Method and device for three-dimensional reconstruction of underground objects

Abstract

This application relates to the field of non-destructive testing technology for infrastructure, and particularly to a method and apparatus for three-dimensional reconstruction of underground objects. The method includes: constructing a target dataset, a neural network architecture, and a target loss function; constructing a target neural network model for three-dimensional reconstruction of underground targets using the target dataset, neural network architecture, and target loss function; inputting three-dimensional ground-penetrating radar (GPR) signal data into the target neural network model, and outputting the three-dimensional reconstruction result of the underground target corresponding to the three-dimensional GPR signal data through the target neural network model. This application can achieve accurate end-to-end identification of underground pipes and cavities, as well as precise classification of underground target categories. It can also reconstruct the geometry of underground targets, significantly improving the efficiency and accuracy of underground space detection. It can be widely applied in the post-processing stage of GPR data and various engineering scenarios, contributing to improved safety and management efficiency of urban infrastructure.

Description

Technical Field

[0001] This application relates to the field of non-destructive testing technology for infrastructure, and in particular to a method and apparatus for three-dimensional reconstruction of underground objects. Background Technology

[0002] Underground road space is an important component of a city. In urban operation and maintenance, two types of underground road structures require key monitoring: first, various pipelines that ensure the operation of the city; and second, various defects such as voids and cavities caused by construction quality, geological conditions, environmental factors, and abnormal loads.

[0003] In related technologies, traditional detection methods mostly rely on manual analysis and two-dimensional signal interpretation, making it difficult to accurately reconstruct the three-dimensional geometry of underground targets. Especially in complex geological environments, traditional methods suffer from low accuracy and low efficiency. Ground penetrating radar (GPR), as a non-destructive testing technology, has the characteristics of strong penetration capability and sensitive signal response.

[0004] However, ground-penetrating radar signal processing technology still faces many challenges in 3D reconstruction, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency. Moreover, most of these challenges require manual operation, resulting in insufficient detection efficiency and accuracy, which urgently need to be addressed. Summary of the Invention

[0005] This application provides a method and apparatus for three-dimensional reconstruction of underground objects, in order to solve the problems that ground-penetrating radar signal processing technology still faces many challenges in three-dimensional reconstruction, such as insufficient modeling of target shape, classification bias caused by data imbalance, and poor geometric consistency. Moreover, most of these problems require manual operation, resulting in certain deficiencies in detection efficiency and accuracy.

[0006] The first aspect of this application provides a method for three-dimensional reconstruction of underground objects, comprising the following steps: constructing a target dataset, a neural network architecture, and a target loss function; constructing a target neural network model for the purpose of three-dimensional reconstruction of underground target bodies using the target dataset, the neural network architecture, and the target loss function; inputting three-dimensional ground-penetrating radar signal data into the target neural network model, so as to output the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model.

[0007] Optionally, in one embodiment of this application, the construction of the target dataset, neural network architecture, and target loss function includes: acquiring simulated ground-penetrating radar signal data of a preset typical underground target and performing data preprocessing to obtain preprocessed data; constructing a three-dimensional matrix corresponding to the three-dimensional model of the typical underground target and extracting the material spatial distribution label of the typical underground target; and constructing the target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution label.

[0008] Optionally, in one embodiment of this application, the step of constructing a target neural network model for the purpose of three-dimensional reconstruction of an underground target using the target dataset, the neural network architecture, and the target loss function includes: inputting the preprocessed data into an initial neural network model to output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data, thereby constructing the target neural network model.

[0009] Optionally, in one embodiment of this application, the step of outputting the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model includes: identifying the type information and geometric information of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the mapping model in the neural network architecture; and determining the three-dimensional reconstruction result based on the type information and the geometric information.

[0010] Optionally, in one embodiment of this application, the step of constructing a target neural network model for the purpose of three-dimensional reconstruction of an underground target using the target dataset, the neural network architecture, and the target loss function includes: constructing a target three-dimensional convolutional neural network architecture based on the target encoder-decoder architecture; and determining the neural network architecture based on the target three-dimensional convolutional neural network architecture to construct the target neural network model.

[0011] Optionally, in one embodiment of this application, the target loss function is:

[0012] ,

[0013] in, Indicates the loss value. Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the regularization term of the total variation.

