A method and system for generating forward simulation data

Abstract

The application provides a method and system for generating forward simulation data, the method comprising: constructing a three-dimensional voxel model corresponding to a simulation scene, dividing a binning region according to the numerical characteristics of electromagnetic parameters corresponding to each voxel contained in the three-dimensional voxel model, and the binning region describing the value range of the numerical characteristics of the electromagnetic parameters; mapping and converting the electromagnetic parameters with different numerical characteristics corresponding to multiple voxels in one binning region into electromagnetic parameters with the same numerical characteristics; determining a target connected domain in the three-dimensional voxel model, wherein the voxels contained in the target connected domain correspond to electromagnetic parameters with the same numerical characteristics and the voxels are connected in space in the three-dimensional voxel model; selecting a local region of interest from the three-dimensional voxel model, generating geometric commands corresponding to the target connected domain included in the local region of interest; and executing the geometric commands to generate forward simulation data. The technical problem of excessive number of geometric commands and low efficiency of simulation data generation is solved.

Description

Technical Field

[0001] This invention relates to the fields of geophysical exploration and underground engineering detection technology, specifically to a method and system for generating forward modeling simulation data. Background Technology

[0002] Ground penetrating radar (GPR) is a non-destructive testing tool that uses high-frequency electromagnetic waves to detect the distribution of underground media. GPR captures the signals that occur when electromagnetic waves propagate through underground media and encounter interfaces with differences in dielectric constant or conductivity, resulting in reflection, refraction, and scattering. Through signal processing and data analysis, it obtains spatial distribution information of underground structures.

[0003] With the rapid development of deep learning models, significant progress has been made in intelligent recognition and 3D reconstruction technologies based on GPR forward simulation data. Training these deep learning models often requires a large amount of high-quality simulation data. For example, voxel-by-voxel modeling of underground structures can accurately describe the physical characteristics of a 3D region of interest (ROI) based on the electromagnetic parameters of each voxel within one or more ROIs. However, since each voxel corresponds to a geometric command, a complete 3D ROI can contain tens of thousands or even hundreds of thousands of geometric commands. This makes it difficult for simulation software such as gprMax to quickly process the massive number of geometric commands, thus severely impacting the efficiency of simulation data generation.

[0004] Therefore, a method is needed to efficiently generate high-precision GPR forward modeling data for multiple scenarios to support the development of deep learning-based GPR intelligent interpretation and 3D reconstruction technologies. Summary of the Invention

[0005] This invention provides a method for generating forward simulation data to solve the technical problem of excessive geometric commands and low efficiency in generating simulation data.

[0006] This invention provides a method for generating forward simulation data, comprising: constructing a three-dimensional voxel model corresponding to a simulation scene; dividing the model into bin regions based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model, wherein the bin regions describe the value range of the electromagnetic parameters; mapping electromagnetic parameters with different numerical characteristics corresponding to multiple voxels belonging to a bin region to electromagnetic parameters with the same numerical characteristics; determining a target connected component in the three-dimensional voxel model, wherein the target connected component contains voxels with the same numerical characteristics of electromagnetic parameters and the voxels are spatially connected in the three-dimensional voxel model; selecting a region of interest from the three-dimensional voxel model and generating a geometric command corresponding to the target connected component included in the region of interest; and executing the geometric command to generate forward simulation data.

[0007] As can be seen, the above-mentioned method for generating forward simulation data can adapt to the needs of multi-scenario detection and has strong practicality and generalization ability: it can construct three-dimensional voxel models of various underground detection scenarios such as metal pipelines, non-metal pipelines, and building structural defects. Whether it is simple single underground medium detection, or complex multi-underground medium combination and special geological environment detection, it can efficiently generate high-precision forward simulation data with strong generalization ability. The method further simplifies the number of geometric commands without affecting the forward modeling accuracy, avoiding unnecessary calculations that consume excessive computing resources. Compared with traditional homogeneous modeling, which cannot distinguish underground medium defects with a small number of geometric commands, and the problem of excessive computational load and impracticality of voxel-by-voxel modeling, this method achieves the best balance between efficiency and accuracy. Large-scale forward simulation data generation can be completed on ordinary computing devices, reducing hardware investment costs.

[0008] In one embodiment of the present invention, binning regions are divided according to the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model, including: determining multiple segmented regions based on the numerical characteristics of the electromagnetic parameters; determining binning regions by applying corresponding binning methods to different segmented regions; and mapping the electromagnetic parameters of the voxels to electromagnetic parameters with the same numerical characteristics, wherein the electromagnetic parameters with the same numerical characteristics are determined based on the median or mean of the value range corresponding to the binning region.

[0009] In one embodiment of the present invention, based on the distribution of the numerical characteristics of electromagnetic parameters, the binning method includes linear equal-space binning or logarithmic scale binning.

[0010] In one embodiment of the present invention, a target connected region is determined in a three-dimensional voxel model, wherein the voxels contained in the target connected region have the same numerical characteristics of electromagnetic parameters and the voxels are connected in space in the three-dimensional voxel model, including: determining the smallest axis-aligned cuboid that completely encloses the voxels with the same numerical characteristics of electromagnetic parameters and is connected as the target connected region in the three-dimensional voxel model.

[0011] In one embodiment of the present invention, the method further includes: merging or eliminating target connected regions that meet preset conditions, wherein the preset conditions are that the volume of the target connected region is less than a preset volume threshold or the number of voxels contained in the target connected region is less than a preset number threshold.

[0012] In one embodiment of the present invention, the local region of interest is determined according to at least one of the following methods: grid extraction, target-oriented extraction, or key location extraction. Grid extraction determines the region of interest by moving a sliding window of a fixed size in a three-dimensional voxel model. Target-oriented extraction determines the local region of interest containing characteristic underground media in the three-dimensional voxel model based on thresholds and / or gradient features of electromagnetic parameters. Key location extraction determines the local region of interest based on known locations in the three-dimensional voxel model.

[0013] In one embodiment of the present invention, executing geometric commands to generate forward modeling simulation data further includes: determining the antenna center frequency, time window, spatial sampling interval, longitudinal resolution, and maximum detection depth corresponding to the simulation scene as system parameters; and using the system parameters and geometric commands as inputs to call the gprMax simulation software to generate forward modeling simulation data.

[0014] In one embodiment of the present invention, the simulation scenario includes at least one of the following: a local disease scenario, a metal pipeline scenario, a non-metal pipeline scenario, and a composite scenario.

