CN106022383B - SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation - Google Patents
SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation Download PDFInfo
- Publication number
- CN106022383B CN106022383B CN201610361515.3A CN201610361515A CN106022383B CN 106022383 B CN106022383 B CN 106022383B CN 201610361515 A CN201610361515 A CN 201610361515A CN 106022383 B CN106022383 B CN 106022383B
- Authority
- CN
- China
- Prior art keywords
- sparse
- test sample
- azimuth
- class
- coefficient vector
- 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 48
- 238000012360 testing method Methods 0.000 claims abstract description 145
- 238000012549 training Methods 0.000 claims abstract description 111
- 239000011159 matrix material Substances 0.000 claims abstract description 57
- 239000013598 vector Substances 0.000 claims description 61
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 12
- 238000004364 calculation method Methods 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 6
- 102100024506 Bone morphogenetic protein 2 Human genes 0.000 description 5
- 101000762366 Homo sapiens Bone morphogenetic protein 2 Proteins 0.000 description 5
- 230000008901 benefit Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000007123 defense Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000010287 polarization Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- PXDJXZJSCPSGGI-UHFFFAOYSA-N palmityl palmitate Chemical compound CCCCCCCCCCCCCCCCOC(=O)CCCCCCCCCCCCCCC PXDJXZJSCPSGGI-UHFFFAOYSA-N 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The present invention provides a kind of SAR target identification methods based on azimuth associated dynamic dictionary rarefaction representation, it estimates the azimuth of the SAR image as test sample first, then a relative orientations arc angular region is calculated according to this azimuth estimated value, the sparse features training sample set constituted to the set of the sparse features matrix based on each training sample, the sparse features arranged in matrix for enabling training sample of the azimuth value except relative orientations arc angular region among sparse features training sample set is null matrix, only retain the sparse features matrix of training sample of the azimuth value within relative orientations arc angular region, thus the corresponding sparse features azimuth associated dynamic dictionary of test sample is constituted, then rarefaction representation Classification and Identification is carried out again, greatly reduce the calculation amount of sparse coding and sparse reconstruct, improve identifying processing efficiency, together When decrease interference of the incoherent training sample in azimuth to test sample target identification so that recognition accuracy is also improved.
Description
Technical Field
The invention relates to the technical field of radar target identification, in particular to an SAR target identification method based on azimuth angle related dynamic dictionary sparse representation.
Background
Synthetic Aperture Radar (SAR for short) technology is a pulse Radar technology which adopts a mobile Radar carried on a satellite or an airplane to obtain a Radar target image of a high-precision geographical area. Synthetic aperture radar is an active microwave imaging system that provides high resolution images of a target area by illuminating the target area with electromagnetic waves and performing signal analysis on the echo signals. It has all-weather and all-day working ability and certain penetrating ability. In view of its advantages, it is widely used in the fields of mineral exploration, marine environmental monitoring, military defense, and the like. The research on the identification of targets is the most extensive in the field of military defense, so that the research on the automatic target identification (ATR) of the SAR has attracted extensive attention of domestic and foreign scholars.
In recent years, with the development of the theory of compressed sensing, Sparse Representation (SR) based on compressed sensing has attracted attention of many researchers in the fields of signal processing and pattern recognition. Sparse representation theory indicates that the signal can be represented by a linear combination of atoms in a dictionary, and the distribution of the atoms is sparse, that is, most of coefficients are zero or close to zero, only the corresponding coefficients of the atoms with larger correlation with the input signal are not zero, and the sparsity of the sparse coefficients contains identification information and can describe the most main characteristic information of the target under the premise of less non-zero data elements. Therefore, the sparse representation and sparse reconstruction theory is widely applied to the fields of face recognition, medical tumor recognition, SAR image target recognition and the like.
However, the existing SAR image target recognition based on sparse representation theory uses training samples of all azimuth angles as dictionary atoms, the sparse representation classification identification method omits the scientific fact that the SAR target image characteristics and azimuth angles are closely related, therefore, in the identification process, the test sample needs to be sparsely encoded based on all training samples in the dictionary atom to obtain the sparse coefficient vector corresponding to the test sample, so as to perform sparse reconstruction to determine the target class of the test sample and realize target identification of the test sample, the sparse coding and sparse reconstruction operation process is complicated, the calculation amount is large, the identification processing efficiency is low, and the training samples of uncorrelated orientation angles actually interfere with the target identification of the test sample, these disturbances are prone to recognition errors, and to some extent lead to insufficient target recognition accuracy.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an SAR target recognition method based on azimuth angle related dynamic dictionary sparse representation, which can improve the recognition processing efficiency and the recognition accuracy of radar target recognition based on an SAR image and is used for solving the problems of complicated sparse coding and sparse reconstruction operation processes and insufficient recognition processing efficiency and target recognition accuracy of the SAR target recognition method based on sparse representation classification in the prior art.
In order to achieve the purpose, the invention adopts the following technical means:
an SAR target recognition method based on azimuth angle correlation dynamic dictionary sparse representation comprises the following steps:
1) for different types of known radar targets, respectively collecting SAR images of a plurality of known radar targets in the azimuth angle range of 0-360 degrees for each type as training samples, and respectively recording the azimuth angle value of each training sample;
2) respectively extracting the sparse feature matrix of each training sample in each category, and taking the sparse feature matrix extracted aiming at each training sample as a sparse feature training sample, so that a sparse feature training sample set is formed by a set of the extracted sparse feature matrices of each training sample in each category;
3) aiming at a radar target to be tested, collecting an SAR image of the radar target to be tested as a test sample, respectively calculating the image pixel distribution correlation of the test sample and each training sample, and taking the azimuth angle value of the training sample with the maximum image pixel distribution correlation of the test sample as the azimuth angle estimated value g of the test sampleyTo determine the associated azimuth range phi (g) of the test sampley)＝[(gy-Δg),(gy+Δg)](ii) a Δ g denotes presetAn azimuth floating range value;
4) the azimuth value in the sparse characteristic training sample set is in the relevant azimuth range phi (g)y) Setting sparse feature matrix of the training samples outside the range as a zero matrix, and only keeping azimuth value in relevant azimuth range phi (g)y) Forming a sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample by using the sparse characteristic matrix of the training sample within;
5) extracting a sparse feature matrix of a test sample, establishing a sparse linear equation by utilizing each sparse feature training sample in a sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, performing sparse linear expression on the sparse feature matrix of the test sample, and solving to obtain a coefficient vector of the sparse linear equation to serve as a sparse coefficient vector of the test sample;
6) and aiming at the sparse coefficient vector of each test sample, respectively extracting the class sparse coefficient vector corresponding to each type of known radar target in the sparse coefficient vector of each test sample, then respectively calculating the reconstruction error of sparse reconstruction of the sparse characteristic matrix of the test sample by using the class sparse coefficient vector corresponding to each type of known radar target and the sparse characteristic azimuth related dynamic dictionary corresponding to the test sample through a sparse linear equation, and judging the known radar target class corresponding to the class sparse coefficient vector with the minimum reconstruction error as the radar target class to which the radar target to be tested corresponding to the test sample belongs, thereby realizing the class identification of the radar target to be tested.
In the above SAR target recognition method based on the azimuth angle-dependent dynamic dictionary sparse representation, specifically, in step 3), a specific way of calculating the image pixel distribution correlation of the test sample and the training sample is as follows:
wherein,n-th representing the test sample and the i-th classiThe image pixel distribution correlation coefficient of each training sample,the larger the value of (A) is, the more the test sample and the nth class of the ith class areiThe greater the image pixel distribution correlation of each training sample; i is more than or equal to 1 and less than or equal to K, wherein K represents the total number of the collected classes of the known radar targets; n is more than or equal to 1i≤Ni，NiRepresenting the number of training samples contained in the ith category;n-th representing the ith categoryiPixel values of n-th row and n-th column pixels in the SAR image of each training sample,n-th representing the ith categoryiSAR image pixel mean, f of individual training samplesy(m, n) represents the pixel value of the m-th row and n-th column of pixels in the SAR image of the test sample,a SAR image pixel mean representing a test sample; m ∈ {1,2, …, M }, N ∈ {1,2, …, N }, M and N respectively represent the number of pixel rows and the number of pixel columns of the SAR image.
In the method for identifying an SAR target based on the sparse representation of the azimuth angle-related dynamic dictionary, as an optimal scheme, in the step 3), the value range of the azimuth angle floating range value Δ g is 5 ° to 10 °.
In the method for identifying an SAR target based on sparse representation of an azimuth angle-related dynamic dictionary, specifically, the step 6) is specifically:
61) for the sparse coefficient vector α of the test sample, each class of the azimuth-dependent dynamic dictionary corresponding to the sparse feature is extracted separatelyClass sparse coefficient vector of known radar target, wherein sparse coefficient vector α corresponds to class sparse coefficient vector delta of ith class known radar targeti(α) is:
wherein the class sparse coefficient vector δiThe dimension of (α) is the same as the dimension of the sparse coefficient vector α of the test sample,the nth class in the sparse coefficient vector α representing the test sample corresponding to the ith classiSparse coefficients of a sparse feature matrix of individual training samples, and a class sparse coefficient vector δiThe values of the sparse coefficients corresponding to the non-ith category known radar targets in (α) are all zero, i is more than or equal to 1 and less than or equal to K, K represents the total number of the categories of the acquired known radar targets, n is more than or equal to 1 and less than or equal to ni≤Ni，NiRepresenting the number of training samples contained in the ith category;
62) respectively calculating reconstruction errors for performing sparse reconstruction on a sparse feature matrix of a test sample through a sparse linear equation by utilizing a category sparse coefficient vector corresponding to each type of known radar target and a sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, and judging a known radar target category corresponding to the category sparse coefficient vector with the minimum reconstruction error as a radar target category to which a radar target to be tested corresponding to the test sample belongs; namely:
wherein lyRepresenting the radar target category to which the radar target to be tested corresponding to the test sample belongs; y represents a sparse feature matrix of the test sample; xyRepresenting sparse feature azimuth phase corresponding to test sampleClosing the dynamic dictionary; xyδi(α) the sparse coefficient vector delta using class is showni(α) sparse feature azimuth correlation dynamic dictionary X corresponding to test sampleyCarrying out sparse reconstruction on a sparse feature matrix of the test sample through a sparse linear equation; i | · | purple wind2Is the L2 norm operator;
therefore, the category identification of the radar target to be detected is realized.
Compared with the prior art, the invention has the following beneficial effects:
the invention relates to an SAR target recognition method based on azimuth angle related dynamic dictionary sparse representation, which firstly estimates the azimuth angle of an SAR image as a test sample, and then calculates a related azimuth angle range according to the azimuth angle estimated value, thereby based on a sparse characteristic training sample set formed by a set of sparse characteristic matrixes of each training sample, setting the sparse characteristic matrix of the training sample of which the azimuth angle value is outside the related azimuth angle range in the sparse characteristic training sample set as a zero matrix, only keeping the sparse characteristic matrix of the training sample of which the azimuth angle value is within the related azimuth angle range, thereby forming a sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample, and then carrying out sparse representation classification recognition, greatly reducing the calculated amount of sparse coding and sparse reconstruction, improving the recognition processing efficiency, and simultaneously reducing the interference of the azimuth angle unrelated training sample on the target recognition of the test sample, and further, the identification accuracy is improved, and the problems that in the prior art, the sparse coding and sparse reconstruction operation process of the SAR target identification method adopting sparse representation classification is complicated, and the identification processing efficiency and the target identification accuracy are insufficient are effectively solved.
Drawings
Fig. 1 is a flowchart of an SAR target recognition method based on an azimuth angle-dependent dynamic dictionary sparse representation according to the present invention.
Fig. 2 shows visible light images of three different radar targets of BMP2, BTR70 and T72.
Fig. 3 is SAR images of three different classes of radar targets, BMP2, BTR70, and T72.
Detailed Description
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
First, the sparse representation theory is introduced.
The theory of sparse representation indicates that the input signal can be represented by a linear combination of a set of basis vectors. Given enough samples of class k targets, make-upAny new test sample y ∈ RmCan be well represented by training samples belonging to that class:
whereinSince it is not known to which class a new test sample belongs, n training samples of different classes of class k are used as basis vectors.
Wherein X ═ X1,X1,...,XK]∈Rm×n，Then y can be linearly represented by all training samples:
y＝X1α1+X2α2+…+XKαK＝Xα； (2)
wherein α ═ α1,α2,...αK]∈Rn. Since m < n, the solution of equation (2) is not unique. It is common practice to find the most sparse solution:
wherein | · | purple sweet0And | · | non-conducting phosphor2L0 norm and L2 norm, respectively, ε is the error threshold, and the solution of equation (3) is the NP-hard problem. The development of the compressed sensing theory shows that if the solution is solvedSufficiently sparse, then equation (3) can be considered as an optimization problem with the smallest L1 norm:
wherein | · | purple1Is the L1 norm. The optimization problem for solving the norm can use an algorithm such as Orthogonal Matching Pursuit (OMP).
The idea of the SAR image target identification method based on sparse representation is as follows: constructing a dictionary by using a set of sparse feature matrices of all training samples of all types of known radar targets, and carrying out sparse representation on the sparse feature matrices of the SAR images of the test targets under the dictionary to obtain sparse coding coefficient vectors; and identifying the test sample according to the minimum error between the reconstructed test sample and the test sample after the intra-class coefficient is reconstructed. That is, sparse code vectors for test samples are obtainedAnd then, judging the data as the class with the minimum reconstruction error of each class of subspace coefficients. Namely:
class i subspace coefficientsIs a sum sparse coding vectorVectors of the same dimension, and in which the coefficients corresponding to the training samples other than the i-th class areThe other coefficients are zero except that the corresponding coefficients in the same.
It can be seen that in the prior art, the SAR image target recognition based on the sparse representation theory all uses training samples of all azimuth angles as dictionary atoms, and the sparse representation classification recognition method omits the scientific fact that the SAR target image characteristics and the azimuth angles are closely related. According to the target scattering characteristic theory, the SAR image only has strong correlation with the training sample SAR image with the azimuth deviation of about 5-10 degrees. For example, the SAR image of a target at an azimuth angle of 16 ° has a strong correlation with the SAR image at an azimuth angle of 6 ° to 26 °, wherein the SAR image at an azimuth angle of 11 ° to 21 ° has a strong correlation, and the SAR image at other azimuth angles has a weak correlation, which causes interference to the target SAR image. Therefore, in the sparse representation, the dictionary atoms used to sparsely represent the test image should also contain only the training SAR images within these relevant azimuth ranges (i.e., with strong correlation), rather than containing the training SAR images at all azimuths. Based on the point, the invention provides an SAR target recognition method based on azimuth angle related dynamic dictionary sparse representation, which comprises the steps of firstly estimating the azimuth angle of an SAR image serving as a test sample, then calculating a related azimuth angle range according to the azimuth angle estimated value (the training sample within the azimuth angle range has stronger correlation with the image of the test sample), then finding out all the training samples within the related azimuth angle range to form an azimuth angle related dynamic dictionary corresponding to the test sample, and then carrying out sparse representation classification recognition.
According to the technical idea, the flow of the SAR target recognition method based on the azimuth angle correlation dynamic dictionary sparse representation is shown in FIG. 1, and the method comprises the following steps:
1) for different types of known radar targets, SAR images of a plurality of known radar targets are respectively distributed and collected in the range of 0-360 degrees of azimuth angles for each type to serve as training samples, and azimuth angle values of the training samples are respectively recorded.
In the step, under the condition that conditions allow, the SAR images for collecting a plurality of known radar targets in each type are distributed more densely and better within the range of 0-360 degrees of azimuth angles, so that the estimation error of the azimuth angle of the test sample can be estimated more favorably, and the target identification accuracy of the test sample is improved.
2) And respectively extracting the sparse feature matrix of each training sample in each category, and taking the sparse feature matrix extracted aiming at each training sample as a sparse feature training sample, so that a sparse feature training sample set is formed by a set of the extracted sparse feature matrices of each training sample in each category.
In the step, sparse features of the radar target image are established based on the feature data of the radar target image, and the problem of selection needs to be carried out according to actual application conditions. For a radar target SAR image, the scattering echoes of the target are approximated as the echo response sum of a plurality of scattering centers according to the principle of physical optical approximation (see prior art document "Potter, l.c.; Ertin, e.; Parker, j.t.; Cetin, m.spacinity and compressed sensing in scattering procedures of the IEEE 2010,98,1006 and 1020."), and these sparse scattering centers provide a concise, physically relevant description of the target characteristics (see prior art document "M. is; karl, w.c.; castanon, d.a. evaluation of a standardized SAR imaging technology-oriented defects, in proc.spie 4053, Algorithms for Synthetic Aperture radar image VII, Orlando, FL, USA,24 April 2000; pp.40-51 "). The classical scattering center parameter modeling method is based on an approximate physical optical model, and parameters of the method comprise pixel value characteristics, pixel frequency spectrum, incidence angle, acceptance angle, polarization and the like; the parameters of the scattering centers can be selected as effective sparse features for SAR target identification to establish a sparse feature matrix.
3) Aiming at a radar target to be tested, collecting an SAR image of the radar target to be tested as a test sample, respectively calculating the image pixel distribution correlation of the test sample and each training sample, and taking the azimuth angle value of the training sample with the maximum image pixel distribution correlation of the test sample as the azimuth angle estimated value g of the test sampleyTo determine the associated azimuth range phi (g) of the test sampley)＝[(gy-Δg),(gy+Δg)](ii) a Δ g represents a preset azimuth float range value.
In this step, the specific way of calculating the image pixel distribution correlation of the test sample and the training sample is as follows:
wherein,n-th representing the test sample and the i-th classiThe image pixel distribution correlation coefficient of each training sample,the larger the value of (A) is, the more the test sample and the nth class of the ith class areiThe greater the image pixel distribution correlation of each training sample; i is more than or equal to 1 and less than or equal to K, and K represents the acquired known radar targetA target category total; n is more than or equal to 1i≤Ni，NiRepresenting the number of training samples contained in the ith category;n-th representing the ith categoryiPixel values of n-th row and n-th column pixels in the SAR image of each training sample,n-th representing the ith categoryiSAR image pixel mean, f of individual training samplesy(m, n) represents the pixel value of the m-th row and n-th column of pixels in the SAR image of the test sample,a SAR image pixel mean representing a test sample; m ∈ {1,2, …, M }, N ∈ {1,2, …, N }, M and N respectively represent the number of pixel rows and the number of pixel columns of the SAR image.
Through the correlation of the image pixel distribution condition, the azimuth angle correlation among SAR images can be embodied to a certain extent; compared with sparse coding and sparse reconstruction operation, the operation of the image pixel distribution correlation coefficient is very simple, the operation can be rapidly executed in computer processing, and a plurality of different image pixel distribution correlation coefficients can be operated in parallel in a batch processing mode, so that the step has higher computer operation processing efficiency. Meanwhile, according to the target scattering characteristic theory, the SAR image only has strong correlation with the SAR image of the training sample with the deviation of the azimuth angle of about 5-10 degrees, so the optimal value range of the azimuth angle floating range value delta g can be set to be 5-10 degrees; the specific value of the azimuth angle floating range value deltag is determined according to the practical application condition.
4) The azimuth value in the sparse characteristic training sample set is in the relevant azimuth range phi (g)y) Setting sparse feature matrix of the training samples outside the range as a zero matrix, and only keeping azimuth value in relevant azimuth range phi (g)y) Sparse feature matrices of training samples within, from which measurements are constructedAnd testing a sparse characteristic azimuth angle related dynamic dictionary corresponding to the sample.
Compared with the basic sparse representation method in the prior art, the sparse feature training sample set is directly used as the dictionary, the sparse feature azimuth related dynamic dictionary corresponding to the obtained test sample is obviously different from the sparse feature training sample set:
a. in the basic sparse representation, the dictionary is composed of training samples of all azimuths; only a few training samples of azimuth angles (near the azimuth angle estimated value of the test sample) are formed in the sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample; the number of atoms in the two dictionaries is the same, because in order to maintain the sparse representation model, the training samples in the sparse feature azimuth angle correlation dynamic dictionary corresponding to the test sample except for the azimuth angle of the test sample are taken as atoms, and the matrix atoms of other azimuth angles are replaced by 0 matrixes.
b. In a basic sparse representation, the dictionary is static; in the method, the sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample is dynamically changed due to different test samples, and because the azimuth angles of different test targets are different, the training sample atoms within the corresponding related azimuth angle range are different, namely, each test sample corresponds to a sparse characteristic azimuth angle related dynamic dictionary only taking the matrix atoms within the related azimuth angle range as a non-zero matrix; when the next test image comes, the azimuth angle of the test image needs to be estimated again, and the corresponding sparse feature azimuth angle related dynamic dictionary needs to be selected again.
c. In the method, the sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample has a lot of matrix atoms of zero matrixes, and only a small number of matrix atoms of non-zero matrixes exist, so that compared with a global dictionary in basic sparse representation, the calculation amount of sparse coding and sparse reconstruction is greatly reduced, and the identification process of the target identification of the test sample is accelerated.
5) Extracting a sparse characteristic matrix of a test sample, establishing a sparse linear equation by utilizing each sparse characteristic training sample in a sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample, carrying out sparse linear expression on the sparse characteristic matrix of the test sample, and solving to obtain a coefficient vector of the sparse linear equation to be used as a sparse coefficient vector of the test sample.
In the same step, the pixel value characteristics, the pixel frequency spectrum, the incidence angle, the receiving angle, the polarization and other scattering center parameters of the test sample image can be selected as effective sparse characteristics for SAR target identification to establish a sparse characteristic matrix. Only, the parameters chosen to build the sparse feature matrix for the training samples and the test samples should be the same.
6) And aiming at the sparse coefficient vector of each test sample, respectively extracting the class sparse coefficient vector corresponding to each type of known radar target in the sparse coefficient vector of each test sample, then respectively calculating the reconstruction error of sparse reconstruction of the sparse characteristic matrix of the test sample by using the class sparse coefficient vector corresponding to each type of known radar target and the sparse characteristic azimuth related dynamic dictionary corresponding to the test sample through a sparse linear equation, and judging the known radar target class corresponding to the class sparse coefficient vector with the minimum reconstruction error as the radar target class to which the radar target to be tested corresponding to the test sample belongs, thereby realizing the class identification of the radar target to be tested.
The method comprises the following steps:
61) for the sparse coefficient vector α of the test sample, respectively extracting a class sparse coefficient vector corresponding to each class of known radar targets in the sparse feature azimuth correlation dynamic dictionary, wherein the sparse coefficient vector α corresponds to the class sparse coefficient vector delta of the ith class of known radar targetsi(α) is:
wherein the class sparse coefficient vector δiThe dimension of (α) is the same as the dimension of the sparse coefficient vector α of the test sample,the nth class in the sparse coefficient vector α representing the test sample corresponding to the ith classiSparse coefficients of a sparse feature matrix of individual training samples, and a class sparse coefficient vector δiThe values of the sparse coefficients corresponding to the non-ith category known radar targets in (α) are all zero, i is more than or equal to 1 and less than or equal to K, K represents the total number of the categories of the acquired known radar targets, n is more than or equal to 1 and less than or equal to ni≤Ni，NiRepresenting the number of training samples contained in the ith category;
62) respectively calculating reconstruction errors for performing sparse reconstruction on a sparse feature matrix of a test sample through a sparse linear equation by utilizing a category sparse coefficient vector corresponding to each type of known radar target and a sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, and judging a known radar target category corresponding to the category sparse coefficient vector with the minimum reconstruction error as a radar target category to which a radar target to be tested corresponding to the test sample belongs; namely:
wherein lyRepresenting the radar target category to which the radar target to be tested corresponding to the test sample belongs; y represents a sparse feature matrix of the test sample; xyRepresenting a sparse feature azimuth related dynamic dictionary corresponding to the test sample; i | · | purple wind2Is the L2 norm operator; xyδi(α) the sparse coefficient vector delta using class is showni(α) sparse feature azimuth correlation dynamic dictionary X corresponding to test sampleyCarrying out sparse reconstruction on a sparse feature matrix of a test sample through a sparse linear equation, namely:
wherein,sparse feature azimuth correlation dynamic dictionary X corresponding to representation test sampleyN of the ith classiSparse feature matrix corresponding to each training sample position, and if the nth class of the ith classiThe azimuth of each training sample is not in the relevant azimuth range phi (g) of the test sampley) Within, thenOnly when the nth class is presentiThe azimuth of each training sample is within the range of the associated azimuth of the test sample phi (g)y) When the content is within the range, the user can select the specific part,is a non-zero value; this makes the non-zero terms in the reconstruction formula very small, which greatly reduces the amount of calculation.
Therefore, the category identification of the radar target to be detected is realized.
The SAR target recognition method based on the azimuth angle correlation dynamic dictionary sparse representation can be applied to a radar target recognition system based on computer programming self-operation, and automatic radar target recognition is realized.
The technical solution of the present invention will be further described by examples.
Example (b):
the present example uses the MSTAR database for the experiment, which is the measured data obtained by the SAR system of the st wave band of the san diego national laboratory, usa, with a resolution of 0.3m × 0.3m, and the pixel density of each SAR image is 128 rows × 128 columns, acquired at an azimuth angle of 0 ° to 360 °. In this embodiment, experiments are performed using three types of targets, namely, BMP2 (infantry tank), BTR70 (armored troops), and T72(T-72 type master tank) in the MSTAR database, visible light images of three radar targets of different types, BMP2, BTR70, and T72, are shown as (2a), (2b), and (2c) in fig. 2, respectively, and SAR images of three radar targets of different types, BMP2, BTR70, and T72, are shown as (3a), (3b), and (3c) in fig. 3, respectively. In this embodiment, the SAR image data of the part of each class of targets with azimuth angles of 0 ° to 360 ° is used as training sample data, and the SAR image data of the remaining part is used as test sample data. The number of training samples and test samples is shown in table 1:
TABLE 1
Then, by adopting the SAR target recognition method based on azimuth angle related dynamic dictionary sparse representation, the training samples are used for recognizing the target types of the test samples; meanwhile, for comparison, a basic sparse representation classification identification method (without distinguishing the estimation azimuth angle of the test sample) in the prior art is adopted, the training samples are used for identifying the target class of the test sample, the identification result confusion matrixes of the basic sparse representation classification identification method in the prior art and the method of the invention are respectively counted, and the number of the three different training sample classes of the test sample identified in each class is recorded in the identification result confusion matrix. The confusion matrix of the identification result of the method is shown in table 2, and the confusion matrix of the identification result of the basic sparse representation classification identification method (without distinguishing the estimation azimuth angle of the test sample) is shown in table 3.
TABLE 2
TABLE 3
Therefore, the recognition accuracy of the method of the present invention and the basic sparse representation classification recognition method in the prior art are obtained through statistics, as shown in table 4.
TABLE 4
Compared with the prior art, the SAR target recognition method based on the azimuth angle related dynamic dictionary sparse representation has the advantage that the recognition accuracy is improved.
In summary, the SAR target recognition method based on the azimuth angle related dynamic dictionary sparse representation of the invention firstly estimates the azimuth angle of the SAR image as the test sample, and then calculates a related azimuth angle range according to the azimuth angle estimated value, thereby based on the sparse feature training sample set formed by the set of the sparse feature matrixes of each training sample, the sparse feature matrix of the training sample with the azimuth angle value outside the related azimuth angle range in the sparse feature training sample set is set as a zero matrix, only the sparse feature matrix of the training sample with the azimuth angle value within the related azimuth angle range is reserved, thereby forming the sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, and then sparse representation classification recognition is carried out, thereby greatly reducing the calculated amount of sparse coding and sparse reconstruction, improving the recognition processing efficiency, and simultaneously reducing the interference of the training sample with irrelevant azimuth angle to the target recognition of the test sample, and further, the identification accuracy is improved, and the problems that in the prior art, the sparse coding and sparse reconstruction operation process of the SAR target identification method adopting sparse representation classification is complicated, and the identification processing efficiency and the target identification accuracy are insufficient are effectively solved.
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (4)
1. The SAR target recognition method based on azimuth angle correlation dynamic dictionary sparse representation is characterized by comprising the following steps:
1) for different types of known radar targets, respectively collecting SAR images of a plurality of known radar targets in the azimuth angle range of 0-360 degrees for each type as training samples, and respectively recording the azimuth angle value of each training sample;
2) respectively extracting the sparse feature matrix of each training sample in each category, and taking the sparse feature matrix extracted aiming at each training sample as a sparse feature training sample, so that a sparse feature training sample set is formed by a set of the extracted sparse feature matrices of each training sample in each category;
3) aiming at a radar target to be tested, collecting an SAR image of the radar target to be tested as a test sample, respectively calculating the image pixel distribution correlation of the test sample and each training sample, and taking the azimuth angle value of the training sample with the maximum image pixel distribution correlation of the test sample as the azimuth angle estimated value g of the test sampleyTo determine the associated azimuth range phi (g) of the test sampley)＝[(gy-Δg),(gy+Δg)](ii) a Δ g represents a preset azimuth floating range value;
4) the azimuth value in the sparse characteristic training sample set is in the relevant azimuth range phi (g)y) Setting sparse feature matrix of the training samples outside the range as a zero matrix, and only keeping azimuth value in relevant azimuth range phi (g)y) Forming a sparse characteristic azimuth angle related dynamic dictionary corresponding to the test sample by using the sparse characteristic matrix of the training sample within;
5) extracting a sparse feature matrix of a test sample, establishing a sparse linear equation by utilizing each sparse feature training sample in a sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, performing sparse linear expression on the sparse feature matrix of the test sample, and solving to obtain a coefficient vector of the sparse linear equation to serve as a sparse coefficient vector of the test sample;
6) and aiming at the sparse coefficient vector of each test sample, respectively extracting the class sparse coefficient vector corresponding to each type of known radar target in the sparse coefficient vector of each test sample, then respectively calculating the reconstruction error of sparse reconstruction of the sparse characteristic matrix of the test sample by using the class sparse coefficient vector corresponding to each type of known radar target and the sparse characteristic azimuth related dynamic dictionary corresponding to the test sample through a sparse linear equation, and judging the known radar target class corresponding to the class sparse coefficient vector with the minimum reconstruction error as the radar target class to which the radar target to be tested corresponding to the test sample belongs, thereby realizing the class identification of the radar target to be tested.
2. The SAR target recognition method based on the azimuth angle correlation dynamic dictionary sparse representation as claimed in claim 1, wherein in the step 3), the specific way of calculating the image pixel distribution correlation of the test sample and the training sample is as follows:
wherein,n-th representing the test sample and the i-th classiThe image pixel distribution correlation coefficient of each training sample,the larger the value of (A) is, the more the test sample and the nth class of the ith class areiThe greater the image pixel distribution correlation of each training sample; i is more than or equal to 1 and less than or equal to K, wherein K represents the total number of the collected classes of the known radar targets; n is more than or equal to 1i≤Ni，NiRepresenting the number of training samples contained in the ith category;n-th representing the ith categoryiPixel values of n-th row and n-th column pixels in the SAR image of each training sample,n-th representing the ith categoryiSAR image pixel mean, f of individual training samplesy(m, n) represents the pixel value of the m-th row and n-th column of pixels in the SAR image of the test sample,a SAR image pixel mean representing a test sample; m is equal to {1,2, …, M }, N is equal to {1,2, …, N }, and M and N respectively represent SAR imagesThe number of pixel rows and the number of pixel columns.
3. The SAR target recognition method based on the azimuth angle correlation dynamic dictionary sparse representation as claimed in claim 1, wherein in the step 3), the value range of the azimuth angle floating range value Δ g is 5-10 °.
4. The SAR target recognition method based on the azimuth angle-dependent dynamic dictionary sparse representation as claimed in claim 1, wherein the step 6) is specifically as follows:
61) for the sparse coefficient vector α of the test sample, respectively extracting a class sparse coefficient vector corresponding to each class of known radar targets in the sparse feature azimuth correlation dynamic dictionary, wherein the sparse coefficient vector α corresponds to the class sparse coefficient vector delta of the ith class of known radar targetsi(α) is:
wherein the class sparse coefficient vector δiThe dimension of (α) is the same as the dimension of the sparse coefficient vector α of the test sample,the nth class in the sparse coefficient vector α representing the test sample corresponding to the ith classiSparse coefficients of a sparse feature matrix of individual training samples, and a class sparse coefficient vector δiThe values of the sparse coefficients corresponding to the non-ith category known radar targets in (α) are all zero, i is more than or equal to 1 and less than or equal to K, K represents the total number of the categories of the acquired known radar targets, n is more than or equal to 1 and less than or equal to ni≤Ni，NiRepresenting the number of training samples contained in the ith category;
62) respectively calculating reconstruction errors for performing sparse reconstruction on a sparse feature matrix of a test sample through a sparse linear equation by utilizing a category sparse coefficient vector corresponding to each type of known radar target and a sparse feature azimuth angle related dynamic dictionary corresponding to the test sample, and judging a known radar target category corresponding to the category sparse coefficient vector with the minimum reconstruction error as a radar target category to which a radar target to be tested corresponding to the test sample belongs; namely:
wherein lyRepresenting the radar target category to which the radar target to be tested corresponding to the test sample belongs; y represents a sparse feature matrix of the test sample; xyRepresenting a sparse feature azimuth related dynamic dictionary corresponding to the test sample; xyδi(α) the sparse coefficient vector delta using class is showni(α) sparse feature azimuth correlation dynamic dictionary X corresponding to test sampleyCarrying out sparse reconstruction on a sparse feature matrix of the test sample through a sparse linear equation; i | · | purple wind2Is the L2 norm operator;
therefore, the category identification of the radar target to be detected is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610361515.3A CN106022383B (en) | 2016-05-26 | 2016-05-26 | SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610361515.3A CN106022383B (en) | 2016-05-26 | 2016-05-26 | SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106022383A CN106022383A (en) | 2016-10-12 |
| CN106022383B true CN106022383B (en) | 2019-05-31 |
Family
ID=57094370
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610361515.3A Active CN106022383B (en) | 2016-05-26 | 2016-05-26 | SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106022383B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109344767B (en) * | 2018-09-29 | 2021-09-28 | 重庆大学 | SAR target identification method based on multi-azimuth multi-feature collaborative representation |
| CN109444880A (en) * | 2018-10-23 | 2019-03-08 | 重庆大学 | A kind of SAR target identification method based on the fusion of multiple features low-rank representation |
| CN110068799B (en) * | 2019-04-22 | 2022-04-22 | 电子科技大学 | Sparse neighborhood center preserving RCS sequence feature extraction method |
| CN112990342B (en) * | 2021-04-08 | 2023-09-19 | 重庆大学 | Semi-supervised SAR target recognition method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012215568A (en) * | 2011-03-31 | 2012-11-08 | Mitsubishi Electric Research Laboratories Inc | High resolution sar imaging using non-uniform pulse timing |
| CN103065162A (en) * | 2013-01-31 | 2013-04-24 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) target azimuth angle estimation method based on sparse description |
| CN103984966A (en) * | 2014-05-29 | 2014-08-13 | 西安电子科技大学 | SAR image target recognition method based on sparse representation |
| CN104899549A (en) * | 2015-04-17 | 2015-09-09 | 重庆大学 | SAR target recognition method based on range profile time-frequency image identification dictionary learning |
-
2016
- 2016-05-26 CN CN201610361515.3A patent/CN106022383B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012215568A (en) * | 2011-03-31 | 2012-11-08 | Mitsubishi Electric Research Laboratories Inc | High resolution sar imaging using non-uniform pulse timing |
| CN103065162A (en) * | 2013-01-31 | 2013-04-24 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) target azimuth angle estimation method based on sparse description |
| CN103984966A (en) * | 2014-05-29 | 2014-08-13 | 西安电子科技大学 | SAR image target recognition method based on sparse representation |
| CN104899549A (en) * | 2015-04-17 | 2015-09-09 | 重庆大学 | SAR target recognition method based on range profile time-frequency image identification dictionary learning |
Non-Patent Citations (2)
| Title |
|---|
| 一种利用方位角信息的SAR目标分类方法;任皓 等;《仪器仪表学报》;20090630;第30卷(第6期);第110-114页 |
| 稀疏表示框架下的SAR目标识别;程建 等;《电子科技大学学报》;20140731;第43卷(第4期);第524-529页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106022383A (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108229404B (en) | Radar echo signal target identification method based on deep learning | |
| CN106096506B (en) | Based on the SAR target identification method for differentiating doubledictionary between subclass class | |
| CN106056070B (en) | Restore the SAR target identification method with rarefaction representation based on low-rank matrix | |
| CN106022383B (en) | SAR target identification method based on azimuth associated dynamic dictionary rarefaction representation | |
| CN103984966B (en) | SAR image target recognition method based on sparse representation | |
| CN103226196B (en) | Radar target recognition method based on sparse feature | |
| CN112990334A (en) | Small sample SAR image target identification method based on improved prototype network | |
| CN105913081B (en) | SAR image classification method based on improved PCAnet | |
| CN103824093B (en) | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods | |
| CN105608465B (en) | Full-polarization SAR coastline terrain classification method | |
| CN107194329B (en) | One-dimensional range profile identification method based on adaptive local sparse preserving projection | |
| Wang et al. | SAR target recognition based on probabilistic meta-learning | |
| CN110231617A (en) | Target obstacle orientation detection method, apparatus, trailer-mounted radar and storage medium | |
| CN106874932B (en) | SAR target model identification method based on rapid sparse description | |
| CN104751183B (en) | Classification of Polarimetric SAR Image method based on tensor MPCA | |
| CN111796272A (en) | Real-time gesture recognition method and computer equipment for through-wall radar human body image sequence | |
| CN109001687A (en) | Airborne radar space-time self-adaptive filtering method based on generalized sidelobe cancellation structure | |
| CN105160353A (en) | Polarimetric SAR data ground object classification method based on multiple feature sets | |
| CN112070151A (en) | Target classification and identification method of MSTAR data image | |
| CN102737253A (en) | SAR (Synthetic Aperture Radar) image target identification method | |
| CN108875798B (en) | Super-pixel-level feature extraction method based on spatial pyramid pooling | |
| CN107403136B (en) | SAR target model identification method based on structure-preserving dictionary learning | |
| CN105373809B (en) | SAR target identification methods based on non-negative least square rarefaction representation | |
| CN110135280B (en) | Multi-view SAR automatic target recognition method based on sparse representation classification | |
| CN108830290B (en) | SAR image classification method based on sparse representation and Gaussian distribution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |