An edge-computing-based method for determining a maximum design safety factor of a mine slope

Abstract

The application discloses a kind of based on edge computing's mine slope maximum design safety factor value method, it is related to image analysis technical field, including, obtain the high-resolution three-dimensional laser point cloud data and image data of slope surface, and carry out time synchronization, geometric correction and tile slicing, form point cloud block sequence and orthophoto sequence;Point cloud block sequence and orthophoto sequence are input into edge computing node, start multi-modal fusion mechanism and self-calibration mechanism, output point cloud-image fusion data block;Based on structured feature information, real-time solution mine slope under current constraint condition instantaneous safety factor;With instantaneous safety factor and structured feature information as benchmark, by analyzing the influence on slope stability under different working condition combinations, determine the maximum design safety factor value.The application improves the fusion quality of point cloud and image data in edge computing node by starting multi-modal fusion mechanism and self-calibration mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image analysis, and in particular to a method for obtaining a maximum design safety factor of a mine slope based on edge computing. BACKGROUND

[0002] In recent years, with the rapid development of unmanned aerial vehicle remote sensing and laser radar technology, the mine slope monitoring method based on three-dimensional point cloud and image data has become a research hotspot. In particular, the combination of multi-sensor fusion technology and edge computing architecture provides a new technical path for realizing real-time analysis of slope stability. In the field of image analysis, the multi-modal data registration, three-dimensional reconstruction and feature extraction algorithms based on deep learning are constantly improving, which improves the accuracy and efficiency of automatic processing of geological information.

[0003] However, the existing technology still has deficiencies in the edge side collaborative processing of massive heterogeneous data. Due to the inherent differences in acquisition frequency, resolution characteristics and physical dimensions of point cloud and image data, traditional methods are difficult to achieve high-precision spatio-temporal synchronization and semantic-level fusion in resource-constrained edge nodes, resulting in poor quality of three-dimensional model reconstruction and decreased feature extraction accuracy, which cannot meet the stringent requirements of real-time slope stability calculation on data quality, and becomes a technical bottleneck restricting the reliability of monitoring. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a method for obtaining a maximum design safety factor of a mine slope based on edge computing, which solves the problem of insufficient multi-modal geological data fusion quality in the existing edge computing environment, which limits the accuracy and real-time performance of slope safety factor analysis.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The application provides a mine slope maximum design safety factor value method based on edge computing, which comprises the following steps: obtaining high-resolution three-dimensional laser point cloud data and image data of a slope surface, and performing time synchronization, geometric correction and tile slicing to form a point cloud block sequence and an orthographic image sequence; inputting the point cloud block sequence and the orthographic image sequence into an edge computing node, starting a multi-modal fusion mechanism and a self-calibration mechanism, and outputting point cloud-image fusion data blocks; performing three-dimensional reconstruction and information fusion on the point cloud-image fusion data blocks, and performing smooth splicing according to the overlapping areas between adjacent data blocks to generate a three-dimensional geological model; constructing and configuring a lightweight multi-scale feature extraction architecture, inputting the three-dimensional geological model into the lightweight multi-scale feature extraction architecture, dynamically assigning a calculation path, extracting rock mass structure surface geometric parameters and micro-deformation vectors, and outputting structured feature information; based on the structured feature information, the instantaneous safety factor of the mine slope under the current constraint condition is solved in real time; and the instantaneous safety factor and the structured feature information are taken as the reference, the influence of different working condition combinations on the slope stability is analyzed, and the maximum design safety factor value is determined.

[0008] As a preferred scheme of the mine slope maximum design safety factor value method based on edge computing, the point cloud block sequence and the orthographic image sequence are formed by the following steps,

[0009] A precise clock synchronization protocol is used to stamp the high-resolution three-dimensional laser point cloud data and the image data of the slope surface with a sub-microsecond level time stamp, and a time-synchronized multi-modal data stream is generated;

[0010] The time-synchronized multi-modal data stream is associated with multi-modal features by taking the initial pose provided by the inertial navigation as a constraint, and a pose graph is constructed for joint optimization to obtain point cloud and orthographic image with accurate geometric position alignment;

[0011] The point cloud and orthographic image with accurate geometric position alignment are subjected to semantic segmentation to generate a geological feature difference area map;

[0012] The point cloud and orthographic image with accurate geometric position alignment are subjected to non-uniform grid division according to the geological feature difference area map to form the point cloud block sequence and the orthographic image sequence.

[0013] As a preferred scheme of the mine slope maximum design safety factor value method based on edge computing, the point cloud block sequence and the orthographic image sequence are formed by the following steps,

[0014] Feature mapping is performed on the point cloud block sequence and the orthographic image sequence by a heterogeneous feature extraction network to generate point cloud geometric feature vectors and image texture feature vectors, and an uncertainty estimation value is assigned to each feature vector;

[0015] The point cloud geometry feature vector and the image texture feature vector are weighted and averaged to generate an initial fusion feature vector;

[0016] Based on the uncertainty estimation value of each feature vector and the cosine similarity between each feature vector and the initial fusion feature vector, dynamic weight distribution is performed, and through iterative fusion and self-calibration refinement, until the initial fusion feature vector meets the convergence condition, the refined fusion feature vector and the uncertainty measure are output;

[0017] The refined fusion feature vector and the uncertainty measure are input into the deconvolution decoder network, the refined fusion feature vector is weighted and modulated using the uncertainty measure, and a point-cloud-image fusion data block is generated.

[0018] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the three-dimensional geological model is generated, and the specific steps are as follows,

[0019] Each point-cloud-image fusion data block is reconstructed in three dimensions by a local implicit neural representation network to generate a local implicit surface model corresponding to each data block;

[0020] For the overlapping regions of adjacent data blocks, the uncertainty measures of the data blocks are used to assign dynamic weights to the local implicit surface models corresponding to the data blocks, and weighted fusion is performed to generate a global implicit surface representation.

[0021] The global implicit surface representation is extracted and texture mapped to generate a three-dimensional geological model.

[0022] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the three-dimensional geological model is generated, and the specific steps are as follows,

[0023] A macro architecture search space is constructed on the edge computing node, and a meta-controller with a recurrent neural network as the core is constructed according to a predefined architecture coding specification;

[0024] Global context descriptors of the three-dimensional geological model are extracted, input into the meta-controller to generate subnet architecture parameters, and based on the subnet architecture parameters, the meta-controller is instantiated from the macro architecture search space to a lightweight feature extraction subnet;

[0025] The performance of the lightweight feature extraction subnet is evaluated through a context-aware evaluation mechanism to generate an evaluation result, and the meta-controller is updated based on the evaluation result to complete the construction and configuration of the lightweight multi-scale feature extraction architecture.

[0026] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the three-dimensional geological model is generated, and the specific steps are as follows,

[0027] transforming a triangular mesh surface of a three-dimensional geological model into a graph structure, and calculating a local geometric descriptor of each vertex in the graph structure;

[0028] inputting the graph structure into a lightweight multi-scale feature extraction architecture, and dynamically assigning an optimal feature extraction path for each vertex according to the local geometric descriptor of each vertex by a built-in routing strategy network;

[0029] extracting rock mass structural surface geometric parameters and micro-deformation vectors along the assigned optimal feature extraction path by a lightweight feature extraction subnetwork;

[0030] aggregating the rock mass structural surface geometric parameters and micro-deformation vectors extracted from all vertices according to the topological connection relationship of the graph structure, and outputting structured feature information.

[0031] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the rock mass structural surface geometric parameters include the dip, dip angle, spacing, trace length and roughness of the rock mass structural surface.

[0032] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the current constraint condition includes the field geological condition of the mine slope, the real-time monitored external load condition and the engineering state condition.

[0033] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the real-time calculation of the instantaneous safety factor of the mine slope under the current constraint condition includes the following specific steps,

[0034] real-time analysis of the structured feature information, extraction of the dip, dip angle, spacing, trace length, roughness and micro-deformation vector of the rock mass structural surface at the current time, and generation of a multi-dimensional input vector combined with the current constraint condition;

[0035] parallel input of the multi-dimensional input vector into different types of safety factor prediction models, calculation of safety factor prediction values, and calculation of fusion weights of different types of safety factor prediction models through a dynamic weighting algorithm;

[0036] weighted fusion of the safety factor prediction values of different types of safety factor prediction models by using the fusion weights, and output of the instantaneous safety factor.

[0037] As a preferred scheme of the mine slope maximum design safety factor value determination method based on edge computing, the determination of the maximum design safety factor value includes the following specific steps,

[0038] Taking the instantaneous safety factor and the structured feature information as the benchmark, a multi-objective optimization algorithm is used to perform parallel search under different preset working condition combinations to obtain an optimal solution set;

[0039] Multi-criteria decision analysis is performed on the optimal solution set to determine the maximum design safety factor value.

[0040] The present application has the beneficial effects that: by starting the multi-modal fusion mechanism and the self-calibration mechanism, the fusion quality of point cloud and image data at the edge computing node is improved; by using the heterogeneous feature extraction network to learn the geometric features of the point cloud and the texture features of the image respectively, the iterative optimization algorithm is used to realize the accurate alignment of the feature space, effectively overcoming the data inconsistency problem caused by sensor differences and external interference, the self-calibration mechanism can dynamically detect and correct the registration error in the fusion process, ensuring that the output point cloud-image fusion data block has both millimeter-level geometric accuracy and rich semantic information, providing reliable input for three-dimensional reconstruction, and ensuring the accuracy and reliability of the instantaneous safety factor calculation result, providing a solid data foundation for slope stability decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0042] Fig. 1 The flowchart of the method for obtaining the maximum design safety factor of the mine slope based on edge computing.

[0043] Fig. 2 The flowchart of the output point cloud-image fusion data block.

[0044] Fig. 3 The flowchart of generating a three-dimensional geological model.

[0045] Fig. 4 The flowchart of outputting structured feature information. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0047] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0048] Second, the "one embodiment" or "an embodiment" referred to herein means a particular feature, structure, or characteristic including an implementation that can be included in at least one implementation of the application. The appearances of "in one embodiment" or "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all referring to a single, alternative embodiment, or a single, alternative implementation.

[0049] Referring to Figs. 1-4 For one embodiment of the application, the embodiment provides a mine slope maximum design safety factor value method based on edge computing, comprising the following steps:

[0050] S1, obtaining high-resolution three-dimensional laser point cloud data and image data of the slope surface, and performing time synchronization, geometric correction and tiling slicing to form point cloud block sequences and orthographic image sequences.

[0051] S1.1: using a precise clock synchronization protocol, time stamps of sub-microsecond level precision are added to the high-resolution three-dimensional laser point cloud data and image data of the slope surface, and a time-synchronized multi-modal data stream is generated;

[0052] It should be noted that by using the precise clock synchronization protocol, the clock of the unmanned aerial vehicle onboard laser radar for collecting high-resolution three-dimensional laser point cloud data and the clock of the multispectral camera for collecting image data are synchronized to a unified time reference source, so that the time of all unmanned aerial vehicle onboard laser radars and multispectral cameras is consistent, reaching sub-microsecond level precision; during data collection, a time stamp is assigned to each data point in the high-resolution three-dimensional laser point cloud data to record the precise time of point cloud collection; at the same time, a time stamp is assigned to each frame of image data to record the precise time of image shooting; according to the correspondence of the time stamps, the high-resolution three-dimensional laser point cloud data and the image data are aligned on the time axis, and a time-synchronized multi-modal data stream is generated.

[0053] S1.2: with the initial pose provided by the inertial navigation as a constraint, the time-synchronized multi-modal data stream is associated with multi-modal features, and a pose graph is constructed for joint optimization solution to obtain point clouds and orthographic images with geometric position accurate alignment;

[0054] Specifically, the initial pose provided by inertial navigation is used as a constraint to implement multi-modal feature association on time-synchronized multi-modal data streams, to identify point cloud feature points from high-resolution three-dimensional laser point cloud data, to identify image feature points from image data, to establish a matching relationship between the point cloud feature points and the image feature points, to form feature correspondence, to construct a pose graph based on the feature correspondence, the nodes of the pose graph representing the pose at the data acquisition time, and the edges of the pose graph representing the relative pose constraints generated by the feature matching; the optimized pose sequence is obtained through joint optimization of the pose graph; and the high-resolution three-dimensional laser point cloud data and the image data are geometrically corrected using the optimized pose sequence, so that the point cloud and the orthophoto are accurately aligned in geometric position.

[0055] S1.3: performing semantic segmentation on the point cloud and the orthophoto accurately aligned in geometric position to generate a geological feature difference region map;

[0056] It should be noted that the pre-training process of the semantic segmentation network is as follows: a historical data set containing a large number of labeled images and a large number of labeled point clouds is called, each pixel of each image in the historical data set is labeled with a corresponding object class label, and each point of each point cloud is labeled with a corresponding object class label; the historical data set is used to train the semantic segmentation network, the historical data set is input into the semantic segmentation network to generate a prediction result, the prediction result is compared with the true class label to obtain a difference result, the internal parameters of the semantic segmentation network are adjusted according to the difference result to reduce the prediction error, and the process is repeated until the prediction performance of the semantic segmentation network reaches a stable state, at which time the parameters of the semantic segmentation network are saved to obtain the pre-trained semantic segmentation network.

[0057] Specifically, the semantic segmentation network is used to assign a geological class label to each point in the point cloud accurately aligned in geometric position, and to assign the same geological class label to each pixel in the orthophoto accurately aligned in geometric position, to obtain a point cloud semantic segmentation result and an image semantic segmentation result, respectively; the point cloud semantic segmentation result and the image semantic segmentation result are fused to generate a geological feature classification map, each region in the geological feature classification map is labeled with a specific geological class; based on the geological feature classification map, the change of the geological class between adjacent regions is identified, and the geological feature difference region map is generated by comparing the change of the geological class.

[0058] It should be noted that the geological class label includes rock, soil and fissure.

[0059] S1.4: performing non-uniform grid division on the point cloud and the orthophoto accurately aligned in geometric position according to the geological feature difference region map to form a point cloud block sequence and an orthophoto sequence.

[0060] Specifically, according to the geological feature difference area map, a region with a dramatic change in geological features and a region with a gentle change in geological features are identified in the map; according to the change intensity of the geological features, a grid division density rule is set, high-density grid division is adopted in the region with a dramatic change in geological features, and low-density grid division is adopted in the region with a gentle change in geological features; the point cloud with accurate geometric position alignment is subjected to a non-uniform grid division rule to generate a point cloud block; the orthographic image with accurate geometric position alignment is subjected to the same non-uniform grid division rule to generate an orthographic image block; the point cloud block and the orthographic image block are arranged in order of grid space to form a point cloud block sequence and an orthographic image sequence.

[0061] S2, input the point cloud block sequence and the orthographic image sequence into an edge computing node, start a multi-modal fusion mechanism and a self-calibration mechanism, and output a point cloud-image fusion data block.

[0062] S2.1: feature mapping is performed on the point cloud block sequence and the orthographic image sequence by a heterogeneous feature extraction network to generate a point cloud geometric feature vector and an image texture feature vector, and an uncertainty estimation value is assigned to each feature vector;

[0063] Specifically, the pre-training process of the heterogeneous feature extraction network is independently performed on a point cloud feature extraction network and an image feature extraction network, the pre-training of the point cloud feature extraction network uses a point cloud data set containing a large amount of labeled point cloud data, each point cloud sample in the point cloud data set is labeled with a corresponding geometric feature class; the mapping relationship between the point cloud data and the geometric feature class is learned by the point cloud feature extraction network, the internal parameters of the point cloud feature extraction network are adjusted until the point cloud feature extraction network can stably generate a point cloud geometric feature vector with distinguishability; the pre-training of the image feature extraction network uses an image data set containing a large amount of labeled images, each image sample in the data set is labeled with a corresponding texture feature class; the mapping relationship between the image data and the texture feature class is learned by the image feature extraction network, the internal parameters of the image feature extraction network are adjusted until the image feature extraction network can stably generate an image texture feature vector with distinguishability, thereby obtaining a pre-trained heterogeneous feature extraction network.

[0064] It should be noted that the uncertainty estimation value is generated in the pre-training process of the heterogeneous feature extraction network and is realized through the uncertainty estimation branch inside the heterogeneous feature extraction network; in the pre-training process, the uncertainty estimation branch is trained together with the feature extraction network (point cloud or image); the uncertainty estimation branch learns by evaluating the volatility of the prediction results of the feature extraction network (point cloud or image) on the training data; specifically, the feature extraction network (point cloud or image) is trained to output an estimation value reflecting the reliability of the output feature vector (point cloud geometry or image texture) while outputting the feature vector (point cloud geometry or image texture); for example, for different training samples of the same category, the more consistent the feature vector (point cloud geometry or image texture) generated by the heterogeneous feature extraction network, the lower the uncertainty estimation value; on the contrary, the higher the uncertainty estimation value; when the pre-trained heterogeneous feature extraction network processes new point cloud blocks or orthophoto blocks, the feature extraction part generates new feature vectors (point cloud geometry or image texture), and the uncertainty estimation branch directly generates the corresponding uncertainty estimation value according to the degree of fit between the current input data and the training data distribution.

[0065] Specifically, the heterogeneous feature extraction network includes a point cloud feature extraction network and an image feature extraction network; the point cloud feature extraction network is applied to each point cloud block in the point cloud block sequence to extract geometric features and generate a point cloud geometry feature vector; at the same time, the image feature extraction network is applied to each orthophoto block in the orthophoto sequence to extract texture features and generate an image texture feature vector; the uncertainty estimation branch attached to the heterogeneous feature extraction network is used to assign uncertainty estimation values to the point cloud geometry feature vector and the image texture feature vector, and the uncertainty estimation values reflect the reliability degree of the point cloud geometry feature vector and the image texture feature vector.

[0066] S2.2: Weighted average of the point cloud geometry feature vector and the image texture feature vector to generate an initial fusion feature vector, the expression is:

[0067] ;

[0068] In the formula, denotes the initial fusion feature vector, denotes the weight of the point cloud geometry feature vector, denotes the point cloud geometry feature vector, denotes the weight of the image texture feature vector, denotes the image texture feature vector.

[0069] It should be noted that the weight of the point cloud geometry feature vector and the weight of the image texture feature vector are based on the initial equal weight setting, and the example value is 0.5, and the value is determined to ensure that the point cloud geometry feature vector and the image texture feature vector have equal contribution in the early stage of fusion.

[0070] S2.3: Based on the uncertainty estimation value of each feature vector and the cosine similarity between each feature vector and the initial fusion feature vector, dynamic weight distribution is performed, and through iterative fusion and self-calibration refinement, until the initial fusion feature vector meets the convergence condition, the refined fusion feature vector and the uncertainty measure are outputted;

[0071] Specifically, based on the uncertainty estimation value of the point cloud geometric feature vector and the uncertainty estimation value of the image texture feature vector, and the cosine similarity between the point cloud geometric feature vector and the initial fusion feature vector and the cosine similarity between the image texture feature vector and the initial fusion feature vector, dynamic weight distribution is performed. The dynamic weight distribution adjusts the weights of the point cloud geometric feature vector and the image texture feature vector through the proportional relationship between the cosine similarity and the uncertainty estimation value. The adjusted weights are used to perform weighted averaging on the point cloud geometric feature vector and the image texture feature vector to generate an updated fusion feature vector. Dynamic weight distribution and weighted averaging are repeated to perform iterative fusion and self-calibration refinement of the weight distribution of the point cloud geometric feature vector and the image texture feature vector in the fusion process. After each iteration, the difference between the updated fusion feature vector and the previous fusion feature vector is compared. When the difference is less than a preset difference threshold, it is considered to meet the convergence condition, and the refined fusion feature vector is outputted. Based on the weight of the point cloud geometric feature vector and the weight of the image texture feature vector in the iterative fusion process, the uncertainty estimation value of the point cloud geometric feature vector and the uncertainty estimation value of the image texture feature vector are combined to generate the refined fusion feature vector and the uncertainty measure.

[0072] It should be noted that the preset difference threshold is set based on historical experimental data and the requirement for fusion accuracy. By running the iterative fusion process on multiple sets of sample data with known accurate fusion results, the feature vector difference value when sufficient convergence is reached is recorded as a benchmark. According to the benchmark data, a candidate threshold range is determined, and the performance of each candidate threshold on independent verification samples is tested, including evaluating the average number of iterations and the deviation of the fusion result from the true value. By balancing the convergence speed and the fusion accuracy, the candidate threshold that can achieve the best balance is selected as the preset difference threshold, with a value range of 0.001 to 0.01. The value is based on balancing fusion accuracy and computational efficiency. A value lower than 0.001 may cause premature convergence and insufficient fusion, and a value higher than 0.01 may cause excessive iterations and resource waste.

[0073] S2.4: The refined fusion feature vector and the uncertainty measure are inputted into the deconvolution decoder network, the refined fusion feature vector is weighted and modulated using the uncertainty measure, and the point-cloud-image fusion data block is generated.

[0074] It should be noted that the pre-training process of the deconvolution decoder network is as follows: using a historical data set containing a large number of samples, each sample in the historical data set includes a feature vector and a corresponding point-cloud-image fusion data block, by inputting each feature vector into the deconvolution decoder network, a predicted data block is generated, the predicted data block is compared with the real point-cloud-image fusion data block to obtain a difference, the internal parameters of the deconvolution decoder network are adjusted according to the difference, and the operation is repeated until the deconvolution decoder network can stably reconstruct the point-cloud-image fusion data block, thereby obtaining the pre-trained deconvolution decoder network.

[0075] Specifically, the refined fusion feature vector and the uncertainty measure are combined with the deconvolution decoder network, the deconvolution decoder network reconstructs the refined fusion feature vector into spatial data through a deconvolution layer and an up-sampling layer; at the same time, the uncertainty measure is used to implement weighted modulation on the refined fusion feature vector, in the weighted modulation, the intensity of the refined fusion feature vector is adjusted according to the numerical value of the uncertainty measure, the feature part with a lower uncertainty measure is enhanced, and the feature part with a higher uncertainty measure is inhibited; after the weighted modulation, the deconvolution decoder network generates a point-cloud-image fusion data block.

[0076] S3, the point-cloud-image fusion data block is three-dimensionally reconstructed and information is fused, and is smoothly spliced according to the overlapping area between adjacent data blocks to generate a three-dimensional geological model.

[0077] S3.1: three-dimensionally reconstructing each point-cloud-image fusion data block through a local implicit neural representation network to generate a local implicit surface model corresponding to each data block;

[0078] It should be noted that the pre-training process of the local implicit neural representation network is as follows: using a historical data set containing a large number of known surface model samples, each sample in the historical data set includes a spatial coordinate point and a corresponding real signed distance value, the specific network structure type of the local implicit neural representation network is a multi-layer perceptron, the mapping relationship between the spatial coordinate point and the signed distance value is learned, the internal parameters of the local implicit neural representation network are iteratively adjusted, the signed distance value predicted by the local implicit neural representation network is gradually close to the real signed distance value through a mean square error loss function, until the local implicit neural representation network can stably predict the signed distance value, thereby obtaining the pre-trained local implicit neural representation network.

[0079] Specifically, the three-dimensional space range occupied by the point cloud-image fusion data block is divided into a regular grid array, for each spatial coordinate point in the grid array, the spatial coordinate point is combined with the corresponding feature information in the point cloud-image fusion data block; the combined spatial coordinate and feature information are sent to the local implicit neural representation network to predict the signed distance value of the coordinate point in the grid array, the signed distance value representing the nearest distance of the coordinate point to the inferred object surface and the inside-outside relationship on the surface; all spatial coordinate points in the entire grid array are traversed to generate a complete signed distance field, and the signed distance field is the local implicit surface model corresponding to the point cloud-image fusion data block.

[0080] S3.2: For the overlapping region of adjacent data blocks, the uncertainty measure of each data block is used to assign a dynamic weight to the local implicit surface model corresponding to each data block, and the weighted fusion is performed to generate a global implicit surface representation;

[0081] Specifically, for the overlapping region of adjacent point cloud-image fusion data blocks, the spatial overlapping part is found, and for each spatial coordinate point in the overlapping region, the signed distance value predicted by the local implicit surface model corresponding to each adjacent point cloud-image fusion data block is obtained; the uncertainty measure of each point cloud-image fusion data block is used to assign a dynamic weight to the signed distance value prediction of each data block, and the data block with lower uncertainty measure obtains higher weight, the assigned weight is used to combine the signed distance values to obtain the fused signed distance value of each coordinate point, and the weighted combination is completed for all spatial coordinate points to generate a global implicit surface representation.

[0082] S3.3: Isosurface extraction and texture mapping are performed on the global implicit surface representation to generate a three-dimensional geological model.

[0083] Specifically, the mobile cube algorithm is used to traverse the three-dimensional space of the global implicit surface representation, the mobile cube algorithm locates the isosurface point with a signed distance value of zero in a small cube composed of adjacent grid points, and connects the isosurface points to form a triangular patch, after traversing the entire three-dimensional space, all triangular patches are combined into a continuous triangular mesh model, and the triangular mesh model represents the stratum surface, the texture color information corresponding to the position of each vertex of the triangular mesh model is obtained from the orthophoto sequence, the texture color information is assigned to the corresponding vertex of the triangular mesh model, and a three-dimensional geological model is generated.

[0084] S4, construct and configure a lightweight multi-scale feature extraction architecture, input the three-dimensional geological model into the lightweight multi-scale feature extraction architecture, dynamically dispatch the calculation path, extract the rock mass structure surface geometric parameters and micro-deformation vector, and output the structured feature information.

[0085] S4.1: Constructing a macro-architecture search space on the edge computing node, and constructing a meta-controller with recurrent neural network as the kernel according to the pre-defined architecture coding specification;

[0086] It should be noted that the pre-defined architecture coding specification refers to the rule of mapping each possible neural network architecture configuration in the macro-architecture search space to a fixed-length digital sequence, which explicitly specifies the specific neural network layer type or parameter value corresponding to each digit in the digital sequence.

[0087] Specifically, a set of optional neural network components are defined as the basic elements of the macro-architecture search space, including convolutional layers, pooling layers, and fully connected layers; configurable parameter ranges are set for each neural network component (for example, the number of filters in a convolutional layer can be selected within a specified range, and the kernel size of a convolutional layer can be selected within a specified size); all possible architecture variants are enumerated according to the combination of all neural network components and configurable parameters, forming the macro-architecture search space; according to the pre-defined architecture coding specification, each architecture variant in the macro-architecture search space is mapped to a fixed-length digital sequence; a meta-controller with recurrent neural network as the kernel is constructed, and the recurrent neural network of the meta-controller is organized as a sequence generator, which is trained to generate a digital sequence conforming to the architecture coding specification according to given conditions.

[0088] S4.2: Extracting global context descriptors of the three-dimensional geological model, inputting the meta-controller to generate subnet architecture parameters, and instantiating a lightweight feature extraction subnet from the macro-architecture search space based on the subnet architecture parameters;

[0089] Specifically, the global context descriptors of the three-dimensional geological model are extracted, the overall size, volume, and surface area of the three-dimensional geological model are obtained by counting the vertex coordinates and triangular facet distribution of the three-dimensional geological model, forming the global context descriptors; the global context descriptors are combined with the meta-controller, and the meta-controller generates a sequence of subnet architecture parameters based on the recurrent neural network kernel; based on the sequence of subnet architecture parameters, the corresponding architecture configuration is matched from the macro-architecture search space, and a lightweight feature extraction subnet is instantiated.

[0090] S4.3: Evaluating the performance of the lightweight feature extraction subnet through a context-aware evaluation mechanism, generating an evaluation result, and updating the meta-controller based on the evaluation result to complete the construction and configuration of the lightweight multi-scale feature extraction architecture.

[0091] It should be noted that the context-aware evaluation mechanism is a method of testing the lightweight feature extraction subnet through a historical validation dataset containing known rock mass structure surface labels, and evaluating the performance by comparing the accuracy of the predicted results with the true labels.

[0092] Specifically, a historical validation dataset containing known rock mass structure surface annotations is used. 3D geological model samples from this dataset are fed into a lightweight feature extraction subnet to obtain prediction results. These predictions are compared with the actual annotations, and the ratio of correct predictions to total predictions is used to generate an accuracy rate as an evaluation result. Based on the accuracy rate, a policy gradient method is employed to adjust the parameters of the recurrent neural network in the meta-controller, making the meta-controller tend to generate subnet architecture parameters that achieve higher accuracy. This evaluation and update process is repeated until the accuracy of the lightweight feature extraction subnet stabilizes, thus completing the construction and configuration of the lightweight multi-scale feature extraction architecture.

[0093] S4.4: Convert the triangular mesh surface of the 3D geological model into a graph structure and compute the local geometric descriptor for each vertex in the graph structure;

[0094] Specifically, a graph structure is constructed by mapping each vertex in the triangular mesh to a node in the graph structure, and mapping the edges of the triangular mesh connecting the vertices to the edges connecting the corresponding nodes in the graph structure. The normal vector of a vertex is calculated by the normal vectors of the triangular faces adjacent to the vertex, and the curvature of the vertex is calculated by fitting the local surface formed by the vertex and its adjacent vertices. The normal vector and curvature together constitute the local geometric descriptor of the vertex.

[0095] The normal vector of a vertex is calculated using the normal vectors of the triangular faces adjacent to the vertex. The expression is:

[0096] ;

[0097] In the formula, Represents vertices The normal vector, Represents the vertex index. Represents vertices An adjacent triangular facet, Represents vertices The set of all adjacent triangular faces, Represents a triangular facet The normal vector, This represents the magnitude of the vector.

[0098] The curvature of a vertex is calculated by fitting a local surface formed by the vertex and its adjacent vertices. The expression is as follows:

[0099] ;

[0100] In the formula, Represents vertices curvature, Represents vertices The number of vertices in the set of adjacent vertices. Represents vertices one adjacent vertex, denotes an adjacent vertex a position coordinate in three-dimensional space, denotes a vertex a position coordinate in three-dimensional space, denotes a vertex the Euclidean distance between the adjacent vertex

[0101] S4.5: input the graph structure into the lightweight multi-scale feature extraction architecture, and dynamically assign the optimal feature extraction path for each vertex according to the local geometric descriptor of each vertex by the built-in routing strategy network;

[0102] Specifically, the graph structure is combined with the lightweight multi-scale feature extraction architecture, and the built-in routing strategy network receives the local geometric descriptor of each vertex; the routing strategy network assigns a path identifier to each vertex based on the similarity of the local geometric descriptor, and the path identifier corresponds to different neural network branches (such as shallow path or deep path) in the lightweight multi-scale feature extraction architecture, thereby realizing dynamic assignment of the optimal feature extraction path for each vertex.

[0103] S4.6: the lightweight feature extraction subnetwork extracts the rock mass structure surface geometric parameters and micro-deformation vector along the assigned optimal feature extraction path;

[0104] Specifically, for each vertex, the lightweight feature extraction subnetwork selects the corresponding neural network branch according to the path identifier assigned by the routing strategy network; based on the normal vector in the local geometric descriptor of the vertex, the distance relationship between the vertices and the local surface variation degree described by the vertex curvature, the rock mass structure surface geometric parameters are obtained; the micro-deformation vector is obtained by comparing the three-dimensional coordinate changes of the same vertex at different time points.

[0105] S4.7: aggregate the rock mass structure surface geometric parameters and micro-deformation vectors extracted for all vertices according to the topological connection relationship of the graph structure, and output the structured feature information.

[0106] Specifically, according to the topological connection relationship of the graph structure, the graph pooling operation is used to aggregate the rock mass structure surface geometric parameters and micro-deformation vectors extracted for all vertices, and the graph pooling operation (such as average pooling) combines the rock mass structure surface geometric parameters and micro-deformation vectors of each vertex to generate global rock mass structure surface geometric parameter summary and micro-deformation vector summary; the global summary is organized in table form to form structured feature information.

[0107] S4.8: the rock mass structure surface geometric parameters include the dip direction, dip angle, spacing, trace length and roughness of the rock mass structure surface.

[0108] ​It should be noted that the tendency of the rock mass structure surface refers to the extension direction of the structure surface in the horizontal plane;

[0109] The inclination of the rock mass structure surface refers to the maximum angle between the structure surface and the horizontal plane;

[0110] The spacing of the rock mass structure surface refers to the vertical distance between adjacent structure surfaces;

[0111] The trace length of the rock mass structure surface refers to the visible length of the structure surface on the outcrop or excavation surface;

[0112] The roughness of the rock mass structure surface refers to the degree of unevenness of the structure surface.

[0113] S5, based on the structured feature information, real-time solving the instantaneous safety factor of the mine slope under the current constraint condition.

[0114] S5.1: The current constraint condition includes the field geological condition of the mine slope, the real-time monitored external load condition and the engineering state condition.

[0115] It should be noted that the field geological condition of the mine slope refers to the inherent geological properties of the rock mass type, geological structure characteristics and groundwater level of the area where the slope is located;

[0116] The real-time monitored external load condition refers to the external forces such as seismic load, rainfall load and blasting vibration load obtained by sensors in real time;

[0117] The engineering state condition refers to the construction and maintenance information such as the excavation stage of the current slope and the working state of the supporting structure.

[0118] S5.2: Real-time analysis of structured feature information, extract the tendency, inclination, spacing, trace length, roughness and micro-deformation vector of the rock mass structure surface at the current time, and generate a multi-dimensional input vector combined with the current constraint condition;

[0119] Specifically, read the specific values of the tendency, inclination, spacing, trace length, roughness and micro-deformation vector of the rock mass structure surface at the current time; At the same time, obtain the parameters of the current constraint condition, including the rock mass type and groundwater level of the field geological condition, the seismic acceleration and rainfall intensity and blasting vibration velocity of the real-time monitored external load condition, and the excavation stage and support state of the engineering state condition; Arrange all parameters in a fixed order and combine them into a multi-dimensional input vector.

[0120] It should be noted that the fixed order refers to the pre-defined arrangement order that all parameters must strictly follow.

[0121] S5.3: Parallel input multi-dimensional input vector to different types of safety factor prediction model, calculate safety factor prediction value, and calculate the fusion weight of different types of safety factor prediction model through dynamic weighting algorithm;

[0122] It should be noted that different types of safety factor prediction model include limit equilibrium method model, finite element method model and discrete element method model;

[0123] Parallel input multi-dimensional input vector to different types of safety factor prediction model, calculate safety factor prediction value, expression is:

[0124] ;

[0125] In the formula, Safety factor prediction value of the i-th safety factor prediction model, Safety factor prediction value of the i-th safety factor prediction model, Regression intercept term (also known as constant term) of the i-th safety factor prediction model, obtained by fitting historical data of the model, Total dimension of multi-dimensional input vector, Dimension index of multi-dimensional input vector, Regression coefficient of the i-th safety factor prediction model corresponding to the j-th dimension, Value of the j-th dimension in multi-dimensional input vector. It should be noted that the regression coefficient of the i-th safety factor prediction model corresponding to the j-th dimension is derived from the fitting of historical data, and the example value is 0.5. The value is optimized from historical data through regression analysis (such as least squares method) to minimize prediction error. Through dynamic weighting algorithm, the fusion weight of different types of safety factor prediction model is calculated, and the expression is:

[0126] ; In the formula,

[0127] Fusion weight of the i-th safety factor prediction model, Historical prediction accuracy (obtained by fitting historical data) of the i-th safety factor prediction model in the loop summation,

[0128] Applicability score of the i-th safety factor prediction model to the current condition in the loop summation,

[0129] ​​​​​​​​​a total number of types of safety factor prediction models, an index of a safety factor prediction model at the time of cyclic summation, a historical prediction accuracy of the safety factor prediction model at the time of cyclic summation, an applicability score of the safety factor prediction model at the time of cyclic summation to the current condition.

[0130] It should be noted that the applicability score of the safety factor prediction model at the time of cyclic summation to the current condition is derived from a pre-defined applicability rule base, which summarizes the performance of different models under different combinations of working condition parameters based on historical data. By matching specific parameters (such as rock type, seismic acceleration value) in the current multi-dimensional input vector with the conditions in the applicability rule base, the applicability score of each model is determined.

[0131] S5.4: The safety factor prediction values of different types of safety factor prediction models are weighted and fused using the fusion weight, and the instantaneous safety factor is output, expressed as:

[0132] ;

[0133] In the formula, represents the instantaneous safety factor.

[0134] S6, taking the instantaneous safety factor and the structured feature information as the reference, the maximum design safety factor value is determined by analyzing the influence on the slope stability under the preset different working condition combinations.

[0135] S6.1: Taking the instantaneous safety factor and the structured feature information as the reference, a multi-objective optimization algorithm is used to perform parallel search under the preset different working condition combinations to obtain an optimal solution set;

[0136] It should be noted that the preset different working condition combinations refer to a series of virtual scenarios for evaluating the stability of the slope formed by arranging and combining the specific parameters based on the on-site geological conditions of the mine slope, the real-time monitored external load conditions, and the engineering state conditions;

[0137] For example, the preset working condition combinations can include: taking "rock type as granite, seismic acceleration as 0.1, rainfall intensity as 50, and excavation stage as the third level" as one combination, and taking "rock type as sedimentary rock, seismic acceleration as 0.05, rainfall intensity as 20, and excavation stage as the fifth level" as another combination, which together constitute a working condition set for comprehensive evaluation.

[0138] Specifically, taking the instantaneous safety factor and the structural feature information as the basis, a multi-objective optimization relationship is established, the instantaneous safety factor is taken as the safety performance target, the geometric parameters of the rock mass structure surface in the structural feature information are taken as the target of improving the quality of the rock mass structure, and the micro-deformation vector in the structural feature information is taken as the target of controlling deformation; a multi-objective optimization algorithm (such as a genetic algorithm) is used to perform parallel search under different preset working condition combinations; the instantaneous safety factor and the structural feature information under each working condition combination are evaluated to generate multiple candidate solutions, and the non-dominated solutions, that is, the solutions that do not exist other solutions that are better in all targets, are screened out to form an optimal solution set.

[0139] S6.2: Multi-criteria decision analysis is performed on the optimal solution set to determine the maximum design safety factor value.

[0140] Specifically, multi-criteria decision analysis is performed on the optimal solution set, the instantaneous safety factor is taken as the safety performance criterion, the geometric parameters of the rock mass structure surface are taken as the rock mass structure quality criterion, and the micro-deformation vector is taken as the deformation control criterion to establish a decision framework; in the decision process, the priority levels of the criteria are comprehensively considered (for example, the safety performance is placed in the highest priority level, and the requirements of rock mass structure quality and deformation control are also considered), the performance of each solution in the optimal solution set in terms of safety performance, rock mass structure quality and deformation control is reviewed one by one, and a horizontal comparison and trade-off are performed to identify the solution that achieves the best balance among multiple criteria; based on the instantaneous safety factor value of the solution that achieves the best balance, the maximum design safety factor value is determined, and it is ensured that the maximum design safety factor value meets the comprehensive safety and stability requirements under the preset working condition combination.

[0141] In summary, the application improves the fusion quality of point cloud and image data at the edge computing node by starting the multi-modal fusion mechanism and the self-calibration mechanism; the heterogeneous feature extraction network learns the geometric features of the point cloud and the texture features of the image respectively, and the iterative optimization algorithm is used to realize accurate alignment of the feature space, effectively overcoming the inconsistency problem caused by sensor differences and external interference; the self-calibration mechanism can dynamically detect and correct registration errors in the fusion process, ensuring that the output point cloud-image fusion data block has both millimeter-level geometric accuracy and rich semantic information, providing reliable input for three-dimensional reconstruction and ensuring the accuracy and reliability of the instantaneous safety factor calculation result, and providing a solid data foundation for slope stability decision-making.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application and not to limit it, although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and they should be covered in the scope of the claims of the application.

Claims

1. An edge computing-based method for determining the maximum design safety factor of a mine slope, characterized in that: The application relates to a method for real-time monitoring of the stability of a mine slope. The method comprises the following steps: high-resolution three-dimensional laser point cloud data and image data of a slope surface are acquired, time synchronization, geometric correction and tile slicing are performed, and point cloud block sequences and orthographic image sequences are formed; the point cloud block sequences and the orthographic image sequences are input into an edge computing node, a multi-modal fusion mechanism and a self-calibration mechanism are started, and point cloud-image fusion data blocks are output; three-dimensional reconstruction and information fusion are performed on the point cloud-image fusion data blocks, and smooth splicing is performed on the basis of overlapping areas between adjacent data blocks to generate a three-dimensional geological model; a lightweight multi-scale feature extraction architecture is constructed and configured, the three-dimensional geological model is input into the lightweight multi-scale feature extraction architecture, a calculation path is dynamically assigned, rock mass structural surface geometric parameters and micro-deformation vectors are extracted, and structured feature information is output; based on the structured feature information, an instantaneous safety factor of a mine slope under current constraint conditions is solved in real time; 2. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: the instantaneous safety factor and the structured feature information are taken as a reference, the influence of different preset working condition combinations on the stability of the slope is analyzed, and a maximum design safety factor value is determined. The point cloud block sequences and the orthographic image sequences are formed through the following specific steps: a precise clock synchronization protocol is used to add sub-microsecond-precision time stamps to the high-resolution three-dimensional laser point cloud data and the image data of the slope surface, and time-synchronized multi-modal data streams are generated; initial poses provided by inertial navigation are used as constraints to perform multi-modal feature correlation on the time-synchronized multi-modal data streams, a pose graph is constructed for joint optimization solving, and point clouds and orthographic images with geometric position accurate alignment are obtained; the point clouds and the orthographic images with geometric position accurate alignment are subjected to semantic segmentation, and a geological feature difference area graph is generated; 3. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: the point clouds and the orthographic images with geometric position accurate alignment are subjected to non-uniform grid division according to the geological feature difference area graph, and point cloud block sequences and orthographic image sequences are formed. The point cloud-image fusion data blocks are output through the following specific steps: feature mapping is performed on the point cloud block sequences and the orthographic image sequences by a heterogeneous feature extraction network, point cloud geometric feature vectors and image texture feature vectors are generated, and an uncertainty estimation value is assigned to each feature vector; weighted average is performed on the point cloud geometric feature vectors and the image texture feature vectors, and an initial fusion feature vector is generated; based on the uncertainty estimation values of the feature vectors and the cosine similarity between the feature vectors and the initial fusion feature vector, dynamic weight distribution is performed, and iterative fusion and self-calibration refinement are performed until the initial fusion feature vector meets a convergence condition, a refined fusion feature vector and an uncertainty measure are output; 4. The edge-computing-based mine slope maximum design safety factor value determination method according to claim 1, characterized in that: the refined fusion feature vector and the uncertainty measure are input into a deconvolution decoder network, the refined fusion feature vector is weighted and modulated by using the uncertainty measure, and a point cloud-image fusion data block is generated. The three-dimensional geological model is generated through the following specific steps: three-dimensional reconstruction is performed on each point cloud-image fusion data block by a local implicit neural representation network, and a local implicit surface model corresponding to each data block is generated. For the overlapping region of adjacent data blocks, the uncertainty measure of each data block is used to assign dynamic weights to the local implicit surface model corresponding to each data block, and the global implicit surface expression is generated by weighted fusion; The global implicit surface expression is extracted and texture mapped to generate a three-dimensional geological model.

5. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: The specific steps of constructing and configuring the lightweight multi-scale feature extraction architecture are as follows, A macro-architecture search space is constructed on the edge computing node, and a meta-controller with a recurrent neural network as the kernel is constructed according to the pre-defined architecture coding specification; Global context descriptors of the three-dimensional geological model are extracted to input the meta-controller to generate subnet architecture parameters, and the subnet architecture parameters are instantiated into a lightweight feature extraction subnet from the macro-architecture search space; The performance of the lightweight feature extraction subnet is evaluated through a context-aware evaluation mechanism to generate an evaluation result, and the meta-controller is updated based on the evaluation result to complete the construction and configuration of the lightweight multi-scale feature extraction architecture.

6. The edge-computing-based mine slope maximum design safety factor value determination method according to claim 1, characterized in that: The specific steps of outputting the structured feature information are as follows, The triangular mesh surface of the three-dimensional geological model is converted into a graph structure, and the local geometric descriptors of each vertex in the graph structure are calculated; The graph structure is input into the lightweight multi-scale feature extraction architecture, and the built-in routing strategy network dynamically assigns the optimal feature extraction path for each vertex based on the local geometric descriptors of each vertex; The lightweight feature extraction subnet extracts the rock mass structure surface geometric parameters and micro-deformation vectors along the assigned optimal feature extraction path; According to the topological connection relationship of the graph structure, the rock mass structure surface geometric parameters and micro-deformation vectors extracted from all vertices are aggregated to output the structured feature information.

7. The edge-computing-based mine slope maximum design safety factor value determination method according to claim 6, characterized in that: The rock mass structure surface geometric parameters include the dip direction, dip angle, spacing, trace length and roughness of the rock mass structure surface.

8. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: The current constraint conditions include the field geological conditions of the mine slope, the real-time monitored external load conditions and the engineering state conditions.

9. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: The specific steps of real-time calculating the instantaneous safety factor of the mine slope under the current constraint conditions are as follows, Real-time analysis of the structured feature information extracts the dip direction, dip angle, spacing, trace length, roughness and micro-deformation vector of the rock mass structure surface at the current time, and generates a multi-dimensional input vector in combination with the current constraint conditions; The multi-dimensional input vector is input into different types of safety factor prediction models in parallel, the safety factor prediction values are calculated, and the fusion weights of different types of safety factor prediction models are calculated through a dynamic weighting algorithm; The safety factor prediction values of different types of safety factor prediction models are weighted and fused using the fusion weights to output the instantaneous safety factor.

10. The edge-computing-based method for determining the maximum design safety factor of a mine slope according to claim 1, characterized in that: The specific steps of determining the maximum design safety factor value are as follows, Taking the instantaneous safety factor and the structured feature information as the benchmark, a multi-objective optimization algorithm is used to perform parallel search under different preset working condition combinations to obtain an optimal solution set; Multi-criteria decision analysis is performed on the optimal solution set to determine the maximum design safety factor value.

Images