A geological radar translation method and system based on deep learning

Abstract

The present invention provides a geological radar translation method and system based on deep learning, which belongs to the technical field of geological radar data processing. The translation method includes the following processes: preprocessing the geological radar raw data to obtain the geological radar waveform, encoding the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform; decoding each element one by one through the sequence information in the wave vector to generate a new target model sequence, and translating the target model sequence into a dielectric constant model diagram through a model generation function; the present invention can translate various dielectric constant models of tunnel linings containing complex defects, and effectively reconstruct complex tunnel lining defects with clear boundaries.

Description

Technical Field

[0001] The present invention relates to the technical field of geological radar data processing, and in particular to a geological radar translation method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Underground structure detection has always been an important problem in the fields of geology, engineering and environmental science. In recent years, with the advancement of science and technology, geological radar has played an increasingly important role in disease detection and underground exploration. Ground penetrating radar inversion is a complex process of inferring the location of hidden targets and obtaining physical parameters based on scanning data. This technology is crucial for road and tunnel maintenance and traffic safety.

[0004] During tunnel operation, due to the influence of complex factors such as geology, hydrology, and construction quality, more than 75% of highway tunnels have lining cracks, water leakage, abnormal deformation, lining back debonding, etc. These diseases not only affect the service life and safety of the tunnel, but the reinforcement costs are often high, sometimes even exceeding the construction costs, resulting in significant economic losses and adverse social impacts.

[0005] The current inversion methods are divided into two categories: traditional and deep learning-based. Traditional methods include tomography and full waveform inversion, while deep learning-based methods have been the forefront of research in recent years. However, the inventors found that the existing methods have some limitations. Many models have complex structures and large computational complexity, and most of them ignore the refined features of ground penetrating radar images, resulting in low inversion performance. Summary of the invention

[0006] In order to address the deficiencies of the prior art, the present invention provides a geological radar translation method and system based on deep learning, which can translate various dielectric constant models of tunnel linings containing complex defects and effectively reconstruct complex tunnel lining defects with clear boundaries.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a geological radar translation method based on deep learning.

[0009] A geological radar translation method based on deep learning includes the following processes:

[0010] Preprocessing the geological radar raw data to obtain a geological radar waveform, encoding the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform;

[0011] Through the sequence information in the wave vector, each element is decoded to generate a new target model sequence, and the target model sequence is translated into a dielectric constant model diagram through the model generation function.

[0012] As a further limitation of the first aspect of the present invention, preprocessing the raw data of the geological radar includes:

[0013] Load geological radar data through the radar reading function, read VTI data through the model reading function and convert it into model label data, and adjust the size of the model label data to match the size of the geological radar data;

[0014] The original geological radar data is resampled to reduce the data volume, the geological radar data after the resampling operation is normalized, and the normalized geological radar data and the resized model label data are saved in the set format.

[0015] As a further limitation of the first aspect of the present invention, encoding the geological radar waveform into a wave vector comprises:

[0016] The geological radar waveform sequence is input into the encoder, which includes a waveform embedding layer and a GRU network. The encoder adjusts the dimension of the input geological radar waveform and converts the data type. The converted waveform data is mapped into a wave vector sequence through the waveform embedding layer, where the wave vector of each data serves as the time step of the input sequence.

[0017] As a further limitation of the first aspect of the present invention, the GRU network calculates a value of a reset gate based on a current input and a hidden state at a previous time step, and calculates a value of an update gate based on an input and a hidden state at a previous time step;

[0018] Use the calculated reset gate to synthesize the candidate hidden state considering the old hidden state and the new input, use the update gate to control the fusion of the old hidden state and the new candidate hidden state, calculate the hidden state of the current time step, and continue to pass it to the next time step;

[0019] The hidden state of each time step captures the data of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state, which is used as the characteristic wave vector of the input sequence.

