

[0055] Therefore, based on the determination in step S213 and the determination in step S214 With the density distribution to be determined The linear constraint relationship between them allows us to calculate the density vector of each voxel. .

[0056] S215. Based on the voxel density vector, projection matrix, and observation vector, construct an optimization objective function.

[0057] For example, the following optimization objective function can be constructed:

[0058] In the formula, Let be the observation consistency term, representing the theoretical projection. Compared with the measured projection The sum of squared errors is used to ensure that the reconstruction results match the observation data, and the measured projection... Observations from each path composition, ,matrix The length of each path through each voxel Composition, path-voxel projection matrix ; The sparse prior is used to constrain the sparsity of the density field in the gradient domain. It can suppress noise and highlight density abrupt changes (such as sliding surfaces or void boundaries). The optimization objective of the solution is... ; The weighting parameter controls the fitting weights of the observation consistency term and the smoothing weights of the sparse prior term.

[0059] In this embodiment, by specifically defining the calculation formula of the objective function, the method of obtaining the density values ​​of each voxel through observation consistency and sparsity optimization is specifically defined, thereby obtaining a more accurate three-dimensional interface.

[0060] Optionally, based on the above objective function, the optimal density vector x is determined using the following formula. :

[0061] The specific calculation process includes: First, set the initial density field x0=0; Furthermore, the gradient is calculated using the following formula:

[0062] In the formula, is the density distribution estimate at the t-th iteration of the iterative optimization; is the inversion variable at the current iteration time, used to progressively approximate the optimal solution. .Right now:

[0063] in, This is the step size (or learning rate).

[0064] In muon imaging inversion Corresponding to the The subsurface density field estimate obtained from the inversion at each data update or acquisition cycle. It can be the first... The intermediate density field obtained from the second iteration update, or the third iteration update... The density reconstruction result of accumulated data over a time period. Therefore, if the algorithm iterates on the same batch of data, then... This indicates the algorithm's iteration step; if the algorithm updates on the time series data, then... Indicates the time step (collection cycle).

[0065] Then, the Iterative Shrinkage Thresholding (ISTA) algorithm is used to solve the problem, as shown in the following formula:

[0066] In the formula, For the first Density estimation vector at the next iteration The iteration step size, This represents the current gradient direction. The first equation uses gradient descent to decrease the objective function, while the second equation... The soft threshold function is used to implement sparsity constraints, and it is defined as follows: Threshold The sparsity intensity is controlled by the equilibrium parameters. With step size The decision is made jointly. Through stepwise iteration, a smooth and sparse density field estimate can be obtained under noisy conditions.

[0067] Then, a convergence check is performed when... The iteration stops when the time is right, where, This is the convergence threshold, used to determine whether the iteration has converged. It is determined when two consecutive density estimation results... and The difference is less than When the inversion results are considered stable, the iteration terminates. The value is set according to the accuracy of the observation data and the inversion resolution, with a typical range of . ~ .

[0068] Next, output the density values ​​of each voxel. This maps the density values ​​of each voxel to a three-dimensional density-volume image, which is beneficial for visualizing the internal structure of the landslide body and obtaining three-dimensional interfaces. Used to represent the stable result when finally converging, that is, the optimal solution or the density vector obtained by inversion. and The relationship is:

[0069] in, This is the stopping time of the iteration (the number of iterations that satisfy the convergence condition). Because... By mapping the density estimates of each voxel in the voxel grid to a three-dimensional voxel coordinate system according to their spatial location, a three-dimensional density volume image of the landslide body can be generated. For regions with abrupt changes in density gradient, the sliding surface or void boundary can be further extracted to visualize the internal structure of the landslide body.

[0070] Therefore, when the convergence condition is met At that time, the final voxel density vector is output. . This represents the stable result after the algorithm converges, indicating the inversion density values ​​of each voxel. By mapping spatial coordinates to a three-dimensional voxel grid, a three-dimensional density volume image of the landslide body can be generated, which can be used to reveal the spatial distribution of sliding surfaces, cavities, or high-density anomaly zones.

[0071] In this embodiment, FISTA is used to gradually approximate the optimal density distribution through gradient descent and soft threshold update. The solution result is a preliminary reconstruction map of the three-dimensional density field of the landslide body, which can be directly used for the detection of three-dimensional interfaces.

[0072] Optionally, over time Over time, the number of Muzi events recorded... As the noise level increases, statistical noise decreases, the signal-to-noise ratio improves, and thus a higher resolution density distribution can be reconstructed. When unchanged, It accumulates and updates over time. The estimation gradually converges. To achieve dynamic evolution monitoring of the internal structure of the landslide, the system continuously collects muon transmission events over time. Assuming that at the acquisition time... The observation vector at that time is:

[0073] in, As the collection time increases The extension, measured muon flux The cumulative increase reduces the variance of the probability estimate, thereby improving the inversion accuracy.

[0074] Therefore, the dynamic reconstruction formula based on time series can be written as:

[0075] In the formula, Based on new data The reconstructed density field, For time-series fusion weights.

[0076] Through this iterative update, the density distribution can be gradually refined over time, forming a dynamic density evolution monitoring system for landslides.

[0077] In another embodiment, the landslide interface detection method further includes: determining the landslide interface based on a three-dimensional Gaussian spline generation algorithm.

[0078] For example, processing the detection data and outer surface data based on the compressed sensing iterative reconstruction algorithm yields a landslide body three-dimensional interface with relatively coarse resolution, but it can improve the speed of interface acquisition. However, by processing with the three-dimensional Gaussian spline generation algorithm, the three-dimensional interface of the landslide body can be further refined, thereby improving the detection accuracy of the interface.

[0079] In the landslide interface detection method provided in this embodiment, a three-dimensional Gaussian spline generation algorithm is introduced, thereby combining the three-dimensional Gaussian spline generation algorithm with the compressed sensing iterative reconstruction algorithm. The two-stage algorithm ensures that both the overall structure and detailed boundaries can be accurately recovered, which is beneficial to improve the detection accuracy of the three-dimensional interface under the condition of limited model computation, and breaks through the limitation of traditional methods that can only perform qualitative analysis or point monitoring.

[0080] For example, combined Figure 4 As shown, methods for determining landslide boundaries also include: S231. Determine the preliminary three-dimensional density distribution of the landslide body based on the density values ​​of each voxel.

[0081] For example, since each voxel density value has three-dimensional position coordinates, according to the mapping relationship between voxel number and coordinates, the corresponding voxel density values ​​are filled into the voxel grid to form a three-dimensional voxel density distribution corresponding to coordinates and density. After reliability processing of outliers in the three-dimensional voxel density distribution, the preliminary three-dimensional density distribution of the landslide body is finally obtained. For example, outlier voxel values ​​that deviate significantly from the regional density range are removed and replaced with the average density of the surrounding voxels.

[0082] Optionally, the credibility score of each voxel is obtained; based on the credibility score of each voxel, a credible preliminary three-dimensional density distribution is determined.

[0083] Because the reliability of different voxels can vary significantly—for example, voxels covered by multiple muon paths have high inversion accuracy, while voxels with less path coverage are more susceptible to noise—using voxels with high noise levels to determine the 3D density distribution may lead to misjudgments in subsequent boundary determinations in low-reliability voxel regions. Therefore, assigning a reliability score to each voxel can improve the accuracy of boundary detection.

[0084] The credibility score for each voxel is determined based on a multi-factor weighted function. For example, the following formula is used to determine the credibility score for each voxel. Credibility rating :

[0085] In the formula, voxels The number of times a path is traversed by a muon path (path coverage). The larger the value, the higher the credibility. The residual between the theoretical and measured projections of the voxel-corresponding path is denoted as . The smaller the value, the more stable the inversion results and the higher the reliability. The density gradient is the most stable, and the higher the reliability. For the estimation of density variance caused by noise propagation, The smaller the value, the higher the credibility.

[0086] Among them, each voxel Arranged according to spatial coordinates, a confidence field consistent with the density field dimension is formed (high-value areas are reliable areas, and low-value areas are suspicious areas). The confidence scoring method can be used to weight the final landslide interface extraction results. That is, the interface is extracted first in high-confidence areas to obtain the preliminary three-dimensional density distribution, while smoothing is appropriately performed in low-confidence areas, or the interface extraction update is delayed (prioritizing the trend of adjacent high-confidence areas). By using weighting processing, the interface extraction results are tilted towards high-confidence areas, thereby significantly reducing the risk of misjudgment of the interface due to insufficient paths or noise. S232, The preliminary three-dimensional density distribution of the landslide body is completed and refined to determine the landslide interface.

[0087] In this step, the completion and refinement of the preliminary three-dimensional density distribution of the landslide body can include supplementing the density values ​​of obviously missing data areas in the preliminary three-dimensional density distribution using simple interpolation (such as nearest neighbor interpolation) to avoid data gaps and achieve completion processing; or it can include reasonably adding data to different distribution areas in the preliminary three-dimensional density distribution to achieve refinement processing.

[0088] In this embodiment, by converting the density values ​​of each voxel into a three-dimensional density distribution, a three-dimensional Gaussian spline generation algorithm is used to process regions with different density distributions. This method has strong sparse adaptability and is suitable for unfavorable detection conditions with limited points and limited acquisition time, thereby improving the detection accuracy of the interface.

[0089] In one embodiment, such as Figure 5 As shown, the landslide interface detection method includes: S2321. Place the Gaussian elements in the preliminary three-dimensional density distribution of the landslide body to determine the density field of the superimposed Gaussian elements.

[0090] In this step, during the 3D Gaussian spline generation model (3DGS) stage, Gaussian elements are small volumetric function units used to approximate continuous density changes in the initial density field. For example, a 3D Gaussian function is initialized at each voxel center or high gradient point. The initial parameters in the preliminary 3D density distribution are determined by the voxel density field, which is the 3D density distribution matrix obtained through CS-IR inversion. Each voxel position (x,y,z) corresponds to a density value ρ(x,y,z), representing the average material density at that spatial point. The Gaussian elements are then superimposed to form the simulated density field.

[0091] Among them, the three-dimensional Gaussian function It is expressed by the following formula:

[0092] In the formula, For any point coordinate vector, corresponding to the coordinates of the current calculation voxel or spatial sampling point; For the first The coordinates of the center positions of each Gaussian element are initialized by the initial voxel density field distribution (such as CS-IR results) and continuously adjusted during optimization. Number the primitives; For the first The covariance matrix of each primitive represents the spatial scale and anisotropy of that primitive, corresponding to the scale (length, width, and height). For the first The density weight of each primitive determines the contribution intensity of that primitive to the overall density field.

[0093] Optionally, density texture features are determined based on the density field of Gaussian element superposition. The density texture features include the density gradient, directional second derivative, local density variance, and layered continuity index of each voxel. Based on the density texture features, the weak interlayer or slip zone structure of the landslide body is determined. Based on the weak interlayer or slip zone structure of at least part of the landslide body, the landslide interface is determined.

[0094] For example, the density gradient is determined by the drastic change in density of each voxel. .

[0095] The directional second derivative is determined by the following formula. :

[0096] In the formula, Density; For angles in different directions.

[0097] Local density variance is characterized by small and uniform density in the slip zone, while large variance exists in the fractured zone. The layered continuity index, calculated along the normal to the slip zone, shows significantly higher density consistency in continuous thin layers compared to surrounding areas. Density texture features allow for the automatic location of weak interlayers within the landslide body within the initial density field, enabling further refinement of the spatial morphology of the landslide interface and improving overall identification accuracy. By integrating the identified weak interlayers or slip zone structures with the preliminary three-dimensional density distribution, the spatial morphology of the landslide interface is further refined.

[0098] S2322. Perform convergence optimization on the loss function of the density field to determine a continuous Gaussian field. The loss function should include at least a projection consistency term and a regularization term.

[0099] In this step, the theoretical transmittance of the density field is calculated, and the loss function for its projection consistency term and regularization term to suppress excessively large or numerous primitives is determined based on the theoretical transmittance. The parameters are then optimized using the loss function to determine the continuous Gaussian field.

[0100] In some embodiments, the method for determining a continuous Gaussian field includes: The theory for determining the density field is through projection. Please refer to the following formula:

[0101] In the formula, For the path along the muon Computational theory through projection; For path The corresponding muon penetration path, this integral can be realized through numerical methods (e.g., path voxel interpolation accumulation), or calculated by simulating muon flux decay through Monte Carlo simulation; This indicates the continuous density field The theoretical transmission vectors projected along different muon paths are considered as the theoretical measurement results that should occur after the simulated muon passes through the current Gaussian field, and the continuous density field formed by the superposition of all Gaussian elements. The following formula is used to determine it:

[0102] Based on theory through projection and observed projection values Determine the projection consistency term in the loss function of the density field. ; Based on the Gaussian elements of the density field, the regularization term in the loss function of the density field is determined. .

[0103] Therefore, the loss function is defined as follows:

[0104] In the formula, Theory through projection; The observed projection values ​​obtained by the actual detector are calculated and converted into path cumulative mass thickness (or transmittance). ; Indicates the projection consistency term; is the regularization coefficient, which is the structural constraint coefficient in the continuous Gaussian field optimization stage. It is used to suppress the situation where the number of primitives is too large or the scale is too large, so as to maintain the sparsity of the density field and the physical rationality of the model. It is the covariance matrix; Number the primitives; This represents a regularization term that suppresses excessively large or numerous primitives.

[0105] The loss function is optimized using gradient descent or the Levenberg-Marquardt algorithm, which minimizes a nonlinear least squares problem, iteratively adjusted until the error converges. By minimizing this loss function, the Gaussian parameters are optimized, achieving convergent reconstruction of the continuous density field.

[0106] S2323. Based on the density gradient, refine or homogenize the continuous Gaussian field to determine the landslide interface.

[0107] For example, the continuous Gaussian field is processed according to the magnitude of the density gradient. For instance, regions with large density gradients are refined to improve the resolution of density abrupt change regions, such as slip surfaces or void boundaries; or regions with small density gradients, such as relatively homogeneous stable rock masses or filled soil layers, are homogenized, and Gaussian elements within single density volume regions are merged to reduce model redundancy and improve computational efficiency.

[0108] In this embodiment, a continuous Gaussian field is obtained by optimizing the density field using a loss function. Then, based on the density gradient, the regions with different density changes are refined or homogenized accordingly, thereby improving the characterization accuracy of the three-dimensional interface, reducing model redundancy, and improving model computation efficiency.

[0109] Methods for determining the density field of Gaussian element superposition include: The first and second regions were used to determine the preliminary three-dimensional density distribution of the landslide mass. The first region was configured as the density variation. Exceeding the first threshold The area, that is The second region represents the density change. Less than the second threshold The area, that is The second threshold is less than or equal to the first threshold; a first number of Gaussian elements are placed in the first region, and a second number of Gaussian elements are placed in the second region to determine the density field of the Gaussian element superposition, wherein the first number is greater than the second number.

[0110] For example, Gaussian primitives are split in regions where the density gradient is greater than a first preset gradient threshold to refine the continuous Gaussian field; Gaussian primitives are averaged in regions where the density gradient is less than a second preset gradient threshold to homogenize the continuous Gaussian field. For instance, one primitive is placed at the geometric center or gradient extremum point (the point with the largest gradient, closer to the actual density abrupt change location) of each voxel in the first region. In the second region, one primitive is placed every 3–5 voxels, with the specific number adjusted according to the voxel size; for example, if the voxel is 5m, one primitive is placed every 15m, with the location chosen as the average center of the group of voxels.

[0111] In this embodiment, by inserting different numbers of primitives into regions with different density distributions, the resolution is improved without adding too many redundant primitives to the model, thus reducing the computational load.

[0112] The above-described contents can be implemented individually or in various combinations, and all such variations are within the scope of this disclosure.

[0113] Combination Figure 6 As shown, the landslide interface detection device 60 includes: The acquisition module 61 is configured to acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device; The processing module 62 is configured to process the detection data and the outer surface data based on the compressed sensing iterative reconstruction algorithm to determine the landslide interface.

[0114] Optionally, the processing module 62 is specifically used for: Based on the detection data and external surface data, an optimization objective function is constructed; The density values ​​of each voxel are determined by minimizing the objective function based on the iterative shrinkage threshold algorithm. The landslide boundary was determined based on the density values ​​of each voxel.

[0115] Optionally, the processing module 62 is specifically used for: The outer surface data is discretized into a three-dimensional voxel mesh based on the preset voxel size to determine the number of voxels; Based on the probe data, determine the number of paths; The projection matrix and observation vector are determined based on the number of paths and the number of voxels; The voxel density vector is determined based on the projection matrix and the observation vector. An optimization objective function is constructed based on the voxel density vector, projection matrix, and observation vector.

[0116] Furthermore, based on the above embodiments, the processing module 62 can also be used for: The landslide interface was determined based on a three-dimensional Gaussian spline generation algorithm.

[0117] Optionally, the processing module 62 is specifically used for: Based on the density values ​​of each voxel, the preliminary three-dimensional density distribution of the landslide body is determined; Based on the 3D Gaussian spline generation algorithm, the preliminary 3D density distribution of the landslide body is completed and refined to determine the landslide interface.

[0118] Optionally, the processing module 62 is specifically used for: By placing Gaussian elements into the preliminary three-dimensional density distribution of the landslide body, the density field of the Gaussian element superposition is determined. The loss function of the density field is convergently optimized to determine the continuous Gaussian field. The loss function must include at least a projection consistency term and a regularization term. Based on the density gradient, the continuous Gaussian field is refined or homogenized to determine the landslide boundary.

[0119] Optionally, the processing module 62 is specifically used for: A first region and a second region are determined to form the preliminary three-dimensional density distribution of the landslide body. The first region is configured as a region where the density change exceeds a first threshold, and the second region is a region where the density change is less than a second threshold. The second threshold is less than or equal to the first threshold.

[0120] Furthermore, based on the above embodiments, the processing module 62 can also be used for: A first number of Gaussian elements are placed in the first region, and a second number of Gaussian elements are placed in the second region. The density field of the superposition of Gaussian elements is determined, with the first number being greater than the second number.

[0121] Optionally, the processing module 62 is specifically used for: The theory that determines the density field is derived through projection; Based on theoretical projection and observed projection values, the projection consistency term in the loss function of the density field is determined. Based on the Gaussian elements of the density field, the regularization term in the loss function of the density field is determined.

[0122] The landslide interface detection device provided in this embodiment can perform the landslide interface detection method of the above embodiment. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0123] In this embodiment of the invention, electronic devices or main control devices can be divided into functional modules according to the above method examples. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional module. It should be noted that the module division in this embodiment of the invention is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0124] In the specific implementation of the aforementioned landslide interface detection device, each module can be implemented as a processor. The processor can execute computer execution instructions stored in the memory, so that the processor executes the aforementioned landslide interface detection method.

[0125] Combination Figure 7 As shown, the electronic device 70 includes: At least one processor 71 and memory 72.

[0126] The electronic device also includes a communication component 73.

[0127] The processor 71, memory 72, and communication component 73 are connected via bus 74.

[0128] In the specific implementation process, at least one processor 71 executes computer execution instructions stored in memory 72, causing at least one processor 71 to execute the landslide interface detection method as described above.

[0129] The specific implementation process of processor 71 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0130] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0131] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage.

[0132] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0133] The above describes the solutions provided by the embodiments of the present invention for the functions implemented by the electronic device and the main control device.

[0134] It is understandable that electronic devices or main control devices include hardware structures and / or software modules that perform the above functions in order to achieve the above functions.

[0135] By combining the units and algorithm steps of the various examples described in the embodiments of this invention, the embodiments of this invention can be implemented in hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the technical solutions of the embodiments of this invention.

[0136] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements a landslide interface detection method.

[0137] The computer program product provided in this embodiment can execute the landslide interface detection method of the above embodiment. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0138] This disclosure also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described landslide interface detection method.

[0139] The computer-readable storage medium provided in this embodiment can execute the landslide interface detection method of the above embodiment. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0140] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0141] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] In this disclosure, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase “comprising…” does not exclude the presence of additional identical elements in the article or device that includes said element.

[0145] Although preferred embodiments of the present disclosure have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this disclosure.

[0146] Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims of this disclosure and their equivalents, the intent of this disclosure also includes these modifications and variations.

Claims

1. A method for detecting landslide interfaces, characterized in that, include: Acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device; Based on the compressed sensing iterative reconstruction algorithm, the detection data and the outer surface data are processed to determine the landslide interface.

2. The landslide interface detection method according to claim 1, characterized in that, The compressed sensing iterative reconstruction algorithm processes the detected data and the outer surface data to determine the landslide interface, including: Based on the detection data and the outer surface data, an optimization objective function is constructed; The density values ​​of each voxel are determined by minimizing the optimization objective function based on the iterative shrinkage threshold algorithm. The landslide interface is determined based on the density values ​​of each voxel.

3. The landslide interface detection method according to claim 2, characterized in that, The step of constructing an optimization objective function based on the probe data and the outer surface data includes: The outer surface data is discretized into a three-dimensional voxel grid based on a preset voxel size to determine the number of voxels; Based on the detection data, the number of paths is determined; Based on the number of paths and the number of voxels, determine the projection matrix and the observation vector; Based on the projection matrix and the observation vector, the voxel density vector is determined; An optimization objective function is constructed based on the voxel density vector, the projection matrix, and the observation vector.

4. The landslide interface detection method according to claim 2, characterized in that, After determining the density values ​​of each voxel, the method for determining the landslide interface further includes: The landslide interface is determined based on a three-dimensional Gaussian spline generation algorithm; and / or, The method for determining the landslide interface based on the three-dimensional Gaussian spline generation algorithm includes: Based on the density values ​​of each voxel, the preliminary three-dimensional density distribution of the landslide body is determined; Based on the aforementioned three-dimensional Gaussian spline generation algorithm, the preliminary three-dimensional density distribution of the landslide body is completed and refined to determine the landslide interface.

5. The landslide interface detection method according to claim 4, characterized in that, The algorithm for generating three-dimensional Gaussian splines is used to complete and refine the preliminary three-dimensional density distribution of the landslide body, and to determine the landslide interface, including: By placing Gaussian elements into the preliminary three-dimensional density distribution of the landslide body, the density field of the superimposed Gaussian elements is determined. The loss function of the density field is convergent and optimized to determine a continuous Gaussian field. The loss function includes at least a projection consistency term and a regularization term. Based on the density gradient, the continuous Gaussian field is refined or homogenized to determine the landslide interface; and / or, The step of placing Gaussian elements in the preliminary three-dimensional density distribution of the landslide body and determining the density field of the superimposed Gaussian elements includes: A first region and a second region are determined to represent the preliminary three-dimensional density distribution of the landslide body. The first region is configured as a region where the density change exceeds a first threshold, and the second region is a region where the density change is less than a second threshold, wherein the second threshold is less than or equal to the first threshold; and / or The convergence optimization of the loss function of the density field to determine the continuous Gaussian field includes: The theory for determining the density field is projected. Based on the aforementioned theory, the projection consistency term in the loss function of the density field is determined through projection and observation of the projection value. Based on the Gaussian elements of the density field, the regularization term in the loss function of the density field is determined.

6. The landslide interface detection method according to claim 5, characterized in that, The method further includes: A first number of Gaussian elements are placed in the first region, and a second number of Gaussian elements are placed in the second region. The density field of the superimposed Gaussian elements is determined, wherein the first number is greater than the second number.

7. The landslide interface detection method according to claim 3, characterized in that, The method further includes: Based on the outer surface data of the landslide and the spatial location of the muon detector, a three-dimensional visible domain model is constructed. Based on the aforementioned three-dimensional visible domain model, the visible voxels are determined; The projection matrix is ​​determined based on the visible voxels.

8. The landslide interface detection method according to claim 6, characterized in that, The method further includes: Based on the density field of the superimposed Gaussian elements, density texture features are determined, including the density gradient of each voxel, the directional second derivative, the local density variance, and the layered continuity index. Based on the density texture features, the weak interlayer or slip zone structure of the landslide body is determined; The landslide interface is determined based on the weak interlayers or slip zones of at least a portion of the landslide body.

9. The landslide interface detection method according to claim 4, characterized in that, The method further includes: Obtain the credibility score for each voxel; Based on the credibility scores of each voxel, a credible preliminary three-dimensional density distribution is determined.

10. A landslide interface detection device, characterized in that, include: The acquisition module is configured to acquire detection data collected by the muon detector and external surface data of the landslide body collected by the topographic mapping device; The processing module is configured to process the detection data and the outer surface data based on the compressed sensing iterative reconstruction algorithm to determine the landslide interface.

Images