[0015] A second aspect of the present invention provides a system for generating forward simulation data. The system includes: a model generation module, an electromagnetic parameter mapping processing module, a local region of interest (ROI) measurement module, and a simulation data calculation module. The model generation module constructs a three-dimensional voxel model corresponding to the simulation scene. The electromagnetic parameter mapping processing module divides the model into bin regions based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model, wherein the bin regions describe the range of values ​​for the numerical characteristics of the electromagnetic parameters. The electromagnetic parameter mapping processing module is also used to map electromagnetic parameters with different numerical characteristics corresponding to multiple voxels belonging to a bin region into electromagnetic parameters with the same numerical characteristics. The ROI measurement module is used to determine a target connected region in the three-dimensional voxel model, wherein the target connected region contains voxels with the same numerical characteristics of the electromagnetic parameters and the voxels are spatially connected in the three-dimensional voxel model. The simulation data calculation module is used to select a ROI from the three-dimensional voxel model, generate geometric commands corresponding to the target connected region included in the ROI, and execute the geometric commands to generate forward simulation data.

[0016] A third aspect of the present invention provides an electronic device 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 implement the steps of the method for generating forward simulation data as provided in the first aspect.

[0017] The beneficial effects of this invention are as follows: By using the above-mentioned method for generating forward simulation data based on ground penetrating radar, the efficiency of forward simulation data generation is greatly improved. By using electromagnetic parameter quantization mapping, the numerically discrete distribution of electromagnetic parameters is mapped into a fixed number of electromagnetic parameters. Then, by combining connected component analysis and geometric optimization, the number of geometric commands corresponding to large voxels is greatly compressed, which solves the problem of low efficiency when forward simulation software such as gprMax processes massive geometric commands. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0019] In the attached diagram: Figure 1 This is an application scenario diagram of the method for generating underground exploration GPR forward simulation data provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a method for generating GPR forward modeling data for underground exploration with multi-scenario adaptive quantization according to an embodiment of the present invention; Figure 3 This is a schematic diagram of various application scenarios provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the quantization mapping of electromagnetic parameters of each voxel in a three-dimensional voxel model provided in one embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the process of extracting and annotating local regions of interest from a three-dimensional voxel model according to an embodiment of the present invention; Figure 6 This is a structural diagram of a forward simulation data generation system provided in one embodiment of the present invention; Figure 7 This is a structural diagram of an electronic device for generating forward simulation data according to an embodiment of the present invention. Detailed Implementation

[0020] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] Please see Figure 1 , Figure 1 This is an application scenario diagram of the method for generating underground exploration GPR forward simulation data provided in an embodiment of the present invention.

[0024] like Figure 1 As shown, to generate the forward simulation data required for model training, technicians used a voxel-by-voxel modeling approach to construct a 3D scene of the underground pipe network distribution. For the 3D region of interest 10 of the underground structure (e.g., a pipe network distribution area within 10 meters underground and with a planar range of 20m × 20m), the area was divided into voxel 101 units of size 0.01m × 0.01m × 0.01m. The electromagnetic parameters (such as dielectric constant and conductivity) of each voxel were obtained to fully reconstruct the physical properties of the underground medium and ensure the high accuracy of the forward simulation data. It can be seen that each voxel corresponds to an independent geometric command (the geometric command is used to define the physical properties of each voxel). For the aforementioned 20m × 20m × 10m detection range, 200 × 200 × 100 = 4 million voxels need to be divided, thereby generating 4 million geometric commands.

[0025] Currently used GPR forward modeling software such as gprMax in engineering has significant performance limitations when handling large-scale geometric commands. It cannot quickly parse and compute the geometric commands corresponding to a large number of voxels, resulting in a significant increase in the forward modeling simulation cycle for a single scene (e.g., with conventional modeling and fewer than 10,000 geometric commands, a single scene simulation takes about 1 hour; while with the aforementioned voxel modeling and 4 million geometric commands, the simulation time can be extended to more than 36 hours). If it is necessary to generate massive simulation data with multiple scenes and multiple parameter combinations (e.g., 240 sets of scenes covering various soil moisture, pipeline burial depths, and pipeline materials), the overall simulation data generation cycle can reach more than ten days, further reducing generation efficiency and failing to meet the needs of rapid training and iterative optimization of deep learning models.

[0026] To address the low generation efficiency of forward simulation data, this invention provides a method for generating forward simulation data based on ground-penetrating radar (GPR), comprising the following steps: Constructing a three-dimensional voxel model corresponding to a scene. This three-dimensional voxel model is a modeling method that uses voxels (volume pixels) as basic units to represent three-dimensional space in a regular discretized manner. Scenes include metal pipeline scenes, non-metal pipeline scenes, building structure defect scenes, composite scenes, and special environment scenes; each scene contains at least one underground medium. Quantizing and mapping the electromagnetic parameters of the three-dimensional voxel model. Dividing the model into segmented regions based on the numerical characteristics of the electromagnetic parameters of each voxel, and applying corresponding binning processing to different segmented regions, outputting a mapping table of electromagnetic parameters. This mapping table maps electromagnetic parameters of multiple voxels with different values ​​to the same electromagnetic parameters. Performing connected component analysis on voxels with the same electromagnetic parameters for geometric optimization, determining the target connected component. In the target connected component, the voxels have the same electromagnetic parameters, and the voxels are spatially connected. A local region of interest (ROI) is identified from the 3D voxel model. Geometric commands corresponding to the target connected components within the ROI are generated, where the number of geometric commands corresponding to the target connected components is less than the number of geometric commands corresponding to all voxels within the ROI. The GPR system parameters corresponding to the gprMax tool are configured, including detection depth and electromagnetic wave propagation characteristics, determining the antenna center frequency, time window, spatial sampling interval, longitudinal resolution, and maximum detection depth, etc. Forward simulation data corresponding to the geometric commands is then generated using gprMax.

[0027] As can be seen, the above-mentioned forward simulation data generation method based on ground-penetrating radar significantly improves the efficiency of forward simulation data generation. By quantizing and mapping electromagnetic parameters, the discrete distribution of electromagnetic parameters is transformed into a fixed number of electromagnetic parameters. Combined with connected component analysis and geometric optimization, the number of geometric commands corresponding to large voxels is significantly compressed, solving the problem of low efficiency in forward simulation software such as gprMax when processing massive geometric commands. It adapts to multi-scenario detection needs and possesses strong practicality and generalization capabilities: it can construct 3D voxel models of various underground detection scenarios, including localized defects, metal pipelines, non-metal pipelines, building structural defects, composite scenarios, and special environments, covering multiple application areas such as roads, urban underground spaces, buildings, and geological disaster prevention and control. Whether it is simple single underground medium detection or complex multi-underground medium combination and special geological environment detection, it can efficiently generate high-precision forward simulation data with strong generalization capabilities, and can be widely applied to geophysical exploration and underground engineering detection scenarios. Balancing efficiency and accuracy while optimizing computational resource allocation: By merging or eliminating tiny connected domains, the number of geometric commands is further reduced without affecting forward modeling accuracy, avoiding unnecessary computations that consume excessive computational resources. Compared to the shortcomings of traditional homogeneous modeling, which cannot distinguish subsurface media with a small number of geometric commands, and the problem of excessive computational load and impracticality of voxel-by-voxel modeling, this method achieves the best balance between efficiency and accuracy. Large-scale forward modeling data generation can be completed on ordinary computing devices, reducing hardware investment costs.

[0028] Please see Figure 2 , Figure 2 This is a flowchart of a method for generating GPR forward modeling data for underground exploration using multi-scenario adaptive quantization, as provided in an embodiment of the present invention. Figure 2 As shown, the generation method includes the following steps.

[0029] Step S201: Construct three-dimensional voxel models corresponding to multiple scenes.

[0030] In some embodiments, the scenario here can be as follows: Figure 3 As shown, the scenarios include: Scene Category I: Localized defects, including road cavities (circular, elliptical, irregular shapes, 0.3-1.5 meters in size), delamination layers (0.05-0.3 meters in thickness), and cracks (1-50 millimeters in width); Electromagnetic parameters: Dielectric constant of dry cavities. (Dry) / (Wet), surrounding soil Scene Category II: Metal Pipeline Scene, including cast iron pipes and steel pipes (diameter 0.05-0.5 meters, burial depth 0.3-1.5 meters); Electromagnetic parameters: Cast iron conductivity S / m, steel pipe S / m. Scene Category III: Non-metallic pipeline scene, including PE water supply pipes ( PVC communication pipe ( ), concrete drainage pipes ( Water in the pipe Scene Category IV: Structural Defect Scene, including internal voids in concrete and steel mesh; Electromagnetic Parameters: Concrete steel bars S / m, void Scene Category V: Composite scene, including combinations of manhole covers, pipelines, and cavities, with multiple intersecting pipe networks, simulating a complex underground environment. Scene Category VI: Special environment scene, including karst / permafrost / high water-bearing areas, soil... Humidity gradient changes.

[0031] Step S202: Perform quantization mapping on the electromagnetic parameters included in the three-dimensional voxel model.

[0032] In some embodiments, the GPR forward modeling scene for metal pipeline scene detection is used as an example for illustration. The process of step S202 can be described by... Figure 4 The process can be described as an adaptive quantization algorithm, which mainly includes: acquiring a three-dimensional voxel model, determining the electromagnetic parameters corresponding to each voxel, dividing the numerical characteristics of the electromagnetic parameters into segmented regions, outputting a medium definition library, and the medium definition library describing the mapping table corresponding to the electromagnetic parameters of the voxels.

[0033] First, the scene is set up by selecting a 3D region of interest (ROI) from a 3D voxel model. The voxel dimensions of the ROI include 40×40×26 (0.4m×0.4m×0.26m) voxels, totaling 41,600 voxels. Each voxel can be a cube with a side dimension of 10mm (0.01m). Each voxel has an independent continuous dielectric constant. Electrical conductivity etc.

[0034] Next, the electromagnetic parameters corresponding to each voxel in the 3D voxel model, such as dielectric constant and conductivity, are segmented and quantized into a finite underground medium. This preserves the differences in physical properties of each voxel while significantly reducing the numerical variations of the electromagnetic parameters, laying the foundation for simplified modeling in subsequent steps. This quantization mapping process includes segmenting the electromagnetic parameters based on their numerical values ​​and performing numerical processing on the parameters.

[0035] Step S2021: Dielectric constant segmentation region division.

[0036] In some embodiments, a distribution histogram corresponding to the dielectric constant is determined based on the numerical value (numerical characteristic) of the dielectric constant. The distribution histogram can represent the numerical distribution of the dielectric constant and can describe and identify the distribution peaks and sparse regions corresponding to the dielectric constant. For example, the numerical value of the dielectric constant can be segmented into regions, including three partition types: low... district( ),middle district( ),high district( ).

[0037] Specifically, we first divide the dielectric constant distribution histogram of 41,600 voxels into three intervals, as shown in Table 1. Each segment can correspond to one or more underground media.

[0038]

[0039] Table 1 Step S2022: Perform boxing processing on the dielectric constant segmented region.

[0040] In some embodiments, step S2022 includes: using linearly equally spaced bins for low dielectric constant regions, using mean-adaptive bins for dielectric constant regions, and using logarithmic-scale bins for high dielectric constant regions. That is, based on further subdividing each segmented region, each segmented region is divided into multiple bins (also called bin regions), so that each bin can correspond to one or fewer similar underground media, and a fixed dielectric constant is set for each bin. The fixed dielectric constant here can be the median or mean of the value range corresponding to the bin.

[0041] Among them, linearly equally spaced bins are used for the low dielectric constant region, including: for The void area is divided into linearly spaced equal-interval boxes. These boxes can also be called interval divisions (ranges of dielectric constant values). The rules are as follows: The quantification formula is: ,in ; ( (This indicates rounding down).

[0042] For example, voxels , It was assigned to box number 6; voxels , It was assigned to box number 13; the final result was low. The area is divided into 15 equal-width boxes, each corresponding to a fixed dielectric constant. This not only allows for precise differentiation of the underground medium corresponding to each box, but also enables the use of the same dielectric constant to describe the voxels corresponding to the underground medium in that box. In other words, it maps dielectric constants with different values ​​belonging to the same box to the same dielectric constant, reducing the distribution of dielectric constant values.

[0043] Mean-adaptive binning is applied to the dielectric constant region, including: bin width Fine binning is used in the peak distribution area.

[0044] right The main roadbed area adopts mean-adaptive binning, with bin width... In the peak distribution area, fine-grained binning is performed. First, the average dielectric constant of this area is calculated to be 7.2. Centered on this average value, the peak area (6-8) is finely divided into bins with a width of 0.5, while the sparse edge areas (3-6, 8-15) are uniformly divided into bins with a width of 0.5. For example: voxels , assigned Box number; voxel , assigned Box No.

[0045] Final result: China The area was divided into 24 boxes, which accurately distinguished the changes in the physical properties of the main roadbed and avoided excessive division of sparse areas.

[0046] Logarithmic-scale binning is used for the high dielectric constant region, including: In the waterlogged disease area, logarithmic scale binning was used, formula (Pick (Suitable for dielectric constant ranges of 15 to 81). For example: voxels. , , It was assigned to box number 13; voxels , , It was assigned to box number 16. Final result: High The region is divided into 9 logarithmic bins to avoid excessive differentiation of high dielectric constant using linear bins, which is consistent with the response characteristics of GPR to high dielectric media.

[0047] Step S2023: Quantize the conductivity on a logarithmic scale.

[0048] In some embodiments, conductivity is quantized using a logarithmic scale, defining the conductivity range. S / m is mapped to 12 levels, and the quantization formula is as follows: .

[0049] For example: voxels Classified as Level 1; Voxels Classified into 3 levels; voxels The conductivity was divided into 4 levels; the final result was that 12 orders of magnitude of conductivity were determined, while preserving the differences in the loss characteristics of the medium.

[0050] Among these, the conductivity of metallic materials is identified through independent processing. Voxels of S / m are assigned independent electromagnetic parameters and are not merged with other subsurface media. Because metals are ideal total reflectors, their GPR response is completely different, allowing for the selection of independent electromagnetic parameters.

[0051] Step S2024: Output the medium definition library and establish a mapping table for the electromagnetic parameters of each voxel for the three-dimensional region of interest.

[0052] In some embodiments, the mapping table here can be as shown in Table 2. The final mapping table can represent the original 41,600 sets of continuously varying electromagnetic parameters ( , , This is compressed into a medium definition library containing N fixed electromagnetic parameters, which means mapping varying electromagnetic parameters into fixed electromagnetic parameters, including: dielectric constant. Electrical conductivity and permeability Here, the fixed values ​​for dielectric constant, conductivity, and permeability can be median, mean, or any user-defined value. The medium definition library here can represent the electromagnetic parameters of different types of underground media. N here can represent the number of each compartment plus the number of special zones (special compartments). For example: the low-density... Zone: 15 sub-containers, middle Section: 12 sub-containers, high Zone: 3 sub-bins and 1 independent level for special media (metal), final N=31. It can be seen that N=31 is not a fixed value, but a result of scenario adaptation. If the scenario changes (such as underground karst layer detection, underground tunnel detection), the media definition library will adjust according to the media type: Karst Detection: High The number of sub-compartments in the middle zone (karst cave filling water / mud) will increase to 5, and the number of sub-compartments in the lower zone (aerated karst cave) will decrease to 8, potentially increasing the total amount of underground media to 28; Tunnel lining detection: Middle The number of concrete (concrete) boxes has increased to 15, low With 10 separate containers, the total amount of underground media may become 35. The relative permeability can be represented by the constant 1.

[0053]

[0054] Table 2 It is understandable that after determining the medium definition library, the three-dimensional region of interest is quantized. That is, according to the mapping table, the dielectric constant, conductivity, etc., of each voxel in the three-dimensional region of interest are replaced with the corresponding dielectric constant and conductivity in the medium definition library. , and )wait.

[0055] Step S203: Perform geometric optimization based on connected component analysis.

[0056] In some embodiments, the 3D voxel model quantized in step S202 still contains a large number of voxels. Directly converting it to the geometric commands corresponding to the tool gprMax would generate too many #box definitions. Here, #box refers to the geometric modeling command corresponding to the tool gprMax, mainly used to define the simulation region or medium volume of a 3D cuboid (cube). Therefore, it is necessary to perform connected component analysis in step S203 to merge spatially connected voxels belonging to the same medium in the medium definition library (that is, voxels with the same electromagnetic parameters and spatial connectivity). This allows for the definition of voxels constituting the region using a single geometric command, achieving a significant reduction in the number of geometric commands.

[0057] For each subsurface medium, an improved Union-Find algorithm is used to label connected components according to neighborhood connectivity rules (adjacent voxels in 3D space are considered connected). The algorithm complexity is O(n log n). ( The total number of voxels. (This is the inverse Ackermann function). For example, for the 41600 voxels mentioned above, the connected component labeling processing time can be <0.1 seconds.

[0058] For example, for a three-dimensional voxel model containing 41,600 voxels, M independent connected domains (three-dimensional connected domains) can be identified, such as: a cavity connected domain, including 2,000 connected voxels, corresponding to the roadbed void area; a metal pipeline connected domain, including 600 connected voxels, corresponding to underground pipelines, etc.

[0059] For each connected component, calculate the minimum envelope box (MBB). Then, calculate the smallest axis-aligned cuboid that completely encloses the connected component to obtain the target connected component. This allows you to represent all voxels included in the target connected component using a single #box geometry command corresponding to gprMax. For example, if the target connected component includes voxels with x-ranges of 0~0.2m, y-ranges of 0~0.05m, and z-ranges of 0.05~0.075m, a single #box:000.050.20.050.075 set command can replace the original 2000 geometry commands.

[0060] It is understandable that, in order to ensure the feasibility of the gprMax simulation tool, the total number of geometric commands Mmax≤2000 can be limited. When the number of target connected components exceeds the limit, the target connected components that have the most significant impact on electromagnetic wave propagation are retained according to priority. For connected components that meet the condition of "volume <0.001m³ or voxel number <100", connected component merging or connected component elimination can be performed.

[0061] Taking the three-dimensional region of interest (0.4m×0.4m×0.26m, resolution (size) 10mm, 41600 voxels) corresponding to the above three-dimensional voxel model as an example: the original voxels are 41600. After performing connected component analysis and calculating the minimum envelope box, 1800 connected components are determined, so that the number of #box geometry commands is 1800, the compression rate is 95%, and the processing time is <0.1 seconds / ROI.

[0062] Step S204: Adaptively configure GPR system parameters.

[0063] In some embodiments, GPR system parameters are adaptively calculated and configured based on the scene category, detection depth, and electromagnetic wave propagation theory, including: antenna center frequency, time window, spatial sampling interval, maximum detection depth, etc.

[0064] The antenna center frequency is selected according to the detection depth: for shallow detection (d<1m), f=900-2600MHz is used, with a resolution Δz=3-8cm; for medium-deep detection (1m≤d≤3m), f=400-900MHz is used, with a resolution Δz=8-19cm; for deep detection (d>3m), f=100-400MHz is used, with a resolution Δz=19-75cm.

[0065] Time window calculation: ,in, For the time window (s), Maximum detection depth (m) The average dielectric constant of the soil, For the speed of light ( m / s).

[0066] Lateral sampling interval: ,in, The horizontal sampling interval (m) is... The wavelength (m) in the medium The center frequency of the antenna is (Hz).

[0067] Vertical resolution: This determines the minimum resolvable layer thickness.

[0068] Maximum detection depth estimation: ,in, It is a dielectric loss factor, with deep penetration at low frequencies and high resolution at high frequencies.

[0069] electromagnetic wave velocity in a medium (For non-magnetic media) ≈1). The time window must cover the two-way travel time of the electromagnetic wave to and from the maximum detection depth, with a safety margin. The lateral sampling interval must satisfy the spatial Nyquist sampling theorem.

[0070] Step S205: Extract and label the local region of interest (ROI).

[0071] In some embodiments, such as Figure 5 As shown, the process of extracting and labeling local regions of interest can include four processes, a to d, corresponding to steps S2051 to S2054. In the three-dimensional region of interest corresponding to the three-dimensional voxel model, local regions of interest are extracted, and structured annotation files are generated to provide a standardized dataset for deep learning training.

[0072] Step S2051: Mesh extraction.

[0073] In some embodiments, the local region of interest (ROI) can be determined with a fixed size (0.4m, 0.4m, 0.26m) and a sliding window overlap rate η = 50%. For a 3D voxel model (a primitive voxel model) "3m × 8m × 1.5m", approximately 5460 ROIs can be extracted. Specifically, the sliding window step size is calculated as follows: with an overlap rate of 50%, the step size = ROI size × (1 - 50%); the step size in the x / y direction is... z-axis step size: Number of ROIs that can be extracted in each direction (standard sliding window formula: , For large scene length, (ROI dimensions): x-axis (3m): One; y-axis (8m): One; z-axis (1.5m): There are 100,000 ROIs, of which the total number of ROIs is: indivual.

[0074] It is understandable that, in addition to the extraction methods mentioned above, the following two extraction methods can also be used to adapt to different training needs.

[0075] Step S2052: Goal-oriented extraction.

[0076] In some embodiments, a local region of interest containing underground media (cavities, pipelines, metals, cracks) is automatically identified based on a dielectric constant threshold. Determining based on gradient features. Target-oriented extraction: Automatic identification based on dielectric constant gradient features: ( , )and( There are significant differences in dielectric constant. The algorithm automatically identifies regions where the dielectric constant gradient exceeds the threshold and extracts regions of interest (b-connected region labels) that contain effective local interest.

[0077] Step S2053: Extract key locations.

[0078] In some embodiments, a local region of interest is extracted directionally in key areas such as known locations or pipeline intersections. Key location extraction involves directionally extracting the local region of interest (c-minimum envelope box) based on the known locations confirmed by detection.

[0079] Step S2054: Automatic annotation.

[0080] In some embodiments, for each extracted region of interest (ROI), a standardized JSON format annotation file is automatically generated, containing parameters corresponding to the ROI (optimized after output), such as size, etc. For example, there are 256 ROIs, of which 87 (34%) contain subsurface media and 169 (66%) contain background.

[0081] Step S206: Perform GPR forward simulation calculations to generate a dataset.

[0082] In some embodiments, the tool gprMax is invoked to perform forward modeling calculations and generate A-scan data, B-scan data, and C-scan data.

[0083] gprMax is a GPR forward modeling simulation software based on the finite-difference time-domain (FDTD) method. Its core input carrier for acquiring forward modeling simulation data is a text input file with the .in extension. All simulation rules, electromagnetic parameters, and geometric models are defined through command lines starting with #. It also supports auxiliary inputs such as Python API and external model / waveform files to achieve batch forward modeling of complex scenes.

[0084] The essential core inputs for gprMax include: #domain #dx_dy_dz; #time_window #material; #box; #waveform and #hertzian_dipole; #rx. #domain:xyz defines the total length of the three-dimensional computational domain in the x, y, and z directions, in meters (m). For example, #domain:0.50.30.2 represents a computational space of 0.5m in the x direction, 0.3m in the y direction, and 0.2m in the z direction. For two-dimensional simulations, only one direction's length needs to be set to a single mesh size. #dx_dy_dz:dxdydz defines the spatial mesh step size in the x, y, and z directions, in meters (m), directly determining the simulation accuracy and computational cost. #time_window:T defines the total simulation time window, in seconds (s). For example, #time_window: This represents a total simulation duration of 10 nanoseconds. `#waveform`: Type, amplitude, dominant frequency, and waveform ID. Common waveform types include: Ricker (the preferred choice for GPR forward modeling, with zero-mean bandpass characteristics, closely matching real radar antennas), Gaussian (Gaussian pulse), and Sine (sine wave). Amplitude: The peak amplitude of the waveform. Dominant frequency: The center frequency of the waveform, in Hertz (Hz), e.g., 1GHz is written as 1e9. Waveform ID: A custom unique identifier used to bind the waveform to the excitation source. `#hertzian_dipole`: Polarization direction xyz. Waveform ID: Hertzian dipole point source, most commonly used in GPR forward modeling, suitable for simulating the radiation characteristics of most antennas. Example: `#hertzian_dipole:z0.250.150.18my_ricker`, defines a dipole source polarized in the z-direction with coordinates (0.25, 0.15, 0.18)m, and binds it to the `my_ricker` waveform. #rx:xyz: Single-point receiver, acquires the full-component time-domain waveforms of the electric and magnetic fields at this location, corresponding to A-scan data; Example: #rx:0.250.150.18, sets a common-offset receiver at the same source location, conforming to the transmit-receive co-location antenna mode of most commercial GPRs. #material: , , Material ID: The four core electromagnetic parameters of the custom medium. All underground soil, concrete, water and other media must be defined through this command. σ: Relative permittivity; σ: Conductivity, measured in Siemens per meter (S / m), which determines electromagnetic wave attenuation; Relative magnetic permeability is fixed at 1 for non-magnetic underground media (rock, soil, concrete, water, etc.); Magnetic loss: Fixed at 0 for non-magnetic underground media; Material ID: Custom unique identifier used to assign values ​​to geometric models. Example: #material:60.00110dry_soil, defines dry soil media. =6. Electrical conductivity 0.001 S / m.

[0085] Specifically, define the computational domain and resolution: #domain:0.40.40.26, #dx_dy_dz:0.0020.0020.002, #time_window:30e-9. Define electromagnetic parameters (dielectric constant, conductivity, etc.): #material:6.00.011.00.0soil, #material:1.00.01.00.0void_dry, #material:15.00.051.00.0void_wet, #material:1.05.96e71.00.0iron. Construct the geometric model (using the #box command to define the cuboid region): #box:0000.40.40.26soil, #box:0.10.10.080.30.250.18void_dry. Configure the antenna source and receiver: #waveform:ricker1.0900e6my_ricker, #hertzian_dipole:z0.10.10.25my_ricker, #rx:0.150.10.25.

[0086] gprMax outputs 3D voxel data (a numerical grid representing electromagnetic field variations over time and space). Three scan types can be extracted from this data: A-scan: Signal variation at a single receiver point in the time dimension (amplitude vs. time), corresponding to the evolution sequence of a point in the 3D voxel data over time. B-scan: Multiple A-scans are merged along a straight line (survey line) to form a 2D profile of distance vs. time (depth). It is a slice of the 3D voxel data in one dimension of space plus time. C-scan: At a fixed time (or depth), the signal intensity of the entire plane (e.g., the xy plane) is extracted, forming a horizontal slice. It is often used to display the planar distribution of subsurface media at a certain depth. Therefore, A / B / C scans are different visualization methods for 3D voxel data, namely, A-scan data, B-scan data, and C-scan data.

[0087] It can be seen from the above Figure 2The method shown can construct scene models of various sizes (e.g., 3m × 8m × 1.5m) containing multiple simulation objects of different sizes (e.g., cavities and metal pipes). Adaptive quantization is used to define N media. M connected regions are identified through connected component analysis, with each connected region (0.4m × 0.4m × 0.26m) generating 1800 geometric commands, a 95% reduction compared to voxel-by-voxel modeling (41600 commands). GPR parameters are configured as follows: antenna frequency 900MHz, time window 30ns, B-scan at 40 locations with a 10cm interval. 256 ROIs (0.4m × 0.4m × 0.26m, 50% overlap) are extracted, 87 of which contain underground media. S5. The simulation is executed using gprMax, taking approximately 8 minutes per ROI and generating 256 sets of training data. In contrast, traditional homogeneous modeling methods cannot distinguish underground media with a small number of geometric commands; the voxel-by-voxel method requires 41,600 commands, making simulation infeasible (>24 hours); the method of this invention requires 1,247 commands, with a simulation time of about 8 minutes per ROI, and a correlation coefficient of 0.95 with the voxel-by-voxel method, achieving the best balance between efficiency and accuracy.

[0088] See Figure 6 , Figure 6 The present invention provides a schematic diagram of a forward simulation data generation system 60. The forward simulation data generation system 60 can be divided into a model generation module 601, an electromagnetic parameter mapping processing module 602, a local region of interest measurement module 603, and a simulation data calculation module 604. The descriptions of each functional module are as follows.

[0089] The model generation module 601 is used to construct a three-dimensional voxel model corresponding to the simulation scene. It divides the area into bin regions based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model. The bin regions describe the range of values ​​of the electromagnetic parameters.

[0090] The electromagnetic parameter mapping processing module 602 is used to determine the target connected domain in the three-dimensional voxel model, wherein the voxels contained in the target connected domain have the same numerical characteristics of electromagnetic parameters and the voxels are connected in the space of the three-dimensional voxel model.

[0091] The Local Region of Interest (ROI) module 603 is used to select a ROI from a 3D voxel model and generate geometric commands corresponding to the target connected domains included in the ROI.

[0092] The simulation data calculation module 604 is used to execute the geometric commands and generate forward simulation data.

[0093] The following are examples of electromagnetic parameters of each voxel in a 3D voxel model for different scenarios, as well as various indices of the 3D voxel model.

[0094] Example 1: Low dielectric constant anomaly scenario (road defect detection application) Taking the detection of road cavities and voids as an example, a 3m×8m×1.5m scene model was constructed.

[0095] Scene composition: Surface layer: Asphalt concrete pavement layer, 0.1m thick. =4-6, σ=0.001S / m.

[0096] Base layer: Cement-stabilized crushed stone layer, 0.2m thick. =6-8, σ=0.005S / m.

[0097] Bottom layer: Roadbed soil, =4-8, σ=0.01-0.05S / m.

[0098] Target configuration: Circular cavity: 0.8m in diameter, 0.5m in depth, dry condition. r=1.0, σ≈0.

[0099] Elliptical cavity: major axis 1.2m × minor axis 0.6m, burial depth 0.7m, water-bearing state. =15-25.

[0100] Irregularly shaped cavity: equivalent volume 0.3m³, burial depth 0.9m, partially filled with water. =8-12.

[0101] Void layer: Horizontally extending 2.5m × 1.5m, thickness 0.08m, burial depth 0.3m. =1.0.

[0102] Longitudinal crack: 1.5m long, 20mm wide, 0.4m deep, inflatable. =1.0.

[0103] Electromagnetic parameter contrast: The ratio of the dielectric constant of the target object to that of the background is 1:4 to 1:8, forming a significant negative contrast.

[0104] GPR parameter configuration: antenna frequency 1500-2600MHz, time window 15-20ns, lateral sampling interval 2-5cm, detection depth 0.3-0.8m.

[0105] Simulation results show that: voids exhibit typical hyperbolic reflection characteristics in B-scan images, voided layers appear as horizontal strong reflection stripes, and cracks show weak signals or diffracted waves.

[0106] Example 2: High conductivity target scenario (metal pipeline detection application) Taking the detection of underground metal pipelines in cities as an example, a 3m×6m×2m scene model is constructed.

[0107] Scene composition: Surface layer: Urban road pavement layer, 0.15m thick. =5-7.

[0108] Backfill layer: backfill soil for pipe trenches, =6-10, σ=0.01-0.03S / m.

[0109] Original soil: clay / sand =4-12, σ=0.005-0.02S / m.

[0110] Target configuration: Cast iron water supply pipe: diameter DN150 (outer diameter 170mm), wall thickness 10mm, burial depth 0.8m, σ=5.96×107S / m.

[0111] Steel gas pipe: diameter DN100 (outer diameter 114mm), wall thickness 6mm, burial depth 0.6m, σ=1.04×107S / m.

[0112] Aluminum alloy cable conduit: 80mm diameter, 3mm wall thickness, 0.4m burial depth, σ=3.77×107S / m.

[0113] Copper grounding grid: flat steel 40mm×4mm, buried depth 0.5m, σ=5.8×107S / m.

[0114] Reinforced concrete pipe culvert: inner diameter 0.4m, wall thickness 80mm (inclusive) 12 steel mesh), buried at a depth of 1.2m.

[0115] Electromagnetic parameter contrast: The conductivity of the metallic material σ>10⁵ S / m is significantly lower than that of the surrounding soil (σ≈10⁵ S / m). The difference between 2S / m and 2S / m is 7-9 orders of magnitude.

[0116] Quantization processing: Logarithmic scaling is used for quantization. Metallic materials are assigned independent material grades (grades 11-12) and are not combined with other materials.

[0117] GPR parameter configuration: antenna frequency 400-900MHz, time window 30-50ns, lateral sampling interval 10-20cm, detection depth 0.5-3.0m.

[0118] Simulation results show that metal pipelines produce strong hyperbolic reflections with multiple waves in B-scan; the larger the pipe diameter and the shallower the burial depth, the stronger the reflected signal.

[0119] Example 3: Target scenario with medium dielectric constant (non-metallic pipeline detection application) Taking the detection of municipal non-metallic pipelines as an example, a 3m×6m×1.8m scene model is constructed.

[0120] Scene composition: Surface layer: Asphalt pavement, 0.1m thick. =4-6.

[0121] Pipe layer: Sand cushion layer wrapped around it. =3-5, σ=0.001S / m.

[0122] Bottom layer: original soil, =6-10, σ=0.01S / m.

[0123] Target configuration: PE water supply pipe: outer diameter 110mm, wall thickness 6mm, burial depth 0.8m, pipe wall... =2.3, water in the pipe =81.

[0124] PVC drainage pipe: outer diameter 200mm, wall thickness 5mm, burial depth 1.0m, pipe wall... =3.0, air inside the pipe =1.0.

[0125] PVC communication conduit (multi-hole): outer diameter 100mm, 4-hole structure, burial depth 0.6m. =3.0, including optical cable.

[0126] Concrete drainage pipe: inner diameter 300mm, wall thickness 50mm, burial depth 1.2m. =6-8.

[0127] HDPE gas pipe: outer diameter 63mm, wall thickness 5.8mm, burial depth 0.9m. =2.3, natural gas in the pipeline ≈1.0.

[0128] Fiberglass cable conduit: outer diameter 150mm, wall thickness 8mm, burial depth 0.7m. =4-5.

[0129] Electromagnetic parameter contrast: Dielectric constant contrast is moderate. The range is 0.3-2.0; the medium (water / air) inside the pipe forms a secondary contrast with the pipe wall.

[0130] GPR parameter configuration: antenna frequency 600-900MHz, time window 30-40ns, lateral sampling interval 10-15cm, detection depth 0.5-1.5m.

[0131] Simulation results show that the reflected signal from non-metallic pipelines is weaker than that from metallic pipelines, the signal from water-containing pipelines is stronger, and the signal from empty pipelines is weaker; a comprehensive judgment needs to be made in conjunction with the pipe diameter and burial depth.

[0132] Example 4: Multi-material composite structure scenario (application for building structure inspection) Taking the detection of internal defects in building concrete structures as an example, a 2m×2m×0.5m scene model is constructed.

[0133] Scene composition: Concrete substrate: C30 concrete. =6-9 (dry) / 9-12 (wet), σ=0.01-0.05S / m.

[0134] Reinforcing mesh: 12@150mm double-layer bidirectional reinforcement, with a protective layer thickness of 30-50mm.

[0135] Embedded parts: metal sleeves, junction boxes, etc.

[0136] Target configuration: Internal holes: 50-150mm in diameter, irregularly distributed. =1.0.

[0137] Honeycomb texture: Areas with insufficient density in certain regions. =3-5.

[0138] Reinforcing steel: HRB400, diameter 8- 25, σ = 1 × 10⁷ S / m.

[0139] Corroded steel bars: Local corrosion and expansion, σ decreases to 105-106S / m.

[0140] Cracks: 0.2-2mm wide, 30-200mm deep, filled with water / air.

[0141] Protective layer delamination: thickness 5-20mm, =1.0.

[0142] Electromagnetic parameter characteristics: Multiple materials coexist, with steel bars (high conductivity) and holes (low dielectric constant) and concrete (medium dielectric constant) forming a complex electromagnetic environment.

[0143] Quantitative processing: Reinforcing steel is classified into independent material grades; concrete is divided into 3-4 grades according to moisture content; voids and cracks are classified as low ε. grade.

[0144] GPR parameter configuration: antenna frequency 1500-2600MHz, time window 10-15ns, lateral sampling interval 2-3cm, detection depth 0.1-0.4m.

[0145] Simulation results show that the reinforcing bars exhibit periodic strong reflections, the holes show local anomalies, and the cracks show linear weak signals; the reinforcing bar mesh may produce reverberation interference.

[0146] Example 5: Multi-objective composite scenario (application in complex underground environment) Taking the exploration of underground space at urban road intersections as an example, a large-scale scene model of 5m×8m×2.5m was constructed.

[0147] Scene composition: Road surface structure: asphalt surface layer + water-stabilized base course + subgrade, with a total thickness of 0.5m.

[0148] Pipeline layer: A densely distributed area of ​​multiple pipelines, with a depth of 0.5-2.0m.

[0149] Structures: Inspection wells, valve wells, cable trenches, etc.

[0150] Target configuration (multiple types coexisting): Cast iron water supply pipe: DN200, buried depth 1.0m, σ=5.96×107S / m.

[0151] PE gas pipe: DN110, buried depth 0.8m. =2.3.

[0152] Communication optical cable conduit group: 6-hole PVC pipe, buried at a depth of 0.6m.

[0153] Sewage concrete pipe: DN400, buried depth 1.5m. =7.

[0154] Power cable trench: 0.6m×0.8m cross-section, burial depth 0.7m.

[0155] Inspection well: 1.2m × 1.2m, depth 2.0m, concrete wall. =8, air inside the well =1.0.

[0156] Cavity under the road surface: irregular shape, volume 0.5m3, burial depth 0.8m.

[0157] Pipeline intersection: Water supply pipe and gas pipe cross each other at a distance of 0.3m.

[0158] Scene complexity: It includes a combination of targets with high conductivity (metal tubes), targets with low dielectric constant (voids), and targets with medium dielectric constant (non-metal tubes); there are multiple layers of overlapping and intersecting pipelines.

[0159] GPR parameter configuration: adopts a multi-frequency combined scanning strategy - high frequency (900MHz) to detect shallow details, low frequency (400MHz) to penetrate deep targets; time window 50-70ns.

[0160] Simulation results show that the B-scan image exhibits superposition of multiple reflection signals, with the strongest signal in metal pipelines, followed by voids, and the weakest signal in non-metallic pipelines; complex diffraction waveforms appear at pipeline intersections.

[0161] Example 6: Heterogeneous background scene (application in special geological environments) Taking geological exploration in karst areas and high water-bearing areas as an example, a large-scale scene model of 10m×10m×5m is constructed.

[0162] Scene composition: Surface cover: residual slope soil layer, 0.5-2m thick. =8-15 (change in water content).

[0163] Bedrock: Limestone / Dolomite =6-9, σ=0.001-0.01S / m.

[0164] Solution zone: karst development area The changes were drastic.

[0165] Target configuration: Caves: Irregularly shaped, 2-5m in size, dry =1.0 / water filled =81.

[0166] Dissolution fissures: 0.1-0.5m wide, extending 3-10m deep, partially filled with clay. =15-25.

[0167] Earthen caves: Collapsed cavities in the soil, 1-3m in diameter. =1.0-10.

[0168] Weak interlayer: saturated clay layer, 0.2-0.5m thick. =20-30, σ=0.1S / m.

[0169] Groundwater level: 2-4m deep, above the water level =6-10, below water level =20-30.

[0170] Frozen soil (cold regions): frozen soil =4-6, Melting Soil =15-25.

[0171] Heterogeneous characteristics: The background medium itself exhibits gradient changes or random distribution, rather than a uniform constant; the contrast between the target object and the background varies with depth and water content.

[0172] Quantization processing: Dielectric constant range =1-81, using segmented quantization: low region ( <3) Linear binning, mean adaptive binning in the middle zone (3-15), logarithmic scale binning in the high zone (>15), considering humidity gradient, and setting a gradient material distribution in the vertical direction.

[0173] GPR parameter configuration: antenna frequency 100-400MHz (low frequency deep penetration), time window 100-200ns, lateral sampling interval 20-50cm, detection depth 5-15m.

[0174] Simulation results show that the groundwater level exhibits a continuous strong reflection interface; karst caves produce complex multiple reflections and diffractions; weak interlayers appear as horizontal reflection stripes; and the signal attenuates rapidly with depth.

[0175] like Figure 7 As shown, the present invention provides a method for running Figure 6 The electronic device 70 of the forward simulation data generation system 60 shown may include a memory 701, a processor 702 and a bus, and may also include computer programs stored in the memory 701 and executable on the processor 702, such as the various functional modules of the forward simulation data generation system.

[0176] The memory 701 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 701 can be an internal storage unit of the electronic device 70, such as the portable hard drive of the electronic device 70. In other embodiments, the memory 701 can also be an external storage device of the electronic device 70, such as a plug-in portable hard drive, SmartMediaCard (SMC), SecureDigital (SD) card, FlashCard, etc., equipped on the electronic device 700. Furthermore, the memory 701 can include both internal and external storage units of the electronic device 70. The memory 701 can be used not only to store application software and various types of data installed on the electronic device 70, such as the task distribution code of a forward simulation data generation system, but also to temporarily store data that has been output or will be output.

[0177] In some embodiments, processor 702 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. Processor 702 is the control unit of electronic device 70, connecting various components of the entire electronic device 70 via various interfaces and lines. It executes programs or modules stored in memory 701 (such as the control program of a forward simulation data generation system) and calls data stored in memory 701 to perform various functions and process data of electronic device 70.

[0178] The processor 702 executes the operating system of the electronic device 70 and various installed application programs. The processor 702 executes the application programs to implement the steps in the method for generating forward simulation data of the electronic device in the aforementioned forward simulation data generation system.

[0179] For example, a computer program may be divided into one or more modules, one or more of which are stored in memory 701 and executed by processor 702 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in electronic device 70.

[0180] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module stored in the storage medium includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute part of the functions of the forward simulation data generation method of the electronic device in the forward simulation data generation system of the various embodiments of this application.

[0181] In one embodiment, a storage medium is provided on which a computer program is stored, and when the computer program is executed by a processor, it can also perform the steps described above when the processor executes the computer program.

[0182] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for generating forward simulation data, characterized in that, include: A three-dimensional voxel model corresponding to the simulation scene is constructed. Based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model, a bin region is divided, wherein the bin region describes the value range of the numerical characteristics of the electromagnetic parameters. The electromagnetic parameters of different numerical characteristics corresponding to multiple voxels belonging to one bin region are mapped and converted into electromagnetic parameters of the same numerical characteristics. In the three-dimensional voxel model, a target connected region is determined, wherein the voxels contained in the target connected region have the same numerical characteristics of the electromagnetic parameters and the voxels are connected in space in the three-dimensional voxel model. Select a local region of interest from the three-dimensional voxel model and generate a geometric command corresponding to the target connected component included in the local region of interest; Execute the geometry command to generate forward simulation data.

2. The generation method according to claim 1, characterized in that, The step of dividing the bin regions based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model includes: Multiple segmented regions were determined based on the numerical characteristics of the electromagnetic parameters. The corresponding binning method is used for different segmented regions to determine the binning regions; The electromagnetic parameters of the voxels are mapped and converted into electromagnetic parameters with the same numerical characteristics, wherein the electromagnetic parameters with the same numerical characteristics are determined based on the median or mean of the value range corresponding to the bin regions.

3. The generation method according to claim 2, characterized in that, Based on the distribution of the numerical characteristics of electromagnetic parameters, the binning method includes linear equal-space binning or logarithmic scale binning.

4. The generation method according to claim 1, characterized in that, The step of determining a target connected region in the three-dimensional voxel model, wherein the voxels contained in the target connected region have the same numerical characteristics of electromagnetic parameters and the voxels are spatially connected in the three-dimensional voxel model, includes: On the three-dimensional voxel model, the smallest axis-aligned cuboid that completely encloses the voxels with the same numerical characteristics of the electromagnetic parameters and is connected is identified as the target connected domain.

5. The generation method according to claim 1, characterized in that, Also includes: Merge or remove target connected components that meet preset conditions, wherein the preset conditions are that the volume of the target connected component is less than a preset volume threshold or the number of voxels contained in the target connected component is less than a preset number threshold.

6. The generation method according to claim 1, characterized in that, The local region of interest is determined using at least one of the following methods: mesh extraction, target-oriented extraction, or key location extraction. Mesh extraction determines the region of interest by moving a fixed-size sliding window in the three-dimensional voxel model. Target-oriented extraction determines the local region of interest containing characteristic subsurface media in the three-dimensional voxel model based on the threshold and / or gradient features of the electromagnetic parameters. Key location extraction determines the local region of interest based on known locations in the three-dimensional voxel model.

7. The generation method according to claim 1, characterized in that, Executing the geometric command to generate forward simulation data further includes: The antenna center frequency, time window, spatial sampling interval, longitudinal resolution, and maximum detection depth corresponding to the simulation scenario are determined as system parameters. The system parameters and geometric commands are used as inputs to call the gprMax simulation software to generate the forward modeling simulation data.

8. The generation method according to claim 1, characterized in that, The simulation scenario includes at least one of the following: localized defect scenario, metal pipeline scenario, non-metal pipeline scenario, and composite scenario.

9. A system for generating forward simulation data, characterized in that, The generation system includes: a model generation module, an electromagnetic parameter mapping processing module, a local region of interest measurement module, and a simulation data calculation module, wherein, The model generation module constructs a three-dimensional voxel model corresponding to the simulation scene; The electromagnetic parameter mapping processing module divides the data into bin regions based on the numerical characteristics of the electromagnetic parameters corresponding to each voxel in the three-dimensional voxel model. The bin regions describe the range of values ​​for the numerical characteristics of the electromagnetic parameters. The electromagnetic parameter mapping processing module is also used to map electromagnetic parameters with different numerical characteristics corresponding to multiple voxels belonging to one bin region into electromagnetic parameters with the same numerical characteristics. The local region of interest measurement module is used to determine the target connected region in the three-dimensional voxel model, wherein the voxels contained in the target connected region have the same numerical characteristics of the electromagnetic parameters and the voxels are connected in space in the three-dimensional voxel model. The simulation data calculation module is used to select a local region of interest from the three-dimensional voxel model, generate a geometric command corresponding to the target connected component included in the local region of interest, and execute the geometric command to generate forward simulation data.

10. An electronic device 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 steps of the method for generating forward simulation data as described in any one of claims 1 to 8.

Images