CN115856882B - Intelligent inversion and imaging method for multi-polarization geological radar data - Google Patents
Intelligent inversion and imaging method for multi-polarization geological radar data Download PDFInfo
- Publication number
- CN115856882B CN115856882B CN202310112866.0A CN202310112866A CN115856882B CN 115856882 B CN115856882 B CN 115856882B CN 202310112866 A CN202310112866 A CN 202310112866A CN 115856882 B CN115856882 B CN 115856882B
- Authority
- CN
- China
- Prior art keywords
- polarization
- data
- geological radar
- inversion
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Radar Systems Or Details Thereof (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses an intelligent inversion and imaging method for multi-polarization geological radar data, which comprises the following steps: constructing an inversion network model, wherein the inversion network model comprises a circularly generated countermeasure network model and a polarization rule constraint convolution semi-supervised depth inversion network model; constructing a training sample, wherein the training sample comprises a simulation data set and a real data set; training the inversion network model through the training sample; acquiring multi-polarization geological radar data, and performing dielectric constant inversion on the multi-polarization geological radar data through a trained inversion network model to obtain a dielectric constant distribution diagram so as to realize disease detection.
Description
Technical Field
The invention relates to the technical field of nondestructive testing, in particular to an intelligent inversion and imaging method for multi-polarization geological radar data.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The geological radar has the advantages of high precision, high detection efficiency, flexible and convenient field work and the like, becomes a main stream detection method for detecting the defects of the infrastructure engineering structure, and is very important for guaranteeing the safe service of the geological radar. Different from the monopole geological radar commonly used in the current engineering, the multipolar geological radar adopts a plurality of directions to polarize the antenna, and different polarized antennas have different sensitivity to targets with different internal properties, trend and morphology of the structure, so that richer information of the targets can be obtained from a plurality of directions, and the information of the targets in the walking direction detection blind area can be detected. Currently, some companies develop and use commercial multi-polarization radars, and multi-polarization geological radars become the preferred method and future development trend of detecting diseases in the infrastructure.
In recent years, with the rapid development of new generation information technology, techniques such as signal processing and artificial intelligence are used for automatically explaining multi-polarization geological radar data at home and abroad. Traditional multi-polarization geological radar data inversion and imaging methods comprise Freeman technology, H-alpha decomposition and the like. With the rapid development of deep learning technology, a deep neural network-based method has been applied to multi-polarization geological radar data inversion and imaging. Two papers have been used to invert and image multi-polarized geological radar data using the deep neural network approach, for example: the method includes processing polarization data using a Convolutional Neural Network (CNN) to identify a subsurface target, and providing a masked pilot multi-polarization integrated neural network (MMITET) to estimate a plurality of root related parameters in non-uniform soil. There is still little research on deep learning methods in terms of multi-polarization geological radar inversion and imaging.
However, the existing multi-polarization geological radar inversion and imaging method based on deep learning has the following problems: (1) At present, most of single-polarization geological radars are adopted, so that more vertical-vertical (VV) polarization data are acquired, and because special multi-polarization radars are required for acquisition, the quantity of vertical-horizontal (VH) and horizontal-horizontal (HH) polarization data is small, so that the problem of data unbalance of the multi-polarization geological radars is caused, and the effect of a deep learning inversion method is poor. (2) The tags of the unipolar geological radar data are difficult to acquire, the tags of the multipolar geological radar data are fewer, a large number of multipolar geological radar data are lacked, supervised training is difficult to perform, and the problem that a network model is difficult to converge in the existing multi-polar geological radar inversion and imaging training process based on deep learning is caused.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an intelligent inversion and imaging method for multi-polarization geological radar data, which adopts a circularly generated countermeasure network to generate the multi-polarization geological radar data, solves the problem of unbalanced data of the multi-polarization geological radar data, provides polarization rule constraint convolution, fully extracts the difference characteristics of different polarization data, adopts a semi-supervised deep learning network, and fully utilizes a large amount of unlabeled multi-polarization data and a small amount of labeled multi-polarization geological radar data.
In order to achieve the technical purpose, the invention provides the following technical scheme:
an intelligent inversion and imaging method for multi-polarization geological radar data, comprising the following steps:
constructing an inversion network model, wherein the inversion network model comprises a circularly generated countermeasure network model and a polarization rule constraint convolution semi-supervised depth inversion network model; constructing a training sample, wherein the training sample comprises a simulation data set and a real data set;
training the inversion network model through the training sample; acquiring multi-polarization geological radar data, and performing dielectric constant inversion on the multi-polarization geological radar data through a trained inversion network model to obtain a dielectric constant distribution diagram so as to realize disease detection.
Optionally, the process of acquiring the simulation dataset includes:
acquiring medium disease information, and randomly composing the medium disease information to generate a simulated dielectric constant distribution diagram;
EDTD forward modeling is carried out on the simulated dielectric constant distribution diagram, simulated multi-polarization geological radar data are generated, and the simulated dielectric constant distribution diagram and the simulated multi-polarization geological radar data are integrated to obtain a simulated data set, wherein the simulated multi-polarization geological radar data comprise vertical-vertical polarization data, vertical-horizontal polarization data and horizontal-horizontal polarization data.
Optionally, the process of acquiring the real dataset includes:
the method comprises the steps of obtaining actual measurement multi-polarization geological radar data, and carrying out random clipping and bilinear interpolation on the actual measurement multi-polarization geological radar data in the horizontal direction based on a dielectric constant model to generate a real data set.
Optionally, the loop generation type countermeasure network model includes two generation type countermeasure networks for respectively migrating the vertical-vertical polarization data to the vertical-horizontal polarization data and migrating the vertical-vertical polarization data to the horizontal-horizontal polarization data, generating the reconstructed training data, and optimizing the generation type countermeasure network through the polarization characteristic constraint loss function.
Optionally, the constructing process of the polarization characteristic constraint loss function includes:
and (3) carrying out inverse Fourier transform based on the data before migration and the data after migration, decomposing an inverse Fourier transform result through a polarization decomposition algorithm, generating feature spaces corresponding to the data before migration and the data after migration, and taking the minimum mean square value deviation of the corresponding feature spaces as a polarization feature constraint loss function.
Optionally, the polarization rule constraint convolution semi-supervised depth inversion network model comprises a polarization rule constraint convolution layer and a semi-supervised depth inversion network model, the semi-supervised depth inversion network model comprises a teacher model and a student model, wherein the multi-polarization geological radar data are processed through the polarization rule constraint convolution by the polarization rule constraint convolution layer, and the processed multi-polarization geological radar data are subjected to inversion processing through the semi-supervised depth inversion network model.
Optionally, the process of polarization rule constraint convolution includes:
processing the multi-polarization geological radar data by adopting regular constraint convolution to generate a multi-polarization feature map; wherein the convolution kernel of the law-constrained convolution is changed based on scattering parameters, different scattering parameters are constructed according to the multi-polarization geological radar data, and when the scattering parameter values of the vertical-vertical polarization data and the horizontal-horizontal polarization data are 1, the scattering parameter values of the vertical-horizontal polarization data are the difference value between the vertical-horizontal polarization data and the average value of the corresponding vertical-vertical polarization data and horizontal-horizontal polarization data.
Optionally, the training the inversion network model includes:
dividing a training sample into labeled data and unlabeled data;
generating label-free data through a circulation generation type countermeasure network model, generating label-free multi-polarization data, extracting characteristics of the label-free data and the label-free data through a polarization rule constraint convolution layer, training a student network by using the extracted label-free result, inputting the extracted label-free result into the trained student network and a teacher network, taking the trained student network result as a label of the teacher network, optimizing the teacher network according to the student network, and inverting the optimized teacher network.
The invention has the following technical effects:
1. the method is suitable for the intelligent inversion and imaging method of the multi-polarization geological radar data, and can be used for more omnibearing and finer inversion and imaging of targets with different trend and morphology based on a small amount of unbalanced multi-polarization geological radar data;
2. the cyclic generation type countermeasure network module embedded with polarization physical feature constraint respectively learns that features of vertical-vertical (VV) polarization data are migrated to unpaired vertical-horizontal (VH) polarization data and learns that features of vertical-vertical (VV) polarization data are migrated to unpaired horizontal-horizontal (HH) polarization data by adopting two cyclic generation type countermeasure networks, so that the problem of data imbalance of multi-polarization geological radar data is solved, a polarization feature constraint loss function is established by introducing a polarization decomposition operator of a scattering matrix, and high-fidelity conversion between different polarization geological radar data is realized;
3. the polarization rule constraint convolution realizes that the scattering parameter K of the convolution kernel is changed according to different polarization data inputs, so that the receptive field is enlarged, and the scattering characteristics of different polarization data are accurately extracted. And the semi-supervised deep learning network method fully utilizes a large amount of unlabeled multi-polarization data and a small amount of labeled multi-polarization geological radar data.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for intelligent inversion and imaging of multi-polarized geological radar data;
FIG. 2 is a schematic diagram of a network architecture of a method for intelligent inversion and imaging of multi-polarized geological radar data;
FIG. 3 is a block diagram of a loop-generated countermeasure network module embedded with polarized physical characteristics constraints;
FIG. 4 is a block diagram of a polarization rule constraint convolution module;
FIG. 5 is a multi-polarized geological radar data result generated by a loop-generated countermeasure network module in accordance with implementation of the embedded polarized physical characteristics constraints shown.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order to achieve the above object, the present invention adopts the following technical scheme: a multi-polarization geological radar data intelligent inversion and imaging method comprises a circularly generated type countermeasure network module embedded with polarization physical feature constraint and a semi-supervised depth inversion network module with polarization rule constraint convolution.
The circularly generated countermeasure network module embedded with polarization physical feature constraints respectively learns that features of vertical-vertical (VV) polarization data migrate to unpaired vertical-horizontal (VH) polarization data and learns that features of vertical-vertical (VV) polarization data migrate to unpaired horizontal-horizontal (HH) polarization data by employing two circularly generated countermeasure networks such as CycleGAN, thereby enabling one set of multi-polarization radar data features to migrate to another set of multi-polarization data and generating a large amount of multi-polarization geological radar data. And a polarization decomposition operator of the scattering matrix is introduced to establish a polarization characteristic constraint loss function, so that high-fidelity conversion between different polarized geological radar data is realized.
The circularly generated type countermeasure network module embedded with the polarization physical feature constraint establishes a polarization feature constraint loss function by introducing a polarization decomposition operator of a scattering matrix, and realizes high-fidelity conversion between different polarization geological radar data. In order to ensure that original polarization characteristics of two converted data fields are still kept unchanged, a polarization decomposition operator of a scattering matrix is introduced to establish a polarization characteristic constraint loss function, firstly, multi-polarization scattering matrix of each measuring point is extracted by performing inverse Fourier transform on multi-geological radar data pairs
By using +.>
Polarization attribute decomposition operator respectively carrying out polarization decomposition on the multi-polarization matrix before and after conversion to obtain polarization decomposition characteristic space of original data and converted data>
And->
And taking the minimum mean square deviation MSE of the two as a loss function +.>
:
The polarization law constraint convolution, which multiplies the convolution kernel by different scattering parameters K according to the input polarization data, wherein VV polarization data (S VV ) And HH polarization data (S) HH ) Is 1 and the convolution K value of the VH polarization data is
Wherein the scattering parameter K value of the VH polarization data is the data by rotation of the monopole antenna>
，/>
And
the multi-polarization scattering matrix calculation is generated, and the process is as follows:
final multi-polarization scattering matrix S (assuming
) The method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for incident wave electric field, +.>
，/>
And->
Is scattering data for three angles. Therefore, the convolution constraint K value of VH polarization data is +.>
The convolution kernel is changed according to different polarization data input, so that different scattering characteristics of multi-polarization data are learned.
The semi-supervised deep inversion network adopting the polarization rule constraint convolution adopts a semi-supervised learning method, utilizes a large amount of unlabeled multi-polarization data and a small amount of labeled multi-polarization geological radar data generated by a cyclic generation type countermeasure network module embedded with polarization physical feature constraint to extract the features of different polarization data through the polarization rule constraint convolution, trains a student network by the features of a small amount of labeled multi-polarization radar data, takes the features of a large amount of unlabeled multi-polarization data as the input of the student network and a teacher network at the same time, takes the output result of the student network as the label of the teacher network, updates the teacher network according to the parameters of the student network, and simultaneously trains the cyclic generation type countermeasure network module embedded with the polarization physical feature constraint and the semi-supervised learning module through the polarization feature constraint loss function, the consistency loss function and the cyclic countermeasure loss function.