[0014] A second aspect of this application provides a three-dimensional reconstruction device for underground objects, comprising: a first construction module for constructing a target dataset, a neural network architecture, and a target loss function; a second construction module for constructing a target neural network model for the purpose of three-dimensional reconstruction of an underground target using the target dataset, the neural network architecture, and the target loss function; and a reconstruction module for inputting three-dimensional ground-penetrating radar signal data into the target neural network model, so as to output the three-dimensional reconstruction result of the underground target corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model.

[0015] Optionally, in one embodiment of this application, the first construction module includes: an acquisition unit, configured to acquire simulated ground-penetrating radar signal data of a preset typical underground target and perform data preprocessing to obtain preprocessed data; an extraction unit, configured to construct a three-dimensional matrix corresponding to the three-dimensional model of the typical underground target and extract the material spatial distribution label of the typical underground target; and a first construction unit, configured to construct the target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution label.

[0016] Optionally, in one embodiment of this application, the second construction module includes: a second construction unit, used to input the preprocessed data into an initial neural network model to output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data, so as to construct the target neural network model.

[0017] Optionally, in one embodiment of this application, the reconstruction module includes: an identification unit, configured to identify the type information and geometric information of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through a mapping model in the neural network architecture; and a determination unit, configured to determine the three-dimensional reconstruction result based on the type information and the geometric information.

[0018] Optionally, in one embodiment of this application, the second construction module includes: a third construction unit, configured to construct a target three-dimensional convolutional neural network architecture based on the target encoder-decoder architecture; and a fourth construction unit, configured to determine the neural network architecture based on the target three-dimensional convolutional neural network architecture to construct the target neural network model.

[0019] Optionally, in one embodiment of this application, the target loss function is:

[0020] ,

[0021] in, Indicates the loss value. Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the regularization term of the total variation.

[0022] A third aspect of this application 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 program to implement the three-dimensional reconstruction method for underground objects as described in the above embodiments.

[0023] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for three-dimensional reconstruction of underground objects.

[0024] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, is used to implement the above-described method for three-dimensional reconstruction of underground objects.

[0025] This application embodiment can construct a target neural network model based on the target dataset, neural network architecture, and target loss function. Three-dimensional ground-penetrating radar (GPR) signal data is input into the target neural network model, thereby outputting the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional GPR signal data. This achieves the goal of training the initial neural network model with highly accurate GPR simulation signal data generated by simulation tools through engineering scenario surveys, improving the practical application capability of the target neural network model. Furthermore, by combining a three-dimensional deep convolutional neural network architecture, residual connections and skip connections effectively enhance the three-dimensional reconstruction performance of the target neural network. Further, this application can further enhance the geometric consistency of the three-dimensional reconstruction results by combining a composite loss function of Dice loss and total variation regularization. Ultimately, this application can achieve end-to-end accurate identification of underground pipes and cavities, and while accurately classifying underground target body categories, it can also reconstruct the geometry of underground target bodies, significantly improving the efficiency and accuracy of underground space detection. It can be widely applied in the data post-processing stage of two-dimensional or three-dimensional GPR, as well as in various engineering scenarios such as urban infrastructure management, road monitoring, and underground defect assessment, contributing to improving the safety and management efficiency of urban infrastructure. This solves the problems that ground-penetrating radar signal processing technology still faces in 3D reconstruction, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency. In addition, most of these problems require manual operation, resulting in certain deficiencies in detection efficiency and accuracy.

[0026] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0027] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0028] Figure 1 This is a flowchart of a three-dimensional reconstruction method for underground objects according to an embodiment of this application;

[0029] Figure 2 This is a flowchart illustrating a three-dimensional reconstruction of underground pipes and cavities according to an embodiment of this application;

[0030] Figure 3 This is a schematic diagram of a neural network architecture according to an embodiment of this application;

[0031] Figure 4 This is a schematic diagram of the structure of the three-dimensional reconstruction device for underground objects provided according to the embodiments of this application;

[0032] Figure 5 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application.

[0033] Figure label:

[0034] 10-Three-dimensional reconstruction device for underground objects: 100-First building module, 200-Second building module and 300-Reconstruction module; 501-Memory, 502-Processor and 503-Communication interface. Detailed Implementation

[0035] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0036] The following description, with reference to the accompanying drawings, outlines a method and apparatus for three-dimensional reconstruction of underground objects according to embodiments of this application. Addressing the challenges faced by ground-penetrating radar (GPR) signal processing technology in three-dimensional reconstruction, as mentioned in the background section, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency, and the fact that most of these methods require manual operation, resulting in deficiencies in detection efficiency and accuracy, this application provides a method for three-dimensional reconstruction of underground objects. In this method, a target neural network model can be constructed based on a target dataset, a neural network architecture, and a target loss function. Three-dimensional GPR signal data is then input into the target neural network model to output the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional GPR signal data. This invention enables the use of simulation tools to generate highly accurate ground-penetrating radar (GPR) simulation signal data for training an initial neural network model through engineering scenario surveys, thereby improving the practical application capability of the target neural network model. Furthermore, by combining a 3D deep convolutional neural network architecture with residual and skip connections, the 3D reconstruction performance of the target neural network is effectively enhanced. Additionally, this application further enhances the geometric consistency of the 3D reconstruction results by incorporating a composite loss function combining Dice loss and total variation regularization. Ultimately, this application achieves end-to-end accurate identification of underground pipes and cavities, accurately classifying underground target categories while reconstructing their geometry, significantly improving the efficiency and accuracy of underground space detection. It can be widely applied in the post-processing stage of 2D or 3D GPR data, as well as in various engineering scenarios such as urban infrastructure management, road monitoring, and underground defect assessment, contributing to improved safety and management efficiency of urban infrastructure. This addresses the challenges still faced by GPR signal processing technology in 3D reconstruction, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency. Moreover, much of this technology requires manual operation, resulting in insufficient detection efficiency and accuracy.

[0037] Specifically, Figure 1 This is a flowchart of a method for three-dimensional reconstruction of underground objects provided in an embodiment of this application.

[0038] like Figure 1 As shown, the method for three-dimensional reconstruction of underground objects includes the following steps:

[0039] In step S101, the target dataset, neural network architecture, and target loss function are constructed.

[0040] In some embodiments, when performing 3D reconstruction of underground objects, this application may, but is not limited to, model the implementation process as a model for practical application. When constructing this model, it can be based on a certain deep learning neural network architecture, trained and tested using a target dataset, and a certain target loss function can be designed to improve the accuracy of the final 3D reconstruction result.

[0041] In this context, the target dataset can be understood as a collection of various types of data used in modeling, which can be used to train and test the model.

[0042] The following section provides a further explanation of the construction process of the target dataset in the embodiments of this application.

[0043] Optionally, in one embodiment of this application, constructing a target dataset, a neural network architecture, and a target loss function includes: acquiring simulated ground-penetrating radar signal data of a preset typical underground target and performing data preprocessing to obtain preprocessed data; constructing a three-dimensional matrix corresponding to a three-dimensional model of a typical underground target and extracting material spatial distribution labels of the typical underground target; and constructing a target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution labels.

[0044] In actual implementation, when constructing the target dataset, this application can first obtain the simulated ground-penetrating radar signal data of a preset typical underground target, and then perform data preprocessing to obtain the preprocessed data as part of the target dataset.

[0045] Here, the preset typical underground target objects can be understood as some relatively typical underground target objects that have been determined in advance, such as cavities and underground pipes. In actual applications, underground objects in different places may vary, and the underground objects that need to be reconstructed in 3D will also be different. Therefore, the preset typical underground target objects can be determined by professionals in this field according to the actual situation, thereby improving the practical application capability of the model and the accuracy of the 3D reconstruction results.

[0046] For example, Figure 2 This is a flowchart illustrating a three-dimensional reconstruction of underground pipes and cavities according to an embodiment of this application. Figure 2 As shown, this application can first combine engineering scenario surveys to summarize the rules of pipeline and defect target modeling, use gprMax to carry out radar signal simulation of typical target bodies of multiple survey lines, obtain ground penetrating radar signals and corresponding model size information of underground pipelines and cavities, obtain two-dimensional B-scan signals and combine them to obtain three-dimensional simulated ground penetrating radar signal data C-scan data, and then perform normalization preprocessing to unify it into the [0,1] interval to obtain preprocessed data.

[0047] It should be noted that the ground-penetrating radar signal data in the embodiments of this application can be generated based on open-source simulation software and actual experiments, etc. The specific acquisition method can be selected or adjusted according to the actual situation. The embodiments of this application are only illustrative and do not impose specific limitations.

[0048] Furthermore, embodiments of this application can simultaneously extract the spatial distribution of materials such as soil, metal pipes, plastic pipes, concrete pipes, and cavities from the input file as different labels.

[0049] Finally, this application can convert the three-dimensional models of these typical underground targets into corresponding three-dimensional matrices. By unifying the preprocessed data, material spatial distribution labels, and three-dimensional matrices into data of a specified dimension, a target dataset can be formed. The specific dimensions can be selected or adjusted by those skilled in the art; this is merely illustrative and not intended to impose specific limitations. It should also be noted that different preprocessed data, material spatial distribution labels, and three-dimensional matrices will result in different target datasets; the embodiments in this application are merely illustrative and not intended to impose specific limitations.

[0050] Step S102: Construct a target neural network model for the purpose of 3D reconstruction of underground target objects using the target dataset, neural network architecture, and target loss function. The target loss function can be, but is not limited to, expressed as:

[0051]

[0052] Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the regularization term of the total variation.

[0053] Based on the descriptions of other embodiments, it is understood that, for practical application purposes, the embodiments of this application can utilize the target dataset, neural network architecture, and target loss function to model the process of three-dimensional reconstruction of underground objects into a certain model.

[0054] In practical implementation, this application can utilize the target dataset, neural network architecture, and target loss function to construct a target neural network model for the purpose of 3D reconstruction of underground target bodies. This target neural network model can vary depending on the target dataset, neural network architecture, and target loss function.

[0055] In this application embodiment, the target loss function can be constructed using, but is not limited to, an improved Dice loss function. Specifically, this application embodiment can use an improved Dice loss function combined with a total variation regularization term to optimize the target neural network model.

[0056] For example, the Dice loss calculation formula can be expressed, but is not limited to, as follows:

[0057]

[0058] in, It is the first The prediction results of individual factors, and It is the first The true results of individual factors, This represents the total number of voxels. To enhance the geometric continuity of the prediction results, embodiments of this application may also add a total variation regularization term to the loss function. Considering that the prediction results are discrete matrices, the total variation regularization term in embodiments of this application may, but is not limited to, be expressed as:

[0059]

[0060] in, This refers to the position as Voxel prediction values. The final objective loss function can be, but is not limited to, expressed as:

[0061]

[0062] in, Indicates the loss value. Indicates the first Loss values ​​for each category, This is the number of categories. In actual use, the label matrix can be one-hot encoded, and then the loss value can be calculated for each category. For example, in the embodiment of this application, it can be taken as 5. is a scaling factor. To avoid the regularization term being too large and adversely affecting training, its value can be set to the number of all elements involved in the calculation. The formula can be, but is not limited to, expressed as follows:

[0063]

[0064] in, , and These represent the dimensions of the three dimensions of the label matrix.

[0065] It should be noted that the prediction results in this embodiment refer to the material distribution in three-dimensional space corresponding to the C-scan data of the three-dimensional simulated ground-penetrating radar signal during the construction of the target neural network model.

[0066] In this embodiment, the target loss function of the target neural network model adopts the Dice loss function with added total variation regularization constraint, and the weight of the regularization term is designed as the reciprocal of the number of elements in the output matrix, thereby optimizing the target neural network model to enhance the accuracy and geometric consistency of the 3D reconstruction of the target neural network model.

[0067] Optionally, in one embodiment of this application, a target neural network model for the purpose of three-dimensional reconstruction of underground target bodies is constructed using a target dataset, a neural network architecture, and a target loss function. This includes: inputting preprocessed data into an initial neural network model to output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data, thereby constructing the target neural network model.

[0068] Based on the descriptions of other embodiments, it is understood that the embodiments of this application can utilize the target dataset, neural network architecture, and target loss function to construct a target neural network model. During the training of the target neural network model, certain prediction results will be obtained, namely, the distribution of materials in three-dimensional space corresponding to the three-dimensional simulated ground-penetrating radar signal C-scan data.

[0069] In actual implementation, please refer to Figure 2 ,like Figure 2 As shown, this application can input the preprocessed data, i.e. the normalized three-dimensional simulated ground-penetrating radar signal C-scan data, into the initial neural network model, and then output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data through the initial neural network model, i.e. the distribution of the material corresponding to the three-dimensional simulated ground-penetrating radar signal C-scan data in three-dimensional space, thereby completing the training of the initial neural network model to construct the target neural network model.

[0070] Optionally, in one embodiment of this application, a target neural network model for the purpose of three-dimensional reconstruction of underground target objects is constructed using a target dataset, a neural network architecture, and a target loss function, including: constructing a target three-dimensional convolutional neural network architecture based on a target encoding-decoding architecture; and determining a neural network architecture based on the target three-dimensional convolutional neural network architecture to construct a target neural network model.

[0071] As one possible implementation, the neural network architecture used to construct the target neural network model in this application embodiment may be, but is not limited to, a target "U"-shaped three-dimensional convolutional neural network architecture based on a target encoder-decoder architecture. The target encoder-decoder architecture can be understood as a relevant encoder-decoder architecture capable of satisfying the construction of the target neural network model, such as the U-Net encoder-decoder structure or the SegNet encoder-decoder structure in a convolutional neural network (CNN). This application embodiment is merely illustrative and does not impose specific limitations.

[0072] For example, Figure 3 This is a schematic diagram of a neural network architecture according to an embodiment of this application, such as... Figure 3 As shown, the target "U"-shaped 3D convolutional neural network architecture based on the target encoder-decoder architecture in this embodiment of the application may, but is not limited to, consist of "conv", "n", "lrelu", "drop", "upscale", and "concat". Here, "conv", "n", "lrelu", "drop", "upscale", and "concat" represent a 3D convolutional layer, an instance normalization layer, a leaky correction linear unit (Leaky ReLU) activation layer, a random dropout layer, an upsampling layer, and a cascaded operation, respectively. Furthermore, the "+" in the illustration indicates a continuous combination of the various operations.

[0073] Furthermore, the neural network architecture in this embodiment also includes skip connections and long-range parallel connections. Skip connections refer to a short-circuit structure that directly adds the transformed tensor to the input tensor. This embodiment uses two types of convolutional modules, both with a kernel size of 3×3×3 and padding of 1×1×1. The strides of "conv" and "conv" are 1×1×1 and 2×2×2 respectively. Therefore, the first type does not change the size of the input tensor, but the second type reduces the size by half.

[0074] The input tensor has 1 channel, which is then processed by the initial convolutional layer to increase the number of channels. Then, using 5 encoders, the number of tensor channels was changed to Simultaneously, the size of the input is gradually reduced to half the size of the input layer by layer. The decoder then restores the feature map to twice the original input size layer by layer, while reducing the number of channels. Finally, after the output is restored to the original size, the softmax activation function can be used for normalization to obtain the prediction result, thus completing the construction of the target neural network model.

[0075] Step S103: Input the three-dimensional ground-penetrating radar signal data into the target neural network model, so as to output the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model.

[0076] In other embodiments, after the target neural network model is constructed, the embodiments of this application can use the target neural network model to complete the three-dimensional reconstruction of the real underground target.

[0077] In this context, underground target bodies can be understood as underground objects that need to be reconstructed in three dimensions. For example, in urban operation and maintenance, it is necessary to focus on monitoring various pipelines that ensure the operation of the city and various defects such as voids and cavities caused by construction quality, geological conditions, environmental factors, and abnormal loads. These are the two types of underground road target bodies: underground pipelines and cavities.

[0078] Specifically, in this embodiment, the three-dimensional ground-penetrating radar signal data corresponding to the underground target body to be reconstructed can be input into the target neural network model, and then the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data can be output through the target neural network model.

[0079] The 3D reconstruction results include, but are not limited to, the material category of each location with underground space (similar to the output of 3D segmentation). Furthermore, by using the material category of each location, this embodiment of the application can identify the type of underground target body at each location and determine the geometry of the underground target body.

[0080] Optionally, in one embodiment of this application, the output of the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model includes: identifying the type information and geometric information of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the mapping model in the neural network architecture; and determining the three-dimensional reconstruction result based on the type information and geometric information.

[0081] In some embodiments, in order to improve the accuracy of the three-dimensional reconstruction results, this application can also build a mapping model between radar signals and the distribution of target categories in underground space in the neural network architecture used to construct the target neural model, so as to use the mapping model to identify the type information and geometric information of underground target bodies corresponding to the three-dimensional ground-penetrating radar signal data, thereby realizing the accurate reconstruction of the type and geometric information of underground target bodies.

[0082] Therefore, the embodiments of this application can perform geometric three-dimensional reconstruction of the underground target body while identifying the category of each voxel of the underground target body. Thus, the three-dimensional reconstruction of the underground target body in the embodiments of this application can essentially be understood as voxel reconstruction.

[0083] In this context, a voxel refers to the smallest unit used to represent and quantify a 3D object or space in the context of 3D reconstruction. A voxel has its own location and typically has one or more attributes related to the reconstruction task, such as density, color, and material.

[0084] The 3D reconstruction method for underground objects proposed in this application can construct a target neural network model based on a target dataset, neural network architecture, and target loss function. 3D ground-penetrating radar (GPR) signal data is input into the target neural network model, thereby outputting the 3D reconstruction result of the underground target body corresponding to the GPR signal data. This achieves improved practical application capability of the target neural network model by using highly accurate GPR simulation signal data generated through engineering scenario surveys and simulation tools to train the initial neural network model. Furthermore, by combining a 3D deep convolutional neural network architecture with residual connections and skip connections, the 3D reconstruction performance of the target neural network is effectively improved. Further, this application can further enhance the geometric consistency of the 3D reconstruction results by combining a composite loss function of Dice loss and total variation regularization. Ultimately, this application can achieve end-to-end accurate identification of underground pipes and cavities, and while accurately classifying underground target body categories, it can also reconstruct the geometry of underground target bodies, significantly improving the efficiency and accuracy of underground space detection. It can be widely applied in the data post-processing stage of 2D or 3D GPR, as well as in various engineering scenarios such as urban infrastructure management, road monitoring, and underground disease assessment, contributing to improved safety and management efficiency of urban infrastructure. This solves the problems that ground-penetrating radar signal processing technology still faces in 3D reconstruction, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency. In addition, most of these problems require manual operation, resulting in certain deficiencies in detection efficiency and accuracy.

[0085] Next, the three-dimensional reconstruction device for underground objects according to the embodiments of this application is described with reference to the accompanying drawings.

[0086] Figure 4 This is a schematic diagram of the structure of the three-dimensional reconstruction device for underground objects according to an embodiment of this application.

[0087] like Figure 4 As shown, the underground object three-dimensional reconstruction device 10 includes: a first construction module 100, a second construction module 200, and a reconstruction module 300.

[0088] The first building module 100 is used to build the target dataset, neural network architecture, and target loss function.

[0089] The second building module 200 is used to construct a target neural network model for the purpose of three-dimensional reconstruction of underground target objects using the target dataset, neural network architecture and target loss function.

[0090] The reconstruction module 300 is used to input three-dimensional ground-penetrating radar signal data into the target neural network model, so as to output the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model.

[0091] Optionally, in one embodiment of this application, the first construction module 100 includes: an acquisition unit, an extraction unit, and a first construction unit.

[0092] The acquisition unit is used to acquire simulated ground-penetrating radar signal data of a preset typical underground target and perform data preprocessing to obtain preprocessed data.

[0093] The extraction unit is used to construct the three-dimensional matrix corresponding to the three-dimensional model of a typical underground target and extract the material spatial distribution labels of the typical underground target.

[0094] The first building unit is used to construct the target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution labels.

[0095] Optionally, in one embodiment of this application, the second construction module 200 includes: a second construction unit, used to input preprocessed data into an initial neural network model to output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data, so as to construct a target neural network model.

[0096] Optionally, in one embodiment of this application, the reconstruction module 300 includes: an identification unit and a determination unit.

[0097] The identification unit is used to identify the type and geometric information of underground targets corresponding to the three-dimensional ground-penetrating radar signal data through the mapping model in the neural network architecture.

[0098] The determination unit is used to determine the 3D reconstruction result based on type and geometric information.

[0099] Optionally, in one embodiment of this application, the second building module 200 includes a third building unit and a fourth building unit.

[0100] The third building unit is used to construct the target three-dimensional convolutional neural network architecture based on the target encoder-decoder architecture.

[0101] The fourth building unit is used to determine the neural network architecture based on the target 3D convolutional neural network architecture in order to build the target neural network model.

[0102] Optionally, in one embodiment of this application, the target loss function may be, but is not limited to, expressed as:

[0103]

[0104] Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the regularization term of the total variation.

[0105] It should be noted that the foregoing explanation of the embodiment of the three-dimensional reconstruction method for underground objects also applies to the three-dimensional reconstruction device for underground objects in this embodiment, and will not be repeated here.

[0106] The 3D reconstruction device for underground objects proposed in this application can construct a target neural network model based on a target dataset, neural network architecture, and target loss function. 3D ground-penetrating radar (GPR) signal data is input into the target neural network model, thereby outputting the 3D reconstruction result of the underground target body corresponding to the GPR signal data. This achieves the improvement of the practical application capability of the target neural network model by using highly accurate GPR simulation signal data generated through engineering scenario surveys and simulation tools to train the initial neural network model. Furthermore, by combining a 3D deep convolutional neural network architecture with residual connections and skip connections, the 3D reconstruction performance of the target neural network is effectively improved. Further, this application can further enhance the geometric consistency of the 3D reconstruction results by combining a composite loss function of Dice loss and total variation regularization. Ultimately, this application can achieve end-to-end accurate identification of underground pipes and cavities, and while accurately classifying underground target bodies, it can also reconstruct the geometry of underground target bodies, significantly improving the efficiency and accuracy of underground space detection. It can be widely applied in the data post-processing stage of 2D or 3D GPR, as well as in various engineering scenarios such as urban infrastructure management, road monitoring, and underground disease assessment, contributing to the improvement of urban infrastructure safety and management efficiency. This solves the problems that ground-penetrating radar signal processing technology still faces in 3D reconstruction, such as insufficient target shape modeling, classification bias caused by data imbalance, and poor geometric consistency. In addition, most of these problems require manual operation, resulting in certain deficiencies in detection efficiency and accuracy.

[0107] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0108] The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0109] When the processor 502 executes the program, it implements the three-dimensional reconstruction method for underground objects provided in the above embodiments.

[0110] Furthermore, electronic devices also include:

[0111] Communication interface 503 is used for communication between memory 501 and processor 502.

[0112] The memory 501 is used to store computer programs that can run on the processor 502.

[0113] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0114] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0115] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0116] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0117] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for three-dimensional reconstruction of underground objects.

[0118] This application also provides a computer program product, including a computer program that can run computer instructions. When the computer instructions are executed by a processor, they implement the three-dimensional reconstruction method for underground objects provided in this application.

[0119] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," 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 this application. 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. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0120] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0121] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0122] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0123] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0124] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0125] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0126] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for three-dimensional reconstruction of underground objects, characterized in that, Includes the following steps: Construct the target dataset, neural network architecture, and target loss function; A target neural network model for the purpose of three-dimensional reconstruction of underground target bodies is constructed using the target dataset, the neural network architecture, and the target loss function. The three-dimensional ground-penetrating radar signal data is input into the target neural network model, so that the target neural network model outputs the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data, wherein the three-dimensional reconstruction result includes the material category of each location in the underground space; The construction of the target dataset, neural network architecture, and target loss function includes: acquiring simulated ground-penetrating radar signal data of a preset typical underground target and its corresponding model size information to obtain a two-dimensional B-scan signal; combining the two-dimensional B-scan signal to obtain three-dimensional simulated ground-penetrating radar signal C-scan data; performing normalization preprocessing on the C-scan data to obtain the preprocessed data; constructing a three-dimensional matrix corresponding to the three-dimensional model of the typical underground target and extracting the material spatial distribution label of the typical underground target; and constructing the target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution label. The step of constructing a target neural network model for the purpose of three-dimensional reconstruction of underground target objects using the target dataset, the neural network architecture, and the target loss function includes: constructing a target three-dimensional convolutional neural network architecture based on the target encoder-decoder architecture; and determining the neural network architecture based on the target three-dimensional convolutional neural network architecture to construct the target neural network model. The step of outputting the 3D reconstruction result of the underground target body corresponding to the 3D ground-penetrating radar signal data through the target neural network model includes: identifying the type information, geometric information, and location information of the underground target body corresponding to the 3D ground-penetrating radar signal data through the mapping model in the neural network architecture; and determining the 3D reconstruction result based on the type information, the geometric information, and the location information. The target loss function is: ， ， Indicates the loss value. Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the total variation regularization term; This refers to the position being ( Voxel prediction values; , and These represent the dimensions of the three dimensions of the label matrix; The neural network architecture includes a 3D convolutional layer, an instance normalization layer, a leakage correction linear unit activation layer, a random dropout layer, and an upsampling layer, which are combined through cascading operations. The neural network architecture also includes skip connections and long-distance parallel connections, where skip connections are short-circuit structures used to directly add the transformed tensor to the input tensor.

2. The method for three-dimensional reconstruction of underground objects according to claim 1, characterized in that, The construction of a target neural network model for the purpose of 3D reconstruction of underground target bodies using the target dataset, the neural network architecture, and the target loss function includes: The preprocessed data is input into the initial neural network model to output the material category of each voxel in the underground space corresponding to the simulated ground-penetrating radar signal data, so as to construct the target neural network model.

3. A three-dimensional reconstruction device for underground objects, characterized in that, include: The first building module is used to construct the target dataset, neural network architecture, and target loss function; The second construction module is used to construct a target neural network model for the purpose of three-dimensional reconstruction of underground target objects using the target dataset, the neural network architecture, and the target loss function. The reconstruction module is used to input three-dimensional ground-penetrating radar signal data into the target neural network model, so as to output the three-dimensional reconstruction result of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the target neural network model, wherein the three-dimensional reconstruction result includes the material category of each location in the underground space; The first construction module includes: an acquisition unit, used to acquire simulated ground-penetrating radar signal data of a preset typical underground target and its corresponding model size information, to obtain a two-dimensional B-scan signal, and to combine the two-dimensional B-scan signal to obtain three-dimensional simulated ground-penetrating radar signal C-scan data, and to perform normalization preprocessing on the C-scan data to obtain the preprocessed data; an extraction unit, used to construct a three-dimensional matrix corresponding to the three-dimensional model of the typical underground target, and to extract the material spatial distribution label of the typical underground target; and a first construction unit, used to construct the target dataset based on the preprocessed data, the three-dimensional matrix, and the material spatial distribution label. The second construction module includes: a second construction unit, used to construct a target three-dimensional convolutional neural network architecture based on the target encoder-decoder architecture; and to determine the neural network architecture based on the target three-dimensional convolutional neural network architecture to construct the target neural network model; The reconstruction module includes: an identification unit, used to identify the type, geometric, and location information of the underground target body corresponding to the three-dimensional ground-penetrating radar signal data through the mapping model in the neural network architecture; and a determination unit, used to determine the three-dimensional reconstruction result based on the type, geometric, and location information. The target loss function is: ， ， in, Indicates the loss value. Indicates the first Loss values ​​for each category, It is the number of categories. It is a scale factor. It is the total variation regularization term; This refers to the position being ( Voxel prediction values; , and These represent the dimensions of the three dimensions of the label matrix; The neural network architecture includes a 3D convolutional layer, an instance normalization layer, a leakage correction linear unit activation layer, a random dropout layer, and an upsampling layer, which are combined through cascading operations. The neural network architecture also includes skip connections and long-distance parallel connections, where skip connections are short-circuit structures used to directly add the transformed tensor to the input tensor.

4. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the three-dimensional reconstruction method for underground objects as described in any one of claims 1-2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the three-dimensional reconstruction method for underground objects as described in any one of claims 1-2.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed, it is used to implement the three-dimensional reconstruction method for underground objects as described in any one of claims 1-2.

Images