Intelligent identification method and system for shield wall back grouting defects based on image processing

Abstract

The application discloses a shield wall post-grouting defect intelligent identification method and system based on image processing, and relates to the field of image recognition.The application realizes accurate detection and repair of grouting defects by establishing a three-layer medium model and combining physical tests and radar signal images.Firstly, the physical accuracy of the model is ensured by setting real electromagnetic parameters and boundary conditions.After the preset defect type, the generated radar image data is manually labeled to construct a defect type and parameter mapping model.After the on-site radar image is preprocessed, the model is inputted to detect the defect type and parameter in real time, record the repair time consumption and compare the threshold value, automatically judge whether the repair is completed or needs further adjustment, and finally construct a time consumption overrun ratio mapping model through multiple parameter optimization and mapping to realize efficient defect repair, thereby significantly improving the detection accuracy and repair efficiency, reducing manual intervention, reducing the repair time cost, and being suitable for complex grouting engineering.

Description

Technical Field

[0001] This invention belongs to the field of image recognition, and more specifically, relates to a method and system for intelligent identification of grouting defects behind shield tunnel walls based on image processing. Background Technology

[0002] Current methods for detecting and repairing grouting defects rely on manual inspection, which is inefficient and has limited accuracy. The combination of physical models and radar images is insufficient, resulting in inaccurate detection results. Furthermore, the lack of a real-time monitoring mechanism for repair time makes it impossible to effectively assess the repair progress. In addition, the process of parameter optimization and mapping model construction is cumbersome, time-consuming, and has unstable results. Summary of the Invention

[0003] To address the problems in related technologies, this invention proposes an intelligent identification method and system for grouting defects behind shield tunnel walls based on image processing, in order to overcome the aforementioned technical problems existing in existing related technologies.

[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: The present invention is an intelligent identification method for grouting defects behind shield tunnel walls based on image processing, comprising the following steps: S1. Establish a three-layer medium model for grouting behind the shield wall and set the actual electromagnetic parameters, boundary conditions and preset defect types; then construct a physical test model. S2. The radar signal image is obtained and preprocessed based on the three-layer medium model and physical test model constructed based on S1; S3. Preset grouting defect types and corresponding parameters. Use the three-layer medium model and physical test model constructed by S1 to generate grouting radar image data. After manually annotating the generated data with features, construct the final grouting defect type and parameter mapping model. S4. Preprocess the on-site radar image and input it into the mapping model in S3 to map the defect type and parameters; record the repair time and compare it with the preset threshold, and perform repeated detection to count the timeout probability; if the timeout probability is less than the threshold, the repair is complete; otherwise, proceed to S5. S5. By repeatedly adjusting the multiple operation parameters preprocessed in S2 and the number of mappings in S4, a grouting defect repair experiment is conducted. During the experiment, the data on the repair time exceeding the limit is recorded, and a final defect repair time exceeding the limit mapping model is constructed based on the recorded data. S6. Repeatedly adjust the preprocessing parameters and mapping times corresponding to the repeated detection process in S4 and input them into the mapping model in S5. Feedback is given on the adjustment effect based on the mapping results until the probability of time exceeding the limit is lower than the threshold.

[0005] Preferably, step S1 includes the following steps: S11. A three-layer medium model including tunnel segments, backfill grout, and surrounding rock and soil is established using the finite-difference time-domain method. Electromagnetic parameters and corresponding boundary conditions that conform to the actual physical properties of each medium layer are set, and simulated defect types are preset in the backfill grout, thus obtaining a shield tunnel segment-grout layer numerical model. The electromagnetic parameters include dielectric constant, conductivity, etc. S12. Construct a full-scale layered medium test model for backfill grouting; the full-scale layered medium test model for backfill grouting includes segments, grouting body, backfill, and fixed supports; by constructing a high-fidelity numerical simulation model and a physical full-scale test model, the organic combination of the two models can not only reveal the reflection echo law of defects inside the grouting layer, but also effectively identify the grouting interface location and evaluate the grouting fullness, thereby significantly improving the quality control level and safety assurance capability of shield tunnel construction.

[0006] Preferably, step S2 includes the following steps: S21. Using the numerical model of the shield tunnel segment-grouting layer and the full-scale layered medium test model of the back wall grouting, the radar signal data of the current back wall grouting of the shield tunnel is obtained, and the current grouting radar image data is obtained. S22. The current grouting radar image data is subjected to denoising, gain correction, background filtering, and depth image enhancement processing to obtain the current processed grouting radar image data. This multi-level and multi-dimensional signal processing strategy not only greatly improves the quality and interpretability of radar images, but also provides high-quality data assurance for the accuracy and robustness of subsequent intelligent recognition algorithms, thereby significantly enhancing the overall technical level of non-destructive testing of grouting quality behind shield tunnel walls.

