Methods, devices, equipment, media and products for calculating co-seismic deformation fields

Abstract

The present application relates to a method, device, equipment, medium and product for solving a co-seismic deformation field, and in particular to the field of satellite signal processing technology. The method includes: obtaining a first D-InSAR deformation field and a POT deformation field; inputting the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model; the noise correction model is obtained by training the POT deformation field and a second D-InSAR deformation field with decoherence regions and unwrapping error regions removed; inputting the entangled phases of the POT low-noise deformation field and the D-InSAR deformation field into a near-field deformation prediction model to obtain near-field deformation; inputting the spatial information of the pixel points and the spatial relationship data of the epicenter into a full-scale deformation prediction model to obtain a full-scale deformation field output by the full-scale deformation prediction model; the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the decoherence missing values ​​of the second D-InSAR deformation field. The embodiment of the present application can solve a complete co-seismic deformation field with low noise and correct phase unwrapping.

Description

Technical Field

[0001] The present application relates to the field of satellite signal processing technology, and in particular to a method, device, equipment, medium and product for calculating a co-seismic deformation field. Background Art

[0002] The coseismic deformation field of an earthquake has an intuitive indicative effect on the distribution of rupture zones, the geometric distribution of seismogenic faults, the location of the hardest-hit areas, and the intensity distribution. It is also an important constraint data for earthquake dynamics simulation and kinematic inversion, and is crucial to understanding the geometric structure of seismogenic faults and analyzing earthquake rupture mechanisms.

[0003] At present, the main means of acquiring large-scale earthquake coseismic deformation fields based on synthetic aperture radar (SAR) satellites include synthetic aperture radar differential interferometry (D-InSAR) technology and pixel offset tracking (POT) technology. Among them, D-InSAR technology solves the coseismic deformation field by differentiating the phase information of radar waves before and after the earthquake. The core operation of this technology is phase unwrapping using the phase continuity assumption. However, if the deformation gradient exceeds half a wavelength, it is likely to cause phase unwrapping errors, and the phase island area caused by the large area of ​​decoherence is also difficult to be correctly unwrapped. POT technology solves the coseismic deformation field by comparing the deviations of the same-name pixel points in the images before and after the earthquake. It may happen that: due to excessive deformation, the ground surface is ruptured, which leads to errors in matching the same-name ground points; in the far-field small deformation area and the homogeneous ground area with small deformation, the ground intensity characteristics are not obviously different, resulting in matching errors, which leads to large noise in the deformation results solved by POT. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a method, device, equipment, medium and product for solving the co-seismic deformation field, which can solve the co-seismic deformation field with low noise and correct phase unwrapping, and can compensate for the deformation value of the decoherence area to a certain extent under the premise of considering the spatial distribution law of the co-seismic deformation field.

[0005] In order to achieve the above purpose, the technical solutions provided by the embodiments of the present application are as follows:

[0006] In a first aspect, the present application provides a method for solving a co-seismic deformation field, comprising: obtaining a first synthetic aperture radar differential interferometry D-InSAR deformation field and a pixel offset tracking POT deformation field; inputting the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model; the noise correction model is obtained by training the POT deformation field and a second D-InSAR deformation field with decoherence regions and unwrapping error regions removed; inputting the entanglement phase of the POT low-noise deformation field and the D-InSAR deformation field into a near-field deformation prediction model to obtain The near-field deformation output by the near-field deformation prediction model; the near-field deformation prediction model is used to resolve the near-field deformation of the unwrapping error in the D-InSAR deformation field; the spatial information of the pixel points and the spatial relationship data of the epicenter are input into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model; the spatial information of the pixel points and the spatial relationship data of the epicenter reflect the spatial distribution law of the co-seismic deformation field, and the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

[0007] As an optional implementation of an embodiment of the present application, the training process of the noise correction model includes: based on the coherence parameter and the interference pattern, removing the decoherent area and the unwrapping error area from the first D-InSAR deformation field to obtain the second D-InSAR deformation field; obtaining the POT range deformation field; constructing an initial noise correction model, taking the POT range deformation field as the input of the initial noise correction model, taking the second D-InSAR deformation field as the true value, and training the initial noise correction model until the model converges to obtain the noise correction model; wherein, the noise correction model is used to correct the high-dimensional noise in the POT range deformation field.

[0008] As an optional implementation of an embodiment of the present application, the training process of the near-field deformation prediction model includes: based on the coherence parameter and the interference pattern, removing the decoherent area and the unwrapping error area from the first D-InSAR deformation field to obtain a second D-InSAR deformation field; obtaining the POT low-noise deformation field output by the noise correction model; constructing an initial near-field deformation prediction model, taking the winding phase of the D-InSAR deformation field and the near-field deformation of the POT low-noise deformation field as input, and taking the second D-InSAR deformation field as the true value, training the initial near-field deformation prediction model until the model converges to obtain the near-field deformation prediction model.

[0009] As an optional implementation of an embodiment of the present application, the training process of the full-scale deformation prediction model includes: obtaining the POT low-noise deformation field output by the noise correction model, and the near-field deformation output by the near-field deformation prediction model; constructing an initial full-scale deformation prediction model, taking the spatial information of the pixel points and the spatial relationship data of the epicenter as input, and taking the fusion result of the POT low-noise deformation field and the near-field deformation as the true value, training the initial full-scale deformation prediction model until the model converges to obtain the full-scale deformation prediction model.

[0010] As an optional implementation of an embodiment of the present application, the D-InSAR deformation field and the POT deformation field are correlated with the spatial information of the pixel points and the spatial relationship of the epicenter; after obtaining the first synthetic aperture radar differential interferometry measurement D-InSAR deformation field and the pixel offset tracking POT deformation field, and before inputting the POT deformation field into the noise correction model to obtain the POT low-noise deformation field output by the noise correction model, the method also includes: based on the correlation and regression prediction algorithm, constructing a noise correction model, a near-field deformation prediction model and a full-scale deformation prediction model and training them separately until the model converges. As an optional implementation of an embodiment of the present application, the unwrapping error region includes a phase island region surrounded by a decoherent region, and an unwrapping error region caused by not satisfying the phase continuity criterion; based on the coherence parameter and the interference pattern, the decoherent region and the unwrapping error region are removed from the first D-InSAR deformation field to obtain a second D-InSAR deformation field, including: extracting the decoherent region from the first D-InSAR deformation field by threshold segmentation based on the coherence parameter; extracting the phase island region surrounded by the decoherent region from the first D-InSAR deformation field; and, based on the phase gradient and the epicenter position reflected by the interference pattern, extracting the unwrapping error region caused by not satisfying the phase continuity criterion in the first D-InSAR deformation field; removing the decoherent region, the phase island region, and the unwrapping error region caused by not satisfying the phase continuity criterion from the first D-InSAR deformation field to obtain the second D-InSAR deformation field.

[0011] In a second aspect, the present application provides a co-seismic deformation field solving device, the device comprising:

[0012] An acquisition module, used for acquiring a first D-InSAR deformation field and a POT deformation field;

[0013] A noise correction module is used to input the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model; the noise correction model is obtained by training based on the POT deformation field and the second D-InSAR deformation field with the decoherence region and the disentanglement error region removed;

[0014] The near-field deformation prediction module is used to input the twisted phase of the POT low-noise deformation field and the D-InSAR deformation field into the near-field deformation prediction model to obtain the near-field deformation output by the near-field deformation prediction model; the near-field deformation prediction model is used to resolve the near-field deformation of the unwrapping error in the D-InSAR deformation field;

[0015] The full-scale deformation prediction module is used to input the spatial information of pixel points and the spatial relationship data of the epicenter into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model; the spatial information of pixel points and the spatial relationship data of the epicenter reflect the spatial distribution law of the co-seismic deformation field. The full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

[0016] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a noise correction model training module, and the noise correction model training module is used to: based on the coherence parameter and the interference pattern, remove the decoherent area and the disentanglement error area from the first D-InSAR deformation field to obtain the second D-InSAR deformation field; obtain the POT range deformation field; construct an initial noise correction model, use the POT range deformation field as the input of the initial noise correction model, use the second D-InSAR deformation field as the true value, and train the initial noise correction model until the model converges to obtain the noise correction model; wherein, the noise correction model is used to correct the high-dimensional noise in the POT range deformation field.

[0017] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a near-field deformation prediction model training module, and the near-field deformation prediction model training module is used to: based on the coherence parameter and the interference pattern, remove the decoherent area and the unwrapping error area from the first D-InSAR deformation field to obtain the second D-InSAR deformation field; obtain the POT low-noise deformation field output by the noise correction model; construct an initial near-field deformation prediction model, take the entanglement phase of the D-InSAR deformation field and the POT low-noise deformation field as input, take the second D-InSAR deformation field as the true value, and train the initial near-field deformation prediction model until the model converges to obtain the near-field deformation prediction model.

[0018] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a training module of a full-scale deformation prediction model, and the training module of the full-scale deformation prediction model is used to: obtain the POT low-noise deformation field output by the noise correction model, and the near-field deformation output by the near-field deformation prediction model; construct an initial full-scale deformation prediction model, take the spatial information of the pixel points and the spatial relationship data of the epicenter as input, take the fusion result of the POT low-noise deformation field and the near-field deformation as the true value, and train the initial full-scale deformation prediction model until the model converges to obtain the full-scale deformation prediction model.

[0019] As an optional implementation of the embodiment of the present application, the D-InSAR deformation field and the POT deformation field are correlated with the spatial information of the pixel points and the spatial relationship of the epicenter; the device also includes a model construction and training module, which is used to construct a noise correction model, a near-field deformation prediction model and a full-scale deformation prediction model based on the correlation and regression prediction algorithm, and train them separately until the model converges.

[0020] As an optional implementation of an embodiment of the present application, the unwrapping error region includes a phase island region surrounded by a decoherent region, and a unwrapping error region caused by not satisfying the phase continuity criterion; the training module of the noise correction model or the training module of the near-field deformation prediction model, when removing the decoherent region and the unwrapping error region from the first D-InSAR deformation field based on the coherence parameter and the interference pattern to obtain the second D-InSAR deformation field, extracts the decoherent region from the first D-InSAR deformation field through threshold segmentation based on the coherence parameter; extracts the phase island region surrounded by the decoherent region from the first D-InSAR deformation field; and, based on the phase gradient reflected by the interference pattern and the epicenter position, extracts the unwrapping error region caused by not satisfying the phase continuity criterion in the first D-InSAR deformation field, thereby removing the decoherent region, the phase island region, and the unwrapping error region caused by not satisfying the phase continuity criterion from the first D-InSAR deformation field to obtain the second D-InSAR deformation field.

[0021] In a third aspect, the present application provides an electronic device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method for solving the co-seismic deformation field as described in the first aspect or any one of its optional embodiments.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium, comprising: a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the method for solving the co-seismic deformation field as described in the first aspect or any one of its optional embodiments.

[0023] In a fifth aspect, the present application provides a computer program product, comprising: the computer program product comprises a computer program, and when the computer program runs on a computer, the computer implements the method for solving the co-seismic deformation field as described in the first aspect or any one of its optional embodiments.

[0024] Compared with the prior art, the technical solution provided by the embodiments of the present application has the following advantages:

[0025] The embodiment of the present application provides a method, device, equipment, medium and product for solving the co-seismic deformation field, wherein the method obtains the D-InSAR deformation field and the POT deformation field, both of which have certain problems. The present application first inputs the POT deformation field into a noise correction model to obtain the POT low-noise deformation field output by the noise correction model to correct the noise contained in the POT deformation field, wherein the noise correction model is obtained by training the POT deformation field and a second D-InSAR deformation field with the decoherence region and the disentanglement error region removed; then the POT low-noise deformation field is compared with the D-I The winding phase of the nSAR deformation field is used as the input of the near-field deformation prediction model to obtain the near-field large gradient deformation, wherein the near-field deformation prediction model is used to resolve the near-field deformation of the winding error in the D-InSAR deformation field; further, the spatial information of the pixel points reflecting the spatial distribution law of the co-seismic deformation field and the epicenter relationship data are input into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model, and the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

[0026] Thus, this application takes into account the problems of high noise in POT technology and phase unwrapping errors in D-InSAR technology. Based on deep learning methods, the D-InSAR deformation field is first used to correct the noisy POT deformation field, and then the near-field large gradient deformation is solved by combining the wrapped phase of the D-InSAR deformation field to solve the unwrapping error problem caused by the deformation gradient of adjacent pixels exceeding half the wavelength and the phase island phenomenon. Then, the POT deformation field and the near-field large gradient deformation are combined to fill in the missing area of ​​the second D-InSAR deformation field, and a complete co-seismic deformation field with low noise and correct unwrapping is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0029] Figure 1 This is a schematic diagram of the main process of D-InSAR technology;

[0030] Figure 2 It is a flowchart of the phase unwrapping method based on deep learning;

[0031] Figure 3 It is a schematic diagram of the main process of POT technology;

[0032] Figure 4 A schematic diagram of a flow chart of a method for solving a co-seismic deformation field provided in an embodiment of the present application;

[0033] Figure 5 It is a schematic diagram of the BP neural network structure;

[0034] Fig. 6A This is a schematic diagram for comparing the co-seismic deformation fields calculated using different methods;

[0035] Figure 6B A schematic diagram of a learning curve and a residual curve of a method for calculating a co-seismic deformation field provided in an embodiment of the present application;

[0036] Figure 7 A schematic diagram of the structure of a co-seismic deformation field solving device provided in an embodiment of the present application;

[0037] Figure 8 A schematic diagram of the structure of an electronic device described in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the technical terms required to be used in the embodiments or the prior art description are briefly introduced below:

[0039] The main process of D-InSAR technology is as follows: Figure 1 As shown, there are seven main steps that are usually followed:

[0040] (1) Single-look imaging: The original echo signals received by the SAR sensor, including pre-earthquake SAR images and post-earthquake SAR images, are processed by SAR imaging algorithms to obtain single-look complex (SLC) images with high spatial resolution and phase information. SLC images are SAR images formed using only a section of synthetic aperture length. SLC images include SLC main images and SLC auxiliary images. In this step, the imaging quality is ensured by precise Doppler parameter estimation, azimuth frequency modulation estimation, etc.

[0041] (2) Pre-filtering: Reduce the noise level of the SLC image through spectral analysis. This step removes the noise generated by wave number shift through a complex bandpass filter and processes the Doppler center frequency difference, which is crucial to maintaining the coherence of the image;

[0042] (3) Registration and resampling of primary and secondary images: To ensure the high spatial consistency of the SLC images obtained from the two observations, advanced algorithms such as the coherence coefficient method or the least squares method based on amplitude images are usually used to achieve sub-pixel registration accuracy.

[0043] (4) Generate differential interference pattern: The main and auxiliary SLC images generate interference patterns through conjugate multiplication, which contain key information of surface deformation. After the interference pattern is generated, it is necessary to estimate the baseline and remove the flat ground phase and terrain phase, so as to separate the real surface deformation signal and generate a coherence coefficient map. Among them, the baseline estimation adopts methods such as orbital parameters, interference fringe frequency or external information, while the removal of flat ground and terrain phase uses the existing digital elevation model (Digital Elevation Model, DEM) data or other geometric relationship models; among them, DEM is a physical ground model that realizes digital simulation of ground terrain through limited terrain elevation data. DEM uses a set of ordered numerical arrays to represent ground elevation.

[0044] (5) Post-filtering: By using various algorithms such as adaptive filtering and low-pass filtering, the coherence coefficient map is filtered to reduce interference noise, improve the signal-to-noise ratio, and make the phase fringes clearer and more continuous;

[0045] (6) Phase unwrapping: The interference phase is restored to a continuous phase through different unwrapping methods. The unwrapping algorithm needs to deal with the periodicity of the main phase value and consider spatial consistency and accuracy. The methods that can be used in this process include path tracking, complex algorithms based on minimum norm and network planning to ensure accurate recovery of phase information;

[0046] Among them, algorithms based on path tracking, such as Goldstein branch cutting method, quality map method, mask method, etc., realize phase unwrapping by selecting appropriate integral paths to integrate the phase gradients of adjacent pixels; algorithms based on minimum norm, such as least squares, realize phase unwrapping by minimizing the difference between the entangled phase gradient and the true phase gradient; network cost flow based on network theory converts the phase unwrapping problem into a problem of calculating the minimum cost network flow, and limits the phase error transmission in low-quality areas by minimizing the difference between the discrete partial derivatives of the unwrapped phase and the entangled phase, so as to obtain the global optimal solution. The unwrapped phase is based on the entangled phase. Add or subtract an integer multiple of 2π cycles. However, the existing methods do not completely solve the phase unwrapping problem. Under the limitation of the phase continuity assumption, unwrapping errors still occur in low-coherence or phase-dense areas.

[0047] (7) Deformation information extraction: The unwrapped phase change is converted into the deformation value of the ground point, and the points with high coherence in the stable area are selected as references to restore the deformation value of each pixel. Finally, geocoding is required to convert the deformation or elevation information from the SAR coordinate system to the geographic coordinate system.

[0048] At present, the phase unwrapping methods widely used in D-InSAR to solve the coseismic deformation field (such as phase unwrapping methods based on path tracking, phase unwrapping methods based on minimum norm, phase unwrapping methods based on network planning, etc.) all adopt the phase continuity assumption. However, if the deformation gradient exceeds half a wavelength, it is likely to cause phase unwrapping errors. In the near-field large gradient deformation area, the phase continuity assumption may not be met; and when the phase information difference between the pre-seismic and post-seismic images is too large, decoherence will occur. Decoherence is the phenomenon that the phase of the ground echo obtained by two imagings changes due to various factors, resulting in reduced coherence. This phenomenon is mainly affected by factors such as spatial baseline, thermal noise and changes in scattering characteristics. Phase coherence is an important indicator for evaluating the reliability of InSAR interferometric phase. Once decoherence occurs, the InSAR deformation monitoring value of the corresponding area is unreliable. When there are many decoherent areas, it is difficult to find a suitable integral path to solve the entangled phase (that is, the phase island phenomenon occurs), because the phase island area surrounded by a large area of ​​decoherent areas is also difficult to be correctly unwrapped. The deformation field unwrapped based on the above method may be wrong. Therefore, in order to solve the problems existing in D-InSAR, a phase unwrapping method based on deep learning has emerged in recent years, which can perform unwrapping without following the phase continuity criterion.

[0049] Deep learning is essentially a data-driven method for solving nonlinear mapping relationships between different parameters, which mainly includes the following four steps:

[0050] (1) Data selection: To prevent the low generalization of the model caused by overfitting and the lack of accuracy caused by underfitting, the experimental samples need to be randomly divided into three parts according to a fixed ratio, namely training set (60%), validation set (20%), and test set (20%).

[0051] (2) Data modeling: Use the training set data to build a simulation model of variable mapping relationships, and apply the simulation model to predict output parameters from input parameters;

[0052] (3) Debugging the model: Based on the validation set data, optimize the model parameters and improve the model performance according to the deviation between the predicted data and the true value;

[0053] (4) Testing model: Use test set data to test model accuracy and evaluate model performance.

[0054] Many current studies apply deep learning-based phase unwrapping methods to phase unwrapping in D-InSAR processing to avoid errors caused by the inconsistency between the continuity assumption and the actual situation. The main process is as follows: Figure 2 As shown:

[0055] (1) Using seismic waves or other prior information to roughly give a deformation field as the true value to simulate the coseismic deformation field;

[0056] (2) Then, random noise is added through mathematical models such as Gaussian noise and salt and pepper noise to simulate decoherence and wrap it back to simulate the wrapping phase (signal received by the radar);

[0057] (3) Then, through the deep learning model, the mapping relationship between the winding phase and the coseismic deformation field is constructed.

[0058] (4) After evaluating the accuracy of the model using test data, the model is applied to actual untangling, the tangling phase is input, and the deformation field is predicted based on the model.

[0059] However, for this type of method, the difference between the simulated deformation and the real deformation field, as well as the difference between the simulated noise and the actual noise, the accumulated errors of the two will have a great interference on the deformation field results, that is, the quality of the simulation results largely determines the reliability of the deformation solution results, but the prior information is difficult to fully and accurately reflect the laws of the deformation field and errors. Therefore, in practical applications, this method has poor generalization and is difficult to be widely used.

[0060] The POT technique calculates the co-seismic deformation field by comparing the deviations of pre- and post-seismic images. Figure 3 As shown:

[0061] (1) Image registration and resampling: Select two SAR images before and after the earthquake event, extract the initial pixel offset value based on the orbital state parameters, and then register the pre- and post-earthquake data with the DEM data, and resample the post-earthquake data to the same size as the pre-earthquake data;

[0062] (2) Rough determination of the overall offset: The offset between the primary and secondary images is estimated using the image intensity cross-correlation algorithm, and the coefficients of the fitted bilinear polynomial are calculated based on this offset field to determine the overall offset between the two images;

[0063] (3) Accurate determination of the offset: oversample the image pixels to improve the accuracy to the sub-pixel level (theoretical accuracy can reach 1 / 10-1 / 30 pixel). The oversampled image is cross-correlated with the set search window size and a certain step size using a window search-based precision registration method to obtain a more accurate overall offset.

[0064] (4) Calculation of local offset: A polynomial is used to fit the track offset and terrain undulation offset. The offset fitted by the polynomial model is then deducted from the overall offset in the previous step to obtain the offset related to the local surface deformation.

[0065] (5) Offset conversion: The real and imaginary parts of the offset residual are converted into surface displacements in the range and azimuth directions, respectively.

[0066] For the deformation field solved by POT, the following may occur: because the near-field deformation is too large, the surface will be cracked, which will lead to errors in matching the same-name ground points; in the far-field small deformation area, the ground object is homogeneous in the area with small deformation, because the difference in the strength characteristics of the ground objects is not obvious, resulting in matching errors, which in turn leads to large noise in the deformation results solved by POT.

[0067] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the scheme of the present application will be further described below. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0068] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only part of the embodiments of the present application, rather than all of the embodiments.

[0069] In order to solve some or all of the technical problems existing in the related art, the embodiments of the present application provide a method, device, equipment, medium and product for solving the co-seismic deformation field, wherein the method takes into account the problems of high noise in POT technology and phase unwrapping errors in D-InSAR technology, and based on deep learning means, the D-InSAR deformation field is first used to correct the noisy POT deformation field, and then the near-field large gradient deformation is solved in combination with the wrapped phase of the D-InSAR deformation field to solve the unwrapping error problem caused by the deformation gradient of adjacent pixels exceeding half the wavelength and the phase island phenomenon, and then the POT deformation field and the near-field large gradient deformation are combined to fill the missing area of ​​the second D-InSAR deformation field, and a complete co-seismic deformation field with low noise and correct unwrapping is obtained by solution.

[0070] A method for solving a co-seismic deformation field provided in an embodiment of the present application can be implemented by a co-seismic deformation field solving device or an electronic device, and the electronic device includes but is not limited to a vehicle terminal, a server, a personal computer, a laptop computer, a tablet computer, a smart phone, etc. The operating system of the electronic device may include Android, a mobile operating system (iOS) developed by Apple, an operating system (Windows) developed by Microsoft Corporation of the United States, etc., and the embodiment of the present application does not limit this. The electronic device can be run alone to implement the present application, or it can be connected to a network and implement the present application through interactive operations with other computer devices in the network. Among them, the network in which the electronic device is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN) network, etc.

[0071] It should be noted that the protection scope of the method for solving the co-seismic deformation field described in the embodiment of the present application is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present application are included in the protection scope of the present application.

[0072] Figure 4 The flowchart of a method for solving a co-seismic deformation field according to an embodiment of the present application is shown in FIG. 1 , which can be executed by a co-seismic deformation field solving device, wherein the device can be implemented by software and / or hardware and can generally be integrated in an electronic device. Figure 4 As shown, the method mainly includes the following steps S401 to S404:

[0073] S401. Acquire a first D-InSAR deformation field and a POT deformation field.

[0074] In some embodiments, the D-InSAR deformation field is obtained by processing the pre-seismic SAR image and the post-seismic SAR image based on the D-InSAR technology; the POT deformation field is obtained by processing the pre-seismic SAR image and the post-seismic SAR image based on the POT technology. Among them, the POT deformation field includes the POT range deformation field and the POT azimuth deformation field. The range direction refers to the direction along which the radar wave is emitted, that is, the straight-line distance direction between the radar and the target object. The azimuth direction refers to the direction along which the radar moves, that is, the direction along the track.

[0075] Specifically, first obtain the pre-earthquake SAR image and the post-earthquake SAR image, and use the D-InSAR technology and the POT technology to solve and obtain the corresponding co-seismic deformation field. Based on the above description, the D-InSAR technology has the problem of phase unwrapping errors, and the POT technology has noise caused by window matching errors because the phase data of satellite monitoring is more sensitive to surface deformation than the intensity data. In response to these problems, the present application performs post-processing on the first D-InSAR deformation field and the POT deformation field as S402 to S404.

[0076] S402, inputting the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model.

[0077] Among them, the noise correction model is obtained by training based on the POT deformation field and the second D-InSAR deformation field with the decoherence region and the disentanglement error region removed; the decoherence region is the region with decoherence in the first D-InSAR deformation field, and the disentanglement error region is the region with phase disentanglement error in the first D-InSAR deformation field, which can be obtained through detection; after removing the decoherence region and the disentanglement error region, the coseismic deformation contained in the second D-InSAR deformation field is correct, which can be used as the theoretical truth value for model training. The noise correction model is used to correct the high-dimensional noise of the POT deformation field, specifically to correct the high-dimensional noise with a certain distribution pattern in the geographic space caused by the matching error of the same-name pixel points in the POT distance deformation field.

[0078] In some embodiments, the D-InSAR deformation field and the POT deformation field have a correlation with the spatial information of the pixel points and the spatial relationship of the epicenter. The D-InSAR deformation field includes a first D-InSAR deformation field and a second D-InSAR deformation field. The spatial information of the pixel points can be the x, y coordinates of each pixel point in the SAR image, and the spatial relationship data of the epicenter of the pixel points can be the distance between each pixel point in the SAR image and the epicenter. The spatial information of the pixel points and the spatial relationship data of the epicenter are used to reflect the spatial distribution law of the co-seismic deformation field.

[0079] After executing step S401 and before step S402, a noise correction model, a near-field deformation prediction model and a full-scale deformation prediction model are first constructed based on the correlation and the regression prediction algorithm. Therefore, unlike the traditional CNN network for images, this application takes into account that the POT deformation field and the D-InSAR deformation field are highly correlated with the spatial information of the pixel points and the spatial relationship with the epicenter, and adopts the idea of ​​regression prediction to process the raster image as a series of pixel points to construct the above model. The constructed model can be used as an initial model, and then trained according to the subsequent corresponding training steps until convergence before application.

[0080] In some embodiments, the process of training the noise correction model includes: first, based on the coherence parameter and the interference pattern, removing the decoherence area and the unwrapping error area from the first D-InSAR deformation field to obtain a second D-InSAR deformation field; obtaining the POT deformation field, specifically, obtaining the POT range deformation field; and then constructing an initial noise correction model, using the POT range deformation field as the input of the initial noise correction model, and using the second D-InSAR deformation field with the decoherence area and the unwrapping error area removed as the true value, and training the initial noise correction model until the model converges to obtain the noise correction model.

[0081] Among them, the coherence parameters include DEM data and coherence coefficient. DEM data and coherence coefficient can refer to the prior art, and this application will not repeat them here. The interference pattern reflects the phase gradient and epicenter position, and also contains surface deformation information. Usually, the interference fringes in the epicenter area are dense.

[0082] In some embodiments, the unwrapping error region includes a phase island region surrounded by a decoherent region and an unwrapping error region caused by not satisfying the phase continuity criterion; the process of obtaining the second D-InSAR deformation field includes: firstly extracting the decoherent region from the first D-InSAR deformation field by threshold segmentation based on the coherence parameter of the image (SAR image), then extracting the phase island region surrounded by the decoherent region, and extracting the unwrapping error region caused by not satisfying the phase continuity criterion based on the phase gradient and epicenter position reflected by the interference pattern, so that all the unwrapping error regions are extracted, and further, removing the decoherent region and all the unwrapping error regions (including the phase island region and the unwrapping error region that does not satisfy the phase continuity criterion) from the first D-InSAR deformation field to obtain the second D-InSAR deformation field. Thus, the second D-InSAR deformation field containing all correct coseismic deformations is used as the theoretical truth value for model training.

[0083] Optionally, an initial noise correction model is constructed based on a deep learning regression model. The POT distance deformation field is input into the initial noise correction model to obtain a predicted value output by the initial noise correction model. The predicted value is the result predicted by the model of removing high-dimensional noise in the POT distance deformation field; the second D-InSAR deformation field after removing the decoherence area and the unwrapping error area is the true value, the error between the true value and the predicted value is calculated, and the model parameters of the initial noise correction model are optimized until a converged noise correction model is obtained. The converged noise correction model reflects the high-dimensional mapping relationship between the noisy value and the true value. Among them, the deep learning regression model can be a multi-layer feedforward neural network (Back Propagation, BP). Figure 5 As shown in the figure, the BP neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm. Its training process mainly includes: network initialization, forward calculation, error back propagation, weight threshold update and judgment whether the iteration is completed.

[0084] Specifically, (1) during the forward pass, the input data (such as Figure 5 x1, x2, x3, x4…, xn) in the input layer are processed through the hidden layer to the output layer, simulating the nonlinear relationship between the variables and obtaining the model prediction results based on the relationship (such as Figure 5 y in the result); (2) The model will then automatically calculate the error between the result and the expected output. If the error exceeds the preset range, the model parameters will be adjusted in the opposite direction; (3) This cycle will be repeated until the error is within an acceptable range and the training is terminated, thus obtaining a model with a better fitting effect; (4) Because the model simulates the mapping relationship between input and output parameters, the known input information can be used to predict the output value of the unknown area.

[0085] After the converged noise correction model is obtained through training, the noisy POT deformation field is used as input to obtain the output POT low-noise deformation field.

[0086] In the above embodiment, after removing the incoherent region and the unwrapping error region of the D-InSAR deformation field, based on the deep learning technology, the noisy POT deformation field is constrained by using the second D-InSAR deformation field with the incoherent region and the unwrapping error region removed, so that the noise problem existing in the POT deformation field is corrected by using the D-InSAR deformation field to correct the near-field error of the POT deformation field, effectively suppress the noise in the POT deformation field, and obtain a POT low-noise deformation field that can reflect the near-field deformation.

[0087] S403, inputting the twisted phases of the POT low-noise deformation field and the D-InSAR deformation field into a near-field deformation prediction model to obtain the near-field deformation output by the near-field deformation prediction model.

[0088] Among them, the near-field deformation prediction model is used to solve the near-field deformation of the D-InSAR deformation field. In the traditional D-InSAR deformation field, the area with a large deformation gradient close to the epicenter is prone to unwrapping errors. The near-field deformation prediction model in this application can be combined with the POT low-noise deformation field with improved accuracy to accurately solve the near-field large gradient deformation of the D-InSAR deformation field. The POT low-noise deformation field contains the correct near-field deformation. The entanglement phase of the D-InSAR deformation field is obtained by monitoring the D-InSAR deformation field. It should be noted that although the D-InSAR deformation field has the problem of unwrapping errors, the information reflected by its entanglement phase is correct.

[0089] In some embodiments, the training process of the near-field deformation prediction model includes: first, based on the coherence parameter and the interference pattern, the decoherent region and the disentanglement error region are removed from the first D-InSAR deformation field to obtain the second D-InSAR deformation field. The specific implementation method of obtaining the second D-InSAR deformation field can refer to the above content and will not be repeated here; the POT low-noise deformation field output by the noise correction model is obtained, and then the initial near-field deformation prediction model is constructed, the POT low-noise deformation field is used as input, the winding phase of the D-InSAR deformation field is used as a constraint condition, and the second D-InSAR deformation field with the decoherent region and the disentanglement error region removed is used as the true value, and the initial near-field deformation prediction model is trained until the model converges to obtain the near-field deformation prediction model. Among them, the initial near-field deformation prediction model can be constructed based on the deep learning regression model; the winding phase of the D-InSAR deformation field is used as a strong constraint to make the near-field deformation prediction model reflect the actual situation as much as possible.

[0090] When training the initial near-field deformation prediction model, the winding phase of the POT low-noise deformation field and the D-InSAR deformation field are input to obtain the predicted value of the model output, which is the near-field large gradient deformation predicted by the model; the error between the true value and the predicted value is calculated, and the model parameters of the initial near-field deformation prediction model are optimized until a converged near-field deformation prediction model is obtained, which can predict the near-field large gradient deformation of the D-InSAR deformation field.

[0091] After the converged near-field deformation prediction model is trained, the entangled phases of the POT low-noise deformation field and the D-InSAR deformation field are taken as input to obtain the near-field deformation output by the near-field deformation prediction model.

[0092] The above embodiment applies the entanglement phase obtained by monitoring and introduces the POT low-noise deformation field obtained in step S402. Based on deep learning technology, the near-field large gradient deformation is accurately solved, thereby solving the D-InSAR deformation field unwrapping error problem caused by the deformation gradient of adjacent pixels exceeding half the wavelength and the phase island phenomenon.

[0093] S404: Inputting the spatial information of the pixel points and the spatial relationship data of the epicenter into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model.

[0094] The spatial information of the pixel point can be the x and y coordinates of each pixel point in the SAR image, and the epicenter spatial relationship data of the pixel point can be the distance between each pixel point and the epicenter in the SAR image. The spatial information of the pixel point and the epicenter spatial relationship data are used to reflect the spatial distribution law of the co-seismic deformation field.

[0095] The full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the missing values ​​of the second D-InSAR deformation field based on the spatial distribution law, which is equivalent to fusing the POT low-noise deformation field and the near-field deformation to fill the previously removed incoherent areas and disentangled error areas, thereby obtaining a complete coseismic deformation field. The full-scale deformation prediction model reflects the mapping relationship between the spatial distribution law of the coseismic deformation field and the complete coseismic deformation field.

[0096] In some embodiments, the training process of the full-scale deformation prediction model includes: obtaining the POT low-noise deformation field output by the noise correction model, and the near-field deformation output by the near-field deformation prediction model; constructing an initial full-scale deformation prediction model, taking the spatial information of the pixel points and the spatial relationship data of the epicenter as input, and taking the fusion result of the POT low-noise deformation field and the near-field deformation as the true value, training the initial full-scale deformation prediction model until the model converges to obtain the full-scale deformation prediction model.

[0097] Optionally, an initial full-scale deformation prediction model is constructed based on a deep learning regression model; the POT low-noise deformation field obtained by executing step S402 and the near-field deformation obtained by executing step S403 are fused, and the fusion result is used as the theoretical true value for model training; the pixel spatial information reflecting the spatial distribution law of the co-seismic deformation field and the epicenter spatial relationship data are used as inputs of the initial full-scale deformation prediction model to obtain a predicted value output by the initial full-scale prediction model, which is the predicted full-scale deformation field; the true value and the predicted value are compared, the error between the true value and the predicted value is calculated, and the model parameters of the initial full-scale deformation prediction model are optimized until a converged full-scale deformation prediction model is obtained.

[0098] After the converged full-scale deformation prediction model is trained, the full-scale deformation prediction model is used to fill the incoherence missing values ​​of the second D-InSAR deformation field to supplement the incoherence area and disentanglement error area of ​​the second D-InSAR deformation field.

[0099] It should be noted that the root cause of the problem of traditional phase unwrapping based on deep learning is that the output of the deep learning model, that is, the true value in the physical sense is unknown, that is, the surface deformation is unknown. The traditional method is to simulate the true value approximately with the prior information, and then simulate the noise, and further add the simulated noise to the simulated true value as the observed value, and then obtain the mapping relationship between the observed value and the true value through network training. However, there is a deviation between the simulated value and the true value after all, and this deviation causes the deep learning model to be difficult to evaluate and poor in generalization. However, the present application uses the second D-InSAR deformation field that removes the decoherent area and the unwrapping error area as the theoretical true value (relatively high accuracy) to correct the POT deformation field with lower precision and noise, and then use the corrected POT low-noise deformation field to constrain the near-field deformation information, and further fuse the POT low-noise deformation field and the near-field deformation to obtain the full-scale deformation field. The whole process does not involve simulating data based on mathematical models, so there is no error caused by simulation, which improves the generalization of the model. The application of such a model can solve the co-seismic deformation field with low noise, accuracy and completeness.

[0100] The above embodiment takes into account the spatial distribution law of the co-seismic deformation field and fills the missing values ​​of incoherence based on deep learning technology to correctly fill the incoherent area and the unwrapping error area of ​​the second D-InSAR deformation field. The obtained co-seismic deformation field is complete and accurate, avoids the decoherence phenomenon, and solves the problem of phase unwrapping errors.

[0101] In summary, the embodiment of the present application provides a method for solving a coseismic deformation field. The method obtains a first D-InSAR deformation field and a POT deformation field. Both deformation fields have certain problems. The POT deformation field is first input into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model to correct the noise contained in the POT deformation field, wherein the noise correction model is obtained by training the POT deformation field and a second D-InSAR deformation field with decoherence regions and disentanglement error regions removed; then the POT low-noise deformation field is compared with the D-InSAR deformation field. The entangled phase of the AR deformation field is used as the input of the near-field deformation prediction model to obtain the near-field large gradient deformation, wherein the near-field deformation prediction model is used to resolve the near-field deformation with unwrapping errors in the D-InSAR deformation field; further, the spatial information of the pixel points and the epicenter relationship data are input into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model, and the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation, and consider the spatial distribution law of the co-seismic deformation field to fill the incoherence missing values ​​of the second D-InSAR deformation field.

[0102] Thus, this application takes into account the problems of high noise in POT technology and phase unwrapping errors in D-InSAR technology. Based on deep learning methods, the D-InSAR deformation field is first used to correct the noisy POT deformation field, and then the near-field large gradient deformation is solved by combining the wrapped phase of the D-InSAR deformation field to solve the unwrapping error problem caused by the deformation gradient of adjacent pixels exceeding half the wavelength and the phase island phenomenon. Then, considering the spatial distribution law of the co-seismic deformation field, the POT deformation field and the near-field large gradient deformation are combined to predict and fill in the missing area of ​​the second D-InSAR deformation field, and a low-noise, correctly unwrapped and complete co-seismic deformation field is solved.

[0103] To further illustrate the technical effect of the embodiment of the present application, POT, D-InSAR, and the co-seismic deformation field solution method provided in the embodiment of the present application are used to solve the co-seismic deformation field of the experimental synthetic aperture radar image (SAR image). The solution results are shown in FIG. Fig. 6A As shown, from a qualitative perspective, it can be seen that the POT noise level is high; the D-InSAR results suffer from unwrapping errors due to excessive near-field deformation gradients and the encirclement of decoherent regions; and this application achieves the complementary advantages of the two, while ensuring that the deformation solution is as correct as possible, reducing the noise level of the results as much as possible.

[0104] From a quantitative perspective, the learning curve and residual curve of the co-seismic deformation field solution method provided in the embodiment of the present application are constructed, such as Figure 6B As shown, it can be seen that the errors of the training set and the validation set converge, indicating that the model does not have the problems of overfitting (using an overly complex model) and underfitting (using an overly simple model); analyzing the residual distribution diagram of the test set, it can be concluded that the distribution of the predicted value and the true value is roughly 45°, and, after statistics, the root mean square error is about 5cm, and 95% of the residual absolute values ​​are within 50cm, indicating that most of the prediction results are good.

[0105] like Figure 7 As shown, Figure 7 A schematic diagram of a co-seismic deformation field solving device provided in an embodiment of the present application, the device comprising:

[0106] An acquisition module 701 is used to acquire a first D-InSAR deformation field and a POT deformation field;

[0107] The noise correction module 702 is used to input the POT deformation field into the noise correction model to obtain the POT low-noise deformation field output by the noise correction model; the noise correction model is obtained by training based on the POT deformation field and the second D-InSAR deformation field with the decoherence region and the disentanglement error region removed;

[0108] The near-field deformation prediction module 703 is used to input the twisted phase of the POT low-noise deformation field and the D-InSAR deformation field into the near-field deformation prediction model to obtain the near-field deformation output by the near-field deformation prediction model; the near-field deformation prediction model is used to resolve the near-field deformation of the unwrapping error in the D-InSAR deformation field;

[0109] The full-scale deformation prediction module 704 is used to input the spatial information of the pixel points and the spatial relationship data of the epicenter into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model; the spatial information of the pixel points and the spatial relationship data of the epicenter reflect the spatial distribution law of the co-seismic deformation field. The full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

[0110] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a noise correction model training module, and the noise correction model training module is used to: based on the coherence parameter and the interference pattern, remove the decoherent area and the disentanglement error area from the first D-InSAR deformation field to obtain the second D-InSAR deformation field; obtain the POT range deformation field; construct an initial noise correction model, use the POT range deformation field as the input of the initial noise correction model, use the second D-InSAR deformation field as the true value, and train the initial noise correction model until the model converges to obtain the noise correction model; wherein, the noise correction model is used to correct the high-dimensional noise in the POT range deformation field.

[0111] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a near-field deformation prediction model training module, and the near-field deformation prediction model training module is used to: based on the coherence parameter and the interference pattern, remove the decoherent area and the unwrapping error area from the first D-InSAR deformation field to obtain the second D-InSAR deformation field; obtain the POT low-noise deformation field output by the noise correction model; construct an initial near-field deformation prediction model, take the entanglement phase of the D-InSAR deformation field and the POT low-noise deformation field as input, take the second D-InSAR deformation field as the true value, and train the initial near-field deformation prediction model until the model converges to obtain the near-field deformation prediction model.

[0112] As an optional implementation of an embodiment of the present application, the co-seismic deformation field solving device also includes a training module of a full-scale deformation prediction model, and the training module of the full-scale deformation prediction model is used to: obtain the POT low-noise deformation field output by the noise correction model, and the near-field deformation output by the near-field deformation prediction model; construct an initial full-scale deformation prediction model, take the spatial information of the pixel points and the spatial relationship data of the epicenter as input, take the fusion result of the POT low-noise deformation field and the near-field deformation as the true value, and train the initial full-scale deformation prediction model until the model converges to obtain the full-scale deformation prediction model.

[0113] As an optional implementation of the embodiment of the present application, the D-InSAR deformation field and the POT deformation field are correlated with the spatial information of the pixel points and the spatial relationship of the epicenter; the device also includes a model construction and training module, which is used to construct a noise correction model, a near-field deformation prediction model and a full-scale deformation prediction model based on the correlation and regression prediction algorithm, and train them separately until the model converges.

[0114] As an optional implementation of an embodiment of the present application, the unwrapping error region includes a phase island region surrounded by a decoherent region, and a unwrapping error region caused by not satisfying the phase continuity criterion; the training module of the noise correction model or the training module of the near-field deformation prediction model, when removing the decoherent region and the unwrapping error region from the first D-InSAR deformation field based on the coherence parameter and the interference pattern to obtain the second D-InSAR deformation field, extracts the decoherent region from the first D-InSAR deformation field through threshold segmentation based on the coherence parameter; extracts the phase island region surrounded by the decoherent region from the first D-InSAR deformation field; and, based on the phase gradient reflected by the interference pattern and the epicenter position, extracts the unwrapping error region caused by not satisfying the phase continuity criterion in the first D-InSAR deformation field; removes the decoherent region, the phase island region, and the unwrapping error region caused by not satisfying the phase continuity criterion from the first D-InSAR deformation field to obtain the second D-InSAR deformation field.

[0115] For the specific definition of the co-seismic deformation field solving device, please refer to the definition of the co-seismic deformation field solving method mentioned above, which will not be repeated here. Each module in the above-mentioned co-seismic deformation field solving device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0116] like Figure 8As shown, the embodiment of the present application provides an electronic device, which includes: a processor 801, a memory 802, and a computer program stored in the memory 802 and executable on the processor 801, and when the computer program is executed by the processor 801, each process of the method for solving the co-seismic deformation field in the above method embodiment is implemented. And the same technical effect can be achieved, so it will not be repeated here to avoid repetition.

[0117] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the method for solving the co-seismic deformation field in the above method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0118] The computer readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0119] An embodiment of the present application provides a computer program product, which stores a computer program. When the computer program is executed by a processor, each process of the method for solving the co-seismic deformation field in the above method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0120] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media that include computer-usable program code.

[0121] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0122] In the present application, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0123] In this application, memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0124] In this application, computer-readable media includes permanent and non-permanent, removable and non-removable storage media. Storage media can be implemented by any method or technology to store information, and the information can be computer-readable instructions, data structures, modules of programs or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. According to the definition in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0125] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0126] The above is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for calculating a co-seismic deformation field, characterized in that: include: Obtaining the first synthetic aperture radar differential interferometry D-InSAR deformation field and pixel offset tracking POT deformation field; Inputting the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model; The noise correction model is obtained by training based on the POT deformation field and the second D-InSAR deformation field with the decoherence region and the unwrapping error region removed; Inputting the twisted phases of the POT low-noise deformation field and the D-InSAR deformation field into a near-field deformation prediction model to obtain the near-field deformation output by the near-field deformation prediction model; the near-field deformation prediction model is used to calculate the near-field deformation of the unwrapping error in the D-InSAR deformation field; The spatial information of the pixel points and the spatial relationship data of the epicenter are input into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model; the spatial information of the pixel points and the spatial relationship data of the epicenter reflect the spatial distribution law of the co-seismic deformation field, and the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation, and fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

2. The method according to claim 1, characterized in that: The training process of the noise correction model includes: Based on the coherence parameter and the interference pattern, removing the decoherence region and the unwrapping error region from the first synthetic aperture radar differential interferometry D-InSAR deformation field to obtain the second D-InSAR deformation field; Get the POT distance deformation field; Constructing an initial noise correction model, taking the POT distance deformation field as the input of the initial noise correction model, taking the second D-InSAR deformation field as the true value, and training the initial noise correction model until the model converges to obtain the noise correction model; The noise correction model is used to correct the high-dimensional noise in the POT distance deformation field.

3. The method according to claim 1, characterized in that The training process of the near-field deformation prediction model includes: Based on the coherence parameter and the interference pattern, removing the decoherence region and the unwrapping error region from the first synthetic aperture radar differential interferometry D-InSAR deformation field to obtain the second D-InSAR deformation field; Obtaining a POT low-noise deformation field output by the noise correction model; An initial near-field deformation prediction model is constructed, the winding phase of the D-InSAR deformation field and the POT low-noise deformation field are taken as input, the second D-InSAR deformation field is taken as the true value, and the initial near-field deformation prediction model is trained until the model converges to obtain the near-field deformation prediction model.

4. The method according to claim 1, characterized in that: The training process of the full-scale deformation prediction model includes: Acquire the POT low-noise deformation field output by the noise correction model and the near-field deformation output by the near-field deformation prediction model; An initial full-scale deformation prediction model is constructed, the spatial information of the pixel points and the spatial relationship data of the epicenter are taken as input, the fusion result of the POT low-noise deformation field and the near-field deformation is taken as the true value, and the initial full-scale deformation prediction model is trained until the model converges to obtain the full-scale deformation prediction model.

5. The method according to claim 1, characterized in that The D-InSAR deformation field and the POT deformation field have a correlation with the spatial information of the pixel points and the spatial relationship of the epicenter; After acquiring the first synthetic aperture radar differential interferometry D-InSAR deformation field and the pixel offset tracking POT deformation field, and before inputting the POT deformation field into a noise correction model to obtain the POT low-noise deformation field output by the noise correction model, the method further includes: Based on the correlation and the regression prediction algorithm, the noise correction model, the near-field deformation prediction model and the full-scale deformation prediction model are constructed and trained respectively until the models converge.

6. The method according to claim 2 or 3, characterized in that: The unwrapping error region includes a phase island region surrounded by a decoherent region and an unwrapping error region caused by failure to meet a phase continuity criterion; The method of removing the decoherent region and the unwrapping error region from the first synthetic aperture radar differential interferometry D-InSAR deformation field based on the coherence parameter and the interference pattern to obtain the second D-InSAR deformation field includes: Based on the coherence parameter, extracting the decoherence region from the first synthetic aperture radar differential interferometry D-InSAR deformation field by threshold segmentation; Extracting the phase island region surrounded by the decoherence region from the first synthetic aperture radar differential interferometry D-InSAR deformation field; and extracting the unwrapping error region caused by not satisfying the phase continuity criterion in the first synthetic aperture radar differential interferometry D-InSAR deformation field based on the phase gradient and epicenter position reflected by the interference pattern; The decoherence region, the phase island region, and the unwrapping error region caused by failure to meet the phase continuity criterion are removed from the first synthetic aperture radar differential interferometry D-InSAR deformation field to obtain the second D-InSAR deformation field.

7. A co-seismic deformation field solving device, characterized in that: include: An acquisition module, used for acquiring a first synthetic aperture radar differential interferometry D-InSAR deformation field and a POT deformation field; A noise correction module, used for inputting the POT deformation field into a noise correction model to obtain a POT low-noise deformation field output by the noise correction model; the noise correction model is obtained by training based on the POT deformation field and a second D-InSAR deformation field with decoherence regions and disentanglement error regions removed; A near-field deformation prediction module is used to input the twisted phase of the POT low-noise deformation field and the D-InSAR deformation field into a near-field deformation prediction model to obtain the near-field deformation output by the near-field deformation prediction model; the near-field deformation prediction model is used to resolve the near-field deformation of the unwrapping error in the D-InSAR deformation field; The full-scale deformation prediction module is used to input the spatial information of the pixel points and the spatial relationship data of the epicenter into the full-scale deformation prediction model to obtain the full-scale deformation field output by the full-scale deformation prediction model; the spatial information of the pixel points and the spatial relationship data of the epicenter reflect the spatial distribution law of the co-seismic deformation field, and the full-scale deformation prediction model is used to combine the POT low-noise deformation field and the near-field deformation to fill the incoherence missing values ​​of the second D-InSAR deformation field based on the spatial distribution law.

8. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method for solving the co-seismic deformation field as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: include: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for solving the co-seismic deformation field according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that include: The computer program product comprises a computer program, and when the computer program is run on a computer, the computer is enabled to implement the method for solving the co-seismic deformation field according to any one of claims 1 to 6.

Images