A terrain-adaptive SAR image data processing method and device, a terminal device and a computer readable storage medium

Abstract

The application discloses a terrain-adaptive SAR image data processing method and device, terminal equipment and a computer readable storage medium, and belongs to the technical field of radar remote sensing data processing. The method comprises the following steps: acquiring multi-source input data containing SAR images and DEM data; extracting electromagnetic scattering features and terrain geometric features to obtain multi-dimensional joint features; quantifying terrain complexity according to the multi-dimensional joint features, and dividing a region to be processed into first and second complexity grids; calling a first geocoding model to iteratively locate and output first geographic coordinates for the first complexity grid, and calling a second geocoding model to non-iteratively map and output second geographic coordinates for the second complexity grid; smoothing a transition area to obtain a transition region, and outputting a target coding result in combination with the coordinates and the transition region. The application solves the problems of power waste and local geometric distortion caused by the mismatch between the existing single processing mode and variable terrain.

Description

Technical Field

[0001] This invention relates to the field of radar remote sensing data processing technology, and in particular to a terrain-adaptive SAR image data processing method, apparatus, terminal equipment, and computer-readable storage medium. Background Technology

[0002] With the rapid development of synthetic aperture radar (SAR) remote sensing technology, SAR image geocoding has been widely used in fields such as surface deformation monitoring, disaster assessment and resource exploration, becoming an important technical means to eliminate radar geometric distortion and realize multi-source spatial data fusion analysis.

[0003] While existing SAR image geocoding methods can perform basic coordinate transformations, they typically employ a fixed, uniform processing granularity and a single solution model when dealing with large-scale, varied terrain. This makes them ill-suited to complex and diverse terrains, leading to a mismatch between the global processing mode and local terrain features. This results in wasted computing resources and geometric distortion in local positioning. Therefore, how to accurately identify terrain differences while ensuring high-precision positioning in complex terrain, thereby improving the overall computational efficiency of large-scale SAR image geocoding, has become a pressing technical problem to be solved in this field. Summary of the Invention

[0004] This invention provides a terrain-adaptive SAR image data processing method, apparatus, terminal device, and computer-readable storage medium, which can solve the problems of wasted computing power and local geometric distortion caused by the mismatch between the existing single processing mode and the variable terrain.

[0005] One embodiment of the present invention provides a terrain-adaptive SAR image data processing method, comprising: acquiring multi-source input data of the area to be processed, wherein the multi-source input data includes SAR image data and digital elevation model (DEM) data; Electromagnetic scattering features are extracted based on the SAR image data, and topographic geometric features are extracted based on the digital elevation model (DEM) data. The electromagnetic scattering features and the topographic geometric features are then combined to obtain multidimensional joint features. The terrain complexity of the region to be processed is quantized based on the multidimensional joint features, and the region to be processed is divided into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. For the first complex grid, a first geocoding model based on radar imaging geometry is invoked, and precise positioning is achieved through nonlinear iterative approximation using the first geocoding model, outputting the first geographic coordinates; for the second complex grid, a second geocoding model based on surface fitting is invoked, and a non-iterative direct mapping solution is achieved using the second geocoding model, outputting the second geographic coordinates. A smooth transition process is performed on the boundary region between the first complexity grid and the second complexity grid to obtain a transition region. Based on the first geographic coordinates, the second geographic coordinates, and the transition region, the geocoding result of the target SAR image of the area to be processed is output.

[0006] Furthermore, electromagnetic scattering features are extracted based on the SAR image data, and terrain geometric features are extracted based on the digital elevation model (DEM) data. The electromagnetic scattering features and the terrain geometric features are combined to obtain multidimensional joint features, including: based on the SAR image data, the interferometric coherence, backscattering intensity texture, and polarization scattering entropy of the SAR image data are extracted, and the interferometric coherence, the backscattering intensity texture, and the polarization scattering entropy are used as initial electromagnetic scattering features. Based on the digital elevation model (DEM) data, the local terrain curvature and elevation variation coefficient of the DEM data are calculated, and the local terrain curvature and elevation variation coefficient are used as initial terrain geometric features. The initial electromagnetic scattering features are spatially aligned with the initial terrain geometry features, and the spatially aligned initial electromagnetic scattering features are concatenated with the initial terrain geometry features to obtain the initial feature tensor; Calculate the covariance matrix between each feature dimension in the initial feature tensor, calculate the Mahalanobis distance between each feature vector in the initial feature tensor based on the covariance matrix, and use the calculated Mahalanobis distance as the normalized multidimensional joint feature.

[0007] Furthermore, the terrain complexity of the region to be processed is quantified based on the multidimensional joint features, and the region to be processed is divided into a first complexity grid and a second complexity grid based on the quantized terrain complexity. This includes: uniformly dividing the region to be processed into multiple initial basic grids; inputting the multidimensional joint features into a preset fuzzy clustering model for calculation to obtain the fuzzy membership degree of each pixel in the region to be processed belonging to the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category, respectively; wherein the complexity of the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category decreases sequentially. Obtain the fuzzy membership degree of the pixels contained in each of the initial basic grids, and calculate the grid comprehensive membership degree of each of the initial basic grids based on the fuzzy membership degree; determine the initial terrain complexity category of each of the initial basic grids based on the grid comprehensive membership degree of each of the initial basic grids. For each initial base grid, based on the initial terrain complexity category of the initial base grid, the adaptive grid reconstruction operation is repeatedly performed until the interferometric coherence variation of the currently divided grid is less than or equal to a preset variation threshold; wherein, the adaptive grid reconstruction operation includes: determining the target spatial resolution of the current base grid based on the current terrain complexity category of the current base grid; wherein, the initial current base grid is the initial base grid, and the initial current terrain complexity category is the initial terrain complexity category; re-dividing the corresponding current base grid based on the target spatial resolution to obtain several current divided grids; for each current divided grid, determining the interferometric coherence variation of the current divided grid according to the SAR image data; If the degree of change in interferometry is greater than the preset change threshold, the complexity level of the current terrain complexity category corresponding to the current grid division is increased to obtain the updated terrain complexity category. The current grid division is then used as the updated current base grid, and the next adaptive grid reconstruction operation is performed based on the updated terrain complexity category and the updated current base grid. If the degree of change in interferometry is less than or equal to the preset change threshold, the current terrain complexity category corresponding to the current grid division is categorized: if the current terrain complexity category is the first terrain complexity category, the current grid division is categorized into the first complexity grid; if the current terrain complexity category is the second terrain complexity category or the third terrain complexity category, the current grid division is categorized into the second complexity grid.

[0008] Furthermore, for the first complex grid, a first geocoding model based on radar imaging geometry is invoked. Precise positioning is achieved through nonlinear iterative approximation using the first geocoding model, outputting the first geographic coordinates. This includes: acquiring satellite orbit data of the area to be processed, and extracting velocity and azimuth acceleration information from the satellite orbit data; constructing a nonlinear Doppler equation based on the velocity information, the azimuth acceleration information, and a preset slant range quadratic correction term; acquiring multi-view observation data of the SAR image data, and combining the multi-view observation data with the nonlinear Doppler equation to construct a geometric observation overdetermined equation set; acquiring the first physical space index of the first complex grid, and converting the first physical space index into a first continuous code; rearranging the SAR image data corresponding to the first complex grid based on the first continuous code to obtain first rearranged data, and storing the first rearranged data in a continuous video memory address space, allocating a first floating-point data bit width for the first complex grid; Based on the first rearranged data stored in the contiguous video memory address space and the geometric observation overdetermined equation set, a nonlinear least squares iterative operation is repeatedly performed using the first floating-point data bit width until the updated target state parameter vector satisfies the preset convergence condition, thereby obtaining the first geographic coordinates; wherein, the nonlinear least squares iterative operation includes: substituting the first rearranged data into the geometric observation overdetermined equation set, and using the first floating-point data bit width to calculate the residual vector between the actual observed value of the multi-view observation data and the model prediction value corresponding to the current state parameter vector; wherein, the initial current state parameter vector is a pre-acquired initial estimated state parameter vector; based on the residual vector, a nonlinear least squares processing is performed on the current state parameter vector using the first floating-point data bit width to obtain the updated target state parameter vector; if the updated target state parameter vector does not satisfy the preset convergence condition, the updated target state parameter vector is used as the current state parameter vector, and the next nonlinear least squares iterative operation is performed based on the current state parameter vector; If the updated target state parameter vector satisfies the preset convergence condition, three-dimensional spatial coordinate components are extracted from the updated target state parameter vector, and the three-dimensional spatial coordinate components are used as the first geographic coordinates.

[0009] Furthermore, for the second complex grid, a second geocoding model based on surface fitting is invoked. A non-iterative direct mapping solution is performed using the second geocoding model to output the second geographic coordinates. This includes: obtaining the second physical space index of the second complex grid and converting it into a second continuous code; rearranging the SAR image data and DEM data corresponding to the second complex grid based on the second continuous code to obtain second rearranged data, and storing the second rearranged data in a continuous video memory address space; allocating a second floating-point data bit width to the second complex grid; extracting the polarization alpha angle and polarization scattering entropy from the second rearranged data; calculating the anisotropic weight coefficients for each data point within the second complex grid based on the polarization alpha angle and the polarization scattering entropy; constructing a weighted surface fitting equation based on the second rearranged data stored in the continuous video memory address space, the second floating-point data bit width, and the anisotropic weight coefficients; and solving the weighted surface fitting equation to obtain the initial mapped coordinates. The fitting residual value is calculated based on the initial mapping coordinates, and the fitting residual value is compared with a preset residual threshold; if the fitting residual value is less than or equal to the preset residual threshold, the initial mapping coordinates are used as the second geographic coordinates; If the fitted residual value is greater than the preset residual threshold, the grid corresponding to the fitted residual value is input into the first geocoding model for solving, and the solved geographic coordinates are used as the second geographic coordinates. Furthermore, a smooth transition process is performed on the boundary region between the first complexity grid and the second complexity grid to obtain a transition region. Based on the first geographic coordinates, the second geographic coordinates, and the transition region, the geocoding result of the target SAR image of the area to be processed is output, including: extracting the boundary nodes of the boundary region between the first complexity grid and the second complexity grid, and constructing an unstructured grid based on the boundary nodes; extracting a first coordinate set corresponding to the boundary nodes from the first geographic coordinates, and extracting a second coordinate set corresponding to the boundary nodes from the second geographic coordinates; performing spatial interpolation processing on the unstructured grid based on the first coordinate set and the second coordinate set to obtain the transition region; extracting the pixel gradient fields corresponding to the first complexity grid, the second complexity grid, and the transition region, and performing weighted smoothing processing on the overlapping region between the first complexity grid, the second complexity grid, and the transition region according to the corresponding pixel gradient fields to obtain an initial stitching result; acquiring ground reference data of the area to be processed, and constructing an error transformation model based on the ground reference data; and performing global correction on the initial stitching result based on the error transformation model to obtain the geocoding result of the target SAR image.

[0010] Furthermore, the terrain-adaptive SAR image data processing method further includes: Obtain the geocoding result of the target SAR image, and extract the initial pixel intensity set of the boundary region between the first complexity grid and the second complexity grid from the geocoding result of the target SAR image; perform data distribution statistics on the initial pixel intensity set to obtain a radiation difference histogram, and determine whether the radiation difference histogram satisfies the preset bimodal distribution characteristics; When the radiation difference histogram satisfies the bimodal distribution characteristic, the fuzzy membership degree of each pixel in the boundary area belonging to the first terrain complexity category is obtained, and the obtained fuzzy membership degree is used as the target fuzzy membership degree; the SAR image data is resampled based on the first geographic coordinates to obtain the first pixel intensity value of each pixel; the SAR image data is resampled based on the second geographic coordinates to obtain the second pixel intensity value of each pixel; according to the target fuzzy membership degree, the first pixel intensity value and the second pixel intensity value are weighted and mixed to obtain the target pixel intensity value; the target pixel intensity value is used to replace the initial pixel intensity set in the target SAR image geocoding result to obtain the updated target SAR image geocoding result, and the updated target SAR image geocoding result is used as the final SAR image geocoding result; If the radiation difference histogram does not satisfy the bimodal distribution characteristic, the geocoding result of the target SAR image is directly used as the final geocoding result of the SAR image.

[0011] Another embodiment of the present invention provides a terrain-adaptive SAR image data processing device, including: a data acquisition module, a feature extraction module, a grid division module, a coordinate solving module, and a result output module; The data acquisition module is used to acquire multi-source input data of the area to be processed, including SAR image data and digital elevation model (DEM) data. The feature extraction module is used to extract electromagnetic scattering features based on the SAR image data, extract terrain geometric features based on the digital elevation model (DEM) data, and combine the electromagnetic scattering features with the terrain geometric features to obtain multidimensional joint features. The grid partitioning module is used to quantize the terrain complexity of the region to be processed based on the multidimensional joint features, and to divide the region to be processed into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. The coordinate solving module is used to, for the first complex grid, call a first geocoding model based on radar imaging geometric relationships, perform nonlinear iterative approximation for accurate positioning through the first geocoding model, and output the first geographic coordinates; for the second complex grid, call a second geocoding model based on surface fitting, perform non-iterative direct mapping solution through the second geocoding model, and output the second geographic coordinates. The result output module is used to perform smooth transition processing on the boundary area between the first complexity grid and the second complexity grid to obtain a transition area, and output the geocoding result of the target SAR image of the area to be processed based on the first geographic coordinates, the second geographic coordinates and the transition area.

[0012] Another embodiment of the present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a terrain-adaptive SAR image data processing method of the present invention.

[0013] Another embodiment of the present invention provides a computer-readable storage medium item, including: a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the terrain-adaptive SAR image data processing method of the present invention.

[0014] The embodiments of the present invention have the following beneficial effects: This invention provides a terrain-adaptive SAR image data processing method. It acquires multi-source input data of the area to be processed, including SAR image data and digital elevation model (DEM) data. Electromagnetic scattering features of the SAR image data and terrain geometric features of the DEM data are extracted to construct multi-dimensional joint features. The terrain complexity of the area to be processed is quantified based on these multi-dimensional joint features, and the area is adaptively divided into a first complexity grid and a second complexity grid according to the quantized terrain complexity. Then, a corresponding geocoding model is configured according to the different complexities of the grids to solve for the three-dimensional geographic coordinates of the grids. Specifically, for the... The system describes a first complex grid, which is then used to call a first geocoding model based on radar imaging geometry. This model performs precise positioning through nonlinear iterative approximation, outputting first geographic coordinates. For the second complex grid, a second geocoding model based on surface fitting is called to perform a non-iterative direct mapping solution, outputting second geographic coordinates. Finally, a smooth transition is applied to the boundary region between the first and second complex grids to obtain a transition region. Based on the first geographic coordinates, the second geographic coordinates, and the transition region, the target SAR image geocoding result for the area to be processed is constructed and output.

[0015] By employing the above technical solution, this invention addresses the shortcomings of existing SAR image geocoding, which suffers from a mismatch between a single processing mode and variable terrain features, leading to ineffective waste of computing resources or geometric distortion in local positioning. It extracts multi-dimensional joint features to quantify terrain complexity and adaptively divides the grid, breaking the limitations of traditional fixed granularity at the data partitioning level and achieving precise matching between processing granularity and local terrain features. Simultaneously, based on the terrain complexity partitioning results, it dynamically configures differentiated geocoding models. In high-complexity areas, iterative models ensure image positioning accuracy, while non-iterative models accelerate mapping solutions in medium- and low-complexity areas. Smooth transitions are implemented in boundary regions, confining complex calculations only to necessary areas. This achieves adaptive synergy between geocoding strategies and local terrain space, improving the accuracy and overall computational efficiency of large-scale SAR image geocoding. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1This is a schematic flowchart of a terrain-adaptive SAR image data processing method provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the apparatus for a terrain-adaptive SAR image data processing method provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0021] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0024] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0025] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0026] Specifically, in the step descriptions of the various method embodiments below, the executing entity involved is collectively referred to as the "system". It should be understood that the "system" does not specifically refer to a single physical computing device, but rather to the overall logical architecture for implementing the solution of this application. In actual physical deployment, the "system" encompasses a business front-end terminal for interaction between surveyors and remote sensing analysts, a local remote sensing data processing server storing massive amounts of SAR image data, digital elevation model (DEM) data, and satellite orbit data, as well as a high-performance heterogeneous computing cluster or spatial information processing cloud platform deployed in remote or local data centers for performing large-scale parallel computing, GPU mixed-precision acceleration, and nonlinear iterative solutions.

[0027] Those skilled in the art should understand that the following steps, including the acquisition and preprocessing of multi-source input data, extraction of multi-dimensional joint features and Mahalanobis distance normalization, adaptive grid partitioning based on dynamic fuzzy clustering and temporal coherence monitoring, solution of differentiated geocoding models based on underlying memory rearrangement and mixed precision, and smooth transition and global correction of boundary areas, can be flexibly configured and executed on mapping data center hosts, GPU parallel acceleration nodes, spaceborne edge computing payloads, or distributed remote sensing collaborative processing platforms, without departing from the core logic of this application, depending on the actual security and confidentiality requirements of remote sensing mapping data, the computing power distribution of GPU heterogeneous computing resources, and the stringent requirements of large-scale high-resolution SAR image processing on memory bandwidth and concurrent throughput.

