A landslide deformation time sequence thermal map generation system

Abstract

This invention relates to the field of remote sensing image processing and temporal InSAR deformation image generation technology, and discloses a landslide deformation temporal thermal map generation system, including a processor and a memory; it performs temporal InSAR processing on SAR single-look complex image sequences covering target landslides, establishes a unified spatiotemporal raster and a fixed eight-neighbor edge set, generates unified reference deformation values, single-phase coherence coefficients, effective observation masks and displacement domain inversion residuals, constructs a time baseline calculation window and common effective observation relationships based on the original effective observations, generates boundary barrier strength values ​​and boundary constraint deformation propagation weights for each edge and each target temporal phase, generates spatial weights based on the boundary constraint deformation propagation weights and pixel center distances, and performs spatiotemporal image reconstruction to obtain a complete temporal deformation raster.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing and time-series InSAR deformation image generation technology, specifically relating to a landslide deformation time-series thermal map generation system. Background Technology

[0002] Temporal InSAR processing typically takes multiple SAR single-view complex images, precise orbit data, and DEMs as input. After registration, interferometry, filtering, unwrapping, atmospheric correction, and reference zone correction, the output is a line-of-sight deformation raster arranged according to the acquisition time. This type of raster is commonly used in landslide deformation monitoring to express the deformation amplitude of different locations on the slope over time, and can be further used to generate heat maps, dynamic layers, or raster products with geographic coordinates. Existing public document CN109884635B has described an InSAR deformation monitoring data processing method, which obtains deformation monitoring results through SAR image processing, point target selection, grid organization, and network solution, demonstrating that InSAR deformation data processing has an engineering workflow suitable for large-scale remote sensing data.

[0003] In the actual processing of landslide areas, SAR side-looking imaging, slope occlusion, coherence variations, vegetation cover changes, unwrapping failures, and atmospheric residuals can cause inconsistent effective pixel distributions across different time phases. Engineering processing typically requires filling in missing pixels and then mapping the filled deformation values ​​to color frames. If deformation values ​​are only passed based on pixel distance, ordinary adjacency relationships, or frame-by-frame smoothing results, deformation differences on both sides of the missing area may be smoothed out, and multiple adjacent deformation hotspots may be merged in the image. If color levels are determined separately for each time phase, image colors are redistributed as the acquisition time changes, and the same color may correspond to different deformation ranges in different frames, affecting cross-time phase readings.

[0004] Furthermore, conventional heatmap outputs primarily display deformation values ​​in color, which is insufficient in expressing the differences in the sources of complete pixels and original valid observation pixels, and also lacks auxiliary layers related to pixel boundary barriers. For landslide deformation time-series products, practical applications require the simultaneous retention of a unified geographic reference, acquisition time index, fixed color scale, deformation units, original observation support information, and boundary barrier line layers to facilitate subsequent frame-sequence viewing, overlay checks, and result comparisons in remote sensing image processing systems or GIS environments. Summary of the Invention

[0005] The present invention aims to provide a landslide deformation time-series thermogram generation system, which solves the technical problems mentioned in the background art.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution.

[0007] A landslide deformation time-series thermal map generation system includes a processor and a memory. The memory stores a computer program, and the processor reads and executes the computer program to perform time-series InSAR processing on a SAR single-look complex image sequence covering a target landslide. When the processor reads and executes the computer program, it performs the following: Establish a unified spatiotemporal grid and a fixed set of eight neighboring edges; Generate a unified reference deformation value, single-phase coherence coefficient, effective observation mask, and displacement domain inversion residual; Based on the original effective observations, a time baseline calculation window is constructed for each target time phase, and a common effective observation relationship is constructed for each edge in the fixed eight-neighbor edge set and each target time phase; For each edge and each target temporal phase in the fixed eight-neighbor edge set, the boundary barrier strength value and boundary constraint deformation propagation weight are generated based on the changes in geomorphic normal, incremental deformation difference, common effective observation ratio, coherence change, and slope direction consistency of adjacent pixels. Spatial weights are generated based on the boundary constraint deformation propagation weights and the pixel center distance, and spatiotemporal image reconstruction is performed using the spatial weights to obtain a complete temporal deformation raster. Deformation color frames are generated using a unified color scale across the entire time series, and observation support transparency frames are generated based on the path with the minimum propagation cost. For each acquisition phase, the boundary barrier intensity value is rasterized into a boundary barrier line layer. The corresponding deformation color frame, observation support transparency frame, and boundary barrier line layer are combined to form a layer composite result. All layer composite results are encoded in ascending order of acquisition time to obtain the landslide deformation time-series thermogram.

[0008] Compared with the prior art, the present invention has the following substantial features and significant progress: This invention uses a unified spatiotemporal grid and a fixed eight-neighbor edge set to process SAR deformation results from different time phases within the same pixel grid. It constrains the propagation of deformation values ​​across steep slopes, valleys, lateral shear zones, and secondary slider boundaries by using boundary barrier strength values ​​and boundary constraint deformation propagation weights. It reduces adjacent hotspot merging and hotspot position shifts caused by missing pixel completion by jointly solving for the complete temporal deformation grid using observation weights, spatial weights, and temporal weights. It generates deformation color frames using a unified color scale across the entire time series, ensuring that deformation images at each acquisition time are read using the same fixed color scale. It distinguishes the support sources of original valid observation pixels and completed pixels using observation support transparency frames. Finally, it outputs a landslide deformation temporal heat map with a unified geographic reference, fixed color scale, deformation unit, and acquisition time index by combining the boundary barrier layer, deformation color frame, and observation support transparency frame generated for each acquisition time phase. Attached Figure Description

[0009] Figure 1 This is a processing architecture diagram of a landslide deformation time-series thermal map generation system according to the present invention. Detailed Implementation

[0010] The present invention will now be described in conjunction with the accompanying drawings. Each module executes according to the data dependency order specified in this embodiment; the same technical features, parameters, and technical terms in this embodiment refer to the same object.

[0011] Example 1: The following combination Figure 1 This paper describes an implementation of a landslide deformation time-series thermal map generation system. This implementation is executed by an electronic device 10. The electronic device 10 includes a processor 100 and a memory 110. The memory 110 stores a computer program, a SAR single-look complex image sequence 120, precise orbit data 130, a DEM 140, intermediate data, and a landslide deformation time-series thermal map 310. The processor 100 reads and executes the computer program, sequentially calling a unified spatiotemporal raster establishment module 200, a fixed eight-neighbor edge set establishment module 210, a unified reference deformation data generation module 220, a time baseline calculation window generation module 230, a boundary barrier strength value generation module 240, a boundary constraint deformation propagation weight generation module 250, a spatiotemporal image reconstruction module 260, a deformation color frame generation module 270, an observation support transparency frame generation module 280, a boundary barrier line layer generation module 290, and a layer compositing and map encoding module 300. Figure 1 Solid lines in the diagram represent data input and output, while dashed lines represent module calls or parameter passing.

[0012] SAR single-view complex image sequence 120 covers the target landslide and the area outside the target landslide, and has the same orbital direction, the same imaging mode, and the same polarization. When the input data provides a peripheral deformation reference area, SAR single-view complex image sequence 120 and DEM 140 also cover the peripheral deformation reference area. Precise orbit data 130 is used for orbit correction. DEM 140 is used to provide elevation data of the target landslide and the area outside the target landslide. The peripheral deformation reference area is the area outside the target landslide used to form a unified reference. The landslide deformation time-series heat map 310 includes a unified geographic reference, fixed color scale, deformation unit, a one-to-one correspondence between frame number and acquisition time, and expresses the spatial distribution of deformation, the transparency of observation support, and the boundary barrier line layer using the composite results of layers from each acquisition phase.

[0013] Each image in the SAR single-view complex image sequence 120 is read according to the image record table. The image record table includes at least the image number, acquisition time, orbit number, imaging mode, polarization mode, number of rows, number of columns, complex pixel matrix, pixel spacing, proximity time, azimuth time, orbital state vector, and geocoding parameters. Precise orbit data 130 is associated with the image record table according to the acquisition time index, and DEM 140 is read according to the target projection coordinate system determined by the unified spatiotemporal raster establishment module 200. After reading the data, the unified spatiotemporal raster establishment module 200 first checks whether the image number, acquisition time, number of rows, number of columns, and complex pixel matrix exist in the image record table, and then checks whether the same orbital direction, the same imaging mode, and the same polarization mode are consistent. Images that pass the check are then entered into the main image selection and resampling process.

[0014] For ease of description, assume that the SAR single-look complex image sequence 120 includes N acquisition phases. The phase index is represented by t. And the data collection time meets the requirements. In a unified spatiotemporal raster, cell indices are represented by p and q; the planar coordinates of cell p are denoted as q. DEM elevation is denoted as The fixed set of eight neighboring edges is denoted as One of the edges is denoted as (p, q); the unified reference deformation value is denoted as... The single-phase coherence coefficient is denoted as The effective observation mask is denoted as The displacement domain inversion residual is denoted as . This indicates that pixel p is the original effective observed pixel at acquisition time t. This indicates that pixel p was not an original valid observation pixel at the time of acquisition (phase t).

[0015] This embodiment , , , , and All numbers are positive. , Pick ; Pick ; Pick With the square of the effective sample median of the displacement domain inversion residual The larger of the multiples; Pick The square of the effective sample median of the difference between adjacent rates The larger of the multiples. Propagation participation threshold. Take 0.05.

[0016] In this implementation, MAD represents the median absolute deviation of a sample from its median. For a sample set S, the calculation method is as follows: first, calculate the median of the sample set; then, calculate the absolute value of the difference between each sample and the median; finally, take the median of all absolute values. If the sample set is empty, the parameters of this function are treated as if there are no valid samples. If the sample set S contains only one sample, MAD is zero, and the calculation remains executable due to the lower limit parameter in the corresponding formula. `clip(x,a,b)` restricts the input x to the closed interval [a,b], calculated as follows: if x is less than a, take a; if x is greater than b, take b; the original value is maintained within the interval.

[0017] Example 2: The unified spatiotemporal grid establishment module 200 acquires SAR single-look complex image sequence 120, precise orbit data 130, and DEM 140. The module reads the acquisition time, orbit number, imaging mode, polarization, row and column number, complex image data, and metadata for each image in the SAR single-look complex image sequence 120, and checks whether each image covers the target landslide and the area outside the target landslide. When the input data provides a peripheral deformation reference area, it also checks whether each image covers the aforementioned peripheral deformation reference area. If any image lacks acquisition time or spatial coverage information, the unified spatiotemporal grid establishment module 200 outputs incomplete data information and stops processing dependent on that image.

[0018] The unified spatiotemporal raster establishment module 200 selects the main image based on the median time of all acquisition times of the SAR single-look complex image sequence 120. The unified spatiotemporal raster establishment module 200 first calculates the median time of all acquisition times, and selects the image with the smallest difference from this median time as the candidate main image. If multiple candidate main images exist, the vertical baselines of the candidate main image and other images at the target landslide center location are obtained from the precise orbit data 130, forming the sum of the absolute values ​​of the vertical baselines of the candidate main images. The ranking value of the candidate main image r is the sum of the absolute values ​​of the vertical baselines of the candidate main image r and all other images. This represents the vertical baseline between the candidate master image r and image s. The unified spatiotemporal raster establishment module 200 determines the master images in ascending order of sorting values; if the sorting values ​​are the same, the candidate master image ranked first in ascending order of acquisition time is selected. After selecting the master image, the unified spatiotemporal raster establishment module 200 uses the master image as the registration reference and performs orbit correction based on the precise orbit data 130.

[0019] The unified spatiotemporal raster creation module 200 uses the projection coordinate system, origin, pixel center arrangement, pixel size, number of rows, and number of columns of the main image as the basis for the unified spatiotemporal raster. Let pixel p in the unified spatiotemporal raster be defined by planar coordinates... and DEM elevation The determined three-dimensional coordinates are For the acquisition phase t, the orbital position is obtained by interpolation based on the orbital state vector in the precise orbital data 130. and orbital velocity The zero Doppler condition is: the dot product of the three-dimensional coordinates of pixel p, the vector of its orbital position at the corresponding azimuth time, and the orbital velocity vector at that time is zero. This condition is used to solve for the azimuth time of pixel p at time phase t. Radar slant range The Euclidean distance between the three-dimensional coordinates of pixel p and its corresponding orbital position at the corresponding azimuth time is given. The unified spatiotemporal grid establishment module 200 uniquely determines the range and azimuth positions of pixel p in each SAR single-look complex image based on the azimuth time, radar slant range, proximity time, azimuth time, pixel spacing, and geocoding parameters in the image record table. If no solution satisfying the zero Doppler condition exists or the position is outside the image range, the corresponding position is recorded as a value without a source. The complex pixel matrix of the SAR single-look complex image sequence 120 uses bilinear interpolation to process the real and imaginary parts separately, and DEM140 is resampled to the corresponding pixel center using bilinear interpolation. If a pixel in the resampled DEM140 has no source value, the pixel is written into a fixed exclusion mask. The unified spatiotemporal grid stores the projection coordinate system identifier, origin coordinates, number of rows, number of columns, pixel size, invalid value code, pixel center coordinate index, and acquisition time index.

[0020] The unified spatiotemporal raster establishment module 200 writes water bodies, permanent no-echo zones, DEM missing regions, and data coverage areas into a fixed exclusion mask. The fixed exclusion mask is formed by the union of water bodies, permanent no-echo zones, DEM missing regions, and data coverage areas. Water bodies are obtained by resampling external water body rasters into the unified spatiotemporal raster; when there are no external water body rasters, the unified spatiotemporal raster establishment module 200 calculates the full-time amplitude median of each pixel with at least one complex pixel source value. Let the number of full-time amplitude medians be M. The max(1, 0.05M)th value of the full-time amplitude medians arranged in ascending order is used as a preset amplitude threshold. Eight-neighbor connected regions with a full-time amplitude median lower than the preset amplitude threshold and a number of pixels not less than four are directly written into the fixed exclusion mask as water bodies. Permanent no-echo zones are formed by pixels with no complex pixel source values ​​or whose amplitudes are all zero in all acquisition phases. Missing DEM regions are formed by pixels with no source value from the resampled DEM140. Regions outside data coverage are obtained by inverting the intersection of the effective coverage areas of each resampled image. The fixed exclusion mask remains consistent across all acquisition phases and is not modified using subsequently generated coherence coefficients or deformation values.

[0021] The fixed eight-neighbor edge set establishment module 210 reads the unified spatiotemporal raster and the fixed exclusion mask, establishes a fixed eight-neighbor edge set outside the fixed exclusion mask according to the cell center adjacency relationship, and records the two endpoints, cell center distance, and edge direction unit vector of each edge. The fixed eight-neighbor edge set establishment module 210 traverses the cells in the unified spatiotemporal raster not covered by the fixed exclusion mask in row-major order. For the current cell, only the four types of adjacent cells—right, bottom, lower right, and lower left—are checked; if the adjacent cell is also not covered by the fixed exclusion mask, an undirected edge is written. The edge number is incremented according to the writing order, the first endpoint of the edge is the current cell, the second endpoint is the checked adjacent cell, and the edge direction unit vector points from the first endpoint to the second endpoint. This rule avoids duplicate writing of the same adjacency relationship and ensures that the edge direction unit vector, cell center distance, and edge number are unique.

[0022] Pixel center distance Let be the planar Euclidean distance between the centers of pixel p and pixel q. The edge direction unit vector. Let be the unit direction vector from pixel p to pixel q, obtained by dividing the difference in coordinates between the two pixels by the distance between their centers. By fixing any edge in the set of eight neighboring edges, connecting two distinct pixel centers, therefore... The fixed eight-neighbor edge set remains unchanged in the generation of boundary barrier strength values, the generation of boundary constraint deformation propagation weights, spatiotemporal image reconstruction, and the generation of boundary barrier line layers.

