Sky-sky-ground landslide monitoring method and system based on deep learning

Abstract

The invention discloses an air-space-ground landslide monitoring method and system based on deep learning, and relates to the technical field of landslide monitoring of mines. The system divides a mine area into a plurality of monitoring sub-areas by establishing a visual man-machine interaction platform and combining a digital geological disaster model; the multi-source data collection covers various sensing data of air-based, space-based and ground-based data, and the consistency and reliability of the data are ensured through standardization, time synchronization and space registration processing; a dynamic landslide monitoring model is constructed by adopting a convolutional neural network, multi-scale spatial-temporal characteristics of a landslide mass are extracted in real time, and accurate identification of landslide physical effect risks is realized; in combination with a geomechanical strain response index, a micro-seismic excitation response coefficient and a fluctuation disturbance abnormal index, a landslide risk is evaluated in multiple dimensions through a threshold judgment mechanism, a risk state and an early warning level are dynamically output, an early warning strategy is generated, mine safety management and emergency response are assisted, and life safety of mining personnel and environmental stability are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of mine landslide monitoring, and specifically to an air-space-ground landslide monitoring method and system based on deep learning. Background Art

[0002] Landslides, a typical geological disaster, frequently occur in mountainous and hilly areas, posing a serious threat to the safety of life and property, as well as the stable operation of regional infrastructure. Traditional landslide monitoring relies primarily on ground observation points and manual inspections. While this method can provide some information on landslide occurrences, it suffers from limitations such as limited coverage, untimely data updates, limited monitoring accuracy, and high labor costs, making it difficult to meet the needs of modern disaster prevention and control.

[0003] Despite rapid developments in remote sensing technology, UAVs, and the Global Navigation Satellite System (GNSS), and the rise of landslide monitoring methods that integrate multi-source data from space, air, and land, existing technologies still have significant shortcomings. They are unable to effectively identify the risk characteristics of geology, landslide masses, and the physical effects of microvibrations, and are unable to identify real-time risks and provide dynamic warnings. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an air-space-ground landslide monitoring method and system based on deep learning to solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method and system for monitoring landslides in air and space based on deep learning, comprising: The landslide monitoring platform establishment module is used to establish a user interface and a visual human-computer interaction window platform for visually displaying digital geological hazard models, receiving risk assessment results, and issuing early warning notifications; and divides the entire mining area into several monitoring sub-areas of uniform size; The multi-source data acquisition module is used to collect multi-source data of the mine monitoring sub-area and establish air-based data groups, space-based data groups and ground-based data groups; The data processing module is used to standardize, time synchronize, and spatially register the collected data to ensure consistency of multi-source data in time and space dimensions; The landslide dynamic monitoring model establishment module uses a convolutional neural network to train a feature extractor that characterizes the physical effect risk of landslides using airborne, space-based, and ground-based data as input. This extracts multi-scale feature vectors and constructs a landslide dynamic monitoring model. The model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. The geomechanical effect identification module is used to extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; The microseismic excitation effect identification module is used to calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitoring sub-area is within the normal range. If not, a strategy is given; The fluctuation disturbance anomaly recognition module is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

[0006] Preferably, the landslide monitoring platform establishment module includes a platform establishment unit and an area division unit; The platform establishment unit is used to establish a user interface and a visual human-computer interaction window platform for visually displaying a digital geological hazard model, receiving risk assessment results and recommended strategies through 5G communication, and issuing early warning notifications through text messages; The area division unit is used to divide the entire mine area into a number of monitoring sub-areas with uniform areas. The M monitoring sub-areas are arranged in the digital geological disaster model according to 、 、 ,..., Mark in sequence.

[0007] Preferably, the multi-source data acquisition module includes: an air-based data acquisition unit, a space-based data acquisition unit and a ground-based data acquisition unit; The air-based data acquisition unit is used to collect data from the monitoring sub-area through a multispectral scanner and a synthetic aperture radar device carried on a remote sensing satellite platform, including: collecting the multispectral reflectance of the observation point in the monitoring sub-area through the multispectral scanner of the remote sensing satellite platform; ; Obtain multi-temporal radar images of the monitoring sub-area by using synthetic aperture radar, and extract historical elevation data using InSAR interferometric processing technology; Collect elevation values in the digital elevation model (DEM) through the lidar system ; and obtain DEM data of multiple historical periods to construct a historical elevation series for the monitoring sub-area; establish an airborne data set; The space-based data acquisition unit is used to collect data including the main crack length Lc and the main crack length growth rate Rlc by using an unmanned aerial vehicle equipped with a high-precision laser radar, an oblique photography camera, and a multispectral camera. The data collected include: collecting the main crack length Lc and the main crack length growth rate Rlc by using orthophoto and stereo image processing technology; collecting the texture features and orthophoto image data of the crack area by using a multispectral and texture recognition camera; and establishing a space-based data set. The ground data acquisition unit is used to collect ground physical response data through fixed sensing equipment such as GNSS displacement monitors, ground strain gauges and seismometers deployed in the landslide area: collecting three-dimensional displacement rate Vd through GNSS displacement monitors; collecting ground strain tensors through ground strain gauges. ; Collect microseismic magnitude M and frequency energy spectrum density change value E through seismometers and accelerometers; establish a foundation data set.

[0008] Preferably, the data processing module is used to perform format standardization, time synchronization and spatial registration processing on the collected air-based, space-based and ground-based data, including: position registration of the multispectral reflectivity obtained by remote sensing satellites, sequence reconstruction and average extraction of historical elevation data obtained by InSAR and laser altimeters; image superposition and time alignment of DEM elevation values and main crack data obtained by drones; denoising and time series completion processing of GNSS three-dimensional displacement rate, ground strain tensor and frequency energy spectral density change values, so as to ensure consistency of multi-source data in time and space dimensions.

[0009] Preferably, the landslide dynamic monitoring model establishment module is used to construct an initial deep learning model using a convolutional neural network, with an air-based data group, a space-based data group and a ground-based data group as input, corresponding to the physical effect risk of geomechanical strain response index, microseismic excitation response coefficient and fluctuation disturbance anomaly index; the initial deep learning model is trained and tested using data to obtain a basic feature extractor that can distinguish the physical effect risk of landslides; the output of the intermediate layer of the initial deep learning model is then intercepted as a multi-dimensional feature vector to characterize the multi-scale spatiotemporal characteristics of the landslide body; the feature vector is used as input and trained again. After the training is completed, it is used as a landslide dynamic monitoring model for real-time data operation, and the landslide risk state vector, local abnormal heat zone map and warning level are output online to provide fast and dynamic risk identification and warning support for the air-ground landslide monitoring system.

[0010] Preferably, the geomechanical effect identification module includes: a first calculation unit and a first analysis unit; The first calculation unit is used to extract the three-dimensional displacement rate Vd and the ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, after dimensionless processing, the geomechanical strain response index DLY is calculated; The first analysis unit is configured to preset a first threshold Q1 and compare and analyze the geomechanical strain response index DLY with the first threshold Q1 to obtain a first evaluation result, including: When the geomechanical strain response index DLY is less than the first threshold Q1, it indicates that the geological deformation controllability of the monitoring sub-area is qualified, there is no sign of landslide, and continuous monitoring is required; When the geomechanical strain response index DLY ≥ the first threshold Q1, it indicates that the geological deformation controllability of the monitored sub-area is unqualified and there is a landslide trend. The first early warning instruction is triggered and the first strategy is generated: mark the abnormal area; notify the mine dispatcher; generate a 24-hour displacement prediction map; report to the landslide monitoring platform, notify the suspension of mining operations and deploy stress relief holes.

[0011] Preferably, the microseismic excitation effect identification module includes a second calculation unit and a second analysis unit; The second calculation unit is used to calculate and obtain the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data group and performing dimensionless processing; The second analysis unit is configured to obtain a second evaluation result by presetting a second threshold Q2 and comparing and analyzing the microseismic excitation response coefficient WJX with the second threshold Q2. The result includes: When the microseismic excitation response coefficient WJX is less than the second threshold Q2, it indicates that the vibration activity in the monitoring sub-area is within the normal range, the rock mass is stable, and continuous monitoring is required; When the microseismic excitation response coefficient WJX ≥ the second threshold Q2, it indicates that the vibration activity in the monitored sub-area is not within the normal range and there is a risk of internal cracking and slippage. The second warning instruction is triggered and the second strategy is generated: increase the sampling frequency of microseismic data; issue an orange warning; draw a hot zone map of the earthquake source; report to the landslide monitoring platform, and notify personnel to evacuate the threatened area.

[0012] Preferably, the fluctuation disturbance anomaly identification module includes a third calculation unit and a third analysis unit; The third calculation unit is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical high-speed series, after dimensionless processing, the fluctuation disturbance anomaly index BRY is calculated.

[0013] Preferably, the third analysis unit is configured to preset a third threshold Q3 and compare and analyze the fluctuation disturbance anomaly index BRY with the third threshold Q3 to obtain a third evaluation result, including: When the fluctuation disturbance anomaly index BRY is less than the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is stable, has no deformation trend, and needs to be continuously monitored; When the fluctuation disturbance anomaly index BRY ≥ the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is unstable and has a deformation trend, triggering the third warning instruction and generating the third strategy: constructing a time series , obtain the growth rate of the fluctuation disturbance anomaly index, analyze the duration of continuous anomalies across time windows, and analyze the extent of the disaster impact; report to the landslide monitoring platform, and provide specific measures such as time-limited inspections, personnel evacuation, and equipment reinforcement.

[0014] Preferably, the air-ground landslide monitoring method based on deep learning comprises the following steps: Step 1: Establish a user interface and a visual human-computer interaction window platform to visualize the digital geological hazard model, receive risk assessment results, and issue early warning notifications; and divide the entire mining area into several monitoring sub-areas of uniform size; Step 2: Collect multi-source data of the mine monitoring sub-area and establish air-based data group, space-based data group and ground-based data group; Step 3: Standardize, synchronize time, and register space for the collected data to ensure consistency of multi-source data in time and space. Step 4: Using a convolutional neural network, with airborne, space-based, and ground-based data as input, a feature extractor is trained to characterize the risk of landslide physical effects. This extracts multi-scale feature vectors and constructs a dynamic landslide monitoring model. This model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. Step 5: Extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; Step 6: Calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitored sub-area is within the normal range. If not, a strategy is given; Step 7: Extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

[0015] The present invention provides a method and system for monitoring landslides in air, space, and land based on deep learning. It has the following beneficial effects: (1) This deep learning-based air-space-ground landslide monitoring method and system, by constructing three types of data groups, namely air-based, space-based and ground-based, and comprehensively utilizing remote sensing satellite, UAV and ground sensor data, realizes the full-space, multi-scale and multi-time data collection and synchronous processing of landslide areas, significantly improving the spatial coverage and data acquisition accuracy of landslide monitoring, and providing solid data support for the accurate identification of geological disasters.

[0016] (2) This deep learning-based air-space-ground landslide monitoring method and system uses a convolutional neural network as the basis to construct a dynamic landslide monitoring model. It uses multi-source data to train a feature extractor to extract high-dimensional vectors that characterize the multi-scale evolution of landslides, thereby realizing rapid identification of landslide physical effect risks, early warning classification, and location of abnormal hot spots, greatly improving the speed and accuracy of landslide early warning response.

[0017] (3) This deep learning-based air-space-ground landslide monitoring method and system constructs three types of physical effect indicators: geomechanical strain response index DLY, microseismic excitation response coefficient WJX, and wave disturbance anomaly index BRY. It integrates multi-source information such as GNSS displacement, ground strain, microseismic, crack evolution, and spectral elevation to establish a risk judgment criterion that is highly coupled with the actual landslide process, effectively realizing the multi-dimensional dynamic identification of the landslide disaster process and the generation of intelligent response strategies.

[0018] (4) This deep learning-based air-space-ground landslide monitoring method and system, by building a landslide monitoring platform, integrates digital geological hazard model display, real-time risk assessment feedback and early warning information release functions to form a complete closed-loop feedback mechanism for early warning information. It not only enhances the coordination ability of mine scheduling and emergency response, but also provides a visual, easy-to-operate and high-efficiency technical path for the normalized intelligent supervision of landslide risks in mining areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a block diagram and flow chart of the air-ground landslide monitoring system based on deep learning of the present invention; Figure 2 This is a schematic diagram of the steps of the air-space-ground landslide monitoring method based on deep learning of the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Example 1 See also Figure 1The present invention provides a method and system for monitoring landslides in air, space and land based on deep learning, including: The landslide monitoring platform establishment module is used to establish a user interface and a visual human-computer interaction window platform for visually displaying digital geological hazard models, receiving risk assessment results, and issuing early warning notifications; and divides the entire mining area into several monitoring sub-areas of uniform size; The multi-source data acquisition module is used to collect multi-source data of the mine monitoring sub-area and establish air-based data groups, space-based data groups and ground-based data groups; The data processing module is used to standardize, time synchronize, and spatially register the collected data to ensure consistency of multi-source data in time and space dimensions; The landslide dynamic monitoring model establishment module uses a convolutional neural network to train a feature extractor that characterizes the physical effect risk of landslides using airborne, space-based, and ground-based data as input. This extracts multi-scale feature vectors and constructs a landslide dynamic monitoring model. The model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. The geomechanical effect identification module is used to extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; The microseismic excitation effect identification module is used to calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitoring sub-area is within the normal range. If not, a strategy is given; The fluctuation disturbance anomaly recognition module is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

[0022] In this embodiment, a human-machine platform with a visual interactive interface is constructed through the landslide monitoring platform establishment module to achieve a unified division of the entire mine area and an intuitive display of the landslide risk status; a multi-source data acquisition module is used to establish three types of data groups: air-based, space-based, and ground-based, to ensure the comprehensiveness and multi-scale nature of the information source; the data processing module is used to achieve standardization, time synchronization, and spatial registration processing of multi-source heterogeneous data, which significantly improves the consistency and fusion quality of the data; on this basis, the landslide dynamic monitoring model establishment module introduces a convolutional neural network structure to integrate the three types of air-space-ground data for The system extracts multi-scale features and outputs real-time risk state vectors, abnormal heat map, and warning level, achieving in-depth identification and dynamic early warning of landslide risks. Furthermore, the system includes a geomechanical effect identification module, a microseismic excitation effect identification module, and a wave disturbance anomaly identification module. These modules construct DLY, WJX, and BRY physical effect risk indicators based on multiple physical quantities, including three-dimensional displacement rate, crack evolution, microseismic signals, and reflectivity fluctuations. These indicators are then dynamically compared and analyzed with corresponding thresholds Q1, Q2, and Q3. This not only enables precise identification of landslide hazards but also provides targeted risk control strategies. The synergistic effect of these modules significantly improves the landslide monitoring system's early warning accuracy, response speed, and regional positioning capabilities for landslide hazards in complex geological environments.

[0023] Example 2 This embodiment is explained in Example 1. Specifically, the landslide monitoring platform establishment module includes a platform establishment unit and an area division unit; The platform establishment unit is used to establish a user interface and a visual human-computer interaction window platform for visually displaying a digital geological hazard model, receiving risk assessment results and recommended strategies through 5G communication, and issuing early warning notifications through text messages; The area division unit is used to divide the entire mine area into a number of monitoring sub-areas with uniform areas. The M monitoring sub-areas are arranged in the digital geological disaster model according to 、 、 ,..., Mark in sequence.

[0024] In this embodiment, by setting up a landslide monitoring platform establishment module, the system realizes the visual construction and efficient human-computer interaction of the monitoring platform; among them, the platform establishment unit supports the reception of risk assessment results and recommended strategies based on 5G communication technology, and quickly issues early warning notifications to relevant responsible persons via text messages, significantly improving the transmission efficiency and coverage of early warning information; at the same time, the platform uses a digital geological hazard model as a carrier to intuitively display the landslide risk status of the mining area, thereby improving the decision-making efficiency of managers; the regional division unit can automatically divide the mining area into M monitoring sub-areas of uniform area, and number and mark them in sequence in the digital model, so that subsequent risk analysis, positioning tracking and response strategies can be finely executed in the sub-area dimension, effectively supporting the zoning management and refined control of landslide risks.

[0025] Example 3 This embodiment is explained in Example 2. Specifically, the multi-source data acquisition module includes: an air-based data acquisition unit, a space-based data acquisition unit, and a ground-based data acquisition unit; The air-based data acquisition unit is used to collect data from the monitoring sub-area through a multispectral scanner and a synthetic aperture radar device carried on a remote sensing satellite platform, including: collecting the multispectral reflectance of the observation point in the monitoring sub-area through the multispectral scanner of the remote sensing satellite platform; ; Obtain multi-temporal radar images of the monitoring sub-area by using synthetic aperture radar, and extract historical elevation data using InSAR interferometric processing technology; Collect elevation values in the digital elevation model (DEM) through the lidar system ; and obtain DEM data of multiple historical periods to construct a historical elevation series for the monitoring sub-area; establish an airborne data set; The space-based data acquisition unit is used to collect data including the main crack length Lc and the main crack length growth rate Rlc by using an unmanned aerial vehicle equipped with a high-precision laser radar, an oblique photography camera, and a multispectral camera. The data collected include: collecting the main crack length Lc and the main crack length growth rate Rlc by using orthophoto and stereo image processing technology; collecting the texture features and orthophoto image data of the crack area by using a multispectral and texture recognition camera; and establishing a space-based data set. The ground data acquisition unit is used to collect ground physical response data through fixed sensing equipment such as GNSS displacement monitors, ground strain gauges and seismometers deployed in the landslide area: collecting three-dimensional displacement rate Vd through the GNSS displacement monitor; collecting tensor Vd through the ground strain gauge ; Collect microseismic magnitude M and frequency energy spectrum density change value E through seismometers and accelerometers; establish a foundation data set.

[0026] In this embodiment, through the coordinated deployment of three types of data acquisition units: air-based, space-based, and ground-based, the present invention can simultaneously obtain key elements such as remote sensing reflectivity, historical elevation series, DEM elevation values, main crack evolution information, GNSS displacement, ground strain, and microseismic energy spectrum, construct a complete air-space-ground multi-source data set, significantly improve the data dimension, spatiotemporal resolution, and accuracy of landslide monitoring, and provide a comprehensive and reliable data foundation for subsequent dynamic risk identification and early warning decisions.

[0027] Example 4 This embodiment is explained in Example 3. Specifically, the data processing module is used to perform format standardization, time synchronization and spatial registration processing on the collected air-based, space-based and ground-based data, including: position registration of the multispectral reflectivity obtained by remote sensing satellites, sequence reconstruction and average extraction of historical elevation data obtained by InSAR and laser altimeters; image superposition and time alignment of DEM elevation values and main crack data obtained by drones; denoising and time series completion processing of GNSS three-dimensional displacement rate, ground strain tensor and frequency energy spectral density change values, so as to ensure consistency of multi-source data in time and space dimensions.

[0028] In this embodiment, by implementing unified format standardization, precise time synchronization and spatial alignment for air-based, space-based and ground-based data, the data processing module of the present invention effectively eliminates the deviations in time resolution and spatial coordinates of multi-source observations, allowing multispectral reflectivity, historical elevation series, DEM and displacement-microseismic data to be seamlessly integrated under the same time and space benchmark, significantly improving the accuracy and reliability of deep learning model training.

[0029] Example 5 This embodiment is an explanation of the fourth embodiment. Specifically, the landslide dynamic monitoring model establishment module is used to construct an initial deep learning model using a convolutional neural network, with an air-based data group, a space-based data group and a ground-based data group as inputs, corresponding to the physical effect risks of the geomechanical strain response index, the microseismic excitation response coefficient and the fluctuation disturbance anomaly index; the initial deep learning model is trained and tested using data to obtain a basic feature extractor that can distinguish the physical effect risks of the landslide; the output of the intermediate layer of the initial deep learning model is then intercepted as a multi-dimensional feature vector to characterize the multi-scale spatiotemporal characteristics of the landslide body; the feature vector is used as input and trained again. After the training is completed, it is used as a landslide dynamic monitoring model for real-time data operation, and the landslide risk state vector, the local abnormal heat zone map and the warning level are output online, providing fast and dynamic risk identification and warning support for the air-ground landslide monitoring system.

[0030] In this embodiment, a dynamic landslide monitoring model jointly trained with air-based, space-based, and ground-based data sets can instantly output landslide risk state vectors and abnormal heat zone maps during online operation, enabling rapid identification and early warning classification of multi-source physical effects of landslides in mining areas, significantly improving the real-time performance and prediction accuracy of the landslide monitoring system.

[0031] Example 6 This embodiment is explained in Example 5. Specifically, the geomechanical effect identification module includes: a first calculation unit and a first analysis unit; The first calculation unit is used to extract the three-dimensional displacement rate Vd and the ground strain tensor of the foundation data set , combined with the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, after dimensionless processing, the geomechanical strain response index DLY is calculated, and the formula is as follows: ; Where A represents the total area of the monitoring sub-region, n represents the total number of sampling points in the monitoring sub-region, represents the GNSS three-dimensional displacement rate of the i-th sampling point in unit time, represents the gradient value of the ground strain tensor at the i-th sampling point, It represents the main crack growth sensitivity coefficient. In the laboratory rock sample tension-shear-compression shear test and the field in-situ crack extension monitoring test, the coupling relationship between the crack length increment dLc / dt and the synchronous GNSS displacement rate and the ground strain gradient was measured. The sensitivity slope of the crack growth to the integrated displacement-strain effect was obtained by least squares fitting. Then, the fractures were grouped according to lithology, joint density and water content. The weighted average of the fitting results was taken and dimensionless processing was performed to obtain , as a quantitative coefficient of the contribution of crack propagation to the geological strain index under different rock mass types, represents the length growth rate of the main crack adjacent to the i-th sampling point at two time points; ; Where, represents the main crack length of the i-th crack at time t2, represents the main crack length of the i-th crack at time t1, Indicates the time interval between two samples; The first analysis unit is configured to preset a first threshold Q1 and compare and analyze the geomechanical strain response index DLY with the first threshold Q1 to obtain a first evaluation result, including: When the geomechanical strain response index DLY is less than the first threshold Q1, it indicates that the geological deformation controllability of the monitoring sub-area is qualified, there is no sign of landslide, and continuous monitoring is required; When the geomechanical strain response index DLY ≥ the first threshold Q1, it indicates that the geological deformation controllability of the monitored sub-area is unqualified and there is a landslide trend. The first early warning instruction is triggered and the first strategy is generated: mark the abnormal area; notify the mine dispatcher; generate a 24-hour displacement prediction map; report to the landslide monitoring platform, notify the suspension of mining operations and deploy stress relief holes.

[0032] The first threshold Q1 is obtained by statistically analyzing the geomechanical strain response index DLY in a large number of historical samples of mine landslide monitoring. The typical distribution intervals of the "deformation controllable stage" and the "landslide incubation stage" are extracted respectively. Combined with the green mine construction specifications, slope engineering safety technical standards and expert experience, the critical dividing point that can distinguish the two stages is determined. Then, with reference to the slope strain alarm limit given by the industry and the safe operation parameters provided by the mine design unit, the upper 95% confidence bound of the joint distribution is taken as the first threshold Q1, which is used to determine whether the geological deformation in the monitored sub-area exceeds the limit.

[0033] In this embodiment, the geomechanical effect identification module enables fine-grained physical response analysis of the geological structure within the landslide area by integrating GNSS three-dimensional displacement rates and ground strain tensor information from ground-based data sets with the main crack length and growth rate from space-based data sets, and constructing a geomechanical strain response index (DLY) based on dimensionless processing. Compared to traditional monitoring methods that rely solely on surface deformation, this module, by leveraging the coupled response of the ground surface and crack evolution, can more realistically reflect internal structural deformation trends and strain concentration areas. Furthermore, combined with a first threshold Q1, it enables real-time, graded assessment of geomechanical conditions. When DLY exceeds the threshold, an intelligent landslide trend identification mechanism is automatically triggered, and specific strategies are generated, including abnormal area marking, displacement prediction map generation, scheduling notifications, and mining suspension instructions. This shifts risk management from a "delayed response" to an "active warning" approach, significantly improving the intelligence, precision, and timely response of the landslide monitoring system in dynamic and complex geological environments, enhancing the mine's overall geological hazard risk prevention and control capabilities and emergency command efficiency.

[0034] Example 7 This embodiment is explained in Example 5. Specifically, the microseismic excitation effect identification module includes a second calculation unit and a second analysis unit; The second calculation unit is used to extract the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data group, and calculate the microseismic excitation response coefficient WJX after dimensionless processing. The formula is as follows:

[0035] Where m represents the number of sampling points of microseismic data in the total sampling time, represents the microseismic magnitude collected in the jth time segment, represents the change value of the frequency energy spectrum density in the jth time segment, w1 and w2 represent the weight coefficients, , ,and ; Method for obtaining w1 and w2: Based on a statistical analysis of the microseismic magnitude M and frequency energy spectral density E in different landslide events, a multivariate regression approach combined with expert experience was used to determine the weighting factors w1 and w2. This weighted combination maximizes the impact of microseismic activity on landslide risk. The weighting factors were set based on relevant geological monitoring standards and research results on landslide physical mechanisms to ensure the sensitivity and stability of the indicators and achieve a scientific and quantitative evaluation of microseismic excitation effects.

[0036] The second analysis unit is configured to obtain a second evaluation result by presetting a second threshold Q2 and comparing and analyzing the microseismic excitation response coefficient WJX with the second threshold Q2. The result includes: When the microseismic excitation response coefficient WJX is less than the second threshold Q2, it indicates that the vibration activity in the monitoring sub-area is within the normal range, the rock mass is stable, and continuous monitoring is required; When the microseismic excitation response coefficient WJX ≥ the second threshold Q2, it indicates that the vibration activity in the monitored sub-area is not within the normal range and there is a risk of internal cracking and slippage. The second warning instruction is triggered and the second strategy is generated: increase the sampling frequency of microseismic data; issue an orange warning; draw a hot zone map of the earthquake source; report to the landslide monitoring platform, and notify personnel to evacuate the threatened area.

[0037] The second threshold, Q2, is determined by long-term monitoring and analysis of microseismic data from a large number of mine landslide areas. The range of microseismic magnitude and frequency energy spectral density variations under normal vibration conditions and during periods of abnormal microseismic excitation is statistically analyzed. This is combined with the experience of geomechanics experts and landslide hazard early warning standards to determine a reasonable microseismic excitation response threshold. This threshold, developed with reference to relevant national microseismic monitoring specifications for geological hazards, technical guidelines for landslide early warning, and mine safety regulations, accurately distinguishes normal vibration from abnormal microseismic excitation, thereby enhancing the scientific nature of landslide risk identification and the timeliness of early warnings.

[0038] In this embodiment, a microseismic excitation response coefficient WJX is constructed based on the change in foundation microseismic magnitude and frequency energy spectral density. The microseismic excitation effect identification module can accurately reflect the dynamic changes in microseismic activity within the monitored sub-area and its impact on rock mass stability. By setting a reasonable second threshold Q2, the module achieves real-time automatic identification and risk classification of microseismic anomalies, effectively distinguishing normal vibration from potential slip risk states, and triggering a rapid response mechanism. This mechanism includes increasing data sampling frequency, issuing orange warnings, generating earthquake source hot zone maps, and issuing timely personnel evacuation instructions. This significantly enhances the sensitivity and response speed of landslide risk warnings, ensures the safety of personnel in the mining area and the stability of the production environment, and improves the system's recognition accuracy and emergency response capabilities for complex microseismic disturbance events.

[0039] Example 8 This embodiment is explained in Example 7. Specifically, the fluctuation disturbance anomaly identification module includes a third calculation unit and a third analysis unit; The third calculation unit is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical high-speed series, after dimensionless processing, the fluctuation disturbance anomaly index BRY is calculated, and the formula is as follows:

[0040] Where N represents the total number of observation points for multispectral reflectance and DEM elevation values in the monitoring sub-area. represents the multispectral reflectance of the k-th observation point, It represents the mean reflectivity of the channel corresponding to the k-th point in the long-term historical monitoring. Indicates the elevation value of the kth point, Represents the historical average elevation value of the k-th location.

[0041] In this embodiment, based on the spatiotemporal variation characteristics of multispectral reflectance and elevation values, the fluctuation disturbance anomaly identification module can scientifically calculate the fluctuation disturbance anomaly index (BRY), accurately reflecting the stability changes of the landslide structure in the monitored sub-area. By combining historical elevation series for dimensionless processing and anomaly index calculation, the module achieves dynamic monitoring and anomaly identification of surface and internal disturbances of the landslide, effectively capturing potential structural damage and deformation trends. This module can issue timely warnings, assist in the development of targeted stability maintenance strategies, improve the accuracy and response efficiency of landslide risk warnings, and ensure mine safety and environmental stability.

[0042] Example 9 This embodiment is explained in Example 8. Specifically, the third analysis unit is used to preset a third threshold Q3 and compare and analyze the fluctuation disturbance abnormality index BRY with the third threshold Q3 to obtain a third evaluation result, including: When the fluctuation disturbance anomaly index BRY is less than the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is stable, has no deformation trend, and needs to be continuously monitored; When the fluctuation disturbance anomaly index BRY ≥ the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is unstable and has a deformation trend, triggering the third warning instruction and generating the third strategy: constructing a time series , obtain the growth rate of the fluctuation disturbance anomaly index, analyze the duration of continuous anomalies across time windows, and analyze the extent of the disaster impact; report to the landslide monitoring platform, and provide specific measures such as time-limited inspections, personnel evacuation, and equipment reinforcement.

[0043] In this embodiment, by setting a reasonable third threshold value Q3, the third analysis unit can accurately determine the stability of the landslide structure in the monitored sub-area, enabling dynamic monitoring and real-time assessment of the fluctuation disturbance anomaly index (BRY). This module not only promptly identifies structural instability and deformation trends but also conducts multi-dimensional disaster impact analysis based on the growth rate and duration of the anomaly index, assisting in the formulation of scientific emergency response strategies. By recommending measures such as time-limited inspections, personnel evacuation, and equipment reinforcement, the risk of landslide disasters can be effectively reduced, ensuring the safety of mine operations and the safety of life and property, and improving the practicality and responsiveness of the early warning system.

[0044] Example 10 For a deep learning-based landslide monitoring method, please refer to Figure 2 , including the following steps: Step 1: Establish a user interface and a visual human-computer interaction window platform to visualize the digital geological hazard model, receive risk assessment results, and issue early warning notifications; and divide the entire mining area into several monitoring sub-areas of uniform size; Step 2: Collect multi-source data of the mine monitoring sub-area and establish air-based data group, space-based data group and ground-based data group; Step 3: Standardize, synchronize time, and register space for the collected data to ensure consistency of multi-source data in time and space. Step 4: Using a convolutional neural network, with airborne, space-based, and ground-based data as input, a feature extractor is trained to characterize the risk of landslide physical effects. This extracts multi-scale feature vectors and constructs a dynamic landslide monitoring model. This model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. Step 5: Extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; Step 6: Calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitored sub-area is within the normal range. If not, a strategy is given; Step 7: Extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

[0045] In this embodiment, by systematically executing the above steps, the present invention realizes all-round, multi-level dynamic monitoring and intelligent early warning of mine landslide risks. Specifically, by utilizing the spatiotemporal fusion and deep learning technology of multi-source heterogeneous data, an accurate dynamic landslide monitoring model is constructed, which can reflect the physical effect risk status of the landslide body in real time; combining a variety of physical response indices and threshold judgment mechanisms, the geomechanical deformation, microseismic excitation and wave disturbance anomalies of the landslide body are scientifically evaluated to realize multi-dimensional identification and graded early warning of risks. This method significantly improves the accuracy and response speed of landslide risk identification, provides timely and effective decision-making support and emergency strategy guarantees for mine safety production, effectively reduces the probability of landslide disasters and their potential losses, and ensures the safety of personnel and the stability of the mine environment.

[0046] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each set of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0047] The above formulas are obtained by collecting a large amount of data and performing software simulation, and a formula close to the actual value is selected. The coefficients in the formula are set by those skilled in the art according to actual conditions. The above is only a preferred specific implementation method of the present invention, but the protection scope of the present invention is not limited to this. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solution and inventive concept of the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The air-ground landslide monitoring system based on deep learning is characterized by: include: The landslide monitoring platform establishment module is used to establish a user interface and a visual human-computer interaction window platform for visually displaying digital geological hazard models, receiving risk assessment results, and issuing early warning notifications; and divides the entire mining area into several monitoring sub-areas of uniform size; The multi-source data acquisition module is used to collect multi-source data of the mine monitoring sub-area and establish air-based data groups, space-based data groups and ground-based data groups; The data processing module is used to standardize, time synchronize, and spatially register the collected data to ensure consistency of multi-source data in time and space dimensions; The landslide dynamic monitoring model establishment module uses a convolutional neural network to train a feature extractor that characterizes the physical effect risk of landslides using airborne, space-based, and ground-based data as input. This extracts multi-scale feature vectors and constructs a landslide dynamic monitoring model. The model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. The geomechanical effect identification module is used to extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; The microseismic excitation effect identification module is used to calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitoring sub-area is within the normal range. If not, a strategy is given; The fluctuation disturbance anomaly recognition module is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

2. The deep learning-based air-space-ground landslide monitoring system according to claim 1, characterized in that: The landslide monitoring platform establishment module includes a platform establishment unit and an area division unit; The platform establishment unit is used to establish a user interface and a visual human-computer interaction window platform for visually displaying a digital geological hazard model, receiving risk assessment results and recommended strategies through 5G communication, and issuing early warning notifications through text messages; The area division unit is used to divide the entire mine area into a number of monitoring sub-areas with uniform areas. The M monitoring sub-areas are arranged in the digital geological disaster model according to 、 、 ,..., Mark in sequence.

3. The deep learning-based air-space-ground landslide monitoring system according to claim 2, characterized in that: The multi-source data acquisition module includes: an air-based data acquisition unit, a space-based data acquisition unit and a ground-based data acquisition unit; The air-based data acquisition unit is used to collect data from the monitoring sub-area through a multispectral scanner and a synthetic aperture radar device carried on a remote sensing satellite platform, including: collecting the multispectral reflectance of the observation point in the monitoring sub-area through the multispectral scanner of the remote sensing satellite platform; ; Obtain multi-temporal radar images of the monitoring sub-area by using synthetic aperture radar, and extract historical elevation data using InSAR interferometric processing technology; Collect elevation values in the digital elevation model (DEM) through the lidar system ; and obtain DEM data of multiple historical periods to construct a historical elevation series for the monitoring sub-area; establish an airborne data set; The space-based data acquisition unit is used to collect data including the main crack length Lc and the main crack length growth rate Rlc by using an unmanned aerial vehicle equipped with a high-precision laser radar, an oblique photography camera, and a multispectral camera. The data collected include: collecting the main crack length Lc and the main crack length growth rate Rlc by using orthophoto and stereo image processing technology; collecting the texture features and orthophoto image data of the crack area by using a multispectral and texture recognition camera; and establishing a space-based data set. The ground data acquisition unit is used to collect ground physical response data through fixed sensing equipment such as GNSS displacement monitors, ground strain gauges and seismometers deployed in the landslide area: collecting three-dimensional displacement rate Vd through GNSS displacement monitors; collecting ground strain tensors through ground strain gauges. ; Collect microseismic magnitude M and frequency energy spectrum density change value E through seismometers and accelerometers; establish a foundation data set.

4. The deep learning-based air-space-ground landslide monitoring system according to claim 3 is characterized in that: The data processing module is used to standardize the format, synchronize time, and perform spatial registration processing on the collected air-based, space-based, and ground-based data. This includes: position registration of multispectral reflectance acquired by remote sensing satellites; sequence reconstruction and average extraction of historical elevation data acquired by InSAR and laser altimeters; image superposition and time alignment of DEM elevation values and main crack data acquired by drones; and denoising and time series completion of GNSS three-dimensional displacement rates, ground strain tensors, and frequency energy spectral density change values, to ensure consistency of multi-source data in time and space.

5. The deep learning-based air-space-ground landslide monitoring system according to claim 4 is characterized in that: The landslide dynamic monitoring model establishment module is used to construct an initial deep learning model using a convolutional neural network, taking an air-based data set, a space-based data set, and a ground-based data set as input, corresponding to the physical effect risk of the geomechanical strain response index, the microseismic excitation response coefficient, and the fluctuation disturbance anomaly index; using the data to train and test the initial deep learning model, a basic feature extractor that can distinguish the physical effect risk of the landslide is obtained; and then intercepting the intermediate layer output of the initial deep learning model as a multi-dimensional feature vector to characterize the multi-scale spatiotemporal characteristics of the landslide body; The system takes the feature vector as input and trains again. After the training is completed, it runs real-time data as a dynamic landslide monitoring model, outputs landslide risk state vectors, local abnormal heat zone maps and warning levels online, and provides fast and dynamic risk identification and warning support for the air-space-ground landslide monitoring system.

6. The deep learning-based air-space-ground landslide monitoring system according to claim 5, characterized in that: The geomechanical effect identification module includes: a first calculation unit and a first analysis unit; The first calculation unit is used to extract the three-dimensional displacement rate Vd and the ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, after dimensionless processing, the geomechanical strain response index DLY is calculated; The first analysis unit is configured to preset a first threshold Q1 and compare and analyze the geomechanical strain response index DLY with the first threshold Q1 to obtain a first evaluation result, including: When the geomechanical strain response index DLY is less than the first threshold Q1, it indicates that the geological deformation controllability of the monitoring sub-area is qualified, there is no sign of landslide, and continuous monitoring is required; When the geomechanical strain response index DLY ≥ the first threshold Q1, it indicates that the geological deformation controllability of the monitored sub-area is unqualified and there is a landslide trend. The first early warning instruction is triggered and the first strategy is generated: mark the abnormal area; notify the mine dispatcher; generate a 24-hour displacement prediction map; report to the landslide monitoring platform, notify the suspension of mining operations and deploy stress relief holes.

7. The deep learning-based air-space-ground landslide monitoring system according to claim 5, characterized in that: The microseismic excitation effect identification module includes a second calculation unit and a second analysis unit; The second calculation unit is used to calculate and obtain the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data group and performing dimensionless processing; The second analysis unit is configured to obtain a second evaluation result by presetting a second threshold Q2 and comparing and analyzing the microseismic excitation response coefficient WJX with the second threshold Q2. The result includes: When the microseismic excitation response coefficient WJX is less than the second threshold Q2, it indicates that the vibration activity in the monitoring sub-area is within the normal range, the rock mass is stable, and continuous monitoring is required; When the microseismic excitation response coefficient WJX ≥ the second threshold Q2, it indicates that the vibration activity in the monitored sub-area is not within the normal range and there is a risk of internal cracking and slippage. The second warning instruction is triggered and the second strategy is generated: increase the sampling frequency of microseismic data; issue an orange warning; draw a hot zone map of the earthquake source; report to the landslide monitoring platform, and notify personnel to evacuate the threatened area.

8. The deep learning-based air-space-ground landslide monitoring system according to claim 5, characterized in that: The fluctuation disturbance anomaly identification module includes a third calculation unit and a third analysis unit; The third calculation unit is used to extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical high-speed series, after dimensionless processing, the fluctuation disturbance anomaly index BRY is calculated.

9. The deep learning-based air-space-ground landslide monitoring system according to claim 8, characterized in that: The third analysis unit is configured to preset a third threshold Q3 and compare and analyze the fluctuation disturbance anomaly index BRY with the third threshold Q3 to obtain a third evaluation result, including: When the fluctuation disturbance anomaly index BRY is less than the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is stable, has no deformation trend, and needs to be continuously monitored; When the fluctuation disturbance anomaly index BRY ≥ the third threshold Q3, it indicates that the landslide structure in the monitoring sub-area is unstable and has a deformation trend, triggering the third warning instruction and generating the third strategy: constructing a time series , obtain the growth rate of the fluctuation disturbance anomaly index, analyze the duration of continuous anomalies across time windows, and analyze the extent of the disaster impact; report to the landslide monitoring platform, and provide specific measures such as time-limited inspections, personnel evacuation, and equipment reinforcement.

10. A method for monitoring air-to-ground landslides based on deep learning, comprising the air-to-ground landslide monitoring system based on deep learning according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Establish a user interface and a visual human-computer interaction window platform to visualize the digital geological hazard model, receive risk assessment results, and issue early warning notifications; and divide the entire mining area into several monitoring sub-areas of uniform size; Step 2: Collect multi-source data of the mine monitoring sub-area and establish air-based data group, space-based data group and ground-based data group; Step 3: Standardize, synchronize time, and register space for the collected data to ensure consistency of multi-source data in time and space. Step 4: Using a convolutional neural network, with airborne, space-based, and ground-based data as input, a feature extractor is trained to characterize the risk of landslide physical effects. This extracts multi-scale feature vectors and constructs a dynamic landslide monitoring model. This model outputs risk state vectors, abnormal heat map, and warning level in real time, enabling rapid identification and dynamic warning of landslide risks. Step 5: Extract the three-dimensional displacement rate Vd and ground strain tensor of the foundation data set , combining the main crack length Lc and the main crack length growth rate Rlc of the space-based data set, calculate the geomechanical strain response index DLY, and compare it with the first threshold Q1 to determine whether the geological deformation controllability of the monitored sub-area is qualified. If it is unqualified, a strategy is given; Step 6: Calculate the microseismic excitation response coefficient WJX by extracting the microseismic magnitude M and the frequency energy spectrum density change value E of the foundation data set, and compare and analyze it with the second threshold Q2 to determine whether the vibration activity in the monitored sub-area is within the normal range. If not, a strategy is given; Step 7: Extract the multispectral reflectance of the spatial data set and elevation values , combined with the historical elevation series, the fluctuation disturbance anomaly index BRY is calculated and compared with the third threshold Q3 to determine whether the landslide structure in the monitored sub-area is stable. If it is unstable, a strategy is given.

Images