Matched software system
The deep learning embedded control module is provided with an extended communication interface, the PCIe interface is connected with the 4G module, and the Wifi or 4G is adopted to transmit real-time geological radar detection data and recognition results to electronic equipment such as a mobile phone terminal, a tablet personal computer and a computer. And constructing a software platform by adopting an intelligent processing algorithm, a Hadoop technology and an Hbase distributed database, and finally forming an intelligent software system integrating a circularly generated countermeasure network module embedded with polarization physical characteristic constraint and a semi-supervised deep inversion network module with polarization rule constraint convolution.
Example two
As shown in fig. 1, the embodiment discloses a multi-polarization geological radar data intelligent inversion and imaging method, which comprises the following steps:
step S1: constructing multi-polarization geological radar data inversion network model
The multi-polarization geological radar data inversion network structure comprises a cyclic generation type countermeasure network module embedded with polarization physical feature constraint and a semi-supervised depth inversion network module convoluting with polarization rule constraint, a large amount of multi-polarization geological radar data generated by the cyclic generation type countermeasure network embedded with the polarization physical feature constraint are utilized, the features of different polarization data are extracted through the polarization rule constraint convoluting, and a semi-supervised learning method is adopted, so that the cyclic generation type countermeasure network module embedded with the polarization physical feature constraint and the semi-supervised learning module are trained simultaneously through a consistency loss function and a countermeasure loss function.
The device specifically comprises two modules:
(1) A loop generation type countermeasure network module embedded with polarized physical characteristics constraint, as shown in fig. 3, includes two generators by employing a loop generation type countermeasure network
And two discriminators->
Image feature migration and conversion between different domains are achieved by employing two generative countermeasure network structures, migrating from vertical-vertical (VV) polarization data features to unpaired vertical-horizontal (VH) polarization data and from vertical-vertical (VV) polarization data features to unpaired horizontal-horizontal (HH) polarization data, respectively, thereby generating a large number of multi-polarization geological radar data, and establishing a polarization feature constraint loss function by introducing a polarization decomposition operator of a scattering matrix, thereby achieving high-fidelity conversion between different polarization geological radar data. In order to ensure that original polarization characteristics of two converted data fields are still kept unchanged, a polarization decomposition operator of a scattering matrix is introduced to establish a polarization characteristic constraint loss function, firstly, multi-geological radar data pairs are subjected to inverse Fourier transformation to extract a multi-polarization scattering matrix +_of each measuring point>
By using +.>
Polarization attribute decomposition operator respectively carrying out polarization decomposition on the multi-polarization matrix before and after conversion to obtain polarization decomposition characteristic space of original data and converted data>
And->
And taking the minimum mean square deviation MSE of the two as a loss function +.>
:
Thus, the total loss function of the loop-generating countermeasure network module embedded with the polarization physical feature constraint is:
wherein, the liquid crystal display device comprises a liquid crystal display device,
and->
MSE loss function, respectively for two loop-generated countermeasure networks embedded with polarization physical feature constraints,/->
And->
Generating a loss function for a discriminator of a countermeasure network for one cycle, likewise ≡>
And->
A loss function of the arbiter of the antagonism network is generated for another cycle,
and->
Generating a round-robin loss function of the countermeasure network for two rounds, respectively,>
and->
For constraint parameters of two cyclic loss functions, A is source domain polarization data, B and C represent two target domain data, F represents a mapping function from source domain to target domain, G represents a mapping function from target domain to source domain, and->
And->
Mapping functions of two discriminators of an countermeasure network are generated for one cycle, +.>
And->
A mapping function is generated for two discriminators of the countermeasure network for another cycle. The specific formula is as follows: />
(2) As shown in fig. 4, the semi-supervised depth inversion network module of polarization rule constraint convolution processes multi-polarization data by first adopting the polarization rule constraint convolution, wherein the polarization rule constraint convolution adopts different scattering parameters K to multiply convolution kernels according to different input polarization data, and VV polarization data (S VV ) And HH (S) HH ) Polarization data K value is 1, while VH polarization data K value is
Wherein the scattering parameter K value of the VH polarization data is the data by rotation of the monopole antenna>
，
And->
The multi-polarization scattering matrix calculation is generated, and the process is as follows:
final multi-polarization scattering matrix S (assuming
) The method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for incident wave electric field, +.>
，/>
And->
Scattering data for three angles 90, 0, 45. Therefore, the convolution constraint K value of VH polarization data is +.>
The convolution kernel is changed according to different polarization data input, so that different scattering characteristics of multi-polarization data are learned.
Secondly, a semi-supervised learning method is adopted, a large amount of unlabeled multi-polarization data and a small amount of labeled multi-polarization geological radar data generated by an embedded polarization physical feature constraint are utilized to generate a circular generation type countermeasure network, firstly, features of different polarization data are extracted through polarization rule constraint convolution, a small amount of labeled multi-polarization radar data train a student network, secondly, the features of a large amount of unlabeled multi-polarization data generated by the generation type countermeasure network are simultaneously used as input of the student network and a teacher network, the student network updates parameters through gradient descent, the output result of the student network is used as a label of the teacher network, meanwhile, parameters of the teacher network are updated according to the parameters of the student network, and finally, the inversion result of the multi-polarization radar data of the relevant data through a polarization rule constraint convolution layer and the teacher network is the inversion result of the whole network structure.
It should be noted that, the teacher network and the student network both need to generate data of the countermeasure network, the student network in the first stage needs to have tag data, the teacher in the second stage needs to have no tag data, and the generated countermeasure network acts to expand the no tag training set, but participates in the training process as a whole, and the test recognition process does not need to be used.
Step S2: establishing a simulation data set, and training a multi-polarization geological radar data inversion network model based on the simulation data set
Aiming at the problem of detecting the diseases of the large infrastructure structure, a corresponding simulation data set is established. The step S2 specifically includes:
step S201: large infrastructure structure dielectric constant profiles of various lengths are constructed.
Specifically, a dielectric constant distribution map of a lining structure section is generated according to each combination mode by randomly combining a background medium, a disease internal medium, the number of diseases, the positions of the diseases and the like. The background medium comprises a plurality of background mediums such as plain concrete, reinforced concrete and the like, the diseases comprise void, incompact and the like, and the internal mediums of the diseases are mediums such as water, air and the like.
Step S202: the Rake wavelet with the same frequency and phase as the actual geological radar wavelet is used as the source wavelet for modeling the simulation data to carry out FDTD forward modeling on each dielectric constant distribution diagram, and the corresponding multi-polarization geological radar data is generated by changing the directions of the transmitting antenna and the receiving antenna.
Step S203: based on the simulation data set, a loop-generated challenge network module (such as fig. 5) and a semi-supervised learning module are employed to simultaneously train and test embedded polarization physical feature constraints with a consistency loss function and a challenge loss function.
Step S3: and establishing a real data set to obtain a polarized geological radar data inversion network model suitable for the actual detection data of the geological radar.
The step S3 specifically includes:
step S301: establishing a real dataset
And carrying out corresponding random clipping and bilinear interpolation in the horizontal direction on geological radar profile data actually detected by the multi-polarization geological radar according to a dielectric constant model established in an actual field so as to carry out data enhancement, and establishing a real data set.
Step S302: training the multi-polarization geological radar data inversion network model based on the real data set.
Step S4: and performing dielectric constant inversion on the actually acquired multi-polarization geological radar data by utilizing a multi-polarization geological radar data inversion network to obtain a corresponding dielectric constant distribution diagram.
Substituting the trained intelligent inversion model parameters suitable for the actual data into the constructed intelligent inversion network to obtain the prediction model capable of being practically applied. Then, the development of the graphical interface is carried out by using the surface emitting tool, an interface for a user to use is generated, the user can randomly select and collect multi-polarization geological radar data to input the graphical interface, and then the prediction model inverts the input data to generate a dielectric constant distribution diagram. And reducing the background medium and the disease morphology of the infrastructure according to the dielectric constant distribution diagram, thereby achieving the purpose of disease detection.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (5)
1. An intelligent inversion and imaging method for multi-polarization geological radar data, which is characterized by comprising the following steps:
constructing an inversion network model, wherein the inversion network model comprises a circularly generated countermeasure network model and a polarization rule constraint convolution semi-supervised depth inversion network model; constructing a training sample, wherein the training sample comprises a simulation data set and a real data set;
training the inversion network model through the training sample; acquiring multi-polarization geological radar data, and performing dielectric constant inversion on the multi-polarization geological radar data through a trained inversion network model to obtain a dielectric constant distribution diagram so as to realize disease detection;
the process of acquiring the simulation dataset includes:
acquiring medium disease information, and randomly composing the medium disease information to generate a simulated dielectric constant distribution diagram;
EDTD forward modeling is carried out on the simulated dielectric constant distribution diagram, simulated multi-polarization geological radar data are generated, and the simulated dielectric constant distribution diagram and the simulated multi-polarization geological radar data are integrated to obtain a simulated data set, wherein the simulated multi-polarization geological radar data comprise vertical polarization data, vertical polarization data and horizontal polarization data;
the circularly generated countermeasure network model comprises two generated countermeasure networks, which are respectively used for migrating vertical and vertical polarization data to vertical and horizontal polarization data, generating reconstruction training data, and optimizing the generated countermeasure network through a polarization characteristic constraint loss function;
the construction process of the polarization characteristic constraint loss function comprises the following steps:
and (3) carrying out inverse Fourier transform based on the data before migration and the data after migration, decomposing an inverse Fourier transform result through a polarization decomposition algorithm, generating feature spaces corresponding to the data before migration and the data after migration, and taking the minimum mean square value deviation of the corresponding feature spaces as a polarization feature constraint loss function.
2. The imaging method of claim 1, wherein:
the process of acquiring the real dataset comprises:
the method comprises the steps of obtaining actual measurement multi-polarization geological radar data, and carrying out random clipping and bilinear interpolation on the actual measurement multi-polarization geological radar data in the horizontal direction based on a dielectric constant model to generate a real data set.
3. The imaging method of claim 1, wherein:
the polarization rule constraint convolution semi-supervised depth inversion network model comprises a polarization rule constraint convolution layer and a semi-supervised depth inversion network model, wherein the semi-supervised depth inversion network model comprises a teacher model and a student model, the multi-polarization geological radar data are processed through the polarization rule constraint convolution by the polarization rule constraint convolution layer, and the processed multi-polarization geological radar data are subjected to inversion processing through the semi-supervised depth inversion network model.
4. The imaging method of claim 1, wherein:
the polarization rule constraint convolution process comprises the following steps:
processing the multi-polarization geological radar data by adopting regular constraint convolution to generate a multi-polarization feature map; the convolution kernel of the law constraint convolution is changed based on scattering parameters, different scattering parameters are constructed according to the multi-polarization geological radar data, and when the scattering parameter value of vertical and horizontal polarization data is 1, the scattering parameter value of the vertical and horizontal polarization data is the difference value between the vertical and horizontal polarization data and the average value of the corresponding vertical and horizontal polarization data.
5. The imaging method of claim 1, wherein:
the process of training the inversion network model comprises:
dividing a training sample into labeled data and unlabeled data;
generating label-free data through a circulation generation type countermeasure network model, generating label-free multi-polarization data, extracting characteristics of the label-free data and the label-free data through a polarization rule constraint convolution layer, training a student network by using the extracted label-free result, inputting the extracted label-free result into the trained student network and a teacher network, taking the trained student network result as a label of the teacher network, optimizing the teacher network according to the student network, and inverting the optimized teacher network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310112866.0A CN115856882B (en) | 2023-02-15 | 2023-02-15 | Intelligent inversion and imaging method for multi-polarization geological radar data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310112866.0A CN115856882B (en) | 2023-02-15 | 2023-02-15 | Intelligent inversion and imaging method for multi-polarization geological radar data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115856882A CN115856882A (en) | 2023-03-28 |
| CN115856882B true CN115856882B (en) | 2023-05-16 |
Family
ID=85658049
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310112866.0A Active CN115856882B (en) | 2023-02-15 | 2023-02-15 | Intelligent inversion and imaging method for multi-polarization geological radar data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115856882B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7167123B2 (en) * | 1999-05-25 | 2007-01-23 | Safe Zone Systems, Inc. | Object detection method and apparatus |
| CN103076606B (en) * | 2013-01-10 | 2014-12-10 | 山东大学 | Three-dimensional fine imaging system and method based on drilling geological radar technology |
| EP4014197A1 (en) * | 2019-08-13 | 2022-06-22 | University of Hertfordshire Higher Education Corporation | Predicting visible/infrared band images using radar reflectance/backscatter images of a terrestrial region |
| CN111152182B (en) * | 2019-10-09 | 2022-02-08 | 山东大学 | Multi-arm robot for tunnel lining detection and disease diagnosis in operation period |
| CN112731377B (en) * | 2020-12-15 | 2024-03-26 | 郑州大学 | Dielectric constant inversion method, roadbed detection method and detection system |
| CN113447536B (en) * | 2021-06-24 | 2022-09-30 | 山东大学 | Concrete dielectric constant inversion and disease identification method and system |
| CN114021422B (en) * | 2021-09-23 | 2024-04-26 | 大连理工大学 | Underground structure internal defect identification method based on cross-hole radar and deep learning |
| CN114578349B (en) * | 2022-05-06 | 2022-08-05 | 山东大学 | Unmanned airborne geological radar system for detecting hidden danger of dam and inspection method |
-
2023
- 2023-02-15 CN CN202310112866.0A patent/CN115856882B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115856882A (en) | 2023-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Condat et al. | Cadzow denoising upgraded: A new projection method for the recovery of Dirac pulses from noisy linear measurements | |
| Tang et al. | A joint tensor completion and prediction scheme for multi-dimensional spectrum map construction | |
| Sun et al. | Convolutional neural network based models for improving super-resolution imaging | |
| CN113156439B (en) | SAR wind field and sea wave joint inversion method and system based on data driving | |
| CN114549925A (en) | Sea wave effective wave height time sequence prediction method based on deep learning | |
| CN113359212A (en) | Reservoir characteristic prediction method and model based on deep learning | |
| CN116343046A (en) | Multi-source remote sensing urban river water quality monitoring method based on self-optimizing machine learning | |
| CN114399073A (en) | Ocean surface temperature field prediction method based on deep learning | |
| Wan et al. | An improved complex multi-task Bayesian compressive sensing approach for compression and reconstruction of SHM data | |
| CN114417693A (en) | Ocean three-dimensional temperature field inversion method based on deep learning | |
| CN115856882B (en) | Intelligent inversion and imaging method for multi-polarization geological radar data | |
| CN109615577A (en) | High spectrum image super-resolution processing method based on convolutional network | |
| CN115598714B (en) | Time-space coupling neural network-based ground penetrating radar electromagnetic wave impedance inversion method | |
| CN116863959A (en) | Dolphin sound generating method based on generating countermeasure network | |
| Wei et al. | Probabilistic optimal interpolation for data assimilation between machine learning model predictions and real time observations | |
| Zhang et al. | Accelerating hydraulic fracture imaging by deep transfer learning | |
| Ozturk et al. | Application of compressive sensing to refractivity retrieval using networked weather radars | |
| Bosma et al. | Estimating solar and wind power production using computer vision deep learning techniques on weather maps | |
| Wang et al. | Deformation prediction of rock cut slope based on long short-term memory neural network | |
| Geng et al. | Data Collection for Mobile Crowd Sensing Based on Tensor Completion. | |
| Zhang et al. | A refined rank set pair analysis model based on wavelet analysis for predicting temperature series | |
| Guo et al. | Microwave inversion for sparse data using descent learning technique | |
| Liu et al. | Physics-informed unsupervised deep learning framework for solving full-wave inverse scattering problems | |
| CN109389052A (en) | Full polarimetric SAR city based on clustering algorithm identifies extracting method | |
| Daningburg et al. | Cost minimization in acquisition for gravitational wave surrogate modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |