Opencast coal mine slope monitoring and early warning method and system based on multi-modal satellite fusion AI

Abstract

The invention discloses an open pit coal mine slope monitoring and early warning method and system based on multi-modal satellite fusion AI, and belongs to the technical field of geological disaster monitoring. The method comprises the following steps: S1, collecting multi-source data; s2, transmitting data in real time; s3, data preprocessing and space-time registration; s4, multi-band InSAR phase reconstruction and deformation field extraction are carried out; s5, deeply fusing the multi-source data; and S6, AI intelligent prediction and grading early warning. According to the open pit coal mine slope monitoring and early warning method and system based on multi-modal satellite fusion AI, slope global millimeter-level precision continuous deformation field monitoring is achieved, vegetation terrain shielding interference is reduced, the monitoring range is expanded, meanwhile, monitoring blind areas are eliminated, data processing efficiency is improved, a high-reliability deformation field is constructed, and the early warning error and missing report rate is reduced.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster monitoring technology, specifically involving a method and system for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI. Background Technology

[0002] Currently, there are over one million artificial high and steep slopes nationwide, including those at open-pit mines, dam shoulders of hydropower projects, and cut slopes along highways and railways. Affected by factors such as engineering excavation, load disturbance, and changes in the natural environment, these slopes generally exhibit characteristics of "high height, steep slope, deep excavation, and strong external disturbance." The risk of slope instability leading to geological disasters such as landslides and collapses continues to rise, threatening not only human lives but also causing significant property losses and ecological damage.

[0003] However, existing technologies have the following shortcomings: traditional total stations and GPS can only acquire discrete point data, which cannot cover the continuous deformation field of slopes and are prone to missing potential slip zones; ground-based radar (GB-InSAR) has a limited monitoring range, is greatly affected by terrain obstruction and vegetation, and has poor applicability; UAV photogrammetry relies on visible light, is limited by strong winds (>8m / s), low clouds, rain and snow, and requires airspace approval, making it difficult to conduct high-frequency continuous monitoring; data needs to be processed offline, resulting in poor real-time performance; conventional spaceborne InSAR is prone to "decoherence" in the rapid and large-gradient deformation areas of open-pit mines, making it impossible to solve the data and creating monitoring blind spots; it relies on manual interpretation, and it takes 3-7 days from data acquisition to report output, which does not meet the requirement of "minute-level early warning"; most solutions are based on a single data source, lack multi-dimensional analysis, have high false alarm and false alarm rates, and are difficult to support emergency decision-making.

[0004] Therefore, a new method is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI. This method achieves continuous deformation field monitoring of the entire slope with millimeter-level accuracy, reduces interference from vegetation and terrain shading, expands the monitoring range, eliminates monitoring blind spots, improves data processing efficiency, constructs a highly reliable deformation field, and reduces the false alarm and missed alarm rates.

[0006] To achieve the above objectives, this invention provides a method and system for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI, comprising the following steps: S1. Collect multiple types of raw data from air-based platforms, space-based platforms, and ground-based and slope-based platforms to obtain multi-source heterogeneous raw data; S2. Through cellular networks, satellite communication links, and heterogeneous converged networks, the multi-source heterogeneous raw data in S1 is transmitted to the cloud intelligent processing center in real time. S3 receives the multi-source heterogeneous raw data transmitted by S2. For multi-band SAR satellite data in the multi-source heterogeneous raw data, it performs format conversion and noise removal. For UAV, GNSS and environmental data, it performs standardization processing. It accurately aligns the processed data in the spatial coordinate system and time dimension. It obtains pre-processed and spatiotemporally registered multi-source data. The noise removal process uses the Lee filtering algorithm, and the formula is: ; in, These are the filtered pixel values. For filtering window, This represents the average grayscale value of the pixels within the filtering window. The variance of pixel grayscale values ​​within the wave window. Normalization factor; To filter the window All pixels within Perform a summation operation; S4. Receive the preprocessed and spatiotemporally registered multi-source data output from S3, and use an improved temporal InSAR algorithm to extract the millimeter-level continuous deformation field of the slope; obtain multi-band InSAR phase reconstruction results and deformation field data. The improved temporal InSAR algorithm includes distributed scatterer extraction, three-dimensional phase unwrapping, and phase adaptive filtering; Distributed scatterer extraction uses amplitude deviation index identification, the formula of which is: ; in, The standard deviation of pixel amplitude. To determine the average pixel amplitude, pixels with smaller ADI values ​​are selected as DS points. S5 receives the multi-band InSAR phase reconstruction results and deformation field data output by S4, assimilates the multi-source deformation data through Kalman filtering, and then combines GCN analysis with the coupling relationship with environmental data to generate the full-scene deformation field; obtain the full-scene deformation field data after multi-source deep fusion. The calculation process of Kalman filtering includes: ; ; ; ; ; in, for The state prediction vector at time t; for The state prediction vector at time t; For multi-source observation vectors; This is the state transition matrix; for The prior covariance matrix at time t; for The posterior covariance matrix at time t; for The transpose of the matrix; The process noise covariance matrix; The Kalman gain matrix; The observation matrix; for The transpose of the matrix; To observe the noise covariance matrix; for The posterior state estimate vector at time t; for The observation vector at time; These are the predicted values ​​based on the observed states; To observe residuals; for The posterior covariance matrix at time t; It is the identity matrix; S6 receives the multi-source deep fusion full-scene deformation field data output by S5, uses the LSTM-TCN hybrid model to predict the deformation trend, assesses the risk based on joint probability, triggers hierarchical early warning and forms closed-loop control. The LSTM-TCN hybrid model employs a combination of long short-term memory networks and temporal convolutional networks to learn the deformation time-series curves of each key region. The formula is as follows: ; ; ; ; ; ; in, This is the input vector for the current time step; This is the hidden state vector from the previous time step; , , , , , , This is the weight matrix; , , , It is the bias vector; It is the sigmoid activation function; It is the hyperbolic tangent activation function; This is an element-wise multiplication operation; The causal convolution formula for TCN is: ; in, At time step Output at; The first convolution kernel One weight parameter; Time step Input signal at the location; The coefficient of thermal expansion; Index within the convolution kernel.

[0007] Preferably, S3 also includes statistical filtering to remove outliers, calculated using the following formula: ; in, For the set of outliers; For point To its neighboring area Distance from the mean; This is the threshold coefficient, with a value of 3. The standard deviation of the point cloud within the neighborhood; The coordinates of a point are calculated using precise single-point positioning technology, and the formula is as follows: ; in, Let be the coordinate vector to be estimated. To design the matrix, The observation weight matrix is... A vector of observations; is the inverse of the coefficient matrix of the normal equation; Let be the vector of constant terms in the normal equation.

[0008] Preferably, the spatial coordinate system and time dimension in S3 are precisely aligned, specifically as follows: Spatial registration, the formula is: ; in, Original coordinates For the registered coordinates, These are the affine transformation parameters; The time registration formula is: ; in, Time to be estimated The value at that location; For the target time point where the value needs to be estimated; For known time The value at that location; For known time The value at that location; For the target time With known time The interval.

[0009] Preferably, the three-dimensional phase unwrapping and phase adaptive filtering in S4 are as follows: The three-dimensional phase unwrapping algorithm has the following formula: ; in, The true phase after untangling; For the entangled phase; The optimal untangling coefficients are obtained using the minimum cost flow algorithm. ; Phase adaptive filtering, the formula is: ; in, The filtered phase value; To Summation operation is performed on the phase values ​​of each neighborhood; For adaptive weights; This represents the phase value of the neighborhood.

[0010] Preferably, the formula for the joint probabilistic risk model in S6 is: ; in, The joint probability of slope instability risk; It is the deformation acceleration; The deformation acceleration threshold; For environmental parameters; For environmental parameter thresholds, This represents the probability of the creep curve reaching an inflection point.

[0011] This invention also provides a multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning system, including: The data acquisition module is used to collect various types of raw data from air-based platforms, space-based platforms, and ground-based and slope-based platforms to obtain multi-source heterogeneous raw data. A network transmission module, connected to the data acquisition module, is used to transmit multi-source heterogeneous raw data to the cloud intelligent processing center in real time via cellular networks, satellite communication links, and heterogeneous fusion networks. The data preprocessing module, connected to the network transmission module, is used to receive multi-source heterogeneous raw data, perform format conversion and noise removal on multi-band SAR satellite data in the multi-source heterogeneous raw data; perform standardization processing on UAV, GNSS and environmental data; accurately align the processed data in spatial coordinate system and time dimension; and obtain preprocessed and spatiotemporally registered multi-source data. The deformation field extraction module, connected to the data preprocessing module, is used to receive preprocessed and spatiotemporally registered multi-source data, and to extract the millimeter-level continuous deformation field of the slope using an improved temporal InSAR algorithm; thereby obtaining multi-band InSAR phase reconstruction results and deformation field data. The data fusion module, connected to the deformation field extraction module, is used to receive multi-band InSAR phase reconstruction results and deformation field data, assimilate multi-source deformation data through Kalman filtering, and then combine GCN analysis with the coupling relationship with environmental data to generate full-scene deformation field; thus obtaining full-scene deformation field data after multi-source deep fusion. The graded early warning module, connected to the data fusion module, is used to receive full-scene deformation field data after deep fusion of multiple sources, predict deformation trends using the LSTM-TCN hybrid model, assess risks based on joint probability, trigger graded early warnings, and form closed-loop control.

[0012] Therefore, the present invention employs the above-mentioned multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) This invention adopts a multi-source collaborative architecture of "air-space-ground", integrating multi-band SAR satellites, UAVs and ground-based sensors; it introduces L-band SAR data to overcome the problems of limited coverage (such as single-point monitoring) and insufficient accuracy of traditional monitoring technologies, and the problem that they are easily affected by vegetation and terrain to form monitoring blind spots, so as to realize continuous deformation field monitoring of the entire slope with accuracy down to the millimeter level; it significantly reduces the interference of vegetation and terrain and greatly expands the scope of monitoring application; (2) This invention utilizes multi-band SAR satellites (with all-weather observation capabilities); deploys UAVs carrying SAR payloads; and constructs a cloud-based real-time data processing system to overcome the problems of conventional monitoring being greatly affected by weather (such as rain and fog), and the need for airspace approval for UAV optical photogrammetry, making it difficult to achieve high-frequency continuous monitoring and data processing relying on offline interpretation, resulting in poor real-time performance. This invention achieves uninterrupted all-weather monitoring without being limited by weather conditions; UAVs can conduct intensive observations during satellite revisit intervals, avoiding the airspace approval restrictions on high-frequency monitoring; and data processing efficiency is improved, achieving "minute-level" output. (3) This invention adopts an improved time-series InSAR algorithm for distributed scatterers; combines L-band phase adaptive filtering with three-dimensional phase unwrapping technology; introduces an AI automated analysis module to overcome the "decoherence" phenomenon that is prone to occur in the large gradient deformation region of conventional InSAR technology, which leads to data failure and the formation of monitoring blind spots; traditional data processing relies on manual interpretation, which is inefficient and prone to errors, effectively suppresses the "decoherence" problem in the large gradient deformation region, and completely eliminates the monitoring blind spots; AI replaces manual data interpretation, greatly improving processing efficiency and avoiding human error; (4) This invention constructs a Kalman filter-GCN deep fusion model (integrating multi-dimensional data such as deformation, meteorology, and microseismic data); adopts an LSTM-TCN hybrid model to establish a joint probability early warning mechanism of "deformation acceleration + multi-physics field anomaly", which overcomes the problems of existing schemes that are mostly based on single data source analysis, have low data fusion and insufficient credibility; and rely on a single threshold for early warning, with high false alarm and false alarm rates, and can only achieve "post-event monitoring", thus realizing the transformation from "post-event monitoring" to "pre-event early warning".

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating an embodiment of the multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system of the present invention; Figure 2 This is an architecture diagram of an embodiment of the multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system of the present invention; Figure 3 This is a data processing flowchart of an embodiment of the multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0016] Example 1 like Figures 1-3 As shown, the multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system of the present invention includes the following steps: S1. Collect multiple types of raw data from air-based platforms, space-based platforms, and ground-based and slope-based platforms to obtain multi-source heterogeneous raw data; Among them, the air-based platform schedules multi-band SAR satellites (X / C / L bands) to collect remote sensing data according to their orbital cycles, and simultaneously triggers collaborative observations of hyperspectral and optical satellites to obtain large-scale multimodal remote sensing data covering the slope area of ​​open-pit coal mines; the space-based platform, according to cloud mission instructions, uses UAV SAR payloads to execute encrypted flight routes and collect high-resolution SAR images; the UAV LiDAR scans along a preset profile to obtain point cloud data for generating high-precision digital elevation models; the ground-based and slope-based miniaturized GB-InSAR equipment continuously transmits and receives microwave signals, the Beidou / GNSS monitoring station records satellite positioning data at a sampling rate of 1Hz, and the micro-weather station, soil moisture sensor, and ground acoustic / microseismic sensor respectively collect meteorological (wind speed, humidity, rainfall), soil moisture content, rock acoustic emission, and microseismic event data; S2 transmits the multi-source heterogeneous raw data from S1 to the cloud intelligent processing center in real time via 5G / 4G cellular networks, satellite communication links, and LoRa / Wi-Fi-Sonde heterogeneous converged networks. In this step, the 5G / 4G cellular network transmits low-to-medium volume data (such as GNSS observations, environmental sensor data, and drone LiDAR point clouds) from ground-based sensors and drones in real time, and uses the TCP / IP protocol to ensure data reliability. The satellite communication link transmits massive image data from space-based SAR satellites via Ka-band satellite links, and uses LDPC coding technology to reduce the channel bit error rate. The LoRa / Wi-Fi-Sonde heterogeneous converged network transmits data from miniature sensors (such as soil moisture sensors) in remote areas over low power and long distances via LoRa, and then aggregates the data through Wi-Fi-Sonde nodes before connecting it to the backbone network. S3 receives the multi-source heterogeneous raw data transmitted by S2. For multi-band SAR satellite data in the multi-source heterogeneous raw data, it performs basic processing such as format conversion and noise removal; for UAV, GNSS and environmental data, it performs standardization processing; it performs spatial registration and temporal registration on the processed data; and obtains pre-processed and spatiotemporally registered multi-source data. In this step, path A uses multi-band SAR satellite data. First, the raw SAR data (such as SLC format) is converted to the common HDF5 format. Then, the Lee filtering algorithm is used to suppress speckle noise in the SAR image. The formula is as follows: ; in, These are the filtered pixel values. For filtering window, This represents the average grayscale value of the pixels within the filtering window. The variance of pixel grayscale values ​​within the wave window. Normalization factor; To filter the window All pixels within Perform a summation operation; Path B uses data from UAVs, GNSS, and the environment. Outliers are first removed using statistical filtering, calculated using the following formula: ; in, For the set of outliers; For point To its neighboring area Distance from the mean; This is the threshold coefficient, with a value of 3. The standard deviation of the point cloud within the neighborhood; Then, precise point positioning (PPP) technology is used to calculate the point coordinates. The formula is as follows: ; in, Let be the coordinate vector to be estimated. To design the matrix, The observation weight matrix is... A vector of observations; is the inverse of the coefficient matrix of the normal equation; Let be the vector of constant terms in the normal equation; Then, the meteorological, soil, and ground acoustic data were normalized. Spatial registration uses the coordinates of BeiDou / GNSS monitoring stations as a reference, and employs an affine transformation model to unify the spatial coordinates of multi-source data. The formula is as follows: ; in, Original coordinates For the registered coordinates, These are the affine transformation parameters; Time registration uses linear interpolation to align data at different time sampling rates. The formula is as follows: ; in, Time to be estimated The value at that location; For the target time point where the value needs to be estimated; For known time The value at that location; For known time The value at that location; For the target time With known time The interval, ; S4. Receive the preprocessed and spatiotemporally registered multi-source data output from S3, and use an improved temporal InSAR algorithm (including distributed scatterer improvement, phase adaptive filtering, and three-dimensional phase unwrapping) to extract the millimeter-level continuous deformation field of the slope; obtain multi-band InSAR phase reconstruction results and deformation field data. In this step, the distributed scatterer (DS) extraction uses the amplitude deviation index (ADI) to identify DS points, with the following formula: ; in, The standard deviation of pixel amplitude. To determine the average pixel amplitude, pixels with smaller ADI values ​​are selected as DS points. Temporal InSAR phase unwrapping employs an improved three-dimensional phase unwrapping algorithm to unwrap the phase differences of multi-band SAR data. The formula is as follows: ; in, The true phase after untangling; For the entangled phase; The optimal untangling coefficients are obtained using the minimum cost flow algorithm. ; L-band enhancement processing introduces a phase adaptive filtering algorithm to effectively suppress phase noise caused by large gradient deformation and volume scattering effects. The formula is as follows: ; in, The filtered phase value; To Summation operation is performed on the phase values ​​of each neighborhood; The adaptive weights are calculated from local phase consistency. Using the neighborhood phase value, usable deformation timing information is finally extracted from the fast deformation region where conventional methods fail. Based on the relationship between InSAR phase and slant range variation, the deformation of the deformation field is calculated using the following formula: ; in, For slant range direction deformation, For radar wavelength, The phase change is then combined with satellite orbital parameters to convert the slant range deformation into vertical deformation. S5 receives the multi-band InSAR phase reconstruction results and deformation field data output by S4, assimilates the multi-source deformation data through Kalman filtering, and then combines GCN analysis with the coupling relationship with environmental data to generate the full-scene deformation field; obtain the full-scene deformation field data after multi-source deep fusion. In this step, a state-space model is established, and the multi-source deformation data is assimilated using Kalman filtering to generate a spatiotemporally continuous "slope full-scene deformation field," the formula of which is: ; ; ; ; ; in, for The state prediction vector at time t; for The state prediction vector at time t; For multi-source observation vectors; This is the state transition matrix; for The prior covariance matrix at time t; for The posterior covariance matrix at time t; for The transpose of the matrix; The process noise covariance matrix; The Kalman gain matrix; The observation matrix; for The transpose of the matrix; To observe the noise covariance matrix; for The posterior state estimate vector at time t; for The observation vector at time; These are the predicted values ​​based on the observed states; To observe residuals; for The posterior covariance matrix at time t; It is the identity matrix; The assimilated full-scene deformation field, along with environmental data such as meteorological, hydrological, and microseismic data, is input into the GCN. The entire slope is treated as a graph structure, with each monitoring point as a node and the spatial relationships and geological connections between nodes as edges. The spatial propagation law of deformation and its coupling relationship with environmental driving forces are learned, as shown in the formula: ; in, For the first Layer nodes The feature vectors are used to fuse deformation and environmental information; For nodes In the Hidden feature vectors of the layer; For nodes neighborhood The normalization coefficient is... This is the weight matrix. For bias vectors, For activation functions; S6 receives the multi-source deep fusion full-scene deformation field data output by S5, uses the LSTM-TCN hybrid model to predict the deformation trend, assesses the risk based on joint probability, triggers a four-level graded early warning and forms a closed-loop control. In this step, the LSTM-TCN hybrid model prediction uses a hybrid model of Long Short-Term Memory (LSTM) network and Temporal Convolutional Network (TCN) to learn the deformation time-series curve of each key region. The formula is as follows: ; ; ; ; ; ; in, This is the input vector for the current time step; This is the hidden state vector from the previous time step; , , , , , , This is the weight matrix; , , , It is the bias vector; It is the sigmoid activation function; It is the hyperbolic tangent activation function; This is an element-wise multiplication operation; The causal convolution formula for TCN is: ; in, At time step Output at; The first convolution kernel One weight parameter; Time step Input signal at the location; The coefficient of thermal expansion; Indices within the convolution kernel; A risk probability model is constructed based on the joint probability of deformation acceleration, creep curve inflection point, and anomalies in multiple physics parameters (such as rainfall and microseismic frequency): ; in, The joint probability of slope instability risk; For deformation acceleration, ; The deformation acceleration threshold; For environmental parameters; For environmental parameter thresholds, This represents the probability of an inflection point appearing in the creep curve. Warning information is divided into four levels: blue, yellow, orange, and red; if If a blue alert is triggered; If a yellow alert is triggered; If an orange alert is triggered; This triggered a red alert. The application layer monitors dashboards and deformation thematic maps in real time via the web interface, sends early warning push notifications and emergency response suggestions via the mobile app, and displays risk areas in 3D visualization on the digital twin platform, providing decision support to managers and forming a closed loop from perception to decision-making; at the same time, it generates control commands based on the early warning level and transmits them back to S1 to achieve closed-loop management.

[0017] Therefore, the present invention adopts the above-mentioned multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method and system. This method realizes continuous deformation field monitoring with millimeter-level accuracy across the entire slope area, reduces interference from vegetation and terrain shading, expands the monitoring range, eliminates monitoring blind spots, improves data processing efficiency, constructs a highly reliable deformation field, and reduces the false alarm and missed alarm rate.

[0018] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0019] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multimodal satellite-integrated AI-based method for monitoring and early warning of slopes in open-pit coal mines, characterized in that, Includes the following steps: S1. Collect multiple types of raw data from air-based platforms, space-based platforms, and ground-based and slope-based platforms to obtain multi-source heterogeneous raw data; S2. Through cellular networks, satellite communication links, and heterogeneous converged networks, the multi-source heterogeneous raw data in S1 is transmitted to the cloud intelligent processing center in real time. S3: Receive the multi-source heterogeneous raw data transmitted by S2, and perform format conversion and noise removal processing on the multi-band SAR satellite data in the multi-source heterogeneous raw data. Standardize the processing of data from drones, GNSS, and the environment; The processed data is precisely aligned in both the spatial coordinate system and the time dimension. Obtain preprocessed and spatiotemporally registered multi-source data; The noise removal process uses the Lee filtering algorithm, and the formula is: ； in, These are the filtered pixel values. For filtering window, This represents the average grayscale value of the pixels within the filtering window. The variance of pixel grayscale values ​​within the wave window. Normalization factor; To filter the window All pixels within Perform a summation operation; S4. Receive the preprocessed and spatiotemporally registered multi-source data output from S3, and use an improved temporal InSAR algorithm to extract the millimeter-level continuous deformation field of the slope; obtain multi-band InSAR phase reconstruction results and deformation field data. The improved temporal InSAR algorithm includes distributed scatterer extraction, three-dimensional phase unwrapping, and phase adaptive filtering; Distributed scatterer extraction uses amplitude deviation index identification, the formula of which is: ； in, The standard deviation of pixel amplitude. To determine the average pixel amplitude, pixels with smaller ADI values ​​are selected as DS points. S5 receives the multi-band InSAR phase reconstruction results and deformation field data output by S4, assimilates the multi-source deformation data through Kalman filtering, and then combines GCN analysis with the coupling relationship with environmental data to generate the full-scene deformation field; obtain the full-scene deformation field data after multi-source deep fusion. The calculation process of Kalman filtering includes: ； ； ； ； ； in, for The state prediction vector at time t; for The state prediction vector at time t; For multi-source observation vectors; This is the state transition matrix; for The prior covariance matrix at time t; for The posterior covariance matrix at time t; for The transpose of the matrix; The process noise covariance matrix; The Kalman gain matrix; The observation matrix; for The transpose of the matrix; To observe the noise covariance matrix; for The posterior state estimate vector at time t; for The observation vector at time; These are the predicted values ​​based on the observed states; To observe residuals; for The posterior covariance matrix at time t; It is the identity matrix; S6 receives the multi-source deep fusion full-scene deformation field data output by S5, uses the LSTM-TCN hybrid model to predict the deformation trend, assesses the risk based on joint probability, triggers hierarchical early warning and forms closed-loop control. The LSTM-TCN hybrid model employs a combination of long short-term memory networks and temporal convolutional networks to learn the deformation time-series curves of each key region. The formula is as follows: ； ； ； ； ； ； in, This is the input vector for the current time step; This is the hidden state vector from the previous time step; , , , , , , This is the weight matrix; , , , It is the bias vector; It is the sigmoid activation function; It is the hyperbolic tangent activation function; This is an element-wise multiplication operation; The causal convolution formula for TCN is: ； in, At time step Output at; The first convolution kernel One weight parameter; Time step Input signal at the location; The coefficient of thermal expansion; Index within the convolution kernel.

2. The method for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI according to claim 1, characterized in that, S3 also includes statistical filtering to remove outliers, and the calculation formula is as follows: ； in, For the set of outliers; For point To its neighboring area Distance from the mean; This is the threshold coefficient, with a value of 3. The standard deviation of the point cloud within the neighborhood; The coordinates of a point are calculated using precise single-point positioning technology, and the formula is as follows: ； in, Let the coordinate vector be to be estimated. To design the matrix, The observation weight matrix, A vector of observations; is the inverse of the coefficient matrix of the normal equation; Let be the vector of constant terms in the normal equation.

3. The method for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI according to claim 1, characterized in that, S3 achieves precise alignment in both the spatial coordinate system and the time dimension, specifically: Spatial registration, the formula is: ； in, Original coordinates For the registered coordinates, These are the affine transformation parameters; The time registration formula is: ； in, Time to be estimated The value at that location; For the target time point where the value needs to be estimated; For known time The value at that location; For known time The value at that location; For the target time With known time The interval.

4. The method for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI according to claim 1, characterized in that, The three-dimensional phase unwrapping and phase adaptive filtering in S4 are as follows: The three-dimensional phase unwrapping algorithm has the following formula: ； in, The true phase after untangling; For the entangled phase; The optimal untangling coefficients are obtained using the minimum cost flow algorithm. ; Phase adaptive filtering, the formula is: ； in, The filtered phase value; To Summation operation is performed on the phase values ​​of each neighborhood; For adaptive weights; This represents the phase value of the neighborhood.

5. The method for monitoring and early warning of open-pit coal mine slopes using multimodal satellite fusion AI according to claim 1, characterized in that, The formula for the joint probabilistic risk model in S6 is: ; in, The joint probability of slope instability risk; It is the deformation acceleration; The deformation acceleration threshold; For environmental parameters; For environmental parameter thresholds, This represents the probability of the creep curve reaching an inflection point.

6. A multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning system, applied to the multimodal satellite fusion AI-based open-pit coal mine slope monitoring and early warning method according to any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect various types of raw data from air-based platforms, space-based platforms, and ground-based and slope-based platforms to obtain multi-source heterogeneous raw data. A network transmission module, connected to the data acquisition module, is used to transmit multi-source heterogeneous raw data to the cloud intelligent processing center in real time via cellular networks, satellite communication links, and heterogeneous fusion networks. The data preprocessing module, connected to the network transmission module, is used to receive multi-source heterogeneous raw data and perform format conversion and noise removal processing on multi-band SAR satellite data in the multi-source heterogeneous raw data. Standardize the processing of data from drones, GNSS, and the environment; The processed data is precisely aligned in both the spatial coordinate system and the time dimension. Obtain preprocessed and spatiotemporally registered multi-source data; The deformation field extraction module, connected to the data preprocessing module, is used to receive preprocessed and spatiotemporally registered multi-source data, and to extract the millimeter-level continuous deformation field of the slope using an improved temporal InSAR algorithm; thereby obtaining multi-band InSAR phase reconstruction results and deformation field data. The data fusion module, connected to the deformation field extraction module, is used to receive multi-band InSAR phase reconstruction results and deformation field data, assimilate multi-source deformation data through Kalman filtering, and then combine GCN analysis with the coupling relationship with environmental data to generate a full-scene deformation field. Obtain full-scene deformation field data after deep fusion of multiple sources; The graded early warning module, connected to the data fusion module, is used to receive full-scene deformation field data after deep fusion of multiple sources, predict deformation trends using the LSTM-TCN hybrid model, assess risks based on joint probability, trigger graded early warnings, and form closed-loop control.

7. A computer device, characterized in that, include: A processor configured to be coupled to a memory, read and execute instructions and / or program code in the memory to perform the method as described in any one of claims 1-5.

8. A computer-readable medium, characterized in that, The computer-readable medium stores computer program code that, when executed on a computer, causes the computer to perform the method as described in any one of claims 1-5.

Images