Method and device for monitoring and forecasting dangerous rock collapse based on infrared thermal imaging technology

Abstract

This invention discloses a method and equipment for monitoring and predicting rockfalls based on infrared thermal imaging technology, comprising the following steps: Step 1: Collecting multi-source raw data of the rockfall area using various hardware devices; Step 2: Preprocessing the multi-source raw data and extracting different types of data features; Step 3: Constructing a spatial topology map of the rockfall based on the different types of data features; Step 4: Applying an improved Persistent Homology algorithm to perform persistence analysis on the spatial topology map of the rockfall, generating topological persistence features; Step 5: Extracting the spatial dependence features of the rock mass based on a spectral convolutional network; Step 6: Generating a rockfall risk warning index based on the spatial dependence features of the rock mass and time-series data of temperature changes; Step 7: Calculating the probability of rockfall occurrence and generating the final rockfall risk level. This invention integrates multi-source data and topology analysis to achieve accurate monitoring and early warning of rockfall risks.

Description

Technical Field

[0001] This invention relates to the field of geological monitoring and early warning technology, and in particular to a method and equipment for monitoring and predicting rockfalls based on infrared thermal imaging technology. Background Technology

[0002] With the advancement of industrialization and urbanization, rockfalls, as a geological hazard, have gradually become a major hidden danger affecting human safety and property. Especially in mountainous areas, mining areas, and construction sites, landslides occur frequently. Traditional methods for monitoring rockfalls typically rely on manual inspections and limited geological exploration means. These methods are not only inefficient but also struggle to obtain real-time information on the dynamic changes of unstable rock formations. Therefore, how to achieve real-time monitoring of unstable rock areas, timely assessment of their stability, and effective early warning of landslide risks has become an urgent technical challenge.

[0003] Existing methods for monitoring rockfalls generally rely on simple, single data sources to assess rock mass stability. For example, traditional infrared thermal imaging technology is mainly used for detecting temperature distribution, but it can only provide basic information on surface temperature changes, lacking a comprehensive analysis of the internal structure and environmental conditions of the rock mass. While lidar and meteorological sensors can provide three-dimensional geometric and meteorological data, this data often only statically reflects the state of the rock mass at a specific moment, lacking real-time tracking of dynamic changes. Furthermore, existing collapse risk assessment methods mostly employ analytical methods based on single features or empirical formulas. These methods have low accuracy and reliability under complex geological conditions and cannot fully consider the interactions of various environmental factors.

[0004] Another limitation of existing technologies lies in their oversimplification of spatial feature analysis, failing to accurately capture the complex spatial dependencies within rock masses. Traditional methods for assessing dangerous rock areas often neglect the interactions and intrinsic connections between different regions, particularly the complex relationships between factors such as thermal stress, temperature variations, stress, and humidity within the rock mass. While some studies have attempted to apply machine learning algorithms to landslide risk prediction, these methods typically lack in-depth analysis of geological data characteristics and spatial correlations, resulting in poor accuracy and reliability when processing complex data.

[0005] Therefore, how to provide methods and equipment for monitoring and predicting rockfalls based on infrared thermal imaging technology is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and device for monitoring and predicting rockfalls based on infrared thermal imaging technology. This invention integrates multi-source data from infrared thermal imaging, lidar, meteorological sensors, and geological sensors, combined with an improved Persistent Homology algorithm and spectral convolutional networks, to achieve accurate monitoring and early warning of rockfalls. Through spatial topology analysis and persistent feature extraction, it can identify vulnerable areas of the rock mass and generate a rockfall risk warning index based on real-time and historical data, providing dynamic and accurate rockfall risk prediction and emergency response support.

[0007] The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to embodiments of the present invention includes the following steps:

[0008] Step 1: Collect multi-source raw data of the dangerous rock area using various hardware devices;

[0009] Step 2: Preprocess the different types of multi-source raw data separately and extract the data features of different types;

[0010] Step 3: Based on temperature change characteristics, three-dimensional geometric features, soil moisture and stress characteristics, construct a spatial topology map of the dangerous rock mass. Each small area of ​​the rock mass is regarded as a node in the spatial topology map of the dangerous rock mass. Edges are established between nodes through spatial adjacency, thermal stress and temperature change relationships.

[0011] Step 4: Apply the improved Persistent Homology algorithm to perform persistence analysis on the spatial topology map of the dangerous rock and generate topological persistence features;

[0012] Step 5: Input the topological persistence features and the spatial topology map of the dangerous rock mass into the spectral convolutional network to extract the spatial dependency features of the rock mass;

[0013] Step 6: Generate a collapse risk warning index based on the spatial dependence characteristics of the rock mass and the time series data of temperature changes;

[0014] Step 7: Calculate the probability of a collapse based on the collapse risk warning index and historical collapse data, and generate the final collapse risk level.

[0015] Optionally, the various hardware devices include infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors; the multi-source raw data includes infrared thermal imaging image data, lidar point cloud data, meteorological data, and geological data.

[0016] Optionally, the preprocessing of different types of multi-source raw data and the extraction of different types of data features specifically include:

[0017] The infrared thermal imaging image is subjected to wavelet transform to remove noise, and the brightness and contrast of the denoised infrared thermal imaging image are enhanced. Temperature change features are extracted and time series data of temperature change are generated.

[0018] Gaussian filtering is applied to the lidar point cloud data, followed by spatial coordinate alignment and registration.

[0019] The steps for reconstructing the three-dimensional geometric features of a rock mass using the Poisson reconstruction method include:

[0020] Calculate the normal vector of each point in the lidar point cloud data;

[0021] Based on the point cloud normal vectors and point cloud distribution, the surface is solved using the Poisson equation, and a smooth three-dimensional geometric model is generated.

[0022] Three-dimensional geometric features are extracted from the generated three-dimensional geometric model, including crack features, fault features, protrusion regions, and surface curvature.

[0023] Z-score standardization was performed on meteorological and geological data to extract soil moisture and stress characteristics.

[0024] Optionally, based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics, a spatial topology map of the unstable rock mass is constructed. Each small region of the rock mass is considered a node in the spatial topology map, and edges are established between nodes through spatial adjacency, thermal stress, and temperature change relationships. Specifically:

[0025] Based on the preset division criteria, the entire dangerous rock area is divided into multiple smaller areas according to geographical location and thermal stress;

[0026] The characteristics of each small region include the temperature variation of the rock mass, three-dimensional geometry, soil moisture and stress state, which serve as nodal attributes for each small region;

[0027] Assign a unique node ID to each small region to form a preliminary set of graph nodes;

[0028] The specific steps for constructing a spatial topology map of dangerous rocks based on the graph node set are as follows:

[0029] For each node's small region, spatial adjacency is determined. If two small regions are spatially adjacent and have a connecting boundary, a spatial adjacency edge is established between these two nodes. The weight of the spatial adjacency edge is the spatial contact area of ​​the adjacent regions.

[0030] If the magnitude of thermal stress change in two small regions is less than a preset value, a thermal stress edge is established between these two nodes, and the weight of the thermal stress edge is the difference in thermal stress between the two small regions.

[0031] If the temperature change between two small regions is less than a preset value, a temperature edge is established between the two nodes, and the weight of the temperature edge is the temperature difference between the two small regions.

[0032] If the humidity change between two small areas is less than a preset value, a humidity edge is established between the two nodes, and the weight of the humidity edge is the humidity difference between the two small areas.

[0033] If the stress fluctuation amplitude of two small regions is less than a preset value, a stress edge is established between these two nodes, and the weight of the stress fluctuation edge is the stress difference between the two small regions.

[0034] Optionally, the improved Persistent Homology algorithm is applied to perform persistence analysis on the spatial topology map of the dangerous rock to generate topological persistence features, specifically:

[0035] Each node in the spatial topology graph of the dangerous rock is converted into a persistent point using an improved Persistent Homology algorithm.

[0036] The transformation step specifically involves mapping the topological features of each node to a high-dimensional space using a Laplacian matrix, and using the appearance and disappearance scales of the nodes as coordinates of the persistence values.

[0037] The coordinates of the persistence value are two-dimensional coordinates reflected by the persistence point in the persistence map. The X coordinate represents the survival time of the node feature at the appearance scale, and the Y coordinate represents the disappearance time of the node feature at the disappearance scale.

[0038] Based on the two-dimensional coordinates of the persistent points, the Euclidean distance between the persistent points is calculated, and the edges in the persistent graph are established.

[0039] Based on the persistence map, persistent points connected by edges with persistence values ​​higher than a preset value are selected. The persistence value is the Euclidean distance between persistent points. Persistent points connected by edges with persistence values ​​higher than the preset value represent vulnerable areas of the rock mass and are potential collapse risk points.

[0040] Topological persistence features are extracted from the persistence points connected to edges with persistence values ​​higher than a preset value. These topological persistence features represent structural changes, crack propagation, and potential collapse risk areas of the rock mass.

[0041] Optionally, the step of inputting the topological persistence features and the spatial topology map of the dangerous rock mass into a spectral graph convolutional network to extract the spatial dependency features of the rock mass specifically involves:

[0042] Based on the spatial adjacency relationships, thermal stress, humidity and temperature changes among nodes in the spatial topology diagram of dangerous rocks, an adjacency matrix of the spatial topology diagram of dangerous rocks is constructed.

[0043] The adjacency matrix and topological persistence features of the spatial topology map of the dangerous rock mass are input into the graph convolution layer of the spectral graph convolutional network. By performing graph convolution operation on the Laplacian matrix corresponding to the adjacency matrix and topological persistence features, the spatial dependence of thermal stress and temperature between different regions in the rock mass is extracted.

[0044] After the graph convolution operation in each layer, the Sigmoid activation function is applied to perform a nonlinear transformation on the spatial dependencies, and the pooled spatial dependency features are generated through the pooling layer of the spectral graph convolutional network.

[0045] Multiple spatial dependency features extracted by the pooling layer are input into the fully connected layer for integration and extraction to generate the final spatial dependency features.

[0046] Optionally, the collapse risk warning index is generated based on the spatial dependence characteristics of the rock mass and the time-series data of temperature changes, specifically as follows:

[0047] The time series data of temperature changes are arranged in chronological order to form a time series vector, which represents the temperature change trend of the rock mass in different time periods.

[0048] Spatial dependency features and temporal data vectors are concatenated to form a comprehensive feature vector that includes both spatial and temporal features;

[0049] Min-Max normalization is applied to the composite feature vector to scale the feature values ​​to the same range.

[0050] Each feature in the comprehensive feature vector is assigned a weight according to a preset allocation standard;

[0051] The standardized feature vectors are weighted and summed with their corresponding weights to obtain the comprehensive feature value of each small region.

[0052] Based on comprehensive characteristic values, a collapse risk warning index is calculated, which represents the collapse risk of the rock mass at a specific point in time.

[0053] Optionally, the step of calculating the probability of a collapse based on the collapse risk warning index and historical collapse data, and generating a final collapse risk level, specifically involves:

[0054] Collect historical landslide events and record the environmental characteristics and time of the landslides;

[0055] The current calculated collapse risk warning index is compared with the collapse risk warning index in historical collapse events, and Bayesian inference is used to calculate the probability of collapse under the current conditions.

[0056] Based on the calculated probability of collapse, a collapse risk level is generated. The specific steps include:

[0057] Multiple risk levels are set based on the probability of collapse.

[0058] A collapse probability of less than 10% is considered low risk and will be subject to routine monitoring.

[0059] A collapse probability between 10% and 40% is considered medium risk; emergency response equipment should be prepared in advance.

[0060] A collapse probability greater than 40% is considered a high risk; the emergency response procedure should be activated immediately, personnel evacuated, and reinforcement measures implemented.

[0061] The system uses a graphical interface to display the collapse risk warning index and collapse risk level in real time.

[0062] The rockfall monitoring and prediction device based on infrared thermal imaging technology according to an embodiment of the present invention includes the following modules:

[0063] The data acquisition and preprocessing module is used to acquire multi-source raw data from dangerous rock areas through various hardware devices, and to preprocess different types of raw data to extract features of various types of data.

[0064] The spatial topology map construction module is used to construct a spatial topology map of dangerous rocks based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics.

[0065] The topology persistence analysis module is used to apply the improved Persistent Homology algorithm to perform topology data analysis and persistence analysis on the spatial topology map of dangerous rocks, and generate topology persistence features.

[0066] The spatial dependency feature extraction module is used to input topological persistence features and spatial topology maps of dangerous rocks into a spectral convolutional network to extract the spatial dependency features of the rock mass.

[0067] The collapse risk early warning calculation module is used to generate a collapse risk early warning index based on the spatial dependence characteristics of the rock mass and time series data of temperature changes.

[0068] The landslide risk level generation module is used to generate the final landslide risk level based on the landslide risk warning index and historical landslide data, and then display it in real time.

[0069] The beneficial effects of this invention are:

[0070] 1. This invention, by combining various hardware devices such as infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors, can collect multi-source raw data, greatly improving the comprehensiveness and accuracy of monitoring rockfalls. Compared with the limitations of traditional technologies that rely on a single data source, this invention, through the fusion of multi-source data, obtains real-time information on rock masses in multiple dimensions such as temperature changes, three-dimensional geometric morphology, meteorological conditions, and soil moisture, thereby enabling a more accurate assessment of the overall stability of unstable rock masses.

[0071] 2. This invention utilizes an improved Persistent Homology algorithm to perform persistent analysis of the spatial topology of unstable rock masses, enabling the analysis of structural features and crack propagation at different scales. Traditional methods often overlook subtle changes in the internal structure of rock masses, while this invention, through persistent analysis of topological features, successfully captures the long-term stability and potential collapse risk points of the rock mass. This allows the invention to effectively identify and predict vulnerable areas within the rock mass, thereby improving the accuracy and foresight of collapse risk warnings.

[0072] 3. This invention utilizes a spectral graph convolutional network to extract the spatial dependency features of rock masses, overcoming the shortcomings of existing technologies in spatial relationship analysis. Through graph convolution operations, it is possible to deeply explore the complex spatial dependencies between different regions of the rock mass, combining multidimensional and spatial features to provide a more accurate assessment of collapse risk. This spatial dependency-based analysis method can better reflect the interaction of different regions of the rock mass under environmental changes, further improving the accuracy of early warning.

[0073] 4. This invention performs real-time dynamic prediction of landslide risk by calculating a landslide risk warning index and using historical data. By combining real-time and historical data, this invention generates a landslide risk index through weighted summation and calculates the probability of a landslide based on this index. This method does not rely on traditional empirical formulas but instead uses a scientific calculation model combined with the real-time state of the rock mass to generate a landslide risk level and display it to decision-makers in real time. This provides data support for rapid response and emergency measures. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1 This is an overall flowchart of the method for monitoring and predicting rockfalls based on infrared thermal imaging technology proposed in this invention;

[0076] Figure 2This is a schematic diagram of the structure of the rockfall monitoring and forecasting device based on infrared thermal imaging technology proposed in this invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0078] refer to Figure 1 A method for monitoring and predicting rockfalls based on infrared thermal imaging technology includes the following steps:

[0079] Step 1: Collect multi-source raw data of the dangerous rock area using various hardware devices;

[0080] Step 2: Preprocess the different types of multi-source raw data separately and extract the data features of different types;

[0081] Step 3: Based on temperature change characteristics, three-dimensional geometric features, soil moisture and stress characteristics, construct a spatial topology map of the dangerous rock mass. Each small area of ​​the rock mass is regarded as a node in the spatial topology map of the dangerous rock mass. Edges are established between nodes through spatial adjacency, thermal stress and temperature change relationships.

[0082] Step 4: Apply the improved Persistent Homology algorithm to perform persistence analysis on the spatial topology map of the dangerous rock and generate topological persistence features;

[0083] Step 5: Input the topological persistence features and the spatial topology map of the dangerous rock mass into the spectral convolutional network to extract the spatial dependency features of the rock mass;

[0084] Step 6: Generate a collapse risk warning index based on the spatial dependence characteristics of the rock mass and the time series data of temperature changes;

[0085] Step 7: Calculate the probability of a collapse based on the collapse risk warning index and historical collapse data, and generate the final collapse risk level.

[0086] This invention, through the acquisition and fusion of multi-source data and employing hardware such as infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors, can comprehensively acquire temperature, geometric, meteorological, and geological characteristics of dangerous rock areas. Compared with traditional methods, this invention not only improves the comprehensiveness of the data but also enables real-time acquisition of dynamic changes in the rock mass, avoiding the inefficiency and lag of traditional monitoring methods. Through an improved Persistent Homology algorithm and spectral convolutional networks, this invention can accurately extract spatially dependent features based on multi-dimensional features, overcoming the limitations of traditional methods in spatial feature analysis. This allows for a more accurate assessment of the stability of dangerous rock formations, early identification of collapse risk areas, and strong support for rapid early warning and decision-making.

[0087] In this embodiment, the various hardware devices include infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors; the multi-source raw data includes infrared thermal imaging image data, lidar point cloud data, meteorological data, and geological data.

[0088] This step utilizes a variety of hardware devices, including infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors, to achieve comprehensive monitoring of the dangerous rock area. Compared to traditional single-data source monitoring methods, the multi-source data acquisition method of this invention can capture information such as rock mass temperature changes, three-dimensional geometric features, meteorological changes, and soil moisture in real time across multiple dimensions, greatly enhancing the accuracy and completeness of the data. Furthermore, by precisely controlling the data acquisition frequency and spatial resolution, this invention further improves the accuracy of landslide risk assessment, providing more reliable raw data support for subsequent data processing and risk warning.

[0089] In this embodiment, the preprocessing of different types of multi-source raw data and the extraction of different types of data features specifically include:

[0090] The infrared thermal imaging image is subjected to wavelet transform to remove noise, and the brightness and contrast of the denoised infrared thermal imaging image are enhanced. Temperature change features are extracted and time series data of temperature change are generated.

[0091] Gaussian filtering is applied to the lidar point cloud data, followed by spatial coordinate alignment and registration.

[0092] The steps for reconstructing the three-dimensional geometric features of a rock mass using the Poisson reconstruction method include:

[0093] Calculate the normal vector of each point in the lidar point cloud data;

[0094] Based on the point cloud normal vectors and point cloud distribution, the surface is solved using the Poisson equation, and a smooth three-dimensional geometric model is generated.

[0095] Three-dimensional geometric features are extracted from the generated three-dimensional geometric model, including crack features, fault features, protrusion regions, and surface curvature.

[0096] Z-score standardization was performed on meteorological and geological data to extract soil moisture and stress characteristics.

[0097] This step employs targeted preprocessing methods for different types of multi-source raw data, and utilizes techniques such as wavelet transform denoising, Poisson reconstruction of 3D geometric features, and Z-score normalization to ensure improved data quality. This refined data processing and feature extraction approach not only effectively eliminates noise but also extracts more accurate temperature change features, 3D geometric features, soil moisture, and stress features, thus providing high-quality data input for subsequent landslide risk assessment. Compared to traditional, simple preprocessing methods, this invention can more accurately reflect the actual state of the rock mass, improving the accuracy and reliability of predictions.

[0098] In this embodiment, a spatial topology map of the unstable rock mass is constructed based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics. Each small region of the rock mass is considered a node in the spatial topology map, and edges are established between nodes through spatial adjacency, thermal stress, and temperature change relationships. Specifically:

[0099] Based on the preset division criteria, the entire dangerous rock area is divided into multiple smaller areas according to geographical location and thermal stress;

[0100] The characteristics of each small region include the temperature variation of the rock mass, three-dimensional geometry, soil moisture and stress state, which serve as nodal attributes for each small region;

[0101] Assign a unique node ID to each small region to form a preliminary set of graph nodes;

[0102] The specific steps for constructing a spatial topology map of dangerous rocks based on the graph node set are as follows:

[0103] For each node's small region, spatial adjacency is determined. If two small regions are spatially adjacent and have a connecting boundary, a spatial adjacency edge is established between these two nodes. The weight of the spatial adjacency edge is the spatial contact area of ​​the adjacent regions.

[0104] If the magnitude of thermal stress change in two small regions is less than a preset value, a thermal stress edge is established between these two nodes, and the weight of the thermal stress edge is the difference in thermal stress between the two small regions.

[0105] If the temperature change between two small regions is less than a preset value, a temperature edge is established between the two nodes, and the weight of the temperature edge is the temperature difference between the two small regions.

[0106] If the humidity change between two small areas is less than a preset value, a humidity edge is established between the two nodes, and the weight of the humidity edge is the humidity difference between the two small areas.

[0107] If the stress fluctuation amplitude of two small regions is less than a preset value, a stress edge is established between these two nodes, and the weight of the stress fluctuation edge is the stress difference between the two small regions.

[0108] This step utilizes a spatial topological map of the unstable rock mass, constructed based on temperature variation characteristics, three-dimensional geometric features, soil moisture, and stress characteristics, to achieve comprehensive analysis of various regions within the rock mass. Unlike traditional methods that rely on only a single feature for evaluation, this invention considers multiple dimensions and establishes a more comprehensive spatial topological structure. By introducing equilateral relationships such as spatial adjacency, thermal stress, and temperature variation, the topological map effectively reveals the interactions and similarities between regions within the rock mass, thereby improving the accuracy of predicting rock mass stability and potential collapse risks.

[0109] In this embodiment, the application of the improved Persistent Homology algorithm to perform persistence analysis on the spatial topology map of the dangerous rock mass and generate topological persistence features specifically includes:

[0110] Each node in the spatial topology graph of the dangerous rock is converted into a persistent point using an improved Persistent Homology algorithm.

[0111] The transformation step specifically involves mapping the topological features of each node to a high-dimensional space using a Laplacian matrix, and using the appearance and disappearance scales of the nodes as coordinates of the persistence values.

[0112] The coordinates of the persistence value are two-dimensional coordinates reflected by the persistence point in the persistence map. The X coordinate represents the survival time of the node feature at the appearance scale, and the Y coordinate represents the disappearance time of the node feature at the disappearance scale.

[0113] Based on the two-dimensional coordinates of the persistent points, the Euclidean distance between the persistent points is calculated, and the edges in the persistent graph are established.

[0114] Based on the persistence map, persistent points connected by edges with persistence values ​​higher than a preset value are selected. The persistence value is the Euclidean distance between persistent points. Persistent points connected by edges with persistence values ​​higher than the preset value represent vulnerable areas of the rock mass and are potential collapse risk points.

[0115] Topological persistence features are extracted from the persistence points connected to edges with persistence values ​​higher than a preset value. These topological persistence features represent structural changes, crack propagation, and potential collapse risk areas of the rock mass.

[0116] This step applies an improved Persistent Homology algorithm to perform persistent analysis on the spatial topology of the unstable rock mass, accurately capturing its structural features and crack propagation at different scales. Unlike traditional methods that neglect spatial features and local variations, this invention, through persistent analysis, can identify and extract important long-term features, such as vulnerable areas and crack zones within the rock mass. This innovative algorithm not only enhances the understanding of rock mass structural evolution but also provides a scientific basis for early warning of collapse risks, significantly improving the accuracy and reliability of monitoring.

[0117] In this embodiment, the step of inputting the topological persistence features and the spatial topology map of the dangerous rock mass into a spectral convolutional network to extract the spatial dependency features of the rock mass specifically involves:

[0118] Based on the spatial adjacency relationships, thermal stress, humidity and temperature changes among nodes in the spatial topology diagram of dangerous rocks, an adjacency matrix of the spatial topology diagram of dangerous rocks is constructed.

[0119] The adjacency matrix and topological persistence features of the spatial topology map of the dangerous rock mass are input into the graph convolution layer of the spectral graph convolutional network. By performing graph convolution operation on the Laplacian matrix corresponding to the adjacency matrix and topological persistence features, the spatial dependence of thermal stress and temperature between different regions in the rock mass is extracted.

[0120] After the graph convolution operation in each layer, the Sigmoid activation function is applied to perform a nonlinear transformation on the spatial dependencies, and the pooled spatial dependency features are generated through the pooling layer of the spectral graph convolutional network.

[0121] Multiple spatial dependency features extracted by the pooling layer are input into the fully connected layer for integration and extraction to generate the final spatial dependency features.

[0122] This step utilizes a spectral graph convolutional network (SGCN) to extract the spatial dependency features of the rock mass, efficiently capturing the interactions between different regions of the rock mass through graph convolution operations. Unlike traditional methods that only analyze local features, this invention extracts the global spatial dependencies of the rock mass through multi-layer convolution and pooling operations of SGCN. By integrating topological persistence features and spatial dependency features, this invention can better identify key risk areas in the rock mass, providing efficient and accurate spatial information support for subsequent collapse risk assessment and improving the reliability of collapse prediction.

[0123] In this embodiment, the generation of a collapse risk warning index based on the spatial dependence characteristics of the rock mass and time-series data of temperature changes specifically involves:

[0124] The time series data of temperature changes are arranged in chronological order to form a time series vector, which represents the temperature change trend of the rock mass in different time periods.

[0125] Spatial dependency features and temporal data vectors are concatenated to form a comprehensive feature vector that includes both spatial and temporal features;

[0126] Min-Max normalization is applied to the composite feature vector to scale the feature values ​​to the same range.

[0127] Each feature in the comprehensive feature vector is assigned a weight according to a preset allocation standard;

[0128] The standardized feature vectors are weighted and summed with their corresponding weights to obtain the comprehensive feature value of each small region.

[0129] Based on comprehensive characteristic values, a collapse risk warning index is calculated, which represents the collapse risk of the rock mass at a specific point in time.

[0130]

[0131] Where a is a constant term, α i f is the coefficient of the i-th comprehensive eigenvalue. i Let n be the total feature value, and n be the total number of total feature values ​​for each small region.

[0132] This step calculates the collapse risk warning index using a weighted summation method. Based on spatially dependent characteristics and time-series data of temperature changes, it can accurately assess the collapse risk of rock masses at specific points in time. Unlike traditional assessment methods based on empirical formulas, this invention generates a comprehensive risk index through weighted integration of multi-source data.

[0133] In this embodiment, the step of calculating the probability of a collapse based on the collapse risk warning index and historical collapse data, and generating the final collapse risk level, specifically involves:

[0134] Collect historical landslide events and record the environmental characteristics and time of the landslides;

[0135] The current calculated collapse risk warning index is compared with the collapse risk warning index in historical collapse events, and Bayesian inference is used to calculate the probability of collapse under the current conditions.

[0136] Based on the calculated probability of collapse, a collapse risk level is generated. The specific steps include:

[0137] Multiple risk levels are set based on the probability of collapse.

[0138] A collapse probability of less than 10% is considered low risk and will be subject to routine monitoring.

[0139] A collapse probability between 10% and 40% is considered medium risk; emergency response equipment should be prepared in advance.

[0140] A collapse probability greater than 40% is considered a high risk; the emergency response procedure should be activated immediately, personnel evacuated, and reinforcement measures implemented.

[0141] The risk level is dynamically adjusted based on changes in real-time data. For example, when certain environmental factors (such as rapid temperature changes or stress concentration) become abnormal, the risk level may be upgraded to medium or high risk to reflect a greater risk of collapse.

[0142] The system uses a graphical interface to display the collapse risk warning index and collapse risk level in real time.

[0143] This step combines a landslide risk warning index with historical landslide data to generate the probability of landslide occurrence, effectively predicting the likelihood of landslide events. Compared to traditional static prediction methods, this step provides a dynamic early warning system based on the fusion of real-time and historical data. This system can make timely adjustments according to actual changes and accurately predict the landslide risk level in different areas. Through this method, decision-makers can quickly obtain landslide risk information and take effective measures to reduce disaster losses.

[0144] refer to Figure 2 The rockfall monitoring and forecasting equipment based on infrared thermal imaging technology includes the following modules:

[0145] The data acquisition and preprocessing module is used to acquire multi-source raw data from dangerous rock areas through various hardware devices, and to preprocess different types of raw data to extract features of various types of data.

[0146] The spatial topology map construction module is used to construct a spatial topology map of dangerous rocks based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics.

[0147] The topology persistence analysis module is used to apply the improved Persistent Homology algorithm to perform topology data analysis and persistence analysis on the spatial topology map of dangerous rocks, and generate topology persistence features.

[0148] The spatial dependency feature extraction module is used to input topological persistence features and spatial topology maps of dangerous rocks into a spectral convolutional network to extract the spatial dependency features of the rock mass.

[0149] The collapse risk early warning calculation module is used to generate a collapse risk early warning index based on the spatial dependence characteristics of the rock mass and time series data of temperature changes.

[0150] The landslide risk level generation module is used to generate the final landslide risk level based on the landslide risk warning index and historical landslide data, and then display it in real time.

[0151] This step integrates multiple modules—data acquisition and preprocessing, spatial topology map construction, topology persistence analysis, spatial dependency feature extraction, and landslide risk early warning calculation—to achieve end-to-end monitoring and prediction of landslide risk from data acquisition to early warning. The collaborative work of these modules efficiently fuses multi-source data and extracts spatial and persistence features, significantly improving the accuracy and real-time performance of landslide risk assessment. The system can monitor changes in unstable rock areas in real time, dynamically adjust risk levels, and assist decision-makers in making rapid responses to reduce the risk of disasters.

[0152] Example 1:

[0153] To verify the feasibility of this invention in practice, it was applied to a mining area in a mountainous region. Due to long-term mining operations and complex geological conditions, this area contains areas of varying degrees of rock instability, particularly around open-pit mining areas, where the stability of the rock mass is increasingly threatened. By applying the rockfall monitoring and prediction method based on infrared thermal imaging technology from this invention, we can monitor the temperature changes, stress state, and other relevant environmental factors of the rock mass in real time, thereby accurately assessing rock mass stability and providing early warnings of potential collapse risks. Some collapses in this mining area have caused significant damage to surrounding facilities, and traditional monitoring methods failed to detect the hidden dangers in a timely manner.

[0154] When applying this method to this mining area, multi-source raw data was first collected using various hardware devices. Infrared thermal imaging equipment scanned the temperature distribution map of the entire mining area, lidar performed high-precision 3D scanning of the rock surface, meteorological sensors collected meteorological data (such as temperature, humidity, and wind speed) in real time, and geological sensors collected soil moisture and stress data. These devices worked together to provide a comprehensive multi-source dataset. During data acquisition, temperature data was collected every 30 minutes, geological data was collected every hour, and real-time monitoring was conducted based on the periodic patterns of meteorological changes. In the data preprocessing stage, wavelet transform denoising was first performed on the infrared thermal imaging images to remove noise, and the brightness and contrast of the denoised images were enhanced to extract temperature change features. Gaussian filtering was applied to the point cloud data collected by lidar to remove noise and achieve spatial coordinate alignment and registration. A 3D geometric model of the rock surface was generated using the Poisson reconstruction algorithm, and geometric features including cracks, faults, and protrusions were extracted. After Z-score normalization, soil moisture and stress characteristics were extracted from meteorological and geological data, which provided a basis for subsequent spatial topology analysis.

[0155] When constructing the spatial topology map of the unstable rock mass, we considered each small region of the rock mass as a node in the graph based on temperature variation characteristics, three-dimensional geometric features, soil moisture, and stress characteristics. Edges were established between nodes based on spatial adjacency, thermal stress, and temperature variation relationships. Using an improved Persistent Homology algorithm, we performed persistence analysis on the spatial topology map, generating topological persistence features. These features represent structural changes, crack propagation, and potential collapse risk areas within the rock mass. The generated topological persistence features and the unstable rock mass spatial topology map were input into a spectral graph convolutional network to extract spatial dependency features of different regions within the rock mass. These features reflect the interactions between different regions within the rock mass under environmental changes, providing crucial data support for collapse risk prediction. By analyzing the spatial dependency features of the rock mass and time-series data on temperature changes, a collapse risk warning index was generated. This warning index can accurately represent the collapse risk of the rock mass at a specific time point and can be dynamically adjusted based on real-time data.

[0156] Based on the collapse risk warning index and historical data, we calculated the probability of a current collapse and generated the final collapse risk level. By displaying the collapse risk level and warning information to mine management personnel in real time, staff can understand the risk status of each area immediately. The warning information displays low-risk, medium-risk, and high-risk areas and provides corresponding emergency response measures for staff based on the risk level.

[0157] This system successfully addresses the shortcomings of traditional monitoring methods in terms of real-time performance, accuracy, and comprehensiveness. Traditional landslide monitoring often relies on manual inspections and limited sensor data, failing to detect potential hazards in a timely manner. This invention, however, combines multi-source data and advanced algorithms to monitor changes in rock masses in real time and provide timely warnings, offering strong support for decision-makers. By comparing historical and real-time data, we can accurately determine the stability of rock masses based on different meteorological conditions, soil moisture, stress, and other factors, avoiding the uncertainties associated with past assessments relying on experience or single characteristics.

[0158] The table below compares the collapse risk warning index and risk level calculated by the method of this invention and three traditional methods in different areas of the mining area:

[0159] Table 1 Comparison of Rockfall Risk Assessment

[0160]

[0161]

[0162] Traditional Method 1, which assesses risks solely based on temperature changes, fails to adequately consider the spatial dependence of rock masses and factors such as thermal stress. This results in regions A and C still being assessed as medium or low risk, failing to accurately capture the potential hazards in these areas. Traditional Method 2, while assessing risks through soil moisture and stress, does not consider the impact of temperature changes on rock masses, thus failing to accurately predict the risk of all regions, particularly regions B and D, leading to some regions having risk levels lower than the actual situation. Traditional Method 3, based on empirical formulas, while simple, relies too heavily on historical data and cannot cope with the complex changes in actual geological environments, resulting in significant biases in the predictions.

[0163] Using the method of this invention, in regions A and C, because the system considers the comprehensive analysis of multi-source data and extracts the spatial dependence characteristics of the rock mass through topological persistence analysis and spectral convolutional networks, a more accurate collapse risk warning index is obtained and successfully adjusted to a high-risk level, providing timely decision support for mine management personnel. In regions B and D, by combining real-time data with historical data for dynamic early warning, the method of this invention can more accurately identify risk levels and provide early warnings for potential collapses.

[0164] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dangerous rock collapse monitoring and forecasting method based on infrared thermal imaging technology, characterized in that, Includes the following steps: Step 1: Collect multi-source raw data of the dangerous rock area using various hardware devices; Step 2: Preprocess the different types of multi-source raw data separately and extract the data features of different types; Step 3: Based on temperature change characteristics, three-dimensional geometric features, soil moisture and stress characteristics, construct a spatial topology map of the dangerous rock mass. Each small area of ​​the rock mass is regarded as a node in the spatial topology map of the dangerous rock mass. Edges are established between nodes through spatial adjacency, thermal stress and temperature change relationships. Step 4: Apply the improved PersistentHomology algorithm to perform persistence analysis on the spatial topology map of the dangerous rock and generate topological persistence features; The improved PersistentHomology algorithm is applied to perform persistence analysis on the spatial topology map of the unstable rock mass, generating topological persistence features, specifically: The improved PersistentHomology algorithm is used to convert each node in the spatial topology of the dangerous rock into a persistent point. The transformation step specifically involves mapping the topological features of each node to a high-dimensional space using a Laplacian matrix, and using the appearance and disappearance scales of the nodes as coordinates of the persistence values. The coordinates of the persistence value are two-dimensional coordinates reflected by the persistence point in the persistence map. The X coordinate represents the survival time of the node feature at the appearance scale, and the Y coordinate represents the disappearance time of the node feature at the disappearance scale. Based on the two-dimensional coordinates of the persistent points, the Euclidean distance between the persistent points is calculated, and the edges in the persistent graph are established. Based on the persistence map, persistent points connected by edges with persistence values ​​higher than a preset value are selected. The persistence value is the Euclidean distance between persistent points. Persistent points connected by edges with persistence values ​​higher than the preset value represent vulnerable areas of the rock mass and are potential collapse risk points. Topological persistence features are extracted from the persistence points connected to edges with persistence values ​​higher than a preset value. These topological persistence features represent structural changes, crack propagation, and potential collapse risk areas in the rock mass. Step 5: Input the topological persistence features and the spatial topology map of the dangerous rock mass into the spectral convolutional network to extract the spatial dependency features of the rock mass; Step 6: Generate a collapse risk warning index based on the spatial dependence characteristics of the rock mass and the time series data of temperature changes; Step 7: Calculate the probability of a collapse based on the collapse risk warning index and historical collapse data, and generate the final collapse risk level.

2. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, The various hardware devices include infrared thermal imaging equipment, lidar, meteorological sensors, and geological sensors; the multi-source raw data includes infrared thermal imaging image data, lidar point cloud data, meteorological data, and geological data.

3. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, The preprocessing of different types of multi-source raw data and the extraction of different types of data features are specifically as follows: The infrared thermal imaging image is subjected to wavelet transform to remove noise, and the brightness and contrast of the denoised infrared thermal imaging image are enhanced. Temperature change features are extracted and time series data of temperature change are generated. Gaussian filtering is applied to the lidar point cloud data, followed by spatial coordinate alignment and registration. The steps for reconstructing the three-dimensional geometric features of a rock mass using the Poisson reconstruction method include: Calculate the normal vector of each point in the lidar point cloud data; Based on the point cloud normal vectors and point cloud distribution, the surface is solved using the Poisson equation, and a smooth three-dimensional geometric model is generated. Three-dimensional geometric features are extracted from the generated three-dimensional geometric model, including crack features, fault features, protrusion regions, and surface curvature. Z-score standardization was performed on meteorological and geological data to extract soil moisture and stress characteristics.

4. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, Based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics, a spatial topology map of the unstable rock mass is constructed. Each small region of the rock mass is considered a node in the spatial topology map, and edges are established between nodes through spatial adjacency, thermal stress, and temperature change relationships. Specifically: Based on the preset division criteria, the entire dangerous rock area is divided into multiple smaller areas according to geographical location and thermal stress; The characteristics of each small region include the temperature variation of the rock mass, three-dimensional geometry, soil moisture and stress state, which serve as nodal attributes for each small region; Assign a unique node ID to each small region to form a preliminary set of graph nodes; The specific steps for constructing a spatial topology map of dangerous rocks based on the graph node set are as follows: For each node's small region, spatial adjacency is determined. If two small regions are spatially adjacent and have a connecting boundary, a spatial adjacency edge is established between these two nodes. The weight of the spatial adjacency edge is the spatial contact area of ​​the adjacent regions. If the magnitude of thermal stress change in two small regions is less than a preset value, a thermal stress edge is established between these two nodes, and the weight of the thermal stress edge is the difference in thermal stress between the two small regions. If the temperature change between two small regions is less than a preset value, a temperature edge is established between the two nodes, and the weight of the temperature edge is the temperature difference between the two small regions. If the humidity change between two small areas is less than a preset value, a humidity edge is established between the two nodes, and the weight of the humidity edge is the humidity difference between the two small areas. If the stress fluctuation amplitude of two small regions is less than a preset value, a stress edge is established between these two nodes, and the weight of the stress fluctuation edge is the stress difference between the two small regions.

5. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, The process of inputting topological persistence features and the spatial topology map of the unstable rock mass into a spectral graph convolutional network to extract the spatial dependency features of the rock mass is specifically as follows: Based on the spatial adjacency relationships, thermal stress, humidity and temperature changes among nodes in the spatial topology diagram of dangerous rocks, an adjacency matrix of the spatial topology diagram of dangerous rocks is constructed. The adjacency matrix and topological persistence features of the spatial topology map of the dangerous rock mass are input into the graph convolution layer of the spectral graph convolutional network. By performing graph convolution operation on the Laplacian matrix corresponding to the adjacency matrix and topological persistence features, the spatial dependence of thermal stress and temperature between different regions in the rock mass is extracted. After the graph convolution operation in each layer, the Sigmoid activation function is applied to perform a nonlinear transformation on the spatial dependencies, and the pooled spatial dependency features are generated through the pooling layer of the spectral graph convolutional network. Multiple spatial dependency features extracted by the pooling layer are input into the fully connected layer for integration and extraction to generate the final spatial dependency features.

6. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, Based on the spatial dependence characteristics of the rock mass and the time-series data of temperature changes, a collapse risk early warning index is generated, specifically as follows: The time series data of temperature changes are arranged in chronological order to form a time series vector, which represents the temperature change trend of the rock mass in different time periods. Spatial dependency features and temporal data vectors are concatenated to form a comprehensive feature vector that includes both spatial and temporal features; Min-Max normalization is applied to the composite feature vector to scale the feature values ​​to the same range. Each feature in the comprehensive feature vector is assigned a weight according to a preset allocation standard; The standardized feature vectors are weighted and summed with their corresponding weights to obtain the comprehensive feature value of each small region. Based on comprehensive characteristic values, a collapse risk warning index is calculated, which represents the collapse risk of the rock mass at a specific point in time.

7. The method for monitoring and predicting rockfalls based on infrared thermal imaging technology according to claim 1, characterized in that, The process involves calculating the probability of a collapse based on the collapse risk warning index and historical collapse data, and generating a final collapse risk level. Collect historical landslide events and record the environmental characteristics and time of the landslides; The current calculated collapse risk warning index is compared with the collapse risk warning index in historical collapse events, and Bayesian inference is used to calculate the probability of collapse under the current conditions. Based on the calculated probability of collapse, a collapse risk level is generated. The specific steps include: Multiple risk levels are set based on the probability of collapse. A collapse probability of less than 10% is considered low risk and will be subject to routine monitoring. A collapse probability between 10% and 40% is considered medium risk; emergency response equipment should be prepared in advance. A collapse probability greater than 40% is considered high risk. Emergency response procedures should be activated immediately, personnel evacuated, and reinforcement measures implemented. The collapse risk warning index and collapse risk level should be displayed in real time through a graphical interface system.

8. A rockfall monitoring and forecasting device based on infrared thermal imaging technology, comprising the rockfall monitoring and forecasting method based on infrared thermal imaging technology as described in any one of claims 1 to 7, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to acquire multi-source raw data from dangerous rock areas through various hardware devices, and to preprocess different types of raw data to extract features of various types of data. The spatial topology map construction module is used to construct a spatial topology map of dangerous rocks based on temperature change characteristics, three-dimensional geometric features, soil moisture, and stress characteristics. The topology persistence analysis module is used to apply the improved PersistentHomology algorithm to perform topology data analysis and persistence analysis on the spatial topology map of dangerous rocks, and generate topology persistence features. The spatial dependency feature extraction module is used to input topological persistence features and spatial topology maps of dangerous rocks into a spectral convolutional network to extract the spatial dependency features of the rock mass. The collapse risk early warning calculation module is used to generate a collapse risk early warning index based on the spatial dependence characteristics of the rock mass and time series data of temperature changes. The landslide risk level generation module is used to generate the final landslide risk level based on the landslide risk warning index and historical landslide data, and then display it in real time.

Images