Rapid automatic comprehensive detection method for urban road underground cavity

Abstract

The invention relates to the field of underground engineering, and discloses a rapid automatic comprehensive detection method for an urban road underground cavity, which comprises the following steps: S1, carrying out rapid general survey scanning on a road through a vehicle-mounted ground penetrating radar system to obtain an underground B-scan image of the road, processing a radar image by using a deep learning segmentation model, and outputting a cavity probability graph, identifying a suspected cavity area; s2, automatically positioning a suspected area; s3, starting signals are transmitted to the seismic wave sensor array and the ground penetrating radar controller at the same time; s4, performing fusion processing on the pavement elasticity modulus subjected to FWD inversion; and S5, a spatial interpolation algorithm based on the digital twinborn model. Through hardware-level cooperative control of impact loading and multi-source sensing, a five-dimensional synchronous controller triggers working mode switching of a seismic wave array and a ground penetrating radar, time-space alignment is carried out, and equipment response time difference and mechanical vibration interference are eliminated, so that rapid and accurate detection of the underground cavity of the road is ensured.

Description

Technical Field

[0001] The present invention relates to the field of underground engineering technology, in particular to a rapid and automated comprehensive detection method for underground cavities in urban roads. Background Art

[0002] With the continued expansion of urban road networks and the increasing aging of infrastructure, road collapse accidents caused by underground cavities are becoming more frequent. Underground cavities are characterized by their concealment, numerous causes, and difficulty in predicting them. Detecting underground cavities and defects not only provides insights into the structural layers of structures but also prevents ground collapses, providing strong safeguards for sustainable urban development and the safety of people's lives and property. Traditional techniques for detecting underground cavities rely on a combination of single-point scanning with ground-penetrating radar and manual verification.

[0003] However, current technologies combine single-point scanning with manual review and verification, resulting in low survey efficiency, poor coordination of multi-source data, and rough verification target positioning. This makes it difficult to meet the needs of rapid and non-destructive detection of urban road networks, leading to inaccurate detection of underground voids in roads. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a rapid, automated, and comprehensive detection method for underground cavities in urban roads, which solves the problems of low survey efficiency, poor coordination of multi-source data, and rough verification target positioning, which lead to inaccurate detection of underground cavities in roads.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a rapid and automated comprehensive detection method for underground cavities in urban roads, comprising the following steps: S1. Use the vehicle-mounted ground-penetrating radar system to quickly scan the road and obtain B-scan images of the road's underground. Then use the deep learning segmentation model to process the radar images and output a cavity probability map to identify suspected cavity areas. S2. Automatically locate the suspected area, control the drop weight deflectometer to impact load the target point, and synchronously trigger the start signal from the five-dimensional synchronous controller; S3, a start signal is simultaneously transmitted to the seismic wave sensor array and the ground penetrating radar controller, causing the seismic wave array to collect vibration wave data generated by the impact load, while the ground penetrating radar switches to a high-frequency focused scanning mode; S4. Fusion processing of the pavement elastic modulus derived from FWD inversion, deep wave velocity data from seismic imaging, and focused radar images to construct a three-dimensional digital twin damage model; S5. Based on the spatial interpolation algorithm of the digital twin model, the risk entropy value of each voxel is calculated, and the drilling path planning is output.

[0006] Through the above scheme, the cavity probability map is generated through the rapid survey scanning and deep learning recognition of the vehicle-mounted ground penetrating radar, the drop hammer impact loading is automatically guided and the five-dimensional synchronous control is triggered. With the help of hardware-level synchronization signals, millisecond-level coordination of seismic wave acquisition and ground penetrating radar focused scanning is achieved. The elastic modulus, seismic wave velocity and radar dielectric image of bending inversion are integrated to construct a three-dimensional digital twin model. Finally, the drilling verification path is dynamically output based on the risk entropy value and spatial constraints, forming remote sensing positioning, impact excitation, multi-field perception, model reconstruction and decision execution, realizing the integrated intelligent detection of urban road cavity survey, detailed inspection and verification, and significantly improving the detection efficiency and accuracy.

[0007] Preferably, the S1 specifically includes the following steps: S101, preprocessing the original B-scan image collected by the ground penetrating radar, including Hilbert transform to enhance the reflection interface, background noise filtering and time-to-depth conversion; S102: Input the pre-processed data into the deep learning segmentation model and perform the following operations: Encoder extraction, extracting multi-scale features through a 4-level downsampling structure, each level contains a convolutional layer and a maximum pooling layer; Atrous feature fusion: Atrous convolutions with dilation rates of 1, 3, and 6 are performed in parallel on the underlying features, and the output feature maps are concatenated along the channel dimension. Decoding and reconstruction, gradually upsampling through deconvolution layers and fusing with encoder features of the same scale; Probability output, perform Sigmoid activation on the final feature map to generate a pixel-level hole probability map; S103: Generate suspected void areas based on the probability map, perform morphological optimization on abnormal areas with an area less than 0.1 m2, and output a vector bounding box with geographic coordinates.

[0008] Preferably, the step S2 specifically includes the following steps: S201, based on the vector anomaly region bounding box, extract the coordinates of the peak point of the void probability (longitude λ, latitude φ) as the impact test target; S202, sending a positioning command to the drop weight deflectometer via the vehicle-mounted PLC controller to drive it to move directly above the target point, with a positioning error of ≤5 cm; S203, trigger the five-dimensional synchronous controller to execute: Start the free fall of the drop weight; Synchronously activate the Doppler laser velocimeter to collect road deformation; Synchronously turn on the dynamic strain gauge array; Simultaneously start a high-definition camera to capture cracks in the road surface; Synchronously record GPS timestamps.

[0009] Preferably, the step S3 specifically includes the following steps: S301, sending a trigger instruction to the seismic wave sensor array through the GPS timestamp signal to start seismic wave data collection; S302, synchronously sending a mode switching instruction to the ground penetrating radar controller, including switching to a 1.6 GHz high-frequency focused scanning mode, focusing the detection range to scan the shallow surface layer, and increasing the scanning line spacing to centimeter-level accuracy; S303 , setting a device startup sequence based on the impact time t0 , including starting sampling by the seismic wave array before t0 and starting scanning by the ground penetrating radar after t0 .

[0010] Preferably, the 1.6 GHz high frequency focused scanning mode includes: Electromagnetic beam focusing control, adjust the antenna array element phase offset Δφ to meet: Where: d is the array element spacing, λ is the wavelength of the electromagnetic wave in the medium, and θ is the beam pitch angle; Layer-fold scanning path planning, generating a spiral scanning path with the impact target as the center; Real-time inversion of dielectric constant, dynamic correction of dielectric constant based on reflected wave amplitude: Where: ε init is the initial dielectric constant, A max is the measured maximum reflected wave amplitude, A theory is the theoretical amplitude value, according to Calculation, k is the antenna gain coefficient, r is the electromagnetic wave propagation distance.

[0011] Preferably, the S4 specifically includes the following steps: S401. Analyze the deflection time history curve collected synchronously in five dimensions and invert the elastic modulus of the pavement structure. Where E is the elastic modulus, v is the Poisson's ratio, which is taken as 0.35, P is the impact load, a is the radius of the pressure plate, and δ is the measured deflection; S402. Extract the first arrival travel time of seismic image data and calculate the deep wave velocity profile Among them, Vi is the wave velocity of the i-th layer, Δhi is the thickness of the layer detected in the geological database, Δt i The travel time difference of seismic waves; S403. Fuse focused radar images, elastic modulus, and wave velocity profiles to construct a three-dimensional digital twin disease model.

[0012] Preferably, the digital twin pest model includes: Unified spatial coordinate system based on GPS timestamp; Multi-scale voxels are used to divide shallow voxel resolution and deep voxel resolution; Fusion of multi-source physical property data via radial basis function.

[0013] Preferably, the S5 specifically includes the following steps: S501, based on the voxel data of the digital twin model, calculate the risk entropy value of each voxel: (k = dielectric constant, elastic modulus, wave velocity) Among them, p k For the abnormal probability S502 of the kth physical property parameter, the risk entropy value and the location information are integrated to generate a drilling priority matrix; S503: Output the drilling path according to the priority matrix.

[0014] Preferably, the drilling path follows the principle of prioritizing traversal of high-risk areas and the minimum spacing between adjacent boreholes is ≥0.3m.

[0015] A rapid and automated comprehensive detection system for underground cavities in urban roads, comprising: Ground penetrating radar scanning module, used to obtain road B-scan images and output cavity probability maps through deep learning models; Automatic positioning impact module, used to drive the drop weight deflectometer to the suspected void point to perform impact loading and trigger the five-dimensional synchronization signal; A multi-sensor synchronization control module is used to simultaneously activate the seismic wave array to collect vibration wave data and switch the ground penetrating radar to a high-frequency focused scanning mode; A multi-source data fusion module is used to fuse elastic modulus derived from deflection inversion, seismic wave velocity, and focused radar images to construct a three-dimensional digital twin damage model; The risk decision module is used to calculate the risk entropy value based on the twin model and generate the drilling path planning.

[0016] The present invention provides a rapid, automated, and comprehensive detection method for underground cavities in urban roads. It has the following beneficial effects: 1. The present invention uses hardware-level coordinated control of impact loading and multi-source sensing, and a five-dimensional synchronous controller to trigger the switching of the seismic wave array and ground-penetrating radar working modes, and align them in time and space to eliminate equipment response time difference and mechanical vibration interference, thereby ensuring rapid and accurate detection of underground cavities in roads.

[0017] 2. Based on a unified spatial coordinate system, the present invention integrates three types of heterogeneous parameters: radar dielectric images, deflection inversion elastic modulus, and seismic wave velocity profiles. It applies radial basis function interpolation to reconstruct a three-dimensional physical property voxel model, and simultaneously characterizes dielectric anomalies, structural stiffness attenuation, and wave propagation characteristic variations, thereby achieving a multi-dimensional holographic expression of underground diseases.

[0018] 3. The present invention integrates the multi-parameter abnormal probability distribution in the twin model through the information entropy algorithm to generate a quantitative risk entropy value field, couples the spatial depth constraint to output the drilling priority matrix, and dynamically plans the drilling path by combining the high-risk area priority traversal and the minimum mechanical spacing principle to achieve dual closed-loop control of verifying target intelligent focusing and operational safety, eliminating blind spots in manual decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a rapid, automated, and comprehensive detection method for underground cavities in urban roads according to the present invention; Figure 2 This is an architectural diagram of a rapid, automated, and comprehensive detection system for underground cavities in urban roads according to the present invention. DETAILED DESCRIPTION

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

[0021] Please see the attached Figure 1 The embodiment of the present invention provides a rapid, automated, and comprehensive detection method for underground cavities in urban roads, comprising the following steps: S1. Use the vehicle-mounted ground-penetrating radar system to quickly scan the road and obtain B-scan images of the road's underground. Then use the deep learning segmentation model to process the radar images and output a cavity probability map to identify suspected cavity areas. S2. Automatically locate the suspected area, control the drop weight deflectometer to impact load the target point, and synchronously trigger the start signal from the five-dimensional synchronous controller; S3, a start signal is simultaneously transmitted to the seismic wave sensor array and the ground penetrating radar controller, causing the seismic wave array to collect vibration wave data generated by the impact load, while the ground penetrating radar switches to a high-frequency focused scanning mode; S4. Fusion processing of the pavement elastic modulus derived from FWD inversion, deep wave velocity data from seismic imaging, and focused radar images to construct a three-dimensional digital twin damage model; S5. Based on the spatial interpolation algorithm of the digital twin model, the risk entropy value of each voxel is calculated and the drilling path planning is output.

[0022] Specifically, S1 uses a vehicle-mounted ground-penetrating radar system to continuously scan along the longitudinal direction of the road, collecting B-scan images of the underground medium (the horizontal axis is the position sequence along the mileage, and the vertical axis is the two-way travel time). The B-scan images are then input into a deep learning segmentation model, which outputs the probability value of the existence of a cavity corresponding to each pixel point, generating a two-dimensional cavity probability map. Based on a preset probability threshold, the connected areas in the map are screened, and the geographic coordinate range of suspected cavity target areas is identified and marked, enabling a rapid survey of road structure and preliminary positioning of abnormal areas. After identifying a suspected cavity area, S2 drives the drop-weight deflectometer to the target coordinate point (based on the peak position of the cavity probability map) via a high-precision positioning slide. It then controls the hydraulic actuator to release the standard mass drop weight to set the impact force vertically on the road surface, and simultaneously triggers the five-dimensional synchronous controller, which generates a start signal with a precise timestamp. This enables subsequent sensing equipment such as the seismic wave array and ground-penetrating radar to start working strictly according to the unified time reference, achieving millisecond-level coordinated control of mechanical loading and multi-source data acquisition. S3 sends a start signal through the five-dimensional synchronous controller and distributes it to the seismic wave sensor array and the ground-penetrating radar controller via the synchronous transmission interface. After receiving the start signal, the seismic wave sensor array immediately collects vibration wave field data under the impact load. The ground-penetrating radar controller synchronously responds to the start signal and switches its working mode to the high-frequency focused scanning state to perform energy-focused detection on the target area. After acquiring multi-source data under impact loading, S4 reads the static elastic modulus inverted by the drop-weight deflectometer (calculated based on the displacement-load relationship of the bearing plate), the wave velocity profile of the layered medium collected by the seismic wave sensor array (interpreted by the travel time difference of the first arrival wave), and the dielectric constant distribution map generated by the high-frequency focused scanning of the ground-penetrating radar. Based on a unified spatial coordinate system, the three types of physical parameters are mapped to a three-dimensional grid space. The radial basis function interpolation algorithm is used to spatially reconstruct the discrete data points, and a three-dimensional voxel model of the diseased volume is constructed, which includes dielectric properties, mechanical modulus, and wave velocity attributes, realizing a full-dimensional digital twin representation of the underground structure. Based on the constructed three-dimensional digital twin disease model, S5 calculates the normalized deviation values ​​of the dielectric constant, elastic modulus, and wave velocity parameters for each voxel in the grid. It fuses the statistical distribution characteristics of these three types of parameters through the information entropy formula to generate an entropy scalar field that represents the degree of cavity risk. Based on the spatial gradient distribution of the entropy value, it uses path search to plan the optimal path sequence for core drilling and outputs a sequence of drilling rig travel instructions consisting of a set of spatial coordinate points. S1 specifically includes the following steps: S101, preprocessing the original B-scan image collected by the ground penetrating radar, including Hilbert transform to enhance the reflection interface, background noise filtering and time-to-depth conversion; S102: Input the pre-processed data into the deep learning segmentation model and perform the following operations: Encoder extraction, extracting multi-scale features through a 4-level downsampling structure, each level contains a convolutional layer and a maximum pooling layer; Atrous feature fusion: Atrous convolutions with dilation rates of 1, 3, and 6 are performed in parallel on the underlying features, and the output feature maps are concatenated along the channel dimension. Decoding and reconstruction, gradually upsampling through deconvolution layers and fusing with encoder features of the same scale; Probability output, perform Sigmoid activation on the final feature map to generate a pixel-level hole probability map; S103: Generate suspected void areas based on the probability map, perform morphological optimization on abnormal areas with an area less than 0.1 m2, and output a vector bounding box with geographic coordinates.

[0023] Specifically, S101 pre-processes the original B-scan images collected by the ground-penetrating radar: it extracts the reflected wave envelope through Hilbert transform, highlights the sudden change characteristics of the dielectric interface, filters to suppress system noise and environmental interference signals, performs time-to-depth conversion (based on the preset relative dielectric constant of the medium), maps the two-way travel time axis to the actual depth axis, and generates a depth profile image; S102 inputs the preprocessed depth profile into the segmentation model: the encoder path extracts multi-scale spatial features through a 4-level downsampling structure, and performs convolution operations and maximum pooling compression at each level in turn; atrous convolution operations with void rates of 1, 3, and 6 are performed in parallel on the underlying feature layer, and multi-receptive field feature maps are spliced ​​along the channel dimension; the decoder path gradually restores the resolution through the deconvolution layer, and fuses the encoder features of the same scale to enhance the boundary details; finally, the Sigmoid function is activated to output the pixel-level void probability value (in the range of 0 to 1), realizing multi-level interpretation of electromagnetic reflection features and outputting quantitative anomaly probability distribution.

[0024] S103 identifies suspected void areas based on probability maps. It extracts connected areas with probability values ​​exceeding a preset threshold as primary anomaly areas, applies morphological closing operations to optimize the outlines of discrete anomaly areas with an area less than 0.1 m2, eliminates fragmentation noise interference, and converts the optimized area boundaries into vector bounding boxes with geographic coordinates (EPSG:4326 coordinate system). It outputs the coordinates of the target area center point and the circumscribed rectangular range to provide precise spatial targets for subsequent mechanical verification.

[0025] S2 specifically includes the following steps: S201, based on the vector anomaly region bounding box, extract the coordinates of the peak point of the void probability (longitude λ, latitude φ) as the impact test target; S202, sending a positioning command to the drop weight deflectometer via the vehicle-mounted PLC controller to drive it to move directly above the target point, with a positioning error of ≤5 cm; S203, trigger the five-dimensional synchronous controller to execute: Start the free fall of the drop weight; Synchronously activate the Doppler laser velocimeter to collect road deformation; Synchronously turn on the dynamic strain gauge array; Simultaneously start a high-definition camera to capture cracks in the road surface; Synchronously record GPS timestamps.

[0026] Specifically, S201 calculates the coordinates of the maximum point (longitude λ, latitude φ) in the void probability distribution map based on the vector anomaly area bounding box data as the impact test target position, thereby locking the spatial coordinates of the abnormal energy concentration area and providing a positioning reference for subsequent impact loading.

[0027] S202 sends movement instructions to the drop hammer deflectometer through the on-board PLC controller, drives the positioning slide to move the impact device horizontally to just above the target, and controls the hydraulic actuator to adjust the height of the hammer to ensure that the position deviation between the impact center point and the target plane is ≤5cm, ensuring the precise spatial alignment of the impact equipment.

[0028] S203 triggers the five-dimensional synchronous controller to perform the following parallel operations: starting the free-fall mechanism of the drop hammer to release the impact load; synchronously activating the Doppler laser velocimeter to collect the dynamic deformation of the road surface; synchronously turning on the dynamic strain gauge array deployed on the road surface; synchronously starting the high-definition camera to capture the crack expansion morphology under the action of impact; and synchronously recording the UTC timestamp output by the GPS module to achieve precise synchronous acquisition of multi-source sensor data in the time domain.

[0029] S3 specifically includes the following steps: S301, sending a trigger instruction to the seismic wave sensor array through the GPS timestamp signal to start seismic wave data collection; S302, synchronously sending a mode switching instruction to the ground penetrating radar controller, including switching to a 1.6 GHz high-frequency focused scanning mode, focusing the detection range to scan the shallow surface layer, and increasing the scanning line spacing to centimeter-level accuracy; S303 , setting a device startup sequence based on the impact time t0 , including starting sampling by the seismic wave array before t0 and starting scanning by the ground penetrating radar after t0 .

[0030] Specifically, S301 responds to the GPS timestamp signal of the five-dimensional synchronous controller, sends a hardware trigger instruction to the seismic wave sensor array, and starts multi-channel shock wave data acquisition, thereby synchronously recording the time scale of the impact shock wave field.

[0031] S302 synchronously sends a mode switching instruction to the ground penetrating radar controller, adjusts its operating frequency to 1.6GHz, sets the scanning area to the shallow surface geological body, and encrypts the scanning line spacing to centimeter-level accuracy, switching the radar detection mode to high-resolution focused scanning.

[0032] S303 sets the equipment startup sequence based on the impact load occurrence time t0. The seismic wave array starts data sampling before t0 (to capture the pre-trigger background state), and the ground penetrating radar starts scanning after t0 (to avoid the mechanical vibration interference period), ensuring that the two types of equipment work together under the same time reference.

[0033] 1.6GHz high frequency focused scanning modes include: Electromagnetic beam focusing control, adjust the antenna array element phase offset Δφ to meet: Where: d is the array element spacing, λ is the wavelength of the electromagnetic wave in the medium, and θ is the beam pitch angle; Layer-fold scanning path planning, generating a spiral scanning path with the impact target as the center; Real-time inversion of dielectric constant, dynamic correction of dielectric constant based on reflected wave amplitude: Where: ε init is the initial dielectric constant, A max is the measured maximum reflected wave amplitude, A theory is the theoretical amplitude value, according to Calculation, k is the antenna gain coefficient, r is the electromagnetic wave propagation distance.

[0034] Specifically, the electromagnetic beam focusing control adjusts the phase offset of the antenna array element To meet the requirements of the formula, the array element spacing d, the wavelength λ in the medium, and the pitch angle θ are key control variables, thereby ensuring the spatial energy focusing of the electromagnetic beam in the shallow target area and enhancing the signal-to-noise ratio and spatial pointing accuracy of the detection signal; Spiral scanning path planning generates a spirally extended continuous scanning trajectory centered on the impact target, ensuring complete coverage of the target area and improving the spatial continuity of high-density detection.

[0035] The real-time inversion of the dielectric constant corrects the dielectric constant based on the proportional relationship between the maximum amplitude of the measured reflected wave and the theoretical amplitude value, where the theoretical amplitude is calculated according to the law of spherical diffusion of electromagnetic waves, thereby achieving in-situ dynamic calibration of the dielectric properties of the medium and eliminating the depth inversion error caused by dielectric parameter drift.

[0036] S4 specifically includes the following steps: S401. Analyze the deflection time history curve collected synchronously in five dimensions and invert the elastic modulus of the pavement structure. Where E is the elastic modulus, v is the Poisson's ratio, which is taken as 0.35, P is the impact load, a is the radius of the pressure plate, and δ is the measured deflection; S402. Extract the first arrival travel time of seismic image data and calculate the deep wave velocity profile Among them, Vi is the wave velocity of the i-th layer, Δhi is the thickness of the layer detected in the geological database, Δt i The travel time difference of seismic waves; S403. Fuse focused radar images, elastic modulus, and wave velocity profiles to construct a three-dimensional digital twin disease model.

[0037] Specifically, S401 analyzes the deflection time history curve data collected synchronously in five dimensions and uses the elastic half-space theory formula to Inverse the pavement elastic modulus to achieve quantitative characterization of pavement stiffness.

[0038] S402 Extracting the first arrival travel time difference Δt using seismic image data i , according to the wave speed calculation formula Layered interpretation of the medium velocity profile, where the thickness of the i-th layer is Δh i Derived from the standard sequence of geological database, Δt i The original waveform of the detector array is processed by the cross-correlation algorithm to obtain and output the elastic wave propagation characteristic parameters of the underground medium.

[0039] S403 integrates the radar dielectric image obtained by high-frequency focused scanning, the elastic modulus E inverted by S401, and the velocity profile V calculated by S402. i , a three-dimensional digital twin model was constructed under a unified spatiotemporal benchmark (EPSG:4979 coordinate system, GPS timestamp in S203) to complete the spatial physical property reconstruction of the diseased body.

[0040] The digital twin disease model includes: Unified spatial coordinate system based on GPS timestamp; Multi-scale voxels are used to divide shallow voxel resolution and deep voxel resolution; Fusion of multi-source physical property data via radial basis function.

[0041] Specifically, the spatial coordinate system of all sensor data is unified through GPS timestamps, the temporal and spatial benchmark deviation of multi-source information is eliminated, and multi-scale grid division of shallow high-resolution voxels and deep low-resolution voxels is adopted to adapt to the accuracy requirements of disease detection at different depths. The radial basis function is used to perform nonlinear spatial interpolation and fusion of the physical parameters of discrete collection points to construct a three-dimensional digital twin disease body model that fully expresses the dielectric properties, mechanical modulus and wave velocity distribution, ensure the geometric consistency matching of multi-source heterogeneous data, optimize the allocation of computing resources and the ability to resolve local anomalies, and realize the continuous reconstruction of void boundaries and physical property gradients.

[0042] S5 specifically includes the following steps: S501, based on the voxel data of the digital twin model, calculate the risk entropy value of each voxel: (k = dielectric constant, elastic modulus, wave velocity) Among them, p k For the abnormal probability S502 of the kth physical property parameter, the risk entropy value and the location information are integrated to generate a drilling priority matrix; S503: Output the drilling path according to the priority matrix.

[0043] Specifically, S501 calculates the abnormal probability value of three physical parameters (dielectric constant, elastic modulus, and wave velocity) for each voxel point based on the voxel data of the 3D digital twin model. It quantifies the degree of joint abnormality of multiple parameters using the information entropy formula and outputs a scalar field of risk entropy value, thus achieving spatial positioning and numerical representation of cavity disease risks. S502 fuses the risk entropy value and voxel depth information (obtained from the spatial coordinates of the twin model) to generate a priority matrix representing the urgency of drilling, thereby establishing a quantitative decision-making basis for risk level and spatial location; S503 outputs a drilling path coordinate sequence prioritized by high-risk areas through the spatial distribution of the priority matrix and combined with the minimum drilling spacing constraint, providing a spatial travel control instruction sequence for the drilling equipment.

[0044] The drilling path follows the principle of prioritizing traversal of high-risk areas and the minimum spacing between adjacent boreholes is ≥0.3m.

[0045] Specifically, the drilling path follows the principle of prioritizing traversal of high-risk areas, driving the drilling equipment to preferentially scan high-entropy abnormal voxels. At the same time, it enforces the mechanical constraint of a minimum spacing of ≥0.3m between adjacent boreholes to ensure the safety of drill rod operation and the completeness of path coverage, achieving zero-omission verification of high-risk diseased areas and zero-collision control during the drilling process.

[0046] Please see the attached Figure 2 , a rapid and automated comprehensive detection system for underground cavities in urban roads, including: Ground penetrating radar scanning module, used to obtain road B-scan images and output cavity probability maps through deep learning models; Automatic positioning impact module, used to drive the drop weight deflectometer to the suspected void point to perform impact loading and trigger the five-dimensional synchronization signal; A multi-sensor synchronization control module is used to simultaneously activate the seismic wave array to collect vibration wave data and switch the ground penetrating radar to a high-frequency focused scanning mode; A multi-source data fusion module is used to fuse elastic modulus derived from deflection inversion, seismic wave velocity, and focused radar images to construct a three-dimensional digital twin damage model; The risk decision module is used to calculate the risk entropy value based on the twin model and generate the drilling path planning.

[0047] Specifically, the detection system uses the ground-penetrating radar scanning module to realize road survey and cavity probability map generation, the automatic positioning impact module completes target precision strike and five-dimensional synchronous triggering, the multi-sensor synchronous control module ensures millisecond-level coordination of seismic wave acquisition and radar focus scanning, and the multi-source data fusion module constructs a three-dimensional twin model that integrates dielectric properties, mechanical modulus and wave velocity. The risk decision module outputs the drilling path based on the entropy algorithm, forming a closed-loop technology chain from remote sensing detection, mechanical verification, multi-dimensional modeling to decision execution, realizing the full-process intelligent detection of underground cavity survey-detailed inspection-verification.

[0048] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A rapid and automated comprehensive detection method for underground cavities in urban roads, characterized in that: The following steps are involved: S1. Use the vehicle-mounted ground-penetrating radar system to quickly scan the road and obtain B-scan images of the road's underground. Then use the deep learning segmentation model to process the radar images and output a cavity probability map to identify suspected cavity areas. S2. Automatically locate the suspected area, control the drop weight deflectometer to impact load the target point, and synchronously trigger the start signal from the five-dimensional synchronous controller; S3, a start signal is simultaneously transmitted to the seismic wave sensor array and the ground penetrating radar controller, causing the seismic wave array to collect vibration wave data generated by the impact load, while the ground penetrating radar switches to a high-frequency focused scanning mode; S4. Fusion processing of the pavement elastic modulus derived from FWD inversion, deep wave velocity data from seismic imaging, and focused radar images to construct a three-dimensional digital twin damage model; S5. Based on the spatial interpolation algorithm of the digital twin model, the risk entropy value of each voxel is calculated, and the drilling path planning is output.

2. A rapid and automated comprehensive detection method for underground cavities in urban roads according to claim 1, characterized in that: The S1 specifically includes the following steps: S101, preprocessing the original B-scan image collected by the ground penetrating radar, including Hilbert transform to enhance the reflection interface, background noise filtering and time-to-depth conversion; S102: Input the pre-processed data into the deep learning segmentation model and perform the following operations: Encoder extraction, extracting multi-scale features through a 4-level downsampling structure, each level contains a convolutional layer and a maximum pooling layer; Atrous feature fusion: Atrous convolutions with dilation rates of 1, 3, and 6 are performed in parallel on the underlying features, and the output feature maps are concatenated along the channel dimension. Decoding and reconstruction, gradually upsampling through deconvolution layers and fusing with encoder features of the same scale; Probability output, perform Sigmoid activation on the final feature map to generate a pixel-level hole probability map; S103: Generate suspected void areas based on the probability map, perform morphological optimization on abnormal areas with an area less than 0.1 m2, and output a vector bounding box with geographic coordinates.

3. The rapid automated comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The S2 specifically includes the following steps: S201, based on the vector anomaly region bounding box, extract the coordinates of the peak point of the void probability (longitude λ, latitude φ) as the impact test target; S202, sending a positioning command to the drop weight deflectometer via the vehicle-mounted PLC controller to drive it to move directly above the target point, with a positioning error of ≤5 cm; S203, trigger the five-dimensional synchronous controller to execute: Start the free fall of the drop weight; Synchronously activate the Doppler laser velocimeter to collect road deformation; Synchronously turn on the dynamic strain gauge array; Simultaneously start a high-definition camera to capture cracks in the road surface; Synchronously record GPS timestamps.

4. The rapid automated comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The S3 specifically includes the following steps: S301, sending a trigger instruction to the seismic wave sensor array through the GPS timestamp signal to start seismic wave data collection; S302, synchronously sending a mode switching instruction to the ground penetrating radar controller, including switching to a 1.6 GHz high-frequency focused scanning mode, focusing the detection range to scan the shallow surface layer, and increasing the scanning line spacing to centimeter-level accuracy; S303 , setting a device startup sequence based on the impact time t0 , including starting sampling by the seismic wave array before t0 and starting scanning by the ground penetrating radar after t0 .

5. The rapid and automated comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The 1.6 GHz high frequency focused scanning mode includes: Electromagnetic beam focusing control, adjust the antenna array element phase offset Δφ to meet: Where: d is the array element spacing, λ is the wavelength of the electromagnetic wave in the medium, and θ is the beam pitch angle; Layer-fold scanning path planning, generating a spiral scanning path with the impact target as the center; Real-time inversion of dielectric constant, dynamic correction of dielectric constant based on reflected wave amplitude: Where: ε init is the initial dielectric constant, A max is the measured maximum reflected wave amplitude, A theory is the theoretical amplitude value, according to Calculation, k is the antenna gain coefficient, r is the electromagnetic wave propagation distance.

6. The rapid and automated comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The S4 specifically includes the following steps: S401. Analyze the deflection time history curve collected synchronously in five dimensions and invert the elastic modulus of the pavement structure. Where E is the elastic modulus, v is the Poisson's ratio, which is taken as 0.35, P is the impact load, a is the radius of the pressure plate, and δ is the measured deflection; S402. Extract the first arrival travel time of seismic image data and calculate the deep wave velocity profile Among them, Vi is the wave velocity of the i-th layer, Δhi is the thickness of the layer detected in the geological database, Δt i The travel time difference of seismic waves; S403. Fuse focused radar images, elastic modulus, and wave velocity profiles to construct a three-dimensional digital twin disease model.

7. The rapid, automated, and comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The digital twin pest model includes: Unified spatial coordinate system based on GPS timestamp; Multi-scale voxels are used to divide shallow voxel resolution and deep voxel resolution; Fusion of multi-source physical property data via radial basis function.

8. The rapid, automated, and comprehensive detection method for underground cavities in urban roads according to claim 1 is characterized in that: The S5 specifically includes the following steps: S501, based on the voxel data of the digital twin model, calculate the risk entropy value of each voxel: Among them, p k is the abnormal probability of the kth physical property parameter S502 fuses the risk entropy value and the location information to generate a drilling priority matrix; S503: Output the drilling path according to the priority matrix.

9. A rapid and automated comprehensive detection method for underground cavities in urban roads according to claim 8, characterized in that: The drilling path follows the principle of prioritizing traversal of high-risk areas and the minimum spacing between adjacent boreholes is ≥0.3m.

10. A rapid and automated comprehensive detection system for underground cavities in urban roads, characterized by: A rapid, automated, and comprehensive detection method for underground cavities in urban roads as claimed in any one of claims 1 to 9, comprising: Ground penetrating radar scanning module, used to obtain road B-scan images and output cavity probability maps through deep learning models; Automatic positioning impact module, used to drive the drop weight deflectometer to the suspected void point to perform impact loading and trigger the five-dimensional synchronization signal; A multi-sensor synchronization control module is used to simultaneously activate the seismic wave array to collect vibration wave data and switch the ground penetrating radar to a high-frequency focused scanning mode; A multi-source data fusion module is used to fuse elastic modulus derived from deflection inversion, seismic wave velocity, and focused radar images to construct a three-dimensional digital twin damage model; The risk decision module is used to calculate the risk entropy value based on the twin model and generate the drilling path planning.

Images