CN105654554A - Parallel computing method for infrared scattering characteristics of non-Lambert surface target - Google Patents

Abstract

The invention discloses a parallel computing method for the infrared scattering characteristics of a non-Lambert surface target. According to the technical scheme of the invention, the surface of a model is formed by a plurality of triangles, and a five-parameter BRDF model is selected for the spectrum BRDF modeling on the material of a target surface. According to a target coordinate system, a surface element coordinate system is established, and then an incidence vector and a scattering vector are converted from the target coordinate system to the surface element coordinate system. In this way, the scattering characteristics of the background radiance of one surface element in all incidence directions are obtained. Compared with the prior art, the computation speed is vastly improved on the premise that the complexity is finitely programmed. Meanwhile, the computation time is reduced, and the practicality of the method in the engineering application is enhanced.

Description

A kind of parallel calculating method of non-lambertian Area Objects ir scattering characteristic

Technical field

The present invention relates to the parallel calculating method of a kind of non-lambertian Area Objects ir scattering characteristic, belong to the identification of target, Detection Techniques field.

Background technology

The scattering properties of world background infrared radiation is all played an important role by non-lambertian Area Objects in fields such as remote sensing, target recognition tracking, precise guidances. Calculating first has to target is carried out Geometric Modeling, on this basis target surface is carried out Triangular patch subdivision; Bidirectional reflectance distribution function (BidirectionalReflectionDistributionFunction, BRDF) is used to characterize the scattering signatures in non-lambertian face; World background radiation and Atmospheric models, aerosol extinction model, observation geographical position, highly, time etc. are all relevant, extremely complex, calculating use MODTRAN (MODerateresolutionatmosphericTRANsmission) software be calculated; Calculating carries out blocking blanking at incident, detection direction respectively, after calculating the scattering brightness of each visible small patches respectively, is overlapped obtaining the scattering brightness of target.

Complex target in the calculating of world background infrared radiation owing to target bin is many, background radiation relates to the total space, and is distributed in very wide wave band, calculate the time longer. Therefore, speed-up computation process just seems necessary. In recent years, along with the high speed development of multi-core CPU (CentralProcessingUnit) and GPU (GraphicsProcessingUnit), general high performance parallel computation technology is widely applied in multiple subjects. But traditional acceleration based on multi-core CPU, owing to the number of core is limited, accelerates also very limited. Programming personnel is required higher by the programming on bottom layer language such as the CUDA (ComputeUnifiedDeviceArchitecture) and OpenCL (OpenComputingLanguage) based on GPU, programming relates to the distribution of equipment end memory element, release, the tissue of thread and the inter-thread synchronization operation of necessity, program more complicated, need on the basis of original serial program, carry out bigger change, have impact on efficiency. Intel Company was proposed XeonPhi many-core processor in 2011, can have at most 61 cores, and each core could support up 4 threads, and it has independent operating system, supported the OpenMP programming model based on compiling instruction being familiar with. Only need to add some compiling leads during programming in original serial code, realize version if had before based on multinuclear OpenMP, only need to recompilate and can run on Phi.It is seen that, based on the realization of Phi platform, programming is simple, greatly enhances the work efficiency of programmer.

Summary of the invention

It is an object of the invention to provide the parallel calculating method of a kind of non-lambertian Area Objects ir scattering characteristic, the problem solving current non-lambertian Area Objects ir scattering property calculation overlong time.

The technical solution used in the present invention is according to following steps:

Step 1: use 3DMAX software to complete the Geometric Modeling of target, selected target coordinate system during modeling; After having modeled, model surface is carried out the subdivision of Triangular patch, makes model surface be made up of several trianglees

Step 2: select BRDF model that targeted surface material carries out spectrum BRDF modeling

Step 3: owing to BRDF defines for small patches, therefore need to set up bin coordinate system according to target-based coordinate system: the normal direction of bin is decided to be the z-axis of bin coordinate system, z-axis and background radiation brightness incidence vector multiplication cross obtain y-axis, and y-axis obtains x-axis with z-axis multiplication cross. So definition, the angle of background radiation brightness incidence vector, scattering vector and z-axis is incidence zenith angle in bin coordinate system, scattering zenith angle, incident orientation angle perseverance is 0 ��, scattering azimuth is relative bearing, and the angle that scattering vector projection rotates to xoy plane counterclockwise around x-axis is relative bearing;

Step 4: incidence, scattering vector are transformed in bin coordinate system from target-based coordinate system, it is thus achieved that certain bin scattering to the background spoke brightness of all directions incidence

In formula,It is BRDF, the �� ' of small patches_s, �� '_i,Respectively the scattering zenith angle in bin coordinate system, incidence zenith angle, relative bearing,It isThe background spoke brightness that wavelength is �� that direction is incident, solid angle

By to wavelength integration, this bin can be obtained at ��₁����₂Scattering brightness in wave band

All of visible bin is overlapped, target can be obtained and existTotal scattering brightness in direction

In formula, (S_kcos�ȡ�_s) exist for certain binThe projected area in direction;

Further, it is contemplated that the calculating of the background radiation of different directions incidence is had concurrency by target, use a thread to calculate the target scattering to the background radiation of a direction incidence, be then overlapped parallel computation.

Compared with prior art, the invention has the beneficial effects as follows: under limitedly programming complexity, drastically increase calculating speed, decrease the calculating time, enhance the practicality in engineer applied.

Accompanying drawing explanation

Fig. 1 is the target scattering schematic diagram to world background radiation;

Fig. 2 is target-based coordinate system and bin coordinate system schematic diagram, Zuo Tuzhong,For target-based coordinate system,Represent incidence, scattering direction unit vector, �� respectively_i Respectively incidence zenith angle, azimuth, ��_s Respectively scattering zenith angle, azimuth,For small patches normal direction unit vector, �� '_i�ȡ�_sRespectively incident, scattering zenith angle in bin coordinate system; In right figureFor bin coordinate system, wherein, ${\hat{x}}^{\′}={\hat{y}}^{\′}\×{\hat{z}}^{\′},$

Fig. 3 is target Geometric Modeling and bin subdivision;

Fig. 4 be 8-12 ��m of sky, background radiation brightness;

Fig. 5 is the scattering Luminance Distribution of target under selected detection mode.

Detailed description of the invention

Below in conjunction with accompanying drawing and example, the present invention is described in detail.

The present invention is based on the parallel speed-up computation of the non-lambertian Area Objects ir scattering characteristic of many-core platform IntelXeonPhi, the computational methods following steps of the present invention:

Step 1, as it is shown in figure 1, calculate the scattering to world background infrared radiation of the non-lambertian Area Objects, it is necessary to target Geometric Modeling, it is necessary to calculate background infrared radiation brightness.

We use 3DMAX software to complete the Geometric Modeling of target, and during modeling, selected target coordinate system is as shown in left in Figure 2, and x-axis points to head, and y-axis points to side, and z-axis is straight up.After having modeled, model surface is carried out the subdivision of Triangular patch, makes model surface be made up of several trianglees, as shown in Figure 3.

Such as: target geometric model data file is divided into 3 parts, Part I provides total bin number and number of vertex, Part II is bin information, is divided into 4 row, respectively the numbering on 3 summits of bin number and this bin of composition, Part III is vertex information, it is divided into 4 row equally, respectively x, the y on summit numbering and each summit, z coordinate, can obtain the normal of each small patches according to coordinate figureWith area (S), formula is as follows respectively:

$\stackrel{\→}{AB}=(x_B-x_A,y_B-y_A,z_B-z_A)$

$\stackrel{\→}{AC}=(x_C-x_A,y_C-y_A,z_C-z_A)$

$\stackrel{\→}{BC}=(x_C-x_B,y_C-y_B,z_C-z_B)$

$\stackrel{\→}{n}=\stackrel{\→}{AB}\×\stackrel{\→}{AC}$

$\stackrel{\→}{n}=\frac{\stackrel{\→}{b}}{\left|\stackrel{\→}{n}\right|}$

$l=0.5\·\left(\right|\stackrel{\→}{AB}|+|\stackrel{\→}{AC}|+|\stackrel{\→}{BC}\left|\right)$

$S=\sqrt{l\·(l-|\stackrel{\→}{AB}\left|\right)\·(l-|\stackrel{\→}{AC}\left|\right)\·(l-|\stackrel{\→}{BC}\left|\right)}$

In formula, (x_A,y_A,z_A)(x_B,y_B,z_B)(x_C,y_C,z_C) it is summit A, B, the C coordinate in target-based coordinate system respectively,The vector that respectively 3 limits of Triangular patch are constituted,For the normal of bin, S is the area of bin.

Use sky in 8-12 ��m of atmospheric window of MODTRAN4.0 computed in software, background infrared radiation brightness, result is as shown in Figure 4. In calculating, Atmospheric models choose latitude summer, aerosol model selects marine aerosol, visibility is 23km, assume that earth's surface is sea, observation place is north latitude 35-degree, east longitude 120 ��, and observation time is 2:30 on July 1 (Greenwich Mean Time), observed altitude is 6km, calculate sky, background radiation brightness time set view zenith angle respectively as 45 ��, 135 ��. My god, background radiation brightness data file structure similar, points 2 row, first is classified as wavelength, and unit be ��m, and second is classified as corresponding spoke brightness, and unit is Watts/ (cm²*str*��m)��

Step 2, in reality, most material surfaces are non-lambertian face, and bidirectional reflectance distribution function (BRDF) can characterize the scattering signatures in non-lambertian face.

We have selected five parameter BRDF models and targeted surface material carry out spectrum BRDF modeling, defines as follows:

In formula, ��_i, ��_s,Respectively incidence zenith angle, scattering zenith angle and relative bearing,It is the distribution function of print surface facet unit normal, exp [b (1-cos ��)^a] it is the approximate description of Fresnel reflection function,Being shadowing function, in this example, we are taken as 1. k_b, k_d, k_r, a, b is undetermined parameter: k_bAnd k_dThe size of relevant (mirror reflecting component) and incoherent (diffuse-reflectance component) scattering component of reflection, relevant with the roughness on print surface and reflectance respectively, k_rThe slope distribution on reflection print surface, relevant with the roughness on print surface and grain distribution; A and b all reflects the Fresnel reflection function on print surface, relevant with the refractive index of print.

Hypothetical target surfacing is the aluminum through japanning, through modeling BRDF five parameter k_b, k_r, k_d, a, b elects 0.135,0.2775,0.2295,1.578 ,-14.2875 respectively as.

Step 3, by step 2 it can be seen that BRDF defines for small patches, we are accomplished by being transformed in the coordinate system of each small patches the incident vector in target-based coordinate system, scattering vector.

Setting up bin coordinate system according to target-based coordinate system, as in figure 2 it is shown, wherein, the normal direction of bin is decided to be the z-axis of bin coordinate system, z-axis and background radiation brightness incidence vector k_iMultiplication cross obtains y-axis, and y-axis and z-axis multiplication cross obtain x-axis, be defined such that incident orientation angle perseverance is 0 ��, scattering azimuth is relative bearing, by scattering vector k_sProject to xoy plane and be relative bearing, background radiation brightness incidence vector k around the angle that x-axis rotates counterclockwise_i, scattering vector k_sIt is incidence zenith angle in bin coordinate system, scattering zenith angle with the angle of z-axis.

Step 4, for specific scattering direction, owing to there is blocking between bin, the scattering of not all bin can be received by a detector;In like manner, for specific incident direction, not all bin can be irradiated to, therefore needs to carry out shadowing. In target-based coordinate system, if angle incident, between scattering vector and bin normal is both less than 90 ��, then this bin is visible bin. Circulate all bins and complete shadowing. Then according to step 3, incidence, scattering vector are transformed in bin coordinate system from target-based coordinate system, then certain specific visible bin pairThe background spoke brightness of direction incidence existsThe scattering spectrum scattering brightness in direction is:

In formula,It is BRDF, the �� ' of small patches_s, �� '_i,Respectively the scattering zenith angle in bin coordinate system, incidence zenith angle, relative bearing,It isThe background spoke brightness that wavelength is �� that direction is incident, solid angle

Consider that background spoke brightness is from whole day ball, is integrated the total space, the scattering brightness to the background radiation of all directions incidence of certain bin can be obtained:

By to wavelength integration, this bin can be obtained at ��₁-��₂Scattering brightness in wave band:

All of visible bin is overlapped, target can be obtained and existTotal scattering brightness in direction:

In formula, (S_kcos�ȡ�_s) exist for certain binThe projected area in direction.

Step 5, it is contemplated that the calculating of the background radiation of different directions incidence is had concurrency by target, we use a thread to calculate the target scattering to the background radiation of a direction incidence, have then been overlapped parallel computation. as shown in table 1, in the realization of serial program, we have write subprogram (Intensity) and have calculated the target scattering brightness to the background radiation of certain specific direction incidence, then different incident directions is circulated superposition and obtains the scattering brightness that target is total, when Parallel Implementation, serial program adds OpenMP lead by distribution of computation tasks to different threads, and use Intel compiler to recompilate, then the input file of executable file and necessity is copied under XeonPhi, this completes the parallel speed-up computation to the scattering properties of world background infrared radiation of the non-lambertian Area Objects based on many-core platform IntelXeonPhi.

The effect of the present invention can be further illustrated by example calculated below:

(1) simulated conditions and emulation content:

The experimental situation host side CPU of the present invention is IntelXeonE5-2620, accelerator card is IntelXeonPhi5110p, there are 60 cores, operating system is 64 CentOS, version is 6.3, the compiler used is IntelCompilericpc13.0.0, MPSS (ManycorePlatformSoftwareStack) version is 2.1.6720-19, and target is made up of 4708 bins.

Add before the circulation of incidence zenith angle OpenMP lead #pragmaompparallelforreduction (+: radiance) lead completes concurrent program, to the scatterometer point counting of different directions incident background radiation, target is fitted on multiple thread and performs.

Table 1 serial program adds OpenMP lead

Being compiled first by Intel compiler, instruction is:

$ icpc-O2-openmp-mmic-oscattering_mp_nativemain.cppsubrout ine.cpp wherein icpc is Intel compiler,-O2 is for optimizing option,-openmp need to be added owing to employing OpenMP lead,-mmic specifies the binary executable generated to run on XeonPhi, what-o specified the name of executable file generated, two cpp file respectively principal functions and subfunction realizes file.

By using virtual TCP/IP technology, accessing the Phi of PCIe slot connection just as an independent computer on network, the IP of acquiescence is 172.31.1.1.After having compiled, by scp order the binary executable (scattering_mp_native) generated, object module bin data (target.mod), the OpenMP library file (libiomp5.so) of world background infrared radiation brightness data (in_ground.datin_sky.dat) and necessity copies respective paths on XeonPhi to, and instruction is:

$scpscattering_mp_nativemic0:/home/shingo

$scptarget.modmic0:/home/shingo

$scpin_ground.datin_sky.datmic0:/home/shingo

$sudoscp/opt/parallel_studio/composerxe/lib/mic/libiomp5.somic0:/lib64

Logging in XeonPhi by ssh, run program, instruction is:

$sshmic0

$./scattering_mp_native

Consider that XeonPhi supports substantial amounts of thread, be that a circulation increases degree of parallelism background incidence zenith angle and azimuthal two-layer loop fusion, as shown in table 2

Table 2 loop fusion is optimized

The performance of XeonPhi is had a great impact by the alignment storage of data, Intel compiler provides _ and mm_malloc () and _ mm_free () function substitute malloc () and free () respectively and carry out the distribution of Dram, release, simultaneously need to use grand _ _ assume_alligned (64) to indicate the data of dynamically distribution in subfunction, to target bin normal, area in principal function, world background radiation data, the distribution of the data such as BRDF parameter, release have all carried out replacement and have reached storage memory access optimization.

(2) simulation result:

Experiment calculate along the scattering Luminance Distribution under target-based coordinate system under the back of the body-side-this detection mode in abdomen-side, use Origin8.0 software that result is displayed, as shown in Figure 5, according to Fig. 4, ground background spoke brightness is more than sky background spoke brightness, therefore the scattering brightness bottom target is bigger, has symmetry additionally, due to target, result also shows that symmetry. Table 3 is added up for Speedup Ratio of Parallel Program.

The speed-up ratio that table 3 concurrent program realizes

The invention provides a kind of non-lambertian Area Objects based on many-core platform IntelXeonPhi parallel speed-up computation method of scattering properties to world background infrared radiation. The method achieves the speed-up ratio of 21 times so that calculates target and dropped to 3 seconds in the time of the scattering brightness value of an observation angle from 65 seconds. And the program passed through realizes partly it can be seen that realizing of concurrent program is only added a line compiling on the basis of former serial program to instruct statement, and replace to be optimized just achieve good speed-up ratio by merging circulation, code. In a word, the method can be greatly enhanced calculating speed under limited programming complexity, reduces the calculating time, enhances the practicality in engineer applied.

Claims (2)

1. the parallel calculating method of a non-lambertian Area Objects ir scattering characteristic, it is characterised in that carry out according to following steps:

Step 1: use 3DMAX software to complete the Geometric Modeling of target, and model surface is carried out the subdivision of Triangular patch, make model surface be made up of several trianglees;

Selected target coordinate system during modeling, x-axis point to head, y-axis point to side, z-axis straight up, k_i, k_sRespectively background radiation brightness incidence is vectorial, scattering is vectorial, model surface carries out after having modeled the subdivision of Triangular patch, makes model surface be made up of several trianglees;

Step 2: select five parameter BRDF models that targeted surface material carries out spectrum BRDF modeling, define as follows

In formula, k_bRepresent the coherent scattering component of print surface BRDF, k_dRepresent the non-dry component of phase scattering, ��_i, ��_s,Respectively incidence zenith angle, scattering zenith angle and relative bearing,It is the distribution function of print surface facet unit normal, exp [b (1-cos ��)^a] it is the approximate description of Fresnel reflection function,It is shadowing function, k_b, k_d, k_r, a, b is undetermined parameter: k_bAnd k_dThe relevant size with incoherent scattering component of reflection, relevant with the roughness on print surface and reflectance respectively, k_rThe slope distribution on reflection print surface, relevant with the roughness on print surface and grain distribution;A and b all reflects the Fresnel reflection function on print surface, relevant with the refractive index of print;

Step 3: set up bin coordinate system according to target-based coordinate system, wherein, the normal direction of bin is decided to be the z-axis of bin coordinate system, z-axis and background radiation brightness incidence vector k_iMultiplication cross obtains y-axis, and y-axis and z-axis multiplication cross obtain x-axis, be defined such that incident orientation angle perseverance is 0 ��, scattering azimuth is relative bearing, background radiation brightness incidence vector k_i, scattering vector k_sIt is incidence zenith angle in bin coordinate system, scattering zenith angle with the angle of z-axis, by k_sProject to the angle that xoy plane rotates counterclockwise around x-axis and be relative bearing;

Step 4: incidence, scattering vector are transformed in bin coordinate system from target-based coordinate system, it is thus achieved that the scattering to the background spoke brightness of all incident directions of certain bin

By to wavelength integration, this bin can be obtained at ��₁����₂Scattering brightness in wave band:

All of visible bin is overlapped, target can be obtained and existTotal scattering brightness in direction:

In formula, (S_kcos�ȡ�_s) exist for certain binThe projected area in direction;

In formula,It is BRDF, the �� ' of small patches_s, �� '_i,Respectively the scattering zenith angle in bin coordinate system, incidence zenith angle, relative bearing,It isThe background spoke brightness that wavelength is �� that direction is incident, solid angle

2. the parallel calculating method of the non-lambertian Area Objects ir scattering characteristic described in claim 1, it is characterized in that, consider that the calculating of the background radiation of different directions incidence is had concurrency by target, use a thread to calculate the target scattering to the background radiation of a direction incidence, be then overlapped parallel computation.