[0023] Example 3: The unified reference deformation data generation module 220 constructs a connected small baseline interferometry network based on the SAR single-look complex image sequence 120. The preset time baseline threshold is three times the median of all adjacent acquisition time intervals; the number of all image pairs is P, and the preset orbital baseline threshold is the max(1, 0.75P)th value of the absolute values ​​of the vertical baselines of all image pairs arranged in ascending order. For any two acquisition time phases i and j, if the acquisition time interval does not exceed the preset time baseline threshold and the absolute value of the orbital baseline does not exceed the preset orbital baseline threshold, they are written into the candidate interferometric edge set. The candidate interferometric edges are sorted in ascending order by acquisition time interval, ascending order by orbital baseline absolute value, and ascending order by the indices of the two acquisition time phases; the candidate interferometric edges are checked sequentially, and when the two acquisition time phases of a candidate interferometric edge are not yet located in the same connected component, the candidate interferometric edge is written into the connected small baseline interferometry network until all acquisition time phases are located in the same connected component; subsequently, the remaining candidate interferometric edges are written into the connected small baseline interferometry network in their original order. If the candidate interference edge set cannot make all acquisition phases lie in the same connected component, the unified reference deformation data generation module 220 outputs information about insufficient connected small baseline interferometric network and stops subsequent processing.

[0024] The unified reference deformation data generation module 220 performs differential interferometry, interference edge coherence coefficient calculation, phase filtering, phase unwrapping, and atmospheric correction on the connected small baseline interferometric network. For the interference edge (i,j) in the connected small baseline interferometric network, the interference phase is formed from the complex image data after registration of two scenes, and based on the precise orbit data 130, DEM 140, and radar slant range... , The sum of the flat terrain phase and the DEM topographic phase is calculated, and the sum of the flat terrain phase and the DEM topographic phase is subtracted from the interferometric phase to obtain the differential interferometric phase. Let the complex pixel value of pixel u after registration at acquisition phase t be... The set of pixels within the fixed window of five rows and five columns of the current pixel p that are not covered by the fixed exclusion mask and whose complex pixel values ​​are valid in both scenes is: The interference edge coherence coefficient at pixel p for interference edge (i,j) is denoted as... . This means limiting the input to a certain value by adding or subtracting an integer multiple of 2π. Original interference phase The argument of pixel p after registration at time phases j and i, multiplied by the conjugate of its complex pixel values. The sum of the flat terrain phase and the DEM topographic phase. The differential interferometric phase is calculated from the difference in slant range between the two time phases and the radar wavelength, and is 4π / λ multiplied by the difference in slant range between the two time phases. The result is the original interferometric phase after subtracting the sum of the flat terrain phase and the DEM topographic phase, wrapped around the principal value interval of (-π,π].

[0025] If there are fewer than nine valid samples, the denominator of the interference edge coherence coefficient formula is zero, or the phase filtering weighted sum is zero, the corresponding result will be recorded as invalid. (Filtered winding phase) The argument is obtained by weighting the complex differential phases within a fixed window using the coherence coefficients of the interference edges as weights.

[0026] Phase unwrapping is performed separately for each interference edge. Pixels with an interference edge coherence coefficient not less than 0.3 and not covered by a fixed exclusion mask are divided into connected regions according to a fixed set of eight neighboring edges. Within each connected region, the pixel with the smallest pixel number is selected as the starting point, and the unwrapped phase position of the starting point is its filtered wrapped phase. The edges within the connected region are sorted in descending order of the minimum coherence coefficient between the two endpoints and in ascending order of edge number. This process is repeated to traverse all edges, selecting edges that can connect unwrapped pixels to unwrapped pixels each time, and adding an integer multiple of 2π to the unwrapped pixels to minimize the absolute value of the phase difference between them and unwrapped pixels. The process stops when no unwrapped pixels are added in a complete traversal. Pixels that cannot be connected to the starting point remain invalid. Atmospheric low-frequency phase components. The atmospheric low-frequency phase component is obtained by weighting the unwrapped phase with the interference edge coherence coefficient within a fixed window; the set of pixels within the fixed window that are not covered by the fixed exclusion mask, have valid unwrapped phases, and have valid interference edge coherence coefficients is denoted as... When the weighted sum of the coherence coefficients of the interference edges is zero, the corresponding atmospherically corrected unwrapped phase is recorded as invalid. All interference edges are processed using the same rules. Atmospherically corrected unwrapped phase The result is the original unwrapped phase minus the atmospheric low-frequency phase component.

[0027] The unified reference deformation data generation module 220 first uses the atmospherically corrected displacement domain observations of each interferometric edge to... Under the given conditions, a connected small baseline interferometric network weighted least squares inversion is performed on each pixel to obtain preliminary temporal InSAR results. If the input data provides a peripheral deformation reference area, the unified reference deformation data generation module 220 uses the peripheral deformation reference area; if the input data does not provide a peripheral deformation reference area, the unified reference deformation data generation module 220 determines the peripheral deformation reference area from the candidate connected regions outside the target landslide. Candidate pixels satisfy the following conditions: not covered by a fixed exclusion mask, located outside the target landslide, and having valid preliminary temporal InSAR results at least in two acquisition phases. Subsequently, candidate connected regions are extracted according to the eight-neighbor connectivity relationship, and candidate connected regions are assigned numbers according to the region extraction order.

[0028] For each candidate connected region, the unified reference deformation data generation module 220 calculates the median absolute value of the deformation rate and the deformation time dispersion of the effective pixels within the candidate connected region. Let the set of pixels in the candidate connected region g be... The median absolute value of the deformation rate of the candidate connected region g is The deformation time discreteness is The number of effective pixels is The effective pixel count threshold is Effective number of pixels This refers to the number of pixels within the region that have valid preliminary time-series InSAR results at least at two acquisition time phases. Valid pixel count threshold. Take the larger value between 9 and the median of the effective number of all candidate connected regions. The median of the absolute value of the deformation rate of candidate connected region g. , which is the median of the absolute values ​​of the relative deformation rates of all valid adjacent time intervals within the region. The deformation time dispersion of a candidate connected region g is the median of the absolute values ​​of the differences between the median deformation rate of each time-phase region and the median deformation rate of that time-phase region. Candidate connected region ranking value. It is the ratio of the median absolute value of the deformation rate in this region to the median deformation rate of all candidate regions, plus the ratio of the deformation time dispersion of this region to the median deformation time dispersion of all candidate regions.

[0029] Unified reference deformation data generation module 220 meets the requirements And there exists at least one candidate connected region that is a valid adjacent acquisition, according to The candidate connected regions are selected as the peripheral deformation reference areas in ascending order, descending order of effective pixel count, and ascending order of candidate connected region number. If no candidate connected regions meet the conditions, the unified reference deformation data generation module 220 outputs information indicating insufficient peripheral deformation reference areas and stops subsequent processing.

[0030] Let the carrier wavelength of the SAR single-look complex image sequence 120 be λ, and the unwrapped phase of the interference edge (i,j) after atmospheric correction at pixel p be... Displacement domain observations The line-of-sight displacement difference obtained after atmospheric correction and unwrapping phase transformation is calculated as -λ / (4π) multiplied by the atmospheric correction and unwrapping phase. The unified reference deformation data generation module 220 uses the square of the interferometric edge coherence coefficient as a weighting factor to perform a connected small baseline interferometric network weighted least squares inversion on the displacement domain observations of each pixel, and sets... Preliminary time-series InSAR results were obtained, and the objective function is as follows: Let the outer deformation reference region be For each acquisition phase t, the mid-range deformation reference in the reference region This represents the median of the effective preliminary temporal InSAR deformation values ​​within the peripheral deformation reference area. (Uniform reference deformation value) The initial temporal InSAR deformation value is calculated by subtracting the median deformation reference value of the corresponding time phase from the initial value. The displacement domain observations are synchronously corrected by subtracting the difference in median deformation reference values ​​between the two ends of the interference edge from the original displacement domain observation value. Pixels that cannot be located in the same unwrapped connected region as the peripheral deformation reference area are marked as invalid.

[0031] Single-phase coherence coefficient This is the median of the coherence coefficients of all valid interference edges connecting this time phase. The unified reference deformation data generation module 220 records pixels with overlapping, shadows, areas covered by fixed exclusion masks, invalid unwrapped areas, or no valid inversion values ​​as missing, without replacing missing pixels with zero deformation. For pixel p and acquisition time phase t, the unified reference deformation data generation module 220 outputs the unified reference deformation value. Continuous numerical form of single-phase coherence coefficient Effective observation mask and displacement domain inversion residuals When pixel p is not fixedly excluded from masking coverage, unwrapping is effective, the unified reference deformation value is effective, and the single-phase coherence coefficient is effective during acquisition phase t, ;otherwise, Displacement domain inversion residuals This is the median of the absolute differences between all observed and fitted values ​​of the interferometric edges connected at this time phase. If pixel p has no available interferometric edge residuals at acquisition time phase t, then... It is marked as invalid, and the corresponding observation weight is reset to zero.

