GPU-based large-breadth SAR image rapid positioning method

Abstract

The invention discloses a GPU (Graphics Processing Unit)-based large-breadth SAR (Synthetic Aperture Radar) image quick positioning method, which adopts a CPU and GPU heterogeneous parallel computing architecture and utilizes the many-core characteristics of a GPU to ensure that the efficiency is higher and the time consumption is greatly reduced in the large-breadth SAR image quick positioning process and can reach a minute level or even a second level; the parallel advantage of the GPU is fully exerted, the generation of a plurality of parameters is parallelized, the degree of parallelism is improved, and the operation efficiency is greatly improved; the characteristics that SAR images obtained through multi-mode imaging are located in different coordinate systems, and transplantation and development are tedious are considered, the relation between a first-level SAR image and a second-level correction image is directly established through fitting, processing in the different coordinate systems can be completed only by updating the coordinate conversion relation, and the processing efficiency is improved. Therefore, positioning processing of the SAR image obtained by multi-mode imaging is completed, transplantation and development are facilitated, and the method is more universal.

Description

Technical Field

[0001] This invention belongs to the field of radar signal processing technology, specifically relating to a GPU-based method for fast localization of wide-swath SAR images. Background Technology

[0002] With the increasing maturity of synthetic aperture radar (SAR) technology, it has high practical value and wide application scenarios in both defense and civilian fields. Due to the influence of side-looking and terrain variations on the geometry of SAR imaging, SAR images exhibit greater geometric distortion compared to optical images. Therefore, SAR images require geometric correction or geocoding before application, all of which are based on high-precision SAR positioning. However, SAR imaging is developing towards ultra-high resolution and ultra-wide imaging swaths from high-orbit (GEO) satellites. Ultra-high resolution imaging and ultra-wide imaging swaths from GEO satellites mean processing massive amounts of data and performing enormous computations; positioning processing of wide-swath SAR images is an extremely time-consuming and labor-intensive task.

[0003] Multi-core CPU and many-core GPU parallel computing is a popular high-performance computing architecture. CPUs are suitable for general-purpose computing such as serial instructions, task scheduling, and program control; GPUs are suitable for high-throughput computing and can handle large-scale data processing. GPUs offer high performance, small size, low power consumption, and a versatile hardware and software architecture, providing high-performance computing capabilities in constrained environments. They are widely used in high-performance data processing and applications in surveying, remote sensing, and geosciences. For complex remote sensing image processing algorithms, by deeply analyzing each step, the algorithm is broken down. The computationally intensive parts are handled by the GPU, while the remaining operations with more logic processing and less computation are handled by the CPU. This CPU-GPU heterogeneous parallel computing model based on algorithm decomposition has become the current means of achieving GPU parallel acceleration for remote sensing image processing. Although GPUs have advantages in existing parallel algorithms for SAR image geometric localization, current parallel algorithms for geometric localization still have shortcomings. They cannot be designed based on the many-core characteristics of GPUs and cannot fully leverage the parallel advantages of GPUs.

[0004] However, current parallel algorithms for geometric positioning have high requirements for single-machine processor resources, especially GPU memory. Moreover, existing algorithms have not been specifically designed according to the characteristics of GPUs, and cannot fully leverage the advantages of GPU high parallelism. Often, the computationally intensive parts of the algorithm cannot achieve full parallelism, resulting in long processing times. The GPU parallel algorithms used in existing geometric positioning can only be applied to single requirements and cannot be ported or extended. For SAR images obtained from different imaging modes, the algorithm needs to be rewritten, resulting in long development cycles. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a GPU-based method for fast localization of wide-swath SAR images. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] This invention provides a GPU-based method for fast localization of wide-swath SAR images, the method comprising:

[0007] The CPU is used to preprocess the first-level SAR image to be located, generating the corresponding parameters.

[0008] The CPU is used to extract metadata information from the first-level SAR image, and a blank latitude and longitude grid is established based on the metadata information and parameters;

[0009] The CPU sends the metadata information and the blank latitude and longitude grid to the GPU, and controls the GPU to process the metadata information accordingly based on the GPU's memory resources.

[0010] Using the GPU, select several pixels from the metadata information of the processed first-level SAR image. Based on the preset coordinate system transformation formula, transform the selected pixels from the original coordinate system to the latitude and longitude coordinate system to obtain the first observation matrix.

[0011] Using the GPU based on a preset coordinate system transformation inverse formula, a number of selected pixels are transformed from the latitude and longitude coordinate system to the pixel coordinate system of the secondary correction image to obtain the second observation matrix;

[0012] Using the GPU, a fitting mapping relationship between the first-level SAR image and the second-level corrected image is obtained based on the first and second observation matrices.

[0013] By utilizing the GPU based on the fitted mapping relationship, the blank latitude and longitude grid, and the GPU's processing method for the metadata information, and employing the corresponding interpolation processing method, the final positioning result is obtained.

[0014] In one embodiment of the present invention, the CPU is used to preprocess the first-level SAR image to be located to generate corresponding parameters, including:

[0015] The CPU is used to process the first-level SAR image to be located using a streaming method to generate the corresponding parameters.

[0016] In one embodiment of the present invention, the parameters include:

[0017] The number of azimuth points (Nan), the number of range points (Nrn), the resolution (Rou), the Doppler parameter (fdc), the elevation (H), and the corner coordinates corresponding to each pixel in the first-level SAR image.

[0018] In one embodiment of the present invention, the metadata information includes:

[0019] Image time information, geographic location, and resolution.

[0020] In one embodiment of the present invention, establishing a blank latitude and longitude grid based on the metadata information and parameters includes:

[0021] Based on the metadata information and parameters, the geographical area covered by the image is obtained;

[0022] The boundary range of the blank latitude and longitude grid to be established is obtained based on the geographical region;

[0023] A blank latitude and longitude grid is established based on the boundary range of the blank latitude and longitude grid to be established.

[0024] In one embodiment of the present invention, the GPU is controlled to perform corresponding processing on the metadata information based on the GPU's memory resources, including:

[0025] The CPU is used to determine whether the GPU's memory resources meet preset conditions. If so, the GPU is controlled to perform image block processing on the metadata information; otherwise, the GPU is controlled to perform normal processing on the metadata information.

[0026] In one embodiment of the present invention, the GPU is used to obtain the fitting mapping relationship between the first-level SAR image and the second-level corrected image based on the first observation matrix and the second observation matrix, including:

[0027] Based on the first observation matrix and the second observation matrix, a polynomial matrix is ​​obtained;

[0028] Based on the polynomial matrix and the coordinates of selected pixels in the secondary correction image, a mapping coefficient vector is obtained;

[0029] Based on the mapping coefficient vector, the fitted mapping relationship is obtained.

[0030] In one embodiment of the present invention, a mapping coefficient vector is obtained based on the polynomial matrix and the coordinates of a selected set of pixels mapped to the secondary correction image, including:

[0031] The pseudo-inverse of the polynomial matrix is ​​multiplied by the coordinates of a selected number of pixels in the secondary correction image to obtain the mapping coefficient vector.

[0032] In one embodiment of the present invention, the final positioning result is obtained by utilizing the GPU based on the fitted mapping relationship, the blank latitude and longitude grid, and the GPU's processing method for the metadata information, and by employing a corresponding interpolation method, including:

[0033] If the GPU processes the metadata information in a conventional manner, the GPU uses a two-dimensional interpolation method based on the fitted mapping relationship to interpolate the pixels in the first-level SAR image to the corresponding positions in the second-level corrected image to obtain the first interpolation result. The first interpolation result is then written into the blank latitude and longitude grid to obtain the final positioning result.

[0034] If the GPU uses image block processing for the metadata information, then based on the fitted mapping relationship, the GPU uses a multi-dimensional parallel interpolation method to interpolate the pixels in each block of the first-level SAR image to the corresponding positions in the second-level correction image to obtain the corresponding second interpolation results. The second interpolation results are written into the blank latitude and longitude grid and merged to obtain the final positioning result.

[0035] The beneficial effects of this invention are:

[0036] The solution provided by this invention adopts a heterogeneous parallel computing architecture of CPU and GPU. By leveraging the many-core characteristics of GPU, it achieves higher efficiency and significantly reduced time consumption in the rapid localization process of wide-swath SAR images, reaching the minute or even second level. It fully utilizes the parallel advantages of GPU to parallelize the generation of multiple parameters, thereby improving the parallelism and greatly enhancing the running efficiency. Considering the cumbersome characteristics of porting and developing SAR images obtained from multi-mode imaging in different coordinate systems, the relationship between the first-level SAR image and the second-level correction image is directly established by fitting. Only the coordinate transformation relationship needs to be updated to complete the processing under different coordinate systems, thereby completing the localization processing of SAR images obtained from multi-mode imaging. This facilitates porting and development and has greater universality. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the CPU and GPU heterogeneous parallel computing architecture of a GPU-based wide-swath SAR image fast localization method provided in an embodiment of the present invention.

[0038] Figure 2 A schematic diagram illustrating the steps of a GPU-based wide-swath SAR image fast localization method provided in an embodiment of the present invention;

[0039] Figure 3 This is a flowchart illustrating a GPU-based fast localization method for wide-swath SAR images, as provided in an embodiment of the present invention. Detailed Implementation

[0040] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0041] This invention provides a method for fast localization of wide-swath SAR images based on GPU, employing a heterogeneous parallel computing architecture of CPU and GPU. The schematic diagram of this CPU-GPU heterogeneous parallel computing architecture can be found in [link to schematic diagram]. Figure 1 ,from Figure 1 As can be seen from the upper part of the diagram, in the heterogeneous parallel computing architecture of CPU and GPU, the CPU architecture is as follows: Figure 1 The upper left section, as shown, may include: a controller (Control), several ALUs (Automatic Logic Units), a cache, and dynamic random access memory (DRAM). The GPU architecture is as follows... Figure 1 The upper right portion, as shown, may include: Dynamic Random Access Memory (DRAM) and several network structures; wherein, each network structure may include: a controller (Control), a cache memory (Cache), and multiple analog logic units (ALUs). Figure 1 As can be seen from the lower half of the diagram, Host can represent the CPU and Device can represent the GPU. During the processing, each kernel in the CPU corresponds to a set of network structures, Grid, thereby realizing a heterogeneous parallel computing architecture between the CPU and GPU.

[0042] It is understood that the embodiments of the present invention adopt a heterogeneous parallel computing architecture of CPU and GPU, which can significantly improve the efficiency and accuracy of SAR image data positioning and processing under the condition of limited hardware resources, overcome the shortcomings of the prior art, and can be widely used.

[0043] Please refer to the schematic diagram of the steps of a GPU-based wide-swath SAR image fast localization method provided in this embodiment of the invention. Figure 2 Please refer to the flowchart. Figure 3 The following will be based on Figure 2 and Figure 3 This paper provides a detailed introduction to each step of the GPU-based method for fast localization of wide-swath SAR images.

[0044] A GPU-based method for fast localization of wide-swath SAR images, such as Figure 2 As shown, it may include:

[0045] S1: The CPU is used to preprocess the first-level SAR image to be located and generate the corresponding parameters.

[0046] Specifically, S1 may include:

[0047] Please see Figure 3As shown, the CPU can be used to process the first-level SAR image to be located using a streaming method to generate the corresponding parameters.

[0048] Parameters may include:

[0049] The number of azimuth points (Nan), the number of range points (Nrn), the resolution (Rou), the Doppler parameter (fdc), the elevation (H), and the corner coordinates corresponding to each pixel in the first-level SAR image.

[0050] Understandably, the corner coordinates corresponding to each pixel in a SAR image are determined using appropriate algorithms and techniques. These algorithms and techniques aim to extract corner information from the SAR image and thus determine the corner coordinates corresponding to each pixel. Specifically, this may include the following steps:

[0051] Key points in SAR images are identified through feature extraction algorithms. These key points can be points in the image that change drastically, such as edges and corners.

[0052] After identifying key points, it is necessary to further determine whether these points are corner points. Corner detection technology can be used to determine the corner points in the image.

[0053] The corner coordinates of each corner point are obtained from the determined corner points in the image. These corner coordinates represent the positions of each corner point in the image and are crucial for subsequent processing such as image registration and target recognition.

[0054] By employing stream processing to increase parallelism and reduce processing time, the CPU is first used to handle logically complex but computationally low parameter generation tasks. This preparatory work can be completed before the actual processing of location data. The CPU, due to its complex logic, is well-suited for such tasks. This approach ensures that all preparations are complete before the data stream begins processing, resulting in a smoother and more efficient integrated image processing workflow. This ensures both rapid system response and improved data processing efficiency.

[0055] S2 utilizes the CPU to extract metadata information from the first-level SAR image, and establishes a blank latitude and longitude grid based on the metadata information and parameters.

[0056] Metadata information for Level 1 SAR images may include:

[0057] Image time information, geographic location, and resolution.

[0058] Creating a blank latitude and longitude grid based on metadata information and parameters can include:

[0059] Based on metadata information and parameters, the geographical area covered by the image is obtained;

[0060] The boundary range of the blank latitude and longitude grid to be established is obtained based on the geographical region;

[0061] A blank latitude and longitude grid is established based on the boundary range of the blank latitude and longitude grid to be established.

[0062] Specifically, when the CPU receives a Level 1 SAR image requiring location processing, it first needs to extract the image's metadata. This metadata includes crucial information such as the image's time, geographic location, and resolution, which is essential for subsequent image processing and analysis. Through in-depth analysis of the metadata and parameters obtained from S1, the geographic area covered by the image is determined, thus defining the boundary of the blank latitude and longitude grid. This blank grid covers the entire image area and provides a precise geographic location for each pixel. This allows the image pixels to correspond to actual geographic coordinates, providing a solid foundation for subsequent image analysis and processing. When creating the blank latitude and longitude grid, the Earth's curvature and the image's projection method must also be considered, as these factors affect the accuracy of the grid's creation and location. By extracting and calculating the metadata and parameters of the Level 1 SAR image, an accurate blank latitude and longitude grid is established, providing a vital foundation for image location and processing.

[0063] S3 uses the CPU to send metadata information and blank latitude and longitude grid to the GPU, and controls the GPU to process the metadata information accordingly based on the GPU's memory resources.

[0064] Specifically, based on the GPU's memory resources, controlling the GPU to process metadata information accordingly can include:

[0065] The CPU is used to determine whether the GPU's memory resources meet the preset conditions. If so, the GPU is controlled to perform image block processing on the metadata information; otherwise, the GPU is controlled to perform normal processing on the metadata information.

[0066] The CPU can determine whether image segmentation is necessary based on the GPU's memory resources. A heterogeneous parallel computing architecture between CPU and GPU can be divided into two parts: CPU and GPU. During processing, the critical parallel computation is performed by the GPU. Compared to CPU memory, GPU memory is more efficient and often less, therefore, image segmentation needs to be considered.

[0067] It is understandable that the processing method selected in step S3 corresponds to the processing method in the subsequent step S7. For details, please refer to the description in the corresponding position below.

[0068] S4. Using the GPU, select several pixels from the metadata information of the processed first-level SAR image. Based on the preset coordinate system transformation formula, transform the selected pixels from the original coordinate system to the latitude and longitude coordinate system to obtain the first observation matrix.

[0069] The positioning model selected in this embodiment of the invention can be a rigorous range-Doppler model. It is understood that the wide-swath SAR image rapid positioning method proposed in this embodiment obtains the precise positioning information of the sampling points through this model, and uses polynomial fitting to reduce the computational load, thereby obtaining the positioning position of the entire matrix. Specifically, within the radar illumination area, points with Doppler frequency shifts caused by relative motion form a hyperbolic bundle along the azimuth direction at the nadir point, while points distributed at equal distances from the target to the satellite form concentric circular bundles centered at the nadir point. Target points located on the same circle will have different Doppler frequency shifts, and target points located on the same hyperbola will have different time delays. Therefore, by determining the range based on the echo signal time delay and the angle with the azimuth direction based on the Doppler frequency shift, the point target can be located at the intersection of a circle and a hyperboloid. This is the so-called range-Doppler positioning principle, or RD positioning principle for short.

[0070] Geometric positioning is performed using the distance-Doppler model positioning method.

[0071] Let (x, y) be any point on the slope distance graph, satisfying the following condition:

[0072] t = y / PRF;

[0073] R = R0 + xM;

[0074] Where t represents time, y represents the row number of the point in the slant range image, PRF represents the pulse repetition frequency of SAR, R represents the slant range from the ground point corresponding to the affected point to the satellite, R0 represents the perigee slant range, x represents the column number of the point in the slant range image, and M represents the slant range resolution of the image.

[0075] As can be seen from the above formula, the spatial position vector and velocity vector can be determined according to the row number y, and the slant distance from the satellite to the ground object can be determined according to the column number x.

[0076] For any point P on the ground, the slant distance condition and the Doppler condition must be satisfied:

[0077] R(t) = |(S(t) - P(t))|;

[0078]

[0079] Where S(t) represents the satellite's position vector at time t, and V SV(t) represents the satellite's velocity vector at time t, λ represents the radar wavelength, P(t) represents the position vector of ground point P at time t, and V P (t) represents the velocity vector of ground point P at time t, f D (t) represents the Doppler frequency shift.

[0080] The ground coordinates of ground point P should satisfy the equation of the ellipsoid of revolution, which is as follows:

[0081]

[0082] Among them, R e Let f represent the major radius of the Earth's ellipsoid, f represent the Earth's oblateness, H represent the geodetic height of point P on the ground, and X represent the minor radius of the ellipsoid. P Y P and Z P This represents the ground coordinates of point P. It's understandable that geodetic height is the distance from a ground point along the normal to the reference ellipsoid to the reference ellipsoid surface.

[0083] The above formula can be used to obtain the one-to-one correspondence between the coordinates of the rows and columns of the SAR image and the coordinates of the ground, thus obtaining the preset coordinate system transformation formula.

[0084] Specifically, the preset coordinate system transformation forward transformation formula can be obtained based on the characteristics of the remote sensing image and the selected correction model. It represents a forward mapping relationship from the distorted image to the target image. In essence, the distorted image is the second-level corrected image obtained from the input first-level SAR image through geometric correction. The preset coordinate system transformation forward transformation formula is as follows:

[0085] (b,l)=f(x1,y1);

[0086] Where b and l represent latitude and longitude, and x1, y1 represent the index of the pixel in the first-level SAR image.

[0087] Based on the preset coordinate system transformation formula, a number of selected pixels are transformed from the original coordinate system to the latitude and longitude coordinate system to obtain the first observation matrix.

[0088] The first observation matrix can represent the pixel location coordinate matrix of a first-level SAR image.

[0089] Understandably, from a geometric positioning perspective, a first-level SAR image is a collection of individual pixels. The correlation between each pixel is weak, allowing for parallel processing. Therefore, the key to geometric positioning algorithms lies in achieving independent positioning of each pixel, and parallel processing can significantly improve processing speed.

[0090] S5 utilizes the GPU based on a preset coordinate system transformation inverse transformation formula to transform a number of selected pixels from the latitude and longitude coordinate system to the pixel coordinate system of the secondary correction image, thus obtaining the second observation matrix.

[0091] Specifically, in order to reverse-map each output pixel to the input image space, a preset inverse coordinate system transformation formula needs to be derived to obtain the inverse mapping relationship from the output image to the distorted image. The preset inverse coordinate system transformation formula can be as follows:

[0092] (x2,y2)=g(b,l);

[0093] Where x2, y2 represent the indices of the pixels in the secondary corrected image.

[0094] The second observation matrix can represent the position coordinate matrix of the corresponding pixel in the secondary correction image.

[0095] S6 utilizes the GPU to obtain the fitting mapping relationship between the first-level SAR image and the second-level corrected image based on the first and second observation matrices.

[0096] Specifically, for S6, it can include:

[0097] S61, based on the first observation matrix and the second observation matrix, yields a polynomial matrix, which may include:

[0098] Based on the first and second observation matrices, the selected pixels in the first-level SAR image are normalized to obtain normalized coordinates, and a polynomial matrix is ​​constructed based on the normalized coordinates.

[0099] S62, based on the polynomial matrix and the coordinates of selected pixels mapped to the secondary corrected image, obtains the mapping coefficient vector, which may include:

[0100] The pseudo-inverse of the polynomial matrix is ​​multiplied by the coordinates of a selected number of pixels in the secondary-corrected image to obtain the mapping coefficient vector.

[0101] S63, based on the mapping coefficient vector, obtains the fitted mapping relationship.

[0102] Using the GPU, a polynomial matrix is ​​obtained based on the first and second observation matrices; based on the polynomial matrix and the coordinates of the second-level corrected image, a mapping coefficient vector is obtained; based on the mapping coefficient vector, a fitted mapping relationship is obtained.

[0103] The pixel position coordinate matrix of the first-level image and the position coordinate matrix of the second-level image are obtained by calculating using the forward transformation formula and the inverse transformation formula.

[0104] Understandably, in step S6, a polynomial matrix is ​​used to describe the geometric position transformation relationship between pixels in the secondary corrected image and pixels in the primary SAR image. Specifically, the pixel position coordinate matrix in the primary SAR image and the pixel position coordinate matrix in the secondary corrected image can be calculated using the forward and inverse transformation formulas, and the corresponding geometric position transformation relationship is obtained as a fitting mapping relationship through fitting.

[0105] The specific geometric position transformation relationship is expressed as follows:

[0106]

[0107] Where x′ represents the x-coordinate of the pixel's location in the secondary SAR image, y′ represents the y-coordinate of the pixel's location in the secondary SAR image, and a ij Represents the coefficient vector, u i v represents the x-coordinate of a pixel's location in a primary image. i b represents the ordinate of the pixel location in the first-level image. ij This represents the coefficient vector, and n represents the number of reference points taken out.

[0108] In this embodiment of the invention, after directly establishing the relationship between the first-level SAR image and the second-level corrected image through fitting, only the coordinate transformation relationship needs to be updated to complete the processing under different coordinate systems, thereby completing the positioning processing of SAR images for multi-mode imaging, which is convenient for porting and development and has greater universality.

[0109] S7 utilizes the GPU's methods for processing metadata information based on fitted mapping relationships, blank latitude and longitude grids, and appropriate interpolation techniques to obtain the final positioning result, which may include:

[0110] If the GPU uses conventional processing for metadata information, it uses a two-dimensional interpolation method based on the fitted mapping relationship to interpolate the pixels in the first-level SAR image to the corresponding positions in the second-level corrected image to obtain the first interpolation result. The first interpolation result is then written into a blank latitude and longitude grid to obtain the final positioning result.

[0111] If the GPU uses image block processing for metadata information, then based on the fitting mapping relationship, the GPU uses multi-dimensional parallel interpolation to interpolate the pixels in each block of the first-level SAR image to the corresponding positions in the second-level correction image to obtain the corresponding second interpolation results. The second interpolation results are written into the blank latitude and longitude grid, merged, and the final positioning result is obtained.

[0112] When processing image data, GPUs can allocate data to thread blocks to achieve parallel computing. By distributing image data and convolution kernel data to different thread blocks and threads, the GPU's parallel computing capabilities can be utilized to achieve fast image processing. This allocation method is designed to maximize the use of the GPU's parallel computing power and accelerate the image processing process.

[0113] Specifically, based on the fitted mapping relationship obtained in step S6, a two-dimensional interpolation method is used to interpolate the pixels in the first-level SAR image to their corresponding positions in the second-level corrected image. This process ensures that pixel information is effectively preserved and accurately mapped during the conversion from first-level to second-level, thereby improving image quality and accuracy. During interpolation, a multi-dimensional parallel interpolation method can be used, greatly increasing the parallelism. This not only achieves point-to-point mapping and positioning between pixels in a single image from first-level to second-level, but also allows for parallel processing of multiple data blocks, significantly improving processing speed. Simultaneously, interpolation can be performed using a lookup table. By pre-storing the interpolation calculation results in the corresponding table, the results are retrieved during interpolation processing, avoiding real-time calculations and reducing computational load and processing time, thus improving processing efficiency. It is understood that if block processing is selected in step S3, corresponding merging processing is also required in step S7. The two-dimensional interpolation fitting method used in this embodiment of the invention for pixel mapping between first-level SAR images and second-level corrected images can leverage the characteristics of GPU many-core processors, resulting in higher efficiency and significantly reduced time consumption during rapid positioning of wide-swath SAR images, reaching minute or even second-level efficiency. Meanwhile, the parallel multi-stream processing method adopted uses the non-blocking stream function of the GPU to parallelize the generation of multiple parameters, thereby improving the parallelism and the efficiency of the algorithm.

[0114] The GPU-based fast localization method for wide-swath SAR images provided in this invention employs a heterogeneous parallel computing architecture of CPU and GPU. Leveraging the many-core characteristics of GPUs, it achieves higher efficiency and significantly reduced time consumption in the fast localization process of wide-swath SAR images, reaching minute or even second-level processing. It fully utilizes the parallel advantages of GPUs, parallelizing the generation of multiple parameters, thereby increasing parallelism and greatly improving operational efficiency. Through optimized computation scheduling, this method achieves fast localization even with massive amounts of data. Considering the cumbersome nature of porting and developing SAR images obtained from multi-mode imaging in different coordinate systems, this invention directly establishes the relationship between the first-level SAR image and the second-level correction image through fitting. Only the coordinate transformation relationship needs to be updated to complete the processing in different coordinate systems, thus completing the localization processing of SAR images obtained from multi-mode imaging. When porting and expansion are needed, only the corresponding transformation formula needs to be replaced to complete the corresponding porting and development, making it convenient for porting and development and more universal.

[0115] It should be noted that, in the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A fast localization method for wide-swath SAR images based on GPU, employing a heterogeneous parallel computing architecture of CPU and GPU, characterized in that, include: The CPU is used to preprocess the first-level SAR image to be located, generating the corresponding parameters. The CPU is used to extract metadata information from the first-level SAR image, and a blank latitude and longitude grid is established based on the metadata information and parameters; The CPU sends the metadata information and the blank latitude and longitude grid to the GPU, and controls the GPU to process the metadata information accordingly based on the GPU's memory resources. Using the GPU, select several pixels from the metadata information of the processed first-level SAR image. Based on the preset coordinate system transformation formula, transform the selected pixels from the original coordinate system to the latitude and longitude coordinate system to obtain the first observation matrix. Using the GPU based on a preset coordinate system transformation inverse formula, a number of selected pixels are transformed from the latitude and longitude coordinate system to the pixel coordinate system of the secondary correction image to obtain the second observation matrix; Using the GPU, a fitting mapping relationship between the first-level SAR image and the second-level corrected image is obtained based on the first and second observation matrices. By utilizing the GPU based on the fitted mapping relationship, the blank latitude and longitude grid, and the GPU's processing method for the metadata information, and employing the corresponding interpolation processing method, the final positioning result is obtained.

2. The method for fast localization of wide-swath SAR images based on GPU according to claim 1, characterized in that, The process of using the CPU to preprocess the first-level SAR image to be located and generate corresponding parameters includes: The CPU is used to process the first-level SAR image to be located using a streaming method to generate the corresponding parameters.

3. The method for fast localization of wide-swath SAR images based on GPU according to claim 2, characterized in that, The parameters include: The number of azimuth points (Nan), the number of range points (Nrn), the resolution (Rou), the Doppler parameter (fdc), the elevation (H), and the corner coordinates corresponding to each pixel in the first-level SAR image.

4. The method for fast localization of wide-swath SAR images based on GPU according to claim 1, characterized in that, The metadata information includes: Image time information, geographic location, and resolution.

5. The method for fast localization of wide-swath SAR images based on GPU according to claim 1, characterized in that, A blank latitude and longitude grid is established based on the aforementioned metadata information and parameters, including: Based on the metadata information and parameters, the geographical area covered by the image is obtained; The boundary range of the blank latitude and longitude grid to be established is obtained based on the geographical region; A blank latitude and longitude grid is established based on the boundary range of the blank latitude and longitude grid to be established.

6. The method for fast localization of wide-swath SAR images based on GPU according to claim 1, characterized in that, The step of controlling the GPU to process the metadata information according to the GPU's memory resources includes: The CPU is used to determine whether the GPU's memory resources meet preset conditions. If so, the GPU is controlled to perform image block processing on the metadata information; otherwise, the GPU is controlled to perform normal processing on the metadata information.

7. The method for fast localization of wide-swath SAR images based on GPU according to claim 1, characterized in that, The step of using the GPU to obtain the fitting mapping relationship between the first-level SAR image and the second-level corrected image based on the first and second observation matrices includes: Based on the first observation matrix and the second observation matrix, a polynomial matrix is ​​obtained; Based on the polynomial matrix and the coordinates of selected pixels in the secondary correction image, a mapping coefficient vector is obtained; Based on the mapping coefficient vector, the fitted mapping relationship is obtained.

8. The GPU-based fast localization method for wide-swath SAR images according to claim 7, characterized in that, Based on the polynomial matrix and the coordinates of selected pixels mapped to the secondary corrected image, a mapping coefficient vector is obtained, including: The pseudo-inverse of the polynomial matrix is ​​multiplied by the coordinates of a selected number of pixels in the secondary correction image to obtain the mapping coefficient vector.

9. A GPU-based method for fast localization of wide-swath SAR images according to claim 6, characterized in that, The method of utilizing the GPU based on the fitted mapping relationship, the blank latitude and longitude grid, and the GPU's processing method for the metadata information, and employing corresponding interpolation methods to obtain the final positioning result, includes: If the GPU processes the metadata information in a conventional manner, the GPU uses a two-dimensional interpolation method based on the fitted mapping relationship to interpolate the pixels in the first-level SAR image to the corresponding positions in the second-level corrected image to obtain the first interpolation result. The first interpolation result is then written into the blank latitude and longitude grid to obtain the final positioning result. If the GPU uses image block processing for the metadata information, then based on the fitted mapping relationship, the GPU uses a multi-dimensional parallel interpolation method to interpolate the pixels in each block of the first-level SAR image to the corresponding positions in the second-level correction image to obtain the corresponding second interpolation results. The second interpolation results are written into the blank latitude and longitude grid and merged to obtain the final positioning result.

Images