[0020] As a further limitation of the first aspect of the present invention, decoding the sequence information in the wave vector element by element to generate a new target model sequence includes:

[0021] The wave vector and label data output by the encoder are input into the decoder, which includes a model embedding layer, a GRU network, and a fully connected layer;

[0022] The decoder adjusts the dimension of the input label data and converts the data type, maps the converted label data into a vector sequence through the model embedding layer, and adjusts the dimension of the wave vector to generate a tensor that matches the decoder input for use by the decoder at each time step;

[0023] The adjusted wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step;

[0024] The decoder gradually generates the target model sequence according to the above process.

[0025] In a second aspect, the present invention provides a geological radar translation system based on deep learning.

[0026] A geological radar translation system based on deep learning, comprising:

[0027] The encoding module is configured to: pre-process the geological radar raw data to obtain a geological radar waveform, and encode the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform;

[0028] The decoding module is configured to: decode element by element through the sequence information in the wave vector to generate a new target model sequence, and translate the target model sequence into a dielectric constant model diagram through a model generation function.

[0029] As a further limitation of the second aspect of the present invention, in the encoding module, encoding the geological radar waveform into a wave vector includes:

[0030] The geological radar waveform sequence is input into the encoder, which includes a waveform embedding layer and a GRU network. The encoder adjusts the dimension of the input geological radar waveform and converts the data type. The converted waveform data is mapped into a wave vector sequence through the waveform embedding layer, where the wave vector of each data serves as the time step of the input sequence.

[0031] As a further limitation of the second aspect of the present invention, the GRU network calculates a value of a reset gate based on the current input and the hidden state of the previous time step, and calculates a value of an update gate based on the input and the hidden state of the previous time step;

[0032] Use the calculated reset gate to synthesize the candidate hidden state considering the old hidden state and the new input, use the update gate to control the fusion of the old hidden state and the new candidate hidden state, calculate the hidden state of the current time step, and continue to pass it to the next time step;

[0033] The hidden state of each time step captures the data of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state, which is used as the characteristic wave vector of the input sequence.

[0034] As a further limitation of the second aspect of the present invention, in the decoding module, decoding element by element through the sequence information in the wave vector to generate a new target model sequence includes:

[0035] The wave vector and label data output by the encoder are input into the decoder, which includes a model embedding layer, a GRU network, and a fully connected layer;

[0036] The decoder adjusts the dimension of the input label data and converts the data type, maps the converted label data into a vector sequence through the model embedding layer, and adjusts the dimension of the wave vector to generate a tensor that matches the decoder input for use by the decoder at each time step;

[0037] The adjusted wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step;

[0038] The decoder gradually generates the target model sequence according to the above process.

[0039] In a third aspect, the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the deep learning-based geological radar translation method as described in the first aspect of the present invention.

[0040] In a fourth aspect, the present invention provides an electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the deep learning-based geological radar translation method as described in the first aspect of the present invention are implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention innovatively proposes a geological radar translation strategy based on deep learning. The wave vector and label data output by the encoder are dimensionally adjusted and mapped into a vector sequence for use by the decoder. The decoder processes the wave vector and label data through a recurrent neural network to generate a predicted output and a new hidden state, and uses a fully connected layer to obtain the final predicted dielectric constant. The decoder gradually generates a target model sequence to complete the translation process of the geological radar data. The present invention can translate various dielectric constant models of tunnel linings containing complex defects and effectively reconstruct complex tunnel lining defects with clear boundaries.

[0043] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0045] Figure 1 A schematic flow chart of a geological radar translation method based on deep learning provided in Example 1 of the present invention;

[0046] Figure 2 The large-scale simple disease test results provided in Example 1 of the present invention;

[0047] Figure 3 The test results of multiple types of complex diseases provided in Example 1 of the present invention;

[0048] Figure 4 This is the micro-disease test result provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0051] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0052] Embodiment 1:

[0053] Embodiment 1 of the present invention provides a geological radar translation method based on deep learning, such as Figure 1 As shown, the following process is included:

[0054] S1: Processing geological radar data and model data through steps such as data loading, resizing, and normalization for further feature analysis and machine learning tasks.

[0055] Specifically, they include:

[0056] S1.1: Load geological radar data (e.g. generated by gprMax simulation) through radar reading function, read VTI data through model reading function and convert it into label data;

[0057] S1.2: Process the obtained label data, adjust the size of the model label data to match the size of the geological radar data, and resample the geological radar data to reduce the data volume;

[0058] S1.3: Normalize the geological radar data through MinMaxScaler, and save the processed geological radar data and model label data as a CSV file.

[0059] Among them, VTI data is a model file generated by GPRMax software, which is convenient for viewing the geometric shape of the simulated model. The file uses the open source visualization toolkit (VTK) format.

[0060] S2: The geological radar waveform is processed by the encoder and converted into a wave vector sequence. The GRU network gradually processes the wave data of each time step and generates a feature wave vector, which contains the entire waveform sequence information and is used by the decoder to generate the target model sequence.

[0061] Specifically, they include:

[0062] S2.1: The geological radar waveform sequence is input into the encoder, which consists of a waveform embedding layer and a GRU network. First, the encoder adjusts the dimension of the input waveform data and converts the data type. Then, the waveform embedding layer maps the converted waveform data into a wave vector sequence, where the wave vector of each data serves as the time step of the input sequence.

[0063] S2.2: Input the wave vector sequence into the GRU network. The GRU network is a variant of a recurrent neural network. The core idea is to introduce two gating mechanisms: update gate and reset gate. These two gating mechanisms help the network decide how to update the hidden state at each time step and can selectively ignore or forget some information. Among them, the update gate controls how much old memory information needs to be retained and determines the weight of the old memory at the current time step and the newly calculated candidate memory. The reset gate determines how to consider the impact of past input on the current time step and helps the network decide whether to ignore the past hidden state and pay more attention to the new input.

[0064] The GRU network calculates the value of the reset gate based on the current input and the hidden state of the previous time step, which controls the impact of the information of the past hidden state on the current time step; similarly, the value of the update gate is calculated based on the input and the hidden state of the previous time step, which determines how to fuse the old hidden state and the new candidate hidden state; then the calculated reset gate is used to synthesize the candidate hidden state considering the old hidden state and the new input; finally, the update gate is used to control the fusion of the old hidden state and the new candidate hidden state, and the hidden state of the current time step is calculated; and it is passed on to the next time step;

[0065] The hidden state of each time step captures the information of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state. This hidden state can be regarded as the feature wave vector of the input sequence, which contains the information of the entire waveform sequence and will be used by the decoder to generate the target model sequence.

[0066] S3: The wave vector and label data output by the encoder are dimensionally adjusted and mapped into a vector sequence for use by the decoder; the decoder processes the wave vector and label data through the GRU network to generate predicted output and new hidden states, and uses the fully connected layer to obtain the final predicted dielectric constant; the decoder gradually generates the target model sequence to complete the translation process of the geological radar data.

[0067] Specifically, they include:

[0068] S3.1: The wave vector and label data output by the encoder are input into the decoder, which consists of a model embedding layer, a GRU network, and a fully connected layer. First, the decoder adjusts the dimension of the input label data and converts the data type. Then, the converted label data is mapped into a vector sequence through the model embedding layer. The dimension of the wave vector is adjusted to generate a tensor that matches the decoder input for use by the decoder at each time step.

[0069] S3.2: The wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and then the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step. The decoder gradually generates the target model sequence according to the above process and finally completes the translation of the entire geological radar data.

[0070] like Figure 2 , Figure 3 and Figure 4As shown, it is a schematic diagram of specific test results. It can be seen that the translation result obtained by adopting the solution of the present invention is slightly different from the real result and can meet the requirements of translation accuracy.

[0071] Embodiment 2:

[0072] Embodiment 2 of the present invention provides a geological radar translation system based on deep learning, including:

[0073] The encoding module is configured to: pre-process the geological radar raw data to obtain a geological radar waveform, and encode the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform;

[0074] The decoding module is configured to: decode element by element through the sequence information in the wave vector to generate a new target model sequence, and translate the target model sequence into a dielectric constant model diagram through a model generation function.

[0075] The encoding module specifically includes:

[0076] The geological radar waveform sequence is input into the encoder, which consists of a waveform embedding layer and a GRU network. First, the encoder adjusts the dimension of the input waveform data and converts the data type. Then, the converted waveform data is mapped into a wave vector sequence through the waveform embedding layer, where the wave vector of each data serves as the time step of the input sequence.

[0077] The wave vector sequence is input into the GRU (Gate Recurrent Unit) network. The GRU network is a variant of a recurrent neural network. The core idea is to introduce two gating mechanisms: update gate and reset gate. These two gating mechanisms help the network decide how to update the hidden state at each time step and can selectively ignore or forget some information. Among them, the update gate controls how much old memory information needs to be retained and determines the weight of the old memory at the current time step and the newly calculated candidate memory. The reset gate determines how to consider the impact of past input on the current time step and helps the network decide whether to ignore the past hidden state and pay more attention to the new input.

[0078] The GRU network calculates the value of the reset gate based on the current input and the hidden state of the previous time step, which controls the impact of the information of the past hidden state on the current time step; similarly, the value of the update gate is calculated based on the input and the hidden state of the previous time step, which determines how to fuse the old hidden state and the new candidate hidden state; then the calculated reset gate is used to synthesize the candidate hidden state considering the old hidden state and the new input; finally, the update gate is used to control the fusion of the old hidden state and the new candidate hidden state, and the hidden state of the current time step is calculated; and it is passed on to the next time step;

[0079] The hidden state of each time step captures the information of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state. This hidden state can be regarded as the feature wave vector of the input sequence, which contains the information of the entire waveform sequence and will be used by the decoder to generate the target model sequence.

[0080] The decoding module specifically includes:

[0081] The wave vector and label data output by the encoder are dimensionally adjusted and mapped into a vector sequence for use by the decoder. The decoder processes the wave vector and label data through the GRU network, generates the predicted output and new hidden state, and uses the fully connected layer to obtain the final predicted dielectric constant. The decoder gradually generates the target model sequence to complete the translation process of the geological radar data.

[0082] The wave vector and label data output by the encoder are input into the decoder, which consists of a model embedding layer, a GRU network, and a fully connected layer. First, the decoder adjusts the dimension of the input label data and converts the data type. Then, the converted label data is mapped into a vector sequence through the model embedding layer. The dimension of the wave vector is adjusted to generate a tensor that matches the decoder input for use by the decoder at each time step.

[0083] The wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and then the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step. The decoder gradually generates the target model sequence according to the above process and finally completes the translation of the entire geological radar data.

[0084] Embodiment 3:

[0085] Embodiment 3 of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the geological radar translation method based on deep learning as described in Embodiment 1 of the present invention are implemented.

[0086] Embodiment 4:

[0087] Embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the geological radar translation method based on deep learning as described in Embodiment 1 of the present invention are implemented.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A geological radar translation method based on deep learning, characterized in that: The process includes: Preprocessing the geological radar raw data to obtain a geological radar waveform, encoding the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform; Through the sequence information in the wave vector, the new target model sequence is generated by decoding element by element, and the target model sequence is translated into a dielectric constant model graph through the model generation function; The geological radar waveform is encoded into a wave vector, including: The geological radar waveform sequence is input to the encoder, which includes a waveform embedding layer and a GRU network. The encoder adjusts the dimension of the input geological radar waveform and converts the data type. The converted waveform data is mapped into a wave vector sequence through the waveform embedding layer, where the wave vector of each data is used as the time step of the input sequence. Among them, through the sequence information in the wave vector, decoding is performed element by element to generate a new target model sequence, including: The wave vector and label data output by the encoder are input into the decoder, which includes a model embedding layer, a GRU network, and a fully connected layer; The decoder adjusts the dimension of the input label data and converts the data type, maps the converted label data into a vector sequence through the model embedding layer, and adjusts the dimension of the wave vector to generate a tensor that matches the decoder input for use by the decoder at each time step; The adjusted wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step; The decoder gradually generates the target model sequence according to the above process.

2. The geological radar translation method based on deep learning according to claim 1, characterized in that: Preprocessing of the raw data of geological radar includes: Load geological radar data through the radar reading function, read VTI data through the model reading function and convert it into model label data, and adjust the size of the model label data to match the size of the geological radar data; The original geological radar data is resampled to reduce the data volume, the geological radar data after the resampling operation is normalized, and the normalized geological radar data and the resized model label data are saved in the set format.

3. The geological radar translation method based on deep learning according to claim 1, characterized in that: The GRU network calculates the value of the reset gate based on the current input and the hidden state of the previous time step, and calculates the value of the update gate based on the input and the hidden state of the previous time step; Use the calculated reset gate to synthesize the candidate hidden state considering the old hidden state and the new input, use the update gate to control the fusion of the old hidden state and the new candidate hidden state, calculate the hidden state of the current time step, and continue to pass it to the next time step; The hidden state of each time step captures the data of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state, which is used as the characteristic wave vector of the input sequence.

4. A geological radar translation system based on deep learning, characterized in that: include: The encoding module is configured to: pre-process the geological radar raw data to obtain a geological radar waveform, and encode the geological radar waveform into a wave vector, wherein the wave vector contains all sequence information in the geological radar waveform; The decoding module is configured to: decode element by element through the sequence information in the wave vector to generate a new target model sequence, and translate the target model sequence into a dielectric constant model graph through a model generation function; Among them, in the encoding module, the geological radar waveform is encoded into a wave vector, including: The geological radar waveform sequence is input to the encoder, which includes a waveform embedding layer and a GRU network. The encoder adjusts the dimension of the input geological radar waveform and converts the data type. The converted waveform data is mapped into a wave vector sequence through the waveform embedding layer, where the wave vector of each data is used as the time step of the input sequence. Among them, in the decoding module, the sequence information in the wave vector is decoded element by element to generate a new target model sequence, including: The wave vector and label data output by the encoder are input into the decoder, which includes a model embedding layer, a GRU network, and a fully connected layer; The decoder adjusts the dimension of the input label data and converts the data type, maps the converted label data into a vector sequence through the model embedding layer, and adjusts the dimension of the wave vector to generate a tensor that matches the decoder input for use by the decoder at each time step; The adjusted wave vector sequence and label data are input into the GRU network to generate output and new hidden state, and the final predicted output is obtained through the fully connected layer. The decoder receives the hidden state of the previous time step and the dielectric constant generated in the previous time step as input at each time step, and outputs the predicted dielectric constant and new hidden state of the current time step.

5. The deep learning-based geological radar translation system according to claim 4, characterized in that: The GRU network calculates the value of the reset gate based on the current input and the hidden state of the previous time step, and calculates the value of the update gate based on the input and the hidden state of the previous time step; Use the calculated reset gate to synthesize the candidate hidden state considering the old hidden state and the new input, use the update gate to control the fusion of the old hidden state and the new candidate hidden state, calculate the hidden state of the current time step, and continue to pass it to the next time step; The hidden state of each time step captures the data of each wave in the input waveform sequence and the information between waves, and encodes this information into a wave vector. After the encoder processes the entire input sequence, it outputs the final hidden state, which is used as the characteristic wave vector of the input sequence.

6. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the deep learning-based geological radar translation method as described in any one of claims 1 to 3 are implemented.

7. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the steps in the deep learning-based geological radar translation method as described in any one of claims 1-3.

Images