[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN108919220B - Missile-borne SAR front-side-view imaging method based on embedded GPU - Google Patents

Missile-borne SAR front-side-view imaging method based on embedded GPU Download PDF

Info

Publication number
CN108919220B
CN108919220B CN201810737271.3A CN201810737271A CN108919220B CN 108919220 B CN108919220 B CN 108919220B CN 201810737271 A CN201810737271 A CN 201810737271A CN 108919220 B CN108919220 B CN 108919220B
Authority
CN
China
Prior art keywords
matrix
echo signal
distance
gpu
signal matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810737271.3A
Other languages
Chinese (zh)
Other versions
CN108919220A (en
Inventor
刘峥
宋超
宋凤博
冉磊
谢荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201810737271.3A priority Critical patent/CN108919220B/en
Publication of CN108919220A publication Critical patent/CN108919220A/en
Application granted granted Critical
Publication of CN108919220B publication Critical patent/CN108919220B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention provides a missile-borne SAR front-side-looking imaging method based on an embedded GPU, which solves the technical problem of poor real-time performance of the existing missile-borne SAR imaging method and comprises the following implementation steps: in an FPGA of a signal processor, performing distance dimension pulse compression and distance dimension interception on an original echo sampling matrix of the missile-borne SAR image; the intercepted echo signal matrix and the SAR parameters are stored in duplicate by using a CPU and a GPU; the following processing is respectively carried out in the embedded GPU: performing inertial navigation motion compensation on the distance dimension echo signal matrix; performing distance dimension processing on the inertial navigation motion compensation echo signal matrix; processing the echo signal matrix in a distance dimension to perform orientation dimension processing; and processing the echo signal matrix in the azimuth dimension for geometric correction, and finally obtaining a high-resolution missile-borne SAR image. The invention improves the imaging parallelism, reduces the imaging computation amount and the computation time, improves the real-time property, is applied to the missile seeker, and is beneficial to the accurate identification and striking of the missile to the target.

Description

Missile-borne SAR front-side-view imaging method based on embedded GPU
Technical Field
The invention belongs to the technical field of digital signal processing, relates to missile-borne SAR front-side-view imaging, and particularly relates to a missile-borne SAR front-side-view imaging method based on an embedded GPU, which can be used for real-time imaging processing of an active radar seeker for imaging and searching.
Background
Synthetic Aperture Radar (SAR) is the main body of Radar imaging, and has the widest application range. The synthetic aperture radar imaging technology obtains an SAR image with two-dimensional high resolution ratio of distance and direction by carrying out two-dimensional processing of distance and direction on a radar echo signal matrix, clearly shows the shape and fine structure characteristics of a target, and greatly improves the detection and identification capability of the target.
With the development of synthetic aperture radar imaging technology, the application range of the synthetic aperture radar imaging technology is further expanded. In recent years, the missile-borne SAR technology combining the synthetic aperture radar technology and the precision guidance technology has great application value. By exerting the advantages of the radar imaging technology, the missile can accurately track and strike the target, and the method has great significance for national defense safety and military affairs. However, the missile-borne SAR inherits the disadvantage of huge computation while exerting the advantages of the synthetic aperture radar imaging technology, so that the real-time performance is difficult to guarantee, and the application of the missile-borne SAR is limited. In practical application, because the maneuvering time of the missile during turning is ensured, the missile-borne SAR needs to work in a front side view mode, compared with a traditional front side view mode, the calculation amount in the mode is much larger, and the application of the missile-borne SAR is further limited.
In order to solve the problems, the current research situation is that on one hand, the imaging quality is sacrificed to obtain real-time performance by simplifying an SAR imaging algorithm, but the SAR image obtained by the method has poor quality, influences the identification of a hit target and is often irreparable; on the other hand, an advanced high-performance hardware platform is adopted, and imaging quality and real-time performance are considered. Document 201410719487.9 discloses a missile-borne SAR front-side-view imaging method based on a GPU, which is used for realizing a missile-borne SAR imaging algorithm on the GPU and greatly improving the real-time performance of the missile-borne SAR imaging algorithm by taking the advantages of a CUDA multi-core framework of the GPU into account. In addition, the missile-borne SAR front-side-looking parallel imaging processing is mostly based on an FPGA-multi-core DSP hardware platform, and the hardware platform makes great progress in real time by exerting the multi-core advantage of the DSP, so that the missile-borne SAR is applied. However, such hardware platforms also have many problems, such as very complex system structure, large power consumption, difficult heat dissipation, high hardware cost, long development period, and the like.
The method of adopting the PC GPU board card has good real-time performance, but cannot be used on the missile; the method adopting the FPGA and the multi-core DSP is applied to the missile, but the system structure is complex, and the software and hardware are difficult to realize, so that the possibility of exploring the method for realizing the missile-borne SAR imaging based on other high-performance hardware platforms has great significance.
Disclosure of Invention
The invention aims to provide a missile-borne SAR front-side-looking imaging method based on an embedded GPU, which greatly improves the real-time performance, aiming at overcoming the defects in the prior art and solving the technical problem of poor real-time performance of the imaging method in the existing missile-borne SAR imaging platform.
The invention relates to a missile-borne SAR front-side imaging method based on an embedded GPU, which is characterized by comprising the following steps:
(1) in a front-end FPGA of a signal processor, performing distance dimension pulse compression and distance dimension interception on an original echo sampling matrix of the missile-borne SAR image:
sampling matrix S for original echo in front-end FPGA of signal processorM×4NPerforming distance dimension pulse compression to obtain an echo signal matrix after distance pulse compression
Figure BDA0001722348680000021
Wherein, M is the number of pulses, namely the number of azimuth sampling points, 4N is the number of sampling points of each pulse, namely the number of distance sampling points, and the superscript P represents the distance pulse pressure; cut off range pulse pressure echo signal matrix
Figure BDA0001722348680000022
Before each row in the
Figure BDA0001722348680000023
Point and back
Figure BDA0001722348680000024
Obtaining the data of points to obtain a distance dimension intercepted echo signal matrix SM×N；
(2) Intercepting echo signal matrix S from distance dimensionM×NAnd double preservation of the synthetic aperture radar parameters:
setting synthetic aperture radar parameters in a CPU at the rear end of the signal processor, and copying the parameters to an embedded GPU memory at the rear end of the signal processor; reading distance dimension intercepting echo signal matrix S in CPUM×NThe CPU memory is accessed and copied to the embedded GPU memory;
(3) distance dimension interception echo signal matrix S by using missile inertial navigation data and radar parametersM×NMotion compensation is performed in the embedded GPU:
(3a) obtaining inertial navigation data matrix I of north, sky and east dimensions from missile inertial navigation system3×LWherein L is the length of the inertial navigation data of a single dimension, and at the end of the CPU, the inertial navigation data matrixes I are respectively aligned3×LIs subjected to cubic spline interpolation, willInterpolating the data length to M to obtain an interpolated inertial navigation data matrix I3×M；
(3b) In an embedded GPU, utilizing an interpolated inertial navigation data matrix I3×MTo obtain the inertial navigation motion compensation matrix RM×NThen setting M multiplied by N GPU threads and compensating the inertial navigation motion matrix RM×NAnd distance dimension intercepting echo signal matrix SM×NPerforming dot multiplication to obtain an inertial navigation motion compensation echo signal matrix
Figure BDA0001722348680000025
Superscript I represents inertial navigation motion compensation;
(4) echo signal matrix for inertial navigation motion compensation
Figure BDA0001722348680000031
Distance dimension processing is carried out in the embedded GPU to obtain an echo signal matrix after distance dimension processing
Figure BDA0001722348680000032
(5) Echo signal matrix processed by distance dimension
Figure BDA0001722348680000033
Performing orientation dimension processing in an embedded GPU to obtain focused SAR image data;
(6) and geometrically correcting the focused SAR image data in the embedded GPU to obtain a high-resolution missile-borne SAR image.
Compared with the prior art, the invention has the following advantages:
(1) the invention carries out distance dimension interception on the echo signal matrix after the distance pulse pressure, intercepts signals at two sides of the distance dimension of the echo signal matrix during interception, and reserves data of the distance dimension center of the echo signal matrix;
(2) the invention adopts the embedded GPU as the imaging processing core, the number of CUDA operation cores is up to 256, and compared with the existing 8-core parallel processing method adopting the DSP as the processing core, the parallelism of the matrix operation and the vector operation of the imaging processing is greatly improved, thereby greatly reducing the imaging processing time and improving the real-time property.
(3) Compared with the existing C language programming method based on a bare board, the C language programming method based on the embedded operating system is adopted in the implementation of the invention, and more C language operation libraries can be called, so the implementation is easy and the development period is short.
Drawings
FIG. 1 is a diagram of a prior art hardware platform architecture;
FIG. 2 is a flow chart of an implementation of the present invention;
FIG. 3 is a diagram of a hardware platform architecture employed by the present invention;
FIG. 4 is a schematic diagram of a time-domain zero-padding matrix according to the present invention;
FIG. 5 is a schematic diagram of a point-and-multiply GPU thread allocation according to the present invention;
FIG. 6 is a block diagram of the present invention;
FIG. 7 is an SAR image obtained by processing actual measurement data by the MATLAB software and the FPGA-multi-core DSP method of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and the specific embodiments.
Example 1
The existing missile-borne SAR imaging method is based on an FPGA-multi-core DSP platform, DSP is used as a calculation core, and the distance Doppler imaging method for realizing full data processing on a single-chip multi-core DSP comprises the following imaging processing steps: distance pulse pressure, inertial navigation motion compensation, distance dimension processing, direction dimension preprocessing, self-focusing processing, direction post-processing and direction pulse pressure are carried out, and a focused SAR image is finally obtained; the structure diagram of the hardware platform adopted by the method is shown in figure 1. As can be seen from fig. 1, the existing missile-borne SAR imaging method has a very complex platform system structure, which results in large power consumption, difficult heat dissipation and high hardware cost. In addition, the method has low calculation parallelism and complex programming. In view of the current situation, the invention provides a missile-borne SAR front-side-view imaging method based on an embedded GPU through research, and referring to fig. 2, the method comprises the following steps:
(1) in a front-end FPGA of a signal processor, performing distance dimension pulse compression and distance dimension interception on an original echo sampling matrix of the missile-borne SAR image:
in the front-end FPGA of the signal processor, the original echo is sampled by a matrix SM×4NPerforming distance dimension pulse compression to obtain an echo signal matrix after distance pulse compression
Figure BDA0001722348680000041
Wherein, M is the number of pulses, namely the number of azimuth sampling points, 4N is the number of sampling points of each pulse, namely the number of distance sampling points, N is the number of quarter points from the number of sampling points, and the superscript P represents the distance pulse pressure; cut off range pulse pressure echo signal matrix
Figure BDA0001722348680000042
Before each row in the
Figure BDA0001722348680000043
Point and back
Figure BDA0001722348680000044
Obtaining the data of points to obtain a distance dimension intercepted echo signal matrix SM×N. Distance dimension interception, the method adopted by the invention is that when the FPGA transmits data to the CPU, the data is directly transmitted from the third line of each line
Figure BDA0001722348680000045
Point starts, and N points of data are transmitted to the CPU.
(2) Intercepting echo signal matrix S from distance dimensionM×NAnd double preservation of the synthetic aperture radar parameters:
setting and initializing synthetic aperture radar parameters in a CPU (central processing unit) at the rear end of the signal processor, and copying the synthetic aperture radar parameters into an embedded GPU (graphics processing unit) memory at the rear end of the signal processor; reading distance dimension intercepting echo signal matrix S in CPUM×NAnd (4) sending the data to a CPU memory and copying the data to an embedded GPU memory. The copying of the invention adopts a mode of copying a data pointer, and the copying result is that a CPU and a GPU respectively hold a pointerAnd the data pointers of the echo signal matrix in the memory are completely the same as the data stored in the CPU and the GPU.
(3) Distance dimension interception echo signal matrix S by using missile inertial navigation data and radar parametersM×NMotion compensation is performed in the embedded GPU:
(3a) obtaining inertial navigation data matrix I of north, sky and east dimensions from missile inertial navigation system3×LWherein L is the length of the inertial navigation data of a single dimension, and at the end of the CPU, the inertial navigation data matrixes I are respectively aligned3×LEach row of the inertial navigation data matrix is subjected to cubic spline interpolation, the data length is interpolated to M, and an interpolated inertial navigation data matrix I is obtained3×M. The invention carries out inertial navigation data interpolation preprocessing in the CPU, and the inertial navigation data interpolation preprocessing is directly obtained through calculation without adding new components.
(3b) In an embedded GPU, utilizing an interpolated inertial navigation data matrix I3×MTo obtain the inertial navigation motion compensation matrix RM×NThen setting M multiplied by N GPU threads and compensating the inertial navigation motion matrix RM×NAnd distance dimension intercepting echo signal matrix SM×NPerforming dot multiplication to obtain an inertial navigation motion compensation echo signal matrix
Figure BDA0001722348680000051
Superscript I denotes the inertial navigation motion compensation process.
(4) Echo signal matrix for inertial navigation motion compensation
Figure BDA0001722348680000052
Distance dimension processing is carried out in the embedded GPU to obtain an echo signal matrix after distance dimension processing
Figure BDA0001722348680000053
(5) Echo signal matrix processed by distance dimension
Figure BDA0001722348680000054
And performing orientation dimensional processing in the embedded GPU to obtain focused SAR image data.
The distance dimension and orientation dimension imaging processing comprises a large amount of vector and matrix operations, which are the largest part of the whole imaging processing.
(6) And geometrically correcting the focused SAR image data in the embedded GPU to obtain a high-resolution missile-borne SAR image.
The technical idea of the invention is as follows: on the premise of considering application scenes of the missile-borne SAR, intercepting an echo signal matrix after the distance pulse pressure to reduce the data volume so as to reduce the overall operation amount; and (3) carrying out distance and direction two-dimensional SAR imaging processing on the original echo sampling matrix of the missile-borne SAR image by adopting a high-performance embedded GPU platform, and finally obtaining a focused SAR image.
The structure diagram of the high-performance embedded GPU hardware platform adopted by the invention is shown in FIG. 3, and as can be seen from FIG. 3, the signal processor hardware platform adopted by the invention comprises a front-end analog-to-digital converter (ADC), a front-end FPGA and a rear-end CPU/GPU chip, and a CPU core and a GPU core are integrated on one chip, so that the hardware platform is simplified, the computing capacity is expanded, and the signal processor hardware platform is simple in structure and easy to realize.
Example 2
The missile-borne SAR front-side view imaging method based on the embedded GPU is the same as that in the embodiment 1, and the distance dimension processing is carried out in the embedded GPU in the step (4), and the method comprises the following steps:
(4a) evaluating an oblique view angle: echo signal matrix compensated by inertial navigation motion
Figure BDA0001722348680000055
Dividing the direction dimension into Q blocks, and obtaining estimated values fdc of Q local Doppler centers by using Q block matrixesiI 1,2, …, Q, then estimates fdc of the Q local doppler centersiCalculating an average value to obtain an estimated value fdc of the global Doppler center, and further calculating an estimated value theta of the squint angle by using an estimated value fdc of the global Doppler centeresqEvaluating the estimated value theta of the squint angleesqThe formula of (1) is as follows:
θesq＝asin(fdc·λ/v/2)，
where λ represents the wavelength of the transmitted signal and v represents the velocity of the radar motion.
(4b) Distance walk correction: in the embedded GPU, the estimated value fdc of the global Doppler center and the estimated value theta of the squint angle are utilizedesqTo the echo signal matrix after inertial navigation motion compensation
Figure BDA0001722348680000061
Carrying out distance walk correction to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure BDA0001722348680000062
The superscript RF indicates that the signal obtained after the distance walk correction process is a frequency domain signal.
(4c) Range curvature correction and second order range pulse pressure: in an embedded GPU, a range frequency domain echo signal matrix after range walk correction
Figure BDA0001722348680000063
Performing range curvature correction and secondary range pulse pressure to obtain echo signal matrix after range dimension processing
Figure BDA0001722348680000064
Superscript R denotes distance dimension processing.
The global Doppler center is estimated in a data partitioning mode, the data partitioning effectively represents the slow change characteristic of Doppler, and compared with the existing global Doppler center estimation method using full data, the global Doppler center is estimated more accurately.
Example 3
The missile-borne SAR front-side view imaging method based on the embedded GPU is the same as that in the embodiment 1-2, and the azimuth dimension processing in the embedded GPU in the step (5) comprises the following steps:
(5a) and (3) high-order phase processing: in embedded GPU, the distance is alignedEcho signal matrix after off-dimensional processing
Figure BDA0001722348680000065
Performing time domain zero padding to obtain an echo signal time domain zero padding matrix
Figure BDA0001722348680000066
Calculating a higher order phase compensation and higher order phase filter matrix C for each element of the zero-padding matrix2M×NBy using C2M×NTime domain zero-filling matrix for echo signal
Figure BDA0001722348680000067
Performing high-order phase compensation and high-order phase filtering and intercepting to obtain a high-order phase compensation matrix and a high-order phase filtered echo signal matrix
Figure BDA0001722348680000068
The superscript H represents the high-order phase processing, and the time-domain zero-filling matrix
Figure BDA0001722348680000069
As shown in fig. 4, it can be seen from fig. 4 that the zero padding is performed on both sides of the azimuth dimension, the azimuth zero padding corresponds to the up-sampling in the azimuth dimension, and the high-order phase processing is performed after the up-sampling, whereby the accuracy of the high-order phase processing can be improved.
(5b) Azimuth declivity: using radar parameters, radar motion parameters, and estimated squint angle θesqSolving an orientation deskew matrix DM×NUsing M N GPU threads will
Figure BDA00017223486800000610
And DM×NDot multiplication to obtain azimuth deskew echo signal matrix
Figure BDA00017223486800000611
Superscript D represents the azimuth deskew process.
(5c) Motion error estimation and correction: using an iterative method to deskew the echo signal matrix of the azimuth
Figure BDA00017223486800000612
And estimating and correcting the motion error, wherein the CPU performs iterative control, and the embedded GPU performs iterative operation.
(5d) Azimuthal pulse pressure: and in the embedded GPU, performing azimuth pulse pressure on the echo signal matrix after motion error estimation and correction to obtain focused SAR image data.
The distance dimension and orientation dimension processing method follows the distance dimension and orientation dimension processing of a front-side-looking range-Doppler imaging algorithm, the front-side-looking range-Doppler imaging algorithm is a typical frequency domain imaging algorithm, and compared with other time domain and frequency domain algorithms, the front-side-looking range-Doppler imaging algorithm is low in calculation amount and is the most common algorithm for missile-borne SAR. Different from the conventional realization of a front-side view range-Doppler imaging algorithm, the method arranges long-time-consuming large matrix and long-vector operation in the GPU for parallel processing, arranges short-vector and single-point operation in the CPU for processing, and arranges the operation threads in the GPU in parallel and the operation between the CPU and the GPU in parallel.
Example 4
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as that in the embodiment 1-3, and the synthetic aperture radar parameters in the step (2) are except for the distance sampling rate fsPulse repetition frequency PRF, bandwidth B, light speed C, pulse width TpCarrier frequency fcVelocity v of missile, height H of missile, angle of declination theta0Synthetic aperture radar beam center line sweep target slant distance RsThe method also comprises an orientation dimension sampling point number M and a distance dimension sampling point number parameter N.
The orientation dimension sampling point number M and the distance dimension sampling point number parameter N can provide boundary conditions for the execution of GPU threads, prevent random memory access errors of certain GPU threads, and improve the robustness of programs.
Example 5
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as the echo signal matrix obtained in the step (4b) and subjected to inertial navigation motion compensation in the embodiments 1 to 4
Figure BDA0001722348680000071
And (3) carrying out distance walking correction, wherein the implementation steps are as follows:
(4b1) initializing a slow time vector t using M GPU threadsmInitializing distance frequency domain vector f using N GPU threadsrUsing slow time vectors tmDistance frequency domain vector frAnd an estimated value theta of squint angleesqTo find a distance walk correction matrix RMM×NThe expression is as follows:
Figure BDA0001722348680000072
wherein fftshift represents fourier spectrum shift operation, superscript T represents transpose operation, and C represents speed of light.
(4b2) Echo signal matrix after inertial navigation motion compensation by using M GPU threads
Figure BDA0001722348680000073
Distance dimension Fourier transform is carried out to obtain a distance frequency domain echo signal matrix after inertial navigation motion compensation
Figure BDA0001722348680000074
The superscript IF represents that the signal after inertial navigation motion compensation processing is a frequency domain signal; using mxn GPU threads will
Figure BDA0001722348680000081
And RMM×NDot multiplication is carried out to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure BDA0001722348680000082
The operation process is represented as:
Figure BDA0001722348680000083
the missile-borne SAR works in a front side view mode, so that distance walking is not negligible, and the distance walking has a complex form, so that distance walking correction has a certain amount of calculation.
Example 6
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as that in the embodiments 1 to 5, and the distance frequency domain echo signal matrix after the distance walk correction is performed in the step (4c)
Figure BDA0001722348680000084
And (3) performing range curvature correction and secondary range pulse pressure, which comprises the following implementation steps:
(4c1) determining a Doppler center reference vector U using the estimated Doppler center fdcMThen, the distance walk corrected distance frequency domain echo signal matrix is subjected to
Figure BDA0001722348680000085
Performing azimuth dimension Fourier transform, performing point multiplication on the obtained result and a reference vector, and performing azimuth dimension Fourier inverse transform on the obtained result to obtain a range frequency domain echo matrix with the azimuth Doppler center removed
Figure BDA0001722348680000086
The superscript F indicates that the signal is a frequency domain signal, and the operation process is represented by the following formula:
Figure BDA0001722348680000087
wherein the IFFTARepresenting an azimuth dimension inverse fourier transform.
(4c2) Utilizing radar parameters to obtain secondary distance pulse pressure and orientation decoupling matrix HM×NUsing M N GPU threads will
Figure BDA0001722348680000088
And HM×NDot multiplication to obtain orientation decoupling echo signal matrix
Figure BDA0001722348680000089
Superscript a denotes the orientation decoupling process.
(4c3) Decoupling an echo signal matrix in azimuth using N GPU threads
Figure BDA00017223486800000810
Performing azimuth dimension Fourier inversion, and then using M multiplied by N threads to perform inverse inversion on the signals and a Doppler center reference vector UMIs a conjugate vector of
Figure BDA00017223486800000811
Dot multiplication is carried out to obtain an echo signal matrix after distance dimension processing
Figure BDA00017223486800000812
A conjugate vector of the reference vector
Figure BDA00017223486800000813
Reference vector U pair using M GPU threadsMThe imaginary part of (c) is inverted.
The missile-borne SAR works in a front side view mode, and extra distance bending is introduced, so that the steps of distance bending correction and secondary distance pulse pressure cannot be omitted.
Example 7
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as that in the embodiments 1 to 6, and the echo signal time domain zero-filling matrix in the step (5a)
Figure BDA0001722348680000091
High-order phase compensation and high-order phase filtering are carried out,the operation steps are as follows:
(5a1) zero-padding matrix using N GPU thread pairs
Figure BDA0001722348680000092
Performing orientation dimension Fourier transform to obtain Doppler frequency domain zero-filling matrix
Figure BDA0001722348680000093
The superscript DF represents the Doppler frequency domain signal, which would be generated using 2 MXN GPU threads
Figure BDA0001722348680000094
And
Figure BDA0001722348680000095
performing dot multiplication to obtain a Doppler frequency domain echo signal matrix after high-order phase compensation and high-order phase filtering
Figure BDA0001722348680000096
The superscript C indicates a high-order phase processing.
(5a2) Obtaining an echo signal matrix after phase compensation and filtering by using 2M GPU threads
Figure BDA0001722348680000097
Performing inverse Fourier transform in azimuth dimension, and cutting off the front of each column of the transformed signal
Figure BDA0001722348680000098
Point and back
Figure BDA0001722348680000099
Obtaining echo signal matrix after high-order phase compensation and high-order phase filtering from point data
Figure BDA00017223486800000910
The interception is to skip the front of each line when reading the echo signal matrix
Figure BDA00017223486800000911
DotAnd then reading the data of M points.
The vibration of the missile-borne SAR platform introduces high-order phases into SAR echo data and needs to be removed. The high-order phase compensation and the high-order phase filtering are the parts with the largest calculation amount of the azimuth dimension processing, and the high-order phase compensation parameters and the high-order phase filtering parameters are calculated into a matrix, namely C2M×NHigh-order phase compensation and high-order phase filtering can be completed only by once matrix operation, and the matrix operation is processed in a GPU in parallel, so that good real-time performance can be obtained under the condition of large operation amount.
A more complete and thorough example is given below to further illustrate the invention:
example 8
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as the missile-borne SAR front-side-view imaging methods in the embodiments 1 to 7, and referring to FIG. 2, the missile-borne SAR front-side-view imaging method based on the embedded GPU comprises the following steps:
step 1), in a front-end FPGA of a signal processor, performing distance dimension pulse compression and distance dimension interception on an original echo sampling matrix:
sampling matrix S for original echo in front-end FPGA of signal processorM×4NPerforming distance dimension pulse compression to obtain an echo signal matrix after distance pulse compression
Figure BDA00017223486800000912
Wherein, M is the number of pulses, namely the number of azimuth sampling points, and 4N is the number of sampling points of each pulse, namely the number of distance sampling points; cut off range pulse pressure echo signal matrix
Figure BDA00017223486800000913
Before each row in the
Figure BDA00017223486800000914
Point and back
Figure BDA00017223486800000915
Obtaining the data of points to obtain a distance dimension intercepted echo signal matrix SM×NIn the embodiment, the value of M is 4096, the value of N is 1024, and the step is performed in the front-end FPGA, which is beneficial to relieving the GPU operation pressure; after the range dimension pulse pressure, the main energy of the echo signal is concentrated in the center of the range dimension, so it is reasonable to intercept both sides of the range dimension of the echo signal matrix.
Step 2) intercepting echo signal matrix S from distance dimensionM×NAnd double preservation of the synthetic aperture radar parameters:
setting synthetic aperture radar parameters in a CPU (central processing unit) and copying the parameters to a GPU (graphics processing unit) memory; reading distance dimension intercepting echo signal matrix S in CPUM×NThe CPU memory is accessed and copied to the GPU memory, and the copying result is that the CPU and the GPU respectively have a data pointer pointing to the data in the memory, and the CPU and the GPU can operate the data through the pointer; the synthetic aperture radar parameters comprise an azimuth dimension sampling point number M, a distance dimension sampling point number N and a distance sampling rate fsPulse repetition frequency PRF, bandwidth B, light speed C, pulse width TpCarrier frequency fcVelocity v of missile, height H of missile, angle of declination theta0Synthetic aperture radar beam center line sweep target slant distance RsAnd the number M of the orientation dimension sampling points and the number N of the distance dimension sampling points provide a memory access boundary condition for the GPU thread.
Step 3) intercepting an echo signal matrix S for a distance dimension by using inertial navigation data and radar parametersM×NAnd (3) performing motion compensation in the embedded GPU:
step 3a) obtaining an inertial navigation data matrix I with three dimensions of north, sky and east from the front end of the radar3×LWherein L is the length of inertial navigation data of a single dimension, and in the CPU, the inertial navigation data matrix I is respectively aligned3×LEach row of the inertial navigation data matrix is subjected to cubic spline interpolation, the data length is interpolated to M, and an interpolated inertial navigation data matrix I is obtained3×MThe interpolation operation is mostly serial operation, parallel processing cannot be performed in the GPU, if interpolation is performed in the GPU, only a single GPU computing core can be used, and the operation speed of interpolation using the single GPU computing core is much lower than that of interpolation using the CPU.
Step 3b) in the embedded GPU, utilizing the plugValued inertial navigation data matrix I3×MTo obtain the inertial navigation motion compensation matrix RM×NThen setting M multiplied by N GPU threads and compensating the inertial navigation motion matrix RM×NAnd distance dimension intercepting echo signal matrix SM×NPerforming dot multiplication to obtain an inertial navigation motion compensation echo signal matrix
Figure BDA0001722348680000101
An mxn thread allocation manner at a GPU end, as shown in fig. 5, where the number of allocated threads corresponds to the number of points in a matrix, and each thread processes operations of a single point, which can greatly increase the operation speed, it should be noted that a GPU only has 256 computational cores, and mxn may be much greater than 256, which means that one computational core can run multiple threads, and threads on the same computational core are processed serially.
Step 4) compensating echo signal matrix for inertial navigation motion
Figure BDA0001722348680000102
And (3) performing distance dimension processing in the embedded GPU:
step 4a) compensating an echo signal matrix by inertial navigation motion
Figure BDA00017223486800001114
Dividing the direction dimension into Q blocks, and obtaining estimated values fdc of Q local Doppler centers by using Q block matrixesiI 1,2, …, Q, then estimates fdc of the Q local doppler centersiCalculating an average value to obtain an estimated value fdc of the global Doppler center, and further calculating an estimated value theta of the squint angle by using an estimated value fdc of the global Doppler centeresqIn this example, the value of Q is 256, the block division method is as shown in fig. 6, and as can be seen from fig. 6, there is an overlap between blocks in the present invention, and in order to improve the estimation accuracy, the overlap between blocks at the time of block division is necessary, and the overlap effectively reflects the slow change characteristic of the doppler center. If the blocks do not overlap, the scheme becomes doppler estimation using global data.
Step 4b) at the GPU end, utilizing the estimated value fdc of the global Doppler center and the estimated value theta of the squint angleesqTo the echo signal matrix after inertial navigation motion compensation
Figure BDA0001722348680000111
Carrying out distance walk correction to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure BDA0001722348680000112
Step 4b1) initializing a slow time vector t using M GPU threadsmInitializing distance frequency domain vector f using N GPU threadsrThen using the slow time vector tmDistance frequency domain vector frAnd an estimated value theta of squint angleesqTo find a distance walk correction matrix RMM×N。
Step 4b2) using M GPU threads to compensate the echo signal matrix after inertial navigation motion compensation
Figure BDA0001722348680000113
Distance dimension Fourier transform is carried out to obtain a distance frequency domain echo signal matrix after inertial navigation motion compensation
Figure BDA0001722348680000114
Using mxn GPU threads will
Figure BDA0001722348680000115
And RMM×NDot multiplication is carried out to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure BDA0001722348680000116
Step 4c) in the embedded GPU, distance frequency domain echo signal matrix after distance walk correction
Figure BDA0001722348680000117
Performing range curvature correction and secondary range pulse pressure to obtain echo signal matrix after range dimension processing
Figure BDA0001722348680000118
Step 4c1) obtaining a Doppler center reference vector U using the estimated Doppler center fdcMThen, the distance walk corrected distance frequency domain echo signal matrix is subjected to
Figure BDA0001722348680000119
Performing azimuth dimension Fourier transform, performing point multiplication on the obtained result and a reference vector, and performing azimuth dimension Fourier inverse transform on the obtained result to obtain a range frequency domain echo matrix with the azimuth Doppler center removed
Figure BDA00017223486800001110
The operation process is represented as:
Figure BDA00017223486800001111
step 4c2) utilizing radar parameters and radar motion parameters to obtain a secondary distance pulse pressure and orientation decoupling matrix HM×NUsing M N GPU threads will
Figure BDA00017223486800001112
And HM×NDot multiplication to obtain orientation decoupling echo signal matrix
Figure BDA00017223486800001113
Step 4c3) decoupling the echo signal matrix for the azimuth using N GPU threads
Figure BDA0001722348680000121
Performing azimuth dimension Fourier inversion, and then using M multiplied by N threads to perform inverse inversion on the signals and a Doppler center reference vector UMIs a conjugate vector of
Figure BDA0001722348680000122
Dot multiplication is carried out to obtain an echo signal matrix after distance dimension processing
Figure BDA0001722348680000123
Step 5) echo signal matrix after distance dimension processing
Figure BDA0001722348680000124
And (3) carrying out orientation dimension processing in the embedded GPU:
step 5a) echo signal matrix after distance dimension processing
Figure BDA0001722348680000125
Performing time domain zero padding to obtain an echo signal time domain zero padding matrix
Figure BDA0001722348680000126
Calculating a higher order phase compensation and higher order phase filter matrix C for each element of the zero-padding matrix2M×NBy using C2M×NTime domain zero-filling matrix for echo signal
Figure BDA0001722348680000127
Performing high-order phase compensation and high-order phase filtering and intercepting to obtain a high-order phase compensation matrix and a high-order phase filtered echo signal matrix
Figure BDA0001722348680000128
Step 5a1) using N GPU-thread pairs for zero-padding matrices
Figure BDA0001722348680000129
Performing orientation dimension Fourier transform to obtain Doppler frequency domain zero-filling matrix
Figure BDA00017223486800001210
Then use 2 mxn GPU threads to execute
Figure BDA00017223486800001211
And C2M×NPerforming dot multiplication to obtain a Doppler frequency domain echo signal matrix after high-order phase compensation and high-order phase filtering
Figure BDA00017223486800001212
Step 5a2) of obtaining an echo signal matrix after phase compensation and filtering
Figure BDA00017223486800001213
Performing inverse Fourier transform in azimuth dimension, and cutting off the front of each column of transformed signal
Figure BDA00017223486800001214
Point and back
Figure BDA00017223486800001215
Obtaining echo signal matrix after high-order phase compensation and high-order phase filtering from point data
Figure BDA00017223486800001216
Step 5b) utilizing the radar parameters, the radar motion parameters and the estimated squint angle thetaesqObtaining an orientation deskew matrix DRM×NUsing mxn GPU threads will
Figure BDA00017223486800001217
And DRM×NDot multiplication to obtain azimuth deskew echo signal matrix
Figure BDA00017223486800001218
Step 5c) adopting an iterative mode to deskew the azimuth echo signal matrix
Figure BDA00017223486800001219
Estimating and correcting the motion error, wherein the CPU performs iterative control, and the embedded GPU performs iterative operation, so that the control advantage of the CPU and the operation advantage of the GPU can be exerted;
step 5d), in the embedded GPU, carrying out azimuth pulse pressure on the echo signal matrix after motion error estimation and correction to obtain focused SAR image data;
and 6) carrying out geometric correction on the focused SAR image data to obtain a high-resolution SAR image.
The invention adopts FPGA to preprocess the missile-borne SAR echo signal and adopts embedded GPU to carry out imaging processing. The existing method takes DSP as a processing core, imaging processing adopts 8-core parallel processing, and the parallelism is low.
The existing missile-borne SAR imaging method with DSP as a processing core adopts a C language programming method based on a bare engine, which needs to be realized by self for some conventional mathematical operations, and the programming is relatively complex.
The technical effects of the present invention will be described below with reference to the measured data processing experiment.
Example 9
The missile-borne SAR front-side-view imaging method based on the embedded GPU is the same as that of the embodiments 1 to 8,
experimental conditions and contents:
the parameters of the radar system used for the actual measurement data imaging experiment data recording are shown in table 1:
TABLE 1 Radar System parameters for recording measured data
Wave band Ku band Oblique angle 80°
Sampling rate 70MHz Bandwidth of 50MHz
Height 500m Wave beam width 7.5°
Pulse width 25us Speed of rotation 50m/s
The actual measurement data are respectively subjected to imaging experiments by using MATLAB software on a PC, an FPGA-multi-core DSP method and the invention, wherein a CPU of the PC is an Intel core I7-4790 processor with a main frequency of 3.6GHz, the multi-core DSP is 8-core TMS320C6678 of a Texas instrument, a high-performance embedded GPU used by the invention is an Intga Tegra Parker, and the obtained imaging results are shown in FIG. 7, wherein FIG. 7(a) is an imaging result of processing the actual measurement data by using MATLAB, FIG. 7(b) is an imaging result of processing the actual measurement data by using an FPGA-multi-core DSP hardware platform, and FIG. 7(C) is an imaging result of processing the actual measurement data by using the invention.
And (3) analyzing an experimental result:
referring to fig. 7, it can be seen that the differences of the three images of fig. 7(a), 7(b) and 7(c) can hardly be distinguished by naked eyes, because the measured data processed by different methods are the same data, the imaging effect is good, and in order to more accurately distinguish the differences of the images of the imaging results of the three processing methods, the image entropies of the three imaged images are shown in table 2.
TABLE 2 image entropy of three processing methods imaged images
Processing method MATLAB FPGA-multi-core DSP method The invention
Entropy of images 7.5744 7.5764 7.5764
Referring to table 2, it can be found that the image entropy of the imaging image processed by using the FPGA-multicore DSP and the present invention is the same, and is slightly higher than the image entropy of the imaging process by MATLAB, which indicates that the imaging image by MATLAB is slightly better because the MATLAB uses double-precision floating point operation in the operation, while the FPGA-multicore DSP platform and the present invention use single-precision floating point operation, and the engineering implementation of numerous radar signal processing algorithms indicates that the operation error caused by the single-precision operation can be ignored in the engineering application.
TABLE 3 imaging time for three treatment methods
Processing method MATLAB FPGA-multi-core DSP method The invention
Time of imaging 12.9091s 926ms 93ms
Acceleration ratio 138.81 9.95 1
Referring to table 3, it can be found that the imaging processing time of the present invention is 93 milliseconds, the imaging time is much shorter than 926 milliseconds of the FPGA-multi-core DSP method, the calculation speed is increased by one order of magnitude, and the real-time performance is greatly improved. Compared with MATLAB imaging processing, the imaging processing time speed-up ratio of the invention is improved to two orders of magnitude, because the imaging processing in MATLAB is single-core serial processing, and the FPGA-multi-core DSP uses 8-core parallel processing, while the invention uses 256 CUDA cores for parallel processing, and the parallelism is much higher than other two processing methods.
In summary, the missile-borne SAR front-side-view imaging method based on the embedded GPU provided by the invention solves the technical problem of poor imaging algorithm real-time performance in the existing missile-borne SAR imaging platform, and comprises the following implementation steps: 1) in a front-end FPGA of a trust place, distance dimension pulse compression and distance dimension interception are carried out on an original echo sampling matrix of the missile-borne SAR image; 2) the echo signal matrix intercepted by the distance dimension and the SAR parameter are subjected to CPU and GPU duplicate storage; 3) performing motion compensation on the distance dimension intercepted echo signal matrix in the embedded GPU by using inertial navigation data and radar parameters; 4) distance dimension processing is carried out on the inertial navigation motion compensation echo signal matrix in the embedded GPU 5) orientation dimension processing is carried out on the echo signal matrix after the distance dimension processing in the embedded GPU to obtain focused SAR image data; 6) and geometrically correcting the focused SAR image data in the embedded GPU to obtain a high-resolution missile-borne SAR image. The invention improves the imaging parallelism, reduces the imaging computation amount and the computation time, improves the real-time property, is applied to the missile seeker, and is beneficial to the accurate identification and striking of the missile to the target.

Claims (7)

1. A missile-borne SAR front-side imaging method based on an embedded GPU is characterized by comprising the following steps:
(1) in a front-end FPGA of a signal processor, performing distance dimension pulse compression and distance dimension interception on an original echo sampling matrix of the missile-borne SAR image:
sampling matrix S for original echo in front-end FPGA of signal processorM×4NPerforming distance dimension pulse compression to obtain an echo signal matrix after distance pulse compression
Figure FDA0003510622390000011
Wherein M is the number of pulses, namely the number of azimuth sampling points, 4N is the number of sampling points of each pulse, namely the number of distance sampling points, N is a positive even number, and the superscript P represents the distance pulse pressure; cut off range pulse pressure echo signal matrix
Figure FDA0003510622390000012
Before each row in the
Figure FDA0003510622390000013
Point and back
Figure FDA0003510622390000014
Obtaining the data of points to obtain a distance dimension intercepted echo signal matrix SM×N；
(2) Intercepting echo signal matrix S from distance dimensionM×NAnd double preservation of the synthetic aperture radar parameters:
setting synthetic aperture radar parameters in a CPU at the rear end of the signal processor, and copying the parameters to an embedded GPU memory at the rear end of the signal processor; reading distance dimension intercepting echo signal matrix S in CPUM×NThe CPU memory is accessed and copied to the embedded GPU memory;
(3) distance dimension interception echo signal matrix S by using missile inertial navigation data and radar parametersM×NMotion compensation is performed in the embedded GPU:
(3a) obtaining inertial navigation data matrix I of north, sky and east dimensions from missile inertial navigation system3×LWherein L is the length of the inertial navigation data of a single dimension, and at the end of the CPU, the inertial navigation data matrixes I are respectively aligned3×LEach row of the inertial navigation data matrix is subjected to cubic spline interpolation, the data length is interpolated to M, and an interpolated inertial navigation data matrix I is obtained3×M；
(3b) In an embedded GPU, utilizing an interpolated inertial navigation data matrix I3×MTo obtain the inertial navigation motion compensation matrix RM×NThen setting M multiplied by N GPU threads and compensating the inertial navigation motion matrix RM×NAnd distance dimension intercepting echo signal matrix SM×NPerforming dot multiplication to obtain an inertial navigation motion compensation echo signal matrix
Figure FDA0003510622390000015
Superscript I represents inertial navigation motion compensation;
(4) echo signal matrix for inertial navigation motion compensation
Figure FDA0003510622390000016
Distance dimension processing is carried out in the embedded GPU to obtain an echo signal matrix after distance dimension processing
Figure FDA0003510622390000017
(5) Echo signal matrix processed by distance dimension
Figure FDA0003510622390000021
Performing orientation dimension processing in an embedded GPU to obtain focused SAR image data;
(6) and geometrically correcting the focused SAR image data in the embedded GPU to obtain a high-resolution missile-borne SAR image.
2. The embedded-GPU-based missile-borne SAR front-side-view imaging method according to claim 1, wherein the distance dimension processing in the embedded GPU in step (4) comprises the following steps:
(4a) evaluating an oblique view angle: echo signal matrix compensated by inertial navigation motion
Figure FDA0003510622390000022
Dividing the direction dimension into Q blocks, and obtaining estimated values fdc of Q local Doppler centers by using Q block matrixesiI 1,2, …, Q, then estimates fdc of the Q local doppler centersiCalculating an arithmetic mean value to obtain an estimated value fdc of a global Doppler center, and further calculating an estimated value theta of an oblique angle of view by using an estimated value fdc of the global Doppler centeresq；
(4b) Distance walk correction: in the embedded GPU, the estimated value fdc of the global Doppler center and the estimated value theta of the squint angle are utilizedesqTo the echo signal matrix after inertial navigation motion compensation
Figure FDA0003510622390000023
Carrying out distance walk correction to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure FDA0003510622390000024
(4c) Range curvature correction and second order range pulse pressure: in an embedded GPU, a range frequency domain echo signal matrix after range walk correction
Figure FDA0003510622390000025
Performing range curvature correction and secondary range pulse pressure to obtain echo signal matrix after range dimension processing
Figure FDA0003510622390000026
Superscript R denotes distance dimension processing.
3. The embedded-GPU-based missile-borne SAR front-side-view imaging method according to claim 1, wherein the orientation dimension processing in the embedding in step (5) comprises the following steps:
(5a) and (3) high-order phase processing: in the embedded GPU, the echo signal matrix after distance dimension processing
Figure FDA0003510622390000027
Performing time domain zero padding to obtain an echo signal time domain zero padding matrix
Figure FDA0003510622390000028
Calculating a higher order phase compensation and higher order phase filter matrix C for each element of the zero-padding matrix2M×NBy using C2M×NTime domain zero-filling matrix for echo signal
Figure FDA0003510622390000029
Performing high-order phase compensation and high-order phase filtering and intercepting to obtain a high-order phase compensation matrix and a high-order phase filtered echo signal matrix
Figure FDA0003510622390000031
Superscript H represents high-order phase processing;
(5b) azimuth declivity: using radar parameters, radar motion parameters, and estimated squint angle θesqSolving an orientation deskew matrix DM×NUsing M N GPU threads will
Figure FDA0003510622390000032
And DM×NDot multiplication to obtain azimuth deskew echo signal matrix
Figure FDA0003510622390000033
The superscript D represents the azimuth deskew process;
(5c) motion error estimation and correction: using an iterative method to deskew the echo signal matrix of the azimuth
Figure FDA0003510622390000034
Estimating and correcting the motion error, wherein the CPU performs iterative control, and the embedded GPU performs iterative operation;
(5d) azimuthal pulse pressure: and in the embedded GPU, performing azimuth pulse pressure on the echo signal matrix after motion error estimation and correction to obtain focused SAR image data.
4. The embedded GPU-based missile-borne SAR front-side view imaging method according to claim 1, wherein the synthetic aperture radar parameters in the step (2) are except for the distance sampling rate fsPulse repetition frequency PRF, bandwidth B, light speed C, pulse width TpCarrier frequency fcVelocity v of missile, height H of missile, angle of declination theta0Synthetic aperture radar beam center line sweep target slant distance RsThe method also comprises an orientation dimension sampling point number M and a distance dimension sampling point number parameter N.
5. The embedded GPU-based missile-borne SAR front-side view imaging method according to claim 2, wherein the echo signal matrix after inertial navigation motion compensation in the step (4b)
Figure FDA0003510622390000035
And (3) carrying out distance walking correction, wherein the implementation steps are as follows:
(4b1) initializing a slow time vector t using M GPU threadsmInitializing distance frequency domain vector f using N GPU threadsrUsing slow time vectors tmDistance frequency domain vector frAnd an estimated value theta of squint angleesqTo find a distance walk correction matrix RMM×N；
(4b2) Echo signal matrix after inertial navigation motion compensation by using M GPU threads
Figure FDA0003510622390000036
Distance dimension Fourier transform is carried out to obtain a distance frequency domain echo signal matrix after inertial navigation motion compensation
Figure FDA0003510622390000037
The superscript IF represents that the signal after inertial navigation motion compensation processing is a frequency domain signal; using mxn GPU threads will
Figure FDA0003510622390000038
And RMM×NDot multiplication is carried out to obtain a distance frequency domain echo signal matrix after distance walk correction
Figure FDA0003510622390000039
6. The embedded GPU-based missile-borne SAR front-side view imaging method according to claim 2, wherein the distance frequency domain echo signal matrix corrected for distance walk in the step (4c)
Figure FDA0003510622390000041
And (3) performing range curvature correction and secondary range pulse pressure, which comprises the following implementation steps:
(4c1) using the estimated Doppler center fdc to find the Doppler center reference vector UMThen, the distance walk corrected distance frequency domain echo signal matrix is subjected to
Figure FDA0003510622390000042
Performing azimuth dimension Fourier transform, performing point multiplication on the obtained result and a reference vector, and performing azimuth dimension Fourier inverse transform on the obtained result to obtain a range frequency domain echo matrix with the azimuth Doppler center removed
Figure FDA0003510622390000043
The superscript F indicates that the signal is a frequency domain signal, and the operation procedure is expressed as:
Figure FDA0003510622390000044
(4c2) calculating two by using radar parametersSub-range pulse pressure and orientation decoupling matrix HM×NUsing M N GPU threads will
Figure FDA0003510622390000045
And HM×NDot multiplication to obtain orientation decoupling echo signal matrix
Figure FDA0003510622390000046
Superscript a represents the orientation decoupling process;
(4c3) decoupling an echo signal matrix in azimuth using N GPU threads
Figure FDA0003510622390000047
Performing azimuth dimension Fourier inversion, and then using M multiplied by N threads to perform inverse inversion on the signals and a Doppler center reference vector UMIs a conjugate vector of
Figure FDA0003510622390000048
Dot multiplication is carried out to obtain an echo signal matrix after distance dimension processing
Figure FDA0003510622390000049
7. The embedded GPU-based missile-borne SAR front-side view imaging method according to claim 3, wherein the echo signal time domain zero-filling matrix in the step (5a)
Figure FDA00035106223900000410
Carrying out high-order phase compensation and high-order phase filtering and intercepting, and the operation steps are as follows:
(5a1) zero-padding matrix using N GPU thread pairs
Figure FDA00035106223900000411
Performing orientation dimension Fourier transform to obtain Doppler frequency domain zero-filling matrix
Figure FDA00035106223900000412
The superscript DF indicates that the signal is a Doppler frequency domain signal, which would then be processed using 2 MXN GPU threads
Figure FDA00035106223900000413
And C2M×NPerforming dot multiplication to obtain a Doppler frequency domain echo signal matrix after high-order phase compensation and high-order phase filtering
Figure FDA00035106223900000414
Superscript C represents high-order phase processing;
(5a2) obtaining an echo signal matrix after phase compensation and filtering by using 2M GPU threads
Figure FDA00035106223900000415
Performing an inverse Fourier transform in the azimuth dimension, where M is an even number, and truncating the front of each column of the transformed signal
Figure FDA00035106223900000416
Point and back
Figure FDA00035106223900000417
Obtaining echo signal matrix after high-order phase compensation and high-order phase filtering from point data
Figure FDA0003510622390000051
CN201810737271.3A 2018-07-06 2018-07-06 Missile-borne SAR front-side-view imaging method based on embedded GPU Active CN108919220B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810737271.3A CN108919220B (en) 2018-07-06 2018-07-06 Missile-borne SAR front-side-view imaging method based on embedded GPU