[0007] Preferably, step S3 includes the following steps: S31. Preset several types of grouting defects behind the shield wall and corresponding defect characterization parameter types to obtain a set of preset grouting defect types and a set of grouting defect characterization parameter types. S32. Based on the preset grouting defect type set and the grouting defect characterization parameter type set, preset different types of grouting defect types and corresponding characterization parameter types for the shield tunnel segment-grouting layer numerical model and the full-scale layered medium test model for backfill grouting. After the preset is completed, execute S21 and S22 to obtain the current experimental grouting radar image dataset, the current experimental grouting defect type dataset, and the current experimental grouting defect characterization parameter dataset. Use the current experimental grouting defect type dataset and the current experimental grouting defect characterization parameter dataset to manually annotate some image data in the current experimental grouting radar image dataset to obtain the current annotated experimental grouting radar image dataset. S33. Based on the current labeled experimental grouting radar image dataset, construct a mapping model that takes grouting radar image data as input and outputs grouting defect types and corresponding characterization parameter data to obtain the final grouting defect type and parameter mapping model. By accurately pre-setting various defects and their parameters in numerical simulation and full-scale test models, a large amount of radar image data containing known defect types and parameters can be generated, greatly enriching the diversity and representativeness of training samples. By manually annotating some images, the trained mapping model can not only identify defect types but also accurately predict their specific geometric and physical parameters, significantly improving detection efficiency and the scientific nature of engineering decisions.

[0008] Preferably, step S4 includes the following steps: S41. Using the same signal image processing method as in S22, the actual radar image of the grouting layer obtained from the current shield tunneling construction site is processed to obtain the current site grouting radar image data; the current site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping to obtain the current site grouting defect type data and the current site grouting defect parameter dataset; based on the current site grouting defect type data and the current site grouting defect parameter dataset, the grouting layer at the current shield tunneling construction site is repaired for defects and the repair time is recorded to obtain the current repair time data; S42. Based on historical experience, set the time threshold for repairing grouting layer defects corresponding to various grouting defect types and their corresponding defect parameters to obtain a set of grouting defect repair time thresholds; and repeat S41 several times. S43. Set a defect repair timeout probability threshold; if the number of times the current repair time exceeds the time threshold corresponding to the grouting defect repair timeout threshold set is greater than or equal to the total number of repetitions in S42, execute S5; otherwise, the repair is complete; through multiple rounds of repeated detection, the repair effect is dynamically verified to avoid misjudgment caused by environmental interference or operational errors, thereby improving the repair success rate; a defect repair timeout probability threshold is introduced, and when the number of times the repair time exceeds the threshold exceeds a preset proportion, an emergency response is automatically triggered to avoid misoperation caused by single detection error.

[0009] Preferably, step S5 includes the following steps: S51. Set several types of operation parameters corresponding to the noise reduction, gain correction, background filtering and depth image enhancement processing operations in S22 to obtain a set of defect image processing operation parameter types. S52. Based on the defect image processing operation parameter type set, using the methods of S41, S42, and S43, and setting the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping to be different each time, record the proportion of the number of times the current repair time data is greater than the corresponding time threshold in the grouting defect repair time threshold set to the total number of repetitions in S42, the number of mappings, and the corresponding defect image processing operation parameter data, to obtain the current recorded time excess ratio dataset, the current recorded mapping number set, and the current recorded operation parameter dataset; S53. Based on the current record time exceeding limit percentage dataset, the current record mapping count dataset, and the current record operation parameter dataset, construct a mapping model with mapping counts and operation parameter data as inputs and time exceeding limit percentage data as outputs to obtain the final defect repair time exceeding limit percentage mapping model; this addresses the issue that only single defects are considered during defect repair, without considering compound defects, leading to an increase in the defect repair time exceeding limit percentage, which in turn affects the progress of the entire shield wall backfilling project; real-time recording and feedback of repair time forms a closed-loop optimization mechanism to continuously improve the robustness and prediction accuracy of the model.

[0010] Preferably, step S6 includes the following steps: S61. Based on the defect image processing operation parameter type set, the operation parameters for denoising, gain correction, background filtering and depth image enhancement processing corresponding to the repeated S41 process in S42, as well as the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping, are repeatedly adjusted to obtain the current adjusted operation parameter set and the current adjusted mapping number. S62. Input the current adjusted operation parameter set and the current adjusted mapping count into the final defect repair time exceeding the limit ratio mapping model to obtain the current adjusted record time exceeding the limit ratio data; until the current adjusted record time exceeding the limit ratio data is less than the defect repair timeout probability threshold, the current final operation parameter set and the current final mapping count are obtained. S63. Based on the current final operation parameter set and the current final mapping count, the operation parameters for denoising, gain correction, background filtering, and depth image enhancement processing corresponding to the repeated S41 process in S42, as well as the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping, are actually set.

[0011] The intelligent identification system for grouting defects behind shield tunnel walls based on image processing includes a double-fit model construction module, an image preprocessing module, a parameter mapping model construction module, a repair timeout judgment module, a proportion mapping model construction module, and a parameter adjustment module. The double-fit model construction module is used to construct a physical test model; the image preprocessing module is used to preprocess radar signal images; the parameter mapping model construction module is used to construct the final grouting defect type and parameter mapping model; the repair timeout judgment module is used to judge the timeout during the repair process of the on-site radar image; the proportion mapping model construction module is used to construct the proportion mapping model of the final defect repair time exceeding the limit; and the parameter adjustment module is used to repeatedly adjust the corresponding preprocessing parameters and the number of mappings during the detection process.

