Three-dimensional flow velocity field inversion method and system based on coherence-assisted sar adaptive offset tracking and storage medium

Abstract

The application provides a three-dimensional flow velocity field inversion method and system based on coherence auxiliary SAR adaptive offset tracking and a storage medium to improve the de-coherence problem of synthetic aperture radar interferometry (InSAR) technology caused by time heterogeneity, and relates to the technical field of power engineering geological disaster monitoring. The method comprises the following steps: establishing a space-time baseline network between SAR images; obtaining the coherence data after geographical coding; selecting a pair of SAR image pairs from the space-time baseline network as a reference unit, and calculating the offset result through a pixel offset tracking technology for the reference unit, and determining the best matching window parameters thereof; establishing a mapping function between the coherence and the best matching window parameters; calculating the best matching window parameters of the remaining SAR image pairs according to the established mapping function; and combining the average deformation rate fields of the line-of-sight of the ascending track and the descending track, and inversely calculating the surface three-dimensional flow velocity deformation field of the target area through a geometric vector decomposition model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power engineering geological disaster monitoring, and in particular to a three-dimensional flow velocity field inversion method and system based on coherence-assisted SAR adaptive offset tracking and a storage medium. BACKGROUND

[0002] In power engineering operation and maintenance, geological disasters such as landslides and glacier surges pose a serious threat to key facilities such as power transmission lines and substations. Power transmission lines often span hundreds of kilometers of complex terrain, and the geological conditions along the line are varied. Traditional manual inspection methods are difficult to achieve comprehensive coverage and timely warning. In particular, in earthquake-prone areas, goaf areas and landslide-prone areas, surface deformation may cause serious consequences such as tower tilting and foundation displacement, directly affecting the safe and stable operation of the power grid.

[0003] Synthetic aperture radar technology, with its large-scale and high-precision monitoring capabilities, provides a new solution for power facility deformation monitoring. Compared with traditional measurement methods, synthetic aperture radar interferometry (InSAR) technology can achieve:

[0004] (1) Regular scanning of wide-area power facilities, with a monitoring range of tens of thousands of square kilometers;

[0005] (2) Penetrating through cloudy and rainy weather to ensure the continuity of monitoring data;

[0006] (3) Identifying millimeter to centimeter level surface deformation to meet the safety warning needs of power facilities.

[0007] Among them, the technical path for measuring surface deformation using SAR data mainly includes two categories: phase-based interferometry and intensity-based offset tracking. Although InSAR technology has excellent precision, its application is strictly limited by "interference coherence". In actual applications, due to the diversity of surface cover types and the complexity of surface features in the distribution area of power facilities, especially in areas with dense vegetation, traditional InSAR technology is prone to lose coherence. When InSAR fails due to loss of coherence, pixel offset tracking (POT) becomes a mature and necessary complementary technology. Offset tracking technology does not rely on phase information, but directly uses the amplitude of SAR images. The main advantage of offset tracking technology is that it is not sensitive to phase loss of coherence, and it can measure large gradient deformation of meters or even tens of meters.

[0008] In the offset tracking algorithm, the matching window size is the core parameter to determine the precision and reliability of the displacement field inversion, and its setting faces the inherent trade-off between signal-to-noise ratio and local deformation fidelity. On the one hand, a larger matching window contains rich surface texture features, which can effectively suppress speckle noise by increasing the statistical sample size, thereby significantly improving the signal-to-noise ratio (SNR) of the cross-correlation peak and the matching robustness; but on the other hand, an excessively large window will introduce a serious spatial averaging effect, resulting in the masking of local non-uniform deformation information. Conversely, a smaller window, while reducing the smoothing error, is extremely susceptible to noise interference, leading to matching failure. Therefore, the standard approach of using a single fixed size window throughout the entire study area is essentially a compromise. Although this strategy performs well in scenarios where the surface features and deformation patterns are uniform, its limitations are particularly pronounced in a strongly heterogeneous environment: spatially, for long-distance power transmission corridors that cross regional terrain, the fixed window has difficulty simultaneously accommodating the feature differences between hilly and flat areas; temporally, in the face of seasonal changes in vegetation growth, a window based on static spatial features cannot adapt to the dynamic evolution of signal quality. Therefore, the traditional fixed window method has been unable to meet the fine adaptation needs of power facility monitoring in complex environments, and the development of an adaptive window strategy is imperative.

[0009] Although the industry has fully recognized the necessity of the adaptive window strategy in offset tracking, previous research has mainly focused on solving spatial heterogeneity problems (such as landslide boundary preservation), and still has limitations in dealing with the complexity of the time dimension. On the other hand, existing methods mostly rely on indirect features such as image texture and estimated gradients, or are based on empirical rules, which not only have high computational costs, but also lack a rigorous physical basis. Essentially, these indicators lack a direct and quantitative connection to the core physical process of the time evolution of SAR signal degradation - namely, the time evolution of scattering properties. Therefore, in view of the significant seasonal changes and multi-temporal observation needs in power facility monitoring, there is an urgent need to establish a physically mechanism-driven and computationally efficient adaptive window method. This method should be able to start from first principles, and quantitatively guide the selection of window parameters through the time decay law of signal quality, thereby overcoming the monitoring challenges in complex time-varying environments. SUMMARY

[0010] The technical problem to be solved by the present application is to provide a three-dimensional flow velocity field inversion method, system and storage medium based on coherence-assisted SAR adaptive offset tracking, to improve the decorrelation problem caused by time heterogeneity in synthetic aperture radar interferometry (InSAR) technology.

[0011] The technical scheme adopted by the present application to solve the technical problem is: a three-dimensional flow velocity field inversion method based on coherence-assisted SAR adaptive offset tracking, comprising the following steps:

[0012] S1, collect and pre-process ascending track SAR data and digital elevation model data covering a target area to establish a time-space baseline network between SAR images;

[0013] S2, based on the time-space baseline network, perform differential interference processing on each pair of SAR images to obtain their geocoded coherence data;

[0014] S3, select a pair of SAR images from the time-space baseline network as a reference unit, and calculate the offset result by pixel offset tracking technology for the reference unit, and determine the best matching window parameters thereof;

[0015] S4, according to the obtained geocoded coherence data of the reference unit and the determined best matching window parameters of the reference unit, establish a mapping function between the coherence and the best matching window parameters;

[0016] S5, according to the established mapping function, substitute the geocoded coherence data of the remaining SAR image pairs in the time-space baseline network to calculate the best matching window parameters of the remaining SAR image pairs, thereby completing the determination of the best matching window parameters of all SAR image pairs in the time-space baseline network;

[0017] S6, using the determined best matching window parameters of all SAR image pairs, perform pixel offset tracking processing on all SAR image pairs in the time-space baseline network to obtain the line-of-sight displacement field of each SAR image pair;

[0018] S7, based on the small baseline set technology, perform time series solution on the obtained line-of-sight displacement field of each SAR image pair to obtain the line-of-sight average deformation rate field of ascending and descending tracks;

[0019] S8, combine the line-of-sight average deformation rate fields of the ascending and descending tracks, and inverse the surface three-dimensional flow deformation field of the target area by a geometric vector decomposition model.

[0020] Further, the pre-processing includes:

[0021] Based on the projection equation, using satellite orbit parameters, the digital elevation model data is cropped, resampled and coordinate system transformed to make its spatial range and reference system consistent with the SAR data;

[0022] Select one SAR image as a super master image, and precisely register the remaining multi-scene SAR images to the radar coordinate system of the super master image;

[0023] Based on the preset time baseline threshold and spatial vertical baseline threshold, all registered SAR images are combined and screened two by two to construct the time-space baseline network.

[0024] Further, the S2 step comprises:

[0025] S21, for each pair of SAR images in the space-time baseline network, determining a master image and a slave image in time sequence, multi-viewing, registering, and interfering the two images to generate a differential interferogram, and removing terrain phase and flat phase by using the digital elevation model data and orbit parameters;

[0026] S22, filtering the generated differential interferogram to obtain a filtered differential interferogram, and calculating the coherence coefficient of each pixel to obtain coherence data;

[0027] S23, based on the satellite orbit parameters and the digital elevation model data, converting the obtained filtered differential interferogram and the obtained coherence data from the radar coordinate system to the target geodetic coordinate system, completing geocoding, and obtaining geocoded coherence data.

[0028] Further, the determination of the optimal matching window parameter by the pixel offset tracking technology comprises:

[0029] On the image corresponding to the reference unit, a series of matching windows of different sizes are used to perform pixel offset tracking calculation respectively;

[0030] Comparing the displacement field result quality under each size window;

[0031] Selecting the window size corresponding to the best quality evaluation result as the optimal matching window parameter.

[0032] Further, the mapping function is:

[0033] ;

[0034] In the formula, : optimal matching window size; : coherence.

[0035] Further, the optimal matching window parameter of the SAR image pair to be processed in the S5 step is calculated by using the following formula:

[0036] ;

[0037] In the formula, : optimal offset tracking matching window of the reference unit; : coherence of the reference unit; : offset tracking optimal matching window of the SAR image pair to be processed; : coherence of the SAR image pair to be processed.

[0038] Further, the S7 step comprises:

[0039] S71, constructing an observation equation group H·V=δ with each SAR image pair line-of-sight displacement as an observation value and each adjacent time interval average deformation rate as an unknown quantity, wherein δ is an observation vector composed of all SAR image pair displacement values, V is a to-be-solved average deformation rate sequence vector, and H is a design matrix;

[0040] S72, solving the observation equation group by using a singular value decomposition method to obtain the ascending track and descending track line-of-sight average deformation rate fields respectively.

[0041] Further, the geometric vector decomposition model is a constraint method based on a parallel flow hypothesis.

[0042] The three-dimensional flow velocity field inversion system based on the coherence-assisted SAR adaptive offset tracking, which is used to execute the three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking described above, comprises:

[0043] A data preprocessing and network construction module configured to execute the S1 step;

[0044] A coherence calculation module configured to execute the S2 step;

[0045] An adaptive window decision module configured to execute the S3, S4 and S5 steps;

[0046] A displacement field extraction and timing solution module configured to execute the S6 and S7 steps; and

[0047] A three-dimensional inversion module configured to execute the S8 step.

[0048] A computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking described above.

[0049] The present application has the beneficial effects that: the present application realizes adaptive and accurate matching of processing parameters to the spatial and temporal changes of surface scattering characteristics by establishing a physical model between coherence and offset tracking optimal window, to a certain extent, overcomes the limitations of the fixed window method that "precision" and "robustness" are difficult to balance in complex scenes, and can simultaneously improve the spatial detail resolution capability of deformation monitoring results and the reliability in low coherence area; through "physical model driven" instead of "manual experience trial and error", the automation and standardization of the processing flow are realized, the surface deformation evaluation precision and accuracy are effectively improved, and the calculation efficiency is greatly improved; the method effectively integrates the technical advantages of InSAR and POT, expands the spatio-temporal range of effective monitoring, is especially suitable for major engineering safety operation such as power facility corridor which crosses multiple landforms and needs to cope with seasonal changes and long-term stable monitoring, and provides a better technical solution for geological disaster identification and infrastructure life cycle safety management. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is a flowchart of the present application. DETAILED DESCRIPTION

[0051] The present application will be further described below in combination with the drawings and examples.

[0052] As Figure 1 shown, the three-dimensional flow velocity field inversion method based on coherence assisted SAR adaptive offset tracking of the present application comprises the following steps:

[0053] S1, collect ascending and descending track synthetic aperture radar (SAR) data and digital elevation model data covering the target area, and pre-process them to establish a time-space baseline network between SAR images;

[0054] S2, based on the time-space baseline network, difference interference processing is performed on each pair of SAR image pairs to obtain their geocoded coherence data;

[0055] S3, a pair of SAR image pairs is selected from the time-space baseline network as a reference unit, and the optimal matching window parameters of the reference unit are determined by pixel offset tracking technology;

[0056] S4, according to the obtained geocoded coherence data of the reference unit and the determined optimal matching window parameters of the reference unit, a mapping function between coherence and optimal matching window parameters is established;

[0057] S5, according to the mapping function established, the coherence data of the geographical coding of the remaining SAR image pairs in the space-time baseline network is substituted, and the optimal matching window parameters of the remaining SAR image pairs are calculated, so that the determination of the optimal matching window parameters of all SAR image pairs in the space-time baseline network is completed.

[0058] S6, using the determined optimal matching window parameters of all SAR image pairs, pixel offset tracking processing is performed on all SAR image pairs in the space-time baseline network, and the line-of-sight displacement field of each SAR image pair is obtained.

[0059] S7, based on the small baseline set technology, the line-of-sight displacement field of each SAR image pair obtained is time series solved, and the line-of-sight average deformation rate field of the ascending track and the descending track is obtained.

[0060] S8, the line-of-sight average deformation rate fields of the ascending track and the descending track are combined, and the ground surface three-dimensional flow deformation field of the target area is inversed through a geometric vector decomposition model.

[0061] The present application realizes the adaptive and accurate matching of the processing parameters to the space-time changes of the ground surface scattering characteristics by establishing a physical model between the coherence and the optimal window of offset tracking, to a certain extent, overcomes the core contradiction that the fixed window method is difficult to balance the "precision" and "robustness" in complex scenes, and can simultaneously improve the spatial detail resolution capability of the deformation monitoring result and the reliability in the low coherence area; through "physical model driven" instead of "manual experience trial and error", the automation and standardization of the processing flow are realized, the ground surface deformation evaluation precision and accuracy are effectively improved, and the calculation efficiency is greatly improved; the method effectively integrates the technical advantages of InSAR and POT, expands the space-time range of effective monitoring, and is especially suitable for major engineering safety operation such as power facility corridor which crosses multiple landforms and needs to cope with seasonal changes and long-term stable monitoring, and provides a better technical solution for geological disaster identification and infrastructure life cycle safety management.

[0062] In the embodiment of the present application, the preprocessing comprises:

[0063] Based on the projection equation, the digital elevation model data is cropped, resampled and coordinate system transformed using satellite orbit parameters, so as to ensure that the spatial range and reference system are consistent with the SAR data;

[0064] One SAR image is selected as a super master image, and the remaining multi-scene SAR images are precisely registered to the radar coordinate system of the super master image;

[0065] Based on the preset time baseline threshold and spatial vertical baseline threshold, all registered SAR images are combined and screened two by two, and the space-time baseline network is constructed.

[0066] The above pre-processing mode ensures that the comparison of each pixel in the subsequent steps (such as differential interference and offset tracking) is performed at the same ground position in a strict sense, which is the most fundamental prerequisite for obtaining high-precision deformation information, and avoids false signals or precision loss caused by data misplacement.

[0067] In the embodiment of the application, the multi-scene SAR image registration to the radar coordinate system of the super master image comprises:

[0068] Super master image selection: selecting an image with good imaging quality (meeting the preset requirements) and a central time-space baseline from all SAR images as a super master image;

[0069] Registration processing: registering each of the remaining SAR images (slave images) to the super master image, specifically including: initial geometric correction based on satellite orbit parameters; on this basis, a large number of homonymic feature points are automatically extracted on the two images and their offsets are calculated to construct a high-precision registration model; finally, the slave images are resampled according to the model to generate a new image that is completely aligned with the super master image in the pixel grid.

[0070] In the embodiment of the application, the S2 step comprises:

[0071] S21, for each pair of SAR images in the time-space baseline network, determining the master image and the slave image in time sequence, performing multi-view, registration and interference on the two images to generate a differential interference graph, and removing terrain phase and flat phase using the digital elevation model data and the orbit parameters; in the embodiment of the application, the multi-view ratio is set to 10:2 to reduce the influence of speckle noise;

[0072] S22, filtering the generated differential interference graph (such as Goldstein filtering) to suppress phase noise and optimize interference fringes, obtaining a filtered differential interference graph; and calculating the coherence coefficient of each pixel to obtain coherence data;

[0073] S23, based on the satellite orbit parameters and the digital elevation model data, converting the filtered differential interference graph obtained in S23 and the coherence data obtained in S22 from the radar coordinate system to the geographic coordinate system to complete geocoding and obtain geocoded coherence data.

[0074] In the embodiment of the application, a pair of SAR images is randomly selected from the time-space baseline network as a reference unit.

[0075] In some embodiments, a pair of images with the shortest (or below a threshold value) spatial baseline and time baseline is selected as the reference unit.

[0076] In this invention, the specific process of establishing the mapping function between coherence and the optimal matching window parameters based on the calculated offset and the determined optimal matching window parameters of the reference unit is as follows:

[0077] S31. Define the total variance of the offset estimation as... ;

[0078]

[0079] in, It is a decoherence / noise error; It is deformation gradient error;

[0080] S32. Based on interferometry theory, Modeling as coherence gamma Inversely proportional to the window area Inversely proportional functions:

[0081] ;

[0082] in, A It is a proportionality constant that includes factors such as sensor noise level and signal brightness;

[0083] S33. Establish deformation gradient error The model; when a deformation gradient exists within the matching window (i.e., the displacement is not constant but varies spatially), the "average displacement" that the offset tracking algorithm attempts to find is itself an approximation; the larger the window, the more drastic the displacement changes it contains, and the greater the error introduced by this approximation; assuming a linear deformation gradient exists in one direction. k (Unit: pixel displacement / pixel distance); in a width of W Within the window, the maximum displacement difference between the two ends is k · W The variance of the error between the estimated average displacement and the actual displacement field can be reasonably assumed to be proportional to the square of the maximum displacement difference.

[0084] ;

[0085] in, B It is a proportionality constant.

[0086] S34. Establish a unified model for total variance.

[0087] Adding the two error sources together gives the total variance with respect to the window size. W coherence gamma and deformation gradient k Functions:

[0088] ;

[0089] wherein, A is a constant;

[0090] Therefore, increasing W will decrease the first term (noise error) but will increase the second term (gradient error); decreasing W will decrease the second term (gradient error) but will increase the first term (noise error).

[0091] S35, solving the optimal window size

[0092] 1) take the derivative of W :

[0093] ;

[0094] 2) set the derivative to zero:

[0095] ;

[0096] 3) solve for W :

[0097] ;

[0098] ;

[0099] ;

[0100] For simplicity, the constant term A , B and the local surface property k can be combined into a new constant because it is relatively fixed for a particular deformation region;

[0101] Thus we get:

[0102] .

[0103] In the present application: for any remaining SAR image pair to be processed in the space-time baseline network, the average coherence is calculated according to the corresponding coherence coefficient map, and the formula for adaptively determining the optimal matching window parameter thereof is combined with the mapping function constructed as follows:

[0104] ;

[0105] In the formula, : the best offset tracking matching window of the reference unit; : the coherence of the reference unit; : The offset tracking optimal matching window of the SAR image pair to be processed; : The coherence of the SAR image pair to be processed.

[0106] The specific process of S7 is as follows:

[0107] S71, construct an observation equation group H·V=δ, in which the line-of-sight displacement of each SAR image pair is taken as an observation value, and the average deformation rate in each adjacent time interval is taken as an unknown quantity, wherein δ is an observation vector composed of the displacement values of all SAR image pairs, V is a sequence vector of the average deformation rate to be solved, and H is a design matrix: assuming that there are N time point SAR images, the solution target is the average deformation rate vector V1, V2,..., V N-1 of N-1 time periods. For M SAR image pairs, the observation displacement vector δ and the deformation rate vector V satisfy a linear relationship H·V=δ. Each row of the design matrix H corresponds to an image pair, and the element value is determined by the start and end time of the image pair.

[0108] S72, solve the observation equation group by using a singular value decomposition method, and obtain the line-of-sight average deformation rate fields of the ascending track and the descending track respectively: Because the space-time baseline network is usually connected by multiple discontinuous time periods (subsets), the matrix H is rank-deficient, and the conventional least square cannot be solved. Singular value decomposition (SVD) is adopted: H=U∑V T ; U: an M×M orthogonal matrix, whose column vectors span the observation space (that is, the space where δ is located). Each column of U is called a left singular vector. ∑: an M×(N-1) matrix, and all elements except the elements on the main diagonal line (called singular values) are 0. V: an (N-1)×(N-1) orthogonal matrix, whose column vectors span the parameter space (that is, the space where V is located). Each column of V is called a right singular vector. By discarding the singular values close to zero, a stable least norm solution v=V∑ + UTδ is obtained by solving the generalized inverse. v: the average deformation rate vector obtained by solving; ∑ + : the pseudo-inverse (generalized inverse) matrix of ∑. The above solving is independently performed for each pixel, and the time sequence line-of-sight deformation rate fields of the ascending track and the descending track are obtained.

[0109] The inversion can be realized by a geometric vector decomposition model, including but not limited to: a joint solution based on multi-source data fusion, a constraint solution based on a specific physical model (such as parallel flow), or a filtering estimation algorithm based on space-time prior, etc. The embodiment preferably adopts a constraint method based on the parallel flow hypothesis.

[0110] In the embodiment of the application, the specific process of the geometric vector decomposition model adopting the constraint method based on the parallel flow hypothesis to invert the target area ground three-dimensional flow velocity deformation field is as follows:

[0111] S81, Constructing observation equations: Obtain the InSAR line-of-sight average deformation rate field of ascending and descending satellites; for each pixel in the target area, use its ascending and descending line-of-sight deformation rate observations, combined with satellite geometric parameters, to establish a linear equation between its three-dimensional deformation rate (east, north and vertical) and two line-of-sight observations:

[0112] ;

[0113] Wherein:

[0114] , are the ascending and descending line-of-sight average deformation rate observations corresponding to the same ground point;

[0115] is the three-dimensional deformation rate vector of the ground point, wherein , , represent the east, north and vertical deformation rate components, respectively;

[0116] and are the direction cosine components of the unit vectors of the ascending and descending satellite line-of-sight directions in the local northeast celestial coordinate system, which are calculated from satellite orbit geometry (incidence angle, azimuth angle);

[0117] , represent the noise errors contained in the ascending and descending observations, respectively;

[0118] S82, Introducing parallel flow hypothesis constraint: According to the topographic features of the target area or known physical model, it is assumed that the ground deformation mainly shows quasi-two-dimensional flow along a certain dominant horizontal direction, that is, there is a fixed proportional relationship between the north and east velocity components: , is the flow constraint coefficient, , and θ is the angle (azimuth angle) between the preset dominant flow direction at this pixel and the due east direction; Substitute this geometric constraint relationship into the equation set in step S81 to reduce the unknowns from three velocity components of each pixel to two;

[0119] S83, Constructing regional overall inversion matrix model: Collect the constraint equation sets of all pixels in the target area into a unified matrix equation, expressed in linear form: d=Gm+e, wherein d is the observation vector composed of all ascending and descending observations, m is the parameter vector composed of the velocity component parameters to be solved of all pixels, G is the coefficient matrix (design matrix) composed of satellite geometry and flow constraints, and e is the observation noise vector;

[0120] S84, truncated singular value decomposition (TSVD) solution: singular value decomposition is performed on the coefficient matrix G; by setting a truncation threshold, singular values and their corresponding singular vectors with too small values are discarded to overcome the matrix ill-conditioning problem caused by poor observation geometry or insufficient constraints; using the truncated singular value decomposition result, a stable parameter vector estimate m is calculated.

[0121] S85, three-dimensional flow velocity field reconstruction and output: according to the parameter vector m obtained by solving step S84, the eastward and vertical deformation rates of each pixel are recovered; combined with the parallel flow constraint relationship introduced in step S82, the northward deformation rate is calculated; finally, the eastward, northward and vertical three-component surface three-dimensional flow velocity deformation field covering the entire target area is synthesized and output.

Claims

1. A three-dimensional flow velocity field inversion method based on coherence-assisted SAR adaptive offset tracking, characterized in that, The method comprises the following steps: S1, collecting and preprocessing ascending / descending track SAR (Synthetic Aperture Radar) data and digital elevation model data covering a target area to establish a time-space baseline network among SAR images; S2, based on the time-space baseline network, performing differential interference processing on each pair of SAR images to obtain geocoded coherence data; S3, selecting a pair of SAR images from the time-space baseline network as a reference unit, and calculating a displacement result by using a pixel displacement tracking technology and determining a best matching window parameter of the reference unit; S4, establishing a mapping function between the coherence and the best matching window parameter according to the geocoded coherence data of the reference unit and the best matching window parameter of the reference unit; ； In the formulae, : best matching window size; : coherence; S5, according to the established mapping function, substituting geocoded coherence data of the remaining SAR image pairs in the time-space baseline network to calculate the best matching window parameter of each SAR image pair, thereby determining the best matching window parameter of all SAR image pairs in the time-space baseline network; S6, using the determined best matching window parameter of all SAR image pairs, performing pixel displacement tracking processing on all SAR image pairs in the time-space baseline network to obtain a line-of-sight displacement field of each SAR image pair; ； In the formula, : Best offset tracking matching window of reference cell; : Coherence of reference cell; : Offset tracking best matching window of SAR image pair to be processed; : Coherence of SAR image pair to be processed; S7, based on a small baseline set technology, performing time series calculation on the obtained line-of-sight displacement field of each SAR image pair to obtain line-of-sight average deformation rate fields of ascending and descending tracks; S8, combining the line-of-sight average deformation rate fields of the ascending and descending tracks, and inversely calculating a three-dimensional flow deformation field of the target area by using a geometric vector decomposition model. The preprocessing comprises:

2. The three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking according to claim 1, wherein, based on a projection equation, using satellite orbit parameters to crop, resample and convert the coordinate system of the digital elevation model data, so that the spatial range and reference system of the digital elevation model data are consistent with the SAR data; selecting one SAR image as a super master image and precisely registering the remaining multi-scene SAR images to the radar coordinate system of the super master image; based on a preset time baseline threshold and a spatial vertical baseline threshold, combining and screening all registered SAR images in pairs to construct the time-space baseline network. The S2 step comprises:

3. The three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking according to claim 1, wherein, S21, for each pair of SAR image pairs in the time-space baseline network, determining a master image and a slave image in time sequence, performing multi-view, registration and interference on the two images to generate a differential interference graph, and removing terrain phase and flat phase by using the digital elevation model data and orbit parameters; S22, filtering the generated differential interference graph to obtain a filtered differential interference graph, and calculating a coherence coefficient of each pixel to obtain coherence data; S23, based on satellite orbit parameters and digital elevation model data, converting the obtained filtered differential interference graph and the obtained coherence data from a radar coordinate system to a target geodetic coordinate system to complete geocoding and obtain geocoded coherence data. ​ 4. The three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking according to claim 1, wherein, The best matching window parameter is determined by the pixel offset tracking technology, and the best matching window parameter includes: A series of matching windows with different sizes are used to perform pixel offset tracking calculation on the image corresponding to the reference unit; The displacement field result quality under each size window is compared; The window size corresponding to the best quality evaluation result is selected as the best matching window parameter.

5. The three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking according to claim 1, wherein, The S7 step includes: S71, constructing an observation equation group H·V=δ, in which V is an unknown quantity of the average deformation rate of each adjacent time interval, and δ is an observation vector composed of all SAR image pair displacement values; S72, solving the observation equation group by using a singular value decomposition method to obtain the average deformation rate field of the ascending track and the descending track.

6. The three-dimensional flow velocity field inversion method based on the coherence-assisted SAR adaptive offset tracking according to claim 1, wherein, The geometric vector decomposition model is a constraint method based on the parallel flow hypothesis.

7. A system for three-dimensional flow velocity field inversion based on coherence-aided SAR adaptive offset tracking, the system being configured to perform the method of three-dimensional flow velocity field inversion based on coherence-aided SAR adaptive offset tracking according to any one of claims 1 to 6, characterized in that It includes: a data preprocessing and network construction module configured to perform the S1 step; a coherence calculation module configured to perform the S2 step; an adaptive window decision module configured to perform the S3, S4 and S5 steps; a displacement field extraction and time series solution module configured to perform the S6 and S7 steps; and a three-dimensional inversion module configured to perform the S8 step.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the three-dimensional flow velocity field inversion method based on the coherence aided SAR adaptive offset tracking according to any one of claims 1 to 6.

Images