Gpu-based sub-image back-projection parallel imaging method

Abstract

The application discloses a GPU-based sub-image backward projection parallel imaging method, comprising the following steps: adaptive block processing of space-borne synthetic aperture radar echo data; GPU parallel backward projection processing; backward projection calculation and asynchronous transmission processing of data blocks; and sub-image generation. The method has the advantages of high practicability, can be used for high-orbit and high-resolution space-borne synthetic aperture radar imaging, and overcomes the defects of high calculation complexity of the traditional synthetic aperture radar backward projection algorithm and long processing time, so that the method has a wide market application prospect.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar imaging and mainly relates to a parallel imaging method based on graphics processing unit (GPU) for sub-image back projection (BP). It utilizes the high concurrency threads of the GPU to adaptively segment and process spaceborne synthetic aperture radar echo data and perform parallel processing of sub-image back projection, thereby achieving efficient and accurate SAR imaging. Background Technology

[0002] Back projection imaging algorithms are suitable for complex observation geometries and can achieve high-precision, approximation-free imaging in spaceborne synthetic aperture radar (SAR). As the swath width / resolution ratio of spaceborne SAR increases, the influence of surface curvature becomes significant. Orbital curvature and surface curvature complicate the spatial variation of signals. For such problems, back projection imaging algorithms can be used to obtain high-longitude focused SAR images.

[0003] To address the computational burden of the BP algorithm, many researchers have improved it using hierarchical techniques, decomposing SAR images into multiple sub-aperture images, also known as sub-images. For example, the Fast Factorized BackProjection (FFBP) algorithm synthesizes sub-aperture images in polar coordinates, while the Cartesian Factorized BackProjection (CFBP) algorithm synthesizes them in Cartesian coordinates, avoiding image geometric distortion. However, for spaceborne high-orbit SAR data, these methods still suffer from high computational cost and significant time consumption when calculating each level of sub-aperture image, thus limiting their practical application. Therefore, it is necessary to develop a GPU-based parallel imaging method for sub-aperture image backprojection specifically for spaceborne high-orbit SAR imaging. This method utilizes the high concurrency of GPU threads to efficiently and accurately achieve high-resolution SAR imaging, possessing significant engineering value and application prospects. Summary of the Invention

[0004] To achieve the above technical objectives, the present invention employs the following technical solution: The present invention provides a GPU-based sub-aperture image backprojection parallel imaging method, which includes the following steps:

[0005] (1) Adaptive block processing of spaceborne synthetic aperture radar (SAR) echo data; (2) GPU parallel back projection processing; (3) Back projection calculation and asynchronous data block transmission processing; (4) Sub-aperture image generation.

[0006] In step (1), the amount of spaceborne synthetic aperture radar (SAR) echo data is large and cannot be transferred to the GPU memory all at once. First, the data at each position in a sub-aperture image grid of size K is set to zero and transferred to the GPU memory. Then, the remaining memory in real time is obtained using the graphics card driver interface function, and the echo data is adaptively divided into blocks based on this.

[0007] This technology maximizes video memory utilization, thereby ensuring maximum data throughput for subsequent concurrent processing. Furthermore, it boasts good versatility and portability, unaffected by graphics card type or echo data size.

[0008] In step (2), the block echo data is transmitted to the GPU memory, T GPU threads are started, each thread reads one data in the echo data block, calculates the echo signal intensity at a point on the sub-aperture grid through the back projection algorithm, and adds the echo signal intensity to the existing value at that position.

[0009] In step (3), when the number of data blocks N > 1, while performing the backward projection processing in step (2), the GPU has independent data transmission hardware, which can simultaneously transfer the next block of data from memory to video memory. Since the backward projection calculation and data block transfer are asynchronous at this time, this technique is called asynchronous transmission.

[0010] In step (4), steps (2) and (3) are repeated until all echo data have been back-projected. At this point, the data in the sub-aperture image grid is the final sub-aperture image data.

[0011] The beneficial effects of the present invention are as follows: The present invention provides an accurate and effective high-resolution synthetic aperture radar imaging calculation method. This method is novel and highly practical, and can be used for high-orbit and high-resolution spaceborne synthetic aperture radar imaging. It overcomes the shortcomings of traditional synthetic aperture radar back projection algorithms, which have high computational complexity and long processing time, and has broad market application prospects. Attached Figure Description

[0012] Figure 1 It is a geometric model of a back-projected sub-aperture image.

[0013] Figure 2 This is a flowchart of parallel computation of back-projected sub-aperture images. Detailed Implementation

[0014] The specific details of the present invention will be further described in detail below with reference to the accompanying drawings. See also... Figure 1 and Figure 2 The specific implementation steps of this invention are as follows:

[0015] Step 1: Adaptive block processing of spaceborne synthetic aperture radar (SAR) echo data. First, an imaging grid is arranged, and the data at each position in a sub-aperture image grid of size K is set to zero and transferred to GPU memory. Then, the real-time remaining memory size R is obtained using the graphics card driver interface function. Assuming the total size of the echo data is E, the number of echo data blocks N can be calculated by the following formula:

[0016] N=(E×2+R-1) / R

[0017] The size M1 of the first N-1 echo data blocks is:

[0018] M1=(E×2+N-1) / N

[0019] The size M2 of the Nth echo data block is:

[0020] M2 = E × 2 - (N - 1) × M1.

[0021] Step 2: GPU performs parallel backward projection processing, transferring the block echo data of size M1 to the GPU memory. If it is the last data block, the block echo data of size M2 is transferred to the GPU memory.

[0022] The number of GPU threads T that are enabled is

[0023] T = K × M1

[0024] Each thread reads one piece of data from the data block and calculates the corresponding azimuth slow time t. a Below, the distance R from the radar phase center to the sub-aperture image grid point p is... p (t a )

[0025]

[0026] Among them, (x s ,y s ,z s ) is t a The position coordinates of the radar in the Earth's fixed coordinate system, (x p ,y p ,z p ) is t a The position coordinates of grid point p on the time aperture image in the Earth fixed coordinate system are determined. Then, the echo signal intensity at point p is calculated.

[0027]

[0028] Among them, t r For distance to fast time, t a For azimuth, slow time, K c =4πfc / c Radar carrier center wave number, f c Let be the radar carrier frequency, π be pi, c be the speed of light, j be the imaginary unit, and B be the transmitter bandwidth. The definition of the sinc function is as follows:

[0029]

[0030] Step 3: When the number of data blocks N > 1, while performing the backward projection processing in step 2, the GPU has independent data transmission hardware, which can simultaneously transfer the next raw echo data block from memory to video memory.

[0031] Step 4: Repeat steps 2 and 3 until all echo data have been back-projected. At this point, the data in the sub-aperture image grid is the final sub-aperture image data.

[0032] This invention is not limited to the specific embodiments described above. Any modifications made by those skilled in the art based on the above concept without creative effort are within the scope of protection of this invention.

Claims

1. A GPU-based sub-image backprojection parallel imaging method, characterized in that, Includes the following steps: (1) Adaptive block processing of spaceborne synthetic aperture radar echo data; First, the data at each position in a sub-image grid of size K is set to zero and passed to the GPU memory. Then, the real-time remaining memory size R is obtained using the graphics card driver interface function. Assuming the total size of the echo data is E, the number of echo data blocks N can be calculated by the following formula: N = (E×2+R-1) / R The size M1 of the first N-1 echo data blocks is: M1 = (E×2 + N-1) / N The size M2 of the Nth echo data block is: M2 = E×2-(N-1)×M1; (2) GPU parallel backward projection processing; The echo data block of size M1 is transferred to the GPU memory. If it is the last data block, then the echo data block of size M2 is transferred to the GPU memory. The number of GPU threads T is: T = K×M1 Each thread reads one piece of data from the echo data block, calculates the echo signal strength at a point on the sub-aperture grid using the back projection algorithm, and adds this echo signal strength to the existing value at that location. (3) Backward projection calculation and asynchronous data block transmission processing; When the number of data blocks N>1, while performing the backward projection processing in step (2), the GPU has independent data transmission hardware and simultaneously transfers the next original echo data block from memory to video memory. At this time, the backward projection calculation and the original echo data block are asynchronously transmitted. (4) Sub-image generation.

2. The GPU-based sub-image backprojection parallel imaging method according to claim 1, characterized in that, In step (4), sub-images are generated. Steps (2) and (3) are repeated until all echo data have been back-projected. At this point, the data in the sub-image grid is the final sub-image data.

Images