A ground-air combined highway slope collapse measuring system and method

Abstract

This invention provides a ground-air integrated system and method for measuring highway slope collapse, relating to the field of highway slope inspection. The system utilizes a scientific-grade CMOS camera and digital image correlation algorithms in the ground detection module to achieve all-weather, non-contact acquisition of millimeter-level displacement data at key slope points, covering steep slopes. The aerial detection module uses a UAV equipped with synthetic aperture radar and multispectral sensors to acquire overall three-dimensional deformation field and soil moisture data of the slope. At the data layer, dual-function markers and the least squares method are used for high-precision spatiotemporal registration; at the feature layer, ground displacement data is used to calibrate aerial deformation data, achieving correlation between macroscopic and microscopic features; at the model layer, an improved U-Net network and a hybrid prediction model are constructed to achieve accurate crack tracking and collapse probability prediction. Visualization is achieved through a digital twin platform, triggering tiered early warnings linked to the traffic management system. This invention overcomes the limitations of single detection methods, realizing accurate perception, intelligent prediction, and dynamic response for highway slope safety.

Description

Technical Field

[0001] This invention relates to the field of highway slope inspection, specifically to a ground-air combined highway slope collapse measurement system and method. Background Technology

[0002] As a key structure in highway engineering, highway slopes not only play a vital role in maintaining road stability and ensuring road smoothness, but also serve as a direct barrier for driving safety. However, due to the combined effects of complex geological structures, extreme weather conditions, and frequent human activities, slope structures are prone to decreased stability, which can then trigger various geological disasters such as landslides, collapses, and uneven settlement.

[0003] To address the aforementioned slope issues, regular inspections and monitoring of slope information are necessary. Traditional manual inspection methods suffer from low levels of automation, poor environmental adaptability, and insufficient ease of operation. Manual inspections require personnel to walk along the slope line, visually inspecting surface cracks and soil spalling. This method is significantly limited by terrain, easily creating blind spots in areas difficult to access, such as steep slopes and canyon sections. Single ground-based detection technologies also have limitations in coverage and data correlation. Existing ground-based detection methods often employ displacement and pressure sensors to detect localized points on the slope. While these technologies can obtain high-precision data at single points, sensor deployment density is limited by cost and terrain, making it difficult to achieve full coverage of large slope areas. Sparse detection points can lead to missed collapse risks. Single aerial detection technologies also suffer from insufficient data accuracy. In some scenarios, drones equipped with visible light cameras or LiDAR are used for aerial inspections. While this can achieve large-scale imaging of slope topography, it cannot accurately identify minute cracks on the slope surface or loosening of shallow soil. Furthermore, aerial inspection can only acquire surface geometric data of the slope and cannot penetrate the surface soil to obtain key information such as deep displacement and stress changes.

[0004] Therefore, this invention provides a ground-air combined highway slope collapse measurement system and method. By combining the advantages of ground and air detection, achieving multi-dimensional data fusion, and adapting to complex working conditions, it detects highway slope collapses. Ground detection uses a scientific-grade CMOS (Complementary Metal Oxide Semiconductor) image sensor to acquire slope images and uses a Digital Image Correlation (DIC) algorithm to measure whether the slope soil has displacement data. By monitoring the displacement data of the slope soil over a long period of time, it predicts whether the highway slope will collapse. Aerial detection uses a UAV equipped with radar and multispectral sensors to establish a three-dimensional terrain model with millimeter-level accuracy. It dynamically detects the deformation trend of high slopes and deep foundation pit areas, introduces an improved U-Net network, realizes automatic crack propagation tracking and collapse probability prediction, and constructs a real-time early warning mechanism for high-risk areas. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is to provide a ground-air combined highway slope collapse measurement system and method, which solves the problems of existing single detection methods, reduces the detection blind spots of manual inspection, and improves the stability and accuracy of ground and aerial detection.

[0006] To address the aforementioned technical problems, the present invention provides a ground-air integrated highway slope collapse measurement system and method. This includes establishing a unified ground-air data calibration system at the data layer, accurately registering spatial coordinates, spatially anchoring two types of detection data using an engineering coordinate system as a reference, synchronizing temporal data, establishing a dynamic time axis calibration mechanism to address the temporal differences between long-term continuous ground detection and periodic aerial inspection, and constructing a unified system to address the format differences between the two types of data. At the algorithm layer, a ground-air feature fusion model is constructed to improve the longitude of risk prediction. Then, local displacement and overall terrain are correlated and analyzed, while crack tracking and displacement data are collaboratively corrected. Finally, the collapse probability prediction is fused temporally. At the application layer, a ground-air linkage decision-making platform is built, simultaneously realizing detection, early warning, and processing functions. A ground-air data fusion visualization platform is developed to intuitively present the slope stability status. Based on the fusion analysis results, the allocation of ground-air detection resources is dynamically adjusted to improve detection efficiency. Finally, a digital twin and early warning platform are output, establishing a hierarchical early warning mechanism based on ground-air integration.

[0007] In a preferred embodiment, the present invention also provides a ground slope detection system, including a scientific-grade CMOS camera to capture minute texture changes in the slope soil, providing a high-resolution image source for the DIC algorithm. During nighttime detection, an LED incandescent lamp fixed at the measurement location is used as a target. Combined with the high light sensitivity of the scientific-grade camera, nighttime measurements are consistent with daytime longitude measurements, thus compensating for the inability to operate at night. The maximum measurement distance reaches 524 meters, covering the long-distance detection needs of steep slopes on deep mountain highways. Simultaneously, utilizing a non-contact design, there is no need to install traditional sensors such as displacement gauges and dial indicators. Measurement is achieved through optical targets or natural textures, avoiding the time-consuming installation / disassembly and susceptibility to slope soil sliding damage associated with traditional sensors, thus reducing the maintenance costs of long-term highway slope detection. The DIC algorithm is used to measure slope soil displacement, transforming the implementation method from slope soil texture matching to slope soil displacement quantification. A specific sub-region of the slope is selected, and the sub-region with the highest similarity is searched in the real-time image. The horizontal / vertical displacement of that point is calculated using the coordinate difference to obtain the slope soil displacement data.

[0008] In a preferred embodiment, the present invention also provides a slope inspection drone system, which includes using the differential interferometry capability of synthetic aperture radar (DLidar-SAR) to compare the phase of data collected at different time points and directly calculate the minute deformation of the ground surface at the centimeter level. This system is suitable for detecting slope soil creep. It can also use specific infrared bands of multispectral sensors to invert the surface soil moisture. Areas with abnormally high moisture levels are associated with seepage points and potential sliding surfaces, enabling the prediction of landslide phenomena. In terms of algorithms, targeted improvements were made to the U-Net network for slope detection. The input was expanded to include optical imagery (RGB), radar intensity maps, digital surface models (DSM), slope maps, and soil moisture index maps as additional input channels. When identifying cracks, the algorithm can detect contextual information such as terrain, humidity, and radar reflection characteristics, achieving multi-source data input fusion. A penalty term was introduced into the loss function, enabling the network to identify cracks with consistent main deformation directions on the slope and suppress irrelevant linear features. Core parameters, including crack length, width, and propagation speed, were automatically calculated from the segmentation results and calibrated with ground displacement vector field data. When crack width changes were detected in the air, displacement data of the corresponding area in the ground displacement vector field was retrieved. If the trends of the two changes were consistent, it was considered an effective crack propagation. Attached Figure Description

[0009] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0010] Figure 1 This is a diagram of the architecture of the ground-air detection combined with highway slope collapse detection system of the present invention;

[0011] Figure 2This is a diagram of the improved U-Net network structure of this invention;

[0012] Figure 3 This is a flowchart of the ground detection module of the present invention;

[0013] Figure 4 This is a flowchart of the aerial inspection module of the present invention;

[0014] Figure 5 This is a calculation example diagram of the ground detection system of the present invention;

[0015] Figure 6 This is a calculation example diagram of the aerial inspection system of the present invention. Detailed Implementation

[0016] To better understand the purpose, system architecture, and functional implementation of this embodiment, the embodiments and features in the embodiments of this application can be combined with each other without conflict. The exemplary embodiments disclosed in this application will be described below with reference to the accompanying drawings, which include specific technical details disclosed in this embodiment to aid understanding; however, these details should be considered exemplary rather than restrictive. Therefore, those skilled in the art should understand that various improvements and adjustments can be made to the embodiments described herein without departing from the scope and core ideas of the invention. Similarly, for clarity, detailed descriptions of well-known technologies, functions, and structures (such as standard image processing algorithms and common communication protocols) are omitted in the following description.

[0017] Example 1

[0018] Figure 1 This is a diagram illustrating the architecture of the ground-air detection combined method of the present invention.

[0019] like Figure 1As shown, this invention proposes a ground-air combined detection method. A scientific-grade CMOS camera at a fixed ground-based detection station captures minute texture changes in the slope soil, providing a high-resolution image source for the DIC algorithm. An unmanned aerial vehicle (UAV) conducts aerial inspection using interferometric radar to detect slope soil creep. Specific infrared bands of a multispectral sensor are used to invert surface soil moisture. Areas with abnormally high moisture levels are correlated with seepage points and potential sliding surfaces to predict landslides. After extracting the slope data, the data is fused and analyzed through data, feature, and model layers to output a digital twin and early warning platform. In the data layer, fusion is achieved through spatiotemporal benchmark unification and data correlation. Permanent, specially designed optical targets with high-precision coordinates are deployed within the slope detection area. These points can be captured by both ground cameras and UAV sensors. Ground displacement vector fields, UAV 3D models, and crack locations are all transformed into the same absolute coordinate system. The coordinate deviations between the ground displacement vector field and the UAV 3D model data on the target are corrected using the least squares method, ensuring that the spatial matching accuracy of the ground-air data is controlled within ±3mm. Establish an event- and periodic plan-based triggering mechanism. The periodic plan requires a full-area scan by UAVs once a week, which triggers all ground monitoring stations to collect data at high frequency. Event-triggered mechanisms will automatically increase the collection frequency of ground monitoring points in the slope area when aerial detection finds significant deformation or crack expansion in the slope soil. When the ground displacement vector field detects abnormal accelerated displacement, UAVs can be automatically dispatched to conduct an emergency detailed investigation of the slope area.

[0020] At the feature layer, a complete perception is formed through complementary advantages. The feature information acquired by the two technologies is stitched together and complemented. First, the UAV 3D model provides the detection area for the ground displacement vector field. The high-precision terrain model acquired by the UAV can initially identify the macroscopic deformation zone and potential sliding surface of the entire slope. The system can automatically optimize or virtually deploy ground detection points in these key areas. The ground displacement vector field provides accuracy verification for the UAV model. The ground-based scientific-grade CMOS camera and DIC algorithm can provide millimeter-level continuous displacement event sequence data. The data is used as a benchmark to verify and calibrate the accuracy of the surface displacement data obtained by UAV radar measurement. At the same time, the small displacements detected by the ground displacement vector field are early microscopic signs of slope instability and are one of the features used to predict slope instability. By correlating multispectral data with displacement, feature analysis is aided. The UAV multispectral sensor can analyze features such as vegetation health and soil moisture, which are indirect indicators of slope internal moisture changes and stability. When the ground displacement vector field detects accelerated displacement at a certain point in the slope soil, the system can link up to view the multispectral data at that location to determine whether it is caused by rainfall infiltration or rising groundwater level, thereby assisting in the analysis of mechanistic characteristics.

[0021] like Figure 2As shown, by inputting two types of data into the same analysis model at the model layer, more accurate slope prediction is achieved. First, an additional input channel is added to the U-Net network. The calculated displacement vector field and UAV optical images are input into the U-Net network together. Considering the special characteristics of slope crack identification and the need for multi-source data fusion, the network's input layer is expanded to multiple channels. Optical images (RGB, 3 channels), radar intensity map (1 channel), digital surface model (1 channel), slope map (1 channel), soil moisture index map (1 channel), and ground displacement vector field (2 channels, horizontal / vertical displacement) are converted into 512×512 pixel feature maps. They are then stitched together according to the channel dimension to form a 10-channel input feature tensor, which solves the limitation of traditional U-Net with only single or 3-channel input and realizes collaborative representation of multi-source data.

[0022] The encoder stage has 5 downsampling layers, each containing two 3×3 convolutions and one 2×2 max pooling (stride 2). The number of feature channels increases from 16 to 128. The decoder stage has 5 upsampling layers, each using a 2×2 transposed convolution (stride 2) to double the feature map size. After upsampling, the feature maps are spliced ​​with the corresponding layer feature maps from the encoder through skip connections to retain detailed features.

[0023] In the standard binary cross-entropy loss function Based on this, add a penalty term based on the slope deformation direction. This penalty term rewards predictions of cracks with consistent direction and penalizes predictions with excessively different direction by calculating the consistency between the predicted crack direction and the potential sliding surface of the slope. This guides the network to identify structural cracks that pose engineering hazards, as shown in the following formula.

[0024]

[0025] in, The weighting coefficient for the penalty term. Standard binary cross-entropy loss, This is a penalty term for the direction of slope deformation. The calculation formula is shown below.

[0026]

[0027] in, For pixel-based true labels, To predict probabilities, Total number of pixels The calculation formula is shown below.

[0028]

[0029] in, To predict the number of pixels in the crack region, For the first The orientation angle of each crack pixel. The potential sliding surface direction angle of the slope. Predict the probability of a crack for this pixel. When the crack direction is consistent with the sliding surface direction, i.e. and The closer, When the value is close to 1, there is no additional penalty; however, when the direction of the crack differs significantly from the direction of the sliding surface... and The angle between them is close to 90°. When the value approaches zero, the penalty term becomes significantly larger. The penalty is calculated only for pixels predicted as cracks, avoiding interference with non-crack areas. Through this loss function, the network can prioritize learning structural cracks related to slope instability, suppressing misidentification of irrelevant linear features. This invention's improved U-Net network can not only learn surface cracks but also perceive and learn unseen strain concentration zones around cracks, significantly improving the accuracy of crack identification, segmentation, and propagation tracking.

[0030] A collapse probability prediction model is established by jointly calculating the collapse probability using a Long Short-Term Memory (LSTM) network and an XGBoost (eXtreme Gradient Boosting) network. The LSTM network learns the long-term dependencies in the displacement sequence and outputs a feature vector representing the current temporal deformation state of the slope. The spatial and environmental features acquired by the aerial detection module, including the difference in three-dimensional terrain displacement between the current and previous inspections, crack expansion area and length, soil moisture index, and vegetation index (NDVI), are constructed into a feature vector. The time-series feature vector output by LSTM Spatial environment feature vector The features are then concatenated to form the final fused feature vector. This fused feature vector is then input into the XGBoost model for training and prediction. The XGBoost model leverages its powerful non-linear fitting capability, based on... The system outputs a collapse probability value between 0 and 1. Ground detection, based on the evolution trend of displacement time series, can obtain the probability of slope collapse. UAV detection, based on the deformation and crack propagation of the 3D model, can also obtain the probability of slope collapse through U-Net network analysis. The system uses a weighted average to fuse the two probability values ​​into a more accurate comprehensive collapse probability prediction. The weight allocation rule of the weighted average algorithm is as follows: when the continuous collection time of ground displacement data is ≥7 days, the weight of the first collapse probability is 0.6 and the weight of the second collapse probability is 0.4; when the continuous collection time of ground displacement data is <7 days, the weight of the first collapse probability is 0.4 and the weight of the second collapse probability is 0.6.

[0031] All fused data, features, and prediction results are ultimately visualized on a digital twin platform for highway slopes. Displacement cloud maps generated from UAV data are displayed in different colors on a 3D terrain model. Slope gradient, slope ratio, and aspect are measured and displayed on the model. Precise displacement calculations are performed for ground-based detection points on the measured slopes. A tiered early warning mechanism based on ground-air integration is established. Three levels of warnings are set based on the fused displacement, crack propagation, and collapse probability. A yellow warning automatically pushes warning information to maintenance personnel, prompting them to strengthen manual verification. An orange warning initiates temporary traffic control, simultaneously dispatching UAVs to increase inspection frequency and ground sensors to collect data. A red warning activates road closure plans, pushes warning information to traffic management departments and emergency response teams, and generates disposal suggestions. After each warning, the raw ground-air integrated data and prediction results are automatically stored for subsequent model optimization.

[0032] Figure 3 This is a flowchart of the ground detection module of the present invention.

[0033] like Figure 3 As shown, specially designed optical targets are installed at key locations on the slope surface, such as potential slip surfaces and both sides of cracks, that need to be monitored. A high-performance scientific-grade CMOS camera is fixed at a stable reference point far from the slope, ensuring its field of view covers all targets. The camera automatically acquires images according to a preset frequency. During the daytime when there is sufficient light, it directly uses natural light for imaging; at night or when there is insufficient light, it automatically triggers LED targets for supplemental lighting, achieving all-weather monitoring. The DIC algorithm continuously tracks the position of each target in the image, performing sub-region search or tracking. By matching the current image with the initial reference image, the displacement vector of each target sub-region is calculated, and the similarity between two sub-regions is quantitatively evaluated. As shown in the following formula.

[0034]

[0035] in, The grayscale value function for the reference sub-region. The grayscale value function for the target sub-region. The coordinates of the center point of the reference sub-region, The coordinates of the center point of the target sub-region are: It is the average gray value of the reference sub-region. It is the average gray value of the target sub-region. The DIC algorithm calculates and visualizes the pixel coordinates contained in a sub-region by quantitatively comparing the similarity of sub-regions in two images, thereby accurately tracking the motion of objects. Figure 5 As shown, three targets, WY19, WY20 and WY21, are set, and the displacement vector of each target sub-region is calculated and uploaded to the system.

[0036] Figure 4 This is a flowchart of the air detection module of the present invention.

[0037] like Figure 4 As shown, the system starts in two modes: a regular survey performed according to a preset schedule, and an emergency detailed survey triggered by ground system alarms or extreme weather events. After determining the mode, automated path planning is performed before flight to ensure standardized flight routes. After the UAV takes off automatically, it simultaneously collects optical images, radar intensity maps, and multispectral image data along the flight route. The collected data is automatically uploaded to the cloud platform for processing. A high-precision digital surface model is generated through 3D modeling. Differential interferometric synthetic aperture radar technology is used to compare the current data with historical data to calculate the surface deformation. Figure 6 As shown, parameters such as slope, slope ratio, and slope aspect are calculated. By improving the U-Net network, optical images are intelligently analyzed to identify and quantify cracks. Multispectral data is used to analyze soil moisture and vegetation health as indirect indicators of slope stability.

[0038] Multi-dimensional features, including deformation, cracks, and environmental characteristics, are fused and input into a hybrid model combining LSTM and XGBoost. The LSTM layer extracts temporal features, while the XGBoost layer fuses spatial features. This addresses the issue that a single model cannot simultaneously capture both temporal trends and spatial differences, resulting in a quantified collapse probability model. The LSTM network learns long-term dependencies in the displacement sequence and outputs a feature vector representing the current temporal deformation state of the slope. The spatial and environmental features acquired by the aerial detection module, including the difference in three-dimensional terrain displacement between the current and previous inspections, crack expansion area and length, soil moisture index, and vegetation index (NDVI), are constructed into a feature vector. ,in, This represents the average value (m) of the three-dimensional terrain displacement difference. The total area of ​​the crack expansion (m²) The maximum crack propagation length (m) is given. The average soil moisture index. The average value of the Normalized Difference Vegetation Index (NDVI) is given. Other related features.

[0039] The time-series feature vector output by LSTM Spatial environment feature vector The features are then concatenated to form the final fused feature vector. ,in, , This indicates a vector concatenation operation.

[0040] This fused feature vector is input into the XGBoost model for training and prediction, and the model outputs... As shown in the following formula.

[0041]

[0042] in, This represents the total number of decision trees in the XGBoost model. To predict the output of the k-th decision tree, each tree focuses on learning the unfitted residuals of the preceding trees, and iteratively optimizes the model using the nonlinear fitting capability of the XGBoost model. Output a collapse probability value between 0 and 1, which intuitively reflects the level of collapse risk.

[0043] The system makes intelligent decisions based on the output of the probability model. When the probability exceeds the threshold, an early warning is issued; if an anomaly is detected but the probability is low, it is marked as important; if no anomaly is detected, the baseline data is updated. When a risk is detected, the system immediately triggers a linkage mechanism, sending the early warning information and precise location to the monitoring center and notifying the ground monitoring system to focus on monitoring the risk area.

[0044] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0045] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A ground-air combined highway slope collapse measurement system, characterized by comprising: a ground detection module that collects high-resolution images of natural textures of the slope and calculates displacement data of the soil body of the slope; an air detection module for collecting data of the slope to establish a three-dimensional terrain model to identify crack data; a data processing module for correlating and analyzing the displacement data obtained by the ground detection module and the three-dimensional terrain model and crack data obtained by the air detection module; a prediction model that integrates ground and air data is constructed to output the collapse probability of the slope; the prediction model is an improved U-Net network, and on the basis of a standard binary cross-entropy loss function, a penalty term based on the deformation direction of the slope is added, the penalty term rewards the prediction of cracks with consistent direction and punishes the prediction of cracks with large direction difference, thereby guiding the network to identify structural cracks with engineering hazards, as shown in the following formula: ； wherein, is a weight coefficient of the penalty term, is a standard binary cross-entropy loss, is a slope deformation direction penalty term; The calculation formula is as follows: ； wherein, is the pixel true label, is the predicted probability, is the total number of pixels; The calculation formula is as follows: ； wherein, is the number of pixels in the crack region, is the direction angle of the th crack pixel, is the direction angle of the potential sliding surface of the slope, is the crack prediction probability of the pixel.

2. The ground-to-air combined highway slope collapse measuring system according to claim 1, wherein, the spatio-temporal registration of the data processing module includes: by arranging double-function marker points with ground DIC target and unmanned aerial vehicle control point functions in the slope monitoring area and correcting errors, the ground displacement data are used to calibrate the accuracy of the air deformation data; a trigger mechanism based on events and regular plans is established to realize the timing synchronization and task linkage of ground continuous monitoring and air regular inspection.

3. The ground-to-air combined highway slope collapse measuring system according to claim 1, wherein, the correlation and complementary analysis of the data processing module includes: the three-dimensional terrain model obtained by the air detection module is used to identify macroscopic deformation zones and potential sliding surfaces to guide the optimal arrangement of monitoring points in the ground detection module; the millimeter-level continuous displacement time series data obtained by the ground detection module are used to verify and calibrate the accuracy of the surface displacement data obtained by the radar measurement of the air detection module.

4. The ground-to-air combined highway slope collapse measurement system according to claim 1, characterized in that, the improved U-Net network, in view of the particularity of slope crack identification and the demand of multi-source data fusion, the input layer of the network is expanded to multiple channels to simultaneously receive different types of data including optical images, radar intensity maps, digital surface models, slope maps, soil moisture index maps and ground displacement vector fields, the above data are standardized and normalized to have the same spatial resolution and be aligned, and the ground displacement vector field is a continuous field with the same resolution as the optical image.

5. The ground-to-air combined highway slope collapse measurement system according to claim 1, characterized in that, the prediction model is a collapse probability prediction model, and the input features include: the historical displacement time series data of the ground detection module, the difference data of the three-dimensional terrain displacement and crack propagation obtained by the air detection module in multiple inspections, and the vegetation index and soil moisture features obtained by multispectral data inversion; the time series feature vector and the spatial environment feature vector are spliced to form the final fusion feature vector, which is input into the collapse probability prediction model for training and prediction, and finally the collapse probability value is output.

6. The ground-to-air combined highway slope collapse measurement system according to claim 1, characterized in that, the system further includes a digital twin and early warning platform, including: visualize the fused ground and air data, and display the displacement vector field in the form of a heat map on the three-dimensional terrain model, and superimpose the crack distribution and monitoring point data; Based on the displacement, crack propagation and collapse probability obtained by fusion analysis, hierarchical early warning is implemented, and corresponding response measures are automatically triggered according to the warning level, including sending an alarm to maintenance personnel, dispatching unmanned aircraft for increased patrol, adjusting ground monitoring frequency or starting traffic control plan.

7. A method for measuring a highway slope collapse in combination with ground and air, characterized in that, The ground-air combined highway slope collapse detection system of any one of claims 1-6, the method comprising the following steps: S1, ground-air data acquisition, through the scientific CMOS camera of the ground detection module, collecting optical target / natural texture images according to the set frequency, generating millimeter-level displacement time series data, through the unmanned aerial vehicle of the air detection module, collecting optical images, synthetic aperture radar deformation data and multispectral data by periodic census once a week or emergency detailed investigation triggered by ground anomalies; S2, data space-time registration, in the data processing and fusion center, taking the double-function landmark point as the reference, unify the ground-air data to the engineering coordinate system, synchronize the data through the time stamp, and complete the missing data by interpolation of ground time series data; S3, feature fusion analysis, input the ground displacement vector field into the improved U-Net network, and fuse with the air multi-source data to realize crack identification, calibrate the air deformation data through the ground displacement data, and associate the macro deformation and local displacement characteristics; S4, collapse probability prediction, input the ground-air fusion features into the LSTM and XGBoost mixed model, and output the collapse probability; S5, visualization and early warning, display the fusion data through the digital twin platform, trigger hierarchical early warning and response measures based on the collapse probability.

8. The ground-to-air combined highway slope collapse measurement method according to claim 7, characterized in that, The feature fusion in step S3 includes: The ground displacement vector field calculated by the DIC algorithm is taken as a feature map, and the optical image of the unmanned aerial vehicle is input into the improved U-Net network to improve the perception ability of crack identification to invisible strain concentration areas; The first collapse probability obtained based on the ground displacement time series data is fused with the second collapse probability obtained based on the three-dimensional model deformation and crack propagation of the air by using the weighted average algorithm to generate a comprehensive collapse probability.

9. The ground-to-air combined highway slope collapse measuring method according to claim 7, characterized in that, The hierarchical early warning in step S5 includes: When the fusion collapse probability exceeds the first threshold, a yellow warning is issued, and information is automatically pushed to maintenance personnel and manual review is prompted; When the fusion collapse probability exceeds a higher second threshold, an orange warning is issued, temporary traffic control is started, and unmanned aerial vehicles and ground sensors are dispatched to increase monitoring frequency; When the fusion collapse probability exceeds the highest third threshold, a red warning is issued, a road closure plan is started, and the traffic management department and rescue team are linked.

Images