CN-121995377-A - Rapid positioning method for large-breadth SAR image based on GPU

CN121995377ACN 121995377 ACN121995377 ACN 121995377ACN-121995377-A

Abstract

The invention discloses a rapid positioning method of a large-breadth SAR image based on a GPU, which adopts a heterogeneous parallel computing architecture of a CPU and the GPU, utilizes the characteristic of a multi-core of the GPU, ensures that the efficiency is higher in the rapid positioning process of the large-breadth SAR image, the time consumption is greatly reduced, and can reach the minute level or even the second level, fully exerts the parallel advantage of the GPU, parallelizes the generation of a plurality of parameters, improves the parallelism degree, greatly improves the operation efficiency, considers the characteristic that SAR images obtained by multi-mode imaging are in different coordinate systems, and has complicated transplanting and developing, and the relationship between a primary SAR image and a secondary correction image is directly established by fitting, so that the processing under different coordinate systems can be completed only by updating the coordinate conversion relationship, thereby completing the positioning processing of SAR images obtained by multi-mode imaging, facilitating the transplanting and developing, and having more universality.

Inventors

SUN GUANGCAI
WU YANYANG
YANG YI
ZHANG ZHIFENG
XING MENGDAO

Assignees

西安电子科技大学

Dates

Publication Date: 20260508
Application Date: 20241106

Claims (9)

1. A rapid positioning method of a large-breadth SAR image based on a GPU adopts a CPU and GPU heterogeneous parallel computing architecture, and is characterized by comprising the following steps: preprocessing a primary SAR image to be positioned by using a CPU to generate corresponding parameters; Extracting metadata information of the primary SAR image by using a CPU, and establishing a blank longitude and latitude grid based on the metadata information and parameters; The CPU is utilized to send the metadata information and the blank longitude and latitude grids to the GPU, and the GPU is controlled to correspondingly process the metadata information according to the memory resources of the GPU; Selecting a plurality of pixel points in metadata information of the processed primary SAR image by using a GPU, converting a positive transformation relation based on a preset coordinate system, and converting the selected plurality of pixel points from an original coordinate system to a longitude and latitude coordinate system to obtain a first observation matrix; Converting the inverse transformation relation by using the GPU based on a preset coordinate system, and converting the selected pixel points from the longitude and latitude coordinate system to a pixel coordinate system of a secondary correction image to obtain a second observation matrix; obtaining a fitting mapping relation between the first-level SAR image and the second-level correction image by using the GPU based on the first observation matrix and the second observation matrix; and obtaining a final positioning result by utilizing a GPU based on the fitting mapping relation, the blank longitude and latitude grid and the processing mode of the GPU on the metadata information and utilizing a corresponding interpolation processing method.
2. The GPU-based rapid positioning method for large-breadth SAR images according to claim 1, wherein the preprocessing of the primary SAR image to be positioned by the CPU to generate the corresponding parameters comprises: And processing the primary SAR image to be positioned by using a CPU in a stream processing mode to generate corresponding parameters.
3. The GPU-based large-breadth SAR image rapid positioning method according to claim 2, wherein said parameters comprise: Azimuth point number Nan, distance point number Nrn, resolution Rou, doppler parameter fdc, elevation H and corner coordinates corresponding to each pixel point in the primary SAR image.
4. The GPU-based large-breadth SAR image rapid positioning method according to claim 1, wherein said metadata information comprises: time information, geographic location, and resolution of the image.
5. The GPU-based large-breadth SAR image rapid positioning method according to claim 1, wherein establishing a blank longitude and latitude grid based on the metadata information and parameters comprises: obtaining a geographic area covered by the image based on the metadata information and the parameters; Obtaining a boundary range of a blank longitude and latitude grid to be established according to the geographic area; and establishing the blank longitude and latitude grid based on the boundary range of the blank longitude and latitude grid to be established.
6. The method for quickly positioning a large-breadth SAR image based on a GPU according to claim 1, wherein said controlling the GPU to perform corresponding processing on the metadata information according to the memory resource of the GPU comprises: And judging whether the memory resource of the GPU meets the preset condition by utilizing the CPU, if so, controlling the GPU to perform image blocking processing on the metadata information, and if not, controlling the GPU to perform conventional processing on the metadata information.
7. The GPU-based rapid positioning method for large-breadth SAR images according to claim 1, wherein the obtaining the fitting mapping relationship between the first-stage SAR image and the second-stage rectification image by using the GPU based on the first observation matrix and the second observation matrix comprises: based on the first observation matrix and the second observation matrix, a polynomial matrix is obtained; Obtaining a mapping coefficient vector based on the polynomial matrix and coordinates in the two-level correction image corresponding to the selected pixel points; And obtaining a fitting mapping relation based on the mapping coefficient vector.
8. The method for quickly positioning a large-breadth SAR image based on a GPU according to claim 7, wherein obtaining a mapping coefficient vector based on the polynomial matrix and coordinates of the selected pixels corresponding to the two-stage correction image comprises: And multiplying the pseudo inverse of the polynomial matrix by coordinates in the second-level correction image corresponding to the selected pixel points to obtain the mapping coefficient vector.
9. The method for quickly positioning a large-breadth SAR image based on a GPU according to claim 6, wherein said obtaining a final positioning result by using a GPU based on the fitting mapping relationship, the blank longitude and latitude grid and the processing mode of the GPU on the metadata information and a corresponding interpolation processing method comprises: If the GPU adopts conventional processing to the metadata information, the GPU interpolates pixel points in the primary SAR image to corresponding positions of the secondary correction image by using a two-dimensional interpolation method based on the fitting mapping relation to obtain a first interpolation result, and the first interpolation result is written into the blank longitude and latitude grid to obtain a final positioning result; If the GPU adopts image blocking processing to the metadata information, the GPU interpolates pixel points in each block of primary SAR image to corresponding positions of the secondary correction image by using a multidimensional parallel interpolation method based on the fitting mapping relation to obtain corresponding second interpolation results, and writes each second interpolation result into the blank longitude and latitude grid to perform merging processing to obtain a final positioning result.

Description

Rapid positioning method for large-breadth SAR image based on GPU Technical Field The invention belongs to the technical field of radar signal processing, and particularly relates to a rapid positioning method of a large-breadth SAR image based on a GPU. Background With the increasing maturity of synthetic aperture radar technology, the method has high practical value and wide application scene in the fields of national defense and civil affairs. Due to the influence of side view and topography variations on SAR (Synthetic Aperture Radar) imaging geometry, there is a greater geometrical distortion of the SAR image compared to the optical image. Therefore, the SAR image needs to be geometrically corrected or geocoded before application, and these processes are all based on SAR positioning with high accuracy. However, SAR imaging is evolving towards ultra-high resolution and ultra-large imaging breadth for high orbit satellites (GEO). The ultra-high resolution imaging and the ultra-large imaging breadth of the high orbit mean that huge data and huge calculation are required to be processed, and the positioning processing of the large-breadth SAR image is a very time-consuming and labor-consuming work. The multi-core CPU and multi-core GPU parallel computing is a popular high-performance computing architecture at present, wherein the CPU is suitable for general computing such as serial instruction, task scheduling, program control and the like, and the GPU is suitable for large-throughput computing, so that large-scale data volume computing can be realized. The GPU has high computing performance, small volume, low power consumption and general software and hardware architecture, can provide high-performance computing capability in a limited environment, and is widely applied to high-performance data processing and application in the fields of mapping, remote sensing, geography and the like. For a remote sensing image processing algorithm with higher complexity, the algorithm is split by carrying out deep analysis on each step in the algorithm, wherein the part with intensive computation is processed by the GPU, and the rest of operations with more logic processing and smaller computation are processed by the CPU. Although the GPU has advantages in the aspect of the existing SAR image geometric positioning parallel algorithm, the existing parallel algorithm for processing geometric positioning still has the defects that the design cannot be carried out according to the many-core characteristics of the GPU, and the parallel advantages of the GPU cannot be fully exerted. However, the current geometric positioning parallel algorithm has high requirements on the resources of a single processor, particularly has high requirements on the GPU, the existing algorithm cannot be specially designed according to the characteristics of the GPU, the advantage of high parallelism of the GPU cannot be fully exerted, the full parallelism cannot be realized in the part with large calculation amount of the algorithm, the time consumption is long, the existing geometric positioning adopted GPU parallel algorithm can only be applied to single requirements, the transplanting and expansion cannot be realized, and the SAR images obtained according to imaging modes of different modes are required to be realized by programming programs again, so that the development period is long. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a rapid positioning method of a large-breadth SAR image based on a GPU. The technical problems to be solved by the invention are realized by the following technical scheme: The invention provides a rapid positioning method of a large-breadth SAR image based on a GPU, which comprises the following steps: preprocessing a primary SAR image to be positioned by using a CPU to generate corresponding parameters; Extracting metadata information of the primary SAR image by using a CPU, and establishing a blank longitude and latitude grid based on the metadata information and parameters; The CPU is utilized to send the metadata information and the blank longitude and latitude grids to the GPU, and the GPU is controlled to correspondingly process the metadata information according to the memory resources of the GPU; Selecting a plurality of pixel points in metadata information of the processed primary SAR image by using a GPU, converting a positive transformation relation based on a preset coordinate system, and converting the selected plurality of pixel points from an original coordinate system to a longitude and latitude coordinate system to obtain a first observation matrix; Converting the inverse transformation relation by using the GPU based on a preset coordinate system, and converting the selected pixel points from the longitude and latitude coordinate system to a pixel coordinate system of a secondary correction image to obtain a second observation matrix; obtaining a