[0032] Example 4: The temporal baseline calculation window generation module 230 constructs a temporal baseline calculation window and a common effective observation relationship based on the original effective observations marked by the effective observation mask. The common effective observation relationship is established only based on the effective observation mask before spatiotemporal image reconstruction and remains unchanged after spatiotemporal image reconstruction.

[0033] The half-width h of the time baseline calculation window is the median of the absolute time baselines of all interference edges. When the connected small baseline interferometric network is empty, the time baseline calculation window generation module 230 outputs information indicating insufficient connected small baseline interferometric network and stops processing. Let the i-th interference edge be connected to the acquisition time phase... and acquisition time phase For the incremental endpoint index k, The adjacent time phase corresponding to the incremental endpoint index k is the acquisition time phase k. 1 and the acquisition time phase k, the relative increment time of adjacent times. This is the midpoint between two adjacent acquisition times. For the target time phase t, the time baseline calculation window for target time phase t is... , is the set of incremental endpoint indices where the difference between the incremental time and the target time phase acquisition time does not exceed half a width h. If Empty, time baseline calculation window generation module 230 is included. The minimum incremental endpoint index is selected; if multiple minimum values ​​exist, the one with the smallest incremental endpoint index is included. The time baseline calculation window generation module 230 does not copy the acquisition phase, does not generate the time interpolation phase, and does not use the complete time-series deformed raster to modify the common effective observation relationship.

[0034] For edges in a fixed eight-neighbor set Effective observation relationship with target time phase t Within the time baseline calculation window, the pixels at both ends of edge (p,q) are the incremental endpoint index set of the original valid observations in adjacent time phases. (Common valid observation ratio) This is the ratio of the number of elements in a common valid observation relationship to the total number of incremental endpoints within the time baseline calculation window. Since each time baseline calculation window contains at least one incremental endpoint index, the denominator of the common valid observation ratio is greater than zero. Empty, edge It remains in the fixed eight-neighbor set, but the incremental deformation difference of the edge at the target time phase t is set to zero.

[0035] The boundary barrier strength value generation module 240 reads DEM140, fixed eight-neighbor edge set, deformation value, single-phase coherence coefficient, effective observation mask, time baseline calculation window and common effective observation relationship, and generates geomorphic normal change, incremental deformation difference, common effective observation ratio, coherence change and slope direction consistency in sequence.

[0036] The boundary barrier strength value generation module 240 calculates the DEM elevation center difference according to the row and column directions of the unified spatiotemporal raster and the actual spacing of the pixel center coordinates. Let the left, right, top, and bottom adjacent pixels of pixel p be... , , and Plane coordinates The first and second coordinate components are denoted as follows: and Lateral elevation gradient The elevation difference between the left and right adjacent pixels is divided by the difference in their lateral coordinates; the vertical elevation gradient is... The elevation difference between adjacent pixels is divided by the longitudinal coordinate difference; boundary pixels are calculated using one-sided difference. If pixel p lacks a left adjacent pixel in the lateral direction, the right adjacent pixel is used for calculation. If the right neighbor cell is missing, the left neighbor cell is used for calculation. When a vertical column lacks an upper or lower adjacent cell, the same calculation rule applies. If the required adjacent pixels are located within a fixed exclusion mask, are missing from DEM140, or have a coordinate difference of zero, then the slope unit normal vector and slope unit vector of that pixel are marked as invalid. For edges with invalid slope unit normal vectors at any endpoint, the change in terrain normal is set to zero, and this edge is not included in the calculation of the median absolute deviation of the effective slope unit normal angle samples. Slope unit normal vector It is obtained by normalizing a three-dimensional vector composed of the negative horizontal gradient, the negative vertical gradient, and the value 1. When When, the gradient unit vector is set to zero; when At that time, the unit vector of slope descent It is obtained by normalizing a two-dimensional vector composed of negative horizontal gradient and negative vertical gradient; when the gradient is zero, the zero vector is taken.

[0037] The boundary barrier strength value generation module 240 generates the terrain normal variation from the slope unit normal angle between adjacent pixels. For the edge unit normal angle of slope The angle between the unit normal vectors of the slope surface of two pixels is calculated using the inverse cosine of the vector dot product, with the dot product pre-truncated to the interval [-1, 1]. The change in the terrain normal is calculated using nonlinear normalization, as shown in the following formula: in, Indicates the change in the normal direction of landforms. This indicates the lower limit of the value. If the number of effective slope unit normal angle samples within the fixed eight-neighbor edge set is zero, the boundary barrier strength value generation module 240 sets the geomorphic normal variation of all edges to zero. In the calculation of geomorphic normal variation, Make the input of the inverse cosine function not less than 1 and not greater than 1.

[0038] For the incremental endpoint index k in the common effective observation relationship, the boundary barrier strength value generation module 240 first calculates the incremental deformation rate based on the relative deformation value of the same pixel in adjacent times and the acquisition time, and then generates the basic value of the incremental deformation difference from the difference of the incremental deformation rates of the two endpoints of edge (p,q): single-time relative incremental deformation rate. Dividing the deformation difference between adjacent time phases by the time interval yields the deformation rate per unit time. Single-time relative incremental deformation difference. This represents the absolute difference in the relative incremental rates between adjacent pixels at the same time. Median of incremental deformation difference. The median of incremental deformation differences within the common effective observation set is used. The incremental deformation difference is calculated using nonlinear normalization, as shown in the following formula: in, Indicates the incremental deformation difference. Let the set of edges representing the non-empty set of common effective observation relationships at time phase t of the target. Indicates the lower limit of the value. If Empty and Set all to zero. If empty, the incremental deformation difference of all edges under the target time phase t is set to zero.

[0039] The boundary barrier strength value generation module 240 generates the coherence variation from the median of the time window coherence coefficients. For pixel p, the median of the time window coherence coefficients... This is the median of the effective single-phase coherence coefficients for that pixel within the time window. For the edge... Coherence change This represents the absolute difference between the median coherence coefficients of adjacent pixels within a time window. If an endpoint has no valid single-phase coherence coefficient within the time baseline calculation window, the boundary barrier strength generation module 240 uses the median of valid single-phase coherence coefficients for that endpoint across all acquisition phases; if there are still no valid single-phase coherence coefficients, the median coherence coefficient for that endpoint within the time window is set to zero. Since the single-phase coherence coefficient lies within a closed interval of zero to one, the coherence variation also lies within a closed interval of zero to one.

[0040] The boundary barrier strength value generation module 240 normalizes the sum of the slope unit vectors of the two endpoints of an edge in the fixed eight-neighbor edge set to obtain the slope direction, and generates a slope direction consistency value from the angle between the edge direction unit vector and the slope direction, as shown in the following formula: in, This represents the slope direction consistency value. If the slope unit vector at either endpoint is invalid, or the sum of the slope unit vectors at both endpoints is zero, the boundary barrier strength value generation module 240 will set the slope direction consistency value of the corresponding side to half.

[0041] Weighted incremental deformation difference The effective observation ratio is multiplied by the incremental deformation difference. The boundary barrier strength value generation module 240 uses the weighted incremental deformation difference and the geomorphic normal change to generate the boundary barrier strength value; when both the geomorphic normal change and the weighted incremental deformation difference are greater than zero, the coherence change is used to adjust the boundary barrier strength value. The final boundary barrier strength value formula is as follows: in, This represents the boundary barrier strength value. Since the landform normal variation, the weighted incremental deformation difference, and the coherence variation are all not less than zero, the boundary barrier strength value is not less than zero. If... or When the product term is zero, the coherence change does not amplify the boundary barrier strength value independently.

[0042] Example 5: Boundary constraint deformation propagation weight The boundary constraint deformation propagation weight is calculated by multiplying the slope direction consistency value by the negative power of the natural constant, with a value range of (0,1]. A larger boundary constraint strength value results in a smaller weight, while a larger slope direction consistency value results in a larger weight. Since the slope direction consistency value is greater than zero and not greater than one, and the boundary constraint strength value is not less than zero, the boundary constraint deformation propagation weight is greater than zero and not greater than one. The boundary constraint deformation propagation weight generation module 250 saves the boundary constraint deformation propagation weights according to the edge number and the acquisition time phase.

[0043] Boundary constraint deformation propagation weight generation module 250 sets propagation participation threshold .like Then the edge During data collection, phase t does not participate in spatial weight formation and minimum propagation cost path search, and the corresponding spatial weight is reset to zero; if Then the edge During collection, phase t participates in propagation.

[0044] The spatiotemporal image reconstruction module 260 generates pre-normalized observation weights based on the single-phase coherence coefficient and the displacement domain inversion residual of the corresponding pixel in the corresponding acquisition phase; it generates pre-normalized spatial weights based on the boundary constraint deformation propagation weights and pixel center distances; and it generates pre-normalized temporal weights based on the positions of the original valid observation pixels in three consecutive acquisition phases of the SAR single-look complex image sequence 120. (Single-phase coherence coefficient) The weights are limited to an interval not less than zero and not greater than one; if the single-phase coherence coefficient or displacement domain inversion residual is invalid, the corresponding observation weights before normalization are reset to zero. Observation weights before normalization The sum of the squared single-phase coherence coefficient and the squared displacement-domain inversion residual (MAD) plus the lower limit of the numerical value is calculated by dividing by the square of the single-phase coherence coefficient and the square of the full-time displacement-domain inversion residual. Invalid pre-normalization observation weights are reset to zero. Pre-normalization observation weights are only applied to... and Effective cell calculation; pre-normalized observation weights at other locations are reset to zero. Pre-normalized spatial weights. When the boundary constraint deformation propagation weight is greater than or equal to the threshold, it is the boundary constraint deformation propagation weight divided by the square of (pixel distance plus the lower limit of the value); when the boundary constraint deformation propagation weight is lower than the threshold, the spatial weight before normalization is reset to zero. Adjacent rate difference This represents the difference in deformation rate of a pixel across three consecutive time phases. Time weights before normalization. This is 1 divided by (the square of the MAD (Matrices of Adjacent Rate Differences) across the entire region plus the lower limit of the numerical value). Wherein, This represents the set of pixels for which all three consecutive acquisition phases are original, valid observation pixels with valid corresponding deformation values. If... If empty, the time weight is reset to zero before normalization.

[0045] Before assembling the objective function, the spatiotemporal image reconstruction module 260 normalizes the observation weights, spatial weights, and temporal weights before non-zero normalization, respectively. Let the median of the observation weights before non-zero normalization be... The median of the spatial weights before non-zero normalization is The median time weights before non-zero normalization are The normalization method for the three categories of weights is as follows: each non-zero weight is divided by the median of its own non-zero values ​​to achieve a balance among the three categories of weights. After normalization, , and The normalized observation weights, normalized spatial weights, and normalized temporal weights are respectively written into the objective function. If a certain class of non-zero pre-normalized weight samples is empty, the constraint term corresponding to that class of weights is not written into the objective function. The normalized weights are used in the subsequent objective function, observation weight matrix, spatial weight matrix, and temporal weight matrix.

[0046] The spatiotemporal image reconstruction module 260 uses the unified reference deformation value of the original effective observation pixels marked by the effective observation mask as the data constraint, the difference in deformation values ​​between the two endpoints of the edge with a spatial weight greater than zero within the same acquisition phase as the spatial constraint, and the difference in adjacent rates on the time-weighted non-equal interval time axis as the time constraint to construct the objective function, which is as follows: The spatiotemporal image reconstruction module 260 assembles the objective function into a sparse linear system, in matrix form as follows: The spatiotemporal image reconstruction module 260 first establishes unknown quantity numbers for all pixels outside the fixed mask and all acquisition time phase combinations. The numbering function is denoted as... Normalized observation weight matrix For a diagonal matrix, the first line on the main diagonal... The elements are The matrix formed by normalized spatial weights. Only for The edge-by-edge assembly is performed, and Laplace terms are written at the corresponding positions of π(p,t) and π(q,t). Adjacent rate difference operator Only for Three coefficients are written in three consecutive acquisition phases. The deformation value matrix D is... Expand by numbering the unknowns.

[0047] The spatiotemporal image reconstruction module 260 establishes a spatiotemporal graph. The nodes of the spatiotemporal graph are fixed (p, t) excluding the mask; when… When, a spatial connection is established between nodes (p,t) and (q,t); when At that time, a temporal connection is established between three consecutive acquisition time phase nodes of the same pixel involved in the adjacent rate difference. Connectivity component labeling is performed on the spatiotemporal map, and only spatiotemporal connected components containing at least two original valid observations from different acquisition time phases are retained. Each retained spatiotemporal connected component is divided into node sets according to the connection relationship corresponding to the non-zero normalized spatial weight and the non-zero normalized temporal weight; each node set is entered into the solution if it contains at least one observation weight greater than zero, otherwise the corresponding unobserved position remains invalid.

[0048] The spatiotemporal image reconstruction module 260 uses sparse Cholesky decomposition to solve the linear system; when the number of unknowns exceeds one million or the sparse Cholesky decomposition fails, the conjugate gradient method is used. The conjugate gradient method uses the zero vector as the initial value, and the stopping condition is that the change in the residual norm between two consecutive iterations is less than 10%. If the iteration count reaches 1000, positions where the solution fails remain invalid, and mask-covered positions are excluded from the solution.

[0049] Example 6: The deformation color frame generation module 270 extracts the complete temporal deformation raster values ​​corresponding to the positions marked as original effective observation pixels by the effective observation mask from each acquisition phase, and calculates the median of the global coherence coefficient for all the original effective observation pixels. If no original effective observation pixels exist in any acquisition phase, the deformation color frame generation module 270 does not calculate the color level endpoints and sets all deformation color frames to transparent. If original effective observation pixels exist, the median of the global coherence coefficient is calculated. This is the median of the single-phase coherence coefficients for all original valid observation pixels.

[0050] The deformation color frame generation module 270 determines the color level endpoints based on the complete temporal deformation raster values ​​corresponding to the original effective observed pixels with a single-phase coherence coefficient greater than or equal to the median of the global coherence coefficient. The candidate color level set consists of the deformation values ​​of the original effective observed pixels with a single-phase coherence coefficient not lower than the median of the global coherence coefficient; negative endpoints... The smaller of the minimum negative deformation and 0 in the candidate set, and the positive endpoint. This is the larger of the maximum value of positive deformation and 0 in the candidate set. If and The deformation color frame generation module 270 maps the effective pixels in the complete temporal deformation raster to zero-value colors, while keeping the excluded mask coverage positions, invalid positions in the complete temporal deformation raster, and unsolved positions transparent. If or The deformable color frame generation module 270 performs truncation and divergence color band mapping. The negative endpoint color vector is denoted as... The zero-value color vector is denoted as The color vector at the positive endpoint is denoted as And it remains unchanged throughout all acquisition phases, fixed at [value]. .

[0051] Cut-off deformation value To limit the deformation value to Within the range, any excess portion is truncated to endpoint values. The deformation color frame generation module 270 uses a three-point diverging color band. The truncated deformation values... ,like Negative color mapping: Negative deformation values ​​are linearly interpolated from blue (0,0,255) to white (255,255,255), with zero being white; if Forward color mapping: positive deformation values ​​are linearly interpolated from white to red (255,0,0); if The color vector takes the zero value. If and Effective pixels are assigned zero-value colors; if only one endpoint is zero, linear interpolation is performed only on the other side. All acquisition phases use the same color level endpoints, zero-value colors, and divergence bands; color level endpoints, zero-value colors, or divergence bands are not updated frame-by-frame. The deformation color frame color vectors for excluded mask coverage locations, invalid locations in the complete temporal deformation raster, and unsolved locations are fixed as zero vectors, and the observation support transparency value is set to zero.

[0052] The observation support transparency frame generation module 280 generates observation support transparency frames based on the minimum propagation cost path. For the acquisition phase t in which the observation support transparency frame is being generated, if there are no original valid observation pixels with observation weights greater than zero in that acquisition phase, the observation support transparency frame generation module 280 sets the observation support transparency frame for that acquisition phase to be fully transparent. If there are original valid observation pixels with observation weights greater than zero, the maximum observation weight is set to [value missing]. This represents the maximum value of the normalized observation weights for all original valid observation pixels in the current time phase. For the original valid observation pixels, the original observation base transparency... The normalized observation weights are divided by the maximum observation weight, truncated to the [0,1] interval. This satisfies the following conditions for the current acquisition phase. and Effective pixel completion, observation support transparency frame generation module 280 satisfies the condition in the fixed eight-neighbor edge set The edges are considered walkable edges, and the cost of edge propagation is... The weight is the negative natural logarithm of the deformation propagation weights for boundary constraints; smaller weights result in higher costs. Since traversable edges satisfy... The path cost is not less than zero. The observation support transparency frame generation module 280 performs a multi-source Dijkstra search for each acquisition phase; all original valid observation pixels with observation weights greater than zero in that acquisition phase are used as the starting point set, and the starting point distance is set to zero. The minimum propagation cost path to any of the original valid observation pixels is searched for among all completed pixels. The boundary constraint deformation propagation weight used for the path cost is the same as the boundary constraint deformation propagation weight used for the final transparency product.

[0053] If the minimum propagation cost path for completing pixel p is The original effective observation pixels at the end of this path are Complete the pixel observation support transparency value The original observation base transparency of the endpoint of the minimum propagation cost path is multiplied by the product of the deformation propagation weights of all boundary constraints along the path, and truncated to the interval [0,1]. If multiple original effective observation pixels result in the same minimum propagation cost, the endpoint with the higher original observation base transparency is selected; if the original observation base transparency is still the same, the endpoint with the first pixel number in the sequence is selected. If the completed pixel cannot satisfy the minimum propagation cost... A passable edge is connected to any original valid observation cell with an observation weight greater than zero, and the observation support transparency value is set to zero. The observation support transparency frame does not reverse-modify the complete temporally deformed raster.

[0054] Example 7: The boundary barrier layer generation module 290 normalizes the boundary barrier strength values ​​of all edges in all acquisition phases and the fixed eight-neighbor edge set using the same global maximum value. If the fixed eight-neighbor edge set is empty, the boundary barrier layer generation module 290 sets the boundary barrier layer of all acquisition phases to full transparency, and the layer composite and map encoding module 300 uses only the deformation color frame and the observation support transparency frame to form the layer composite result. If the fixed eight-neighbor edge set is not empty, the boundary barrier strength value is the global maximum value. This represents the maximum boundary barrier strength value across all time phases and all edges. Boundary line transparency. The single-sided boundary barrier strength value is divided by the global maximum value and normalized to the [0,1] interval. If The boundary barrier line layer generation module 290 sets the boundary barrier line layer of all acquired phases to full transparency. If... The boundary barrier line layer generation module 290 rasterizes each acquisition phase according to the line connecting the pixel centers corresponding to the two endpoints of each edge in the fixed eight-neighbor edge set, with the line width fixed as one output pixel.

[0055] The boundary barrier line layer generation module 290 uses an integer grid connection method for rasterization. Each output cell is represented as a closed rectangle with its center as the center and its side lengths being the horizontal and vertical cell dimensions of a unified spatiotemporal raster, respectively. When the closed line segment between the centers of two cells on edge (p, q) intersects the closed rectangle, the output cell is determined to be covered by the edge. If multiple edges cover the same output cell, the cell-level boundary line transparency of the output cell is taken as the maximum value of the corresponding edge-level boundary line transparency, and the boundary barrier line layer color vector is taken as the preset boundary line color vector.

[0056] The pixel-level boundary line transparency of the boundary barrier layer at pixel p and time phase t is denoted as . When multiple edges are rasterized to the same cell, Take the edge corresponding to the rasterized cell The maximum value; cells not covered by edge rasterization satisfy Boundary barrier layer color vector This represents the color vector of the boundary barrier layer at time phase t for pixel p.

[0057] The layer compositing and map encoding module 300 performs compositing for each acquisition phase, using the corresponding deformation color frame as the bottom layer, the observation support transparency frame as the transparency layer, and the boundary barrier line layer as the top layer. Layer compositing: First, the deformation color and gray background (192, 192, 192) are mixed using the observation support transparency. Then, the black boundary line is superimposed using the pixel-level boundary line transparency to obtain the layer compositing pixel value, as shown in the following formula: Background color vector Fixed at (192, 192, 192), negative endpoint color vector Fixed at (0,0,255), zero-value color vector Fixed at (255, 255, 255), positive endpoint color vector Fixed at (255,0,0), boundary barrier layer color vector Fixed at (0,0,0), each color vector component takes integer values ​​from zero to 255. If pixel p is not covered by edge rasterization at acquisition time t, then If pixel p is covered by edge rasterization during acquisition, the boundary blocking line layer is applied according to the pixel-level boundary line transparency. Overlay on the composite result of the distorted color frame and the background color.

[0058] The layer compositing and map encoding module 300 uses the layer compositing results of each acquisition phase as frame data, and encodes all frame data in ascending order of acquisition time according to the SAR single-view complex image sequence 120, without generating time interpolation frames. The landslide deformation time-series heat map 310 stores a unified geographic reference, fixed color scale, deformation unit, frame number, and a one-to-one correspondence between the frame number and acquisition time. The geographic reference information 320 stores the projected coordinate system, geographic transformation, pixel size, origin coordinates, and number of rows and columns. The fixed color scale 330 stores negative endpoints, positive endpoints, zero-value color, and divergent color bands. The acquisition time index 340 stores the frame number, acquisition time, and source image identifier.

[0059] The landslide deformation time-series thermal map 310 is organized into four data categories: frame data, geographic reference information 320, fixed color scales 330, and acquisition time index 340. Frame data stores the composite pixel values ​​of the layers in the composite result corresponding to each acquisition time. It also saves the observation support transparency frames and boundary blocking line layers under the same row and column index. Georeferenced information 320 saves the projected coordinate system, geographic transformation, cell size, origin coordinates, and number of rows and columns. Fixed color scale 330 saves the negative endpoints, positive endpoints, zero-value color, and divergence color band. Acquisition time index 340 saves the frame number, acquisition time, and source image identifier. The frame numbers are arranged in ascending order of acquisition time and correspond to the image numbers in the SAR single-look complex image sequence 120.

[0060] In one implementation, the landslide deformation time-series heat map 310 is encoded as a NetCDF dataset. The NetCDF dataset sets up acquisition phase indices, row indices, column indices, and color channel indices; layer composite results are saved in the order of acquisition phase index, row index, column index, and color channel index; observation support transparency frames and boundary barrier line layers are saved in the order of acquisition phase index, row index, and column index respectively; acquisition times are saved in the order of acquisition phase index; projected coordinate system, geographic transformation, pixel size, origin coordinates, number of rows and columns, deformation unit, negative endpoints, positive endpoints, zero-value color, and divergent color band are saved as attributes of the corresponding data; invalid locations are saved using a unified invalid value encoding. This data organization ensures a one-to-one correspondence between frame numbers and acquisition times.

[0061] Example 8: Memory 110 stores a unified spatiotemporal raster, a fixed exclusion mask, a fixed eight-neighbor edge set, deformation values, single-phase coherence coefficients, effective observation masks, displacement domain inversion residuals, time baseline calculation windows, common effective observation relationships, boundary barrier strength values, boundary constraint deformation propagation weights, observation weights, spatial weights, temporal weights, a complete temporal deformation raster, deformation color frames, observation support transparency frames, boundary barrier line layers, and landslide deformation temporal heatmaps 310. The fixed eight-neighbor edge set is stored as an edge table, which includes the edge number, two endpoints, pixel center distance, and edge direction unit vector. Boundary barrier strength values ​​and boundary constraint deformation propagation weights are stored by edge number and acquisition phase.

[0062] When the SAR single-look complex image sequence 120 has fewer than two acquisition phases, the electronic device 10 outputs an insufficient acquisition phase information and stops processing. When the SAR single-look complex image sequence 120 includes two acquisition phases, the electronic device 10 does not generate time weights, does not write time constraints into the objective function, and checks whether each node set has a greater than zero observation weight according to the connection relationship corresponding to the non-zero spatial weights. When the coverage of DEM140 is inconsistent with that of the SAR single-look complex image sequence 120, the electronic device 10 writes the missing region of DEM140 into a fixed exclusion mask and does not establish a fixed eight-neighbor edge set across the missing region of DEM140.

[0063] When the common effective observation relationship is empty, the electronic device 10 keeps the common effective observation relationship empty and sets the incremental deformation difference of the corresponding edge in the corresponding acquisition phase to zero. When there is no effective single-phase coherence coefficient within the time baseline calculation window, the electronic device 10 uses the median of the effective single-phase coherence coefficients of the cell in all acquisition phases; if there is still no effective single-phase coherence coefficient, the median of the coherence coefficients of the corresponding time window is set to zero. When the boundary constraint deformation propagation weight is less than the propagation participation threshold, the spatiotemporal image reconstruction module 260 resets the corresponding spatial weight to zero, and the observation support transparency frame generation module 280 does not search for the minimum propagation cost path through this edge.

[0064] When a spatiotemporal connected component does not contain at least two original valid observations from different acquisition phases, or when a set of nodes divided according to the connection relationship corresponding to non-zero spatial weights and non-zero temporal weights does not contain observation weights greater than zero, the corresponding unobserved position remains invalid. When an acquisition phase has no original valid observation pixel observation weights greater than zero, the observation support transparency frame for that acquisition phase is set to full transparency. When the candidate value set of the unified color level across the entire time series is empty, all deformation color frames are set to transparency. When both negative and positive endpoints are zero, the deformation color frame generation module 270 maps the valid pixels in the complete temporal deformation raster to zero-value colors. When the fixed eight-neighbor edge set is empty or the global maximum value of all boundary barrier strength values ​​is zero, the boundary barrier line layer for all acquisition phases is set to full transparency.

[0065] In one specific embodiment, the electronic device 10 reads twelve acquisition phases of SAR single-look complex image sequence 120, precise orbit data 130, and DEM 140. Interferometric edge coherence coefficient calculation and phase filtering use a five-row, five-column fixed window, atmospheric correction uses a twenty-one-row, twenty-one-column fixed window, the effective threshold for the interferometric edge coherence coefficient is 0.3, and the propagation participation threshold... The stopping threshold for the conjugate gradient method is set to 0.05. The maximum number of iterations is set to 1000. The unified spatiotemporal raster establishment module 200 completes main image selection, orbit correction, and resampling. The fixed eight-neighbor edge set establishment module 210 establishes the fixed eight-neighbor edge set in row-priority order. The unified reference deformation data generation module 220 executes the generation rules determined in this embodiment for connected small baseline interferometry network, phase unwrapping, atmospheric correction, preliminary temporal InSAR results, peripheral deformation reference area, unified reference deformation value, single-phase coherence coefficient, effective observation mask, and displacement domain inversion residual. As a verification of the boundary constraint deformation propagation weight calculation, the first edge is... , , , and Take the second edge , , , and According to the calculation formula of this implementation method, the boundary constraint deformation propagation weight of the first edge is less than the propagation participation threshold, and the boundary constraint deformation propagation weight of the second edge is greater than the propagation participation threshold. The verification example formula is as follows: The remaining modules run sequentially according to the formulas and judgment conditions disclosed in this embodiment, and finally output a landslide deformation time series thermal map 310 containing frame data, geographic reference information 320, fixed color mark 330 and acquisition time index 340.

[0066] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A landslide deformation time-series thermal map generation system, comprising a processor and a memory, wherein the memory stores a computer program, and the processor reads and executes the computer program to perform time-series InSAR processing on a SAR single-look complex image sequence covering a target landslide, characterized in that, When the processor reads and executes the computer program, it performs the following: Establish a unified spatiotemporal grid and a fixed set of eight neighboring edges, and record the two endpoints of each edge, the distance between the cell centers, and the unit vector of the edge direction from the first endpoint to the second endpoint; Generate a unified reference deformation value, single-phase coherence coefficient, effective observation mask, and displacement domain inversion residual; Based on the original effective observations, a time baseline calculation window is constructed for each target time phase, and a common effective observation relationship is constructed for each edge in the fixed eight-neighbor edge set and each target time phase; The index of each acquisition phase after the first acquisition phase in the SAR single-look complex image sequence is used as the incremental endpoint index, and the acquisition phase corresponding to the incremental endpoint index is combined with the previous acquisition phase to form an adjacent time phase. For each edge and each target time phase in the fixed eight-neighbor edge set, the incremental endpoint index is written into the corresponding common effective observation relationship only when the two endpoints of the edge are both marked as original effective observation pixels by the effective observation mask in the previous and subsequent acquisition time phases of the adjacent time phases corresponding to the same incremental endpoint index. For each edge and each target temporal phase in the fixed eight-neighbor edge set, the boundary barrier strength value and boundary constraint deformation propagation weight are generated based on the changes in geomorphic normal, incremental deformation difference, common effective observation ratio, coherence change, and slope direction consistency of adjacent pixels. The generation of boundary barrier strength values ​​and boundary constraint deformation propagation weights includes: The slope unit normal vector and slope unit vector are calculated by performing central difference based on the actual spacing of the pixel center coordinates from the DEM elevation. The change in landform normal is generated by the slope unit normal angle between adjacent pixels; For each incremental endpoint index within the common effective observation relationship, the difference between the relative deformation values ​​at adjacent times is divided by the interval between the corresponding two acquisition times to obtain the incremental deformation rate of each of the two endpoints, and the incremental deformation difference is generated from the difference between the incremental deformation rates of the two endpoints. The proportion of common valid observations is generated by the number of incremental endpoint indices in the common valid observation relationship and the number of incremental endpoint indices in the time baseline calculation window. Calculate the median of the coherence coefficients of the two endpoints of an edge for each time window, and generate the coherence change from the absolute difference between the median of the coherence coefficients of the two time windows. The slope direction is obtained by normalizing the sum of the slope unit vectors of the two endpoints of an edge, and the slope direction consistency quantity is generated by the angle between the edge direction unit vector of the edge in the fixed eight-neighbor edge set and the slope direction. The incremental deformation difference is weighted using the common effective observation ratio, and the weighted incremental deformation difference is used together with the geomorphic normal change to generate the boundary barrier strength value. When both the change in the normal direction of the landform and the weighted incremental deformation difference are greater than zero, the boundary barrier strength value is adjusted using the coherence change. Based on the boundary barrier strength value and the slope direction consistency, a boundary constraint deformation propagation weight greater than zero and not greater than one is generated. When the boundary barrier strength value increases, the boundary constraint deformation propagation weight decreases, and when the slope direction consistency increases, the boundary constraint deformation propagation weight increases. Spatial weights are generated based on the boundary constraint deformation propagation weights and the pixel center distance, and spatiotemporal image reconstruction is performed using the spatial weights to obtain a complete temporal deformation raster. Deformation color frames are generated using a unified color scale across the entire time series, and observation support transparency frames are generated based on the path with the minimum propagation cost. For each acquisition phase, the boundary barrier intensity value is rasterized into a boundary barrier line layer. The corresponding deformation color frame, the observation support transparency frame, and the boundary barrier line layer are combined to form a layer composite result. All layer composite results are encoded in ascending order according to the acquisition time of the SAR single-view complex image sequence to obtain the landslide deformation time-series thermal map.

2. The landslide deformation time-series thermal map generation system according to claim 1, characterized in that, Establishing a unified spatiotemporal grid and a fixed set of eight neighboring edges includes: Acquire SAR single-look complex image sequences, precise orbit data, and DEM covering the target landslide and the area outside the target landslide, with the same orbital direction, the same imaging mode, and the same polarization. When the input data provides a peripheral deformation reference area, the SAR single-look complex image sequences and DEM also cover the peripheral deformation reference area. Calculate the median time of all acquisition times in the SAR single-view complex image sequence, and select the SAR single-view complex image with the smallest difference between the acquisition time and the median time as the main image; when there are multiple SAR single-view complex images with the same difference, select the main images in ascending order of the sum of the absolute values ​​of the vertical baseline and the acquisition time. Based on precise orbit data, orbit correction is performed on the SAR single-view complex image sequence. Using the main image as the registration reference, the orbit-corrected SAR single-view complex image sequence and DEM are resampled to a unified spatiotemporal grid with the same projection coordinate system, origin, pixel size, number of rows and columns, and one-to-one corresponding pixel center coordinates. The water body, permanent no-echo zone, DEM missing area and data coverage area are written into a fixed exclusion mask, and the fixed exclusion mask remains unchanged throughout all acquisition phases. Outside the fixed exclusion mask, a fixed eight-neighborhood edge set is established according to the adjacency relationship of the cell centers, and the two endpoints, the distance between the cell centers, and the edge direction unit vector from the first endpoint to the second endpoint of each edge are recorded.

3. The landslide temporal thermal pattern generation system of claim 2, wherein, The generation of unified reference deformation values, single-phase coherence coefficients, effective observation masks, and displacement domain inversion residuals includes: Based on the SAR single-look complex image sequence, a connected small baseline interferometric network covering all acquisition phases is constructed; Differential interferometry, interference edge coherence coefficient calculation, phase filtering, phase unwrapping, and atmospheric correction are performed on the connected small baseline interferometric network to obtain the displacement domain observations and interference edge coherence coefficients of each interference edge; Based on the displacement domain observations, the preliminary time-series InSAR results of each acquisition phase are inverted on the connected small baseline interferometric network. An outer deformation reference area is selected outside the target landslide, and all acquisition phases adopt the same outer deformation reference area and the same atmospheric correction rule. The median of the effective preliminary temporal InSAR deformation values ​​within the outer deformation reference area is used to perform reference correction on the preliminary temporal InSAR results of the corresponding acquisition phase to obtain a unified reference deformation value, and the unwrapped regions that cannot be connected to the outer deformation reference area are marked as invalid. Pixels that are overlaid, shadowed, covered by fixed exclusion mask, ineffective unwrapped area, or have no effective inversion value are recorded as missing, and the original effective observation pixels are marked by effective observation mask; The median of the coherence coefficients of the interference edges connected to the same acquisition time phase is taken as the single-phase coherence coefficient of that acquisition time phase, and the median of the absolute value of the residuals between the displacement domain observations of each interference edge and the fitted values ​​obtained by substituting back the uniform reference deformation values ​​is taken as the displacement domain inversion residual of that acquisition time phase.

4. The landslide temporal thermal pattern generation system of claim 3, wherein, The time baseline calculation window and common effective observation relationships constructed based on the original valid observations include: The median of the absolute time baselines of all interference edges in the connected small baseline interferometric network is used as the half-width of the time baseline calculation window; The index of each acquisition phase after the first acquisition phase in the SAR single-look complex image sequence is used as the incremental endpoint index. The acquisition phase corresponding to the incremental endpoint index is combined with the previous acquisition phase to form an adjacent time phase, and the midpoint of the two acquisition times of the adjacent time phase is used as the incremental time. For each target time phase, the incremental endpoint index whose distance from the incremental time to the target time phase acquisition time does not exceed half the width of the time baseline calculation window is included in the time baseline calculation window. For each edge and each target time phase in the fixed eight-neighbor edge set, the incremental endpoint index is written into the corresponding common effective observation relationship only when the two endpoints of the edge are both marked as original effective observation pixels by the effective observation mask in the previous and subsequent acquisition time phases of the adjacent time phases corresponding to the same incremental endpoint index. The common effective observation relationship is established solely based on the original effective observations marked by the effective observation mask and remains unchanged after spatiotemporal image reconstruction.

5. The landslide temporal thermal pattern generation system of claim 4, wherein, Generating spatial weights based on the boundary constraint deformation propagation weights and pixel center distances, and performing spatiotemporal image reconstruction includes: The observation weights are generated based on the single-phase coherence coefficient and the displacement domain inversion residual of the corresponding pixel in the corresponding acquisition phase; the observation weights of the original valid observation pixels with invalid displacement domain inversion residuals are reset to zero. When the boundary constraint deformation propagation weight is not less than the propagation participation threshold, a spatial weight is generated based on the boundary constraint deformation propagation weight and the pixel center distance; when the boundary constraint deformation propagation weight is less than the propagation participation threshold, the corresponding spatial weight is reset to zero. For the third acquisition phase and each subsequent acquisition phase, a time weight is generated based on the adjacent rate difference of the positions marked as the original effective observation pixels by the effective observation mask in the three consecutive acquisition phases. The data constraint is the uniform reference deformation value of the original effective observation pixels marked by the effective observation mask, the spatial constraint is the difference in deformation values ​​of the two endpoints of the edge with a spatial weight greater than zero within the same acquisition phase, and the objective function is constructed using the difference in adjacent rates on the time-weighted non-equal interval time axis. A spatiotemporal graph is constructed with fixed pixels outside the mask and acquisition phase combinations as nodes. Spatial connections are established between two nodes with spatial weights greater than zero, and temporal connections are established between three consecutive acquisition phase nodes of the same pixel involved in the adjacent rate difference with temporal weights greater than zero. The complete temporal deformable raster is solved only for the original valid observations containing at least two different acquisition phases, and each set of nodes interconnected by non-zero spatial weights or non-zero temporal weights contains at least one spatiotemporally connected component with an observation weight greater than zero. For spatiotemporal connected components that do not meet the solution conditions, fixed exclusion mask coverage locations, and locations where the solution fails, unobserved locations remain invalid.

6. The landslide temporal thermal pattern generation system of claim 5, wherein, Generating distorted color frames according to a unified color level across the entire time sequence includes: Extract the complete temporal deformation raster values ​​corresponding to the positions marked as original effective observation pixels by the effective observation mask from each acquisition phase, and calculate the median of the single-phase coherence coefficient of all the original effective observation pixels as the global coherence coefficient median; From the complete temporal deformed raster values ​​corresponding to the original effective observation pixels whose single-phase coherence coefficient is greater than or equal to the median of the global coherence coefficient, negative endpoints and positive endpoints are determined respectively, and the negative endpoints and positive endpoints constitute color level endpoints; All valid cell values ​​in the complete temporal deformable raster are truncated to the same set of color level endpoints, and mapped to the same divergent color band according to the negative endpoint to zero and the zero to positive endpoint respectively; the positions of the exclusion mask, invalid positions of the complete temporal deformable raster, and unsolved positions are kept transparent; All acquisition phases use the same color level endpoints, zero-value colors, and diverging color bands, without updating color level endpoints, zero-value colors, or diverging color bands frame by frame.

7. The landslide temporal thermal pattern generation system of claim 6, wherein, The observation support transparency frames generated based on the minimum propagation cost path include: For the original valid observation pixels within the acquisition phase in which the observation support transparency frame is being generated, the observation weight is divided by the maximum observation weight of all original valid observation pixels in that acquisition phase, and the result is used as the original observation base transparency; when the maximum observation weight does not exist or is zero, the observation support transparency frame of that acquisition phase is set to full transparency. Pixels that do not have original valid observations in the current acquisition phase but whose valid deformation values ​​are provided by the complete temporal deformation raster are used as filler pixels. For filler pixels, edges in the fixed eight-neighbor edge set whose boundary constraint deformation propagation weight is not less than the propagation participation threshold are used as traversable edges. The negative logarithm of the boundary constraint deformation propagation weight is used as the path cost. The minimum propagation cost path for any original valid observation pixel is searched within the same acquisition phase. The observation support transparency value of the completed pixel is obtained by multiplying the boundary constraint deformation propagation weight product on the minimum propagation cost path with the original observation base transparency of the original effective observation pixel at the endpoint; when there is no passable path from the completed pixel to the original effective observation pixel, the observation support transparency value of the completed pixel is set to zero. The original observation base transparency and observation support transparency values ​​are restricted to a closed interval between zero and one, and the observation support transparency frame does not reverse the modification of the complete temporal deformation raster.

8. The landslide temporal thermal pattern generation system of claim 7, wherein, For each acquisition phase, a boundary barrier line layer is generated, and a layer composite result is formed, including: The boundary barrier strength values ​​of all edges in all acquisition phases and the fixed eight-neighbor edge set are normalized using the same global maximum value, and the normalization result is used as the transparency of the edge-level boundary line; when the global maximum value does not exist or is zero, the boundary barrier line layer of all acquisition phases is set to full transparency. For each acquisition phase, the corresponding edge-level boundary line is rasterized into a boundary barrier line layer for that acquisition phase by connecting the pixel centers corresponding to the two endpoints of each edge in the fixed eight-neighbor edge set. For each acquisition phase, the corresponding deformation color frame is used as the bottom layer, the observation support transparency frame is used as the transparency layer, and the boundary barrier line layer is used as the top layer to form the layer composite result of that acquisition phase; The composite results of all layers are encoded in ascending order according to the acquisition time of the SAR single-view complex image sequence; The landslide deformation time-series heat map preserves a one-to-one correspondence between a unified geographic reference, fixed color scale, deformation unit, frame number and acquisition time; the fixed color scale includes negative value endpoints, positive value endpoints, zero value color and divergent color band.

Images