Tunnel axial ground penetrating radar stratum global perception method

Abstract

This invention relates to the fields of tunnel engineering and geophysical exploration technology, specifically disclosing a method for full-domain strata perception using tunnel axial ground-penetrating radar (GPR). The method includes steps S101: establishing a geological model and performing radar forward modeling; S102: constructing a GPR image sample set and training anomaly recognition model; S103: performing energy feature analysis and establishing classification criteria for geological anomalies; S104: borehole GPR axial detection and data acquisition; and S105: spatial registration and geological anomaly early warning. This method comprehensively applies electromagnetic wave propagation theory, radar forward modeling, deep learning recognition algorithms, and RMS amplitude energy analysis technology. By drilling horizontal detection holes and deploying GPR along the tunnel axis, it achieves high-precision identification, classification, and risk visualization of the strata and geological anomalies (including boulders, karst caves, and debris) ahead of the tunnel, providing scientific geological safety decision support for tunnel excavation.

Description

Technical Field

[0001] This invention relates to the fields of tunnel engineering and geophysical exploration technology, specifically to a method for axial strata perception of tunnels based on borehole ground-penetrating radar. Background Technology

[0002] With the continuous development of underground space, tunnel construction is gradually extending to areas with great depth, long distance, and complex geological conditions. During tunnel excavation, unfavorable geological features such as isolated boulders, karst caves, and fractured zones hidden in the strata ahead are the main sources of risk for major safety accidents such as collapses and water inrushes. Therefore, achieving high-precision, long-distance, and advanced perception and risk warning of the geological conditions ahead of the tunnel has become a key technical link in ensuring construction safety and controlling engineering risks.

[0003] Currently, commonly used methods for advanced geological exploration of tunnels mainly include drilling and sampling, seismic wave methods, and ground penetrating radar (GPR). Among them, drilling has a long construction cycle, high cost, low efficiency, and causes structural damage to the surrounding rock, making it difficult to meet the needs of rapid and economical construction. Although seismic wave methods are effective for detecting large-scale stratigraphic interfaces, their resolution is low, making it difficult to accurately identify adverse geological bodies such as isolated boulders and small caves. Furthermore, the interpretation results are ambiguous, and their reliability is insufficient under complex geological conditions. Electromagnetic wave-based ground penetrating radar also faces severe challenges in practical applications. When conducting detection from the ground, the long propagation path of the radar signal, severe attenuation, and susceptibility to interference often lead to a significant loss of detection accuracy. When conducting detection at the tunnel face, the radar signal is also affected by the metal components of the tunnel construction and the excavation disturbance zone, resulting in signal clutter, rapid attenuation, and a short effective detection distance, failing to meet the requirements for long-distance, high-precision prediction.

[0004] To overcome the aforementioned bottlenecks and improve detection accuracy and long-range identification capabilities, borehole ground-penetrating radar (BPR) technology is increasingly being applied to tunnel forward detection. This technology involves drilling horizontal holes along the tunnel's axis, inserting the BPR into the borehole, and advancing it along the tunnel's axis to achieve long-range detection of the geological structure ahead of the tunnel.

[0005] However, existing borehole radar methods still have the following shortcomings: anomaly identification relies on experience, resulting in insufficient accuracy. Traditional radar images are susceptible to noise and clutter interference, and different types of anomalies (isolated rocks, caves, debris) show only slight differences in imaging, lacking intelligent identification methods. Currently, manual interpretation is mostly used, leading to low identification efficiency and strong subjectivity in the results; energy analysis and anomaly classification are incomplete. Existing research focuses more on morphological identification, lacking quantitative energy analysis; spatial registration and risk classification mechanisms are also lacking; detection results are mostly presented in two-dimensional profile form, making it difficult to intuitively reflect the distribution and influence range of anomalies in space.

[0006] Therefore, there is an urgent need to develop a method for full-domain axial geological sensing of tunnels, which can achieve high-precision identification, attribute differentiation, and spatial early warning of anomalies ahead of the tunnel under complex geological conditions, thereby providing scientific geological safety assurance for tunnel excavation. Summary of the Invention

[0007] The present invention aims to overcome at least one of the shortcomings of the prior art and provide a method for full-domain stratum perception of tunnel axial ground-penetrating radar, so as to realize intelligent identification, accurate positioning, type discrimination and graded early warning of geological anomalies in front of the tunnel, and solve the problems of low identification accuracy, poor anomaly type discrimination ability and lack of spatial early warning in the existing tunnel front detection.

[0008] The technical solution adopted in this invention is to provide a method for full-domain stratum perception using tunnel axial ground-penetrating radar, comprising the following steps:

[0009] S101. Based on the tunnel design cross-section and geological exploration data, establish a geological model including normal strata and geological anomalies and perform radar forward modeling to obtain a signal dataset.

[0010] S102. The signal dataset is preprocessed and converted into image results. The image results are compared with the geological exploration data and then manually labeled. A training sample set is constructed and the model is trained to obtain a recognition model that can identify the location of geological anomalies. The recognition model outputs the coordinates of the anomaly area and the corresponding radar reflection wave amplitude.

[0011] S103. The identification model performs energy feature analysis based on the coordinates of the anomalous body region and the amplitude of radar reflected waves to obtain a classification criterion for geological anomalies based on energy distribution characteristics.

[0012] S104. Axial detection using borehole ground-penetrating radar to obtain measured data;

[0013] S105. The identification model performs spatial registration based on the measured data obtained in step S104, and generates a risk distribution map according to the early warning classification standard.

[0014] The geological anomalies include boulders, caves, and fragmented bodies.

[0015] The above scheme comprehensively applies multiple technical means, and through radar forward modeling and deep learning model training, it realizes the intelligent identification of geological anomalies in the strata in front of the tunnel by ground-penetrating radar; through energy feature analysis, it automatically classifies the discovered geological anomalies; through spatial registration, it achieves accurate positioning, and finally builds a risk early warning mechanism, which solves the problems of low identification accuracy, poor anomaly type discrimination ability and lack of spatial early warning in the existing tunnel front detection.

[0016] Furthermore, the radar forward modeling simulation described in step S101 uses gprMax numerical simulation software to solve the propagation process of electromagnetic waves in the geological model through the finite difference time-domain method.

[0017] Furthermore, the preprocessing in step S102 includes time zero-point correction, background removal, bandpass filtering, and envelope equalization algorithms; the resulting image is a radar profile in the form of a standardized grayscale image.

[0018] Furthermore, the tool for model training in step S102 is the YOLOv8 deep learning network, and the loss function for model training includes CIoU localization loss and multi-class cross-entropy classification loss.

[0019] Further, the energy characteristic analysis in step S103 includes the following: setting a vertical sampling window containing multiple sampling points above the coordinates of the anomalous body region obtained in step S102, with the window height being 1 to 1.5 times the maximum diameter of the geological anomaly; calculating the RMS amplitude energy based on the radar reflection wave amplitude of the sampling points, and comparing the RMS amplitude energy of the anomalous body region with that of adjacent normal strata to obtain energy distribution characteristics and establish classification criteria for geological anomalies.

[0020] Furthermore, the classification criteria include energy intensity, energy ratio, and energy fluctuation characteristics; areas with high energy concentration are classified as isolated rock anomalies, areas with significantly weakened or missing energy are classified as cave anomalies, and areas with obvious energy fluctuations are classified as fragmented body anomalies.

[0021] Further, step S104 includes the following: drilling a horizontal detection hole along the tunnel axis; placing the borehole ground-penetrating radar probe into the borehole; advancing the radar probe segment by segment along the borehole axis at a set step distance, transmitting and receiving electromagnetic wave signals once at each step distance position, and recording and correcting the echo data in real time; processing the data after acquisition to form measured data. Existing technologies drill multiple holes forward on the tunnel face to house the radar transmitting and receiving devices respectively; however, this solution only requires drilling one horizontal detection hole along the tunnel axis to place the borehole ground-penetrating radar probe, saving time and effort. The probe integrates a transmitting antenna, a receiving antenna, and a main control acquisition unit; the transmitting and receiving antennas are connected to the control terminal via optical fiber or coaxial cable.

[0022] Furthermore, the step distance is 0.05–0.1 m. To facilitate detection ahead of the tunnel, adaptive adjustments need to be made according to different types of strata and geological anomalies, including adjusting the step distance. Smaller step distances are beneficial for improving detection accuracy, while larger step distances are beneficial for saving detection time.

[0023] Further, step S105 includes the following: based on the measured data obtained in step S104, establish a one-to-one correspondence between radar sampling points and the tunnel centerline; when a geological anomaly is identified, use the coordinates of the center point of the geological anomaly as a reference to divide warning zones of different risk levels along the front and rear directions of the tunnel, and render and generate a risk distribution map; the system automatically outputs the geological anomaly number, spatial location, equivalent size and corresponding warning level, and generates a geological risk assessment and safety warning report in front of the tunnel.

