CN103217676B - Radar target identification method under noise background based on bispectrum de-noising - Google Patents
Radar target identification method under noise background based on bispectrum de-noising Download PDFInfo
- Publication number
- CN103217676B CN103217676B CN201310161379.XA CN201310161379A CN103217676B CN 103217676 B CN103217676 B CN 103217676B CN 201310161379 A CN201310161379 A CN 201310161379A CN 103217676 B CN103217676 B CN 103217676B
- Authority
- CN
- China
- Prior art keywords
- mrow
- distance image
- average
- dimension
- bispectrum
- 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
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000001228 spectrum Methods 0.000 claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 230000003595 spectral effect Effects 0.000 claims description 13
- 230000006835 compression Effects 0.000 claims description 10
- 238000007906 compression Methods 0.000 claims description 10
- 238000002592 echocardiography Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000000087 stabilizing effect Effects 0.000 abstract 1
- 230000035945 sensitivity Effects 0.000 description 8
- 238000013519 translation Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 238000004088 simulation Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 201000004569 Blindness Diseases 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a radar target identification method under a noise background based on bispectrum de-noising and the method is mainly used for solving the problems in the prior art that target identification is carried out by directly utilizing a distance image which is not de-noised under the noise background so that the identification rate is poor. The realization process comprises the following steps of: normalizing a training sample and extracting the power spectrum characteristic of the normalized training sample; utilizing a power spectrum characteristic training classifier of the training sample to obtain a weight coefficient of the classifier; utilizing a bispectrum de-noising method to carry out de-noising treatment on a testing sample and recover an average distance image which is de-noised, and normalizing the distance image; extracting a power spectrum characteristic of the normalized average distance image; and utilizing the trained classifier to classify the power spectrum characteristic of the normalized de-noised average distance image, and determining a target class mark number. The radar target identification method disclosed by the invention has the advantages of stabilizing noises and recovering the average distance image which is de-noised, and can be used for radar target identification.
Description
Technical Field
The invention belongs to the technical field of radars, and relates to a target identification method, which can be used for identifying targets such as airplanes and vehicles under a noise background.
Background
The high-resolution range image is the vector sum of the target scattering point echo of the broadband radar signal projected in the radar ray direction, can provide the approximate distribution condition of the target scattering point echo in the range direction, has important value for the identification of the target, and thus becomes a hotspot of research in the field of radar automatic target identification.
In practical application, due to the randomness of the high-resolution range profile in a range window, the range profile has the problem of translation sensitivity, and therefore translation alignment is needed by directly utilizing range profile identification. Common range image alignment methods include a sliding correlation method and an absolute alignment method, the sliding correlation method is high in accuracy but complex in calculation, and the absolute alignment method is simple in calculation but low in accuracy. The bispectrum features have translation invariance, and the high-resolution range profile can be directly averaged in a bispectrum domain without performing translation alignment on the bispectrum domain, so that the calculation problem caused by range profile alignment is avoided. The bispectrum characteristic has blindness to any noise with a symmetrical probability density function and a zero mean value, and the denoising problem of the high-resolution range profile under the noise background can be solved by utilizing the inhibition effect of the bispectrum characteristic on Gaussian white noise. The bispectral feature retains all phase information of the signal except linear phase, and except position uncertainty, the original range image can be uniquely recovered from the bispectral feature.
The bispectrum has a restraining effect on white noise, but the bispectrum denoising of the complex distance image must consider the problem of initial phase sensitivity of the complex distance image. The phase of the complex range profile includes a phase caused by target rotation and a phase caused by target translation, the phase caused by target rotation includes identification information of the target and has a certain identification value, and the phase caused by target translation, that is, the initial phase of the complex range profile, is determined by the radial distance between the target and the radar and does not include identification information of the target under the condition that the wavelength of the radar signal is constant, so that the phase is not valuable for identification. For a C-band radar, if the carrier frequency is 6GHz and the wavelength is about 5cm, a translation of 5cm will cause a 4 pi change in the initial phase. This means that a small translation causes a large initial phase change, which makes the phase of the complex range profile difficult to use, which is a problem with the initial phase sensitivity of the complex range profile. If the initial phase can be corrected well, the problem of initial phase sensitivity can be eliminated.