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810737271.3A CN108919220B (en) 2018-07-06 2018-07-06 Missile-borne SAR front-side-view imaging method based on embedded GPU

Publications (2)

Publication Number Publication Date
CN108919220A CN108919220A (en) 2018-11-30
CN108919220B true CN108919220B (en) 2022-05-17

Family

ID=64425187

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810737271.3A Active CN108919220B (en) 2018-07-06 2018-07-06 Missile-borne SAR front-side-view imaging method based on embedded GPU

Country Status (1)

Country Link
CN (1) CN108919220B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110913227B (en) * 2019-11-28 2021-12-21 山东浪潮科学研究院有限公司 Edge-end image compression system and method of heterogeneous computing architecture
CN111175749B (en) * 2020-01-19 2022-02-25 中国科学院电子学研究所 Satellite-borne SAR imaging processing method
DE102020001157A1 (en) * 2020-02-21 2021-08-26 Diehl Defence Gmbh & Co. Kg Method for targeting a missile, missile control and missile
CN111381217B (en) * 2020-04-01 2022-10-18 上海无线电设备研究所 Missile-borne SAR motion compensation method based on low-precision inertial navigation system
CN112764029B (en) * 2020-12-16 2024-03-22 北京无线电测量研究所 SAR real-time imaging realization method and device based on GPU
CN113156435B (en) * 2021-02-08 2023-09-15 西安电子科技大学 Missile-borne SAR front side view time domain imaging method based on embedded GPU
CN114325703B (en) * 2021-12-27 2024-10-01 西安电子科技大学 Real-time imaging processing device and method for high-frame-rate video SAR

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0583972A1 (en) * 1992-08-17 1994-02-23 Texas Instruments Incorporated Improvements in and relating to precision targeting
CN103454635A (en) * 2013-09-08 2013-12-18 西安电子科技大学 Front squint SAR imaging method based on hypersonic aircraft at level flight section
CN104459693A (en) * 2014-12-01 2015-03-25 西安电子科技大学 Missile-borne SAR forward-squint imaging method based on GPU
CN104777479A (en) * 2015-05-05 2015-07-15 西安电子科技大学 Front-side-looking SAR real-time imaging method based on multi-core DSP

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0583972A1 (en) * 1992-08-17 1994-02-23 Texas Instruments Incorporated Improvements in and relating to precision targeting
CN103454635A (en) * 2013-09-08 2013-12-18 西安电子科技大学 Front squint SAR imaging method based on hypersonic aircraft at level flight section
CN104459693A (en) * 2014-12-01 2015-03-25 西安电子科技大学 Missile-borne SAR forward-squint imaging method based on GPU
CN104777479A (en) * 2015-05-05 2015-07-15 西安电子科技大学 Front-side-looking SAR real-time imaging method based on multi-core DSP

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
弹载前斜视SAＲ BP成像算法与实现;张筱靖 等;《现代防御技术》;20171231;第45卷(第6期);第48-52页 *

Also Published As

Publication number Publication date
CN108919220A (en) 2018-11-30

Similar Documents

Publication Publication Date Title
CN108919220B (en) Missile-borne SAR front-side-view imaging method based on embedded GPU
CN111142105B (en) ISAR imaging method for complex moving target
Borden Radar Imaging of Airborne Targets: A primer for applied mathematicians and physicists
CN113156435B (en) Missile-borne SAR front side view time domain imaging method based on embedded GPU
CN109597072B (en) Imaging processing method and device of bistatic Synthetic Aperture Radar (SAR) system
CN111538005B (en) SAR front-side-looking imaging method based on FPGA and multiple multi-core DSPs
CN103197315A (en) Flow structure non-uniform linear motion synthetic aperture radar imaging method
CN104237886B (en) High-precision synthetic aperture radar imaging method
CN116520335A (en) Multi-receiving array element synthetic aperture sonar wave number domain imaging method
US5488373A (en) Signal processing apparatus for synthetic aperture radar
CN117169882A (en) Shipborne radar sea wave information inversion method
CN113219433B (en) Knowledge-aided SR-STAP method and storage medium
CN101819274B (en) Stretching nonlinear scaling method for imaging processing of forward squint-looking sub-aperture of synthetic aperture radar
CN105242264B (en) A kind of planar array multiple target angle high-resolution implementation method
CN116719027A (en) Star-machine double-base SAR imaging method in bidirectional sliding beam focusing mode
CN113376632B (en) Large strabismus airborne SAR imaging method based on pretreatment and improved PFA
CN116106914A (en) Multi-receiving-array SAS six-degree-of-freedom motion error compensation and imaging method and system
Gong et al. High resolution 3D InISAR imaging of space targets based on PFA algorithm with single baseline
CN110320492A (en) A kind of low complex degree DOA estimation method based on steering vector symmetry characteristic
CN115877380A (en) SAR multi-moving-target imaging method and device and storage medium
Shao et al. Noise-robust interferometric ISAR imaging of 3-D maneuvering motion targets with fine image registration
CN114002666A (en) Method and equipment for extracting satellite-borne ATI-SAR ocean current flow velocity under any antenna configuration
CN115542320A (en) Rapid real-time sub-aperture imaging method and device for ground-based synthetic aperture radar
CN115267706A (en) Synthetic aperture radar distance space-variant phase error estimation method, device and medium
CN113655458B (en) Space-time adaptive processing method and device based on dictionary correction and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant