Roadside slope stability image monitoring and risk warning system

Abstract

The present application relates to the field of highway engineering safety monitoring, in particular to a highway slope stability image monitoring and risk early warning system, the system comprises an image acquisition module, a slope parameterization representation module, a multi-scale curvature flow analysis module, a Riemann manifold learning module and a risk early warning module, the system regards the slope as a Riemann manifold, extracts differential features by calculating the metric tensor and the curvature tensor of the surface; the multi-scale representation of the slope surface is constructed by applying the curvature flow theory, and the deformation characteristics under different scales are identified; senior features are extracted by using Riemann manifold learning and geodesic convolution network, and precursors of geological disasters such as landslides, collapses and debris flows are accurately identified, the system has an adaptive monitoring frequency adjustment function, automatically switches the monitoring mode according to the risk level, greatly reduces the slope disaster risk, and provides a strong guarantee for highway safety operation.

Description

Technical Field

[0001] This invention relates to the field of highway engineering safety monitoring, specifically to a highway slope stability image monitoring and risk early warning system, which is used for real-time monitoring of highway slopes and early warning of potential geological disasters. Background Technology

[0002] Slope stability is crucial for the safe operation of highways. Slope instability leading to landslides, collapses, and debris flows not only threatens lives and property but also causes traffic disruptions and economic losses. Traditional slope monitoring methods rely primarily on manual inspections and single-point measurement techniques, such as displacement gauges, inclinometers, and stress gauges, which suffer from limited coverage, poor real-time performance, and low automation. In recent years, with the development of image processing and artificial intelligence technologies, image-based slope monitoring methods have been gradually applied, but the following technical challenges remain:

[0003] 1. Most existing technologies are based on Euclidean geometry and linear analysis methods, which are difficult to effectively characterize the complex nonlinear deformation process of slopes;

[0004] 2. Traditional methods do not adequately consider the multi-scale characteristics of slope disaster precursors and struggle to distinguish between environmental noise and actual deformation;

[0005] 3. Existing pattern recognition methods are ineffective in handling the nonlinear manifold structure of slope features, affecting the accuracy of early warning;

[0006] 4. The field environment is complex and ever-changing, and existing systems face engineering challenges such as power supply, communication and equipment stability during long-term monitoring.

[0007] Therefore, there is an urgent need to develop a slope monitoring system that can comprehensively capture slope deformation characteristics, accurately identify disaster precursors, and provide reliable early warnings. Summary of the Invention

[0008] The purpose of this invention is to provide a highway slope stability image monitoring and risk early warning system, which deeply integrates differential geometry theory with slope monitoring technology to achieve accurate identification and risk early warning of slope disaster precursors.

[0009] This invention proposes a highway slope stability image monitoring and risk early warning system, comprising:

[0010] The image acquisition module is used to acquire high-resolution image data and 3D point cloud data of highway slopes;

[0011] The slope parameterization representation module is communicatively connected to the image acquisition module and is used to represent the highway slope as a two-dimensional Riemannian manifold and calculate the metric tensor and curvature tensor of the slope surface.

[0012] The multi-scale curvature flow analysis module is communicatively connected to the slope parameterization characterization module. It is used to receive the metric tensor and curvature tensor of the slope surface, construct a multi-scale representation of the slope surface, and extract multi-scale feature maps.

[0013] The Riemannian manifold learning module, which is communicatively connected to the multi-scale curvature flow analysis module, is used to represent slope features in the Riemannian manifold space, extract high-level features through geodesic convolutional networks, and identify precursors of disasters such as landslides, collapses, and debris flows.

[0014] The risk warning module is communicatively connected to the Riemannian manifold learning module and is used to generate risk level assessments and warning information based on the identified disaster precursors.

[0015] Preferably, the image acquisition module includes:

[0016] A high-resolution camera array is used to acquire image data of highway slopes;

[0017] Three-dimensional LiDAR is used to acquire point cloud data of highway slopes;

[0018] GPS positioning system is used to obtain spatial location information;

[0019] The data preprocessing unit is used to perform noise reduction, registration and fusion processing on the image data and point cloud data to generate a textured 3D slope model.

[0020] Preferably, the slope parametric characterization module includes:

[0021] Parametric mapping units are used to construct the mapping relationship from the parameter domain to the three-dimensional coordinates of the slope;

[0022] The differential property calculation unit is used to calculate the metric tensor, curvature tensor, principal curvature, and principal direction of the slope surface;

[0023] Differential invariant generation unit, used to calculate the Gaussian curvature, mean curvature, and shape index of the slope surface;

[0024] The time rate of change analysis unit is used to calculate the rate of change of the differential property over time and to mark abnormal regions where the rate of change exceeds a preset threshold.

[0025] Preferably, the multi-scale curvature flow analysis module includes:

[0026] Curvature flow evolution unit, used to construct multi-scale representations of slope surfaces;

[0027] Feature operator units are used to apply differential feature operators at different scales to extract local features;

[0028] The feature map generation unit is used to organize features at different scales into feature maps.

[0029] Scale-space analysis units are used to analyze the evolution of features in scale space and identify feature patterns related to disaster precursors.

[0030] Preferably, the Riemannian manifold learning module includes:

[0031] Manifold building blocks are used to define appropriate Riemannian manifolds based on the inherent structure of slope characteristics;

[0032] Geodesic convolutional network units are used to perform convolution operations on Riemannian manifolds to extract high-level features;

[0033] Disaster precursor classification unit, used to identify different types of disaster precursors;

[0034] The risk probability estimation unit is used to calculate the probability distribution and spatiotemporal risk map of different disaster types.

[0035] Preferably, the risk warning module includes:

[0036] Risk assessment unit, used to assess risk level based on disaster precursor identification results;

[0037] The early warning information generation unit is used to generate early warning information that includes risk type, location, level, and handling recommendations;

[0038] A multi-channel notification unit is used to send early warning information via SMS, email, and monitoring platforms;

[0039] The response strategy recommendation unit is used to recommend appropriate response measures based on the risk level.

[0040] Preferably, the system also includes a monitoring mode control module, which is used to automatically adjust the monitoring frequency and analysis depth according to the slope risk status, including a regular monitoring mode, a warning monitoring mode and an emergency monitoring mode.

[0041] Preferably, the geodesic convolutional network unit includes:

[0042] Geodesic convolutional layers are used to perform convolution operations on Riemannian manifolds;

[0043] The tangent space pooling layer is used to perform feature aggregation and dimensionality reduction in the tangent space.

[0044] The parallel transport layer is used to align and fuse features from different layers through parallel transport.

[0045] The manifold batch normalization layer is used to perform batch normalization processing adapted to Riemannian geometry.

[0046] Preferably, the curvature flow evolution unit includes:

[0047] A flow evolution equation solver is used to solve the curvature flow evolution equations for the smoothing process of control surfaces at different scales.

[0048] Feature-preserving smoother is used to preserve key features during the smoothing process;

[0049] A scale space builder for generating sequences of surface representations covering different feature scales;

[0050] Critical point tracker is used to track the evolution trajectory of feature points during the evolution of curvature flow.

[0051] Preferably, the system also includes a data management module for storing historical monitoring data, model parameters, and early warning records, and providing data query, analysis, and visualization functions; the data management module establishes communication connections with the image acquisition module, the slope parametric characterization module, the multi-scale curvature flow analysis module, the Riemannian manifold learning module, and the risk early warning module for data sharing and collaborative analysis.

[0052] This invention employs an innovative method for characterizing slope surface features based on differential geometry, multi-scale curvature flow analysis technology, and a disaster precursor identification model based on Riemannian manifold learning. Combined with a hardware design that is highly adaptable to the field, it has the following beneficial effects:

[0053] 1. By treating the slope as a Riemannian manifold, this invention can comprehensively capture the complex nonlinear deformation characteristics of the slope, improving the accuracy and comprehensiveness of monitoring;

[0054] 2. Based on multi-scale curvature flow analysis technology, this invention can effectively distinguish between environmental noise and real disaster precursors, significantly improving the accuracy and reliability of early warning;

[0055] 3. By employing the Riemannian manifold learning method, this invention can adapt to the nonlinear distribution of slope characteristics, thereby improving the accuracy of disaster precursor identification;

[0056] 4. The system design fully considers the complexity of the field environment and achieves long-term stable monitoring through solar power supply, wireless data transmission and modular design.

[0057] This invention enables early warning of landslide precursors in 3-7 days, early warning of collapse precursors in 1-3 days, and early warning of debris flow precursors in 12-24 hours, which is of great value in ensuring the safe operation of highways and reducing casualties and property losses. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention;

[0059] Figure 2 This is a schematic diagram of the image acquisition module of the present invention;

[0060] Figure 3 This is a flowchart of the slope parametric characterization module of the present invention;

[0061] Figure 4 This is a flowchart of the multi-scale curvature flow analysis module of the present invention;

[0062] Figure 5 This is a network structure diagram of the Riemannian manifold learning module of the present invention;

[0063] Figure 6 This is a flowchart of the risk warning module of the present invention;

[0064] Figure 7 This is a schematic diagram of the structure of the geodesic convolutional network unit of the present invention;

[0065] Figure 8 This is a schematic diagram of the curvature flow evolution unit of the present invention. Detailed Implementation

[0066] Please refer to Figures 1-8 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0067] Reference Figure 1 The highway slope stability image monitoring and risk early warning system provided by this invention includes an image acquisition module 1, a slope parametric characterization module 2, a multi-scale curvature flow analysis module 3, a Riemannian manifold learning module 4, a risk early warning module 5, a monitoring mode control module 6, and a data management module 7. The modules are connected via a communication interface to form a complete data processing link.

[0068] In a preferred embodiment of the present invention, the system adopts a layered architecture design, divided from bottom to top into a data acquisition layer, a data processing layer, and an application service layer. The data acquisition layer is mainly handled by the image acquisition module 1, the data processing layer includes a slope parametric characterization module 2, a multi-scale curvature flow analysis module 3, and a Riemannian manifold learning module 4, and the application service layer consists of a risk warning module 5, a monitoring mode control module 6, and a data management module 7. This layered design gives the system good scalability and maintainability.

[0069] Reference Figure 2 The image acquisition module 1 of the present invention includes a high-resolution camera array 11, a three-dimensional lidar 12, a GPS positioning system 13, and a data preprocessing unit 14.

[0070] A high-resolution camera array 11 is used to acquire image data of the highway slope. In one embodiment of the invention, the high-resolution camera array employs multiple industrial cameras with resolutions of 20 megapixels or higher, arranged along different directions of the slope to form an observation network covering the entire slope. The sampling frequency of the cameras is automatically adjusted according to the monitoring mode: once every 12 hours in normal monitoring mode, once every hour in alert monitoring mode, and once every 10 minutes in emergency monitoring mode.

[0071] A 3D LiDAR 12 is used to acquire point cloud data of the highway slope. Preferably, the 3D LiDAR uses a device with a point density of not less than 100 points / square meter and an accuracy better than 5 millimeters to ensure the acquisition of fine structure of the slope surface. The scanning frequency of the LiDAR is synchronized with the sampling frequency of the camera to ensure temporal consistency between the point cloud data and the image data.

[0072] The GPS positioning system 13 is used to acquire spatial location information. This invention employs a differential GPS system with centimeter-level accuracy to provide precise geographic reference for image and point cloud data. The GPS system is not only used for spatial registration of data but also for monitoring the device's own positional stability, ensuring long-term spatial consistency.

[0073] The data preprocessing unit 14 is used to perform noise reduction, registration, and fusion processing on the image data and point cloud data to generate a textured 3D slope model. Specifically, the data preprocessing includes the following steps:

[0074] Data cleaning: Remove outliers from point clouds using statistical filtering, and remove noise from images using morphological processing;

[0075] Coordinate transformation: Converting data from different sources to a unified coordinate system;

[0076] Time alignment: Time alignment of data collected at different times;

[0077] Data fusion: fusing point cloud data with image data to generate textured 3D models;

[0078] Region segmentation: Extract the slope area of ​​interest and remove the background and irrelevant areas.

[0079] During data preprocessing, special attention must be paid to the normalization of different data sources. For example, for point cloud data, the spatial coordinates need to be normalized to the [-1,1] interval according to the actual measurement range; for image data, brightness correction and color balance are required to ensure that images collected at different times and under different weather conditions are comparable. In addition, during data fusion, control point matching is needed to ensure accurate registration of point clouds and images, with errors controlled at the pixel level.

[0080] Reference Figure 3The slope parameterization characterization module 2 of the present invention includes a parameterization mapping unit 21, a differential characteristic calculation unit 22, a differential invariant generation unit 23, and a time rate of change analysis unit 24.

[0081] The parameterized mapping unit 21 is used to construct the mapping relationship from the parameter domain to the three-dimensional coordinates of the slope. In this invention, the slope is regarded as a two-dimensional Riemannian manifold embedded in three-dimensional Euclidean space. By establishing a parameterized representation system, the foundation for subsequent differential geometric analysis is laid.

[0082] Specifically, the parameterization mapping unit 21 performs the following operations:

[0083] Construct a local coordinate system, with the center of the slope as the origin, and establish a local rectangular coordinate system;

[0084] The slope area is divided into grid-like units, with each unit preferably measuring 0.5 meters × 0.5 meters.

[0085] Construct a parameter mapping, defining the mapping relationship from the parameter domain (u,v) to the actual three-dimensional coordinates (x,y,z) of the slope;

[0086] Adding a time dimension creates a spatiotemporal parameterized representation, capturing the dynamic characteristics of slope changes over time.

[0087] The parameter domain is selected using a regular parameter domain to ensure a good correspondence between the parameter space and the physical space. The grid resolution can be adjusted according to the monitoring accuracy requirements, generally within the range of 0.1 to 1 meter, with a default value of 0.5 meters. This value achieves a good balance between accuracy and computational efficiency.

[0088] The differential property calculation unit 22 is used to calculate the metric tensor, curvature tensor, principal curvature, and principal direction of the slope surface. These differential properties are fundamental quantities describing the local geometric properties of the surface and are crucial for identifying slope deformation.

[0089] In the calculation of differential properties, the first fundamental form of the surface (metric tensor) is calculated first. :

[0090] .

[0091] in: A vector function representing a parametric surface maps the parameters (u,v) to points (x,y,z) in three-dimensional space; u and v are the parametric coordinates; E, F, and G are the components of the metric tensor, representing the metric relationships between the parametric curves; "." indicates a vector dot product operation. The metric tensor is a fundamental tool for measuring distances and areas on a surface.

[0092] Next, calculate the second fundamental form of the surface (curvature tensor) b:

[0093] .

[0094] in: It is the unit normal vector of the surface, calculated as follows: , where × represents the vector cross product operation; L, M, and N are the components of the curvature tensor, which describe the degree of curvature of the surface.

[0095] Then, by solving the characteristic equation, the principal curvature is calculated. and And the corresponding main direction:

[0096] .

[0097] in: `det` represents the principal curvature; `det` represents the determinant operation of a matrix. Principal curvature and These are the two roots of the aforementioned characteristic equation, representing the maximum and minimum curvature of the surface at a certain point along a specific direction, and are important indicators characterizing the local shape of the surface.

[0098] In the calculation process, a high-order central difference scheme is used to ensure the accuracy of numerical differentiation. Singularities that may occur in curvature calculation are regularized, and natural boundary conditions are used to avoid the influence of boundary effects on the calculation results.

[0099] Differential invariant generation unit 23 is used to calculate the Gaussian curvature, mean curvature, and shape index of the slope surface. These differential invariants are intrinsic properties of the surface, unaffected by the parameterization method, and are stable indices characterizing the surface features.

[0100] Gaussian curvature It is the product of the two principal curvatures:

[0101] .

[0102] in: and L, M, and N are the principal curvatures; L, M, and N are the components of the curvature tensor; E, F, and G are the components of the metric tensor. Gaussian curvature is a built-in property of a surface, representing the total curvature of the surface.

[0103] The mean curvature H is the average of the two principal curvatures:

[0104] .

[0105] The meanings of the symbols are the same as above. Mean curvature represents the average degree of curvature of the surface.

[0106] Shape Index It is a dimensionless parameter based on the principal curvature, used to characterize the local shape type of a surface:

[0107] .

[0108] Where π is the mathematical constant pi; arctan is the arctangent function; ≥ express Not less than The shape index ranges from [-1, 1], with different values ​​corresponding to different surface types: S=-1 represents a spherical concave surface, S=-0.5 represents a groove shape, S=0 represents a saddle surface, S=0.5 represents a ridge shape, and S=1 represents a spherical convex surface.

[0109] The time-rate-of-change analysis unit 24 is used to calculate the rate of change of the differential characteristics over time and to mark abnormal areas where the rate of change exceeds a preset threshold. This step is a crucial step in identifying abnormal slope deformation.

[0110] Calculating the rate of change over time requires first precisely aligning the curvature data acquired at different times, and then calculating the time derivatives of each differential characteristic. For Gaussian curvature, the rate of change ΔK is calculated as follows:

[0111] .

[0112] Where: K(u,v,t) represents the Gaussian curvature value of the parameter coordinates (u,v) at time t; Δt is the sampling time interval in days. Similarly, the average rate of change of curvature ΔH and the rate of change of shape exponent ΔS can also be calculated using the same method.

[0113] Based on empirical data and experimental results, this invention sets warning thresholds for the rate of change of different slope types: for sandy slopes, the threshold for the rate of change of Gaussian curvature is 0.05-0.1 / day; for rocky slopes, the threshold is 0.02-0.05 / day. When an area with a rate of change exceeding the threshold is detected, the system marks it as a potential abnormal area and proceeds to the next step of analysis.

[0114] The time window length is automatically adjusted according to the monitoring mode: 7 days in standard mode and 3 days in accelerated deformation mode. Baseline selection can use either a fixed baseline or a sliding baseline mode; the sliding baseline mode is more adaptable to seasonal changes.

[0115] Reference Figure 4 The multi-scale curvature flow analysis module 3 of the present invention includes a curvature flow evolution unit 31, a feature operator unit 32, a feature map generation unit 33, and a scale space analysis unit 34.

[0116] The curvature flow evolution unit 31 is used to construct a multi-scale representation of the slope surface. In this invention, the curvature flow theory in differential geometry is applied to generate a series of surface representations at different scales by solving the curvature flow evolution equation, thereby capturing the slope characteristics at different scales.

[0117] Curvature flow refers to the process by which a curved surface evolves along its normal direction with a velocity proportional to its curvature. In the simplest case, mean curvature flow can be used, and its evolution equation is:

[0118] .

[0119] in: It is a surface position vector, representing a point on the parametric surface; It is a scale parameter that controls the degree of smoothness; It is the mean curvature; It is the unit normal vector. This equation describes the surface as a function of the scale parameter. A process that gradually increases and then becomes smoother.

[0120] In practical applications, to preserve key features, this invention employs feature-preserving curvature flow, whose evolution equation is:

[0121] .

[0122] in: Based on Gaussian curvature The modulation function is used to preserve key features during the smoothing process. A typical modulation function has the following form:

[0123] .

[0124] in: These are control parameters used to adjust the degree of preservation of areas with high curvature. Represents Gaussian curvature The absolute value of. Preferably, The value ranges from 1 to 10, with larger values ​​being... The value will more strongly preserve the high curvature characteristics.

[0125] In the curvature flow evolution process, this invention sets 5 to 8 scale levels, covering a feature scale of 0.5 to 10 meters. The evolution step size is set to 0.01-0.05 according to numerical stability requirements. The evolution terminates when the rate of curvature change is less than a preset threshold (usually 0.001) or when the maximum number of iterations (usually 100) is reached.

[0126] Feature operator unit 32 is used to apply differential feature operators at different scales to extract local features. At each scale, a series of feature operators are applied to extract local features of the surface, including curvature operator, normal change operator, and shape change operator.

[0127] Curvature operators, based on curvature tensors, are used to extract the bending features of surfaces.

[0128] .

[0129] in: Represents the curvature operator; , and Representing the parameter coordinates respectively At scale Gaussian curvature, mean curvature, and shape index are given; square brackets indicate vector combinations.

[0130] The normal change operator, based on the rate of change of the normal vector, is used to detect abnormal changes in the surface normal:

[0131] .

[0132] in: Indicates the normal change operator; Normal vector scale The derivative; The Euclidean norm of a vector is denoted by .

[0133] Shape transformation operators, based on changes in shape exponents, are used to identify changes in the local shape of a surface.

[0134] .

[0135] in: Represents shape transformation operators; Represents shape index scale The derivative; It represents the absolute value.

[0136] The feature response threshold is automatically adjusted based on the noise level, and is typically set to 2-3 times the standard deviation. This means that regions with feature response values ​​exceeding the background noise by 2-3 standard deviations are considered valid features.

[0137] The feature map generation unit 33 is used to organize features at different scales into feature maps. A feature map is a multidimensional data structure that records the distribution of features in spatial and scale dimensions. Formally, a feature map can be represented as:

[0138] .

[0139] in: Representation of feature maps; These are parametric coordinates; It is a scale parameter; Indicates the type of feature (such as curvature, normal variation, or shape variation); It is the response value of the corresponding feature operator.

[0140] When constructing feature maps, the weights of features at different scales need to be allocated appropriately. Generally, larger-scale features have higher weights because they represent more significant structural changes. A typical weight allocation method is:

[0141] .

[0142] in: Representing scale The weights; Indicates the first Each scale value; It is the number of scale levels; Indicates from arrive The summation operation.

[0143] Scale-space analysis unit 34 is used to analyze the evolution of features in scale space and identify feature patterns related to disaster precursors. This step is crucial for extracting disaster precursor features from multi-scale features.

[0144] First, feature persistence analysis is performed to evaluate the persistence of features in scale space. Persistence refers to the ability of a feature to remain present across continuous scales, used to filter transient noise. The persistence metric can be defined as:

[0145] .

[0146] in: Representation of features A measure of persistence; It is an indicator function that takes the value 1 when the condition is true and 0 otherwise. Characteristic Operator In parameter coordinates and scale The response value at the location; It is the characteristic response threshold; It is the number of scale levels; Indicates all scales The summation operation is used. The persistence threshold is usually set to 0.6-0.8, meaning that a feature needs to remain significant in 60% to 80% of the scales to be considered persistent.

[0147] Then, the evolution path of features in scale space is extracted to form a feature spectrum. The feature spectrum describes the behavior of features as they change with scale and is an important basis for identifying specific patterns. The feature spectrum can be represented as:

[0148] .

[0149] in: Representation of features Characteristic spectrum; Characteristic Operator In parameter coordinates and scale The response value at the location; curly braces indicate a set.

[0150] Finally, the extracted feature spectra are matched with known disaster precursor patterns to identify potential disaster precursors. Pattern matching employs appropriate distance metrics in the feature space, such as geodesic distance or Mahalanobis distance. Simultaneously, anomalies that significantly deviate from normal evolutionary patterns are detected; these anomalies may be indicators of disaster precursors.

[0151] During pattern matching, the feature spectrum templates for different types of disasters are established based on historical cases and expert knowledge. For example, landslide precursors are usually manifested as a sustained increase in the average curvature of a local area; collapse precursors are manifested as a sudden change in the local Gaussian curvature; and debris flow precursors may manifest as a gradual change in the shape index of the gully area.

[0152] Reference Figure 5 The Riemannian manifold learning module 4 of the present invention includes a manifold construction unit 41, a geodesic convolutional network unit 42, a disaster precursor classification unit 43, and a risk probability estimation unit 44.

[0153] Manifold building unit 41 is used to define an appropriate Riemannian manifold based on the intrinsic structure of the slope features. In this invention, the slope features are considered as points distributed on the Riemannian manifold, and a mathematical basis for nonlinear feature learning is provided by defining an appropriate Riemannian metric.

[0154] Specifically, manifold building unit 41 performs the following operations:

[0155] Manifold structure definition: Define an appropriate Riemannian manifold based on the inherent structure of the slope characteristics;

[0156] Metric Tensor Design: Design Riemannian metric tensors that reflect feature similarity;

[0157] Geodesic calculation: Calculate the geodesics between points on the manifold as a distance metric;

[0158] Tangent space mapping: Establishes a mapping from the manifold to the local tangent space, which facilitates local analysis.

[0159] In this invention, the dimension of the Riemannian manifold is set according to the feature complexity, typically 3-8 dimensions. To prevent ill-conditioned cases, upper and lower bounds are set on the curvature of the manifold, typically [-10, 10]. For the selection of local coordinates, an intrinsic coordinate system that reflects the essence of the features is preferred.

[0160] Riemannian metric tensor It is a positive definite symmetric matrix at every point on the manifold, defining the distance measure on the manifold:

[0161] .

[0162] in: Represents the square of a small distance; Point Components of the Riemannian metric tensor; and Represents coordinate differential; This indicates all indicators and The summation operation.

[0163] In this invention, an adaptive Riemann metric is designed based on the characteristics of the feature distribution:

[0164] .

[0165] in: It is a point The Riemannian metric tensor at that location; It is the Jacobian matrix of the feature map, which reflects the local variation of the features; express The transpose of the matrix; This is a regularization parameter that prevents metric degradation; its value is typically between 0.01 and 0.1. It is an identity matrix, with dimensions equal to... same.

[0166] Geodesic convolutional network unit 42 is used to perform convolution operations on Riemannian manifolds to extract high-level features. Traditional convolutional neural network designs are based on Euclidean space and are difficult to apply directly to Riemannian manifolds. This invention designs a geodesic convolutional network based on Riemannian geometry, which is suitable for feature extraction requirements of non-Euclidean geometries.

[0167] The geodesic convolutional network unit 42 includes a geodesic convolutional layer, a tangent spatial pooling layer, a parallel transport layer, and a manifold batch normalization layer.

[0168] Geodesic convolution layers perform convolution operations on Riemannian manifolds. Unlike convolution in Euclidean space, convolution on Riemannian manifolds needs to consider curvature and geodesic distance. The mathematical expression for geodesic convolution is:

[0169] .

[0170] in: It is the input feature function, defined on the manifold superior; It is a convolution kernel function, defined on the tangent space; From point Time The inverse exponential mapping of maps points on the manifold to the tangent space; It is a volume element, in which Represents the metric tensor The determinant of; In manifold Integrals on; This represents a convolution operation on a manifold. In practical computation, geodesic convolution is approximated by defining the convolution kernel on the tangent space and using geodesic neighborhood sampling.

[0171] The tangent-space pooling layer performs feature aggregation and dimensionality reduction in the tangent space. For points on the manifold... First, map its neighborhood points to the tangent space. Then, standard pooling operations are performed in the tangent space. This method avoids the difficulties of performing pooling operations directly on the surface.

[0172] Parallel transport layers achieve the alignment and fusion of features from different layers through parallel transport. In Riemannian geometry, parallel transport is the operation of transporting vectors along geodesics while maintaining their parallelism. Parallel transport can be represented as:

[0173] .

[0174] in: Indicates from point Time Parallel transmission operation; It is a parallel transfer matrix, which describes how the tangent space is... Vectors are transferred in parallel to the tangent space. ; It is the vector to be transmitted, located in the tangent space. middle.

[0175] The manifold batch normalization layer performs batch normalization adapted to Riemannian geometry. Unlike batch normalization in Euclidean space, batch normalization on Riemannian manifolds needs to consider the local structure of the manifold. The form of manifold batch normalization is:

[0176] .

[0177] in: These are the features after normalization; These are input features; It is the mean calculated on the manifold; It is the variance calculated on the manifold; It is a small constant to prevent division by zero errors, and is usually set to a value of [value missing]. ; and These are learnable parameters; This represents a multiplication operation adapted to the manifold structure.

[0178] In network design, the depth of geodesic convolutional networks is set according to task complexity, typically 6-12 layers. The kernel size is defined in geodesic distance space, usually covering a local neighborhood, with a radius of 0.1-0.5 geodesic units. The learning rate adopts an adaptive strategy, with an initial value of 0.001, which is gradually reduced as training progresses.

[0179] The disaster precursor classification unit 43 is used to identify different types of disaster precursors. In this invention, the focus is mainly on the precursor characteristics of three common geological disasters: landslides, collapses, and debris flows.

[0180] The classification decision uses the softmax function to output the probabilities of different classes:

[0181] .

[0182] in: Indicates the given input When the output category is The probability of; It corresponds to the category Network output; express The exponential function; This represents the sum of exponential functions for all categories. The classification decision threshold is set according to risk preference. In this invention, priority is given to reducing the false negative rate. Therefore, the identification threshold is set to 0.3-0.4. That is, when the probability of a certain type of disaster exceeds this threshold, the system considers that there are precursors to that type of disaster.

[0183] The risk probability estimation unit 44 is used to calculate the probability distribution and spatiotemporal risk map for different disaster types. The risk probability distribution includes not only the probability of the disaster type, but also estimates of the disaster's occurrence time and intensity.

[0184] Time-based prediction constructs a time model of disaster development based on feature evolution rate:

[0185] .

[0186] in: It is the predicted time of the disaster; It is the current time; It is the threshold for characteristic changes that trigger disasters, determined based on historical data; These are the currently observed feature changes; It is the characteristic rate of change, that is, the ratio of the characteristic change to time.

[0187] The time prediction windows are set to 24 hours, 72 hours, and 7 days, corresponding to short-term, medium-term, and long-term predictions, respectively. The confidence level of the prediction decreases as the time window increases, and the system will provide confidence intervals for predictions within different time windows.

[0188] Spatial risk distribution is represented by generating risk heatmaps. These heatmaps visually display the location and extent of risk areas based on the spatial distribution and intensity of characteristic anomalies. Risk levels are divided into four categories: low risk (green), moderate risk (yellow), high risk (orange), and extremely high risk (red), corresponding to different early warning and response measures.

[0189] Reference Figure 6 The risk warning module 5 of the present invention includes a risk assessment unit 51, a warning information generation unit 52, a multi-channel notification unit 53, and a response strategy recommendation unit 54.

[0190] Risk assessment unit 51 is used to assess the risk level based on the results of disaster precursor identification. The risk assessment comprehensively considers factors such as disaster type, probability of occurrence, potential impact range, and severity, and provides a comprehensive risk assessment result.

[0191] Risk level calculation uses the risk matrix method, which maps the probability of disaster occurrence and the severity of potential impact onto the risk matrix:

[0192] .

[0193] Where: Risk represents the risk value; Probability represents the probability of a disaster occurring, divided into 5 levels (1 - very low, 2 - low, 3 - medium, 4 - high, 5 - very high); Severity represents the severity of a disaster, also divided into 5 levels (1 - minor, 2 - moderate, 3 - severe, 4 - very severe, 5 - catastrophic). The final risk level is divided into four levels based on the risk value: low risk (1-4), moderate risk (5-9), high risk (10-16), and very high risk (17-25).

[0194] The early warning information generation unit 52 is used to generate early warning information that includes risk type, location, level, and handling recommendations. The early warning information adopts a structured format and includes the following elements:

[0195] Warning Number: A unique identifier for the warning event;

[0196] Warning time: The time when the warning is issued;

[0197] Risk type: The type of disaster for which an early warning is issued (landslide, collapse, or debris flow).

[0198] Risk Location: Geographical description and coordinates of the risk area;

[0199] Risk level: Low risk, moderate risk, high risk, or extremely high risk;

[0200] Estimated time of occurrence: The predicted timeframe within which a disaster may occur;

[0201] Scope of impact: The area expected to be affected;

[0202] Recommended measures to address this risk.

[0203] The multi-channel notification unit 53 is used to send early warning information via SMS, email, and the monitoring platform. To ensure timely delivery of early warning information, the system adopts a multi-channel redundant sending mechanism, including SMS, email, dedicated app push notifications, and display on the monitoring platform. Different notification strategies correspond to different risk levels: low-risk levels are only displayed on the monitoring platform, general-risk levels also receive email notifications, and high-risk and extremely high-risk levels activate full-channel notifications, including SMS and telephone alerts.

[0204] The response strategy recommendation unit 54 is used to recommend corresponding response measures based on the risk level. For different risk levels and disaster types, the system provides customized response strategy suggestions:

[0205] Low risk: Enhanced monitoring and shorter sampling intervals are sufficient; no special measures are required.

[0206] General risks: Dispatch engineering personnel to conduct on-site inspections, prepare emergency supplies, and develop preliminary emergency plans;

[0207] High risk: Implement traffic control, evacuate surrounding personnel, activate emergency plans, and deploy emergency response measures;

[0208] Extremely high risk: close roads, evacuate completely, activate the highest level of emergency response, and mobilize rescue forces.

[0209] The formulation of response strategies takes into account various factors such as disaster characteristics, terrain conditions, weather factors, and available resources, providing decision-makers with a scientific reference.

[0210] The present invention also includes a monitoring mode control module 6, which is used to automatically adjust the monitoring frequency and analysis depth according to the slope risk status, including a regular monitoring mode, a warning monitoring mode and an emergency monitoring mode.

[0211] The standard monitoring mode is the system's default operating mode, suitable for slope monitoring under normal conditions. In this mode, the system samples every 12 hours, performs comprehensive analysis, calculates resource allocation balance, and completes data processing and analysis within 24 hours.

[0212] The alert and monitoring mode is activated when a potential risk is detected, and is suitable for situations with a general risk level. In this mode, the system sampling interval is shortened to 1 hour, focusing on abnormal areas and prioritizing the allocation of computing resources to ensure that key data analysis is completed within 1 hour.

[0213] The emergency monitoring mode, activated under high-risk or extremely high-risk conditions, is the highest level of monitoring in the system. In this mode, the system sampling interval is further shortened to 10 minutes, providing real-time analysis of high-risk areas. It utilizes edge computing and cloud collaboration to ensure real-time analysis and early warning are completed within 10 minutes.

[0214] The transition between modes is automatic based on risk level assessment results, but manual intervention is also supported. Transitions from lower to higher levels can be triggered automatically, while transitions from higher to lower levels require manual confirmation to ensure safety.

[0215] Reference Figure 7 The geodesic convolutional network unit 42 of the present invention includes a geodesic convolutional layer 421, a tangent spatial pooling layer 422, a parallel transport layer 423, and a manifold batch normalization layer 424.

[0216] Geodesic convolutional layer 421 is used to perform convolution operations on Riemannian manifolds. Traditional convolutional neural network designs are based on Euclidean space, making them difficult to directly apply to Riemannian manifolds with non-zero curvature. This invention designs convolution operations based on geodesics and tangent space, achieving feature extraction on Riemannian manifolds.

[0217] The core idea of ​​geodesic convolution is to extend the convolution operation in Euclidean space to Riemannian manifolds. The specific implementation includes the following steps:

[0218] Define geodesic neighborhood: For a point x on a manifold, define its geodesic neighborhood. To and The geodetic distance is less than the preset radius The set of all points;

[0219] Construct a local coordinate system: establish a tangent space at point x. And define a local orthogonal coordinate system;

[0220] Mapping to the tangent space: via inverse exponential mapping Map neighborhood points to the tangent space;

[0221] Perform convolution: Perform standard convolution operations in the tangent space;

[0222] Feature aggregation: Maps the convolution result back to a manifold.

[0223] The mathematical expression for geodesic convolution is:

[0224] .

[0225] Where: f*gW represents the result of geodesic convolution; f is the input feature function; W is the convolution kernel function; It is the geodesic neighborhood of point x; It is an inverse exponential mapping that maps a point y on a manifold to the tangent space of a point x; * denotes the dot product operation.

[0226] The tangent-space pooling layer 422 is used for feature aggregation and dimensionality reduction in the tangent space. Pooling operations are performed in the tangent space, avoiding the complexity of pooling directly on the manifold. Specific steps include:

[0227] Define the pooling region: Define a geodesic neighborhood as the pooling region on the manifold;

[0228] Mapping to tangent space: Maps points within the pooling region to the tangent space of the center point;

[0229] Perform pooling: Perform standard pooling operations (such as max pooling or average pooling) in the tangent space.

[0230] Feature aggregation: The pooling result is used as the feature representation of the center point.

[0231] Parallel transport layer 423 is used to align and fuse features from different layers through parallel transport. On a Riemannian manifold, feature vectors at different points lie in different tangent spaces and cannot be directly compared or combined. Parallel transport provides a method to transport vectors from one tangent space to another while maintaining vector parallelism.

[0232] The mathematical expression for parallel transmission has been given above. In practical calculations, parallel transmission can be achieved by solving the parallel transmission equation or by using discrete approximation methods.

[0233] Manifold batch normalization layer 424 is used to perform batch normalization processing adapted to Riemannian geometry. Traditional batch normalization stabilizes network training in Euclidean space by normalizing the feature distribution. On Riemannian manifolds, normalization needs to consider the local structure of the manifold, especially the Riemannian metric.

[0234] The steps of manifold batch normalization include:

[0235] Calculate the mean on the manifold: use the Fréchet mean under the Riemannian metric.

[0236] Calculate the variance on the manifold: based on the geodesic distance to the mean;

[0237] Standardize features: Map the features to the tangent space, perform standardization, and then map them back to the manifold;

[0238] Scale and offset: Apply learnable scale and offset parameters.

[0239] The mathematical expression for manifold batch normalization has been given above.

[0240] Reference Figure 8 The curvature flow evolution unit 31 of the present invention includes a flow evolution equation solver 311, a feature preservation smoother 312, a scale space builder 313, and a critical point tracker 314.

[0241] The flow evolution equation solver 311 is used to solve the curvature flow evolution equations that govern the smoothing process of a surface at different scales. Curvature flow is an important tool for studying the geometric evolution of surfaces. By solving the curvature flow equations, multi-scale representations of surfaces can be constructed.

[0242] In this invention, mean curvature flow is used as the basic evolution equation, the mathematical expression of which has been given above. Mean curvature flow has the property of minimizing area, gradually smoothing the surface during evolution and reducing high-frequency changes in the surface. This characteristic makes it suitable for constructing multi-scale representations from detailed to coarse.

[0243] In numerical solutions, the evolution equations are discretized using explicit or implicit finite difference methods. To ensure numerical stability, the time step must satisfy the CFL condition, typically set to 0.01–0.05. The solution to the evolution equations can be represented as an iterative process:

[0244] .

[0245] in, The scale parameter is represented as The surface position vector at time; The scale parameter is represented as The surface position vector at time; It is the scale step size; It is a scale The average curvature at that time; It is a scale The unit normal vector at that time.

[0246] Feature-preserving smoother 312 is used to retain critical features during the smoothing process. Standard average curvature flow uniformly reduces all curvature during smoothing, failing to distinguish between noise and important features. To preserve the critical features of the slope, this invention designs a feature-preserving curvature flow.

[0247] The key to maintaining feature smoothness is introducing a modulation function that adjusts the smoothing rate based on local geometric characteristics. The modulation function can be based on Gaussian curvature, shape exponent, or other geometric features. In this invention, a modulation function based on Gaussian curvature is used, the mathematical expression of which has been given above.

[0248] The scale space builder 313 is used to generate a sequence of surface representations covering different feature scales. The scale space is a set of parameterized surface sequences representing versions of the original surface at different levels of smoothness. By analyzing the behavior of features in the scale space, noise and true deformation can be distinguished.

[0249] When constructing a scale space, it is necessary to determine the number and distribution of scale levels. This invention uses 5 to 8 scale levels, covering feature scales of 0.5 to 10 meters. The distribution of scale parameters can be linear or logarithmic, with logarithmic distribution being more effective in capturing multi-scale characteristics.

[0250] Scale space can be represented as:

[0251] .

[0252] in: Represents scale space; Represents parameter coordinates Located at scale parameter The lower curved surface point; It is the number of scale levels.

[0253] The critical point tracker 314 is used to track the evolution trajectory of feature points during the evolution of curvature flow. A critical point is a point where curvature or its derivative reaches an extreme value, and is an important indicator of surface features. By tracking the evolution trajectory of critical points in scale space, the scale-space behavior of features can be extracted.

[0254] The steps of critical point tracking include:

[0255] Identify critical points at each scale, such as Gaussian curvature extrema, mean curvature extrema, or shape exponential feature points;

[0256] Establish the correspondence between critical points between adjacent scales to form a tracking trajectory;

[0257] Analyze the behavior of the trajectory, such as duration, displacement, and intensity changes;

[0258] Features are classified based on trajectory characteristics.

[0259] A key challenge in critical point tracking is handling bifurcation and merging events in scale space. In these events, features may split into multiple branches or multiple features may merge into one. This invention addresses this problem using a nearest neighbor and similarity metric-based approach, ensuring both continuity and accuracy in tracking.

[0260] The present invention also includes a data management module 7, which stores historical monitoring data, model parameters, and early warning records, and provides data query, analysis, and visualization functions. The data management module 7 establishes communication connections with the image acquisition module 1, the slope parametric characterization module 2, the multi-scale curvature flow analysis module 3, the Riemannian manifold learning module 4, and the risk early warning module 5 for data sharing and collaborative analysis.

[0261] The data management module 7 includes a data storage unit, a data query unit, a data analysis unit, and a data visualization unit.

[0262] The data storage unit is responsible for storing and managing raw data, processing results, and alert records. The storage adopts a hierarchical structure, including:

[0263] Raw data layer: Stores raw data such as images and point clouds, using an incremental backup strategy;

[0264] Feature data layer: Stores extracted features and analysis results, supporting fast retrieval;

[0265] Early warning record layer: Stores historical early warning information and handling records for traceability and analysis;

[0266] Model parameter layer: Stores trained model parameters and configuration information, and supports model updates.

[0267] The data query unit provides multi-condition query and data export functions, supporting flexible queries based on conditions such as time, location, feature type, and risk level. The query interface adopts a standardized design, supporting integration and interaction with other systems.

[0268] The data analysis unit is responsible for in-depth analysis of historical data, including trend analysis, correlation analysis, and predictive analysis. Trend analysis automatically generates time-series charts of key indicators, visually displaying trends; correlation analysis uncovers potential relationships between different factors, such as the correlation between weather conditions and slope deformation; predictive analysis uses historical data to forecast future risk changes, providing a basis for long-term planning.

[0269] The data visualization unit transforms complex monitoring data and analysis results into intuitive visual representations, including 3D models, heatmaps, trend charts, and risk maps. The visualization interface supports multi-level displays, from macro overviews to micro details, meeting the needs of users at different levels.

[0270] Through data management module 7, this invention realizes full lifecycle management of monitoring data, providing a solid data foundation for slope risk assessment and decision support.

[0271] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A highway slope stability image monitoring and risk early warning system, characterized in that, Comprising: an image acquisition module for obtaining high-resolution image data and three-dimensional point cloud data of the highway slope; a slope parameterization representation module in communication with the image acquisition module, for representing the highway slope as a two-dimensional Riemannian manifold, and calculating the metric tensor and curvature tensor of the slope surface; a multi-scale curvature flow analysis module in communication with the slope parameterization representation module, for receiving the metric tensor and curvature tensor of the slope surface, constructing a multi-scale representation of the slope surface, and extracting a multi-scale feature atlas; a Riemannian manifold learning module in communication with the multi-scale curvature flow analysis module, for representing slope features in a Riemannian manifold space, extracting high-level features through a geodesic convolution network, and identifying landslide, collapse and debris flow disaster precursors; a risk warning module in communication with the Riemannian manifold learning module, for generating risk level assessment and warning information based on the identified disaster precursors; the slope parameterization representation module comprises: a parameterization mapping unit for constructing a mapping relationship from a parameter domain to a three-dimensional coordinate of the slope; a differential characteristic calculation unit for calculating the metric tensor, curvature tensor, principal curvature and principal direction of the slope surface; a differential invariant generation unit for calculating the Gaussian curvature, mean curvature and shape index of the slope surface; a time variation rate analysis unit for calculating the variation rate of the differential characteristics in the time dimension, and marking abnormal areas with a variation rate exceeding a preset threshold; the parameterization mapping unit performs the following operations: construct a local coordinate system with the center of the slope as the origin; divide the slope area into grid units, each unit size is 0.5m x 0.5m; construct a parameter mapping, define a mapping relationship from the parameter domain (u, v) to the actual three-dimensional coordinates (x, y, z) of the slope; add a time dimension to form a spatio-temporal parameterization representation, capture the dynamic characteristics of the slope over time.

2. The system of claim 1, wherein, the image acquisition module comprises: a high-resolution camera array for obtaining image data of the highway slope; a three-dimensional laser radar for obtaining point cloud data of the highway slope; a GPS positioning system for obtaining spatial position information; a data preprocessing unit for denoising, registration and fusion processing of the image data and point cloud data, generating a textured three-dimensional model of the slope.

3. The system of claim 1, wherein, the multi-scale curvature flow analysis module comprises: a curvature flow evolution unit for constructing a multi-scale representation of the slope surface; a feature operator unit for applying differential feature operators at different scales to extract local features; a feature atlas generation unit for organizing features at different scales into a feature atlas; a scale space analysis unit for analyzing the evolution law of features in the scale space, identifying feature patterns related to disaster precursors.

4. The system of claim 1, wherein, the Riemannian manifold learning module comprises: a manifold construction unit for defining appropriate Riemannian manifolds according to the internal structure of slope features; a geodesic convolution network unit for performing convolution operations on Riemannian manifolds to extract high-level features; a disaster precursor classification unit for identifying different types of disaster precursors; a risk probability estimation unit for calculating the probability distribution and spatio-temporal risk map of different disaster types.

5. The system of claim 1, wherein, the risk warning module comprises: A risk assessment unit is configured to assess a risk level based on the disaster precursor identification result. An early warning information generation unit is configured to generate early warning information containing a risk type, a location, a level, and a disposal suggestion. A multi-channel notification unit is configured to send the early warning information through short messages, emails, and monitoring platforms. A response strategy recommendation unit is configured to recommend corresponding disposal measures according to the risk level.

6. The system of any one of claims 1 to 5, wherein, The system further comprises a monitoring mode control module configured to automatically adjust a monitoring frequency and an analysis depth according to a slope risk state, including a regular monitoring mode, a warning monitoring mode, and an emergency monitoring mode.

7. The system of claim 4, wherein, The geodesic convolution network unit comprises: A geodesic convolution layer configured to perform a convolution operation on a Riemannian manifold; A tangent space pooling layer configured to perform feature aggregation and dimension reduction in a tangent space; A parallel transmission layer configured to realize alignment and fusion of features of different layers through parallel transmission; A manifold batch normalization layer configured to perform batch normalization processing adapted to Riemannian geometry.

8. The system of claim 3, wherein, The curvature flow evolution unit comprises: A flow evolution equation solver configured to solve a curvature flow evolution equation controlling a smoothing process of a surface at different scales; A feature preserving smoother configured to preserve key features in the smoothing process; A scale space constructor configured to generate a sequence of surface representations covering different feature scales; A critical point tracker configured to track an evolution trajectory of a feature point in the curvature flow evolution process.

9. The system of claim 1, wherein, The system further comprises a data management module configured to store historical monitoring data, model parameters, and early warning records, and to provide data query, analysis, and visualization functions; the data management module is communicatively connected with the image acquisition module, the slope parameterization representation module, the multi-scale curvature flow analysis module, the Riemannian manifold learning module, and the risk early warning module, for data sharing and collaborative analysis.

Images