[0012] The present invention has the following beneficial effects: 1. This invention achieves accurate detection and repair of grouting defects by establishing a three-layer medium model and combining physical experiments with radar signal images. First, the physical accuracy of the model is ensured by setting real electromagnetic parameters and boundary conditions. After presetting the defect type, the generated radar image data is manually annotated to construct a defect type and parameter mapping model. The preprocessed on-site radar image is input into the model to detect the defect type and parameters in real time, record the repair time and compare it with the threshold, and automatically determine whether the repair is complete or requires further adjustment. Through multiple parameter optimizations and mappings, a time-exceeding ratio mapping model is finally constructed to achieve efficient defect repair. This significantly improves detection accuracy and repair efficiency, reduces manual intervention, and lowers repair time costs, making it suitable for complex grouting projects. 2. In this invention, by repeatedly adjusting the operating parameters in the radar image data processing process and repeatedly inputting the acquired and preprocessed radar images into the final grouting defect type and parameter mapping model, the intensity of radar image preprocessing is adjusted and the optimal number of repeated mappings is determined to best ensure the comprehensiveness of the mapping results. This reduces the proportion of time exceeding the limit for repairing grouting layer defects to below the threshold, ensuring the progress of the entire shield wall grouting project. In particular, by using a pre-constructed mapping model of the proportion of time exceeding the limit for final defect repair corresponding to the adjusted radar image preprocessing parameters and the number of repeated mappings for real-time mapping, there is no need to repeat the previous actual image acquisition and mapping steps; only the mapping in the model needs to be input, thus ensuring adjustment efficiency. 3. In this invention, a defect repair timeout probability threshold is introduced. When the number of repair times exceeding the threshold exceeds a preset proportion, an emergency response is automatically triggered to avoid misoperation caused by single detection error.

[0013] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to the present invention. Figure 2 This is a schematic diagram of the process for preprocessing the current grouting radar image according to the present invention; Figure 3 This is a schematic diagram of the process for finally constructing the grouting defect type and parameter mapping model of the present invention; Figure 4 This is a schematic diagram of an image for identifying grouting void defects according to the present invention; Figure 5 This is a schematic diagram of the grouting cavity area of ​​the present invention (left half of the image) and the corresponding label (right half of the image); Figure 6 This is a schematic diagram of the grout layer thickness defect identification results of the present invention; Figure 7 This is a schematic diagram of the grouting crack defect identification results of the present invention; Figure 8 This is a schematic diagram of the grouting leakage defect image of the present invention; Figure 9 For the present invention Figure 8 A schematic diagram of the leakage labels corresponding to the images of grouting leakage defects; Figure 10 This is a schematic diagram of the process for determining the timeout in the repair of current grouting defects according to the present invention; Figure 11 A flowchart illustrating the process of constructing the final defect repair time exceeding the limit mapping model for this invention; Figure 12 This is a line graph showing the trend of the probability of exceeding the time limit for grouting defect repair in this invention as a function of the number of mappings; Figure 13 This is a schematic diagram illustrating the process of adjusting the defect image processing parameters and the number of mappings according to the present invention. Figure 14 This is a schematic diagram of the module of the intelligent identification system for grouting defects behind shield tunnel walls based on image processing according to the present invention. Detailed Implementation

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0018] Example 1 Please see Figure 1 This embodiment is an intelligent identification method for grouting defects behind shield tunnel walls based on image processing, including the following steps: S1, establish a three-layer medium model of grouting behind shield tunnel walls and set real electromagnetic parameters, boundary conditions and preset defect types; then construct a physical test model.

[0019] Please see Figure 2 S1 includes the following steps: S11, using the finite-difference time-domain method (FDTD, an effective tool for simulating electromagnetic wave propagation) and GPRmax software to establish a three-layer medium model including tunnel segments, backfill grout, and surrounding rock and soil. To closely approximate engineering practice, electromagnetic parameters and corresponding boundary conditions that conform to the actual physical properties of each medium layer are set, and simulated defect types are preset in the backfill grout, to obtain a shield tunnel segment-grout layer numerical model; the electromagnetic parameters include dielectric constant, conductivity, etc.

[0020] For example, the detailed construction steps of the numerical model of the shield tunnel segment-grouting layer are as follows: (1) Establish a three-dimensional spatial model, the core structure of which consists of three layers of media, from the inside out: Tunnel segment layer: simulating the precast concrete lining ring of a shield tunnel, its thickness is set to 300 mm according to common engineering dimensions; Backfill grout layer: simulating the hardened grout layer filled between the tunnel segments and the soil after synchronous grouting, its designed thickness is 150 mm, and simulated defects are preset in it, such as a spherical cavity with a diameter of 100 mm, or a locally thinned area with a thickness of only 50 mm; Surrounding rock and soil layer: simulating the natural strata around the tunnel, its extent is set to avoid boundary reflection interference, for example, extending 2 meters in each direction around the model; (2) Assign electromagnetic parameters to each layer of medium that conform to its real physical properties. This is the key to the accuracy of the simulation. The specific parameter settings refer to common engineering materials and literature data: Tunnel segments (concrete): The relative permittivity is set to 6.5 and the conductivity is set to 0.01S / m to simulate the electromagnetic properties of dry concrete; The backfill grout: The relative permittivity is set to 9.0 and the conductivity is set to 0.02S / m; The preset void defects are filled with air, and their electromagnetic parameters are: relative permittivity is 1.0 and conductivity is 0S / m; The surrounding rock and soil: Based on the assumption of typical clay strata, the relative permittivity is set to 12.0 and the conductivity is set to 0.05S / m. (3) To simulate an infinite space and avoid unnecessary boundary reflections, all outer surfaces of the model use Perfectly Matched Layer (PML, a numerical method for simulating an infinite space and avoiding boundary reflections, commonly used in simulations of electromagnetic fields, acoustics, etc.; it effectively absorbs outward-propagating waves by setting a special absorption layer outside the boundary of the computational region, thereby reducing the impact of reflections on the simulation results) as boundary conditions; the number of PML layers is set to 10 to effectively absorb outward-propagating electromagnetic waves; the entire computational region is discretized using a uniform Yee grid; to ensure simulation accuracy and meet the stability conditions of the FDTD algorithm (usually requiring the grid size to be less than one-tenth of the minimum wavelength), the grid step size is uniformly set to 0.005 meters; for a radar antenna with a center frequency of 900MHz, its wavelength in the grouting body is about 0.1 meters, and this grid size meets the requirements; (4) The simulated radar excitation source uses Ricker wavelet (a commonly used time-domain pulse), and its center frequency is set to 400MHz and 900MHz respectively according to the detection requirements, so as to compare and analyze the resolution capability of different frequency antennas for defects; the antenna model is simplified to a dipole antenna; in the model, it is placed on the inner wall surface of the tunnel segment to simulate the process of the antenna moving and scanning close to the inner wall of the segment in actual detection; the transmit and receive mode is set to the common offset mode, that is, the transmitting antenna and the receiving antenna maintain a fixed distance (e.g., 0.1 meters) and move synchronously; the detection process is carried out by a pre-set measurement line along the inner wall of the segment. A measurement line with a length of 3 meters is set, and a series of detection points are set along this measurement line at intervals of 0.02 meters. The radar echo signal is recorded at each point, and finally a radar profile is synthesized. (5) Input all the above parameters (3D spatial model, material properties, mesh, boundary, excitation source and survey line) into the GPRmax software and perform finite-difference time-domain calculation; the software will calculate the entire process of electromagnetic wave propagation in the model, reflection and transmission when it encounters various medium interfaces and defects; the final output result is the time series signal of each detection point, i.e., synthetic radar data (B-scan image); by analyzing the arrival time, amplitude and phase axis continuity of the reflected wave in the image, the radar response characteristics of defects such as voids of different sizes and thickness reduction can be studied, the detection depth and resolution of GPR can be evaluated, and accurate theoretical reference can be provided for subsequent physical experiment design and selection of actual engineering detection parameters.

[0021] S12. Construct a full-scale layered medium test model for backfill grouting; the full-scale layered medium test model for backfill grouting includes tunnel segments, grout body, backfill, and fixed supports; wherein, the tunnel segments adopt small-scale tunnel segments with the same material as those used in the shield tunneling construction site, grouting cavities are simulated by pre-embedded PVC pipes, grout layer thickness changes are simulated by manually thinning the grout body, and actual scanning is simulated by dragging GPR. By setting simulated cavities and local thinning areas in the grout body, the reflection echo characteristics of the grout layer thickness and defects can be observed; to verify the numerical simulation results and study the actual detection effect, this invention proposes to build a full-scale layered medium test model for backfill grouting. At the same time, the experiment combines different GPR frequencies and parameter settings to analyze the imaging effect and the influence of parameters on detection accuracy, providing a basis for the timing of backfill grouting detection and the selection of GPR parameters in actual engineering; by combining the construction of a high-fidelity numerical simulation model with a physical full-scale test model, the accuracy and reliability of backfill grouting defect detection in shield tunnels are significantly improved; specifically, the finite-difference time-domain method is used in conjunction with GPRmax software. The established three-layer medium model can accurately simulate the propagation behavior of electromagnetic waves in tunnel segments, grouting layers, and surrounding rock. By setting electromagnetic parameters such as dielectric constant and conductivity that conform to the actual material properties, the simulation results are closer to the real response at the engineering site, providing a reliable theoretical basis for the electromagnetic wave reflection characteristics analysis of grouting defects. At the same time, the constructed full-scale layered medium test model, by using segments of the same material as those at the construction site, pre-embedded PVC pipes to simulate cavities, and manually thinning areas to simulate thickness changes, realistically reproduces the complex situation of the grouting structure behind the wall, making the test results more reliable in engineering. The organic combination of the two models can not only deeply reveal the reflection echo law of defects inside the grouting layer, effectively identify the grouting interface location and evaluate the grouting fullness, but also provide a scientific basis for optimizing ground penetrating radar detection parameters and improving on-site detection efficiency, thereby significantly improving the quality control level and safety assurance capability of shield tunnel construction. S2, the three-layer medium model and physical test model based on S1 acquire radar signal images and preprocess them.

[0022] S2 includes the following steps: S21. Using the numerical model of the shield tunnel segment-grouting layer and the full-scale layered medium test model of the back wall grouting, the radar signal data of the current back wall grouting of the shield tunnel is obtained, and the current grouting radar image data is obtained. S22. The current grouting radar image data is processed using time-domain, frequency-domain, and wavelet transform methods for denoising, gain correction, background filtering, and depth image enhancement to obtain the processed grouting radar image data. By comprehensively utilizing a combination of numerical simulation and physical experiments, radar signal data of the grouting state behind the shield wall can be obtained comprehensively and accurately, laying a solid foundation for subsequent high-quality image processing. In the data acquisition stage, a high-fidelity three-layer medium numerical model and a full-scale experimental model closely resembling engineering reality can realistically reproduce the complex electromagnetic wave propagation environment inside the grouting layer, ensuring that the acquired radar signal data has good representativeness and reliability. In the image processing stage, the organic combination of various advanced signal processing techniques such as time-domain analysis, frequency-domain analysis, and wavelet transform can effectively filter out... Environmental noise interference in radar data significantly improves the signal-to-noise ratio; precise gain correction processing can compensate for energy attenuation of electromagnetic waves propagating in different media, enabling effective recovery of deep defect signals; background filtering technology can effectively eliminate strong interference signals such as direct waves and rebar reflection waves, highlighting the weak reflection characteristics of grouting layer interfaces and internal defects; depth image enhancement processing further optimizes image contrast and clarity, making key information such as grouting layer thickness changes and void boundaries more distinct and easily identifiable; this multi-level, multi-dimensional signal processing strategy not only significantly improves the quality and interpretability of radar images, but also provides high-quality data assurance for the accuracy and robustness of subsequent intelligent recognition algorithms, thereby significantly enhancing the overall technical level of non-destructive testing of grouting quality behind shield tunnel walls; S3. Preset grouting defect types and corresponding parameters. Use the three-layer medium model and physical test model constructed in S1 to generate grouting radar image data. After manually annotating the generated data, construct the final grouting defect type and parameter mapping model.

[0023] Please see Figures 3-9 S3 includes the following steps: S31, pre-setting several types of grouting defects behind the shield wall and corresponding defect characterization parameter types to obtain a pre-set set of grouting defect types and a set of grouting defect characterization parameter types; the contents of the pre-set set of grouting defect types and the set of grouting defect characterization parameter types can be referred to as Table 1 below:

[0024] For grouting voids, the main treatment principle is filling and reinforcement. Conventional measures involve using pre-existing grouting holes in the tunnel segments for secondary or simultaneous grouting to directly fill the void area. If the void is large and located in a shallow layer, it can be backfilled in layers with materials such as lime-soil and compacted before grouting reinforcement. For deep voids or voids in complex strata, it is recommended to use sleeve valve pipes for precise deep-hole grouting to ensure dense filling and restore the synergistic stress between the strata and the tunnel segments. Regarding uneven grout thickness, the focus is on process optimization and reinforcing grouting. Prevention is key; dynamically adjust grouting pressure, speed, and grout mix ratio to maintain synchronization between grouting and tunnel boring and avoid grout deviation. For sections with uneven grouting, timely reinforcement grouting should be carried out, with secondary grouting performed through the grouting holes in the segments to fill areas with poor compaction. If uneven grouting has caused localized tunnel deformation, a micro-disturbance grouting method of "uniform, multi-point, small-volume, and multiple times" should be adopted, with precise control from both sides of the tunnel to correct thickness deviations and control stratum deformation. For segment cracks caused by uneven grouting pressure or stress, repair measures should be graded according to crack width and depth. Small surface cracks can be simply filled and sealed with cement mortar or fine aggregate concrete. For deeper cracks that affect structural waterproofing or structural integrity, pressure injection of modified epoxy resin or other chemical grouts should be used to achieve structural repair. If the segments suffer from severe damage such as chipped corners or large-area breakage, the damaged surface must be thoroughly cleaned, filled in layers, cured with formwork, and surface beautification treatment should be performed to restore the overall strength and durability of the segments. When dealing with tunnel leakage defects, the principle of "combining plugging and drainage, and comprehensive management" should be followed.For leakage at the joints (circumferential and longitudinal joints) of the tunnel segments, isolation columns are usually first drilled along both sides of the leakage point, grouting nozzles are then embedded with quick-sealing material, modified epoxy resin or polyurethane chemical grout is injected for leakage sealing and structural reinforcement, and finally, high-elasticity sealant is used to fill and seal the leak. For leakage at the grouting hole location, the sealing ring should be tightened or replaced first, and re-sealing and grouting should be carried out if necessary. If leakage is caused by water pockets behind the wall, backfilling grouting behind the wall is required, especially in water-rich strata, where water-reactive polyurethane chemical grout is used to force the cavitary water into fissure water, fundamentally eliminating water pressure and leakage channels. S32. Based on the preset grouting defect type set and the grouting defect characterization parameter type set, the numerical model of the shield tunnel segment-grouting layer and the full-scale layered medium test model of the backfill grouting are tested. Different types of grouting defects and their corresponding characterization parameter types are preset respectively. After the preset is completed, S21 and S22 are executed to obtain the current experimental grouting radar image dataset, the current experimental grouting defect type dataset, and the current experimental grouting defect characterization parameter dataset. The current experimental grouting defect type dataset and the current experimental grouting defect characterization parameter dataset are used to manually annotate some image data in the current experimental grouting radar image dataset to obtain the current annotated experimental grouting radar image dataset. S33: Based on the current annotated experimental grouting radar image dataset, a mapping model is constructed with grouting radar image data as input and grouting defect type and corresponding characterization parameter data as output to obtain the final grouting defect type and parameter mapping model.

[0025] S33 includes the following steps: S331. Construct an initial grouting defect type and parameter mapping model; use the current labeled experimental grouting radar image dataset as the first training dataset, and use the image data in the current experimental grouting radar image dataset that has not been labeled with features as the first test dataset; S332. Set a first training error threshold (10%~15%, which can be adjusted adaptively according to the actual training situation); input the first training dataset into the initial grouting defect type and parameter mapping model for training; during the training process, if the training error is less than the first training error threshold, stop training and obtain the trained grouting defect type and parameter mapping model; otherwise, continue training until the training error is less than the first training error threshold. S333. Set a first test accuracy threshold (90%~95%, which can be adjusted adaptively according to the actual test situation); input the first test dataset into the trained grouting defect type and parameter mapping model for testing; after the test is completed, obtain the first test accuracy data; if the first test accuracy data is greater than or equal to the first test accuracy threshold, use the trained grouting defect type and parameter mapping model as the final grouting defect type and parameter mapping model; otherwise, return to S332 to continue training the trained grouting defect type and parameter mapping model and repeat S333 until the first test accuracy data is greater than or equal to the first test accuracy threshold.

[0026] The structure of the initial grouting defect type and parameter mapping model can be referred to in Table 2 below:

[0027] The aforementioned structure can effectively learn discriminative features from grouting radar images to distinguish different defect types, and utilize these shared features to simultaneously classify defect types and predict specific parameters, achieving end-to-end multi-task intelligent mapping. By systematically constructing a standard system for grouting defect types and their characterization parameters, and deeply integrating it with high-fidelity numerical models and physical experimental models, the accuracy and generalization ability of intelligent identification of grouting defects behind the shield wall are significantly improved. The solution first establishes a complete defect type library covering geometric defects, thickness defects, crack defects, and leakage defects, and precisely defines key characterization parameters for each type of defect, providing a standardized labeling system for subsequent model training. By accurately pre-setting various defects and their parameters in numerical simulation and full-scale experimental models, a large amount of radar image data containing known defect types and parameters can be generated, greatly enriching the diversity and representativeness of training samples. More importantly, by manually annotating some images, the trained mapping model can not only identify defect types, but also accurately predict their specific geometric and physical parameters (such as cavity diameter, crack length, thickness variation range, etc.), realizing a direct and accurate mapping from radar images to defect quantification information. This provides strong technical support for the refined evaluation and intelligent operation and maintenance of shield tunnel construction quality, significantly improving detection efficiency and the scientific nature of engineering decisions. S4: The on-site radar images are preprocessed and input into the mapping model in S3 for mapping defect types and parameters; the repair time is recorded and compared with a preset threshold, and the probability of timeout during repeated detection is statistically analyzed; if the probability of timeout is less than the threshold, the repair is completed; otherwise, proceed to S5.

[0028] Please see Figure 10 S4 includes the following steps: S41. Using the same signal image processing method as in S22, the actual radar image of the grouting layer obtained from the current shield tunneling construction site is processed to obtain the current site grouting radar image data; the current site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping to obtain the current site grouting defect type data and the current site grouting defect parameter dataset; based on the current site grouting defect type data and the current site grouting defect parameter dataset, the grouting layer at the current shield tunneling construction site is repaired for defects and the repair time is recorded to obtain the current repair time data; S42. Based on historical experience, set the time threshold for repairing grouting layer defects for various grouting defect types and corresponding defect parameters to obtain the grouting defect repair time threshold set; repeat S41 several times (the number of repetitions can be adaptively set according to the actual situation, the more the better). S43. Set a defect repair timeout probability threshold; if the number of times the current repair time exceeds the timeout threshold corresponding to the grouting defect repair timeout threshold set is greater than or equal to the total number of repetitions in S42, execute S5; otherwise, the repair is complete; through real-time signal image processing, the defect type (such as voids, cracks) and its parameters (such as size, depth) of the grouting layer can be captured instantly, and the repair operation can be automatically triggered, greatly shortening the response time; based on historical repair data, a scientific repair time threshold is set for each defect type (such as void repair requiring ≤30 minutes) to ensure the rationality and safety of the repair operation; through multiple rounds of repeated detection The system dynamically verifies the repair effect, avoiding misjudgments caused by environmental interference or operational errors, thus improving the repair success rate. It introduces a defect repair timeout probability threshold; when the number of repair times exceeding the threshold exceeds a preset proportion (e.g., ≥80%), an emergency response is automatically triggered to avoid misoperation due to single-detection errors. Through statistical analysis of historical repair time data, the system adaptively adjusts the threshold set to achieve dynamic optimization (e.g., automatically extending the crack repair threshold when humidity increases during the rainy season), ensuring efficient repair in complex environments. Through multiple rounds of detection and threshold correction, the system can effectively cope with dynamic changes in the construction environment (e.g., temperature fluctuations, humidity changes), ensuring the stability and reliability of repair operations. S5. By repeatedly adjusting the multiple operation parameters preprocessed in S2 and the number of mappings in S4, a grouting defect repair experiment is conducted. During the experiment, the data on the repair time exceeding the limit is recorded, and a final defect repair time exceeding the limit mapping model is constructed based on the recorded data.

[0029] Please see Figure 11S5 includes the following steps: S51, setting several operation parameter types corresponding to the denoising, gain correction, background filtering, and depth image enhancement processing operations in S22, to obtain a defect image processing operation parameter type set; the contents of the defect image processing operation parameter type set can be referred to as Table 3 below:

[0030] S52. Based on the defect image processing operation parameter type set, using methods S41, S42, and S43, and setting the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping to be different each time, record the proportion of multiple sets of current repair time data exceeding the corresponding time threshold in the grouting defect repair time threshold set to the total number of repetitions in S42, the number of mappings, and the corresponding defect image processing operation parameter data, to obtain the current recorded time exceeding limit percentage dataset, the current recorded mapping number set, and the current recorded operation parameter dataset; S53. Based on the current recorded time exceeding limit percentage dataset, the current recorded mapping number set, and the current recorded operation parameter dataset, construct a mapping model with the input being the mapping number and operation parameter data, and the output being the time exceeding limit percentage data, to obtain the final defect repair time exceeding limit percentage mapping model.

[0031] S53 includes the following steps: S531. Construct an initial defect repair time exceeding limit ratio mapping model and set a first training data ratio (such as 8:2 or 7:3, which can be adjusted adaptively according to the actual training situation); divide the current record time exceeding limit ratio dataset, the current record mapping number set and the current record operation parameter dataset according to the first training data ratio to obtain a second training dataset and a second test dataset. S532. Set a second training error threshold (10%~15%, which can be adjusted adaptively according to the actual training situation); input the second training dataset into the initial defect repair time exceeding the limit ratio mapping model for training; during the training process, if the training error is less than the second training error threshold, stop training and obtain the trained defect repair time exceeding the limit ratio mapping model; otherwise, continue training until the training error is less than the second training error threshold. S533. Set a second test accuracy threshold (90%~95%, which can be adjusted adaptively according to the actual test situation); input the second test dataset into the trained defect repair time exceeding the limit ratio mapping model for testing; after the test is completed, obtain the second test accuracy data; if the second test accuracy data is greater than or equal to the second test accuracy threshold, use the trained defect repair time exceeding the limit ratio mapping model as the final defect repair time exceeding the limit ratio mapping model; otherwise, return to S532 to continue training the trained defect repair time exceeding the limit ratio mapping model and repeat S533 until the second test accuracy data is greater than or equal to the second test accuracy threshold.

[0032] The structure of the initial defect repair time exceeding the limit ratio mapping model can be referred to in Table 4 below:

[0033] MLP regression models have a simple structure and high training efficiency, making them ideal for handling mixed input data consisting of scalar and encoded categorical features for regression prediction tasks. Through multi-layer ReLU activation functions, they can capture the non-linear relationship between input parameters (such as the AGC exponent and kernel size) and the proportion of time exceeding limits. The Sigmoid activation function ensures that the output is always within the [0, 1] interval, perfectly matching the physical meaning of "proportion." Regularization techniques such as Dropout effectively prevent overfitting of the model to the training data, improving its predictive stability on new data. Through multiple rounds of detection and parameter combinations, parameters for denoising, gain correction, background filtering, and image enhancement (such as the AGC exponent and kernel size) are dynamically adjusted to achieve accurate processing of complex environments (such as strong noise and low contrast). A mapping model is constructed to map the number of mappings and operation parameter data to the proportion of time exceeding limits, predicting the reliability of repair operations. For example, when the number of mappings increases, if the parameter combination (such as high gain correction) leads to an increase in the proportion of time exceeding limits, the parameters are automatically adjusted (such as reducing the AGC exponent). The model, constructed using the C-index, can automatically identify environmental changes (such as increased humidity) and adjust parameters (such as increasing the convolution kernel size) to ensure efficient repair even under complex working conditions. By considering the number of mapping iterations, it avoids the incompleteness of results caused by a single mapping iteration, which could lead to only considering single defects and neglecting composite defects during defect repair, thus increasing the proportion of defect repair time exceeding the limit and affecting the progress of the entire shield wall grouting project. Real-time recording and feedback of repair time form a closed-loop optimization mechanism to continuously improve the robustness and prediction accuracy of the model. Seamless integration of signal processing, model mapping, repair operation, and threshold decision-making enables full-process automation from defect detection to repair completion. For example, multi-round detection (S41) dynamically verifies the repair effect to avoid misjudgments caused by environmental interference or operational errors. In S6, the preprocessing parameters and mapping iterations corresponding to the repeated detection process in S4 are repeatedly adjusted and input into the mapping model in S5 for mapping. Feedback on the adjustment effect is provided based on the mapping results until the probability of time exceeding the limit is lower than the threshold.

[0034] Please see Figure 12 , Figure 13 S6 includes the following steps: S61. Based on the defect image processing operation parameter type set, the operation parameters for denoising, gain correction, background filtering and depth image enhancement processing corresponding to the repeated S41 process in S42, as well as the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping, are repeatedly adjusted to obtain the current adjusted operation parameter set and the current adjusted mapping number. S62. Input the current adjusted operation parameter set and the current adjusted mapping count into the final defect repair time exceeding the limit ratio mapping model to obtain the current adjusted record time exceeding the limit ratio data; until the current adjusted record time exceeding the limit ratio data is less than the defect repair timeout probability threshold, the current final operation parameter set and the current final mapping count are obtained. S63. Based on the current final operation parameter set and the current final mapping count, the operation parameters for denoising, gain correction, background filtering, and depth image enhancement processing corresponding to the repeated S41 process in S42, as well as the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model, are actually set. By repeatedly adjusting the operation parameters in the radar image data processing process and repeatedly inputting the acquired and preprocessed radar images into the final grouting defect type and parameter mapping model, the intensity of radar image preprocessing is adjusted and the optimal number of repeated mappings is determined to best ensure the comprehensiveness of the mapping results. This reduces the proportion of time exceeding the limit for grouting layer defect repair to below the threshold, ensuring the progress of the entire shield wall grouting project. In particular, by using the pre-constructed final defect repair time exceeding the limit mapping model corresponding to the adjusted radar image preprocessing parameters and the number of repeated mappings for real-time mapping, there is no need to repeat the previous actual image acquisition and mapping steps; only the mapping in the model needs to be input, thus ensuring adjustment efficiency.

[0035] Example 2 Please see Figure 14 This embodiment discloses an intelligent identification system for grouting defects behind shield tunnel walls based on image processing. The system can implement the methods of the above embodiments, including a double-fit model construction module, an image preprocessing module, a parameter mapping model construction module, a repair timeout determination module, a proportion mapping model construction module, and a parameter adjustment module. The double-fit model construction module is used to construct a physical test model; the image preprocessing module is used to preprocess radar signal images; the parameter mapping model construction module is used to construct the final grouting defect type and parameter mapping model; the repair timeout determination module is used to determine the timeout during the repair process of the on-site radar image; the proportion mapping model construction module is used to construct the proportion mapping model of the final defect repair time exceeding the limit; and the parameter adjustment module is used to repeatedly adjust the corresponding preprocessing parameters and the number of mappings during the detection process.

[0036] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0037] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present invention.

Claims

1. A method for intelligent identification of grouting defects behind shield tunnel walls based on image processing, characterized in that, Includes the following steps: S1. Establish a three-layer medium model for grouting behind the shield wall and set the actual electromagnetic parameters, boundary conditions and preset defect types; then construct a physical test model. S2. The radar signal image is obtained and preprocessed based on the three-layer medium model and physical test model constructed based on S1; S3. Preset grouting defect types and corresponding parameters. Use the three-layer medium model and physical test model constructed by S1 to generate grouting radar image data. After manually annotating the generated data with features, construct the final grouting defect type and parameter mapping model. S4. Preprocess the on-site radar image and input it into the mapping model in S3 to map the defect type and parameters; record the repair time and compare it with the preset threshold, and perform repeated detection to count the timeout probability; if the timeout probability is less than the threshold, the repair is complete. Otherwise, execute S5; S5. By repeatedly adjusting the multiple operation parameters preprocessed in S2 and the number of mappings in S4, a grouting defect repair experiment is conducted. During the experiment, the data on the repair time exceeding the limit is recorded, and a final defect repair time exceeding the limit mapping model is constructed based on the recorded data. S6. Repeatedly adjust the preprocessing parameters and mapping times corresponding to the repeated detection process in S4 and input them into the mapping model in S5 for mapping. Feedback on the adjustment effect is given based on the mapping results until the probability of time exceeding the limit is lower than the threshold.

2. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 1, characterized in that: The physical test models described in S1 include a numerical model of shield tunnel segments-grouting layer and a full-scale layered medium test model for backfill grouting; the electromagnetic parameters include dielectric constant and conductivity.

3. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 2, characterized in that, S2 includes the following steps: S21, using the shield tunnel segment-grouting layer numerical model and the full-scale layered medium test model for back wall grouting to obtain the radar signal data of the current shield back wall grouting and perform preprocessing.

4. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 3, characterized in that: The preprocessing described in S21 includes denoising, gain correction, background filtering, and depth image enhancement.

5. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 4, characterized in that: The input to the final grouting defect type and parameter mapping model in S3 is grouting radar image data, and the output is the grouting defect type and the corresponding characterization parameter data.

6. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 5, characterized in that, S4 includes the following steps: S41. Using the preprocessing method in S21, the actual radar image of the grouting layer obtained from the current shield tunneling construction site is processed in the same way to obtain the current site grouting radar image data; the current site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping; based on the mapping result, the grouting layer at the current shield tunneling construction site is repaired for defects and the repair time is recorded to obtain the current repair time data; S42. Set the time threshold for repairing grouting layer defects for various grouting defect types and corresponding defect parameters, and repeat S41 several times. S43. If the proportion of the number of times the current repair time is greater than the corresponding time threshold to the total number of repetitions in S42 is greater than or equal to the defect repair timeout probability threshold, execute S5; otherwise, the repair is complete.

7. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 6, characterized in that, S5 includes the following steps: S51. Set several types of operation parameters corresponding to the preprocessing operation in S22 to obtain a set of defect image processing operation parameter types. S52. Based on the defect image processing operation parameter type set, repeat S41, S42, and S43 multiple times and record the proportion of the number of times the current repair time data is greater than the corresponding time threshold to the total number of repetitions in S42, the number of mappings, and the corresponding defect image processing operation parameter data; the number of mappings represents the total number of times the processed radar image data is input into the final grouting defect type and parameter mapping model for mapping during the repetition process.

8. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 7, characterized in that: The input to the final defect repair time exceeding the limit mapping model described in S5 is the number of mappings and operation parameter data, and the output is the time exceeding the limit data.

9. The intelligent identification method for grouting defects behind shield tunnel walls based on image processing according to claim 8, characterized in that, S6 includes the following steps: S61. The operation parameters for denoising, gain correction, background filtering and depth image enhancement processing corresponding to the process of repeating S41 several times in S42, as well as the number of times the current on-site grouting radar image data is input into the final grouting defect type and parameter mapping model for mapping, are repeatedly adjusted to obtain the current adjusted operation parameter set and the current adjusted mapping number. S62. Input the current adjusted operation parameter set and the current adjusted mapping count into the final defect repair time-out ratio mapping model until the mapping result is less than the defect repair timeout probability threshold.

10. A system for implementing the intelligent identification method for grouting defects behind shield tunnel walls based on image processing as described in any one of claims 1-9, characterized in that: It includes a double-fit model construction module, an image preprocessing module, a parameter mapping model construction module, a repair timeout determination module, a proportion mapping model construction module, and a parameter adjustment module; the double-fit model construction module is used to construct physical experimental models; the image preprocessing module is used to preprocess radar signal images; The parameter mapping model construction module is used to construct the final grouting defect type and parameter mapping model; the repair timeout judgment module is used to judge the timeout during the repair process of the on-site radar image. The proportion mapping model construction module is used to construct the final defect repair time exceeding the limit proportion mapping model; the parameter adjustment module is used to repeatedly adjust the corresponding preprocessing parameters and mapping times during the detection process.

Images