CN103913724B - Based on the clutter suppression method of priori landform cover data - Google Patents

Abstract

The invention belongs to radar clutter suppression technology field, particularly based on the clutter suppression method of priori landform cover data.Should comprise the following steps based on the clutter suppression method of priori landform cover data: utilize airborne radar to obtain echo data, obtain L original training sample; By L maximum for object matching coefficient _out1individual original training sample is rejected, and is once rejected rear training sample; After once rejecting in training sample, according to covariance matrix and the R of each training sample ₀similarity, draw secondary reject after training sample, after rejecting according to secondary, the expectation of covariance matrix of training sample, calculates adaptive weight, then carries out filtering according to adaptive weight to the echo data of pending range gate.

Description

Based on the clutter suppression method of priori landform cover data

Technical field

The invention belongs to radar clutter suppression technology field, particularly based on the clutter suppression method of priori landform cover data, for carrying out clutter recognition to the echo data of radar in non-homogeneous clutter environment.

Background technology

In the Airborne Pulse Doppler Radar in modern times, the phased array antenna that Space time pattern technology can prompt become because of extensive choice for use wave beam and digital signal samples, treatment technology become possibility.This technology produces by utilizing clutter statistical characteristics the oblique recess mated with clutter, effective raising airborne radar is to the detection perform of moving-target, both can be applied in airborne radar, also can be applied in battle space awareness radar and airborne fire control radar detect ground at a slow speed target time.

Because clutter statistical characteristics is with the non-homogeneous clutter environment of distance change, airborne radar is difficult to obtain independent identically distributed clutter sample data.The clutter that heterogeneity shows as different distance ring has different clutter spectrum, and one of its reason is the difference of the reflection characteristic of ground scatter body.Another important aspect is the interference of some the isolated clutters comprising echo signal, and these clutters just appear at other distance range gate, and the disturbance calculated weight vector is clearly.Space-time adaptive process needs the statistical property of the clutter plus noise estimating pending range gate with independent identically distributed training sample.When clutter plus noise sample is non-homogeneous, the statistical property of training sample can not meet independent same distribution.This will cause the covariance matrix estimated by training sample different with the statistical property of the background clutter plus noise of pending range gate, make space-time adaptive process can not effective Background suppression clutter plus noise, even can make the power drop of target, cause radar detedtion probability not high.

For the heterogeneity of above-mentioned clutter, develop a lot of method and avoided or weaken.Such as suppose that clutter data is the staging treating of local uniform, sliding window method, sliding hole method, recursive algorithm etc. in little distance range.Also usefully select the larger sample architecture covariance matrix of clutter power and form weight vector, darker depression can be formed along two-dimentional clutter spectrum like this, strengthening system to the adaptive faculty of clutter.And for comprising the situation of echo signal and some isolated clutters in clutter, conventional non-homogeneous detection method has broad sense inner product (GIP) method and adaptive power residue (APR) method.Broad sense inner product approach can only pick out the Uniform Sample in training sample, when the background clutter plus noise of pending range gate and the clutter plus noise of most of training sample are not independent same distribution, the training sample selected by broad sense inner product can not estimate the statistical property of the background of pending range gate, causes the handling property of space-time adaptive process to decline.Adaptive power residual basis is relatively more rough, can not choose moving-target very accurately, there is false dismissal and false-alarm, and need to repeat to select, calculated amount is large.These the two kinds methods selecting training sample only utilize original training sample statistical property, and not with the data characteristic of pending range gate, select Uniform Sample is not immediate with the statistical property of the background clutter plus noise of pending range gate, so the Background statistic characteristic of pending range gate can not be estimated well, the clutter of space-time adaptive process is caused to remain larger, this can cause false-alarm on the one hand, can raise constant false alarm rate detection threshold on the other hand, reduces detection probability.

Summary of the invention

The object of the invention is to propose the clutter suppression method based on priori landform cover data.The present invention can carry out clutter recognition to the echo data of radar in non-homogeneous clutter environment.

For realizing above-mentioned technical purpose, the present invention adopts following technical scheme to be achieved.

Clutter suppression method based on priori landform cover data comprises the following steps:

S1: utilize airborne radar obtain echo data, in radar return data, choose the echo data of L range gate, L be greater than 1 natural number; Using L the original training sample of the echo data of L range gate as correspondence; Draw the object matching coefficient of each original training sample, in L original training sample, by L maximum for object matching coefficient _out1individual original training sample is rejected, and is once rejected rear training sample; L _out1for the once rejecting number of setting;

S2: after once rejecting in training sample, draws the covariance matrix of each training sample; Draw the covariance matrix R of the echo data of pending range gate ₀, draw covariance matrix and the R of each training sample ₀similarity, the covariance matrix of g training sample and R ₀similarity be D _g, g gets 1 to L1, L1=L-L _out1; At D ₁to D _l1in choose maximum L _out2individual numerical value, after once rejecting in training sample, by described L _out2the L that individual numerical value is corresponding _out2individual training sample is rejected, and obtains secondary and rejects rear training sample; L _out2for the secondary of setting rejects number;

S3: show that secondary rejects the expectation of the covariance matrix of rear training sample

$\hat{R}=\frac{1}{L2}\underset{h=1}{\overset{L2}{\mathrm{\Σ}}}{x}_h{x}_h^H$

Wherein, L2=L-L _out1-L _out2, x _hrepresent that secondary rejects the echo data of h training sample in rear training sample, the conjugate transpose of H representing matrix;

Then basis calculate adaptive weight wherein, μ is normalization coefficient, s _tfor steering vector during target empty; Then according to adaptive weight w _optcarry out filtering to the echo data of pending range gate, the result exported after filtering is: the clutter recognition result of the echo data of pending range gate.

Feature of the present invention and further improvement are:

In step sl, range gate corresponding to each original training sample differs with pending range gate and is no more than A range gate, and A is the natural number of setting.

In step sl, in L original training sample, the object matching coefficient p of l original training sample _lfor:

$p_l=\left|\frac{s_t^H{x}_l}{\sqrt{x_l^H{x}_l{s}_t^H{s}_t}}\right|$

Wherein, l gets 1 to L, x _lit is the echo data of l original training sample; The conjugate transpose of H representing matrix.

In step s 2, after once rejecting in training sample, the covariance matrix R of g training sample _gfor

$R_g=\underset{g\left(i\right)=1}{\overset{N_g}{\mathrm{\Σ}}}{\mathrm{\ρ}}_{g\left(i\right)}{\left({\mathrm{\ρ}}_{g\left(i\right)}\right)}^H(s_{a,g\left(i\right)}\⊗s_{b,g\left(i\right)}){(s_{a,g\left(i\right)}\⊗s_{b,g\left(i\right)})}^H$

Wherein, ρ _gi () represents the echo data amplitude once rejecting g training sample i-th scattering point in rear training sample, s _{a, g (i)}represent the spatial domain steering vector of the target that g training sample i-th scattering point is corresponding in training sample after rejecting with first time, s _{b, g (i)}represent the time domain steering vector of the target that g training sample i-th scattering point is corresponding in training sample after rejecting with first time; N _gfor once rejecting in rear training sample the number of scattering point corresponding to g training sample; the direct product of representing matrix;

Draw the covariance matrix R of the echo data of pending range gate ₀:

$R_0=\underset{i=1}{\overset{N_0}{\mathrm{\Σ}}}{\mathrm{\ρ}}_i{\left({\mathrm{\ρ}}_i\right)}^H(s_{a,i}\⊗s_{b,i}){(s_{a,i}\⊗s_{b,i})}^H$

Wherein, ρ _irepresent the echo data amplitude of pending range gate, s _a,ito represent in the echo data of pending range gate the spatial domain steering vector of the target that i-th scattering point is corresponding, s _b,ito represent in the echo data of pending range gate the time domain steering vector of the target that i-th scattering point is corresponding;

Then covariance matrix and the R of g training sample is determined according to following formula ₀similarity D _g:

$D_g={\left|\right|R_0^{-1}{R}_g-I\left|\right|}_F$

Wherein, I be with with the unit matrix of dimension, ‖ ‖ _fthe quadratic sum of all elements in representing matrix.

In step s3, adaptive weight w is being drawn _optafterwards, the clutter recognition result of the echo data of pending range gate is drawn according to following formula wherein, x _l0represent the echo data of pending range gate.

Beneficial effect of the present invention is: the matching degree of the present invention when selecting training sample first according to training sample and goal orientation vector, reject likely by the training sample of target stains, the problem that when so just can avoid space-time adaptive process, target disappears mutually, keep target output constant, then the training sample the poorest with pending range gate similarity is rejected according to the covariance matrix similarity of training sample and pending range gate, thus pick out and do not make training sample containing target with the most similar sample of pending range gate, estimate that obtaining adaptive weight can send out and adaptive power residual basis clutter reduction better than broad sense inner product by the method, improve detection probability.When the statistical property of pending range gate is different from most training sample, the present invention can pick out and the immediate training sample of pending range gate statistical property from training sample, thus the clutter of this pending range gate can be suppressed well, reduce false-alarm.

Accompanying drawing explanation

Fig. 1 is the geometric relationship schematic diagram of even linear array radar and target;

Fig. 2 is the process flow diagram of the clutter suppression method based on priori landform cover data of the present invention;

Fig. 3 is the distance-Doppler image of the single passage of airborne radar in emulation experiment one;

Fig. 4 is the range Doppler image of the echo data of the single spatial domain passage obtained according to landform cover data and clutter model simulation in emulation experiment one;

Fig. 5 is the filtered distance-Doppler image of space-time adaptive process choosing sample in emulation experiment one based on broad sense Law of Inner Product;

Fig. 6 be in emulation experiment one the present invention through the filtered distance-Doppler image of space-time adaptive process;

Fig. 7 is the mean value contrast schematic diagram of the filtering output power that in emulation experiment one, two kinds of methods obtain;

Fig. 8 is the filtered distance-Doppler image of space-time adaptive process choosing sample in emulation experiment two based on broad sense Law of Inner Product;

Fig. 9 be in emulation experiment two the present invention through the filtered distance-Doppler image of space-time adaptive process;

Figure 10 is the mean value contrast schematic diagram of the filtering output power that in emulation experiment two, several method obtains.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described:

First the echo data model of airborne radar is introduced.The receiving array of airborne radar is even linear array, and the geometric relationship of even linear array radar and target as shown in Figure 1.Carrier aircraft is parallel to ground flying with speed v along X-axis positive dirction, and the angle of radar front and carrier aircraft speed is α, target relative to the position angle of radar, the angle of pitch, cone angle and air line distance be respectively θ, φ and R _l.In Fig. 1, be described for any one range gate, in this range gate, target echo data s can be expressed as:

s=ρ ₀s _t

Wherein, ρ ₀represent the amplitude of target data, steering vector when representing target empty, represent operation of direct product symbol (being multiplied by the element of the same position of two matrixes), s _a0and s _b0represent spatial domain steering vector and the time domain steering vector of target respectively, can be expressed as:

$s_{a0}=\frac{1}{\sqrt{N}}{[1,e^{{j2\mathrm{\πu}}_t},...,e^{{j2\mathrm{\π}(N-1)u}_t}]}^T$

$s_{b0}=\frac{1}{\sqrt{K}}{[1,e^{{j2\mathrm{\πv}}_t},...,e^{{j2\mathrm{\π}(K-1)v}_t}]}^T$

Wherein, N is the array number of the receiving array of airborne radar, and K is the umber of pulse in airborne radar coherent processing inteval, u _tfor normalization spatial frequency, d is the array element distance of the receiving array of airborne radar, and λ is that airborne radar transmits wavelength, ν _tfor normalization Doppler frequency, f _rfor the pulse repetition rate that airborne radar transmits.

The clutter echoed signal of this range gate can be expressed as:

$c=\underset{q=1}{\overset{N_c}{\mathrm{\Σ}}}{\mathrm{\ρ}}_q{s}_{\mathrm{aq}}\⊗s_{\mathrm{bq}}$

In formula, N _cfor the number of current distance door clutter scattering point, s _aqfor the spatial domain steering vector of current distance family status q clutter scattering point, s _bqfor the time domain steering vector of current distance family status q clutter scattering point, ρ _qfor the echo data amplitude of current distance family status q clutter scattering point.ρ _qrelevant with the scattering coefficient of clutter scattering point, clutter scattering point area, radar transmission power, the factor such as radar emission directional diagram and receiving pattern.

Whether according to containing echo signal, echo has two kinds of hypothesis forms, and wherein under a kind of hypothesis, echo data comprises clutter information and noise, and under another kind hypothesis, echo data comprises clutter, noise and echo signal.

With reference to Fig. 2, it is the process flow diagram of the clutter suppression method based on priori landform cover data of the present invention.Should comprise the following steps based on the clutter suppression method of priori landform cover data:

S1: utilize airborne radar obtain echo data, near pending range gate, select L range gate, L be greater than 1 natural number.In this L range gate, range gate corresponding to each original training sample differs with pending range gate and is no more than A range gate, and A is the natural number of setting.Using L the original training sample of the echo data of an above-mentioned L range gate as correspondence.

An above-mentioned L original training sample, likely by target stains, at this moment, can cause target to disappear mutually with required adaptive weight when space-time adaptive process, make the output power of target be less than real power, reduce the detection probability of radar.So first we will reject by the original training sample of target stains.In the embodiment of the present invention, reject by the training sample of target stains according to the matching degree of steering vector when original training sample and target empty.

Draw the object matching coefficient of each original training sample, in L original training sample, the object matching coefficient p of l original training sample _lfor:

$p_l=\left|\frac{s_t^H{x}_l}{\sqrt{x_l^H{x}_l{s}_t^H{s}_t}}\right|$

Wherein, l gets 1 to L, x _lit is the echo data of l original training sample; The conjugate transpose of H representing matrix.

In L original training sample, by L maximum for object matching coefficient _out1individual original training sample is rejected, and (now, the number of training sample is L1=L-L once to be rejected rear training sample _out1); L _out1for the once rejecting number of setting; L _out1<L.

S2: after once rejecting in training sample, draws the covariance matrix of each training sample; Draw the covariance matrix R of the echo data of pending range gate ₀, draw covariance matrix and the R of each training sample ₀similarity, the covariance matrix of g training sample and R ₀similarity be D _g, g gets 1 to L1; At D ₁to D _l1in choose maximum L _out2individual numerical value, after once rejecting in training sample, by described L _out2the L that individual numerical value is corresponding _out2individual training sample is rejected, and obtains secondary and rejects rear training sample; L _out2for the secondary of setting rejects number; L _out2<L1.Be described as follows:

After once rejecting in training sample, the covariance matrix R of g training sample _gfor

$R_g=\underset{g\left(i\right)=1}{\overset{N_g}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_{g\left(i\right)}{\left({\mathrm{\&rho;}}_{g\left(i\right)}\right)}^H(s_{a,g\left(i\right)}\&CircleTimes;s_{b,g\left(i\right)}){(s_{a,g\left(i\right)}\&CircleTimes;s_{b,g\left(i\right)})}^H$

Wherein, ρ _gi () represents the echo data amplitude once rejecting g training sample i-th scattering point in rear training sample, ρ _gi () is drawn by airborne radar backscattering coefficient model, airborne radar clutter model and landform cover data.S _{a, g (i)}represent the spatial domain steering vector of the target that g training sample i-th scattering point is corresponding in training sample after rejecting with first time, s _{b, g (i)}represent the time domain steering vector of the target that g training sample i-th scattering point is corresponding in training sample after rejecting with first time; N _gfor once rejecting in rear training sample the number of scattering point corresponding to g training sample; the direct product of representing matrix.

Draw the covariance matrix R of the echo data of pending range gate ₀:

$R_0=\underset{i=1}{\overset{N_0}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_i{\left({\mathrm{\&rho;}}_i\right)}^H(s_{a,i}\&CircleTimes;s_{b,i}){(s_{a,i}\&CircleTimes;s_{b,i})}^H$

Wherein, ρ _irepresent the echo data amplitude of pending range gate, s _a,ito represent in the echo data of pending range gate the spatial domain steering vector of the target that i-th scattering point is corresponding, s _b,ito represent in the echo data of pending range gate the time domain steering vector of the target that i-th scattering point is corresponding;

Then covariance matrix and the R of g training sample is determined according to following formula ₀similarity D _g:

$D_g={\left|\right|R_0^{-1}{R}_g-I\left|\right|}_F$

Wherein, I be with with the unit matrix of dimension, ‖ ‖ _fthe quadratic sum of all elements in representing matrix.

S3: show that secondary rejects the expectation of the covariance matrix of rear training sample

$\hat{R}=\frac{1}{L2}\underset{h=1}{\overset{L2}{\mathrm{\&Sigma;}}}{x}_h{x}_h^H$

Wherein, L2=L-L _out1-L _out2, x _hrepresent that secondary rejects the echo data of h training sample in rear training sample, the conjugate transpose of H representing matrix;

Then basis calculate adaptive weight wherein, μ is normalization coefficient, and such as μ equals 1, s _tfor steering vector during target empty; Then according to adaptive weight w _optcarry out filtering to the echo data of pending range gate, the result exported after filtering is: the clutter recognition result of the echo data of pending range gate.Particularly, adaptive weight w is being drawn _optafterwards, the clutter recognition result y of the echo data of pending range gate is drawn according to following formula _l0, wherein, x _l0represent the echo data of pending range gate.

Advantage of the present invention can be verified further by following emulation experiment:

Emulation experiment one:

1) experiment parameter and experiment condition:

This experiment is verified according to certain measured data, the receiving array of airborne radar is positive side battle array, and front antenna adopts the linear array of 2 × 11, is separated with 128 pulses between each coherent processing, useful range gate number is 500, peak transmitted power is 1.5kW, and fire pulse width (before pulse pressure) is 50.4 μ s, instant bandwidth 800kHz, the pulse repetition rate that airborne radar transmits is 1984Hz, array element interval d is 0.109m, and radar carrier frequency is 1.24GHz, and range resolution is 120m.

2) experiment content and interpretation of result

With reference to Fig. 3, it is the distance-Doppler image of the single passage of airborne radar in emulation experiment one.With reference to Fig. 4, it is the range Doppler image of the echo data of single spatial domain passage obtained according to landform cover data and clutter model simulation in emulation experiment one.In figs. 3 and 4, the energy of gray-scale value representative data, gray-scale value is larger, and energy is higher.For verifying the performance of this method, test as a comparison with the space-time adaptive process that the broad sense inner product of classics selects training sample herein.In emulation experiment one, target is added in No. 250 distance family status No. 100 Doppler's passages.With reference to Fig. 5, for choosing the filtered distance-Doppler image of space-time adaptive process of sample in emulation experiment one based on broad sense Law of Inner Product.With reference to Fig. 6 be in emulation experiment one the present invention through the filtered distance-Doppler image of space-time adaptive process.In fig. 5 and fig., the energy of gray-scale value representative data, gray-scale value is larger, and energy is higher.Significantly can find out that conventional art broad sense inner product approach clutter reduction residue is more by comparison diagram 5 and Fig. 6, and clutter reduction residue of the present invention is relatively little many, the latter's effect is more a lot of than the former, can see there is a target at No. 250 range gate place of the 100th Doppler's passage.In order to contrast the superior of the inventive method and prior art further, Fig. 7 provides explanation and compares.With reference to Fig. 7, it is the mean value contrast schematic diagram of the filtering output power that two kinds of methods in emulation experiment one obtain.The mean value (7.007dB) of filtering output power of the present invention 4.002dB lower than the mean value (11.09dB) selecting the filtering output power of training sample method based on broad sense inner product as seen from Figure 7, effective suppression of background clutter plus noise is conducive to the detection of target, improve the detection probability of target, illustrate that the present invention can obtain better clutter recognition performance than traditional broad sense inner product approach.We it can also be seen that, for discrete strong clutter point, the inventive method can be good at suppressing, and this can reduce false-alarm probability.

Emulation experiment two:

1) experiment parameter and experiment condition:

In emulation experiment two, radar system parameters is identical with emulation experiment one, just changes target component.Target is added in No. 100 doppler cells of No. 150 range gate.With reference to Fig. 8, for choosing the filtered distance-Doppler image of space-time adaptive process of sample in emulation experiment two based on broad sense Law of Inner Product.With reference to Fig. 9 be in emulation experiment two the present invention through the filtered distance-Doppler image of space-time adaptive process.In Fig. 8 and Fig. 9, the energy of gray-scale value representative data, gray-scale value is larger, and energy is higher.From Fig. 8 with Fig. 9 can draw the conclusion identical with emulation experiment one.With reference to Figure 10, it is the mean value contrast schematic diagram of the filtering output power that several method in emulation experiment two obtains.Mean value (6.24dB) as seen from Figure 10 based on the radar training sample selection of priori landform cover data and the filtering output power of covariance matrix estimation method compares the low 3.769dB of filtering output power mean value (9.009dB) selecting training sample method based on broad sense inner product.

By above-mentioned experiment and analysis, we may safely draw the conclusion: when clutter plus noise is non-homogeneous, chooses the self-adaptation power clutter reduction effect that training sample obtains better than classic method by the inventive method.

Obviously, those skilled in the art can carry out various change and modification to the present invention and not depart from the spirit and scope of the present invention.Like this, if these amendments of the present invention and modification belong within the scope of the claims in the present invention and equivalent technologies thereof, then the present invention is also intended to comprise these change and modification.

Claims (4)

1. based on the clutter suppression method of priori landform cover data, it is characterized in that, comprise the following steps:

S1: utilize airborne radar obtain echo data, in radar return data, choose the echo data of L range gate, L be greater than 1 natural number; Using L the original training sample of the echo data of L range gate as correspondence; Draw the object matching coefficient of each original training sample, in L original training sample, by L maximum for object matching coefficient _out1individual original training sample is rejected, and is once rejected rear training sample; L _out1for the once rejecting number of setting;

S2: after once rejecting in training sample, draws the covariance matrix of each training sample; Draw the covariance matrix R of the echo data of pending range gate ₀, draw covariance matrix and the R of each training sample ₀similarity, the covariance matrix of g training sample and R ₀similarity be D _g, g gets 1 to L1, L1=L-L _out1; At D ₁to D _l1in choose maximum L _out2individual numerical value, after once rejecting in training sample, by described L _out2the L that individual numerical value is corresponding _out2individual training sample is rejected, and obtains secondary and rejects rear training sample; L _out2for the secondary of setting rejects number;

S3: show that secondary rejects the expectation of the covariance matrix of rear training sample

$\hat{R}=\frac{1}{L2}\underset{h=1}{\overset{L2}{\&Sigma;}}{x}_h{x}_h^H$

Wherein, L2=L-L _out1-L _out2, x _hrepresent that secondary rejects the echo data of h training sample in rear training sample, the conjugate transpose of H representing matrix;

Then basis calculate adaptive weight w _opt: wherein, μ is normalization coefficient, s _tfor steering vector during target empty; Then according to adaptive weight w _optcarry out filtering to the echo data of pending range gate, the result exported after filtering is: the clutter recognition result of the echo data of pending range gate.

2. as claimed in claim 1 based on the clutter suppression method of priori landform cover data, it is characterized in that, in step sl, range gate corresponding to each original training sample differs with pending range gate and is no more than A range gate, and A is the natural number set.

3., as claimed in claim 1 based on the clutter suppression method of priori landform cover data, it is characterized in that, in step sl, in L original training sample, the object matching coefficient p of l original training sample _lfor:

$p_l=\left|\frac{s_t^H{x}_l}{\sqrt{x_l^H{x}_l{s}_t^H{s}_t}}\right|$

Wherein, l gets 1 to L, x _lit is the echo data of l original training sample; The conjugate transpose of H representing matrix.

4., as claimed in claim 1 based on the clutter suppression method of priori landform cover data, it is characterized in that, in step s3, drawing adaptive weight w _optafterwards, the clutter recognition result y of the echo data of pending range gate is drawn according to following formula _l0, wherein, x _l0represent the echo data of pending range gate.