CN108490414A - A kind of radar target identification method based on time-frequency distributions instantaneous frequency edge feature - Google Patents
A kind of radar target identification method based on time-frequency distributions instantaneous frequency edge feature Download PDFInfo
- Publication number
- CN108490414A CN108490414A CN201810429367.3A CN201810429367A CN108490414A CN 108490414 A CN108490414 A CN 108490414A CN 201810429367 A CN201810429367 A CN 201810429367A CN 108490414 A CN108490414 A CN 108490414A
- Authority
- CN
- China
- Prior art keywords
- frequency
- time
- distributions
- matrix
- feature
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Abstract
The invention discloses a kind of radar target identification methods based on time-frequency distributions instantaneous frequency edge feature, belong to Technology of Radar Target Identification field, are related to Radar High Range Resolution time-frequency distributions characteristic, the technology of instantaneous frequency local edge.This method first, pre-processes radar range profile, eliminates it and translates sensibility, amplitude sensitive;Then, the Wigner Ville distributions of Range Profile are found out, then the cross-interference terms in Wigner Ville distributions are inhibited by time-frequency reordering technique;Next, seeking the instantaneous frequency edge feature of the time-frequency distributions after time-frequency is reset, it is used in combination and Principal Component Analysis Algorithm obtains the eigenmatrix of instantaneous frequency edge feature;Finally, classified to test sample using k nearest neighbor grader.The information of the time and frequency variation of Range Profile had both been utilized in this method simultaneously in this way, in turn avoided directly improving object recognition rate using the time-frequency distributions problem larger as information redundancy, calculation amount caused by identification feature.
Description
Technical field
The invention belongs to Technology of Radar Target Identification field, it is related to Radar High Range Resolution time-frequency distributions characteristic, instantaneous
The technology of frequency edges characteristic.
Background technology
The frequency of stationary signal is single and constant, distributed constant or distribution law not changed letter at any time
Number.Fourier transformation is carried out to stationary signal, its frequency domain information can be analyzed, but also frequency can be incited somebody to action by inverse Fourier transform
Domain signal transforms to time domain.With that can change over time, frequency is meeting for unstable signal (transient signal) its frequency and statistical property
Variation.Fourier transformation gives signal frequency distribution situation on frequency spectrum, and provides the temporal information of frequency, i.e. transformation results
It only shows without some frequency content, but does not provide the information that corresponding frequencies change over time, and this information is exactly
Reflect most basic and most critical the property of non-stationary signal.This also means that it is necessary to study non-stationary signal m- frequencies at any time
The information that rate changes simultaneously.
The energy of non-stationary signal is the function of T/F.Numerous Time-Frequency Distribution Algorithms have studied non-stationary signal
Energy and T/F physical relationship, more profoundly describe the variation characteristic of signal.With simple time domain or frequency domain
Change information is compared, and time-frequency distributions give the change information of signal energy m- frequency at any time simultaneously, is provided more comprehensive
Recognize the information of non-stationary signal.
Wigner-Ville distribution gives the time-frequency distributions of signal, in numerous research signal Time-Frequency Distribution Algorithms
The time-frequency locality highest of Wigner-Ville distributions, but there are cross-interference terms for the algorithm.Cross-interference terms are seriously done
The analysis to signal time-frequency distributions has been disturbed, error analysis results are frequently can lead to.It is special using Wigner-Ville distribution as identification
Sign, will certainly influence recognition effect.By carrying out time-frequency rearrangement to Wigner-Ville distribution, can effectively inhibit therein
Time-frequency cross-interference terms, and then the analysis of the time-frequency distributions conducive to signal, the Wigner-Ville distribution after being reset with time-frequency are made
For identification feature, it will improve discrimination.
Different target time-frequency distributions than it is more similar on the whole when, using the time-frequency distributions of suppressing crossterms as identification
Feature recognition not rate will not be too ideal, therefore considers certain features of extraction time-frequency distributions, and time-frequency distributions have abundant wink
Shi Tezheng.Comparing from data structure angle, temporal characteristics are often one-dimensional vector, and time-frequency distributions are two-dimensional matrixes,
When thus the former is as identification, calculation amount will greatly reduce.Although contained data volume reduces, due to different target
Certain temporal characteristics difference from time-frequency distributions it is obvious that individually extract, and as the foundation of identification, often obtains preferably
Recognition effect.
Radar High Range Resolution (HRRP) is typical non-stationary signal, traditional based on HRRP recognition methods, is mostly
Using the simple time domain or frequency domain of HRRP as identification feature, although also there is the time-frequency distributions feature based on HRRP to be identified
, but it is HRRP time-frequency distributions features redundancy, computationally intensive, recognition effect needs to be further increased.
Invention content
The invention discloses a kind of radar HRRP target identification methods based on time-frequency distributions instantaneous frequency edge feature, it
Using the instantaneous frequency edge feature of time-frequency distributions as identification feature, therefore the time of Range Profile had both been utilized in this method simultaneously
With the information of frequency variation, in turn avoid directly larger using time-frequency distributions as information redundancy, calculation amount caused by identification feature
Problem, improve object recognition rate.
The invention discloses a kind of radar target identification methods based on time-frequency distributions instantaneous frequency edge feature:Its feature
It is, this approach includes the following steps:
Step 1, training Range Profile sample Y is pre-processed, eliminates translation sensibility and amplitude sensitive:
Training Range Profile sample matrix Y=[y1,y2,…,yi,…,yN]∈Rn×N, yi∈RnIt is n dimensional vectors apart from decent
This, i=1,2 ..., N, Rn×NIndicate n × N-dimensional matrix;
In formula, τ is time delay, and f is frequency;
Step 3, time-frequency rearrangement is carried out to sample Wigner-Ville distribution, inhibits cross-interference terms in distribution:
WithThe time after Wigner-Ville distribution carries out time-frequency rearrangement and frequency, t and f are indicated respectively
Indicate that the time of Wigner-Ville distribution and frequency, t ', f ' respectively represent time, frequency respectively;Time-frequency distributions after rearrangement
It is denoted as Wy′;
Step 4, the instantaneous frequency edge feature of the time-frequency distributions after resetting is sought:
Step 5, the nonlinear characteristic contained in instantaneous frequency edge spy is linearized:
Using nonlinear function φ by instantaneous frequency edge featureIt is mapped to higher dimensional space, mapping result is
And then the nonlinear characteristic contained in instantaneous frequency edge spy is linearized;For i-th of training sample Range Profile instantaneous frequency
Edge feature, i=[1,2 ..., N];
Step 6, according to kernel function principle, the nuclear matrix K of frequency edges Feature Mapping result is constructed:
By kernel function principle, frequency edges Feature Mapping resultInner product withKernel function κ value is equal, i.e.,
Build kernel matrix K:
Step 7, by the nuclear matrix K of training sample, projective transformation matrix X is trained:
Using two-dimensional principal component analysis algorithm, dimensionality reduction is carried out to nuclear matrix K, obtains projective transformation matrix X;X=[x1,
x2,…,xd]∈RN×d, x1,x2,…,xd∈RNFor the feature vector corresponding to kernel matrix K characteristic values, d is feature vector
Number;
Step 8, test sample is sought according to the projective transformation matrix X acquiredProjection properties matrix
Wherein,
Step 9:Using k nearest neighbor grader, classify to test sample, identifies target.
The two-dimentional time-frequency distributions of different target are more similar on the whole, and as identification feature, not only discrimination is relatively low, but also
Operand is bigger, and the present invention uses the frequency edges feature of time-frequency distributions as identification feature, more due to frequency edges feature
It can reflect the geometry and physical characteristic of different target, and it is one-dimensional vector, not only obtains higher discrimination in this way,
And reduce calculation amount.
Description of the drawings
Fig. 1 is implementation steps flow chart of the present invention;
Fig. 2 be " 260 width Range Profiles of amp- 26 ", abscissa indicate per range from range cell, ordinate indicate distance
The number of picture;
Fig. 3 be " diploma " 260 width Range Profiles, abscissa indicate per range from range cell, ordinate indicate distance
The number of picture;
Fig. 4 be " Ya Ke -42 " 260 width Range Profiles, abscissa indicate range from range cell, ordinate indicate often away from
Number from picture.
Specific implementation mode
Step 1:Training sample Y is built, and carries out eliminating translation sensibility and amplitude sensitive processing.
The HRRP data that the present invention uses are to use high-resolution wideband radar outfield measured data, packet by certain domestic research institute
Include " Ya Ke -42 " medium-sized jet airplane, " diploma " miniature jet machine, " totally three kinds of targets such as amp- 26 " light propeller airplane
HRRP data.The HRRP data of each target are the matrixes of 520 rows 256 row.Wherein, the representative of line number 520 has 520 ranges
From picture, columns 256, which represents every width Range Profile, 256 range cells.
In order to probe into the ratio (instruction surveys ratio) of number of training and test sample number to the influence degree of recognition effect, difference
It is surveyed in instruction than being 1:3、1:2、1:1、2:1、3:Experiment is identified under the conditions of 1.Ratio 1 is surveyed with instruction:1, i.e., each target is by 260 width
Range Profile is for training, training sample Y=[y1,y2,…,y780], 260 width Range Profiles for testing for following the steps below.
The translation sensibility that Range Profile y is eliminated using gravity model appoach calculates the center of gravity of Range Profile first:
Then, by the central region of Range Profile to center of gravity W.
(2) l is used2Norm normalized cumulant eliminates the amplitude sensitive of Range Profile as y.
Step 2, the Wigner-Ville distribution W of training sample Y is calculatedy(t,f):
In formula, τ is time delay, and f is frequency.
Step 3, time-frequency rearrangement is carried out to sample Wigner-Ville distribution, inhibits cross-interference terms in distribution:
WithFor the center of gravity of Wigner-Ville distribution, t and the center that f is Wigner-Ville distribution, t ', f ' are respectively
Time, frequency are represented, the time-frequency distributions after rearrangement are denoted as Wy′。
Step 4, the instantaneous frequency edge feature of the time-frequency distributions after resetting is sought:
Step 5, the nonlinear characteristic contained in instantaneous frequency edge spy is linearized:
Using nonlinear function φ by instantaneous frequency edge featureIt is mapped to higher dimensional space, mapping result is
And then the nonlinear characteristic contained in instantaneous frequency edge spy is linearized.For i-th of training sample Range Profile instantaneous frequency
Edge feature, i=[1,2 ..., 780].
Step 6, according to kernel function principle, the nuclear matrix K of frequency edges Feature Mapping result is constructed:
By kernel function principle, frequency edges Feature Mapping resultInner product withCertain kernel function κ value is equal, i.e.,
Build kernel matrix K:
Step 7, by the nuclear matrix K of training sample, projective transformation matrix X is trained:
Using two-dimensional principal component analysis algorithm, dimensionality reduction is carried out to nuclear matrix K, obtains projective transformation matrix X.X=[x1,
x2,…,x20]∈R780×20For projective transformation matrix, x1,x2,…,x20∈R780For the spy corresponding to kernel matrix K characteristic values
Sign vector, the number of feature vector take 20;
Step 8, test sample is soughtProjection properties matrix
Wherein,Neighbour's number parameter of KNN graders is set as 3.
Amp- 26 that the present invention uses, the diploma, Ya Ke -42 three kinds of Aircraft Targets are set forth in Fig. 2, Fig. 3, Fig. 4
HRRP.As can be seen from the figure its different width HRRP of target even of the same race, difference is also bigger, this, which is also meaned that, directly makes
Use HRRP as identification feature, recognition effect will not be too ideal.It is special it is therefore desirable to further excavate the time-frequency wherein contained
Sign is more favorable for identification classification because time-frequency characteristics give the time domain and frequency domain information of HRRP simultaneously.Three width figures are compared may be used also
To find, range cell number that every width HRRP of amp- 26 and Ya Ke -42 includes to be more than that the every width HRRP of the diploma include away from
From number of unit.
Table 1 is based on Range Profile initial data object recognition rate;The table first row is meant that respectively with HRRP, HRRP
Wigner-Ville is distributed, the instantaneous frequency edge feature of the Wigner-Ville distribution of HRRP is identification feature.In order to make pair
Convincingness is had more than effect, projective transformation is carried out using the highest pivot analysis algorithm of each feature recognition rate is made.Wherein, feature
HRRP and instantaneous frequency edge feature carry out projection properties transformation using KPCA algorithms, and Wigner-Ville distribution uses bilateral two
It ties up principal component analysis and carries out projection properties transformation.The first row represents instruction and surveys ratio.It is discrimination (%) in table.
Table -1
Claims (1)
1. a kind of radar target identification method based on time-frequency distributions instantaneous frequency edge feature:It is characterized in that, this method packet
Include following steps:
Step 1, training Range Profile sample Y is pre-processed, eliminates translation sensibility and amplitude sensitive:
Training Range Profile sample matrix Y=[y1,y2,…,yi,…,yN]∈Rn×N, yi∈RnFor n dimensional vector Range Profile samples, i
=1,2 ..., N, Rn×NIndicate n × N-dimensional matrix;
In formula, τ is time delay, and f is frequency;
Step 3, time-frequency rearrangement is carried out to sample Wigner-Ville distribution, inhibits cross-interference terms in distribution:
WithThe time after Wigner-Ville distribution carries out time-frequency rearrangement and frequency, t and f difference are indicated respectively
The time and frequency, t ', f ' for indicating Wigner-Ville distribution respectively represent time, frequency;Time-frequency distributions after rearrangement are denoted as
Wy′;
Step 4, the instantaneous frequency edge feature of the time-frequency distributions after resetting is sought:
Step 5, the nonlinear characteristic contained in instantaneous frequency edge spy is linearized:
Using nonlinear function φ by instantaneous frequency edge featureIt is mapped to higher dimensional space, mapping result isIn turn
The nonlinear characteristic contained in instantaneous frequency edge spy is linearized;For i-th of training sample Range Profile instantaneous frequency edge
Feature, i=[1,2 ..., N];
Step 6, according to kernel function principle, the nuclear matrix K of frequency edges Feature Mapping result is constructed:
By kernel function principle, frequency edges Feature Mapping resultInner product withKernel function κ value is equal,
I.e.
Build kernel matrix K:
Step 7, by the nuclear matrix K of training sample, projective transformation matrix X is trained:
Using two-dimensional principal component analysis algorithm, dimensionality reduction is carried out to nuclear matrix K, obtains projective transformation matrix X;X=[x1,x2,…,
xd]∈RN×d, x1,x2,…,xd∈RNFor the feature vector corresponding to kernel matrix K characteristic values, d is the number of feature vector;
Step 8, test sample is sought according to the projective transformation matrix X acquiredProjection properties matrix
Wherein,
Step 9:Using k nearest neighbor grader, classify to test sample, identifies target.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810429367.3A CN108490414B (en) | 2018-05-08 | 2018-05-08 | Radar target identification method based on time-frequency distribution instantaneous frequency edge characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810429367.3A CN108490414B (en) | 2018-05-08 | 2018-05-08 | Radar target identification method based on time-frequency distribution instantaneous frequency edge characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108490414A true CN108490414A (en) | 2018-09-04 |
CN108490414B CN108490414B (en) | 2021-07-06 |
Family
ID=63354074
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810429367.3A Active CN108490414B (en) | 2018-05-08 | 2018-05-08 | Radar target identification method based on time-frequency distribution instantaneous frequency edge characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108490414B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110133600A (en) * | 2019-06-17 | 2019-08-16 | 电子科技大学 | A kind of lifting airscrew physical parameter extracting method |
CN110738248A (en) * | 2019-09-30 | 2020-01-31 | 朔黄铁路发展有限责任公司 | State perception data feature extraction method and device and system performance evaluation method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050021260A1 (en) * | 2003-06-26 | 2005-01-27 | Robertson Naysen Jesse | Use of I2C programmable clock generator to enable frequency variation under BMC control |
CN102645649A (en) * | 2012-05-14 | 2012-08-22 | 重庆大学 | Radar target recognition method based on radar target range profile time-frequency feature extraction |
CN103033801A (en) * | 2012-12-18 | 2013-04-10 | 中国民航大学 | Air maneuvering target detection method based on virtual time sampling and Wigner-Ville distribution (WVD) transformation |
CN103913736A (en) * | 2014-04-11 | 2014-07-09 | 四川大学 | Laser micro Doppler parameter estimation method based on spectrogram rearrangement |
CN104021400A (en) * | 2014-06-16 | 2014-09-03 | 电子科技大学 | SAR image target recognition method based on fractional spectrum sparse representation |
CN105783974A (en) * | 2014-12-25 | 2016-07-20 | 中国科学院声学研究所 | Chirp signal detection, parameter estimation method, and system thereof |
CN106370403A (en) * | 2016-08-22 | 2017-02-01 | 南京信息工程大学 | Instant frequency estimation method based on edge detection |
-
2018
- 2018-05-08 CN CN201810429367.3A patent/CN108490414B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050021260A1 (en) * | 2003-06-26 | 2005-01-27 | Robertson Naysen Jesse | Use of I2C programmable clock generator to enable frequency variation under BMC control |
CN102645649A (en) * | 2012-05-14 | 2012-08-22 | 重庆大学 | Radar target recognition method based on radar target range profile time-frequency feature extraction |
CN103033801A (en) * | 2012-12-18 | 2013-04-10 | 中国民航大学 | Air maneuvering target detection method based on virtual time sampling and Wigner-Ville distribution (WVD) transformation |
CN103913736A (en) * | 2014-04-11 | 2014-07-09 | 四川大学 | Laser micro Doppler parameter estimation method based on spectrogram rearrangement |
CN104021400A (en) * | 2014-06-16 | 2014-09-03 | 电子科技大学 | SAR image target recognition method based on fractional spectrum sparse representation |
CN105783974A (en) * | 2014-12-25 | 2016-07-20 | 中国科学院声学研究所 | Chirp signal detection, parameter estimation method, and system thereof |
CN106370403A (en) * | 2016-08-22 | 2017-02-01 | 南京信息工程大学 | Instant frequency estimation method based on edge detection |
Non-Patent Citations (2)
Title |
---|
XUELIAN YU ET AL.: "Radar Target Recognition Using Kernel Uncorrelated Discriminant Local Tangent Space Alignment", 《IEEE》 * |
王丽霞 等: "一种基于重排二次型时频分布的电能质量检测新方法", 《四川电力技术》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110133600A (en) * | 2019-06-17 | 2019-08-16 | 电子科技大学 | A kind of lifting airscrew physical parameter extracting method |
CN110133600B (en) * | 2019-06-17 | 2022-11-04 | 电子科技大学 | Helicopter rotor physical parameter extraction method |
CN110738248A (en) * | 2019-09-30 | 2020-01-31 | 朔黄铁路发展有限责任公司 | State perception data feature extraction method and device and system performance evaluation method |
CN110738248B (en) * | 2019-09-30 | 2022-09-27 | 朔黄铁路发展有限责任公司 | State perception data feature extraction method and device and system performance evaluation method |
Also Published As
Publication number | Publication date |
---|---|
CN108490414B (en) | 2021-07-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105574534B (en) | Conspicuousness object detection method based on sparse subspace clustering and low-rank representation | |
CN109766835B (en) | SAR target recognition method for generating countermeasure network based on multi-parameter optimization | |
CN103824088B (en) | SAR target variant recognition method based on multi-information joint dynamic sparse representation | |
CN107271965B (en) | Method for extracting true and false target one-dimensional range profile features in cluster subspace | |
CN105869146B (en) | SAR image change detection based on conspicuousness fusion | |
CN108122008A (en) | SAR image recognition methods based on rarefaction representation and multiple features decision level fusion | |
CN103955701A (en) | Multi-level-combined multi-look synthetic aperture radar image target recognition method | |
CN107133648B (en) | One-dimensional range profile identification method based on adaptive multi-scale fusion sparse preserving projection | |
Guan et al. | Multifractal correlation characteristic for radar detecting low-observable target in sea clutter | |
CN105068062A (en) | Range profile data extrapolation method based on extraction of sparse scattering center | |
Weixing et al. | A fast pedestrian detection via modified HOG feature | |
CN106951822B (en) | One-dimensional range profile fusion identification method based on multi-scale sparse preserving projection | |
CN108490414A (en) | A kind of radar target identification method based on time-frequency distributions instantaneous frequency edge feature | |
Zhu et al. | Improving hyperspectral anomaly detection with a simple weighting strategy | |
CN106908774B (en) | One-dimensional range profile identification method based on multi-scale nuclear sparse preserving projection | |
CN104680536A (en) | Method for detecting SAR image change by utilizing improved non-local average algorithm | |
CN109461127B (en) | SAR image sparse regularization feature enhancement method with interpretation as purpose | |
Ding et al. | Feature extraction based on 2D compressive sensing for SAR automatic target recognition | |
CN109886315A (en) | A kind of Measurement of Similarity between Two Images method kept based on core | |
CN106709428B (en) | One-dimensional range profile robust identification method based on Euler kernel principal component analysis | |
Yan et al. | SAR image target recognition based on Hu invariant moments and SVM | |
CN109766899B (en) | Physical feature extraction and SVM SAR image vehicle target recognition method | |
CN107678007B (en) | Method for extracting radar true and false target one-dimensional range profile features in exponential domain compact subspace | |
Zhu et al. | Target classification by conventional radar based on bispectrum and deep CNN | |
CN104732190A (en) | Synthetic aperture sonar target detection method based on orthogonal texture correlation analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |