CN111562976B - GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target - Google Patents
GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target Download PDFInfo
- Publication number
- CN111562976B CN111562976B CN201910113169.0A CN201910113169A CN111562976B CN 111562976 B CN111562976 B CN 111562976B CN 201910113169 A CN201910113169 A CN 201910113169A CN 111562976 B CN111562976 B CN 111562976B
- Authority
- CN
- China
- Prior art keywords
- radar
- imaging
- electrically large
- gpu
- calculation
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
Abstract
The invention provides a GPU (graphics processing unit) acceleration method for radar imaging of an electrically large target, which is used for quickly imaging an electrically large complex target detected by a radar by using a graphics processor and is characterized by comprising the following steps of: s1, establishing a geometric model of an electrically large-size complex target according to radar working parameters acquired by a radar; s2, performing initial mesh generation on the surface of the geometric model through GMSH to form a plurality of triangular structures; s3, calculating a scattered field received by the electrically large-size complex target at a receiving antenna of the radar according to the triangular structure and a plane wave incident field obtained in advance by a physical optical method, and distributing a calculation task in the calculation process to a GPU for parallel calculation; and S4, combining all scattered fields and the geometric model by a fast back projection method to form an image so as to obtain a radar imaging result.
Description
Technical Field
The invention belongs to the field of radar imaging, and particularly relates to a GPU (graphics processing unit) acceleration method and system for radar imaging of an electrically large-size target.
Background
For a complex target in a radar monitoring multi-interference environment, the method has the characteristic of complex physical structure, belongs to an electric large-size range, and has the advantages that a high-frequency approximation method is selected in the past, such as a Kirchhoff tangent plane method, a geometric optics method, a physical optics method, a geometric diffraction theory, a physical diffraction theory, a ray tracing method and the like. However, the accuracy is not comparable to the numerical method. The numerical method requires a large computer memory and a long calculation time, and the bottleneck of the computer hardware level limits the application of the numerical method for a long time.
On the other hand, the computational advantages of Graphics Processing Units (GPUs), which have powerful computational power and very high memory bandwidth, have been highlighted over the last few years, all of which have led GPUs to be successfully applied to other complex computational problems from special parallel multi-core processors. In particular, the TESLA system product CUDA (computer Unified Device Architecture) developed by Nvidia provides a simple and efficient way to massively utilize resources on a GPU in parallel.
It is reported in the literature that GPU-based accelerated computing in the field of electromagnetic computing has done a lot of work. In the work on accelerated computational electromagnetism using a GPU, graphical electromagnetic computation (GRECO) is the first to accelerate the computation of the first order fringing field of the visible surface and wedges of the target using graphical hardware. However, the method is used for estimating the radar scattering cross section of a single complex target, and the universality is not enough; even a certain usage right is specified for the user.
In order to better adapt to the problem diversity of multi-interference environment composite scattering in practical engineering research, a quick and effective real-time target monitoring system simulation software is developed, and GPU hardware acceleration and parallel computation are adopted to realize the full polarization radar imaging simulation of a complex target from the mechanism of electromagnetic scattering of the complex target.
The imaging method of the GPU-based rear projection dual-station synthetic aperture radar is designed in Chinese patent No. 201210334304.2, and the imaging processing efficiency is improved by optimizing the grading parameter of the traditional quick factorization rear projection method compared with the traditional quick factorization rear projection method; on the other hand, the imaging task is separated into a plurality of independent subtask groups, so that the GPU carries out parallel processing on the subtask groups, and the processing efficiency is improved.
However, this approach has several drawbacks: 1. the back projection algorithm adopted by the method is high in complexity, and for a complex target, the subdivision number of a target surface element is extremely large, so that for a certain calculation accuracy requirement, the memory of a computer is still obviously insufficient under the method, and the calculation time is too long; 2. the patent does not contain a display platform for displaying geometric data, illumination and texture characteristics of a complex three-dimensional target, so that a high-resolution imaging effect is difficult to obtain; 3. this patent copies the distance-compressed data into the memory space of the graphics processor GPU, resulting in a reduced processing speed.
Disclosure of Invention
In order to solve the problems and overcome the defects of the high-resolution radar imaging technology in the existing multi-interference environment, the invention provides a GPU acceleration method and a system which can more efficiently and quickly carry out radar imaging on electrically large-size complex targets, and the invention adopts the following technical scheme:
the invention provides a GPU (graphics processing unit) acceleration method for radar imaging of an electrically large target, which is used for quickly imaging an electrically large complex target detected by a radar by using a graphics processor and is characterized by comprising the following steps of: s1, establishing a geometric model of an electrically large-size complex target according to radar working parameters acquired by a radar; s2, performing initial mesh generation on the surface of the geometric model through GMSH to form a plurality of triangular structures; s3, calculating a scattered field received by the electrically large-size complex target at a receiving antenna of the radar according to the triangular structure and a plane wave incident field obtained in advance by a physical optical method, and distributing a calculation task in the calculation process to a GPU for parallel calculation; and S4, combining all scattered fields and the geometric model by a fast back projection method to form an image so as to obtain a radar imaging result. Wherein, step S3 includes the following substeps: s3-1, calculating a triangular structure irradiated by incident waves in a plurality of triangular structures as a triangular structure to be calculated according to a plane wave incident field acquired in advance; s3-2, sequentially acquiring a triangular structure to be calculated and distributing the triangular structure to each thread of the GPU; s3-3, the thread acquires the calculation parameters of the triangular structure according to the distributed triangular structure to be calculated; and S3-4, calculating the scattered field parameters of the triangular structure to be calculated in the directions of all receiving antennas by the thread according to the distributed triangular structure to be calculated and the corresponding calculation parameters.
The GPU acceleration method for radar imaging of electrically large-sized targets provided by the present invention may further have the technical feature that, in step S3, the principle of calculating the scattered field by a physical optical method is as follows: noting the frequency of the incident wave as f, the corresponding incident wave wavelength λ = c/f, the angular wave number k =2 π/λ, and the form of the incident wave is:
in the formula (I), the compound is shown in the specification,
is an incident electric field, unit vector->
For the direction of propagation of the incident wave, is greater or less>
Is an initial value dependent on the direction of polarization of the incident wave>
Is an incident magnetic field, is greater or less>
Is the initial value of the incident magnetic field, j denotes the imaginary part, mu 0 The magnetic permeability is vacuum magnetic permeability, c is free space light velocity, r is distance parameter, further, the calculation of the incident wave of the electrically large complex target is carried out in the appointed direction->
The far-field scattered wave generated at the upper part is recorded as S on the surface of the geometric model, and the calculated gravity center of the triangular structure is->
The unit normal vector is ^ 5>
Then the pair->
In a computing environment that employs PO approximation, if->
Is directly irradiated by the incident wave, and the area current density at the calculated gravity center of the triangular structure is->
Can be approximated as:
in the entire surface S, the portion irradiated with the incident wave is denoted as S 1 Due to the existence of the area current density, in
Field point in the direction->
(r->Infinity) generated far-field scattered waves are:
accordingly, the fringe field parameters
Comprises the following steps:
in the formula, epsilon 0 Is the vacuum dielectric constant.
The GPU acceleration method for radar imaging of the electrically large target further has the technical characteristics that in the step S3-3, the thread acquires the calculation parameters of the triangular structure according to the distributed triangular structure to be calculated and stores the calculation parameters in the register of the GPU.
The GPU acceleration method for imaging of the electrically large-size target radar provided by the invention can also have the technical characteristics that in the step S2, the surface of the geometric model is subjected to initial mesh subdivision by setting subdivision types, surface element subdivision sizes and calculating frequency parameters through GMSH.
The GPU acceleration method for radar imaging of the electrically large-size target provided by the invention also has the technical characteristics that in the step S4, when the imaging is carried out through a fast back projection method, the edge filtering is carried out on the geometric model through a Hanning window function.
The GPU acceleration method for radar imaging of the electrically large-size target can also have the technical characteristic that the effect of edge filtering is optimized by adjusting and testing the parameters of the Hanning window function.
The GPU acceleration method for radar imaging of the electrically large-size target can also have the technical characteristics that the electrically large-size complex target is a target with the physical size far larger than the wavelength of the radar working frequency band in a high-frequency area, and the ratio of the physical size to the wavelength is larger than 10.
The GPU acceleration method for electrically large-size target radar imaging, provided by the invention, can also have the technical characteristics that the application range of a high-frequency area is 1-2GHz in an L wave band and 2-4GHz in an S wave band.
The invention provides a GPU (graphics processing unit) acceleration system for radar imaging of an electrically large target, which is characterized by comprising the following components: the radar is used for detecting the electrically large-size complex target so as to generate radar working parameters; the central processing unit is used for processing the radar working parameters so as to image the electrically large-size complex target; the graphic processor is provided with a plurality of thread processing units and is used for calculating calculation tasks generated when the central processor processes radar working parameters, wherein the central processor is provided with a working parameter acquisition part, a geometric model construction part, a model mesh division part, a calculation task generation part, a central side communication part and an imaging result display part, the graphic processor is also provided with a graphic side control part, a radar imaging part and a graphic side communication part, the working parameter acquisition part acquires radar working parameters from the radar, the geometric model construction part constructs a geometric model of an electrically large-size complex target according to the radar working parameters, the model mesh division part performs initial mesh division on the surface of the geometric model to form a plurality of triangular structures, the calculation task generation part sequentially generates corresponding calculation tasks according to the triangular structures, the central processing communication part sequentially transmits the calculation tasks to the graphic processor, the graphic processing control part sequentially receives the calculation tasks and distributes the calculation tasks to idle thread processing units, the thread processing units calculate the corresponding triangular structures according to obtain a plurality of scatter field parameters corresponding to the triangular structures, the radar imaging part obtains all scatter field parameters of the electrically large-size complex target and rapidly displays the radar side imaging result through a radar side backward projection imaging algorithm, and the radar imaging result display method.
Action and Effect of the invention
According to the GPU acceleration method and system for radar imaging of the electrically large-size target, a composite scattering model of the target in a multi-interference environment is established, the model is subdivided into smaller blocks by applying a CUDA program of an Nvidia TESLA system product, so that data to be processed by an algorithm is refined, the calculation complexity of each block in calculation is further simplified through a physical optical method, and finally data obtained through calculation are synthesized through accumulation so that the target is rapidly imaged; meanwhile, the GPU is adopted, a plurality of threads are created in the initialization parallel environment of the GPU to process each block of the target model, each thread can calculate the block and store the calculation result into the shared cache, and therefore the calculation efficiency of the full polarization SAR imaging process is further improved.
Drawings
Fig. 1 is a block diagram of a supporting facility planning assistance system based on real population indexes according to an embodiment of the present invention;
FIG. 2 is a block diagram of a CPU according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a geometric model of a Global Hawk drone in an embodiment of the present invention after being split;
FIG. 4 is a block diagram of a graphics processor in an embodiment of the invention;
FIG. 5 is a flow chart of a GPU acceleration method for radar imaging of electrically large targets according to an embodiment of the invention;
FIG. 6 is a schematic diagram showing a comparison of radar imaging results based on a GPU acceleration method and an original CPU calculation method in the embodiment of the present invention; and
FIG. 7 is a schematic diagram illustrating the effect of example calculation acceleration on 4 electrically large complex objects by the GPU acceleration system in the embodiment of the present invention.
Detailed Description
In order to make the technical means, the creation features, the achievement purposes and the effects of the invention easy to understand, the GPU acceleration method for radar imaging of electrically large-sized targets of the invention is specifically described below with reference to the embodiments and the accompanying drawings.
< example >
Fig. 1 is a block diagram of a GPU acceleration system for radar imaging of an electrically large target in an embodiment of the present invention.
As shown in fig. 1, a GPU acceleration system 100 for radar imaging of an electrically large target has a radar apparatus 1, a central processor 2, and a graphic processor 3.
The radar device 1 is used for detecting electrically large-sized complex targets so as to generate radar working parameters. In the present embodiment, the radar selected by the radar apparatus 1 is a Synthetic Aperture Radar (SAR).
In this embodiment, the electrically large complex target is a target having a physical size in a high frequency region far larger than a wavelength of a radar operating frequency band, and a ratio of the physical size to the wavelength is larger than 10, for example, a Global Hawk drone, a stealth aircraft, an aircraft carrier, or other targets. The application range of the high-frequency region is 1-2GHz in an L wave band and 2-4GHz in an S wave band.
The central processing unit 2 is used for processing the radar working parameters generated by the radar device 1 so as to image the large-size complex target. In this embodiment, the CPU 2 and the graphic processor 3 are a CPU and a GPU provided in one computer and connected to each other, and the computer is connected to the radar device 1 and capable of data communication.
The graphic processor 3 is used for calculating the calculation tasks generated when the central processing unit 2 processes the radar working parameters, so that the most time-consuming step in the whole calculation process is accelerated to calculate through the graphic processor 3. In this embodiment, the graphics processor 3 has a plurality of thread processing units, and can perform parallel processing on the calculation tasks generated by the central processing unit 2.
In this embodiment, the GPU selected by the graphics processor 3 is GeForce210 of Nvidia. The GPU is programmed by adopting CUDA, the CUDA computer Capability supported by the GPU is 1.2, and the theoretical floating point operation speed is 67GFLOPS. The GPU requires that the computational process be broken up into multiple threads (threads) to be performed in parallel, each 32 threads being in a group (warp) (i.e. thread processing unit) that execute the same instructions synchronously in SIMD fashion, with different groups running independently, but can access the high-speed shared memory in units of thread blocks (blocks) comprising multiple groups.
Fig. 2 is a block diagram of a central processing unit according to an embodiment of the present invention.
As shown in fig. 2, the central processor 2 includes a working parameter acquisition section 21, a geometric model construction section 22, a model mesh segmentation section 23, a calculation task generation section 24, an imaging result display section 25, a central-side communication section 26, and a central-side control section 27.
The central communication unit 26 exchanges data between the respective components of the central processing unit 2 and between the central processing unit 2 and other devices, and the central control unit 27 controls operations of the respective components of the central processing unit 2.
The operation parameter acquiring unit 21 is configured to acquire the radar operation parameter generated by the radar device 1. In this embodiment, the working parameter acquiring unit 21 acquires a radar working parameter generated when the radar device 1 detects a Global Hawk drone.
The geometric model constructing unit 22 is configured to construct a geometric model of the electrically large and complicated target based on the radar operating parameters acquired by the operating parameter acquiring unit 21.
The model mesh subdivision section 23 is used to perform initial mesh subdivision on the surface of the geometric model to form a plurality of triangular structures.
In this embodiment, the model mesh division unit 23 performs initial mesh division on the surface of the geometric model by using a division type, a surface element division size, and a calculation frequency parameter which are set in advance in the GMSH, thereby forming a plurality of triangular structures. The triangular structures are triangles that constitute the surface of the geometric model, each having three vertices, wherein the triangular structure formed by the subdivision requires much less than a wavelength (i.e., less than 1/10 wavelength) due to computational requirements. Fig. 3 is a schematic diagram of a geometric model of a Global Hawk drone (hereinafter referred to as a Global Hawk model) after subdivision, and data obtained by performing statistics on the Global Hawk model are as follows: there are 18620 nodes, 55658 edges and 37120 bins.
The calculation task generating unit 24 is configured to sequentially generate corresponding calculation tasks from the respective triangular structures.
In this embodiment, the calculation task generating unit 24 gives the trigonometric function, the floating point addition, and the multiplication, which are mainly used in the calculation process, to the GPU as the calculation task to perform the calculation.
The imaging result display section 25 is for displaying the radar imaging result received from the graphic processor 3. In this embodiment, the central processing unit 2 is further connected to a display, and the imaging result display unit 25 displays the radar imaging result through the display.
Fig. 4 is a block diagram of a graphics processor according to an embodiment of the present invention.
As shown in fig. 4, the graphics processor 3 has a plurality of thread processing units 31, a radar imaging section 32, a graphics-side communication section 33, and a graphics-side control section 34.
The graphics-side communication unit 33 exchanges data between the components of the graphics processor 3 and between the graphics processor 3 and other devices, and the graphics-side control unit 34 controls the operations of the components of the graphics processor 3.
In this embodiment, the process based on GPU calculation includes: firstly, converting geometric data and pixel data in a geometric model of an electrical large-size complex target into fragments (namely rasterization); secondly, processing in graphic application (namely a rendering program) through vertexes and fragment programs of all triangular structures of the geometric model; the processing of the fragments can flexibly select the register combiner and the fragment program, thereby realizing complex operation. The two main programmable components in the graphics Processor 3 are Vertex processors (Vertex Processor Units) and Fragment Processor (Fragment Processor Units). During the rendering process, each vertex and fragment processor executes the same vertex and fragment program copy, and computes multiple vertex and fragment data in parallel, so as to realize the data-level parallel of Single Instruction Multiple Data (SIMD). In the graphics pipeline, vertex processing converts 3D vertices to screen 2D space, rasterization finds pixels corresponding to each primitive to generate uncolored fragments, fragment processing steps color each fragment, and finally blending each fragment to generate the final displayed image.
The thread processing unit 31 is configured to process the calculation tasks generated by the central processing unit 2.
In this embodiment, the graphics-side controller 34 allocates the thread processing units 31 according to the computation tasks received from the central processing unit 2, that is, the graphics-side controller 34 schedules between the thread groups currently loaded in the GPU, so as to execute other groups of instructions during the period when a certain group waits for the result of the computation or the memory access, thereby implementing the process of computing by fully utilizing the computation resources.
In this embodiment, the process of calculating the scattered field of the electrically large and complex target is analyzed by a Physical Optical (PO) method to obtain a corresponding calculation formula (i.e., a calculation task) in the calculation process, so that the graphic processor 3 performs calculation according to the corresponding calculation formula to obtain the scattered field parameters, and the principle is as follows:
if the frequency of the incident wave (i.e., the incident plane wave) is f, the corresponding incident wave wavelength λ = c/f, the angular wave number k =2 π/λ, and the incident wave has the form:
in the formula (I), the compound is shown in the specification,
for an incident electric field, a unit vector>
For the direction of propagation of the incident wave, is greater or less>
Is an initial value dependent on the direction of polarization of the incident wave>
Is an incident magnetic field, is greater or less>
Is the initial value of the incident magnetic field, j denotes the imaginary part, mu 0 The magnetic permeability is vacuum magnetic permeability, c is free space light velocity, and r is distance parameter.
Further, calculating the direction of the incident wave of the electrically large-size complex target in a specified direction
The far-field scattered wave generated at the upper part,
noting that the surface of the geometric model is S, the calculated gravity center of the triangular structure is
The unit normal vector is ^ 5>
Then the pair->
In a computing environment that employs PO approximation, if->
Is directly irradiated by the incident wave, and the area current density at the calculated gravity center of the triangular structure is->
Can be approximated as:
in the entire surface S, the portion irradiated with the incident wave is denoted as S 1 Then, due to the existence of the area current density, in
Field point in the direction->
(r->Infinity) generated far-field scattered waves are:
accordingly, the fringe field parameters
(i.e., the scattered electric field) is:
in the formula, epsilon 0 Is the vacuum dielectric constant.
Will be given in formula (6)
Factor e in -jkr Omitted,/4 π r, i.e. a signal which is dependent only on the frequency f, the direction of incidence->
And scattering direction>
The vector of (2) and the polarization direction vector of the scattered wave are subjected to dot product, and the result can be used for imaging.
In this embodiment, the calculation task generated by the central processing unit 2 includes data used for calculation, specifically, the data includes
Equal individual parameters, and the following array:
nodes: dividing the x, y and z coordinates of each vertex after the target surface is divided;
elems: subscripts of three vertexes of each triangular structure in the nodes array after the target surface is split;
polE _ s: each scattering direction
Two directions of polarization of the scattered waves need to be considered.
When the thread processing unit 31 acquires a calculation task and calculates a corresponding one of the triangle structures, the calculation process includes two loops, i.e., a loop for each of the triangle structures in S1 and a loop for each of the triangle structures
The cycle of (2). For each triangle structure, thread processing unit 31 reads the respective vertex coordinates from memory and calculates the center of gravity +>
The normal vector pick>
Therefore, in the embodiment, when the GPU is subjected to the parallelization preprocessing, the triangle structures are averagely allocated to the threads, and after each thread completes the preprocessing on each divided triangle structure, the preprocessing intermediate result is stored in the shared memory, and the/is stored in the shared memory>
Stored in a register so that the thread processing unit 31 can only perform a preprocessing step for reading the parameter once, thereby avoiding the waste of efficiency due to multiple reads, and then starts to execute all/every ≥ s for the triangle structure>
The scattered field contributed in the direction is calculated. Specifically, during preprocessing, each thread processes a triangle structure and stores the results in shared memory, after which a double loop (assigned to each thread in a block of threads) is performed with an outer loop pair ÷ based on>
And reading the triangle structure into a register, and performing the preprocessing on each triangle structure by inner-layer circulation.
In this embodiment, the process of processing a triangle structure by each thread processing unit 31 is specifically shown in steps S3-1 to S3-4.
And S3-1, calculating a triangular structure irradiated by incident waves in the plurality of triangular structures as a triangular structure to be calculated according to the plane wave incident field acquired in advance.
And S3-2, sequentially acquiring the triangular structures to be calculated and distributing the triangular structures to each thread of the GPU.
And step S3-3, the thread processing unit 31 obtains the calculation parameters of the triangular structure according to the distributed triangular structure to be calculated and stores the calculation parameters into the register of the GPU.
In step S3-4, the thread processing unit 31 calculates the fringe field parameters of the triangular structure to be calculated in the directions of the radar receiving antennas according to the allocated triangular structure to be calculated and the corresponding calculation parameters, and stores the fringe field parameters in the shared memory of the GPU.
In this embodiment, the scattering characteristics of the current electrically large and complex target can be synthesized by the scattered field parameters of the respective triangular structures stored in the shared memory in different directions, so that imaging is completed by a fast back projection method.
The radar imaging part 32 is used for acquiring all scattered field parameters of the electrically large complex target from the shared memory and imaging through a fast back projection imaging algorithm and a Hanning window edge filtering method so as to obtain a radar imaging result
In this embodiment, the radar imaging unit 32 performs combined imaging on all scattered fields and the geometric model by a fast back projection method (FBP), and further performs edge filtering on the geometric model by a hanning window function, thereby obtaining a radar imaging result. The parameters of the hanning window function used when the radar imaging section 32 performs the edge filtering are preset, and the effect of the edge filtering is optimized by performing a parameter adjustment test in advance. The parameter adjustment test is to select proper parameters to extract the characteristics of the large-size complex target, establish the corresponding relation between the scattering characteristics of the target and the geometric structural characteristics of the target, thereby providing image data verification for target identification and obtaining proper parameters according to the verification result.
FIG. 5 is a flowchart of a GPU acceleration method for radar imaging of electrically large targets according to an embodiment of the invention.
As shown in fig. 5, the GPU acceleration method for radar imaging of electrically large-sized targets includes the following steps:
s1, establishing a geometric model of an electrically large complex target according to radar working parameters acquired from a radar device 1, and then entering S2;
s2, performing initial mesh generation on the surface of the geometric model through GMSH and preset generation related parameters to form a plurality of triangular structures, and then entering the step S3;
s3, analyzing a calculation process for calculating a scattered field received by the electrically large-size complex target at a receiving antenna of the radar according to the triangular structure and a plane wave incident field obtained in advance through a physical optical method, so as to obtain a calculation task in the calculation process, distributing the calculation task to a GPU for parallel calculation, and then entering S4;
and S4, combining all scattered fields and the geometric model through a fast backward projection method to obtain a radar imaging result, and ending the step.
Fig. 6 is a schematic diagram illustrating comparison of radar imaging results based on a GPU acceleration method and an original CPU calculation method in the embodiment of the present invention.
FIG. 7 is a schematic diagram illustrating the effect of example calculation acceleration on 4 electrically large complex objects by the GPU acceleration system in the embodiment of the present invention.
As shown in fig. 6 and fig. 7, the effect of the acceleration method for imaging the electrically large target radar by the GPU in this embodiment is shown by taking a Global Hawk Global eagle drone and several classical examples as examples.
Figure 6 shows a comparison graph of radar imaging results of the GPU hardware acceleration algorithm followed by the original CPU-based algorithm. The computer configuration environment for performing the comparison operation is as follows: a CPU of Intel Core i5-2300 (2.8 GHz); and a block of GPU for NVIDIA GeForce210. As can be seen from the results of fig. 6, there is no difference in the operation results of the two operation methods (GPU-based and CPU-based), but the operation of the GPU is 15.85 times faster than that of the CPU in terms of the operation speed.
In addition, fig. 7 shows the effect of 4 electrically large-sized complex target calculations (Global Hawk unmanned plane, F-16 fighter, F-117 stealth plane, CVN-76 rengen aircraft carrier) on calculation acceleration by using the GPU (Speed Up cube column in the figure represents acceleration multiple), and it can be seen from the figure that the calculation Speed of the GPU is kept about 15 times the CPU Speed in the hardware environment, and the acceleration result is very obvious, and meanwhile, tests show that the radar imaging effect is also very ideal when the calculation formula is introduced into the GPU for calculation imaging by the method of the present invention.
Examples effects and effects
According to the GPU acceleration method and system for radar imaging of the electrically large-size target, provided by the embodiment, the composite scattering model of the target in a multi-interference environment is established, the CUDA program of an Nvidia TESLA system product is applied to subdivide the model into smaller blocks (namely triangular structures), so that data to be processed by an algorithm is refined, the calculation complexity of each block in calculation is further simplified through a physical optical method, and finally the data obtained by calculation is synthesized through accumulation so that the target is rapidly imaged; meanwhile, the GPU is adopted, a plurality of threads are created in the initialization parallel environment of the GPU to process each block of the target model, each thread can calculate the block and store the calculation result into the shared cache, and therefore the calculation efficiency of the full-polarization SAR imaging process is further improved.
In the embodiment, the imaging process of the target is analyzed and calculated by the physical optical method, so that the calculation complexity of the calculation process is effectively reduced, the calculation process which is the most time-consuming in the radar imaging process is simplified, the speed of the whole imaging process is accelerated, and meanwhile, when the scattered field of the electrically large and complex target is calculated by the physical optical method, the addition and multiplication of a trigonometric function and a floating point are mainly operated, so that the method is more suitable for being realized by using a GPU.
In this embodiment, since each thread stores the currently processed triangle structure in the register, the time for calling the parameter of the triangle structure in the calculation process is reduced.
In the embodiment, the geometrical model during imaging is subjected to edge filtering through the Hanning window function, so that the final imaging of the electrically large-size complex target is optimized.
The above-described embodiments are merely illustrative of specific embodiments of the present invention, and the present invention is not limited to the description of the above-described embodiments.
Claims (9)
1. A GPU acceleration method for radar imaging of an electrically large target is used for rapidly imaging an electrically large complex target detected by a radar by using a graphics processor, and is characterized by comprising the following steps:
s1, establishing a geometric model of the electrically large-size complex target according to radar working parameters acquired by the radar;
s2, performing initial mesh generation on the surface of the geometric model through GMSH to form a plurality of triangular structures;
s3, calculating a scattered field received by the electrically large-size complex target at a receiving antenna of the radar according to the triangular structure and a plane wave incident field obtained in advance through a physical optical method, and distributing a calculation task in the calculation process to a GPU for parallel calculation;
s4, combining and imaging all the scattered fields and the geometric model by a fast back projection method to obtain a radar imaging result,
wherein the step S3 comprises the following substeps:
s3-1, calculating the triangular structure irradiated by the incident wave in the plurality of triangular structures as a triangular structure to be calculated according to a plane wave incident field acquired in advance;
s3-2, sequentially acquiring the triangular structure to be calculated and distributing the triangular structure to each thread of the GPU;
s3-3, the thread acquires the calculation parameters of the triangular structure according to the distributed triangular structure to be calculated;
and S3-4, the thread calculates the scattered field parameters of the triangular structure to be calculated in the direction of each receiving antenna according to the distributed triangular structure to be calculated and the corresponding calculation parameters.
2. The GPU acceleration method for radar imaging of electrically large objects according to claim 1, characterized in that:
in step S3, the principle of calculating the scattered field by the physical optical method is as follows:
noting the frequency of the incident wave as f, the corresponding incident wave wavelength λ = c/f, and the angular wave number k =2 π/λ, the form of the incident wave is:
in the formula (I), the compound is shown in the specification,
is an incident electric field, unit vector->
For the direction of propagation of the incident wave, is greater or less>
Is an initial value dependent on the direction of polarization of the incident wave>
Is an incident magnetic field, is greater or less>
Is the initial value of the incident magnetic field, j denotes the imaginary part, mu 0 Is vacuum magnetic conductivity, c is free space light velocity, r is distance parameter,
further, calculating the direction of the electrically large-size complex target in a specified direction relative to the incident wave
The far-field scattered wave generated at the upper part,
noting the surface of the geometric model as S, the calculated center of gravity of the triangular structure as
The unit normal vector is ^ 5>
Then pair
In the context of a computing environment that employs the PO approximation, if +>
Is directly irradiated by the incident wave, the area current density at the calculated center of gravity of the triangular structure->
Can be approximated as:
in the entire surface S, a portion irradiated with the incident wave is denoted as S 1 Due to the presence of the area current density, in
Field point in the direction->
(r->Infinity) generated far-field scattered waves are: />
Accordingly, the fringe field parameters
Comprises the following steps:
in the formula, epsilon 0 Is the vacuum dielectric constant.
3. The GPU acceleration method for radar imaging of electrically large objects according to claim 1, characterized in that:
in step S3-3, the thread acquires the calculation parameters of the triangle structure to be calculated according to the allocated triangle structure to be calculated, and stores the calculation parameters in the register of the GPU.
4. The GPU acceleration method for radar imaging of electrically large objects according to claim 1, characterized in that:
in the step S2, the initial mesh generation of the surface of the geometric model is completed by setting a generation type, a surface element generation size and calculating a frequency parameter through the GMSH.
5. The GPU acceleration method for radar imaging of electrically large objects according to claim 1, characterized in that:
in step S4, when imaging is performed by the fast back-projection method, edge filtering is performed on the geometric model by using a hanning window function.
6. The GPU acceleration method for radar imaging of electrically large objects according to claim 5, characterized in that:
and adjusting and testing parameters of the Hanning window function so as to optimize the effect of the edge filtering.
7. The GPU acceleration method for radar imaging of electrically large objects according to claim 1, characterized in that:
the electrically large complex target is a target with a physical size far larger than the wavelength of the radar working frequency band in a high-frequency region, and the ratio of the physical size to the wavelength is larger than 10.
8. The GPU acceleration method for radar imaging of electrically large objects according to claim 7, characterized in that:
wherein the application range of the high-frequency region is 1-2GHz in an L wave band and 2-4GHz in an S wave band.
9. A GPU acceleration system for radar imaging of electrically large-sized targets, comprising:
the radar is used for detecting the electrically large-size complex target so as to generate radar working parameters;
the central processing unit is used for processing the radar working parameters so as to image the electrically large-size complex target;
a graphics processor having a plurality of thread processing units for computing computation tasks generated when the central processor processes the radar operating parameters,
wherein the central processing unit comprises a working parameter acquisition unit, a geometric model construction unit, a model mesh division unit, a calculation task generation unit, a central communication unit, and an imaging result display unit,
the graphics processor further has a graphics-side control section, a radar imaging section, and a graphics-side communication section,
the operating parameter acquiring unit acquires the radar operating parameter from the radar,
the geometric model constructing part constructs a geometric model of the electrically large complex target according to the radar working parameters,
the model mesh subdivision part carries out initial mesh subdivision on the surface of the geometric model to form a plurality of triangular structures,
the calculation task generation unit sequentially generates corresponding calculation tasks based on the respective triangular structures,
the central processing communication section sequentially sends the calculation tasks to the graphic processor,
the graphics processing control unit receives the computing tasks in sequence and assigns them to the idle thread processing units,
the thread processing unit calculates the corresponding triangular structure according to the calculation task so as to obtain a plurality of scattered field parameters corresponding to the triangular structure,
the radar imaging part acquires all the scattered field parameters of the electrically large-size complex target and performs imaging through a fast back projection imaging algorithm and a Hanning window edge filtering method to obtain a radar imaging result,
the graphic-side communication section transmits the radar imaging result to the central processing unit,
the imaging result display part displays the received radar imaging result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910113169.0A CN111562976B (en) | 2019-02-13 | 2019-02-13 | GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910113169.0A CN111562976B (en) | 2019-02-13 | 2019-02-13 | GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111562976A CN111562976A (en) | 2020-08-21 |
| CN111562976B true CN111562976B (en) | 2023-04-18 |
Family
ID=72071340
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910113169.0A Active CN111562976B (en) | 2019-02-13 | 2019-02-13 | GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111562976B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101430376A (en) * | 2008-12-24 | 2009-05-13 | 北京航空航天大学 | Target radar scattering cross-section pre-estimation system with graphics electromagnetic computation accelerated by index information |
| CN102901965A (en) * | 2012-09-26 | 2013-01-30 | 同济大学 | Radar imaging method of electrically large size target in ocean clutter environment |
| WO2016101279A1 (en) * | 2014-12-26 | 2016-06-30 | 中国海洋大学 | Quick detecting method for synthetic aperture radar image of ship target |
| CN106991215A (en) * | 2017-03-15 | 2017-07-28 | 西北工业大学 | A kind of high-precision graphical electromagnetic scattering computational methods based on GPU |
| WO2017223386A1 (en) * | 2016-06-22 | 2017-12-28 | Duke University | Multiple-input-multiple-output (mimo) imaging systems and methods for performing massively parallel computation |
-
2019
- 2019-02-13 CN CN201910113169.0A patent/CN111562976B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101430376A (en) * | 2008-12-24 | 2009-05-13 | 北京航空航天大学 | Target radar scattering cross-section pre-estimation system with graphics electromagnetic computation accelerated by index information |
| CN102901965A (en) * | 2012-09-26 | 2013-01-30 | 同济大学 | Radar imaging method of electrically large size target in ocean clutter environment |
| WO2016101279A1 (en) * | 2014-12-26 | 2016-06-30 | 中国海洋大学 | Quick detecting method for synthetic aperture radar image of ship target |
| WO2017223386A1 (en) * | 2016-06-22 | 2017-12-28 | Duke University | Multiple-input-multiple-output (mimo) imaging systems and methods for performing massively parallel computation |
| CN106991215A (en) * | 2017-03-15 | 2017-07-28 | 西北工业大学 | A kind of high-precision graphical electromagnetic scattering computational methods based on GPU |
Non-Patent Citations (1)
| Title |
|---|
| 李宁 ; 张民 ; 王欣 ; 聂丁 ; .船海复合场景面元化快速电磁建模方法及SAR成像仿真.国防科技大学学报.2016,(04),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111562976A (en) | 2020-08-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tao et al. | GPU-based shooting and bouncing ray method for fast RCS prediction | |
| CN104427325B (en) | Fast integration image generating method and the naked eye three-dimensional display system with user mutual | |
| US8384723B2 (en) | Method and system of rendering parallel global illumination | |
| KR101353303B1 (en) | Graphics processing unit and image processing apparatus having graphics processing unit and image processing method using graphics processing unit | |
| Feng et al. | A parallel algorithm for viewshed analysis in three-dimensional Digital Earth | |
| CN106991215A (en) | A kind of high-precision graphical electromagnetic scattering computational methods based on GPU | |
| Archirapatkave et al. | GPGPU acceleration algorithm for medical image reconstruction | |
| Chen et al. | Fast parallel algorithm for three-dimensional distance-driven model in iterative computed tomography reconstruction | |
| Zhang et al. | A deep collaborative computing based SAR raw data simulation on multiple CPU/GPU platform | |
| CN106646664B (en) | Human body microwave echoes analogy method based on GPU and system | |
| Valdetaro et al. | Understanding shader model 5.0 with directx11 | |
| Cakir et al. | An FDTD-based parallel virtual tool for RCS calculations of complex targets | |
| CN111562976B (en) | GPU (graphics processing unit) acceleration method and system for radar imaging of electrically large target | |
| Shi et al. | Geocomputation over the emerging heterogeneous computing infrastructure | |
| Stojanović et al. | Performance improvement of viewshed analysis using GPU | |
| CN109558568B (en) | Target RCS calculation method based on CUDA | |
| Tao et al. | From CPU to GPU: GPU-based electromagnetic computing (GPUECO) | |
| Nagano et al. | Massively parallel inverse rendering using Multi-objective Particle Swarm Optimization | |
| Li et al. | GPU accelerated non-illuminated graphical electromagnetic computing method with high accuracy | |
| Deng et al. | Visualization of vector data on global scale terrain | |
| Bural et al. | GPU-Accelerated Shooting and Bouncing Ray Method for Inverse Synthetic Aperture Radar Imaging | |
| Doi et al. | Development of a volume rendering system using 3D texture compression techniques on general-purpose personal computers | |
| CN113671502B (en) | Distributed compressed sensing parallel imaging method of synthetic aperture radar | |
| Fangerau et al. | Parallel volume rendering implementation on graphics cards using CUDA | |
| Islam et al. | Real-time 3D integral imaging system using a faster elemental image generation method using GPU parallel processing |
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 |