A deep learning-based InSAR co-seismic 3D deformation field recovery method

Abstract

The present invention provides a method for recovering the co-seismic three-dimensional deformation field of InSAR based on deep learning, including: sampling the source parameters according to a preset parameter interval, and obtaining the three-dimensional deformation data of the surface based on the elastic dislocation model; projecting the three-dimensional deformation data of the surface to the LOS direction of the SAR satellite to obtain the one-dimensional deformation data of the satellite ascending and descending LOS directions; training and verifying the deep neural network model according to the three-dimensional deformation data of the surface and the one-dimensional deformation data in the LOS direction; collecting the images of the ascending and descending orbits before and after the earthquake; processing the images of the ascending and descending orbits by using the D-InSAR technology; inputting the one-dimensional deformation data of the ascending and descending LOS directions into the three-dimensional deformation prediction model to obtain the three-dimensional deformation results of the surface predicted by the model. The present invention uses deep learning technology to establish a relationship model between satellite observations and surface deformation, and can recover the three-dimensional deformation of the surface using only the one-dimensional LOS observations of the SAR satellite, without the need for additional observations or prior knowledge constraints.

Description

Technical Field

[0001] The present invention relates to the field of earthquake technology, and in particular to an InSAR co-seismic three-dimensional deformation field recovery method based on deep learning. Background Art

[0002] Earthquakes are extremely destructive natural disasters that can not only cause the collapse of buildings and damage to infrastructure, but can also induce secondary disasters such as landslides and collapses, forming a disaster chain that seriously threatens human life and property. Accurately extracting and analyzing the three-dimensional surface deformation caused by earthquakes (i.e., vertical, east-west, and north-south directions) is of great significance for inferring earthquake-causing structures, fault morphology, and activity characteristics, and helps to gain a deeper understanding of the earthquake-causing mechanism and characteristics of earthquake activity.

[0003] Interferometric Synthetic Aperture Radar (InSAR) technology is a remote sensing technology used to monitor surface changes. It has the advantages of high resolution, all-weather observation and large-area monitoring. It has been widely used in monitoring crustal movement, volcanic activity, surface subsidence and glacier changes.

[0004] However, InSAR technology can only obtain one-dimensional surface deformation along the line of sight (LOS), which is the projection of the three-dimensional surface deformation in the LOS direction. Therefore, at least three different sets of InSAR observations are required to solve the three-dimensional surface deformation. In order to obtain co-seismic three-dimensional deformation, many scholars have proposed adding additional observation equations based on ascending and descending orbit observations, such as adding pixel offset tracking (POT) or multiple aperture radar interferometry (MAI) observation results, adding SAR satellite observation results with different radar imaging geometries (such as Sentinel satellite and ALOS satellite joint observations), introducing GPS observation data or geophysical model constraints, etc. However, the above methods are restricted by many factors in practical applications, such as the limited number of SAR satellites, different monitoring accuracy of different technical methods and differences in the resolution of the acquired data, inconsistent satellite observation time and imaging range, and sparse GPS stations.

[0005] Estimating the actual three-dimensional surface deformation from the one-dimensional deformation observations of a single-orbit InSAR is essentially a rank-deficient problem. At present, there are two main strategies for solving the three-dimensional surface deformation: (1) Based on the conventional ascending and descending InSAR observations, the deformation observations from multiple sources are integrated. The fusion of multi-source heterogeneous observations usually includes: the fusion of multiple (three or more) InSAR line-of-sight data from different angles, the fusion of InSAR line-of-sight and azimuth data, the fusion of InSAR line-of-sight and GPS observation data, etc. (2) For specific observation objects, the requirements for InSAR multi-angle observations are reduced by introducing the kinematic model or prior assumptions of the observation object as constraints. For example, if it is assumed that the glacier moves parallel to the surface, then only the ascending and descending data are needed to invert the three-dimensional deformation rate of the glacier.

[0006] The existing strategies for solving the three-dimensional deformation of the surface mainly use the pure geometric relationship between InSAR observations and three-dimensional deformation to solve the three-dimensional deformation of the surface from a purely mathematical perspective. Therefore, in the process of solving the three-dimensional deformation, additional observations or prior knowledge must be used as constraints. In practical applications, additional observations are usually difficult to obtain due to factors such as the limited number of SAR satellites, inconsistent satellite observation time and imaging range, and sparse GPS stations. Prior knowledge or geophysical models are only applicable to certain specific scenarios and are difficult to apply to unknown research areas. In addition, they have the disadvantages of strong subjectivity and high uncertainty. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide an InSAR co-seismic three-dimensional deformation field recovery method based on deep learning, so as to overcome the stringent requirements of the existing InSAR three-dimensional deformation solution method on observation data, and restore the co-seismic three-dimensional deformation field from a completely new perspective without the need for additional observations or prior knowledge constraints.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] A deep learning-based InSAR co-seismic three-dimensional deformation field recovery method, including:

[0010] The source parameters are sampled according to the preset parameter intervals, and the three-dimensional surface deformation data are obtained based on the Okada elastic dislocation model.

[0011] According to the SAR satellite imaging geometric parameters, the three-dimensional surface deformation data is projected to the SAR satellite LOS direction to obtain the SAR satellite ascending and descending LOS direction one-dimensional deformation data;

[0012] The deep neural network model is trained and verified according to the three-dimensional deformation data of the surface and the one-dimensional deformation data in the LOS direction to obtain a three-dimensional deformation prediction model;

[0013] For actual earthquake events, collect SAR images of ascending and descending orbits before and after the earthquake;

[0014] The D-InSAR technology is used to process the SAR images of ascending and descending orbits, and the one-dimensional deformation data in the ascending and descending LOS directions are inverted.

[0015] The one-dimensional deformation data in the ascending orbit and descending orbit LOS directions are input into the three-dimensional deformation prediction model to obtain the three-dimensional surface deformation results predicted by the model.

[0016] Preferably, the deep neural network model is a U-NET deep neural network model.

[0017] Preferably, the deep neural network model is trained and verified according to the three-dimensional deformation data of the ground surface and the one-dimensional deformation data in the LOS direction to obtain a three-dimensional deformation prediction model, including:

[0018] Dividing the three-dimensional surface deformation data and the one-dimensional deformation data in the LOS direction according to a preset ratio to obtain a training data set and a test data set of the model;

[0019] Building a deep neural network model; the input of the deep neural network model is one-dimensional deformation data in the ascending orbit and descending orbit LOS directions, and the output of the deep neural network model is three-dimensional deformation data of the surface;

[0020] Use the training data set to train the deep neural network model and optimize the parameters of the deep neural network model;

[0021] The deep neural network model is tested using a test data set, and the trained three-dimensional deformation prediction model is saved.

[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0023] The invention provides an InSAR co-seismic three-dimensional deformation field recovery method based on deep learning, comprising: sampling source parameters according to preset parameter intervals, and obtaining three-dimensional surface deformation data based on the Okada elastic dislocation model; projecting the three-dimensional surface deformation data to the LOS direction of the SAR satellite according to the SAR satellite imaging geometric parameters, and obtaining one-dimensional deformation data in the LOS direction of the SAR satellite ascending and descending orbits; training and verifying a deep neural network model according to the three-dimensional surface deformation data and the one-dimensional deformation data in the LOS direction, and obtaining a three-dimensional deformation prediction model; for actual earthquake events, collecting SAR images of ascending and descending orbits before and after the earthquake; processing the SAR images of ascending and descending orbits by using the D-InSAR technology, and inverting to obtain one-dimensional deformation data in the LOS direction of the ascending and descending orbits; inputting the one-dimensional deformation data in the LOS direction of the ascending and descending orbits into the three-dimensional deformation prediction model, and obtaining a three-dimensional surface deformation result predicted by the model. The present invention uses deep learning technology to establish a relationship model between satellite observations and surface deformation. The relationship model can restore the three-dimensional surface deformation using only the one-dimensional LOS observations of the SAR satellite, without the need for additional observations or prior knowledge constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0025] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0026] Figure 2 A technical flow chart provided for an embodiment of the present invention;

[0027] Figure 3 A structural diagram of the U-NET deep neural network model provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for recovering InSAR co-seismic three-dimensional deformation field based on deep learning, comprising:

[0031] Step 100: sampling source parameters according to a preset parameter interval, and obtaining three-dimensional surface deformation data based on the Okada elastic dislocation model;

[0032] Step 200: projecting the three-dimensional deformation data of the ground surface to the LOS direction of the SAR satellite according to the SAR satellite imaging geometric parameters, and obtaining the one-dimensional deformation data in the LOS direction of the SAR satellite ascending and descending orbits;

[0033] Step 300: training and verifying a deep neural network model according to the three-dimensional deformation data of the ground surface and the one-dimensional deformation data in the LOS direction to obtain a three-dimensional deformation prediction model;

[0034] Step 400: for an actual earthquake event, collect SAR images of ascending and descending orbits before and after the earthquake;

[0035] Step 500: Process the SAR images of the ascending and descending orbits using the D-InSAR technology, and invert to obtain the one-dimensional deformation data in the ascending and descending orbit LOS directions;

[0036] Step 600: input the one-dimensional deformation data in the ascending orbit and descending orbit LOS directions into the three-dimensional deformation prediction model to obtain the three-dimensional surface deformation results predicted by the model.

[0037] like Figure 2 As shown, the technical route of this embodiment is as follows:

[0038] A deep learning-based InSAR co-seismic three-dimensional deformation field recovery method includes the following steps:

[0039] Process 1: Sample the source parameters according to a certain parameter interval, and obtain the three-dimensional surface deformation data based on the Okada elastic dislocation model;

[0040] Process 2: According to the SAR satellite imaging geometric parameters, the three-dimensional deformation obtained by forward modeling is projected to the SAR satellite LOS direction to obtain the one-dimensional deformation data of the SAR satellite ascending and descending LOS directions;

[0041] Process 3: Divide the three-dimensional deformation data of the surface and the one-dimensional deformation data in the LOS direction according to a certain ratio as the training data set and test data set of the deep learning model;

[0042] Process 4: Figure 3As shown, in this embodiment, a U-NET deep neural network model is built, the model input is the one-dimensional deformation data in the ascending and descending LOS directions, and the model output is the three-dimensional deformation data of the surface;

[0043] Process 5: Use the training data set to train the deep neural network model and optimize the parameters of the deep neural network model;

[0044] Process 6: Use the test data set to test the deep neural network model and save the trained deep neural network model;

[0045] Process 7: For actual earthquake events, collect SAR images of ascending and descending orbits before and after the earthquake;

[0046] Process 8: Use D-InSAR technology to process the SAR images of ascending and descending orbits, and invert the one-dimensional deformation data in the ascending and descending LOS directions;

[0047] Process 9: Input the one-dimensional deformation data in the ascending and descending LOS directions into the deep neural network model to obtain the three-dimensional surface deformation results predicted by the model.

[0048] The beneficial effects of the present invention are as follows:

[0049] The present invention establishes a relationship model between SAR satellite observations and real surface three-dimensional deformation. Without introducing additional observation quantities or prior knowledge, it can achieve rapid estimation of surface three-dimensional deformation, overcome the strict requirements of traditional methods for observation data, improve the efficiency of solving surface three-dimensional deformation, and reduce the uncertainty of solution results.

[0050] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0051] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A deep learning-based InSAR co-seismic three-dimensional deformation field recovery method, characterized in that: include: The source parameters are sampled according to the preset parameter intervals, and the three-dimensional surface deformation data are obtained based on the Okada elastic dislocation model. According to the SAR satellite imaging geometric parameters, the three-dimensional surface deformation data is projected to the SAR satellite LOS direction to obtain the SAR satellite ascending and descending LOS direction one-dimensional deformation data; The deep neural network model is trained and verified according to the three-dimensional deformation data of the surface and the one-dimensional deformation data in the LOS direction to obtain a three-dimensional deformation prediction model; For actual earthquake events, collect SAR images of ascending and descending orbits before and after the earthquake; The D-InSAR technology is used to process the SAR images of ascending and descending orbits, and the one-dimensional deformation data in the ascending and descending LOS directions are inverted. Inputting the one-dimensional deformation data in the ascending orbit and descending orbit LOS directions into the three-dimensional deformation prediction model to obtain the three-dimensional surface deformation results predicted by the model; The deep neural network model is trained and verified according to the three-dimensional deformation data of the surface and the one-dimensional deformation data in the LOS direction to obtain a three-dimensional deformation prediction model, including: Dividing the three-dimensional surface deformation data and the one-dimensional deformation data in the LOS direction according to a preset ratio to obtain a training data set and a test data set of the model; Building a deep neural network model; the input of the deep neural network model is one-dimensional deformation data in the ascending orbit and descending orbit LOS directions, and the output of the deep neural network model is three-dimensional deformation data of the surface; Use the training data set to train the deep neural network model and optimize the parameters of the deep neural network model; The deep neural network model is tested using a test data set, and the trained three-dimensional deformation prediction model is saved.

2. The InSAR co-seismic three-dimensional deformation field recovery method based on deep learning according to claim 1, characterized in that: The deep neural network model is a U-NET deep neural network model.

Images