[0024] The present invention also provides a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the tunnel axial ground-penetrating radar method for full-domain stratum perception.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1) The present invention provides a method for full-domain stratum perception of tunnel axial ground-penetrating radar, which obtains standard radar response characteristics of different geological anomalies through electromagnetic wave numerical forward modeling, and establishes the theoretical basis for anomaly imaging from a physical level.

[0027] 2) A deep learning model was constructed to realize the automatic identification, classification and localization of anomalies in ground-penetrating radar images, overcoming the subjectivity and inefficiency of traditional manual interpretation, and significantly improving the intelligent processing level of tunnel ground-penetrating radar data.

[0028] 3) By combining deep learning detection results with RMS amplitude energy feature calculation, quantitative identification and classification of geological anomalies were achieved, solving the problem that isolated rocks, caves and debris images are similar and difficult to distinguish in traditional radar imaging.

[0029] 4) The detection process of drilling multiple holes in front of the tunnel face is simplified to drilling holes along the tunnel axis, which saves construction time and realizes long-distance non-destructive detection of the strata in front. The collected reflection data has high accuracy and wide coverage.

[0030] 5) Through spatial registration and early warning zone division, a linear hierarchical early warning mechanism based on the influence range of anomalies along the tunnel direction is realized. It is intuitive and quantifiable, providing a visualized basis for geological risk assessment for safe tunneling. Attached Figure Description

[0031] Figure 1 The flowchart illustrates a method for full-domain strata perception using tunnel axial ground-penetrating radar, as provided by this invention.

[0032] Figure 2 The diagram illustrates geological body modeling, radar forward modeling, and image segmentation. a represents geological body modeling, b represents radar forward modeling, and c represents image segmentation.

[0033] Figure 3 This is a schematic diagram of YOLOv8. SPPF: Hollow pyramid pooling layer parameters. C2f: Block parameters. Conv: Convolutional layer parameters.

[0034] Figure 4 A schematic diagram for identifying and locating abnormal points.

[0035] Figure 5 This is a schematic diagram illustrating the energy characteristic analysis and classification of anomalous bodies.

[0036] Figure 6 This is a schematic diagram of borehole ground-penetrating radar axial detection and data acquisition.

[0037] Figure 7 The diagram shows the spatial registration and three-level early warning of geological anomalies along the tunnel axis. a is the spatial registration diagram, and b is the three-level early warning diagram. Detailed Implementation

[0038] 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.

[0039] This embodiment provides a method for full-domain strata perception using tunnel axial ground-penetrating radar, the process of which is as follows: Figure 1 As shown, it includes the following steps:

[0040] Step S101: Establish a geological model and perform radar forward modeling. Based on the tunnel design cross-section and geological exploration data, construct a two-dimensional or three-dimensional model including typical geological bodies such as surrounding rock layers, boulders, karst caves, and debris. Use the electromagnetic wave propagation equation to perform forward modeling with ground-penetrating radar to obtain the reflection characteristics of different media and anomalies, providing standardized data samples for subsequent identification model training. Figure 2 As shown in the figure, different colors or grayscale changes represent the reflection intensity and phase changes of electromagnetic waves at different medium interfaces.

[0041] Step S102: Construct a ground-penetrating radar image sample set and train anomaly recognition model. Based on forward modeling and measured data, radar images are preprocessed, enhanced, and labeled to form a training sample set containing multiple types of anomalies. A YOLOv8 deep learning network is used for model training. Through feature extraction, convolutional fusion, and detection head output, automatic identification and localization of geological anomalies such as isolated boulders, caves, and debris are achieved. Figure 3 , 4As shown, the YOLOv8 deep learning network model automatically detects and identifies the locations of geological anomalies such as isolated rocks, caves, and loose bodies in the image. The identification results are displayed as rectangular boxes or annotation symbols, indicating the type of anomaly and the confidence level, thus realizing intelligent identification and spatial positioning of geological anomalies. Figures 2-4 This intuitively demonstrates the automation and accuracy advantages of the present invention in the forward modeling and identification stages, providing an accurate input data foundation for subsequent energy analysis and space early warning.

[0042] Step S103: Perform energy characteristic analysis and type identification on the anomalous body region. For example... Figure 5 As shown, within the identified anomalous body region, the RMS root mean square energy of the radar reflected wave is calculated and normalized. Based on parameters such as energy ratio and energy variance, three anomalous types are distinguished: isolated boulders (high-energy concentrated reflection), karst caves (low energy or energy deficiency), and fragmented bodies (energy fluctuation type), thus realizing the identification of the physical properties of geological bodies. Figure 5 This invention can intuitively illustrate the principle and effect of using energy characteristic analysis to classify geological anomalies, providing a reliable physical quantitative basis for subsequent spatial early warning classification.

[0043] Step S104: Axial detection and data acquisition using borehole ground-penetrating radar. (For example...) Figure 6 As shown, horizontal detection holes are drilled along the tunnel axis, and a borehole ground-penetrating radar probe is placed inside the hole. The radar probe advances along the borehole at fixed steps, transmitting and receiving electromagnetic wave signals segment by segment, and collecting time-domain waveforms and B-scan image data to provide basic data for subsequent spatial registration and early warning.

[0044] Step S105: Spatial Registration and Geological Anomaly Early Warning. The location ahead of each sampling point in the tunnel is determined based on the drilling ground-penetrating radar's advance distance. If a geological anomaly exists at a certain location, a graded early warning is issued, extending three times the diameter along the tunnel axis before and after the center of the anomaly as the base point. For example... Figure 7 As shown, the anomaly itself is the first-level warning zone (red, high risk), the area within three times its diameter before and after it is the second-level warning zone (yellow, medium risk), and the area beyond three times its diameter is the third-level warning zone (green, low risk). The linear risk distribution is displayed through pseudo-color visualization, achieving a visual effect for early warning. This demonstrates the practicality and intuitiveness of this method in achieving quantitative early warning based on the results of geological anomaly identification.

[0045] In step S101, the specific operations are as follows:

[0046] S11: Establish a model of the strata and geological anomalies ahead of the tunnel. Based on the tunnel geological exploration data, determine the electromagnetic parameters of each stratum, including the dielectric constant. magnetic permeability Electrical conductivity Establish a two-dimensional or three-dimensional geological model containing typical anomalies such as isolated boulders, caves, and loose materials. Model parameters are set as follows: soil dielectric constant... Dielectric constant of rock strata The dielectric constant of a monolith Void dielectric constant (Air), dielectric constant of the mixture of fragments (Depending on the water-to-soil ratio), electrical conductivity Pick magnetic permeability The mesh cell division satisfies ,in The wavelength is used to ensure numerical stability and calculation accuracy.

[0047] S12: Electromagnetic field forward modeling based on Maxwell's equations. The propagation process of electromagnetic waves in the model is solved using the gprMax numerical simulation software and the finite-difference time-domain (FDTD) method. The propagation of electromagnetic waves in an isotropic medium is described by Maxwell's equations: , .in The electric field intensity vector, The magnetic field strength vector. , , These represent the dielectric constant, magnetic permeability, and electrical conductivity, respectively. Time step Determined by the Courant condition: .in It is the speed of light. Normally... Pick To ensure numerical stability, the radar source uses Ricker wavelets as the transmitted waveform, with the function expression as follows: The center frequency The corresponding detection depth is approximately 10 m.

[0048] S13: Simulation Result Analysis and Feature Extraction. After forward modeling, radar response features of various types of geological anomalies are extracted to provide a sample basis for subsequent deep learning identification and energy analysis. gprMax outputs A-scan signals showing the electric field intensity changing over time and B-scan images along the survey line. Simulation results show that boulders exhibit a downward-opening hyperbolic reflection characteristic on the radar profile, cavities show signal weakening or disappearance zones, and fragmented areas generate high-frequency disturbances and irregular reflection bands. Preprocessing operations such as time zero-point correction, background removal, and gain compensation are performed on the output data to enhance signal quality. Parameters such as reflected wave envelope, peak amplitude, and arrival time are extracted to form a standardized dataset, providing input samples for subsequent model training.

[0049] The core of this step is to obtain the standard radar response characteristics of different geological anomalies through electromagnetic wave numerical forward modeling, establish the theoretical basis for anomaly imaging from a physical perspective, and provide high-quality data support for the deep learning recognition model training in the subsequent step S102.

[0050] In step S102, the specific operations are as follows:

[0051] S21: Data Preprocessing and Image Sample Generation. First, the forward modeling results from the ground-penetrating radar and the measured B-scan images are normalized and noise suppressed. Time zero-point correction, background removal, bandpass filtering (Butterworth filtering), and envelope equalization algorithms are employed to enhance the target signal characteristics. Let the original radar signal be... Its filtered signal It can be represented as: .in For Fourier transform operators, The filter transfer function. The signal amplitude envelope is obtained through envelope detection. : .in for The Hilbert transform is applied. Finally, each radar profile is uniformly converted into a standardized grayscale image of size 640×640, which is then used as input to the deep learning model.

[0052] S22: Manual Annotation and Sample Set Construction. Based on the results of forward modeling and on-site geological comparison, anomalous areas in the images were manually annotated. The annotation categories included four types: isolated boulders, caves, debris, and normal strata. Annotations used rectangular bounding boxes, and the label data was saved in Pascal VOC or YOLO format, with the following structure: All coordinate parameters have been normalized, so that The final result is no less than 1,000 sample images, in which the ratio of positive samples (including anomalous objects) to negative samples (without anomalous objects) is approximately 3:1, which is used to enhance the classification ability of the model.

[0053] S23: Training of a YOLOv8-based geological anomaly recognition model. The YOLOv8 (You Only Look Once Version 8) object detection network was used for training. The YOLOv8 network consists of a backbone feature extraction module, a neck feature fusion module, and a head. The input image size is 640×640×3, and the network extracts image information at different resolutions through a multi-scale feature pyramid.

[0054] YOLOv8's detection output is ,in Indicates the confidence level. Let be the class probability. The training loss function consists of three parts: localization loss, confidence loss, and classification loss. The overall loss function is defined as: .in, Using CIoU (Complete IoU) loss, This is a multi-class cross-entropy loss function used for class recognition. Training employs the Adam optimizer with an initial learning rate of 0.001, a batch size of 16, and 100–300 training epochs. The learning rate is dynamically adjusted based on the validation set accuracy. After training, the optimal model parameter weight file is output.

[0055] S24: Model Validation and Field Data Testing. The model is validated using ground-penetrating radar data from unused tunnel boreholes. The model's predictions are compared with human interpretations, and the average accuracy is calculated. (and detection accuracy indicators) The calculation formula is: .in, For the first The average accuracy of the target class, The number of target categories.

[0056] The deep learning model constructed in this step enables the automatic identification, classification, and localization of anomalies in ground-penetrating radar images, laying the foundation for subsequent energy feature analysis and spatial registration. This method overcomes the subjectivity and inefficiency of traditional manual interpretation, significantly improving the intelligent processing level of tunnel ground-penetrating radar data.

[0057] In step S103, the specific operations are as follows:

[0058] S31: Extract the radar reflection signal above the anomalous object. Using the bounding box coordinates of the anomalous object detected in S102, determine its spatial range in the radar profile. A vertical sampling window is preset above each anomalous object, with a height generally 1 to 1.5 times the maximum diameter of the anomalous object to ensure coverage of the main reflected wave energy. Assume this window contains... There are 10 sampling points, and their corresponding time series are: The amplitude of the reflected signal is The signal is extracted and subjected to denoising and time window normalization to obtain an effective reflected wave sequence. .

[0059] S32: Calculate the RMS amplitude energy. The RMS (Root Mean Square) amplitude is a quantitative indicator characterizing the strength of radar reflected signal energy. Its calculation formula is: .in, For the reflected wave at the 1st The amplitude of each sampling point This represents the total number of sampling points. To eliminate gain differences between different radar survey lines, the RMS value is normalized: .in, and These are the maximum and minimum RMS values ​​within the same survey line. This normalized RMS value directly reflects the relative strength of the reflected energy of the geological body.

[0060] S33: Anomaly Type Identification Based on Energy Characteristics. Based on forward modeling and experimental results, energy characteristic classification criteria are established. The statistical patterns of reflected energy in boulders, caves, and debris are as follows: the reflected wave energy is highest above boulders, exhibiting a high-amplitude concentrated peak; the reflected energy is lower above caves due to a sudden drop in dielectric constant; the reflected wave energy in debris areas fluctuates continuously over time. Based on these characteristics, the energy ratio method can be used for classification. Let the RMS energy of the target area be... The RMS energies of its upper and lower neighborhoods are respectively , Then the energy ratio Defined as: .when It was determined to be an isolated rock anomaly; when When it is determined to be a void anomaly; when Furthermore, a large energy variance indicates a fragmented anomaly. The formula for calculating energy variance is: .in This represents the average amplitude of the reflected wave within the window.

[0061] This step enables quantitative identification and classification of geological anomalies based on RMS amplitude energy characteristics, solving the problem of difficulty in distinguishing isolated boulders, caves, and debris in traditional radar imaging due to their similar image features. This method, combining deep learning detection results with energy feature calculations, significantly improves the accuracy and reliability of geological anomaly identification.

[0062] In step S104, the specific operations are as follows:

[0063] S41: Drilling Layout and Radar Equipment Installation. Horizontal detection holes are drilled along the tunnel axis. The hole length is determined based on detection requirements, and the hole diameter is determined based on the size of the ground-penetrating radar probe. A borehole ground-penetrating radar system is used, including a transmitting antenna, a receiving antenna, and a main control acquisition unit. The antenna frequency is typically selected at 200 MHz to balance detection depth and resolution. The transmitting and receiving antennas are connected to the main control unit via fiber optic or coaxial cables, and the system is synchronized.

[0064] S42: Axial Advancement and Data Acquisition. During data acquisition, the radar probe advances segment by segment along the borehole axis at a set step distance. A common step distance is 0.05–0.1 m to ensure spatial sampling accuracy. Each advance completes one transmission and reception cycle, recording the reflected electromagnetic wave signal. The sampling frequency is typically set to 2–4 GHz, and the sampling time window is 100–300 ns. The acquired data includes the time-domain reflection waveform and its amplitude information. The radar signal is recorded as A-scan data, which is then stitched together to form a continuous B-scan profile image.

[0065] S43: Preliminary Data Processing and Storage. After acquisition, the raw waveform undergoes time zero-point correction, background removal, and gain compensation. Time-depth conversion is performed using standard radar signal processing formulas. .in, To detect depth, The speed of electromagnetic wave propagation in the medium. This is the two-way propagation time. Finally, the data for each survey line is stored as a standard SEG-Y or gpr format file, providing foundational data for subsequent energy analysis, spatial registration, and geological risk assessment.

[0066] Through this step, the borehole ground-penetrating radar can achieve long-distance non-destructive detection of the strata ahead along the tunnel axis. The collected reflection data has high accuracy and wide coverage, providing reliable raw information support for the whole-domain perception of the strata.

[0067] In step S105, the specific operations are as follows:

[0068] S51: Spatial registration based on advance distance. The borehole ground-penetrating radar detection path is consistent with the tunnel centerline. Let the detection starting point be point A, and define the starting point coordinates. During radar propulsion, the system records the propulsion distance in real time. Then the first The distance ahead of the tunnel corresponding to this sampling is: .in, This is the radar's advance distance for each step (typically 0.05–0.1 m). By measuring the advance distance, the precise position of each data point on the tunnel axis can be determined, achieving a one-to-one correspondence between the data and the spatial position within the tunnel.

[0069] S52: Linear early warning classification based on anomalies along the tunnel direction. When a geological anomaly is identified at point B on the tunnel axis, a three-level early warning zone is divided along the tunnel direction (front and back) using the center point of the anomaly as a reference: The first-level early warning zone (red, high-risk zone) is the area where the geological anomaly is located, i.e., at the center of point B. Within the range ( The area with the anomaly's equivalent diameter directly corresponds to unfavorable geological features such as boulders, cavities, and fracture zones, posing an extremely high construction risk. The secondary warning zone (yellow, medium-risk zone) is the area within three times the diameter before and after the anomaly's center point. This area may be affected by stress disturbances from anomalies or geological structures, and protection and monitoring should be strengthened; the Level III warning zone (green, low-risk zone) is the area exceeding three times the diameter before and after the anomaly, i.e. The geological strata in this area are stable, allowing for normal tunneling operations.

[0070] The mathematical expression for the warning level is as follows:

[0071]

[0072] S53: Early Warning Visualization and Result Output. Based on the above-mentioned risk classification results, a risk distribution map is generated along the tunnel axis. The image uses distance as the horizontal axis. Vertically, the warning level or energy intensity is indicated. Pseudo-color rendering is used: red represents a Level 1 high-risk area, yellow represents a Level 2 medium-risk area, and green represents a Level 3 safe area. The system automatically outputs a "Tunnel Ahead Stratum Full-Area Perception and Early Warning Report," including anomaly number, center location, and equivalent diameter. Warning levels for each section And construction recommendations and measures.

[0073] This step establishes a linear, hierarchical early warning mechanism based on the influence range of anomalies along the tunnel direction, enabling continuous expression of geological anomaly risks along the tunnel axis. This method is intuitive, quantifiable, and easy to implement in engineering, providing a visualized basis for geological risk assessment for tunneling safety.

[0074] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific implementation of the present invention.

Claims

1. A method for full-domain stratigraphic perception using tunnel axial ground-penetrating radar, characterized in that, Includes the following steps: S101. Based on the tunnel design cross-section and geological exploration data, establish a geological model including normal strata and geological anomalies and perform radar forward modeling to obtain a signal dataset. S102. The signal dataset is preprocessed and converted into image results. The image results are compared with the geological exploration data and then manually labeled. A training sample set is constructed and the model is trained to obtain a recognition model that can identify the location of geological anomalies. The recognition model outputs the coordinates of the anomaly area and the corresponding radar reflection wave amplitude. S103. The identification model performs energy feature analysis based on the coordinates of the anomalous body region and the amplitude of radar reflected waves to obtain a classification criterion for geological anomalies based on energy distribution characteristics. S104. Axial detection using borehole ground-penetrating radar to obtain measured data; S105. The identification model performs spatial registration based on the measured data obtained in step S104, and generates a risk distribution map according to the early warning classification standard. The geological anomalies include isolated boulders, caves, and rubble; The energy characteristic analysis in step S103 includes the following: A vertical sampling window containing multiple sampling points is set above the coordinates of the anomaly region obtained in step S102, with the window height being 1 to 1.5 times the maximum diameter of the geological anomaly; the RMS amplitude energy is calculated and normalized based on the radar reflection amplitude values ​​of the sampling points, and the RMS amplitude energy of the anomaly region is compared with that of adjacent normal strata to obtain energy distribution characteristics, establishing classification criteria for geological anomalies; the classification criteria include energy ratio and energy variance, assuming the RMS energy of the target region is... The RMS energies of its upper and lower neighborhoods are respectively , Then the energy ratio Defined as: The formula for calculating energy variance is: ;in The average amplitude of the reflected wave within the window. For the reflected wave at the 1st The amplitude of each sampling point The total number of sampling points; when It was determined to be an isolated rock anomaly; when When it is determined to be a void anomaly; when Furthermore, when the energy variance is large, it is classified as a fragmented anomaly; Step S105 includes the following: Based on the measured data obtained in step S104, establish a one-to-one correspondence between radar sampling points and the tunnel centerline; when a geological anomaly is identified, use the coordinates of the center point of the geological anomaly as a reference to divide warning zones of different risk levels along the front and rear directions of the tunnel, and render and generate a risk distribution map; the system automatically outputs the geological anomaly number, spatial location, equivalent size and corresponding warning level, and generates a geological risk assessment and safety warning report in front of the tunnel.

2. The method for full-domain strata perception using tunnel axial ground-penetrating radar according to claim 1, characterized in that, The radar forward modeling in step S101 uses gprMax numerical simulation software to solve the propagation process of electromagnetic waves in the geological model using the finite difference time-domain method.

3. The method for full-domain strata perception using tunnel axial ground-penetrating radar according to claim 1, characterized in that, The preprocessing described in step S102 includes time zero-point correction, background removal, bandpass filtering, and envelope equalization algorithms; the resulting image is a radar profile in the form of a standardized grayscale image.

4. The method for full-domain strata perception using tunnel axial ground-penetrating radar according to claim 1, characterized in that, The tool for model training in step S102 is the YOLOv8 deep learning network, and the loss function for model training includes CIoU localization loss and multi-class cross-entropy classification loss.

5. The method for full-domain strata perception using tunnel axial ground-penetrating radar according to claim 1, characterized in that, Step S104 includes the following: drilling horizontal detection holes along the tunnel axis; placing the borehole ground-penetrating radar probe into the borehole; advancing the radar probe along the borehole axis in segments at a set step distance, transmitting and receiving electromagnetic wave signals once at each step distance position, and recording and correcting the echo data in real time; and processing the data after the acquisition is completed to form measured data.

6. The method for full-domain strata perception using tunnel axial ground-penetrating radar according to claim 5, characterized in that, The set step distance is 0.05 to 0.1 m.

7. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the stratum-wide perception method of tunnel axial geological radar as described in any one of claims 1 to 6.

Images