[0028] Therefore, the phrase "system performing an operation" used in the following text for the purpose of concise description includes both situations where a single hardware entity or underlying software performs the operation independently, and situations where multiple devices, such as remote sensing data center servers, spatial information cloud platforms, spaceborne processing payloads, and mapping data receiving terminals, perform the operation in a distributed and collaborative manner. It should not be narrowly interpreted as a strict limitation on the specific physical execution location of the algorithm or a specific hardware entity.

[0029] To address the problems of wasted computing power and local geometric distortion caused by the mismatch between existing single processing modes and varied terrain, an embodiment of the present invention provides a terrain-adaptive SAR image data processing method, comprising: Step S1: Obtain multi-source input data for the area to be processed, including SAR image data and digital elevation model (DEM) data; In a preferred embodiment, acquiring multi-source input data of the area to be processed, including SAR image data and digital elevation model (DEM) data, specifically includes: acquiring raw synthetic aperture radar (SAR) images covering the area to be processed, and sequentially performing radiometric calibration and noise suppression processing on the raw SAR images to output the SAR image data; acquiring initial digital elevation model (DEM) data covering the area to be processed, and performing spatial registration and hole filling processing on the initial DEM data to output the DEM data.

[0030] Preferably, the original synthetic aperture radar (SAR) image is subjected to radiometric calibration and noise suppression processing sequentially through the following steps to output the SAR image data: First, the area to be processed refers to a pre-defined specific three-dimensional geographic spatial range for which geocoding target tasks need to be performed. Since the original SAR image recorded by the radar sensor is usually a dimensionless digital quantization value, this value is easily affected by systematic factors such as gain drift within the sensor system and inhomogeneous antenna patterns, and cannot directly and objectively reflect the true physical scattering characteristics of surface targets. Therefore, this embodiment first performs radiometric calibration processing on the original SAR image. Specifically, based on the calibration constant and absolute incident angle parameter of the radar sensor, a radiometric calibration equation is constructed to convert the digital quantization value into a standardized backscattering coefficient, thereby eliminating the radiometric distortion caused by differences in sensor hardware. Second, for the speckle noise inherent in the SAR coherent imaging mechanism, an adaptive spatial filtering algorithm is used for noise suppression processing. In this embodiment, an enhanced Lee filter or a Frost filter is preferably used. This processing method calculates the local statistical characteristics of pixels (such as local mean and local variance) within a sliding window to adaptively adjust the filter weights: the smoothing intensity is increased in smooth homogeneous regions (such as water bodies and large farmlands) to fully suppress speckle noise, while the smoothing intensity is reduced in heterogeneous regions (such as building edges and corner reflectors). This effectively filters out speckle noise while preserving high-frequency edge and texture structure information in the image to the maximum extent, and finally outputs SAR image data with realistic radiation characteristics and controlled noise.

[0031] Preferably, the initial digital elevation model (DEM) data is spatially registered and filled with holes through the following steps to output the DEM data: First, initial DEM data covering the area to be processed is acquired (e.g., publicly available SRTM or ASTER GDEM datasets). To address potential coordinate reference system differences between multi-source data, spatial registration is performed on the initial DEM data, projecting and uniformly transforming it to a reference geographic coordinate system (e.g., WGS84) consistent with subsequent geocoding processing. Resampling is then used to align the grid division with the target processing framework, ensuring accurate location references for elevation information at a macroscopic spatial scale. Second, due to factors such as geometric obstruction from complex mountainous terrain, radar shadows, or optical cloud cover, the initial DEM data often contains areas lacking effective elevation observations (i.e., data holes). To avoid numerical calculation errors or division-to-zero anomalies caused by these voids in subsequent 3D coordinate calculations, this embodiment utilizes the effective elevation points of the void boundaries and surrounding neighborhoods, and employs spatial interpolation algorithms (such as bilinear interpolation, inverse distance weighted interpolation (IDW), or Kriging interpolation) to fill the voids in the invalid areas. This process estimates and reconstructs the missing elevation surfaces by fitting the changing trends of the local terrain, ensuring the spatial continuity and topological integrity of the elevation data within the processed area, and ultimately outputting the digital elevation model (DEM) data.

[0032] In this embodiment, through the targeted preprocessing operations for multi-source input data, on the one hand, radiometric calibration and adaptive noise suppression are used to recover the true backscattering physical quantities of SAR images, eliminating the interference of speckle noise on subsequent electromagnetic feature (such as coherence and polarization entropy) extraction and ensuring the purity of feature extraction; on the other hand, spatial registration and hole filling are used to ensure the spatial continuity and coordinate reference consistency of DEM data, avoiding divergence and local geometric distortion in 3D geographic coordinate calculation caused by missing or misaligned elevation information from the data source. The resulting SAR image data and digital elevation model (DEM) data have high accuracy and internal consistency in physical radiometric characterization and spatial geometric topology, thus providing a seamless and high-confidence data benchmark for subsequent extraction of multi-dimensional joint features and construction of an adaptive high-precision geocoding model.

[0033] Step S2: Extract electromagnetic scattering features based on the SAR image data, extract terrain geometric features based on the digital elevation model (DEM) data, and combine the electromagnetic scattering features with the terrain geometric features to obtain multidimensional joint features; In a preferred embodiment, electromagnetic scattering features are extracted based on the SAR image data, and topographic geometric features are extracted based on the digital elevation model (DEM) data. The electromagnetic scattering features and the topographic geometric features are combined to obtain multidimensional joint features, including: based on the SAR image data, the interferometric coherence, backscattering intensity texture, and polarization scattering entropy of the SAR image data are extracted, and the interferometric coherence, the backscattering intensity texture, and the polarization scattering entropy are used as initial electromagnetic scattering features. Based on the digital elevation model (DEM) data, the local terrain curvature and elevation variation coefficient of the DEM data are calculated, and the local terrain curvature and elevation variation coefficient are used as initial terrain geometric features. The initial electromagnetic scattering features are spatially aligned with the initial terrain geometry features, and the spatially aligned initial electromagnetic scattering features are concatenated with the initial terrain geometry features to obtain the initial feature tensor; Calculate the covariance matrix between each feature dimension in the initial feature tensor, calculate the Mahalanobis distance between each feature vector in the initial feature tensor based on the covariance matrix, and use the calculated Mahalanobis distance as the normalized multidimensional joint feature.

[0034] Preferably, the multidimensional joint features are constructed through the following implementation details: In a specific implementation, the SAR image sampling resolution of the processing area is set to 3m / pixel. First, for the extraction of electromagnetic scattering features: The interferometric coherence is calculated using two Sentinel-1 satellite SAR images with a time baseline of 12 days and a vertical spatial baseline of 150 meters; within a preset 7×7 pixel sliding window, the interferometric correlation is calculated using the complex domain correlation coefficient formula: ; In the formula, The value represents the interference coherence. The multiple pixel values ​​of the first SAR image within a preset sliding window; The multiple pixel values ​​of the second SAR image within a preset sliding window; for The complex conjugate of ; E[·] is the expectation operator; The polarization scattering entropy is calculated by measuring the coherence matrix of the fully polarimetric SAR data. eigenvalues The formula for obtaining and calculating this is: ; ; In the formula, H is the polarization scattering entropy; is the normalized probability corresponding to the i-th polarization scattering mechanism; , A 3×3 polarimetric coherence matrix (i.e., constructed for fully polarimetric SAR data) matrix); The backscattering intensity texture is obtained by calculating the average contrast of the gray-level co-occurrence matrix in four directions: 0°, 45°, 90°, and 135°. The gray-level co-occurrence matrix is ​​quantized to 256 levels.

[0035] Secondly, regarding the extraction of terrain geometric features: The local terrain curvature is based on the elevation values ​​of the current pixel and its surrounding 8 neighbors in the DEM data. The second-order partial derivative is calculated using quadratic surface fitting to construct a Gaussian curvature calculation model. ; Where K is the Gaussian curvature; E, F, G are the first fundamental form coefficients of the three-dimensional terrain surface; L, M, N are the second fundamental form coefficients of the three-dimensional terrain surface; The elevation variation coefficient is calculated based on an 11×11 pixel local statistical window to eliminate the dimensional influence of absolute altitude on the measurement of topographic relief. ; In the formula, The elevation variation coefficient is mentioned above. This represents the standard deviation of DEM elevation values ​​within a local statistical window. This represents the average elevation values ​​of the DEM within the local statistical window.

[0036] Finally, spatial alignment and feature combination are performed: By establishing a geometric mapping matrix from radar slant range projection to the geographic coordinate system, the five feature dimension matrices extracted above are uniformly resampled to the same geographic raster projection space as the DEM data (sampling interval of 5m); a three-dimensional feature tensor of shape H×W×5 is constructed, where H and W are the number of rows and columns of the geographic raster. The 5×5 covariance matrix Σ of the feature tensor in the channel dimension is calculated. For each pixel feature vector xi in the feature tensor, the Mahalanobis distance formula is used for calculation. ; in, The Mahalanobis distance is the feature vector of the i-th pixel. is the original feature vector of the i-th pixel (a column vector containing five dimensions: interference coherence, texture, polarization entropy, curvature, and coefficient of variation); μ is the global feature mean vector of all pixel feature vectors in the initial feature tensor; Let T be the covariance matrix of the initial feature tensor in the channel dimension; T is the matrix transpose symbol. This calculation process maps five types of features with different spatial scales and dimensions to a unified dimensionless distribution space, thereby eliminating the differences in the original data distribution and obtaining the final normalized multidimensional joint features.

[0037] In this embodiment, by configuring the specific implementation parameters and calculation model described above, the originally abstract electromagnetic physical quantities and topographic geometric quantities are transformed into standardized numerical descriptive features. For example, by calculating the interferometric coherence of a preset pixel window (such as a 7×7 window), the interference of speckle noise on the detection of dynamic surface features is effectively smoothed; by calculating the elevation variation coefficient of a local statistical window (such as an 11×11 window), a clear numerical magnitude distinction is established between gently undulating areas and drastically fluctuating mountainous areas. The multidimensional joint features constructed based on the above specific operational parameters can not only objectively characterize the differences in scattering mechanisms of SAR images under different land features, but also effectively solve the problem of quantization weight imbalance caused by differences in physical dimensions (such as a large range of elevation values ​​and a small range of polarization entropy values) in multi-source data fusion through the normalization of the covariance matrix and Mahalanobis distance. This technical solution provides a high-quality data foundation with uniform numerical distribution and decoupled feature space for subsequent fuzzy clustering, thereby effectively improving the accuracy and robustness of terrain complexity quantization and adaptive grid partitioning in complex scenarios.

[0038] Step S3: Quantize the terrain complexity of the region to be processed based on the multidimensional joint features, and divide the region to be processed into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. In a preferred embodiment, the terrain complexity of the region to be processed is quantified based on the multidimensional joint features, and the region to be processed is divided into a first complexity grid and a second complexity grid based on the quantized terrain complexity, including: uniformly dividing the region to be processed into multiple initial basic grids; inputting the multidimensional joint features into a preset fuzzy clustering model for calculation to obtain the fuzzy membership degree of each pixel in the region to be processed belonging to the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category, respectively; wherein the complexity of the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category decreases sequentially; Obtain the fuzzy membership degree of the pixels contained in each of the initial basic grids, and calculate the grid comprehensive membership degree of each of the initial basic grids based on the fuzzy membership degree; determine the initial terrain complexity category of each of the initial basic grids based on the grid comprehensive membership degree of each of the initial basic grids. For each initial base grid, based on the initial terrain complexity category of the initial base grid, the adaptive grid reconstruction operation is repeatedly performed until the interferometric coherence variation of the currently divided grid is less than or equal to a preset variation threshold; wherein, the adaptive grid reconstruction operation includes: determining the target spatial resolution of the current base grid based on the current terrain complexity category of the current base grid; wherein, the initial current base grid is the initial base grid, and the initial current terrain complexity category is the initial terrain complexity category; re-dividing the corresponding current base grid based on the target spatial resolution to obtain several current divided grids; for each current divided grid, determining the interferometric coherence variation of the current divided grid according to the SAR image data; If the degree of change in interferometry is greater than the preset change threshold, the complexity level of the current terrain complexity category corresponding to the current grid division is increased to obtain the updated terrain complexity category. The current grid division is then used as the updated current base grid, and the next adaptive grid reconstruction operation is performed based on the updated terrain complexity category and the updated current base grid. If the degree of change in interferometry is less than or equal to the preset change threshold, the current terrain complexity category corresponding to the current grid division is categorized: if the current terrain complexity category is the first terrain complexity category, the current grid division is categorized into the first complexity grid; if the current terrain complexity category is the second terrain complexity category or the third terrain complexity category, the current grid division is categorized into the second complexity grid.

[0039] Preferably, the terrain complexity is quantified and adaptive mesh reconstruction is performed through the following implementation details: In a specific implementation, the area to be processed is first uniformly divided, and the spatial size of the initial basic mesh is set to 100m × 100m. For the calculation of pixel-level fuzzy membership, the preset fuzzy clustering model adopts the Dynamic Fuzzy C-means Clustering Model (DFCM). Specifically, the core objective of the DFCM model is to minimize the sum of weighted distances from all pixel feature vectors to the cluster center, while considering the dimensional differences of different terrain features. The mathematical expression of the clustering objective function JT constructed in this embodiment is: ; In the formula, N is the total number of pixels in the input data; C is the preset number of clustering categories. In this embodiment, C=3, which corresponds to the first, second, and third terrain complexity categories, respectively. This represents the multidimensional joint feature vector of the i-th pixel; This represents the cluster center vector of the k-th category; represents the fuzzy membership degree of pixel i belonging to category k, with a value range of [0,1]; m is the fuzzy index, used to control the "softness" of the membership degree distribution, and in this embodiment, the preferred value is m=2; Represents pixel feature vector With cluster center The Mahalanobis distance between them.

[0040] During the solution process, the objective function is iteratively optimized to continuously update the fuzzy membership degree between cluster centers and pixels. The fuzzy membership degree... The update formula is: ; Cluster Center The iterative update formula is: ; The iteration stops when the change in the objective function JT is less than the preset convergence threshold, and the final fuzzy membership degree of each pixel is output. .

[0041] To calculate the overall membership degree of the grid and determine the initial category, in order to avoid interference from extreme pixels within the grid, the fuzzy membership degree of all pixels in the initial basic grid is obtained, and the interference coherence value corresponding to each pixel is extracted as a weighting factor. The overall membership degree of the 100m×100m initial basic grid is calculated using a coherence-weighted average algorithm. Subsequently, the initial terrain complexity category is determined based on the preset membership degree classification threshold: if it belongs to the first category, the grid's comprehensive membership degree is... If the value is greater than 0.8, it is classified as the first category of terrain complexity; if 0.3 ≤ If the value is ≤0.8, it is classified as the second terrain complexity category; if If the value is less than 0.3, it is classified as the third terrain complexity category.

[0042] Furthermore, regarding the resolution mapping and re-partitioning in the adaptive mesh reconstruction operation: a baseline mapping relationship between terrain complexity category and target spatial resolution is established. If the current terrain complexity category is the third terrain complexity category, the target spatial resolution remains at 100m×100m; if it is the second terrain complexity category, the target spatial resolution is set to 50m×50m; if it is the first terrain complexity category, the target spatial resolution is set to 25m×25m. Based on the determined target spatial resolution, a spatial segmentation algorithm is used to physically divide the current basic mesh, generating a current partitioned mesh of the corresponding size.

[0043] The mechanism for calculating and iteratively judging the temporal interferometric coherence variation is as follows: Three consecutive frames of SAR image data, including the current processing time phase, are acquired. The average interferometric coherence value of the currently divided grid region between adjacent time phases is calculated, and the decrease in coherence is defined as the interferometric coherence variation. In this embodiment, the preset variation threshold is set to 30%. If the calculated interferometric coherence variation is greater than 30%, it indicates that a sudden change such as a landslide or surface subsidence may have occurred in the local area. At this time, a grid level transition mechanism is forcibly triggered: the grid originally belonging to the third category is promoted to the second category, or the grid of the second category is promoted to the first category. Subsequently, the segmented grid is used as the updated current base grid, and the promoted complexity category is substituted. The smaller target spatial resolution is called again for segmentation; until the coherence variation within the grid converges to below 30%. Finally, the grids of the first terrain complexity category are assigned to the first complexity grid, and the grids of the second and third terrain complexity categories are assigned to the second complexity grid, serving as the basic units for subsequent model solving.

[0044] In this embodiment, the present invention introduces a specific dynamic fuzzy C-means clustering model (DFCM) and its underlying objective function and membership update formula into the grid partitioning process, and adopts the fuzzy exponent m and Mahalanobis distance. The objective function allows individual pixels to have different probability distributions (membership degrees). It simultaneously belongs to multiple terrain categories, solving the quantification problem of differences in the dimensions of multi-source features and blurred terrain boundaries. Combined with the dynamic fault-tolerant closed-loop mechanism that triggers re-division when the temporal coherence decreases by more than 30%, the system can generate large-size grids in flat areas to save computing power, while adaptively splitting into high-resolution fine-grained grids in complex deformation areas. This effectively improves the scientific nature and fault tolerance of SAR image processing granularity allocation in complex scenes.

[0045] Step S4: For the first complex grid, call the first geocoding model based on radar imaging geometric relationships, and perform precise positioning through nonlinear iterative approximation using the first geocoding model to output the first geographic coordinates; for the second complex grid, call the second geocoding model based on surface fitting, and perform direct mapping solution through the second geocoding model to output the second geographic coordinates. In a preferred embodiment, for the first complex grid, a first geocoding model based on radar imaging geometry is invoked, and precise positioning is achieved through nonlinear iterative approximation using the first geocoding model, outputting first geographic coordinates. This includes: acquiring satellite orbit data of the area to be processed, and extracting velocity information and azimuth acceleration information from the satellite orbit data; constructing a nonlinear Doppler equation based on the velocity information, the azimuth acceleration information, and a preset slant range quadratic correction term; acquiring multi-view observation data of the SAR image data, and combining the multi-view observation data with the nonlinear Doppler equation to construct a geometric observation overdetermined equation set; acquiring a first physical space index of the first complex grid, and converting the first physical space index into a first continuous code; rearranging the SAR image data corresponding to the first complex grid based on the first continuous code to obtain first rearranged data, and storing the first rearranged data in a continuous video memory address space, allocating a first floating-point data bit width for the first complex grid; Based on the first rearranged data stored in the contiguous video memory address space and the geometric observation overdetermined equation set, a nonlinear least squares iterative operation is repeatedly performed using the first floating-point data bit width until the updated target state parameter vector satisfies the preset convergence condition, thereby obtaining the first geographic coordinates; wherein, the nonlinear least squares iterative operation includes: substituting the first rearranged data into the geometric observation overdetermined equation set, and using the first floating-point data bit width to calculate the residual vector between the actual observed value of the multi-view observation data and the model prediction value corresponding to the current state parameter vector; wherein, the initial current state parameter vector is a pre-acquired initial estimated state parameter vector; based on the residual vector, a nonlinear least squares processing is performed on the current state parameter vector using the first floating-point data bit width to obtain the updated target state parameter vector; if the updated target state parameter vector does not satisfy the preset convergence condition, the updated target state parameter vector is used as the current state parameter vector, and the next nonlinear least squares iterative operation is performed based on the current state parameter vector; If the updated target state parameter vector satisfies the preset convergence condition, three-dimensional spatial coordinate components are extracted from the updated target state parameter vector, and the three-dimensional spatial coordinate components are used as the first geographic coordinates.

[0046] Preferably, for the second complexity grid, a second geocoding model based on surface fitting is invoked, and a non-iterative direct mapping solution is performed through the second geocoding model to output the second geographic coordinates. This includes: obtaining the second physical space index of the second complexity grid and converting the second physical space index into a second continuous code; rearranging the SAR image data and digital elevation model (DEM) data corresponding to the second complexity grid based on the second continuous code to obtain second rearranged data, and storing the second rearranged data in a continuous video memory address space; allocating a second floating-point data bit width to the second complexity grid; extracting the polarization alpha angle and polarization scattering entropy from the second rearranged data; calculating the anisotropic weight coefficients of each data point in the second complexity grid based on the polarization alpha angle and the polarization scattering entropy; constructing a weighted surface fitting equation based on the second rearranged data stored in the continuous video memory address space, the second floating-point data bit width, and the anisotropic weight coefficients; and solving the weighted surface fitting equation to obtain the initial mapped coordinates. The fitting residual value is calculated based on the initial mapping coordinates, and the fitting residual value is compared with a preset residual threshold; if the fitting residual value is less than or equal to the preset residual threshold, the initial mapping coordinates are used as the second geographic coordinates; If the fitted residual value is greater than the preset residual threshold, the grid corresponding to the fitted residual value is input into the first geocoding model for solution, and the solved geographic coordinates are used as the second geographic coordinates. Preferably, the following implementation details are used to perform nonlinear iterative approximation for precise positioning of the first complex grid: In specific implementation, a data-level reconstruction based on a heterogeneous computing architecture is first performed. The first physical space index is a two-dimensional grid row and column coordinate, which is converted into a one-dimensional Morton code (i.e., Z-order curve code) as the first continuous code through bit cross-operation. Based on this code, the elevation, scattering features, and other data corresponding to the first complex grid are rearranged so that adjacent grid data in physical space are stored continuously in the graphics processor's video memory physical address to improve cache hit rate. Since the terrain of the first complex grid (such as steep mountainous areas) is drastic and extremely sensitive to computational accuracy, the first floating-point data bit width allocated to it is specified as FP64 (double-precision floating-point number, 64 bits) to avoid the accumulation of rounding errors in nonlinear calculations. Subsequently, the first geocoding model (i.e., nonlinear enhanced distance-Doppler model, ERDM-NL) is constructed. Traditional Doppler equations often neglect satellite azimuth acceleration, leading to a significant amplification of positioning errors in high-dynamic regions. This embodiment, based on target kinematics principles, explicitly introduces a satellite azimuth acceleration term and a slant range quadratic correction term into the Doppler equations, constructing the following nonlinear Doppler equation: ; In the formula, Let λ represent the Doppler frequency, λ represent the radar wavelength, v and a represent the satellite velocity and acceleration vectors extracted from satellite orbit data, respectively, u represent the target point direction unit vector, and R represent the slant range of the radar to the target. Indicates the time interval in the azimuth direction. This represents the second-order correction term for the slant distance.

[0047] Furthermore, to overcome the initial value sensitivity problem caused by single-satellite orbital errors, and to obtain the joint observation data of the two satellites in the dual-satellite SAR system (i.e., the multi-look observation data), the distance equations of the two satellites are combined with the aforementioned nonlinear Doppler equations to construct the following set of geometric observation overdetermined equations: ; In the formula, , Represents the position vector of the binary stars. , Represents the velocity vector of the binary stars. This represents the target state parameter vector (three-dimensional geographic coordinates) to be solved.

[0048] In the iterative solution phase, the Levenberg-Marquardt (LM) algorithm is employed. The first rearranged data is substituted into the overdetermined equations to calculate the residual vector *r* between the observed values ​​and the model predictions. The regularization equations are then constructed and solved. Nonlinear least squares processing is performed, where J is the Jacobian matrix of the residuals with respect to the parameters, μ is the damping factor used to balance the gradient descent method and the Gauss-Newton method, and Δθ is the parameter update amount. The preset convergence condition is specifically set as follows: the coordinate change between two consecutive iterations is less than 0.1 pixel error, or the maximum number of iterations is reached (e.g., preset to 8). After satisfying the convergence condition, the three-dimensional spatial coordinate components of the target state parameter vector T are extracted. As the first geographic coordinate output.

[0049] Preferably, a non-iterative direct mapping solution is performed for the second complexity grid through the following implementation details: In terms of underlying data reconstruction, Morton coding is also used as the second continuous coding to perform continuous rearrangement of the data in video memory. The difference is that for medium-to-low complexity grids (such as plains and water bodies), considering their gentle terrain and high computational requirements, the second floating-point data bit width allocated to them is specified as FP16 (half-precision floating-point, 16 bits), and high-throughput parallel acceleration is achieved by combining the tensor cores in the graphics processor. In terms of model solving, a polynomial surface fitting model based on anisotropic weights (i.e., the second geocoding model) is adopted. To address the problem of spurious geometric distortion caused by the direction sensitivity of scattering in homogeneous regions, polarization entropy is extracted from the second rearranged data. With polarization Alpha angle Calculate the anisotropy weighting coefficients This weighting coefficient automatically reduces the confidence weight of scattering direction-sensitive feature points. Subsequently, a weighted surface fitting equation is constructed: v, in the formula These are the actual elevation values ​​of the DEM. The values ​​are predicted by a polynomial fitting, where M is the total number of sample points participating in the surface fitting. A direct, non-iterative mapping method is used to quickly output the initial mapped coordinates. To prevent fitting failures caused by half-precision calculations or abnormal terrain, a strict residual monitoring loop is designed. The preset residual threshold is set to 1.2 pixels. The geometric residual of the fitting result is calculated. If the residual is ≤1.2 pixels, the initial mapped coordinates are directly output as the second geographic coordinates. If the residual is >1.2 pixels, it indicates that there may be unidentified terrain abrupt changes or half-precision rounding errors in the local area. In this case, the system automatically triggers a model degradation and rollback mechanism: the grid is reassigned to the first geocoding model, and FP64 high-precision resources are used to perform ERDM-NL nonlinear iterative solutions, thereby ensuring the absolute accuracy of the global coordinates.

[0050] In this embodiment, through the aforementioned two-layer differentiated processing architecture designed for grids with varying terrain complexity, the present invention achieves an adaptive balance between algorithm accuracy and hardware performance. On the one hand, at the software algorithm level, a nonlinear Doppler equation incorporating acceleration and slant range quadratic correction is introduced for high-complexity grids, and an overdetermined equation is constructed using multi-view constraints for LM iteration. This theoretically eliminates perspective shrinkage errors in steep mountainous areas, achieving sub-pixel-level (less than 0.1 pixels) extreme terrain localization. For medium- and low-complexity grids, polarization-weighted polynomial fitting is introduced to avoid blind iteration, improve the mapping speed in flat areas, and ensure algorithm accuracy through a 1.2-pixel residual monitoring closed loop. On the other hand, at the hardware acceleration level, by deeply coupling Z-order data memory rearrangement and mixed-precision computation (FP64 and FP16) into the specific geocoding model, not only is the memory bandwidth bottleneck eliminated through continuous memory access, but on-demand allocation of computing resources is also achieved. This allows for high-precision geocoding processing of massive SAR images to improve overall computational efficiency while maintaining mapping-grade accuracy.

[0051] Step S5: Perform smooth transition processing on the boundary region between the first complexity grid and the second complexity grid to obtain a transition region, and output the geocoding result of the target SAR image of the area to be processed based on the first geographic coordinates, the second geographic coordinates and the transition region.

[0052] In a preferred embodiment, a smooth transition processing is performed on the boundary region between the first complexity grid and the second complexity grid to obtain a transition region. Based on the first geographic coordinates, the second geographic coordinates, and the transition region, the geocoding result of the target SAR image of the area to be processed is output, including: extracting the boundary nodes of the boundary region between the first complexity grid and the second complexity grid, and constructing an unstructured grid based on the boundary nodes; extracting a first coordinate set corresponding to the boundary nodes from the first geographic coordinates, and extracting a second coordinate set corresponding to the boundary nodes from the second geographic coordinates; performing spatial interpolation processing on the unstructured grid based on the first coordinate set and the second coordinate set to obtain the transition region; extracting the pixel gradient fields corresponding to the first complexity grid, the second complexity grid, and the transition region, and performing weighted smoothing processing on the overlapping region between the first complexity grid, the second complexity grid, and the transition region according to the corresponding pixel gradient fields to obtain an initial stitching result; acquiring ground reference data of the area to be processed, and constructing an error transformation model based on the ground reference data; and performing global correction on the initial stitching result based on the error transformation model to obtain the geocoding result of the target SAR image.

[0053] Preferably, the transition region is obtained by smoothing the boundary area through the following implementation details: In specific implementation, since the first complexity grid and the second complexity grid use completely different spatial resolutions and geocoding solution models, direct splicing will lead to severe geometric misalignment and coordinate jumps. Therefore, firstly, an overlapping buffer area is defined, and the boundaries of each grid are extended outward by a preset proportion (e.g., 10% spatial width), so that adjacent first complexity grids and second complexity grids form a boundary area with overlapping coverage. Subsequently, all boundary nodes in this boundary area are extracted, and the Delaunay triangulation algorithm is used to connect these boundary nodes to construct an unstructured grid that adapts to arbitrary boundary shapes, which serves as the skeleton for resolution gradient. In the coordinate assignment stage, a first coordinate set is extracted from the first geographic coordinates that have been solved with high precision, which is closer to the boundary nodes on the high complexity side, and a second coordinate set is extracted from the second geographic coordinates that have been solved by fast mapping, which is closer to the boundary nodes on the medium and low complexity side. Using the first coordinate set and the second coordinate set as known control points, the three-dimensional spatial coordinates of all unknown vertices inside the unstructured grid are calculated using a bilinear interpolation algorithm. This interpolation process smooths out the coordinate differences between different models, generating the transition region with continuous coordinate gradient characteristics.

[0054] Preferably, the initial stitching result is obtained by weighted smoothing based on the pixel gradient field through the following implementation details: To eliminate the inconsistency in backscattered radiation and geometric stitching gaps caused by model differences, this embodiment performs seamless fusion in the gradient domain. Specifically, the pixel gradient vectors (i.e., the spatial rate of change of image grayscale in the horizontal and vertical directions) of the first complexity grid, the second complexity grid, and the transition region in the overlapping region are calculated respectively and aggregated to form the overall pixel gradient field. A discrete Poisson fusion algorithm is adopted to construct a functional equation with the objective of minimizing the gradient difference between the fusion region and the source image. (Where ∇f is the gradient of the image to be fused, and v is the gradient field of the source pixel). Using the outer boundary pixel values ​​of the overlapping region as Dirichlet boundary conditions, the Laplace equation corresponding to the above functional is solved. This process forces a smooth transition of pixel values ​​within overlapping areas while maintaining the high-frequency texture structure within each grid. After gradient-domain weighted smoothing, the initial stitching result, which eliminates visual seams and radial tortuosity, is output.

[0055] Preferably, global correction and geocoding results of target SAR images are output through the following implementation details: Due to the inherent theoretical biases of different solution models, floating-point rounding errors in GPU mixed-precision calculations, and residual satellite orbit errors, slight systematic coordinate shifts will occur during large-scale stitching. Therefore, an external reference must be introduced for global control. First, uniformly distributed natural features or artificial landmarks (such as road intersections and building roof corners) with stable scattering characteristics are obtained from public geographic databases (such as OpenStreetMap vector road networks) or high-precision optical orthophotos as ground control points (GCPs), i.e., the ground reference data. Using feature matching algorithms such as Scale Invariant Feature Transform (SIFT), the initial stitching results are spatially registered with the ground reference data at the sub-pixel level to extract corresponding homonymous point pairs. Second, based on the extracted homonymous point pairs, a two-dimensional affine transformation model (i.e., the error transformation model) reflecting the mapping relationship between pixel coordinates and real geographic coordinates is constructed. The affine transformation model includes transformation parameters such as translation, rotation, scaling, and shearing. The objective function is optimized using the least squares method, minimizing the sum of squared spatial residuals between all control point pairs to obtain the optimal affine transformation parameter matrix. Finally, using the obtained optimal affine transformation model, a global coordinate transformation (i.e., global correction) is performed on all pixel coordinates in the initial stitching result. This correction process uniformly eliminates systematic coordinate drift across the entire map, causing the absolute positioning error of the final image to converge to the sub-pixel level (e.g., less than 1.2 pixels in urban areas and less than 2.0 pixels in mountainous areas), and outputs geocoding results for the target SAR image that meet mapping-grade accuracy requirements.

[0056] In this embodiment, through the aforementioned smooth transition and global correction mechanism for boundary areas, the present invention solves the problem of seam fragmentation and error accumulation caused by the stitching of multi-scale grids and heterogeneous solution models. On the one hand, by introducing overlapping extension and Delaunay unstructured grids, combined with bilinear spatial interpolation and Poisson gradient domain fusion, buffers are built in both geometric coordinate and radiation intensity dimensions, which can resolve the resolution abrupt changes and coordinate jumps caused by model differences, achieving seamless mosaicking across complex regions. On the other hand, by combining the underlying algorithm solution with high-precision ground reference data, a global systematic correction is performed through a least-squares optimized affine transformation model, effectively absorbing and eliminating residual systematic offsets caused by single-satellite orbit errors, DEM interpolation errors, and floating-point rounding. This not only ensures high fidelity of local complex terrain but also ensures a high degree of uniformity in visual continuity and absolute positioning accuracy of large-scale global image products, ultimately outputting reliable and high-quality SAR image geocoding results.

[0057] Preferably, after outputting the geocoding result of the target SAR image of the area to be processed, the method further includes: obtaining the geocoding result of the target SAR image; extracting the initial pixel intensity set of the boundary region between the first complexity grid and the second complexity grid from the geocoding result of the target SAR image; performing data distribution statistics on the initial pixel intensity set to obtain a radiation difference histogram; and determining whether the radiation difference histogram satisfies the preset bimodal distribution characteristics. When the radiation difference histogram satisfies the bimodal distribution characteristic, the fuzzy membership degree of each pixel in the boundary area belonging to the first terrain complexity category is obtained, and the obtained fuzzy membership degree is used as the target fuzzy membership degree; the SAR image data is resampled based on the first geographic coordinates to obtain the first pixel intensity value of each pixel; the SAR image data is resampled based on the second geographic coordinates to obtain the second pixel intensity value of each pixel; according to the target fuzzy membership degree, the first pixel intensity value and the second pixel intensity value are weighted and mixed to obtain the target pixel intensity value; the target pixel intensity value is used to replace the initial pixel intensity set in the target SAR image geocoding result to obtain the updated target SAR image geocoding result, and the updated target SAR image geocoding result is used as the final SAR image geocoding result; If the radiation difference histogram does not satisfy the bimodal distribution characteristic, the geocoding result of the target SAR image is directly used as the final geocoding result of the SAR image.

[0058] Preferably, the data distribution statistics and weighted hybrid calculation based on the radiation difference histogram are performed through the following implementation details to output the final SAR image geocoding result: In specific implementation, although the aforementioned steps have performed weighted smoothing of geometric boundaries through pixel gradient fields, due to the essential differences between the first geocoding model (nonlinear iterative model) and the second geocoding model (polynomial fitting model) in the systematic solution mechanism of radar backscattering intensity, the pixel brightness (i.e., radiation intensity) in the boundary area may still have radiation faults that do not conform to the physical characteristics of real ground objects. Therefore, this embodiment introduces a radiation quality verification and feedback closed-loop mechanism. First, the frequency distribution statistics of gray level or physical scattering coefficients are performed on the initial pixel intensity set in the boundary area to obtain a radiation difference histogram reflecting the change of pixel number with radiation intensity. Then, it is determined whether the histogram satisfies the preset bimodal distribution feature. Specifically, the preset bimodal distribution feature refers to the detection of two independent local maxima with significant peaks on the probability density curve, and the depth of the trough between the two peaks exceeds a preset extreme value ratio threshold. In a physical sense, a bimodal distribution means that the radiation intensity on both sides of the boundary region does not transition smoothly, but rather exhibits a step-like brightness jump (i.e., there is a clear splicing gap and inconsistency in radiation).

[0059] When the radiation difference histogram is detected to meet the bimodal distribution characteristics, the radiation smoothing feedback repair process is triggered: The system cross-module calls the original parameters output by the fuzzy clustering model in the previous step to accurately obtain the fuzzy membership degree of each pixel in the boundary area belonging to the first terrain complexity category (i.e., the high complexity category). This is used as the target fuzzy membership degree. This parameter naturally possesses a smooth gradient characteristic between 0 and 1, perfectly matching the weight allocation requirements of the transition zone. Next, reverse resampling of the spatial coordinates is performed. For this pixel, the already calculated high-precision first geographic coordinates are back-projected onto the original SAR image data (i.e., slant range / azimuth complex image), and the corresponding backscattering intensity value is extracted using bilinear interpolation or bicubic interpolation algorithms, denoted as the first pixel intensity value. Similarly, the second pixel intensity value is extracted by back-projecting the second geographic coordinates onto the original SAR image data and resampling. Subsequently, a weighted hybrid calculation is performed based on the target fuzzy membership degree, and the mathematical expression of its fusion equation is as follows: ; In the formula, This is the calculated target pixel intensity value. Using this formula, the region closer to the first complexity grid ( The closer a pixel is to 1, the more its intensity relies on the results of the first geocoding model; in regions closer to the second complexity grid ( The closer the pixel intensity is to 0, the more it relies on the results of the second geocoding model, thus forcing a smooth transition in radiometric intensity at the pixel scale. Finally, the initial set of pixel intensities with bimodal transitions is replaced one by one with the calculated target pixel intensity values, eliminating visual faults, and the updated target SAR image geocoding result is output and confirmed as the final SAR image geocoding result. Conversely, if the radiometric difference histogram shows a unimodal distribution or a uniform transition distribution (i.e., does not satisfy the bimodal distribution characteristic), it indicates that the preceding geometric and gradient smoothing has sufficiently ensured radiometric consistency. In this case, the system directly skips the computationally expensive resampling and weighted calculation steps and directly confirms the original target SAR image geocoding result as the final result.

[0060] In this embodiment, by employing the post-processing quality control closed loop based on the radiometric difference histogram and fuzzy membership, the present invention solves the problems of radiometric inconsistency and visual discontinuity caused by multi-source heterogeneous model stitching. On the one hand, by introducing a detection mechanism based on the bimodal distribution characteristics of the histogram, the system possesses self-checking capabilities, enabling it to identify local areas with genuine radiometric distortion and trigger repair as needed, avoiding the waste of computational power and loss of original texture caused by blind global smoothing. On the other hand, by reusing the fuzzy membership used for grid division in the early stage as a dynamic weighting factor for radiometric fusion, not only is efficient cross-stage transfer of internal system parameters achieved, but also the smooth curve of radiometric transition is physically consistent with the spatial gradient trend of real terrain complexity. While ensuring the geometric accuracy of large-scale geocoding results, the final SAR image product is endowed with extremely high radiometric physical fidelity and seamless visual continuity, meeting the requirements of high-precision mapping and quantitative remote sensing analysis of the land surface.

[0061] like Figure 2 As shown, another embodiment of the present invention also provides a terrain-adaptive SAR image data processing device, including: a data acquisition module, a feature extraction module, a grid division module, a coordinate solving module, and a result output module; The data acquisition module is used to acquire multi-source input data of the area to be processed, including SAR image data and digital elevation model (DEM) data. The feature extraction module is used to extract electromagnetic scattering features based on the SAR image data, extract terrain geometric features based on the digital elevation model (DEM) data, and combine the electromagnetic scattering features with the terrain geometric features to obtain multidimensional joint features. The grid partitioning module is used to quantize the terrain complexity of the region to be processed based on the multidimensional joint features, and to divide the region to be processed into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. The coordinate solving module is used to, for the first complex grid, call a first geocoding model based on radar imaging geometric relationships, perform nonlinear iterative approximation for accurate positioning through the first geocoding model, and output the first geographic coordinates; for the second complex grid, call a second geocoding model based on surface fitting, perform non-iterative direct mapping solution through the second geocoding model, and output the second geographic coordinates. The result output module is used to perform smooth transition processing on the boundary area between the first complexity grid and the second complexity grid to obtain a transition area, and output the geocoding result of the target SAR image of the area to be processed based on the first geographic coordinates, the second geographic coordinates and the transition area.

[0062] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the terrain-adaptive SAR image data processing method provided by any of the above-described method embodiments of the present invention.

[0063] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0064] Based on the above-described embodiment of a terrain-adaptive SAR image data processing method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements any one of the terrain-adaptive SAR image data processing methods of the present invention.

[0065] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0066] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0067] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0068] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the terrain-adaptive SAR image data processing method described in any of the above-described method embodiments of the present invention.

[0069] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0070] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A terrain-adaptive SAR image data processing method, characterized in that, include: Acquire multi-source input data for the area to be processed, including SAR image data and digital elevation model (DEM) data; Electromagnetic scattering features are extracted based on the SAR image data, and topographic geometric features are extracted based on the digital elevation model (DEM) data. The electromagnetic scattering features and the topographic geometric features are then combined to obtain multidimensional joint features. The terrain complexity of the region to be processed is quantized based on the multidimensional joint features, and the region to be processed is divided into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. For the first complex grid, a first geocoding model based on radar imaging geometry is invoked, and precise positioning is achieved through nonlinear iterative approximation using the first geocoding model, outputting the first geographic coordinates; for the second complex grid, a second geocoding model based on surface fitting is invoked, and a non-iterative direct mapping solution is achieved using the second geocoding model, outputting the second geographic coordinates. A smooth transition process is performed on the boundary region between the first complexity grid and the second complexity grid to obtain a transition region. Based on the first geographic coordinates, the second geographic coordinates, and the transition region, the geocoding result of the target SAR image of the area to be processed is output.

2. The terrain-adaptive SAR image data processing method as described in claim 1, characterized in that, Electromagnetic scattering features are extracted based on the SAR image data, and topographic geometric features are extracted based on the digital elevation model (DEM) data. The electromagnetic scattering features and the topographic geometric features are combined to obtain multidimensional joint features, including: based on the SAR image data, the interferometric coherence, backscattering intensity texture, and polarization scattering entropy of the SAR image data are extracted, and the interferometric coherence, the backscattering intensity texture, and the polarization scattering entropy are used as initial electromagnetic scattering features. Based on the digital elevation model (DEM) data, the local terrain curvature and elevation variation coefficient of the DEM data are calculated, and the local terrain curvature and elevation variation coefficient are used as initial terrain geometric features. The initial electromagnetic scattering features are spatially aligned with the initial terrain geometry features, and the spatially aligned initial electromagnetic scattering features are concatenated with the initial terrain geometry features to obtain the initial feature tensor; Calculate the covariance matrix between each feature dimension in the initial feature tensor, calculate the Mahalanobis distance between each feature vector in the initial feature tensor based on the covariance matrix, and use the calculated Mahalanobis distance as the normalized multidimensional joint feature.

3. The terrain-adaptive SAR image data processing method as described in claim 2, characterized in that, The terrain complexity of the region to be processed is quantified based on the multidimensional joint features, and the region to be processed is divided into a first complexity grid and a second complexity grid based on the quantized terrain complexity. This includes: uniformly dividing the region to be processed into multiple initial basic grids; inputting the multidimensional joint features into a preset fuzzy clustering model for calculation to obtain the fuzzy membership degree of each pixel in the region to be processed belonging to the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category, respectively; wherein the complexity of the first terrain complexity category, the second terrain complexity category, and the third terrain complexity category decreases sequentially. Obtain the fuzzy membership degree of the pixels contained in each of the initial basic grids, and calculate the grid comprehensive membership degree of each of the initial basic grids based on the fuzzy membership degree; determine the initial terrain complexity category of each of the initial basic grids based on the grid comprehensive membership degree of each of the initial basic grids. For each initial base grid, based on the initial terrain complexity category of the initial base grid, the adaptive grid reconstruction operation is repeatedly performed until the interferometric coherence variation of the current grid is less than or equal to a preset variation threshold. The adaptive grid reconstruction operation includes: determining the target spatial resolution of the current base grid based on the current terrain complexity category; wherein the initial current base grid is the initial base grid, and the initial current terrain complexity category is the initial terrain complexity category; re-dividing the corresponding current base grid based on the target spatial resolution to obtain several current grids; for each current grid, determining the interferometric coherence variation of the current grid based on the SAR image data; if the interferometric coherence variation is greater than the preset variation threshold, increasing the complexity level of the current terrain complexity category corresponding to the current grid to obtain an updated terrain complexity category, and using the current grid as the updated current base grid, performing the next adaptive grid reconstruction operation based on the updated terrain complexity category and the updated current base grid. When the degree of change in interference coherence is less than or equal to the preset change threshold, the current terrain complexity category corresponding to the current grid division is used for classification: if the current terrain complexity category is the first terrain complexity category, the current grid division is classified into the first complexity grid; if the current terrain complexity category is the second terrain complexity category or the third terrain complexity category, the current grid division is classified into the second complexity grid.

4. The terrain-adaptive SAR image data processing method as described in claim 3, characterized in that, For the first complex grid, a first geocoding model based on radar imaging geometry is invoked. Precise positioning is achieved through nonlinear iterative approximation using the first geocoding model, outputting the first geographic coordinates. This includes: acquiring satellite orbit data of the area to be processed, and extracting velocity and azimuth acceleration information from the satellite orbit data; constructing a nonlinear Doppler equation based on the velocity information, the azimuth acceleration information, and a preset slant range quadratic correction term; acquiring multi-view observation data of the SAR image data, and combining the multi-view observation data with the nonlinear Doppler equation to construct a geometric observation overdetermined equation set; acquiring the first physical space index of the first complex grid, and converting the first physical space index into a first continuous code; rearranging the SAR image data corresponding to the first complex grid based on the first continuous code to obtain first rearranged data, and storing the first rearranged data in a continuous video memory address space, allocating a first floating-point data bit width for the first complex grid; Based on the first rearranged data stored in the contiguous video memory address space and the geometric observation overdetermined equation set, a nonlinear least squares iterative operation is repeatedly performed using the first floating-point data bit width until the updated target state parameter vector satisfies the preset convergence condition, thereby obtaining the first geographic coordinates; wherein, the nonlinear least squares iterative operation includes: substituting the first rearranged data into the geometric observation overdetermined equation set, and using the first floating-point data bit width to calculate the residual vector between the actual observed value of the multi-view observation data and the model prediction value corresponding to the current state parameter vector; wherein, the initial current state parameter vector is a pre-acquired initial estimated state parameter vector; based on the residual vector, performing nonlinear least squares processing on the current state parameter vector using the first floating-point data bit width to obtain the updated target state parameter vector; if the updated target state parameter vector does not satisfy the preset convergence condition, the updated target state parameter vector is used as the current state parameter vector, and the next nonlinear least squares iterative operation is performed based on the current state parameter vector; If the updated target state parameter vector satisfies the preset convergence condition, three-dimensional spatial coordinate components are extracted from the updated target state parameter vector, and the three-dimensional spatial coordinate components are used as the first geographic coordinates.

5. The terrain-adaptive SAR image data processing method as described in claim 4, characterized in that, For the second complexity grid, a second geocoding model based on surface fitting is invoked. A non-iterative direct mapping solution is then performed using this second geocoding model to output the second geographic coordinates, including: Obtain the second physical space index of the second complexity grid, and convert the second physical space index into a second continuous code; Based on the second continuous encoding, the SAR image data and digital elevation model (DEM) data corresponding to the second complexity grid are rearranged to obtain the second rearranged data, and the second rearranged data is stored in a continuous video memory address space. Allocate a second floating-point data bit width to the second complexity grid; Extract the polarization alpha angle and polarization scattering entropy from the second rearranged data; Based on the polarization Alpha angle and the polarization scattering entropy, calculate the anisotropic weight coefficients of each data point in the second complexity grid; Based on the second rearranged data stored in the continuous video memory address space, the second floating-point data bit width, and the anisotropic weighting coefficients, a weighted surface fitting equation is constructed. The initial mapped coordinates are obtained by solving the weighted surface fitting equation. The fitting residual value is calculated based on the initial mapping coordinates, and the fitting residual value is compared with a preset residual threshold. If the fitted residual value is less than or equal to a preset residual threshold, then the initial mapped coordinates are used as the second geographic coordinates; If the fitted residual value is greater than the preset residual threshold, the grid corresponding to the fitted residual value is input into the first geocoding model for solving, and the solved geographic coordinates are used as the second geographic coordinates.

6. The terrain-adaptive SAR image data processing method as described in claim 5, characterized in that, The process involves: smoothing the boundary region between the first and second complexity grids to obtain a transition region; and outputting the geocoding result of the target SAR image of the area to be processed based on the first geographic coordinates, the second geographic coordinates, and the transition region. This includes: extracting the boundary nodes of the boundary region between the first and second complexity grids and constructing an unstructured grid based on the boundary nodes; extracting a first coordinate set corresponding to the boundary nodes from the first geographic coordinates and a second coordinate set corresponding to the boundary nodes from the second geographic coordinates; performing spatial interpolation on the unstructured grid based on the first and second coordinate sets to obtain the transition region; extracting the pixel gradient fields corresponding to the first, second, and transition regions, and performing weighted smoothing on the overlapping regions between the first, second, and transition regions based on the corresponding pixel gradient fields to obtain an initial stitching result; acquiring ground reference data for the area to be processed and constructing an error transformation model based on the ground reference data; and performing global correction on the initial stitching result based on the error transformation model to obtain the geocoding result of the target SAR image.

7. The terrain-adaptive SAR image data processing method as described in claim 6, characterized in that, Also includes: Obtain the geocoding result of the target SAR image, and extract the initial pixel intensity set of the boundary region between the first complexity grid and the second complexity grid from the geocoding result of the target SAR image; perform data distribution statistics on the initial pixel intensity set to obtain a radiation difference histogram, and determine whether the radiation difference histogram satisfies the preset bimodal distribution characteristics; When the radiation difference histogram satisfies the bimodal distribution feature, the fuzzy membership degree of each pixel in the boundary area belonging to the first terrain complexity category is obtained, and the obtained fuzzy membership degree is used as the target fuzzy membership degree; the SAR image data is resampled based on the first geographic coordinates to obtain the first pixel intensity value of each pixel. The SAR image data is resampled based on the second geographic coordinates to obtain the second pixel intensity value of each pixel. Based on the target fuzzy membership degree, the first pixel intensity value and the second pixel intensity value are weighted and mixed to obtain the target pixel intensity value; Using the target pixel intensity value, the initial pixel intensity set in the geocoding result of the target SAR image is replaced to obtain the updated geocoding result of the target SAR image, and the updated geocoding result of the target SAR image is used as the final geocoding result of the SAR image. If the radiation difference histogram does not satisfy the bimodal distribution characteristic, the geocoding result of the target SAR image is directly used as the final geocoding result of the SAR image.

8. A terrain-adaptive SAR image data processing device, characterized in that, include: The module includes a data acquisition module, a feature extraction module, a mesh generation module, a coordinate solving module, and a result output module. The data acquisition module is used to acquire multi-source input data of the area to be processed, including SAR image data and digital elevation model (DEM) data. The feature extraction module is used to extract electromagnetic scattering features based on the SAR image data, extract terrain geometric features based on the digital elevation model (DEM) data, and combine the electromagnetic scattering features with the terrain geometric features to obtain multidimensional joint features. The grid partitioning module is used to quantize the terrain complexity of the region to be processed based on the multidimensional joint features, and to divide the region to be processed into a first complexity grid and a second complexity grid based on the quantized terrain complexity; wherein the terrain complexity of the first complexity grid is higher than that of the second complexity grid. The coordinate solving module is used to, for the first complex grid, call a first geocoding model based on radar imaging geometric relationships, perform nonlinear iterative approximation for accurate positioning through the first geocoding model, and output the first geographic coordinates; for the second complex grid, call a second geocoding model based on surface fitting, perform non-iterative direct mapping solution through the second geocoding model, and output the second geographic coordinates. The result output module is used to perform smooth transition processing on the boundary area between the first complexity grid and the second complexity grid to obtain a transition area, and output the geocoding result of the target SAR image of the area to be processed based on the first geographic coordinates, the second geographic coordinates and the transition area.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the terrain-adaptive SAR image data processing method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the terrain-adaptive SAR image data processing method as described in any one of claims 1-7.

Images