The self-focusing algorithm used in the existing inverse synthetic aperture radar imaging can realize the initial phase correction of a complex range profile theoretically, but is influenced by various factors such as the distance between a target and a radar, the signal-to-noise ratio of the target is often low, and if the influence of noise is not removed, the self-focusing algorithm cannot realize better initial phase correction under the condition of low signal-to-noise ratio, so that the target identification effect is influenced.
Disclosure of Invention
The invention aims to provide a radar target identification method under a noise background based on bispectrum denoising, which aims to solve the problem that the prior art cannot recover the average distance image after denoising under the noise background, thereby causing the low identification rate.
The basic idea for realizing the invention is as follows: the method comprises the steps of training a linear correlation vector machine classifier by using power spectrum features of a training sample, calculating an average bispectrum of a test sample by using a group of deformation range images with range image amplitude and phase twice as much as the range image phase, recovering a denoised average range image from the average bispectrum, inputting the power spectrum features of the denoised average range image into the linear correlation vector machine classifier, and determining a target class label. The method comprises the following specific steps:
(1) taking out various targets from a radar echo database as training targets, performing pulse compression on echoes of the targets to obtain range profile samples x of the training targets, normalizing the range profile samples x, and calculating a weight coefficient W of a linear correlation vector machine classifier through the normalized range profile samples;
(2) the radar system takes a detected unknown target as a test target, and pulse compression is respectively carried out on R continuous echoes of the test target to obtain a range profile sample set of the test target: x = { X1,x2,…,xd,…,xRWhere d =1,2, …, R, xdPerforming pulse compression on the d echoes to obtain a range profile sample;
(3) obtaining a denoised average range image:
3a) sequentially taking out the distance image samples in the distance image sample set of the test target to obtain the bispectrum of the distance image samples:
3a1) for the taken range image sample xd={xd(0),xd(1),…xd(e),…xd(N-1)), generating a deformed range image using the following equation: y isd={yd(0),yd(1),…yd(e),…yd(N-1)}:
Wherein xd(e) Is xdAn e-th dimension element, e =0,1, …, N-1, N being a range profile dimension;
yd(e) is ydThe middle e-dimension element, e =0,1, …, N-1, |, represents the modulus value;
3a2) for the distance image sample xdPerforming fast Fourier transform to obtain xdSpectrum S ofdFor the deformed distance image ydPerforming fast Fourier transform to obtain ydSpectrum T ofd;
3a3) By xdSpectrum S ofd,ydSpectrum T ofdTo obtain xdBispectrum B ofdThe bispectrum BdIs an N-dimensional square matrix, BdRow p +1, column q +1 elements: b isd(p,q)=Sd(p)Sd(q)Td *(p + q) wherein Sd(p) is SdP-th dimension element of (1), Sd(q) is SdQ-th dimension element of (1), Td *(p + q) is TdConjugate of the p + q-th dimension element, p =0,1, …, N-1, q =0,1, …, N-1;
3b) forming a bispectrum feature set B = (B) by the R bispectrums obtained in the step 3a)1,B2,…Bd,…,BRCalculating average bispectral featuresWherein d =1,2, …, R, BdThe bispectrum characteristic of the d-th range profile sample is obtained;
3c) calculating the amplitude U and the phase V of the denoised average distance image spectrum according to the average bispectrum characteristic B';
3d) carrying out inverse fast Fourier transform on the amplitude U and the phase V of the denoised average distance image frequency spectrum to obtain a denoised average distance image x';
(4) and normalizing the denoised average range profile x', and obtaining the target class label by the normalized denoised average range profile and the weight coefficient W of the linear correlation vector machine classifier.
According to the method, the average bispectrum of the distance image of the test target is calculated through the deformed distance image with the amplitude as the distance image amplitude and the phase twice as large as the distance image phase, and the average bispectrum is used for recovering the average distance image after denoising.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a comparison graph of a mean range profile, a noisy mean range profile and a noiseless mean range profile after bispectral denoising;
fig. 3 is a comparison graph of the recognition rate of target recognition by the power spectrum features of the average range profile before and after de-noising.
Detailed Description
Referring to fig. 1, the implementation steps of the invention are as follows:
step 1, calculating a weight coefficient of a linear correlation vector machine classifier.
1a) And taking out various targets from the radar echo database as training targets, and performing pulse compression on the echoes of the targets by using a matched filtering method to obtain range profile samples x of the training targets.
The method of matched filtering is a common method of pulse compression.
1b) Eliminating the intensity sensitivity of a range profile sample x of a training target by adopting a 2-norm intensity normalization method to obtain a normalized training target range profile sample z:
in the practical application environment, due to the influence of various factors such as the distance between a target and a radar, the strength of distance images of the training target is different, and the characteristic is called strength sensitivity; the signal-to-noise ratio of the training target range profile sample can reach more than 30dB, and the noise part of the training target range profile sample can be basically ignored, so that for an initial range profile sample x of a training target, the intensity sensitivity of the initial range profile sample x is eliminated by adopting a 2-norm intensity normalization method, and a normalized training target range profile sample z is obtained:
where | x | is the 2-norm of the training target initial range profile sample x.
1c) And (3) expressing the normalized distance image sample z of the training target as { z (i), i =0,1, …, N-1), wherein z (i) is the ith dimension element in z, N is the dimension of the distance image sample, carrying out Fourier transform on the normalized distance image sample z of the training target, and taking the square of the modulus value to obtain the power spectrum characteristic of z:
Da=(Da(0),Da(1),…,Da(p),…Da(K-1)},
wherein D isa(p) is the training sample power spectrum feature DaP =0,1, …, K-1, K representing the dimension of the power spectral feature;
1d) training a linear correlation vector machine classifier by using the power spectrum characteristics of the training sample to obtain the weight coefficient of the linear correlation vector machine classifier: w = { ω (0), ω (1), …, ω (q), … ω (C-1) },
wherein ω (q) = { ω (0, q), ω (1, q), …, ω (K-1, q) }TQ =0,1, …, C-1, C being the number of object classes,
in the embodiment of the present invention, a linear correlation vector machine classifier is used, but the classifier is not limited to this, and a nonlinear correlation vector machine classifier, a support vector machine classifier, or the like may also be used.
And 2, obtaining a range profile sample of the test target.
The radar system takes a detected unknown target as a test target, and pulse compression is respectively carried out on R continuous echoes of the test target to obtain a test target range profile sample set X = { X =1,x2,...,xd,…,xRWhere d =1,2, …, R, xdAnd (4) performing pulse compression on the d echoes to obtain range image samples.
And 3, obtaining the denoised average range image of the test target.
3a) Sequentially taking out the range profile samples in the range profile sample set of the test target, and taking out the range profile samples xd={xd(0),xd(1),…xd(e),…xd(N-1) }, generating a deformed distance image using the following equation:
yd={yd(0),yd(1),…yd(e)…yd(N-1)}:
wherein,is denoted by ydElement of dimension e in (e =0,1, …, N-1, x)d(e) Is xdThe middle e-dimension element, e =0,1, …, N-1, where N is the distance image dimension, |, represents the modulus value;
3b) for the distance image sample xdPerforming fast Fourier transform to obtain xdSpectrum S ofdFor the deformed distance image ydPerforming fast Fourier transform to obtain ydSpectrum T ofd;
3c) By xdSpectrum S ofd,ydSpectrum T ofdTo obtain xdBispectrum B ofdThe bispectrum BdIs an N-dimensional square matrix, BdTo (1)Row p +1, column q +1 elements: b isd(p,q)=Sd(p)Sd(q)Td *(p + q) wherein Sd(p) is SdP-th dimension element of (1), Sd(q) is SdQ-th dimension element of (1), Td *(p + q) is TdConjugate of the p + q-th dimension element, p =0,1, …, N-1, q =0,1, …, N-1;
3d) forming a bispectrum feature set B = (B) by the R bispectrums obtained in the step 3c)1,B2,…Bd,…,BRCalculating average bispectral featuresWherein d is 1,2, …, R, BdThe bispectrum characteristic of the d-th range profile sample is obtained;
3c) the method comprises the following steps of solving the amplitude U and the phase V of the denoised average distance image spectrum according to the average bispectrum characteristic B', and the method comprises the following specific steps:
3c1) taking the amplitude G of the average bispectral feature B', initializing the average distance image spectral amplitude U after denoising as an N-dimensional vector with all elements of 1, initializing the deformation distance image spectral amplitude F as an N-dimensional vector with all elements of 1, and initializing the iteration number s as 0;
3c2) updating each element of U in turn: updating the q-th dimension element U (q) in U into;
3c3) updating each element of F in turn: updating the q-th dimension element F (q) in F to be:
wherein q =0,1, … N-1, G (j, q-j) is the j +1 th row of G, the q-j +1 th column element, U (j) is the j-th dimension element of U; u (q-j) is a q-j dimension element for removing U;
3c4) updating the iteration number to s +1, if the updated s is less than 100, returning to the step 3c2), otherwise, entering the step 3c5)
3c5) Taking a phase H of the average bispectral feature B', initializing an average distance image spectrum phase V after denoising as an N-dimensional vector with elements of 1, initializing a deformation distance image spectrum phase T as an N-dimensional vector with elements of 1, and initializing the number of iterations s as 0;
3c6) updating each element of V in turn, namely updating the q-th dimension element V (q) in V as:
wherein q =0,1, … N-1, H (q, j) is the q +1 th row of H, the j +1 th column element, V (j) is the jth dimension element of V, and T (q + j) is the q + j dimension element of T;
3c7) updating each element of T in turn, namely updating the q-th dimension element T (q) in T into
Wherein q =0,1, … N-1, H (j, q-j) is the j +1 th row of H, the q-j +1 th column element, V (j) is the j-th dimension element of V, and V (q-j) is the q-j-th dimension element of V;
3c8) updating the iteration number to be s = s +1, if s is less than 100 after updating, returning to the step 3c6), otherwise, entering the step 3 d);
the method described in 3c) is used in the embodiment of the present invention to find the amplitude U and the phase V of the denoised average distance image spectrum, but the method is not limited to this method, and other methods for solving the problem of incomplete data may be used.
3d) And solving the average distance image x' IFFT (UV) l after denoising according to the amplitude U and the phase V of the average distance image frequency spectrum after denoising, wherein the IFFT (question mark) l represents that inverse fast Fourier transform is performed and a module value is taken.
And 4, acquiring a target class label.
4a) Normalizing the denoised average range profile x' by adopting a 2-norm intensity normalization method to obtain a normalized denoised average range profileIn order to eliminate the problem of intensity sensitivity of the denoised average distance image x ' obtained in the step 3, wherein | x ' | is a 2-norm of the training sample x ';
4b) representing the normalized denoised average range image z 'as { z' (i), i =0,1, …, N-1}, wherein z '(i) is the ith dimension element in z', and N is the range image sample dimension; carrying out Fourier transformation on the normalized range profile sample z ', and taking the square of the module value to obtain the power spectrum characteristic of z':
Db={Db(0),Db(I),...,Db(p),...Db(K-1) }, wherein Db(p) is the power spectral feature DbP =0,1, …, K-1, K representing the dimension of the power spectral feature;
4c) characterizing the power spectrum of z' by DbInputting the data into a trained linear correlation vector machine classifier, and calculating the output of the classifier through a weight coefficient W: y = DbW due to power spectral feature DbThe vector is K-dimensional vector, the weight coefficient W is K.C-dimensional matrix, so the output y of the classifier is C-dimensional vector;
4d) and taking the class label corresponding to the maximum value element in the output y ═ y (0), y (1),.. y, (g),. y (C-1) } of the classifier as the target class label. For example, if y (0) is the maximum element in y, the target class is class 1, and if y (g) is the maximum element in y, the target class is class g +1, where g =0,1, …, C-1.
The effect of the invention can be further explained by combining simulation experiments.
1. Simulation conditions
The simulation experiment is carried out in MATLAB7.0 software, the used data are actual data acquired by radar, and the simulation experiment comprises three types of airplanes: jack-42 aircraft, trophy aircraft, and An-26 aircraft, the data is very noisy and negligible.
2. Content of the experiment
Experiment one: in order to test the denoising effect of the range profile, 10 continuously observed ampere-26 high-resolution range profile samples are selected, 0dB of noise is added to the 10 range profile samples, and the denoised range profile is obtained. The method of the present invention is used to obtain the denoised average range profile, and the denoised average range profile is compared with the denoised average range profile and the denoised average range profile, with the result shown in fig. 2. Wherein
FIG. 2(a) shows a noise-free average range profile;
FIG. 2(b) shows the average range image after denoising;
fig. 2(c) shows the average distance image of the noise.
Experiment two: in order to verify the improvement of the recognition performance of the denoised range profile, 50 range profile samples of an Jack-42 airplane, a trophy airplane and an-26 airplane are respectively selected, and 150 samples are used as range profile samples of a training target; 12000 distance image samples of the Jack-42 airplane, 20000 distance image samples of the trophy airplane and 20000 distance image samples of the An-26 airplane are selected to be 52000 samples, and 0dB, 5dB, 10dB, 15dB and 20dB of noise are respectively added into the samples to be used as the distance image samples of the test target.
Denoising every 10 range image samples in the range image samples of 52000 test targets by using the method of the invention to obtain 1 denoised average range image sample, finally obtaining 5200 denoised average range image samples, and performing a target identification experiment by using the power spectrum characteristics of the 5200 denoised average range image samples; and then directly averaging every 10 distance image samples in the 52000 distance image samples of the test target to obtain 1 noise-added average distance image sample, finally obtaining 5200 noise-added average distance image samples, and performing a target identification experiment by using the power spectrum characteristics of the 5200 noise-added average distance image samples. The results are shown in FIG. 3.
In fig. 3, a solid line indicates a change in the recognition rate of the target recognition using the power spectrum feature of the denoised average range profile with the signal-to-noise ratio, and a dotted line indicates a change in the recognition rate of the target recognition using the power spectrum feature of the denoised average range profile with the signal-to-noise ratio.
2. And (3) analyzing an experimental result:
as can be seen from fig. 2: 1) the range of the noise area is basically zero, and the range image of the noise area is smooth; 2) the amplitude of the signal area is slightly changed, the amplitude of the noise area is integrally improved, and the distance image of the noise area is not smooth; 3) in the average distance image of the dual-spectrum denoising, the amplitude of a signal area is slightly changed, the maximum peak value is slightly higher than the distance image without noise, the smaller peak value is slightly lower than the distance image without noise, the amplitude of a noise area is integrally and slightly improved, but the improvement is not obvious, and the distance image of the noise area is not smooth.
As can be seen from fig. 3: when the signal-to-noise ratio is 0dB, the power spectrum characteristic of the average distance image after noise addition is completely unidentifiable, and the power spectrum characteristic of the average distance image after noise removal is improved by 30 percentage points; when the signal-to-noise ratio is within the range of 5dB to 15dB, the recognition rate of the power spectrum characteristic of the average distance image after denoising is 10-15 percentage points higher than that of the power spectrum characteristic of the average distance image after denoising; the recognition rate of the power spectrum characteristic of the average distance image after noise removal is recovered to nearly 80% when the signal-to-noise ratio is 20dB, and the recognition rate of the power spectrum characteristic of the average distance image after noise removal is slightly higher than 80%.
In conclusion, the invention can realize the noise suppression of the distance image under the condition of low signal-to-noise ratio and has better identification effect.
Claims (2)
1. A radar target identification method under a noise background based on bispectrum denoising comprises the following steps:
(1) taking out various targets from a radar echo database as training targets, performing pulse compression on echoes of the targets to obtain range profile samples x of the training targets, and performing 2-norm intensity normalization on the range profile samples x to obtain normalized training samplesNormalization, where x is the training distance2-norm of outlier sample x;
(2) calculating a weight coefficient W of the linear correlation vector machine classifier through the normalized distance image sample:
(2a) carrying out Fourier change on the normalized training sample z, and taking the square of the modulus value to obtain the power spectrum characteristic of z: da={Da(0),Da(1),…,Da(p),…Da(K-1) }, wherein Da(p) is the training sample power spectrum feature DaP-th dimension element of (1), p is 0, …, K-1, K is the dimension of the power spectrum characteristic;
(2b) power spectral feature D using normalized training samplesaTraining a linear correlation vector machine classifier to obtain a weight coefficient of the linear correlation vector machine classifier: w ═ ω (0), ω (1), …, ω (q), … ω (C-1) }, where:
ω(q)={ω(0,q),ω(1,q),…,ω(K-1,q)}Tq-dimension element of weight coefficient, q is 0,1, …, C-1, C is training target class number;
(3) the radar system takes a detected unknown target as a test target, and pulse compression is respectively carried out on R continuous echoes of the test target to obtain a range profile sample set of the test target: x ═ X1,x2,...,xd,…,xRWhere d is 1,2, …, R, xdPerforming pulse compression on the d echoes to obtain a range profile sample;
(4) obtaining a denoised average range image:
4a) sequentially taking out the distance image samples in the distance image sample set of the test target to obtain the bispectrum of the distance image samples:
4a1) for the taken range image sample xd={xd(0),xd(1),…xd(e),…xd(N-1) }, generating a deformed distance image using the following equation: y isd={yd(0),yd(1),…yd(e),…yd(N-1)}:
Wherein xd(e) Is xdThe middle e-dimension element, e is 0,1, …, N-1, N is the distance image dimension;
yd(e) is ydThe middle e-dimension element, e is 0,1, …, N-1, |, represents the module value;
4a2) for the distance image sample xdPerforming fast Fourier transform to obtain xdSpectrum S ofdFor the deformed distance image ydPerforming fast Fourier transform to obtain ydSpectrum T ofd;
4a3) By xdSpectrum S ofd,ydSpectrum T ofdTo obtain xdBispectrum B ofdThe bispectrum BdIs an N-dimensional square matrix, BdRow p +1, column q +1 elements: b isd(p,q)=Sd(p)Sd(q)Td *(p + q) wherein Sd(p) is SdP-th dimension element of (1), Sd(q) is SdQ-th dimension element of (1), Td *(p + q) is TdConjugate of the element of dimension p + q, p-0, 1, …, N-1, q-0, 1, …, N-1;
4b) forming a bispectrum characteristic set B ═ B from the R bispectrums obtained in the step 4a)1,B2,…Bd,…,BRCalculating average bispectral featuresWherein d is 1,2, …, R, BdThe bispectrum characteristic of the d-th range profile sample is obtained;
4c) calculating the average distance image spectral amplitude U and phase V after denoising according to the average bispectrum characteristic B';
4d) carrying out inverse fast Fourier transform on the amplitude U and the phase V of the denoised average distance image frequency spectrum to obtain a denoised average distance image x';
(5) normalizing the denoised mean range image x', and obtaining a target class label by the normalized denoised mean range image and a weight coefficient W of a linear correlation vector machine classifier:
5a) carrying out Fourier change on the normalized denoised average range profile, and taking the square of the module value to obtain the normalized denoised average range profilePower spectrum characteristics of (a): db={Db(0),Db(1),…,Db(p),…Db(K-1) }, wherein Db(p) is the power spectral feature DbP-th dimension element of (1), p ═ 0, …, K-1, K denotes the dimension of the power spectrum feature;
5b) normalizing the power spectrum characteristic D of the denoised average range profilebInputting the data into a trained linear correlation vector machine classifier, and calculating the output of the classifier through a weight coefficient W: y ═ DbW due to power spectral feature DbThe vector is K-dimensional vector, the weight coefficient W is K.C-dimensional matrix, so the output y of the classifier is C-dimensional vector, and C is the number of training target classes;
5c) and determining a target class label according to the output of the classifier, namely taking the class label corresponding to the maximum value element in y as the target class label.
2. The method for identifying radar targets under the noise background according to claim 1, wherein the step 4c) of calculating the spectral amplitude U and the phase V of the denoised average range image is performed according to the following steps:
4c1) taking the amplitude G of the average bispectral feature B', initializing the average distance image spectral amplitude U after denoising as an N-dimensional vector with all elements of 1, initializing the deformation distance image spectral amplitude F after denoising as an N-dimensional vector with all elements of 1, and initializing the iteration number s as 0;
4c2) updating each element of U in turn: updating the q-th dimension element U (q) in U to be:
wherein q is 0,1, … N-1, G (q, j) is the q +1 th row of G, the j +1 th column element, F (q + j) is the q + j th dimension element of F, U (j) is the j th dimension element of U;
4c3) updating each element of F in turn: updating the q-th dimension element F (q) in F to be:
wherein q is 0,1, … N-1, G (j, q-j) is the j +1 th row of G, the q-j +1 th column element, U (j) is the j-th dimension element of U; u (q-j) is a q-j dimension element for removing U;
4c4) updating the iteration number to s-s +1, if the updated s is less than 100, returning to the step 4c2), otherwise, entering the step 4c5),
4c5) taking a phase H of the average bispectral feature B', initializing an average distance image spectrum phase V after denoising as an N-dimensional vector with elements of 1, initializing a deformed distance image spectrum phase T after denoising as an N-dimensional vector with elements of 1, and initializing the number of iterations s as 0;
4c6) updating each element of V in turn: updating q-th dimension element V (q) in V to
Where q is 0,1, … N-1, H (q, j) is the q +1 th row of H, the j +1 th column element, V (j) is the jth dimension element of V, and T (q + j) is the q + j dimension element of T;
4c7) updating each element of T in turn: updating q-th dimension element T (q) in T to
Where q is 0,1, … N-1, H (j, q-j) is the j +1 th row of H, the q-j +1 th column element, V (j) is the jth dimension element of V, and V (q-j) is the q-j dimension element of V;
4c8) and updating the iteration number to s-s +1, if s is less than 100 after updating, returning to the step 4c6), and otherwise, entering the step 4 d).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310161379.XA CN103217676B (en) | 2013-05-06 | 2013-05-06 | Radar target identification method under noise background based on bispectrum de-noising |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310161379.XA CN103217676B (en) | 2013-05-06 | 2013-05-06 | Radar target identification method under noise background based on bispectrum de-noising |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103217676A CN103217676A (en) | 2013-07-24 |
CN103217676B true CN103217676B (en) | 2014-08-20 |
Family
ID=48815632
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310161379.XA Active CN103217676B (en) | 2013-05-06 | 2013-05-06 | Radar target identification method under noise background based on bispectrum de-noising |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103217676B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106597400A (en) * | 2016-11-15 | 2017-04-26 | 北京无线电测量研究所 | Ground moving vehicle target classification and recognition method and system based on high-resolution distance image |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103454623B (en) * | 2013-09-08 | 2015-06-17 | 西安电子科技大学 | Radar target recognition method based on noise backgrounds of plural AGC models |
CN105223552B (en) * | 2015-09-15 | 2018-03-06 | 上海无线电设备研究所 | Disturbance identification method based on the identification of radio spectrum line |
CN106338722A (en) * | 2016-08-25 | 2017-01-18 | 电子科技大学 | High-resolution radar one-dimension range profile target identification method based on multi-order samples |
CN108416393A (en) * | 2018-03-16 | 2018-08-17 | 西安电子科技大学 | A kind of target identification optimization method based on high-resolution lattice image |
CN109448750B (en) * | 2018-12-20 | 2023-06-23 | 西京学院 | Speech enhancement method for improving speech quality of biological radar |
CN109670484B (en) * | 2019-01-16 | 2022-03-25 | 电子科技大学 | Mobile phone individual identification method based on bispectrum characteristics and lifting tree |
CN111243237A (en) * | 2020-01-16 | 2020-06-05 | 珠海格力电器股份有限公司 | Drowning monitoring method, equipment, device and storage medium |
CN112731399B (en) * | 2020-12-22 | 2023-06-02 | 中国人民解放军空军预警学院 | Broadband radar aircraft target frequency identification method based on two-dimensional sparse structure |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102279390A (en) * | 2011-05-06 | 2011-12-14 | 西南交通大学 | Intra-pulse modulation and recognition method of low signal-to-noise radar radiation source signal |
CN102916917A (en) * | 2012-09-25 | 2013-02-06 | 哈尔滨工程大学 | Individual identification method of FSK (frequency-shift keying) signal based on slice bi-spectrum and wavelet transformation |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050195103A1 (en) * | 2004-01-13 | 2005-09-08 | Davis Dennis W. | Phased arrays exploiting geometry phase and methods of creating such arrays |
-
2013
- 2013-05-06 CN CN201310161379.XA patent/CN103217676B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102279390A (en) * | 2011-05-06 | 2011-12-14 | 西南交通大学 | Intra-pulse modulation and recognition method of low signal-to-noise radar radiation source signal |
CN102916917A (en) * | 2012-09-25 | 2013-02-06 | 哈尔滨工程大学 | Individual identification method of FSK (frequency-shift keying) signal based on slice bi-spectrum and wavelet transformation |
Non-Patent Citations (2)
Title |
---|
基于多模态平稳序列建模的雷达高分辨距离像有限样本目标识别方法;王鹏辉等;《光学学报》;20120531;全文 * |
王鹏辉等.基于多模态平稳序列建模的雷达高分辨距离像有限样本目标识别方法.《光学学报》.2012, |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106597400A (en) * | 2016-11-15 | 2017-04-26 | 北京无线电测量研究所 | Ground moving vehicle target classification and recognition method and system based on high-resolution distance image |
Also Published As
Publication number | Publication date |
---|---|
CN103217676A (en) | 2013-07-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103217676B (en) | Radar target identification method under noise background based on bispectrum de-noising | |
CN107633486B (en) | Structural magnetic resonance image denoising method based on three-dimensional full-convolution neural network | |
CN107301432B (en) | Self-adaptive radiation source modulation identification method based on time-frequency analysis | |
Du et al. | Robust statistical recognition and reconstruction scheme based on hierarchical Bayesian learning of HRR radar target signal | |
CN104280724B (en) | High Resolution Range Profile Identification of Radar method based on scattering point coupling | |
CN107392123B (en) | Radio frequency fingerprint feature extraction and identification method based on coherent accumulation noise elimination | |
CN112087774A (en) | Communication radiation source individual identification method based on residual error neural network | |
Vitale et al. | A new ratio image based CNN algorithm for SAR despeckling | |
Huang et al. | Radar waveform recognition based on multiple autocorrelation images | |
Zhu et al. | A probe-feature for specific emitter identification using axiom-based grad-CAM | |
CN104021400B (en) | The SAR image target recognition method of rarefaction representation is composed based on fraction | |
CN103454623B (en) | Radar target recognition method based on noise backgrounds of plural AGC models | |
Ilhan et al. | Automatic directional masking technique for better sperm morphology segmentation and classification analysis | |
CN104463245B (en) | A kind of target identification method | |
CN115169406A (en) | Instantaneous phase fingerprint feature enhancement method based on empirical mode decomposition | |
CN109766899B (en) | Physical feature extraction and SVM SAR image vehicle target recognition method | |
CN104021399A (en) | SAR object identification method based on range profile time-frequency diagram non-negative sparse coding | |
CN109272515B (en) | Unsupervised SAR image segmentation method based on high-order multi-scale CRF | |
CN104463222B (en) | Polarimetric SAR image classification method based on feature vector distribution characteristics | |
Wang et al. | Synthetic aperture radar image segmentation using fuzzy label field‐based triplet Markov fields model | |
CN113095353B (en) | Underdetermined blind source separation method based on AP clustering | |
Yu et al. | Radar‐based human activity recognition using denoising techniques to enhance classification accuracy | |
US7180442B1 (en) | Target indentification method using cepstral coefficients | |
Chan et al. | Speckle noise reduction in SAR images using information theory | |
Khalili et al. | A deep learning approach for Wi-Fi based people localization |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |