CN105607049A - Data partitioning method based on homogeneity test - Google Patents

Abstract

The invention provides a data partitioning method based on a homogeneity test, relating to a data partitioning method based on a homogeneity test. The invention aims to solve the problem that in the prior art when a cluttered environment is non-uniform in an actual condition, a classical method has low robustness performance, low target detection performance and only the homogeneity of partial data is considered and is not applicable to other classical methods. The method comprises the steps of (1) starting, (2) obtaining radar echo amplitude data, and setting carry, (3) carrying out segmentation of the radar echo amplitude data according to needs, and initializing R, (4) calculating the homogeneity level of different segments, and assigning a value to R, (5) adjusting the carry, and adding 1 to the carry, (6) searching a data element which is larger than the carry in the R, and carrying out homogeneity level calculation and going to the step (5) if the data element can be found, and going to a step (7) if the data element can not be found, (7) generating a new partition matrix tt according to the partition matrix R, and (8) ending. The method is applied to the field of signal processing.

Description

A kind of data partition method based on homogeneity test

Technical field

The present invention relates to the data partition method based on homogeneity test.

Background technology

Classical signal processing algorithm and detection technique, all suppose that whole clutter background meets same probability distribution, with skillArt development, has now found that the clutter background of the especially airborne head-down radar of clutter background of common reality does not meet same probabilityDistribute, this otherness is brought the reduction of classical way robust performance. For example space-time adaptive treatment technology (STAP), if will obtainObtain robustness, require accurately estimate covariance matrix. Covariance matrix adopts maximal possibility estimation to obtain, and estimates usedSample is generally from both sides, unit to be detected, for obtaining superperformance (loss of actual output letter miscellaneous noise ratio is no more than 3dB), trainingSample need meet following two conditions: 1, training sample and sample to be detected meet independent same distribution (I.I.D); 2, training sampleNumber is greater than 2 times of degree of freedom in system. But under actual conditions, clutter environment is heterogeneous, be due to landform itself on the one handFeature cause; Because the array structure of airborne radar causes on the other hand. In the case, be difficult to the condition that is metSample, can make STAP performance significantly reduce like this. For traditional target detection technique, there is equally such problem, passThe CFAR detection technology (CFAR) of system is applicable to the region that statistical property is identical. But under actual conditions, clutter environment right and wrongCan bring uniformly the decline of target detection performance. The people such as Jeong in 2013 have proposed CI-CFAR detector, excellent by matchingDegree test selection homogeneity unit, as with reference to unit, carries out CFAR detection, although improved the detection probability of target, the methodOnly consider the homogeney of local data, for for example STAP technology of other classical ways inapplicable.

Summary of the invention

The object of the invention is when solving prior art clutter environment to be non-homogeneous under actual conditions classical wayRobust performance is low, target detection performance is low and only consider the homogeney of local data, for other classical ways inapplicableProblem, and propose a kind of data partition method based on homogeneity test.

Above-mentioned goal of the invention is achieved through the following technical solutions:

Step 1, beginning;

Step 2, obtain radar return amplitude data, carry is set;

Step 3, to the segmentation as required of radar return amplitude data, and initialize subregion matrix R;

Step 4, the different point intersegmental homogeneity levels of calculating, to subregion matrix, R carries out assignment;

Step 5, adjustment carry, by carry+1;

In step 6, searching subregion matrix R, be greater than the data element of carry, if can find, carry out homogeneity levelCalculate and carry out again step 5, if can not find, carry out step 7;

Step 7, generate new subregion matrix tt according to subregion matrix R;

Step 8, end.

Invention effect

This method has proposed a kind of data partition algorithm based on homogeneity test, realizes according to homogeney real dataEntirety carry out subregion, make the data that are positioned at same subregion meet same probability distribution, reach classical signal processing algorithmAnd the data of detection technique meet the requirement of same probability distribution, be convenient to classical signal processing algorithm and the reality of detection techniqueExecute, realize the raising of robust performance, target detection performance improves and is applicable equally for additive method. Solve classical letterNumber Processing Algorithm and detection technique are under actual conditions, and clutter environment is the problem that meeting heterogeneous brings the decline of robust performanceAnd in CI-CFAR, K sample Anderson-Darling checks asking of the homogeney relation between data entirety not being testedTopic. Based on this method, radar return data are carried out to subregion, after subregion, utilize the data that belong to same subregion to carry out covariance squareBattle array is estimated to carry out STAP processing again, and the signal to noise ratio obtaining is somewhere than not utilizing this method directly to carry out covariance matrix againCarry out the STAP processing high 13.68dB of signal to noise ratio that a same place obtains again.

In experiment, adopt emulated data, emulated data form as shown in Figure 1. Have 1396 range gate, 128 DopplerUnit, wherein logn1 is logarithm normal distribution 1: subregion meets logarithm normal distribution, and logarithm average is 7, and logarithm standard deviation is0.7. Wbl1 is Weibull distribution 1: subregion meets Weibull distribution, and scale parameter is 400, and form parameter is 2. Logn2 be logarithm justState distributes 2: subregion meets logarithm normal distribution, and logarithm average is 5, and logarithm standard deviation is 0.4. Wbl2 is Weibull distribution 2: subregionMeet Weibull distribution, scale parameter is 200, and form parameter is 8. Emulated data dB value as shown in Figure 2.

What this emulated data was simulated is certain channel data of airborne head-down radar, and the object of these data being carried out to subregion is justIn the enforcement of follow-up space-time adaptive Processing Algorithm. Because STAP algorithm adopts 3DT-STAP fast algorithm, and the radar of simulationEcho data comprises 20 passages, should be more than or equal to 120 for meeting the data volume that the subregion of RMB criterion minimum comprises. Institute in order toUse the data partition method based on homogeneity test, adopt the reference windows type of 120*1 (range gate * Doppler unit). ?The subregion result arriving as shown in Figure 3, whole process share time 8.0852s. Subsequently each subregion is carried out to rayleigh distributed, logarithm justState distributes, the Probability Distribution Fitting of Weibull distribution, and calculating parameter size, utilizes the KS test of fitness of fot more each subregion one by oneWith the degree of closeness of three kinds of probability distribution, select the immediate probability distribution as this subregion. Obtain result with actualSituation more as shown in table 1.

Table 1 actual conditions and testing result contrast

Fig. 3 demonstration, the data partition method based on homogeneity test can be realized the differentiation that different probability is distributed. But thisIn subregion result and actual conditions incomplete same. In emulated data, range gate unit 1-698 belongs to probability distribution of the same race,But the result that the method obtains is range gate unit 1-720 belongs to probability distribution of the same race. This is because the selection of reference windows is madeBecome, K-AD homogeneity test algorithm first just supposes that in reference windows, data meet same probability distribution, i.e. every 120 distancesUnit is a segmentation sample, and in sample, data belong to same probability distribution. The 720th, from 698 nearest segmentations, so can obtainAbove-mentioned situation. Observe Fig. 4 to Fig. 7, show the Probability Distribution Fitting situation of different subregions, wherein with actual conditions exist discrepancyBe second subregion, this subregion of actual conditions meets Weibull distribution, meets logarithm normal distribution but testing result shows this distribution.The reason that this situation produces is also that reference windows is selected to cause, between the 699th to 720 range gate of each Doppler unitData are subdivided into the subregion different from self probability distribution, make the probability distribution of this subregion change simultaneously. As Fig. 4 and TuShown in 5, first subregion (Fig. 4) is subject to this impact littlely Probability Distribution Fitting not to be caused to too large impact, and the second subregion is subject toThis impact is larger, and probability distribution is originally changed, and has been turned to and has been more prone to lognormal subregion by Weibull distribution. ButThis species diversity does not affect the checking of K-AD homogeneity test subregion validity, can realize equally dividing of K-AD homogeneity testDistrict's ability.

Brief description of the drawings

Fig. 1 is emulated data form figure, and wbl1 is Weibull distribution 1, and wbl2 is Weibull distribution 2, and logn1 is lognormal1, the logn2 that distributes is logarithm normal distribution 2;

Fig. 2 is emulated data dB value figure;

Fig. 3 is emulated data K-AD subregion result figure;

Fig. 4 is subregion 1 Probability Distribution Fitting schematic diagram;

Fig. 5 is subregion 2 Probability Distribution Fitting schematic diagrames;

Fig. 6 is subregion 3 Probability Distribution Fitting schematic diagrames;

Fig. 7 is subregion 4 Probability Distribution Fitting schematic diagrames;

Fig. 8 is flow chart of the present invention.

Detailed description of the invention

Detailed description of the invention one: in conjunction with Fig. 8, present embodiment is described, present embodiment a kind of is based on homogeneity testData partition method is specifically prepared according to following steps:

Step 1, beginning;

Step 2, obtain radar return amplitude data, carry is set;

Step 3, to the segmentation as required of radar return amplitude data, and initialize subregion matrix R;

Step 4, the different point intersegmental homogeneity levels of calculating, to subregion matrix, R carries out assignment;

Step 5, adjustment carry, by carry+1;

In step 6, searching subregion matrix R, be greater than the data element of carry, if can find, carry out homogeneity levelCalculate and carry out again step 5, if can not find, carry out step 7;

Step 7, generate new subregion matrix tt according to subregion matrix R;

Step 8, end.

Detailed description of the invention two: present embodiment is different from detailed description of the invention one: obtain thunder in described step 2Reach echo amplitude data, carry is set; Detailed process is:

Obtain radar return data matrix, echo data element is plural number, gets amplitude and obtains radar return amplitude data squareBattle array, it is 0 that carry is set.

Other step and parameter are identical with detailed description of the invention one.

Detailed description of the invention three: present embodiment is different from detailed description of the invention one or two: right in described step 3Radar return amplitude data piecemeal as required, and initialize subregion matrix R; Detailed process is:

By actual demand, reference windows matrix is selected, reference windows matrix is m'*n', and m' represents reference windows rowNumber, n' represents reference windows columns; If get amplitude, to obtain radar return amplitude data matrix be m*n, and m obtains radar for getting amplitudeEcho amplitude data matrix line number, n obtains radar return amplitude data matrix columns for getting amplitude; By reference windows to getting amplitudeObtain radar return amplitude data matrix and carry out segmentation, be divided intoSection, arranges subregion matrix R, subregion matrix R line numberForColumns isInitializing all data elements of subregion matrix R is 0; M' span isN' getsValue scope isM span is 1≤m≤10000, and n span is 1≤n≤10000;

Described * is multiplication sign.

Other step and parameter are identical with detailed description of the invention one or two.

Detailed description of the invention four: present embodiment is different from one of detailed description of the invention one to three: described by referenceWindow obtains radar return amplitude data matrix and carries out segmentation getting amplitude, is divided intoSection; Detailed process is:

IfTo get amplitude and obtain in radar return amplitude data matrixBe incorporated into m elementDuan Zhong, will get amplitude and obtain in radar return amplitude data matrixArriveN element is incorporated intoDuan Zhong; Described * is multiplication sign.

Other step and parameter are identical with one of detailed description of the invention one to three.

Detailed description of the invention five: present embodiment is different from one of detailed description of the invention one to four: described step 4The different point intersegmental homogeneity levels of middle calculating, to subregion matrix, R carries out assignment; Detailed process is:

Make subregion matrix R (1,1)=1; Calculate the data that (1,1) section is got amplitude and obtain radar return amplitude data matrixDuan Yu (i, j) section is got amplitude and obtains the data segment homogeneity level of radar return amplitude data matrix, Calculate the two-symbol value with being subregion matrix R (i, j) of gained homogeneity level and step.

Other step and parameter are identical with one of detailed description of the invention one to four.

Detailed description of the invention six: present embodiment is different from one of detailed description of the invention one to five: described calculating(1,1) section is got data segment that amplitude obtains radar return amplitude data matrix and is got amplitude with (i, j) section and obtain radar return widthThe data segment homogeneity level of Value Data matrix,Detailed process is:

Calculating (1,1) section gets the data segment that amplitude obtains radar return amplitude data matrix and gets amplitude with (i, j) sectionObtain the standardized test statistics T of the data segment of radar return amplitude data matrix_kN; The 1st row the 1st row of subregion matrix RElement value is 1;

According to standardized test statistics T_kNAnd K-AD homogeneity test tables of critical values arranges homogeneity level, setting-5<T_kN≤ 0.326 homogeneity level is 1,0.326 < T_kN≤ 1.225 homogeneity levels are 2,1.225 < T_kN≤ 1.960 homogeneity waterPutting down is 3,1.960 < T_kN≤ 2.719 homogeneity levels are 4,2.719 < T_kN≤ 3.752 homogeneity levels are 5,3.752 < T_kN≤100 homogeneity levels are 6; K-AD is K sample Anderson-Darling.

Other step and parameter are identical with one of detailed description of the invention one to five.

Detailed description of the invention seven: present embodiment is different from one of detailed description of the invention one to six: described step 6In middle searching subregion matrix R, be greater than the data element of carry, if can find, carry out homogeneity level calculation and walk againRapid five, if can not find, carry out step 7; Detailed process is:

Start from (1,1) individual element of subregion matrix R, find first data element that is greater than carry in step 5Position, if can search out and position that first is greater than the data element of carry in step 5 is positioned at subregion matrix R x capableY row, carry out homogeneity level calculation, and x span isY span isCarry out stepRapid five;

If can not search out the data element that is greater than carry in step 5, perform step seven.

Other step and parameter are identical with one of detailed description of the invention one to six.

Detailed description of the invention eight: present embodiment is different from one of detailed description of the invention one to seven: described from subregion(1,1) individual element of matrix R starts, and finds the position that first is greater than the data element of carry in step 5, if can findTo and its be positioned at the capable y of x row, carry out homogeneity level calculation, detailed process is:

Determine that according to the capable y of the x of subregion matrix R row getting (x, the y) that amplitude obtains radar return amplitude data matrix locates numberAccording to section, the capable y column data of regeneration block matrix R x element is carry+1 in step 5, continues to find the data of subregion matrix RElement is greater than all positions of carry in step 5, and positional representation is (i1, j1),I1 tableShow place, position line number, j1 represents place, position columns, determines that according to the capable j1 row of i1 getting amplitude obtains radar return amplitude data(i1, the j1) of matrix locates data segment, calculates (a, b) and locate data segment and (i1, j1) and locate the homogeneity level of data segment, moreThe new capable j1 column data of subregion matrix R i1 element is carry in homogeneity level+step 5 of calculating.

Other step and parameter are identical with one of detailed description of the invention one to seven.

Detailed description of the invention nine: present embodiment is different from one of detailed description of the invention one to eight: described step 7Middlely generate new subregion matrix tt according to subregion matrix R; Detailed process is:

To obtain radar return amplitude data matrix size identical with getting amplitude for new subregion matrix tt size, and for m is capable, n is listed as;If a is not equal toB is not equal toMake in tt matrix a (m'-1)+1 capable to am', b (n'-1)+1 is to bn' rowThe value of data element equal the value of the capable b column data of a element in subregion matrix R;

If a equalsB is not equal toMake in tt matrix a (m'-1)+1 capable to m, b (n'-1)+1 arrivesThe value of the data element of bn' row equals the value of the capable b column data of a element in subregion matrix R;

If a is not equal toB equalsMake in tt matrix a (m'-1)+1 capable to am', b (n'-1)+1The value of the data element being listed as to n equals the value of the capable b column data of a element in subregion matrix R;

If a equalsB equalsMake in tt matrix a (m'-1)+1 capable to m, b (n'-1)+1 is to n rowThe value of data element equal the value of the capable b column data of a element in subregion matrix R.

Other step and parameter are identical with one of detailed description of the invention one to eight.

Adopt following examples to verify beneficial effect of the present invention:

Embodiment mono-:

A kind of data partition method based on homogeneity test of the present embodiment is specifically prepared according to following steps:

In experiment, adopt emulated data, emulated data form as shown in Figure 1. Have 1396 range gate, 128 DopplerUnit, wherein logn1 is logarithm normal distribution 1: subregion meets logarithm normal distribution, and logarithm average is 7, and logarithm standard deviation is0.7. Wbl1 is Weibull distribution 1: subregion meets Weibull distribution, and scale parameter is 400, and form parameter is 2. Logn2 be logarithm justState distributes 2: subregion meets logarithm normal distribution, and logarithm average is 5, and logarithm standard deviation is 0.4. Wbl2 is Weibull distribution 2: subregionMeet Weibull distribution, scale parameter is 200, and form parameter is 8. Emulated data dB value as shown in Figure 2.

What this emulated data was simulated is certain channel data of airborne head-down radar, and the object of these data being carried out to subregion is justIn the enforcement of follow-up space-time adaptive Processing Algorithm. Because follow-up STAP algorithm adopts 3DT-STAP fast algorithm, and simulationRadar return packet, containing 20 passages, should be more than or equal to 120 for meeting the RMB criterion data volume that subregion of minimum comprises.So utilize the data partition method based on homogeneity test, adopt the reference window of 120*1 (* Doppler unit, range gate unit)Mouth type. The subregion result obtaining as shown in Figure 3, whole process share time 8.0852s. Subsequently each subregion being carried out to Rayleigh dividesCloth, logarithm normal distribution, the Probability Distribution Fitting of Weibull distribution, calculating parameter size, utilize the KS test of fitness of fot one by one thanThe degree of closeness of more each subregion and three kinds of probability distribution, selects the immediate probability distribution as this subregion. ObtainResult and actual conditions more as shown in table 1.

Table 1 actual conditions and testing result contrast

Fig. 3 demonstration, the data partition method based on homogeneity test can be realized the differentiation that different probability is distributed. But thisIn subregion result and actual conditions incomplete same. In emulated data, range gate unit 1-698 belongs to probability distribution of the same race,But the result that the method obtains is range gate unit 1-720 belongs to probability distribution of the same race. This is because the selection of reference windows is madeBecome, K-AD homogeneity test algorithm first just supposes that in reference windows, data meet same probability distribution, i.e. every 120 distancesUnit is a segmentation sample, and in sample, data belong to same probability distribution. The 720th, from 698 nearest segmentations, so can obtainAbove-mentioned situation. Observe Fig. 4 to Fig. 7, show the Probability Distribution Fitting situation of different subregions, wherein with actual conditions exist discrepancyBe second subregion, this subregion of actual conditions meets Weibull distribution, meets logarithm normal distribution but testing result shows this distribution.The reason that this situation produces is also that reference windows is selected to cause, between the 699th to 720 range gate of each Doppler unitData are subdivided into the subregion different from self probability distribution, make the probability distribution of this subregion change simultaneously. As Fig. 4 and TuShown in 5, first subregion (Fig. 4) is subject to this impact littlely Probability Distribution Fitting not to be caused to too large impact, and the second subregion is subject toThis impact is larger, and probability distribution is originally changed, and has been turned to and has been more prone to lognormal subregion by Weibull distribution. ButThis species diversity does not affect the checking of K-AD homogeneity test subregion validity, can realize equally dividing of K-AD homogeneity testDistrict's ability.

K-AD homogeneity test is called again K sample Anderson-Darling inspection and can be understood as broad sense AD inspection, itsCan the in the situation that of the unknown of sample distribution function expression, check each sample whether with distributing, H₀:F₁＝F₂＝F₃…F_N，H₀ForNull hypothesis, wherein F₁…F_NFor each sample distribution function (CDF). For sample vector X=[B₁,B₂,…B_K], K represents data segmentNumber (K >=2), this method K gets 2,, use B_i＝{x₁,x₂,…x_nExpression data segment, n_iWithRepresent data segment B_iInterior data sampleGiven figure and empirical distribution function thereof (EDF), N=∑ n_iRepresent all data sample numbers in vectorial X, H_N(x) be all data samplesEDF. K-AD homogeneity test statistic can be expressed as:

$A_K^2=\underset{i=1}{\overset{K}{\&Sigma;}}{n}_i\underset{D_N}{\&Integral;}\frac{{\{F_{{\mathrm{in}}_i}-H_N\left(x\right)\}}^2}{H_N\left(x\right)\{1-H_N\left(x\right)\}}{\mathrm{dH}}_N\left(x\right)---\left(1\right)$

Wherein, D_N＝{x∈R:H_N(x)<1}，D_NFor sample, all samples in vectorial X are launched, arrange from small to largeTo Z₁<…<Z_NMake M_ijFor data segment B_iIn (1≤i≤K), be not more than Z_jThe data sample number of (1≤j≤N). The K-of discrete formAD homogeneity test statisticCan be expressed as:

$A_{KN}^2=\frac{1}{N}\underset{i=1}{\overset{K}{\&Sigma;}}\frac{1}{n_i}\underset{j=1}{\overset{N-1}{\&Sigma;}}\frac{{({\mathrm{NM}}_{ij}-{\mathrm{jn}}_i)}^2}{j(N-j)}---\left(2\right)$

K-AD homogeneity test statisticMathematic expectaionAnd varianceAs follows respectively:

$E\&lsqb;A_{KN}^2\&rsqb;\&ap;K-1---\left(3\right)$

${\mathrm{\&sigma;}}_N^2=\mathrm{var}\left(A_{KN}^2\right)=\frac{{\mathrm{aN}}^3+{\mathrm{bN}}^2+cN+d}{(N-1)(N-2)(N-3)}---\left(4\right)$

Wherein abcd is calculated by following formula:

$\begin{array}{c}H=\underset{i=1}{\overset{K}{\&Sigma;}}\frac{1}{n_i},h=\underset{i=1}{\overset{N-1}{\&Sigma;}}\frac{1}{i},g=\underset{i=1}{\overset{N-2}{\&Sigma;}}\underset{j=i+1}{\overset{N-1}{\&Sigma;}}\frac{1}{(N-i)j}\\ a=(4g-6)(K-1)+(10-6g)H\\ b=(2g-4)K^2+8hK+(2g-14h-4)H-8h+4g-6\\ c=(6h+2g-2)K^2+(4h-4g-4)K+(2h-6)H+4h\\ d=(2h+6)K^2-4hK\end{array}---\left(5\right)$

In formula, the sample size inverse that H is all data segments and;

Can be by weighingCheck null hypothesis H0 with the similarity degree of Gaussian distribution.Through normalizedAfter obtain T_KN, use σ_NRepresentStandard deviation.

$T_{KN}=\frac{A_{KN}^2-E\&lsqb;A_{KN}^2\&rsqb;}{{\mathrm{\&sigma;}}_N}\&ap;\frac{A_{KN}^2-(K-1)}{{\mathrm{\&sigma;}}_N}---\left(6\right)$

T_kNFor standardization test statistics;

Similar with the identification of AD family of distributions, K-AD homogeneity test is by searching tables of critical values, and whether judgement sample is with distributing.If T_KNBe less than the critical value t of Gaussian distribution under level of confidence α_K-1(α), accept null hypothesis H0, otherwise refuse former vacationIf.

Table 2K-AD homogeneity test tables of critical values t_K-1(α)

For the homogeneous property inspection of K sample AD corresponding to other K values critical value, can obtain by regression equation interpolation calculationArrive

$t_{K-1}\left(\mathrm{\&alpha;}\right)=\left\{\begin{array}{cc}2.326+\frac{1.822}{\sqrt{K-1}}-\frac{0.396}{\sqrt{K-1}}& \mathrm{\&alpha;}=0.01\\ 1.960+\frac{1.146}{\sqrt{K-1}}-\frac{0.391}{\sqrt{K-1}}& \mathrm{\&alpha;}=0.025\\ 1.645-\frac{0.678}{\sqrt{K-1}}-\frac{0.362}{\sqrt{K-1}}& \mathrm{\&alpha;}=0.05\end{array}\right.---\left(7\right)$

t_K-1(α) be K-AD inspection critical flow velocity dividing value corresponding under level of confidence α, α is level of confidence, simulation sequence aFor Weibull distribution, parameter is 3,2; Sequence b is for being respectively Weibull distribution (3,2) Gaussian distribution (3,2), Gaussian distribution (3,1), WeiUncle's distribution (2,3). The length of changing two sequences is respectively 30,60, and 90,120,150,1200. Carry out 10000 Monte Carlos realTest, calculate accuracy separately. Simulation result is as shown in table 3.

Table 3 simulation result

Simulation result shows: K-AD homogeneity test can realize in most cases distinguishes different probability distribution;But for close probability resolution capability a little less than. Wherein weber (3,2) approaches with Gauss's (3,2) average, and variance is close, K-ADHomogeneity test is lower to the correctness of the two homogeneity test. The performance of K-AD homogeneity test is along with the increase of sequence lengthAnd increase, in the time utilizing the algorithm of K-AD homogeneity test subregion, the selection of reference windows size is also very important, ensure referenceThe length of the sequence that window converts to has certain confidence level. On the whole, reference windows size is selected 120 K-AD homogeney inspectionTest and there is the ability that different probability distributes of distinguishing.

Mainly be divided into several steps based on K-AD homogeneity test partitioning algorithm:

(1) by reference windows, real data is carried out to subregion processing,

(2) different segmentations are carried out to homogeneity horizontal check, determine homogeneity level, regeneration block matrix R square according to table 2Battle array.

(3) judge whether to proceed homogeneity horizontal check according to carry information and flag, need to return to(2) step, does not need to carry out (4) step.

(4) subregion matrix R is adjusted, be convenient to follow-up other steps and use.

The present invention also can have other various embodiments, in the situation that not deviating from spirit of the present invention and essence thereof, and this areaTechnical staff is when making according to the present invention various corresponding changes and distortion, but these corresponding changes and distortion all should belong toThe protection domain of the appended claim of the present invention.

Claims (9)

1. the data partition method based on homogeneity test, is characterized in that a kind of data partition based on homogeneity testMethod is specifically carried out according to following steps:

Step 1, beginning;

Step 2, obtain radar return amplitude data, carry is set;

Step 3, to the segmentation as required of radar return amplitude data, and initialize subregion matrix R;

Step 4, the different point intersegmental homogeneity levels of calculating, to subregion matrix, R carries out assignment;

Step 5, adjustment carry, by carry+1;

In step 6, searching subregion matrix R, be greater than the data element of carry, if can find, carry out homogeneity level calculationCarry out again step 5, if can not find, carry out step 7;

Step 7, generate new subregion matrix tt according to subregion matrix R;

Step 8, end.

2. a kind of data partition method based on homogeneity test according to claim 1, is characterized in that: described step 2In obtain radar return amplitude data, carry is set; Detailed process is:

Obtain radar return data matrix, echo data element is plural number, gets amplitude and obtains radar return amplitude data matrix, establishesPutting carry is 0.

3. a kind of data partition method based on homogeneity test according to claim 2, is characterized in that: described step 3In to radar return amplitude data piecemeal as required, and initialize subregion matrix R; Detailed process is:

By actual demand, reference windows matrix is selected, reference windows matrix is m'*n', and m' represents reference windows line number,N' represents reference windows columns; If get amplitude, to obtain radar return amplitude data matrix be m*n, and m obtains radar and returns for getting amplitudeWave amplitude data matrix line number, n obtains radar return amplitude data matrix columns for getting amplitude; Obtain getting amplitude by reference windowsCarry out segmentation to radar return amplitude data matrix, be divided intoSection, arranges subregion matrix R, and subregion matrix R line number isColumns isInitializing all data elements of subregion matrix R is 0; M' span isN' valueScope isM span is 1≤m≤10000, and n span is 1≤n≤10000.

4. a kind of data partition method based on homogeneity test according to claim 3, is characterized in that: described by referenceWindow obtains radar return amplitude data matrix and carries out segmentation getting amplitude, is divided intoSection; Detailed process is:

If $1\&le;a\&le;\&lsqb;\frac{m}{m^{\&prime;}}\&rsqb;,1\&le;b\&le;\&lsqb;\frac{n}{n^{\&prime;}}\&rsqb;,$ To get amplitude and obtain in radar return amplitude data matrix $m-m^{\&prime;}*\&lsqb;\frac{m}{m^{\&prime;}}\&rsqb;+1$ Be incorporated into m elementDuan Zhong, will get amplitude and obtain in radar return amplitude data matrixArriveN element is incorporated intoDuan Zhong.

5. a kind of data partition method based on homogeneity test according to claim 4, is characterized in that: described step 4The different point intersegmental homogeneity levels of middle calculating, to subregion matrix, R carries out assignment; Detailed process is:

Make subregion matrix R (1,1)=1; Calculate (1,1) section get data segment that amplitude obtains radar return amplitude data matrix withThe data segment homogeneity level that (i, j) section is got amplitude and obtain radar return amplitude data matrix,Calculate the two-symbol value with being subregion matrix R (i, j) of gained homogeneity level and step.

6. a kind of data partition method based on homogeneity test according to claim 5, is characterized in that: described calculating(1,1) section is got data segment that amplitude obtains radar return amplitude data matrix and is got amplitude with (i, j) section and obtain radar return widthThe data segment homogeneity level of Value Data matrix,Detailed process is:

Calculating (1,1) section gets data segment that amplitude obtains radar return amplitude data matrix and gets amplitude with (i, j) section and obtainThe standardized test statistics T of the data segment of radar return amplitude data matrix_kN; The 1st row the 1st column element of subregion matrix RValue is 1;

According to standardized test statistics T_kNAnd K-AD homogeneity test tables of critical values arranges homogeneity level, setting-5 < T_kN≤ 0.326 homogeneity level is 1,0.326 < T_kN≤ 1.225 homogeneity levels are 2,1.225 < T_kN≤ 1.960 homogeneity levelsBe 3,1.960 < T_kN≤ 2.719 homogeneity levels are 4,2.719 < T_kN≤ 3.752 homogeneity levels are 5,3.752 < T_kN≤100Homogeneity level is 6.

7. a kind of data partition method based on homogeneity test according to claim 6, is characterized in that: described step 6In middle searching subregion matrix R, be greater than the data element of carry, if can find, carry out homogeneity level calculation and walk againRapid five, if can not find, carry out step 7; Detailed process is: start from (1,1) individual element of subregion matrix R, seekLook for first position that is greater than the data element of carry in step 5, if can search out and first is greater than carry in step 5The position of data element be positioned at the capable y row of subregion matrix R x, carry out homogeneity level calculation, x span isY span isExecution step five;

If can not search out the data element that is greater than carry in step 5, perform step seven.

8. a kind of data partition method based on homogeneity test according to claim 7, is characterized in that: described from subregion(1,1) individual element of matrix R starts, and finds the position that first is greater than the data element of carry in step 5, if can findTo and its be positioned at the capable y of x row, carry out homogeneity level calculation, detailed process is:

Determine that according to the capable y of the x of subregion matrix R row getting (x, the y) that amplitude obtains radar return amplitude data matrix locates data segment,The capable y column data of regeneration block matrix R x element is carry+1 in step 5, and the data element that continues searching subregion matrix R is largeAll positions of carry in step 5, positional representation is (i1, j1),I1 represents positionPlace line number, j1 represents place, position columns, determines that according to the capable j1 of i1 row getting amplitude obtains radar return amplitude data matrix(i1, j1) locates data segment, calculates (a, b) and locate data segment and (i1, j1) and locate the homogeneity level of data segment, regeneration blockThe capable j1 column data of matrix R i1 element is carry in homogeneity level+step 5 of calculating.

9. a kind of data partition method based on homogeneity test according to claim 8, is characterized in that: described step 7Middlely generate new subregion matrix tt according to subregion matrix R; Detailed process is:

To obtain radar return amplitude data matrix size identical with getting amplitude for new subregion matrix tt size, and for m is capable, n is listed as; IfA is not equal toB is not equal toMake in tt matrix a (m'-1)+1 capable to am', b (n'-1)+1 is to the number of bn' rowEqual the value of the capable b column data of a element in subregion matrix R according to the value of element;

If a equalsB is not equal toMake in tt matrix a (m'-1)+1 capable to m, b (n'-1)+1 is to bn' rowThe value of data element equal the value of the capable b column data of a element in subregion matrix R;

If a is not equal toB equalsMake in tt matrix a (m'-1)+1 capable to am', b (n'-1)+1 is to n rowThe value of data element equal the value of the capable b column data of a element in subregion matrix R;

If a equalsB equalsMake in tt matrix a (m'-1)+1 capable to m, b (n'-1)+1 is to the number of n rowEqual the value of the capable b column data of a element in subregion matrix R according to the value of element.