Highway slope deformation intelligent identification and prediction system based on space-air-ground three-dimensional monitoring

Abstract

The application relates to the technical field of artificial intelligence, and discloses a highway side slope deformation intelligent identification and prediction system based on space-air-ground stereoscopic monitoring, which comprises a space-air-ground stereoscopic monitoring module, a data fusion processing module, an intelligent identification module, a deformation prediction module and a central control module; the space-air-ground stereoscopic monitoring module cooperatively collects side slope deformation data by using multi-source heterogeneous sensors, including space satellite remote sensing data, sky unmanned aerial vehicle aerial photography data and ground Internet of Things sensor data, realizes all-weather and multi-scale monitoring of side slope surface displacement, internal stress, crack expansion and environmental factors, guarantees the comprehensiveness and real-time performance of the monitoring data, and eliminates the limitations of a single data source; the data fusion processing module performs space-time registration, denoising and normalization processing on the multi-source monitoring data, guarantees the accuracy and reliability of real-time deformation risk diagnosis, and reduces the false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an intelligent identification and prediction system for highway slope deformation based on three-dimensional air-space-ground monitoring. Background Technology

[0002] Highway slope deformation refers to the displacement, deformation, or damage of highway cutting or embankment slopes caused by natural factors and engineering activities. It is one of the major geological hazards affecting the safe operation of highways. Its manifestations are diverse and its causes are complex, usually involving multiple factors such as soil and rock structure, hydrological conditions, and external loads.

[0003] Currently, highway slope deformation monitoring mainly relies on a single data source or traditional monitoring methods, making it difficult to achieve all-weather, multi-scale collaborative perception of slope surface displacement, internal stress, crack propagation, and environmental factors. This results in limitations and biases in the monitoring data. Furthermore, due to the lack of a mechanism for effectively fusing and processing multi-source heterogeneous monitoring data, it is difficult to generate a consistent deformation dataset, which affects the accuracy of real-time intelligent identification of deformation anomaly areas and disaster precursor characteristics.

[0004] Therefore, a smart identification and prediction system for highway slope deformation based on three-dimensional air-space-ground monitoring is proposed to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent identification and prediction system for highway slope deformation based on three-dimensional air-space-ground monitoring, which solves the problem mentioned in the background technology that the monitoring data has limitations and biases, affecting the accuracy of intelligent identification.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a highway slope deformation intelligent identification and prediction system based on air-space-ground three-dimensional monitoring, the system comprising an air-space-ground three-dimensional monitoring module, a data fusion processing module, an intelligent identification module, a deformation prediction module, and a central control module; The air-ground-space three-dimensional monitoring module is used to collect deformation data of highway slopes through multi-source heterogeneous sensors, including space satellite remote sensing data, sky drone aerial photography data and ground Internet of Things sensor data, to realize all-weather, multi-scale monitoring of slope surface displacement, internal stress, crack propagation and environmental factors. The data fusion processing module is used to perform spatiotemporal registration, denoising, and normalization on multi-source monitoring data, and to generate a consistent slope deformation dataset through data fusion algorithms, thereby eliminating the limitations of a single data source. The intelligent identification module is used to automatically identify abnormal slope deformation areas and disaster precursor features based on a deep learning model, including landslide segmentation, crack detection, and stability assessment, to achieve real-time diagnosis of deformation risk. The deformation prediction module is used to predict the future deformation trend and probability of disaster occurrence of the slope based on historical deformation data and real-time monitoring information, through time series analysis and machine learning algorithms, and outputs short-term, medium-term and long-term prediction results. The central control module is used to coordinate the operation of various modules in the system, schedule data flow and computing resources, control the operation of the early warning feedback unit according to the preset early warning threshold, and provide a human-machine interface to support monitoring parameter setting, real-time data visualization and prediction report export.

[0007] Preferably, the air-space-ground integrated monitoring module includes a satellite remote sensing unit, an unmanned aerial vehicle (UAV) aerial photography unit, and a ground sensing unit; The satellite remote sensing unit acquires large-scale, periodic deformation information of the slope through synthetic aperture radar satellites and optical satellites, supporting millimeter-level displacement monitoring; The drone aerial photography unit, equipped with a multispectral camera and a lidar sensor, enables high-resolution 3D modeling and local deformation capture of the slope. The ground sensing unit consists of tilt sensors, strain gauges, crack gauges, and temperature and humidity sensors deployed at key locations on the slope, which collect internal deformation and environmental parameters in real time.

[0008] Preferably, the satellite remote sensing unit is also equipped with an interferometric radar data processing component, used to extract phase information of small deformations of the slope through differential interferometry. The drone aerial photography unit has autonomous path planning and obstacle avoidance mechanisms, making it suitable for dynamic monitoring tasks in complex terrain. The ground sensing unit adopts a wireless sensor network architecture, which supports low-power wide-area communication, real-time data transmission, and node self-organizing network.

[0009] Preferably, the data fusion processing module includes a data preprocessing unit, a fusion algorithm unit, and a quality assessment unit; The data preprocessing unit is used to perform outlier removal, missing value interpolation, and coordinate unification on the raw monitoring data. The fusion algorithm unit employs the Kalman filtering method to fuse uncertainties from multiple data sources. The quality assessment unit verifies the accuracy of the fused data through consistency indicators and error analysis, and provides data credibility labels for subsequent modules.

[0010] Preferably, the data preprocessing unit integrates adaptive filtering technology, which dynamically adjusts the filtering parameters according to the data characteristics; The fusion algorithm unit supports a multi-scale fusion strategy, which processes global satellite data, regional UAV data and local ground data respectively, and retains feature information at different levels. The quality assessment unit introduces the entropy weight method to quantify the contribution of data sources and optimize the allocation of fusion weights.

[0011] Preferably, the intelligent recognition module includes a feature extraction unit, a model training unit, and an anomaly detection unit; The feature extraction unit is used to extract deformation-related features from the fused data, namely displacement rate, curvature change and spectral characteristics; The model training unit constructs a recognition model based on a convolutional neural network and trains model parameters using historical disaster data. The anomaly detection unit identifies signs of slope instability, namely accelerated displacement and crack penetration, through real-time reasoning, and outputs a risk level score.

[0012] Preferably, the feature extraction unit combines principal component analysis dimensionality reduction technology to reduce data redundancy; The model training unit adopts the transfer learning method, which uses the pre-trained model to accelerate convergence and improve the recognition accuracy under small sample conditions. The anomaly detection unit integrates an attention mechanism to focus on high-risk areas and reduce the false alarm rate.

[0013] Preferably, the deformation prediction module includes a trend analysis unit, a probability prediction unit, and a risk assessment unit; The trend analysis unit predicts the deformation over time using support vector machine regression. The probability prediction unit calculates the probability of disaster occurrence based on Monte Carlo simulation and outputs a confidence interval. The risk assessment unit combines geological environmental factors and prediction results to generate a comprehensive slope stability rating.

[0014] Preferably, the trend analysis unit has multi-step prediction capability, supporting rolling prediction and dynamic correction; The probability prediction unit introduces extreme value theory to handle rare events, providing deformation samples for identifying and processing rare events; The risk assessment unit integrates fuzzy logic methods to handle uncertainty and subjective judgment, and outputs a visualized risk map.

[0015] Preferably, the central control module specifically includes a task scheduling unit, a resource coordination unit, and a human-computer interaction unit; The task scheduling unit is used to dynamically coordinate the working mode and acquisition frequency of each sensor unit in the air-space-ground integrated monitoring module according to the preset monitoring task priority and the status of the data stream, and automatically trigger the analysis and calculation process of the intelligent identification module and the deformation prediction module. The resource coordination unit is used to dynamically allocate the computing resources and manage the storage of the slope deformation dataset. The human-computer interaction unit is used to provide the human-computer interaction interface to set monitoring parameters and the early warning threshold, and to visualize real-time monitoring data, deformation recognition results and risk prediction reports.

[0016] Compared with existing technologies, this invention provides an intelligent identification and prediction system for highway slope deformation based on three-dimensional air-space-ground monitoring, which has the following advantages: 1. In this invention, a three-dimensional air-space-ground monitoring module is used to collaboratively collect slope deformation data using multi-source heterogeneous sensors, including space satellite remote sensing data, aerial drone data, and ground IoT sensor data. This enables all-weather, multi-scale monitoring of slope surface displacement, internal stress, crack propagation, and environmental factors, ensuring the comprehensiveness and real-time nature of the monitoring data and eliminating the limitations of a single data source.

[0017] 2. In this invention, the data fusion processing module performs spatiotemporal registration, denoising, and normalization on multi-source monitoring data, and generates a consistent slope deformation dataset based on the data fusion algorithm. At the same time, the intelligent identification module uses a deep learning model to automatically identify abnormal slope deformation areas and disaster precursor features, ensuring the accuracy and reliability of real-time deformation risk diagnosis and reducing the false alarm rate.

[0018] 3. In this invention, the deformation prediction module predicts the future deformation trend and probability of disaster occurrence of the slope based on historical deformation data and real-time monitoring information, using time series analysis and machine learning algorithms. It outputs short-term, medium-term and long-term prediction results. The central control module coordinates the operation of each module of the system, schedules data flow and computing resources, and controls the early warning operation according to the preset early warning threshold, so as to ensure the timeliness and systematicness of early warning feedback and improve the efficiency of disaster prevention and control. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground, according to the present invention. Detailed Implementation

[0020] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] For specific implementation examples, please refer to: Figure 1The intelligent identification and prediction system for highway slope deformation based on air-space-ground three-dimensional monitoring includes an air-space-ground three-dimensional monitoring module, a data fusion and processing module, an intelligent identification module, a deformation prediction module, and a central control module. The air-ground-space integrated monitoring module is used to collect deformation data of highway slopes through multi-source heterogeneous sensors, including space satellite remote sensing data, sky drone aerial photography data and ground IoT sensor data, to achieve all-weather, multi-scale monitoring of slope surface displacement, internal stress, crack propagation and environmental factors. The data fusion processing module is used to perform spatiotemporal registration, denoising, and normalization on multi-source monitoring data. It generates a consistent slope deformation dataset through data fusion algorithms, eliminating the limitations of a single data source. The intelligent identification module is used to automatically identify abnormal slope deformation areas and disaster precursor characteristics based on deep learning models, including landslide segmentation, crack detection, and stability assessment, to achieve real-time diagnosis of deformation risks. The deformation prediction module is used to predict the future deformation trend and probability of disasters of slopes based on historical deformation data and real-time monitoring information, through time series analysis and machine learning algorithms, and outputs short-term, medium-term and long-term prediction results. The central control module coordinates the operation of various modules in the system, schedules data flow and computing resources, controls the operation of the early warning feedback unit according to the preset early warning threshold, and provides a human-machine interface to support the setting of monitoring parameters, real-time data visualization and export of prediction reports.

[0022] The air-space-ground integrated monitoring module includes a satellite remote sensing unit, an unmanned aerial vehicle (UAV) aerial photography unit, and a ground sensing unit. The satellite remote sensing unit acquires large-scale, periodic deformation information of slopes through synthetic aperture radar satellites and optical satellites, supporting millimeter-level displacement monitoring; The drone aerial photography unit, equipped with a multispectral camera and a lidar sensor, enables high-resolution 3D modeling and local deformation capture of slopes; The ground sensing unit consists of tilt sensors, strain gauges, crack gauges, and temperature and humidity sensors deployed at key locations on the slope to collect internal deformation and environmental parameters in real time.

[0023] The satellite remote sensing unit is also equipped with an interferometric radar data processing component, which is used to extract phase information of minute deformations of slopes through differential interferometry. When implementing differential interferometry: First, the two acquired SAR images are precisely registered and de-flattened to generate a differential interferogram; then, the minimum cost flow algorithm is used for phase unwrapping to obtain the absolute phase field; finally, the phase values ​​are converted into physical deformation values ​​using the deformation phase calculation formula. ; in This indicates the phase difference caused by slope deformation. This represents the total interferometric phase difference observed. This represents the phase introduced by satellite orbital errors. Phase representing the contribution of topographic relief. This represents the phase introduced by the atmospheric delay effect; by accurately estimating and subtracting the latter three error phases, the pure deformation phase can be separated. The drone aerial photography unit has autonomous path planning and obstacle avoidance mechanisms, making it suitable for dynamic monitoring tasks in complex terrain. Specific steps for implementing the obstacle avoidance mechanism: The obstacle avoidance mechanism is based on the potential field principle. When an obstacle is detected, the repulsive force is calculated to dynamically correct the trajectory. The formula for calculating the repulsive force is: ; in This represents the vector of the repulsive force exerted by the obstacle on the drone. It is the repulsive force gain coefficient. It is the real-time distance between the drone and the obstacle. It is the maximum influence distance of the obstacle. It is the unit direction vector from the drone pointing towards the obstacle; The ground sensing unit adopts a wireless sensor network architecture, supporting low-power wide-area communication, real-time data transmission, and node self-organizing network.

[0024] The data fusion processing module includes a data preprocessing unit, a fusion algorithm unit, and a quality assessment unit; The data preprocessing unit is used to perform outlier removal, missing value imputation, and coordinate standardization on the raw monitoring data; Outlier removal employs a box plot method based on statistical quantiles, identifying and removing data points that exceed the upper quartile plus 1.5 times the interquartile range or fall below the lower quartile minus 1.5 times the interquartile range as outliers. Missing value imputation uses the K-nearest neighbor algorithm for time series data. ; in Indicates the missing moment The estimated value to be interpolated, , These represent the moments immediately following the missing time. The previous valid time and the next valid time, , They represent in and Valid monitoring data collected in real time; Coordinate unification is achieved by transforming all spatial data into a unified coordinate system using the seven-parameter Bursa model: ; in This represents the three-dimensional coordinates in the unified coordinate system after transformation. This represents three translation parameters. It indicates that it consists of three rotation angles , , The rotation matrix formed, Represents the parameters of scale variation. Represents the three-dimensional coordinates of raw monitoring data from different sensors in their own local coordinate system; The fusion algorithm unit uses the Kalman filter method to fuse the uncertainties of multi-source data, thereby improving the reliability and integrity of deformation data; Specific implementation steps of Kalman filtering: First, the system is modeled as state equations and measurement equations: ; ; in express The system state vector at time t. It is the state transition matrix. It's process noise. express The observation vector at time t, It is the observation matrix. It is observation noise; The filtering process recursively executes prediction and update: the prediction step calculates the prior state estimate and covariance, the update step calculates the Kalman gain, and uses actual observations to correct and obtain the posterior state estimate, thus achieving optimal fusion of multi-source data in the sense of minimum mean square error; The quality assessment unit verifies the accuracy of the fused data through consistency indicators and error analysis, and provides data credibility labels for subsequent modules. Monitoring data from a stable region over a period of time is selected as a benchmark. Consistency indices between the fused data and the data from each independent source are calculated, namely root mean square error and correlation coefficient. At the same time, error propagation analysis is performed to quantify the impact of the uncertainty of each data source on the final fusion result. A quality assessment report is generated, and the data confidence level is marked to provide a reliable basis for subsequent intelligent identification and prediction.

[0025] The data preprocessing unit integrates adaptive filtering technology, which dynamically adjusts the filtering parameters according to the data characteristics; To dynamically adjust filter parameters and suppress non-stationary noise, a recursive least squares algorithm is used, the core of which is iteratively updating the filter weights: ; in express The filter weight coefficient vector at time 1. This is an adaptive step size factor. for The error between the expected output and the actual filter output at time step [time]. for The input data vector at time step; The fusion algorithm unit supports multi-scale fusion strategies, processing global satellite data, regional UAV data, and local ground data separately, while retaining feature information at different levels. The multi-scale fusion strategy fuses monitoring data at different resolutions at the feature level. First, wavelet transform is used to decompose global satellite data, regional UAV data, and local ground data into low-frequency approximate sub-bands and high-frequency detail sub-bands, respectively. Fusion rules are then formulated for each sub-band. The low-frequency subband adopts a weighted average rule to preserve the large-scale deformation trend; the high-frequency subband adopts a selection rule based on maximizing local gradients to preserve the edge and detailed features of each data source; finally, the fused subbands are subjected to wavelet inverse transform to reconstruct a fused deformation field that has both global consistency and local fine features. The quality assessment unit introduces the entropy weight method to quantify the contribution of data sources and optimize the allocation of fusion weights. Specifically, the fusion weight of each data source is calculated using the following formula: ; ; in Indicates the first The weight of each data source in the fusion process Indicates the first Information entropy of a data source Indicates the first Information entropy of a data source This indicates the total number of data sources participating in the integration. For the first The sample at the th Standardized values ​​under each indicator To evaluate the sample size.

[0026] The intelligent recognition module includes a feature extraction unit, a model training unit, and an anomaly detection unit; The feature extraction unit is used to extract deformation-related features from the fused data, namely displacement rate, curvature change and spectral characteristics; The model training unit builds a recognition model based on a convolutional neural network and trains the model parameters using historical disaster data; First, a convolutional neural network model is constructed, whose input layer receives multi-channel data, and contains multiple convolutional layers and pooling layers in the middle to automatically extract features from shallow to deep. The end is connected to a fully connected layer and a Softmax classification layer. Then, the model is trained using a labeled historical dataset. The network weight parameters are continuously adjusted through the backpropagation algorithm and the gradient descent optimizer Adam to minimize the cross-entropy loss function between the predicted label and the true label until the model converges. The anomaly detection unit identifies signs of slope instability, namely accelerated displacement and crack penetration, through real-time reasoning, and outputs a risk level score.

[0027] The feature extraction unit combines principal component analysis dimensionality reduction technology to reduce data redundancy; The original multidimensional feature data is used to construct a sample matrix; the covariance matrix of the matrix and its eigenvalues ​​and eigenvectors are calculated; the eigenvalues ​​are sorted from largest to smallest, and the top M principal components are selected so that the cumulative variance contribution rate exceeds a preset threshold; the original data is projected onto the low-dimensional space composed of these M principal components to achieve data dimensionality reduction, remove redundancy and noise, and retain the most important change information. The model training unit uses transfer learning to accelerate convergence and improve recognition accuracy with small samples by utilizing pre-trained models. Implementation of transfer learning: Model initialization: Load the pre-trained model, which is the weights of ResNet50 in the convolutional neural network model. These weights contain general features learned from massive images.

[0028] Model fine-tuning: Network structure adjustment: Remove the fully connected classification layer at the top of the original ResNet50 and replace it with a new structure adapted to the specific task of this invention. This can be achieved by adding a global average pooling layer, followed by one or more fully connected layers. Finally, the number of nodes in the output layer is set to match the number of deformation categories and the deformation value regression requirements. Training strategy: A phased training strategy is adopted. First, most of the bottom convolutional layers of the pre-trained model are frozen, and only the newly added top-level network part is trained. Then, some and all convolutional layers are unfrozen, and the entire network is jointly fine-tuned with a small learning rate, so that the model can adaptively learn deformation-related specific features from slope monitoring data. The anomaly detection unit integrates an attention mechanism to focus on high-risk areas and reduce false alarm rates; Attention mechanism: An attention module is embedded in the convolutional neural network. This module learns and automatically generates a weight map. This map “pays attention” to and amplifies the spatial location and channel features in the input feature map that are strongly correlated with disaster signs such as landslides and cracks, while suppressing the response of irrelevant background areas, thereby improving the accuracy and robustness of the model in identifying key signs.

[0029] The deformation prediction module includes a trend analysis unit, a probability prediction unit, and a risk assessment unit; The trend analysis unit uses support vector machine regression to predict the deformation over time. A support vector machine regression model is employed, using radial basis functions as the kernel to map historical time series to a high-dimensional space. A 2ε-width interval is then sought, where ε is the interval width parameter, to ensure that most samples fall within this interval while minimizing the total bias. For a new time point, the predicted deformation value is determined by the support vectors and their weights. ; in To predict the output value for the model, It is the number of support vectors. and It is a Lagrange multiplier. It is a radial basis kernel function. These are features of historical support vector samples. It is a feature of the moment to be predicted. It is a bias term; The probability prediction unit calculates the probability of disaster occurrence based on Monte Carlo simulation and outputs confidence intervals; First, identify the key random variables affecting slope stability and their probability distributions. Then, conduct N Monte Carlo random samplings, obtaining a set of variable values ​​in each sampling. Substitute these values ​​into the limit equilibrium method to calculate the safety factor. The probability of disaster occurrence, Pf, is estimated by the proportion of simulations with a safety factor less than 1. ; in This represents the probability of slope instability. For the number of simulations with a safety factor less than 1, Total number of Monte Carlo simulations; The risk assessment unit combines geological environmental factors and prediction results to generate a comprehensive slope stability rating.

[0030] The trend analysis unit has multi-step forecasting capabilities, supporting rolling forecasting and dynamic correction; The system uses a rolling time window for prediction: each time, a support vector machine regression model is trained based on the latest data of a fixed time length to predict the deformation in the next 7 days; when new actual monitoring data is obtained, it is included in the data window, while the oldest data is removed, and the model is retrained using the updated window for the next round of prediction. The probability prediction unit introduces extreme value theory to handle rare events, providing deformation samples for identifying and processing rare events; The risk assessment unit integrates fuzzy logic methods to handle uncertainty and subjective judgment, and outputs a visual risk map. First, define a fuzzy set and membership function for each input indicator; then, based on domain expert knowledge, establish a fuzzy rule base in the form of "IF-THEN"; finally, through a fuzzy inference engine, fuzzify the clear input values, apply the rule base for inference, and then defuzzify the output fuzzy conclusion to obtain a clear comprehensive early warning level.

[0031] The central control module specifically includes a task scheduling unit, a resource coordination unit, and a human-computer interaction unit; The task scheduling unit is used to dynamically coordinate the working mode and acquisition frequency of each sensor unit in the air-space-ground integrated monitoring module according to the preset monitoring task priority and data stream status, and automatically trigger the analysis and calculation process of the intelligent identification module and deformation prediction module. The resource coordination unit is used to dynamically allocate computing resources and manage the storage of slope deformation datasets; The human-computer interaction unit provides a human-computer interaction interface to set monitoring parameters and early warning thresholds, and to visualize real-time monitoring data, deformation recognition results and risk prediction reports.

[0032] The operating steps of this system are as follows: First, the air-space-ground integrated monitoring module collects deformation data of highway slopes through multi-source heterogeneous sensors, including space satellite remote sensing data, aerial drone data, and ground IoT sensor data, to achieve all-weather, multi-scale monitoring of slope surface displacement, internal stress, crack propagation, and environmental factors. Secondly, the data fusion processing module performs spatiotemporal registration, denoising, and normalization on multi-source monitoring data, and generates a consistent slope deformation dataset through data fusion algorithms, eliminating the limitations of a single data source. Then, the intelligent identification module automatically identifies abnormal slope deformation areas and disaster precursor characteristics based on a deep learning model, including landslide segmentation, crack detection, and stability assessment, to achieve real-time diagnosis of deformation risk; next, the deformation prediction module predicts the future deformation trend of the slope and the probability of disaster occurrence based on historical deformation data and real-time monitoring information through time series analysis and machine learning algorithms, and outputs short-term, medium-term, and long-term prediction results. Finally, the central control module coordinates the operation of each module in the system, schedules data flow and computing resources, controls the operation of the early warning feedback unit according to the preset early warning threshold, and provides a human-machine interface to support the setting of monitoring parameters, real-time data visualization and export of prediction reports.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A highway slope deformation intelligent identification and prediction system based on three-dimensional air-space-ground monitoring, characterized in that: The system includes an air-space-ground integrated monitoring module, a data fusion processing module, an intelligent identification module, a deformation prediction module, and a central control module. The air-ground-space three-dimensional monitoring module is used to collect deformation data of highway slopes through multi-source heterogeneous sensors, including space satellite remote sensing data, sky drone aerial photography data and ground Internet of Things sensor data, to realize all-weather, multi-scale monitoring of slope surface displacement, internal stress, crack propagation and environmental factors. The data fusion processing module is used to perform spatiotemporal registration, denoising, and normalization on multi-source monitoring data, and to generate a consistent slope deformation dataset through data fusion algorithms, thereby eliminating the limitations of a single data source. The intelligent identification module is used to automatically identify abnormal slope deformation areas and disaster precursor features based on a deep learning model, including landslide segmentation, crack detection, and stability assessment, to achieve real-time diagnosis of deformation risk. The deformation prediction module is used to predict the future deformation trend and probability of disaster occurrence of the slope based on historical deformation data and real-time monitoring information, through time series analysis and machine learning algorithms, and outputs short-term, medium-term and long-term prediction results. The central control module is used to coordinate the operation of various modules in the system, schedule data flow and computing resources, control the operation of the early warning feedback unit according to the preset early warning threshold, and provide a human-machine interface to support monitoring parameter setting, real-time data visualization and prediction report export.

2. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 1, characterized in that: The air-space-ground integrated monitoring module includes a satellite remote sensing unit, an unmanned aerial vehicle (UAV) aerial photography unit, and a ground sensing unit. The satellite remote sensing unit acquires large-scale, periodic deformation information of the slope through synthetic aperture radar satellites and optical satellites, supporting millimeter-level displacement monitoring; The drone aerial photography unit, equipped with a multispectral camera and a lidar sensor, enables high-resolution 3D modeling and local deformation capture of the slope. The ground sensing unit consists of tilt sensors, strain gauges, crack gauges, and temperature and humidity sensors deployed at key locations on the slope, which collect internal deformation and environmental parameters in real time.

3. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 2, characterized in that: The satellite remote sensing unit is also equipped with an interferometric radar data processing component, which is used to extract phase information of small deformations of the slope through differential interferometry. The drone aerial photography unit has autonomous path planning and obstacle avoidance mechanisms, making it suitable for dynamic monitoring tasks in complex terrain. The ground sensing unit adopts a wireless sensor network architecture, which supports low-power wide-area communication, real-time data transmission, and node self-organizing network.

4. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 1, characterized in that: The data fusion processing module includes a data preprocessing unit, a fusion algorithm unit, and a quality assessment unit; The data preprocessing unit is used to perform outlier removal, missing value interpolation, and coordinate unification on the raw monitoring data. The fusion algorithm unit employs the Kalman filtering method to fuse uncertainties from multiple data sources. The quality assessment unit verifies the accuracy of the fused data through consistency indicators and error analysis, and provides data credibility labels for subsequent modules.

5. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 4, characterized in that: The data preprocessing unit integrates adaptive filtering technology, which dynamically adjusts the filtering parameters according to the data characteristics. The fusion algorithm unit supports a multi-scale fusion strategy, which processes global satellite data, regional UAV data and local ground data respectively, and retains feature information at different levels. The quality assessment unit introduces the entropy weight method to quantify the contribution of data sources and optimize the allocation of fusion weights.

6. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 1, characterized in that: The intelligent recognition module includes a feature extraction unit, a model training unit, and an anomaly detection unit; The feature extraction unit is used to extract deformation-related features from the fused data, namely displacement rate, curvature change and spectral characteristics; The model training unit constructs a recognition model based on a convolutional neural network and trains model parameters using historical disaster data. The anomaly detection unit identifies signs of slope instability, namely accelerated displacement and crack penetration, through real-time reasoning, and outputs a risk level score.

7. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 6, characterized in that: The feature extraction unit combines principal component analysis dimensionality reduction technology to reduce data redundancy; The model training unit adopts the transfer learning method, which uses the pre-trained model to accelerate convergence and improve the recognition accuracy under small sample conditions. The anomaly detection unit integrates an attention mechanism to focus on high-risk areas and reduce the false alarm rate.

8. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 1, characterized in that: The deformation prediction module includes a trend analysis unit, a probability prediction unit, and a risk assessment unit. The trend analysis unit predicts the deformation over time using support vector machine regression. The probability prediction unit calculates the probability of disaster occurrence based on Monte Carlo simulation and outputs a confidence interval. The risk assessment unit combines geological environmental factors and prediction results to generate a comprehensive slope stability rating.

9. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 8, characterized in that: The trend analysis unit has multi-step prediction capabilities and supports rolling prediction and dynamic correction. The probability prediction unit introduces extreme value theory to handle rare events, providing deformation samples for identifying and processing rare events; The risk assessment unit integrates fuzzy logic methods to handle uncertainty and subjective judgment, and outputs a visualized risk map.

10. The intelligent identification and prediction system for highway slope deformation based on three-dimensional monitoring of air, space, and ground as described in claim 1, characterized in that: The central control module specifically includes a task scheduling unit, a resource coordination unit, and a human-computer interaction unit; The task scheduling unit is used to dynamically coordinate the working mode and acquisition frequency of each sensor unit in the air-space-ground integrated monitoring module according to the preset monitoring task priority and the status of the data stream, and automatically trigger the analysis and calculation process of the intelligent recognition module and the deformation prediction module. The resource coordination unit is used to dynamically allocate the computing resources and manage the storage of the slope deformation dataset. The human-computer interaction unit is used to provide the human-computer interaction interface to set monitoring parameters and the early warning threshold, and to visualize real-time monitoring data, deformation recognition results and risk prediction reports.

Images