CN103824093B - It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods - Google Patents
It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods Download PDFInfo
- Publication number
- CN103824093B CN103824093B CN201410103639.2A CN201410103639A CN103824093B CN 103824093 B CN103824093 B CN 103824093B CN 201410103639 A CN201410103639 A CN 201410103639A CN 103824093 B CN103824093 B CN 103824093B
- Authority
- CN
- China
- Prior art keywords
- msub
- mrow
- sample
- training objective
- kfda
- 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.)
- Expired - Fee Related
Links
Abstract
The present invention provides a kind of based on KFDA（Kernel Fisher Discriminant Analysis）And SVM（Support Vector Machine）SAR image target's feature-extraction and recognition methods, comprise the following steps：The test target sample of training objective sample and unknown classification to known class carries out amplitude data normalized；Feature extraction is carried out using KFDA criterions training objective sample respectively to the known class after normalization and the test target sample of unknown classification；The training objective sample characteristics for the known class extracted using KFDA criterions, are trained to SVM classifier, produce optimal classification surface；Finally by optimal classification surface, the feature of the test target sample for the unknown classification extracted to KFDA criterions is identified；Present invention reduces the requirement to preprocessing process, the azimuthal sensitivity of SAR image is overcome, the dimension of sample characteristics is have compressed, and obtains higher object recognition rate, with good generalization.
Description
Technical field
The invention belongs to SAR image processing and area of pattern recognition, it is related to a kind of based on KFDA (Kernel Fisher
Discriminant Analysis) and SVM (Support Vector Machine) SAR image target's feature-extraction with know
Other method.
Background technology
Synthetic aperture radar (Synthetic Aperture Radar, SAR) is a kind of the active of utilization microwave perception
Sensor, can carry out round-the-clock, round-the-clock scouting to interesting target or region, be obtained with various visual angles, many angle of depression data
Take ability and the penetration capacity to some atural objects.So-called radar target recognition, is exactly detected and is determined to target in radar
On the basis of position, according to target and the radar echo signal of environment, target signature is extracted, attribute, classification or the type of target is realized
Number judgement.With the continuous maturation of SAR imaging techniques, the target identification based on SAR image has more and more important meaning.
In target identification flow based on SAR image, most important two step is characterized extraction and identification.For SAR image,
Due to its special imaging mode so that its optical imagery unlike can be than being described more completely the overall shape of target
Shape, but sparse scattering center distribution is shown as, and it is more sensitive to the orientation of imaging.Therefore, target is efficiently extracted special
Levy and be particularly important.After the clarification of objective of SAR image has been obtained, ensuing main task is exactly to unknown object
It is identified.
In the target's feature-extraction method of SAR image, most commonly principal component analysis (PCA), the master of kernel function divide
The methods such as amount analysis (KPCA), the shortcoming of wherein principal component method is can not to extract nonlinear characteristic present in image,
And the shortcoming of the principal component method of kernel function is the dimension that extracted feature does not have good class discriminating power and feature
Number is higher；In the target identification method of SAR image, most commonly maximal correlation grader and nearest neighbor classifier etc., its
The shortcoming of middle maximal correlation grader is that algorithm complex is also higher when sample dimension is higher, and nearest neighbor classifier is scarce
Point is that the optimal classification surface chosen is not global optimum.
The content of the invention
The technical problem to be solved in the present invention is：There is provided a kind of SAR image target's feature-extraction based on KFDA and SVM with
Recognition methods, this method carries out target's feature-extraction using KFDA criterions, then realizes target identification by SVM classifier.This hair
It is bright by the way that KPCA criterions are combined with SVM classifier, the present invention can very well complete the target's feature-extraction of SAR image with
Identification, reduces the requirement to preprocessing process, overcomes the azimuthal sensitivity of SAR image, have compressed the dimension of sample characteristics,
And higher object recognition rate is obtained, with good generalization.
The technical solution adopted for the present invention to solve the technical problems is：
It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods, including following steps：
Step (1) carries out amplitude data to the training objective sample of known class and the test target sample of unknown classification and returned
One change is handled；
Step (2) is using KFDA criterions to the training objective sample of the known class after normalization and the test of unknown classification
Target sample data carry out feature extraction；
The training objective sample characteristics for the known class that step (3) is extracted using KFDA criterions, to SVM (Support
Vector Machine) grader is trained, and produces optimal classification surface；
Step (4) is by optimal classification surface, and the feature of the test target sample for the unknown classification extracted to KFDA criterions is entered
Row identification.
Further, to the training objective sample and the test target of unknown classification of known class in described step (1)
Sample carry out amplitude data normalized process be specially：
Normalizing formula is：
Wherein, x for any one known class training objective sample or unknown classification test target sample vector table
Show and (image array is arranged in vector form by row), xNormalizedTraining objective sample for corresponding known class or not
Know classification test target sample amplitude data normalization after vector representation.
Further, training objective sample of the KFDA criterions to the known class after normalization is utilized in described step (2)
The process that the test target sample data of this and unknown classification carries out feature extraction is specially：First seek Scatter Matrix K in classwWith class
Between Scatter Matrix Kb, then askThe corresponding characteristic vector of nonzero eigenvalue, finally seek known class under KFDA criterions
The feature of training objective sample and the test target sample of unknown classification；Scatter Matrix K wherein in classwFor：
Wherein, N is the number of the training objective sample of known class, and c is the classification of the training objective sample of known class
Number,For N × NiMatrix, xp(p=1,2 ..., N) for p-th of known class training objective samples normalization
Data afterwards,For the data after j-th of training objective samples normalization in the i-th class, NiKnown to the i-th class
The sample number of the training objective sample of classification, k1() represents kernel function, and I is Ni×NiUnit matrix,It is for elementNi×NiSquare formation；If KwMatrix is a singular matrix, then makes Kw≈Kw+ κ I are to solve KwSingularity, I is and KwTogether
The unit matrix of rank, κ is a very little and the disturbance constant more than zero, generally can use κ≤10-2；
Class scatter matrix KbFor：
Wherein,For i-th
Data in class after j-th of training objective samples normalization,
Ask againThe corresponding characteristic vector α of nonzero eigenvalue, i.e.,：
Wherein, λ represents characteristic value；To the training objective sample or unknown classification of the known class after any one normalization
Test target sample, the training objective sample for the known class that KFDA criterions are finally extracted or the test target sample of unknown classification
Feature is a c-1 dimensional vector, is represented by z=[z1,z2,…,zc-1]T, it is represented by per one-dimensional element：
Wherein, t=1,2 ..., c-1,RepresentT-th of nonzero eigenvalue corresponding to j-th of characteristic vector
Element.
Further, the training objective sample characteristics for the known class extracted in described step (3) using KFDA criterions,
SVM classifier is trained, the process for producing optimal classification surface is specially：
Using Lagrange multiplier methods, functional is maximized：
ai>=0, i=1 ..., n
Wherein, yi∈ {+1, -1 }, corresponds to the training objective sample of two different class known class respectively；ziFor KFDA criterions
The feature of the training objective for i-th of the known class extracted；k2() represents kernel function, with above-mentioned k1() is completely only
It is vertical；aiFor the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked.
Further, by optimal classification surface in described step (4), the test for the unknown classification extracted to KFDA criterions
The process that the feature of target sample is identified is specially：
The function representation of optimal classification surface is：
Wherein, aiFor the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked；yi∈
{+1, -1 }, corresponds to the training objective sample of two different class known class respectively；k2() represents kernel function；ziFor KFDA
The feature of the training objective for i-th of known class that criterion is extracted；Z is the test target sample for the unknown classification that KFDA criterions are extracted
This feature；F (x) ∈ {+1, -1 }, that is, determine the test target sample of current unknown classification
This generic.
The present invention principle be：Composite copolymer criterion is applied to feature extraction, i.e., it is empty in higher-dimension by kernel method
Between in extract the nonlinear characteristic of SAR image, in the hope of obtaining more preferable class discriminating power；What SVM was found is a satisfaction classification
It is required that optimal hyperlane, it is adaptable to small sample, it is non-linear the problems such as；So the two superiority is combined, it can realize well
The target's feature-extraction of SAR image and identification.
The advantage of the present invention compared with prior art is：
1. for target's feature-extraction part, compared to PCA criterions, the present invention can obtain the nonlinear characteristic in image；
2. for target's feature-extraction part, compared to KPCA criterions, the present invention can obtain lower intrinsic dimensionality, and tool
There is more preferable robustness；
3. for target identification portion, compared to maximal correlation grader, the present invention dexterously solves problem of dimension, calculates
Method complexity is unrelated with sample dimension；
4. for target identification portion, compared to nearest neighbor classifier, what the present invention was obtained is globally optimal solution；
5. by the way that KPCA criterions are combined with SVM classifier, the target that the present invention can very well complete SAR image is special
Extraction and identification are levied, the requirement to preprocessing process is reduced, overcomes the azimuthal sensitivity of SAR image, sample is have compressed special
The dimension levied, and higher object recognition rate is obtained, with good generalization.
Brief description of the drawings
Fig. 1 is target's feature-extraction and identification process figure of the invention；
Fig. 2 is to carry out target's feature-extraction and the process of identification to example.
Wherein：
Fig. 2 (a) is two dimension of the class tank of a, b, c tri- when the angle of pitch is 17 ° as the training objective sample of known class
Characteristic profile；
Two dimensional character distribution map when the c class tanks that Fig. 2 (b) is when being 15 ° to the angle of pitch are identified；
(i.e. model is identical, configures different mesh by the modification target c# for the c class tanks that Fig. 2 (c) is when being 15 ° to the angle of pitch
Mark) two dimensional character distribution map when being identified.
Embodiment
Below in conjunction with the accompanying drawings and the present invention is discussed in detail in embodiment.
As shown in figure 1, the present invention SAR image target's feature-extraction based on KFDA and SVM and recognition methods it is specific
Implementation steps are as follows：
The test target sample of step (1), the training objective sample to known class and unknown classification carries out amplitude data
Normalized, normalizing formula is：
Wherein, x for any one known class training objective sample or unknown classification test target sample vector table
Show and (image array is arranged in vector form by row), xNormalizedTraining objective sample for corresponding known class or not
Know classification test target sample amplitude data normalization after vector representation.
Step (2), using KFDA criterions to the training objective sample of the known class after normalization and the survey of unknown classification
Try target sample data xNormalizedFeature extraction is carried out, Scatter Matrix K in class is first soughtwWith class scatter matrix Kb, then ask
KbThe corresponding characteristic vector of nonzero eigenvalue, finally seek the training objective sample and unknown class of known class under KFDA criterions
The feature of other test target sample；Scatter Matrix K wherein in classwFor：
Wherein, N is the number of the training objective sample of known class, and c is the classification of the training objective sample of known class
Number,For N × NiMatrix, xp(p=1,2 ..., N) for p-th of known class training objective samples normalization
Data afterwards,For the data after j-th of training objective samples normalization in the i-th class, NiKnown to the i-th class
The sample number of the training objective sample of classification, k1() represents kernel function, and I is Ni×NiUnit matrix,It is for elementNi×NiSquare formation；If KwMatrix is a singular matrix, then makes Kw≈Kw+ κ I are to solve KwSingularity, I is and KwTogether
The unit matrix of rank, κ is a very little and the disturbance constant more than zero, generally can use κ≤10-2；
Class scatter matrix KbFor：
Wherein,For i-th
Data in class after j-th of training objective samples normalization,
Ask againThe corresponding characteristic vector α of nonzero eigenvalue, i.e.,：
Wherein, λ represents characteristic value；To the training objective sample or unknown classification of the known class after any one normalization
Test target sample, the training objective sample for the known class that KFDA criterions are finally extracted or the test target sample of unknown classification
Feature is a c-1 dimensional vector, is represented by z=[z1,z2,…,zc-1]T, it is represented by per one-dimensional element：
Wherein, t=1,2 ..., c-1,RepresentT-th of nonzero eigenvalue corresponding to j-th of characteristic vector
Element.
The training objective sample characteristics of step (3), the known class extracted using KFDA criterions, are carried out to SVM classifier
Training, produces optimal classification surface, you can utilize Lagrange multiplier methods, maximizes functional：
ai>=0, i=1 ..., n
Wherein, yi∈ {+1, -1 }, corresponds to the training objective sample of two different class known class respectively；ziFor KFDA criterions
The feature of the training objective for i-th of the known class extracted；k2() represents kernel function, with above-mentioned k1() is completely only
It is vertical；aiFor the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked.
Step (4), by optimal classification surface, the feature of the test target sample for the unknown classification extracted to KFDA criterions is entered
Row identification, wherein, the function representation of optimal classification surface is：
Wherein, aiFor the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked；yi∈
{+1, -1 }, corresponds to the training objective sample of two different class known class respectively；k2() represents kernel function；ziFor KFDA
The feature of the training objective for i-th of known class that criterion is extracted；Z is the test target sample for the unknown classification that KFDA criterions are extracted
This feature；F (x) ∈ {+1, -1 }, that is, determine the test target sample of current unknown classification
This generic.
(a) is the class tank of a, b, c tri- when the angle of pitch is 17 ° as the two of the training objective sample of known class in Fig. 2
Dimensional feature distribution map；(b) it is two dimensional character distribution map when c class tanks when being 15 ° to the angle of pitch are identified；(c) for pair
When modification target c# classes (model is identical, configures different targets) tank of c class tanks when the angle of pitch is 15 ° is identified
Two dimensional character distribution map.As can be seen that in figure (b), as with configuration target, the test target sample c classes of unknown classification are smooth
Gram two dimensional character be gathered in well around the two dimensional character of training objective sample c class tanks of known class；Scheme in (c),
As Morph Target, the two dimensional character of the test target sample c# class tanks of unknown classification is also gathered in known class well
Around the two dimensional character of training objective sample c class tanks.Prove that this method has superior target identification ability, realize SAR
Image object Feature extraction and recognition.
The content not being described in detail in description of the invention belongs to prior art known to professional and technical personnel in the field.
Although disclosing highly preferred embodiment of the present invention and accompanying drawing for the purpose of illustration, those skilled in the art can be with
Understand：Without departing from the spirit and scope of the invention and the appended claims, it is various replace, to change and modifications all be possible
's.Therefore, the technical scheme that the present invention is protected should not be limited to most preferred embodiment and accompanying drawing disclosure of that.
Claims (1)
1. it is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods, it is characterised in that this method is utilized
KFDA criterions carry out target's feature-extraction, then realize target identification by SVM classifier, by by KPCA criterions and svm classifier
Device is combined, and can very well be completed target's feature-extraction and the identification of SAR image, be reduced the requirement to preprocessing process,
The azimuthal sensitivity of SAR image is overcome, the dimension of sample characteristics is have compressed, and obtains higher object recognition rate, with good
Good generalization, including following steps：
Step (1) carries out amplitude data normalization to the training objective sample of known class and the test target sample of unknown classification
Processing；
Line amplitude number is entered to the training objective sample of known class and the test target sample of unknown classification in described step (1)
It is specially according to the process of normalized：
Normalizing formula is：
<mrow>
<msub>
<mi>x</mi>
<mrow>
<mi>N</mi>
<mi>o</mi>
<mi>r</mi>
<mi>m</mi>
<mi>a</mi>
<mi>l</mi>
<mi>i</mi>
<mi>z</mi>
<mi>e</mi>
<mi>d</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mi>x</mi>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mi>x</mi>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>2</mn>
</msub>
</mrow>
</mfrac>
</mrow>
Wherein, x for any one known class training objective sample or unknown classification test target sample vector representation, i.e.,
Image array is arranged in vector form, x by rowNormalizedFor the training objective sample or unknown classification of corresponding known class
Test target sample amplitude data normalization after vector representation；
Step (2) is using KFDA criterions to the training objective sample of the known class after normalization and the test target of unknown classification
Sample data carries out feature extraction；
Using KFDA criterions to the training objective sample of the known class after normalization and unknown classification in described step (2)
Test target sample data carry out feature extraction process be specially：First seek Scatter Matrix K in classwWith class scatter matrix Kb, then
AskThe corresponding characteristic vector of nonzero eigenvalue, finally ask known class under KFDA criterions training objective sample and
The feature of the test target sample of unknown classification；Scatter Matrix K wherein in classwFor：
<mrow>
<msub>
<mi>K</mi>
<mi>w</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>c</mi>
</munderover>
<msub>
<mi>K</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<msub>
<mi>N</mi>
<mi>i</mi>
</msub>
</mfrac>
<mo>)</mo>
</mrow>
<msubsup>
<mi>K</mi>
<mi>i</mi>
<mi>T</mi>
</msubsup>
</mrow>
Wherein, N is the number of the training objective sample of known class, and c is the classification number of the training objective sample of known class,For N × NiMatrix, xpAfter the training objective samples normalization of (p=1,2 ..., N) for p-th of known class
Data,For the data after j-th of training objective samples normalization in the i-th class, NiFor the i-th class known class
The sample number of other training objective sample, k1() represents kernel function, and I is Ni×NiUnit matrix,It is for element's
Ni×NiSquare formation；If KwMatrix is a singular matrix, then makes Kw≈Kw+ κ I are to solve KwSingularity, I is and KwThe list of same order
Position battle array, κ is a very little and the disturbance constant more than zero, generally can use κ≤10-2；
Class scatter matrix KbFor：
<mrow>
<msub>
<mi>K</mi>
<mi>b</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>c</mi>
</munderover>
<msub>
<mi>N</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>G</mi>
<mn>0</mn>
</msub>
<mo>)</mo>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>G</mi>
<mn>0</mn>
</msub>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
</mrow>
Wherein, For in the i-th class
Data after j-th of training objective samples normalization,
Ask againThe corresponding characteristic vector α of nonzero eigenvalue, i.e.,：
<mrow>
<mi>&lambda;</mi>
<mi>&alpha;</mi>
<mo>=</mo>
<msubsup>
<mi>K</mi>
<mi>w</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<msub>
<mi>K</mi>
<mi>b</mi>
</msub>
<mi>&alpha;</mi>
</mrow>
Wherein, λ represents characteristic value；Training objective sample or the test of unknown classification to the known class after any one normalization
Target sample, the training objective sample for the known class that KFDA criterions are finally extracted or the test target sample characteristics of unknown classification
It is a c-1 dimensional vector, is represented by z=[z1,z2,…,zc-1]T, it is represented by per one-dimensional element：
<mrow>
<msub>
<mi>z</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<msubsup>
<mi>&alpha;</mi>
<mi>j</mi>
<mi>t</mi>
</msubsup>
<msub>
<mi>k</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
Wherein, t=1,2 ..., c-1,RepresentT-th of nonzero eigenvalue corresponding to characteristic vector j-th of element；
The training objective sample characteristics for the known class that step (3) is extracted using KFDA criterions, are trained to SVM classifier,
Produce optimal classification surface；
The training objective sample characteristics for the known class extracted in described step (3) using KFDA criterions, are entered to SVM classifier
Row is trained, and the process for producing optimal classification surface is specially：
Using Lagrange multiplier methods, functional is maximized：
<mrow>
<mi>Q</mi>
<mrow>
<mo>(</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>a</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>a</mi>
<mi>i</mi>
</msub>
<msub>
<mi>a</mi>
<mi>j</mi>
</msub>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<msub>
<mi>k</mi>
<mn>2</mn>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>z</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>z</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<msub>
<mi>a</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mn>0</mn>
</mrow>
ai>=0, i=1 ..., n
Wherein, yi∈ {+1, -1 }, corresponds to the training objective sample of two different class known class respectively；ziExtracted for KFDA criterions
I-th of known class training objective feature；k2() represents kernel function, with above-mentioned k1() is completely independent；ai
For the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked；
Step (4) is by optimal classification surface, and the feature of the test target sample for the unknown classification extracted to KFDA criterions is known
Not；
By optimal classification surface in described step (4), the spy of the test target sample for the unknown classification extracted to KFDA criterions
Levying the process being identified is specially：
The function representation of optimal classification surface is：
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>s</mi>
<mi>g</mi>
<mi>n</mi>
<mo>{</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>a</mi>
<mi>i</mi>
</msub>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<msub>
<mi>k</mi>
<mn>2</mn>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>z</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>z</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msup>
<mi>b</mi>
<mo>*</mo>
</msup>
<mo>}</mo>
</mrow>
Wherein, aiFor the Lagrange multiplier corresponding with i-th of known class training objective sample to be asked；yi∈{+1,-
1 }, the training objective sample of two different class known class is corresponded to respectively；k2() represents kernel function；ziCarried for KFDA criterions
The feature of the training objective of i-th of the known class taken；Z is the spy of the test target sample for the unknown classification that KFDA criterions are extracted
Levy；F (x) ∈ {+1, -1 }, that is, determine the institute of the test target sample of current unknown classification
Belong to classification；
This method, compared to PCA criterions, can obtain the nonlinear characteristic in image for target's feature-extraction part；
This method is for target's feature-extraction part, compared to KPCA criterions, can obtain lower intrinsic dimensionality, and with more preferable
Robustness；
This method is for target identification portion, compared to maximal correlation grader, dexterously solves problem of dimension, and algorithm is complicated
Degree is unrelated with sample dimension；
This method is for target identification portion, and compared to nearest neighbor classifier, what is obtained is globally optimal solution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410103639.2A CN103824093B (en) | 2014-03-19 | 2014-03-19 | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410103639.2A CN103824093B (en) | 2014-03-19 | 2014-03-19 | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103824093A CN103824093A (en) | 2014-05-28 |
| CN103824093B true CN103824093B (en) | 2017-10-13 |
Family
ID=50759145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410103639.2A Expired - Fee Related CN103824093B (en) | 2014-03-19 | 2014-03-19 | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103824093B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104050489B (en) * | 2014-06-27 | 2017-04-19 | 电子科技大学 | SAR ATR method based on multicore optimization |
| CN104268552B (en) * | 2014-09-04 | 2017-06-13 | 电子科技大学 | One kind is based on the polygonal fine classification sorting technique of part |
| CN105629210A (en) * | 2014-11-21 | 2016-06-01 | 中国航空工业集团公司雷华电子技术研究所 | Airborne radar space and ground moving target classification and recognition method |
| CN106845489B (en) * | 2015-12-03 | 2020-07-03 | 中国航空工业集团公司雷华电子技术研究所 | SAR image target feature extraction method based on improved Krawtchouk moment |
| CN106054189B (en) * | 2016-07-17 | 2018-06-05 | 西安电子科技大学 | Radar target identification method based on dpKMMDP models |
| TWI617175B (en) * | 2016-11-18 | 2018-03-01 | 國家中山科學研究院 | Image detection acceleration method |
| CN109784356B (en) * | 2018-07-18 | 2021-01-05 | 北京工业大学 | Matrix variable limited Boltzmann machine image classification method based on Fisher discriminant analysis |
| CN109753887B (en) * | 2018-12-17 | 2022-09-23 | 南京师范大学 | SAR image target identification method based on enhanced kernel sparse representation |
| CN111400565A (en) * | 2020-03-19 | 2020-07-10 | 北京三维天地科技股份有限公司 | Visualized dragging online data processing method and system |
| CN116776209A (en) * | 2023-08-28 | 2023-09-19 | 国网福建省电力有限公司 | Method, system, equipment and medium for identifying operation state of gateway metering device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339244A (en) * | 2008-08-01 | 2009-01-07 | 北京航空航天大学 | On-board SAR image automatic target positioning method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101373519B (en) * | 2007-08-20 | 2011-01-19 | 富士通株式会社 | Device and method for recognizing character |
| CN102567742A (en) * | 2010-12-15 | 2012-07-11 | 中国科学院电子学研究所 | Automatic classification method of support vector machine based on selection of self-adapting kernel function |
| CN103164689A (en) * | 2011-12-16 | 2013-06-19 | 上海移远通信技术有限公司 | Face recognition method and face recognition system |
-
2014
- 2014-03-19 CN CN201410103639.2A patent/CN103824093B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339244A (en) * | 2008-08-01 | 2009-01-07 | 北京航空航天大学 | On-board SAR image automatic target positioning method |
Non-Patent Citations (4)
| Title |
|---|
| "KPCA和SVM在人脸识别中的应用";李莉莉 等;《山西电子技术》;20060530(第5期);第44页第1节 * |
| "基于Curvelet和快速稀疏LSSVM的目标识别";张一凡;《中国优秀硕士学位论文全文数据库 信息科技辑》;20101215(第12期);第5.3节、第2.1-2.2节 * |
| "基于KFD+ICA特征提取的SAR图像目标实现";宦若虹;《系统工程与电子技术》;20080731;第30卷(第7期);第1237页摘要、第1238页第1节、第1239页第4.2.2节、第4.3节 * |
| 刘爱平，等."一种有效的高分辨率SAR目标特征提取与识别方法".《武汉大学学报 信息科学版》.2010,第35卷(第8期),第946-950. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103824093A (en) | 2014-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103824093B (en) | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods | |
| CN105184309B (en) | Classification of Polarimetric SAR Image based on CNN and SVM | |
| Jia et al. | Feature mining for hyperspectral image classification | |
| Al Bashish et al. | A framework for detection and classification of plant leaf and stem diseases | |
| CN106228201B (en) | A kind of anti-Deceiving interference method of synthetic aperture radar based on shade characteristic | |
| Jordanov et al. | Classifiers accuracy improvement based on missing data imputation | |
| Le et al. | Deep gabor neural network for automatic detection of mine-like objects in sonar imagery | |
| CN101968850B (en) | Method for extracting face feature by simulating biological vision mechanism | |
| CN105069478B (en) | High-spectrum remote-sensing terrain classification method based on super-pixel tensor sparse coding | |
| CN104036289A (en) | Hyperspectral image classification method based on spatial and spectral features and sparse representation | |
| Mishne et al. | Graph-based supervised automatic target detection | |
| CN105894013B (en) | Classification of Polarimetric SAR Image method based on CNN and SMM | |
| CN103116744A (en) | Fake fingerprint detection method based on markov random field (MRF) and support vector machine-k nearest neighbor (SVM-KNN) classification | |
| CN102999762B (en) | Decompose and the Classification of Polarimetric SAR Image method of spectral clustering based on Freeman | |
| CN103996047A (en) | Hyperspectral image classification method based on compression spectrum clustering integration | |
| CN108257151A (en) | PCANet image change detection methods based on significance analysis | |
| CN104103033A (en) | Image real-time processing method | |
| CN107358214A (en) | Polarization SAR terrain classification method based on convolutional neural networks | |
| Chen et al. | Locating crop plant centers from UAV-based RGB imagery | |
| Swadzba et al. | Indoor scene classification using combined 3D and gist features | |
| CN102999761A (en) | Method for classifying polarimetric SAR (synthetic aperture radar) images on the basis of Cloude decomposition and K-wishart distribution | |
| Hu et al. | Wood species recognition based on SIFT keypoint histogram | |
| Du et al. | SPOT: Selective point cloud voting for better proposal in point cloud object detection | |
| CN104134073B (en) | One kind is based on the normalized remote sensing image list class sorting technique of a class | |
| CN104077610A (en) | Method for recognizing SAR target image with two-dimensional nonlinear projection characteristic |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171013 Termination date: 20210319 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |