CN106485709A - EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection - Google Patents
EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection Download PDFInfo
- Publication number
- CN106485709A CN106485709A CN201610902077.7A CN201610902077A CN106485709A CN 106485709 A CN106485709 A CN 106485709A CN 201610902077 A CN201610902077 A CN 201610902077A CN 106485709 A CN106485709 A CN 106485709A
- Authority
- CN
- China
- Prior art keywords
- antibody
- wave band
- band
- selection
- force coefficient
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
A kind of EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection, it is related to the dimension reduction method of high spectrum image.Solve the problem that during high spectrum image waveband selection, band image be cannot be carried out with rational evaluation.The step of the present invention is：First, read in high spectrum image, generate initial antibodies wave band, select optimum antibody wave band using the affine force coefficient of Tsallis entropy redundancy calculating antibody wave band and according to affinity coefficient magnitude.2nd, clone's optimum antibody wave band generates interim antibody wave band, carries out high frequency closedown operation and selects optimum antibody wave band again.3rd, relatively low to affine force coefficient antibody wave band is replaced, and is iterated calculating, and stops until reaching given number of iterations.The present invention, by the use of Tsallis entropy redundancy as the criterion function of waveband selection, efficiently achieves the waveband selection of high spectrum image it is adaptable to the field such as high-spectrum image dimensionality reduction, data compression.
Description
Technical field
The present invention relates to the dimension reduction method of high spectrum image is and in particular to be based on Tsallis entropy redundancy and Immune Clone Selection
EO-1 hyperion band selection method.
Background technology
High-spectrum seems a kind of mass data source of collection of illustrative plates, and it had not only comprised image information but also had included spectral information,
The spectrum intensity data of each pixel on each spectral coverage can either be provided, and there is very high spectral resolution.With remote sensing skill
Art and the development of hyperspectral imager, application in military field for the high spectrum image is more and more extensive, is mainly reflected in unmanned
Machine search identification, satellite stability control, the aspect such as anti-stealthy detection and mapping, but its have wave band number excessively, data
Measure excessive etc. feature and bring very big difficulty to the classification of high spectrum image, identification.For example, information redundance's height, data
Storage requisite space is big, process time is long, and because the wave band number of high spectrum image is excessive, dimension disaster phenomenon easily,
I.e. nicety of grading declines.Therefore, in the case of ensureing terrain classification discrimination, at minimizing data volume, the dimensionality reduction economizing on resources
Reason is very important.
Feature extraction and feature selection are two kinds of main methods of high spectrum image.Processed using feature extraction
When, algorithm is complicated, computationally intensive, and is to bring, by certain change, the purpose realizing dimensionality reduction, because that it changes the thing of initial data
Manage meaning and be unfavorable for the interpretation of data.By contrast, feature selection be select from all wave bands of high spectrum image main
The band combination of effect, can not only substantially reduce the data dimension of high spectrum image, and can ratio more fully remain with
Information, more has special meaning.The main task of feature selection is to select to represent original graph under setting criterion
Feature as information.In EO-1 hyperion feature selection field, the feature extracting method of main flow is waveband selection.Waveband selection is made
For a kind of direct dimension reduction method, the suitable band combination of feature selection of Main Basiss high spectrum image, retaining original image
Physical significance and spectral characteristic reduce data dimension simultaneously, can be effectively to EO-1 hyperion dimensionality reduction.Although waveband selection
Sample is to randomly choose from original sample, but it must comprise high spectrum image all of atural object characteristic.
According to whether priori, hyperspectral image band selection method can be divided into be had supervision waveband selection and no supervises
Superintend and direct waveband selection.Supervision waveband selection is had to make full use of the prior information of known target, during waveband selection to greatest extent
Ground retains target information to be detected or makes target and background separate as much as possible, for example spectral modeling drawing, target clutter
Than etc. method.But before processing high spectrum image, in practical application, often there is no the prior information of correlation, at this moment unsupervised ripple
Section selects just can preferably complete task, selects that data volume is big and redundancy is low in the case of unknown object prior information
Band combination.
Key link in waveband selection is selection criterion and 2 points of search strategy, after the former mainly affects waveband selection
Nicety of grading, the latter mainly affects the speed searched for.Although Bayes error can be used to the error producing in selection course is entered
Row definition, but its expression formula that can not parse.Therefore, many alternative methods are suggested and are applied, and such as originate
Divergence distance in probability statistics aspect etc., they are all the upper bounds of Bayes error in the case of Gauss distribution.In addition, also one
Not strict Bayes error substituting rule, Tsallis entropy redundancy such as statistically etc. in a little mathematical meaning.
Tsallis entropy redundancy（Tsallis entropy redundancy）Letter due to its both energetic band combination
Breath amount can calculate the redundancy size within combination again, therefore has fairly obvious diving in EO-1 hyperion waveband selection field
Power, but its potentiality is not excavated completely.Because Tsallis entropy redundancy has meter in the case of calculating multiband combination
Calculate fireballing feature, and have can multi input advantage it can be seen that it in numerous band selection methods in have substantially
Advantage.
Clonal selection algorithm（Clonal Selection algorithm）It is derived from the clone in human immune system
A kind of system of selection of choosing principles.In recent years, with the development of intelligent optimization method, many searching algorithms are also used for dimensionality reduction
During, representative as clonal selection algorithm, ant group algorithm, ant colony algorithm etc..In terms of simulation result, this kind of method
Effect is better than additive method, and provides new direction for high spectrum image waveband selection research.Clonal selection algorithm introduces
Affine force coefficient, clone operations and high frequency closedown operate it is ensured that algorithm can quickly converge to globally optimal solution.Typically come
Say, antigen can regard object function and the restrictive condition of problem to be solved as, antibody can regard the solution to be selected of problem to be solved, antigen as
It is that the candidate solution of problem to be solved meets the degree of problem object function to be solved that the affine force coefficient of antibody can be regarded as.
Content of the invention
It is an object of the invention to proposing the EO-1 hyperion waveband selection side based on Tsallis entropy redundancy and Immune Clone Selection
Method, there is provided one kind, in EO-1 hyperion waveband selection, proposes and make use of Tsallis entropy redundancy to calculate and Immune Clone Selection iteration
The method calculating.It can solve the problems, such as cannot accurately calculate band combination quality degree during waveband selection.
The purpose of the present invention is achieved through the following technical solutions：Generate multiple arrays at random and form initial antibodies wave bands
Combination, each array is made up of multiple randoms number, and each random number represents one of high spectrum image wave band.Select
Tsallis entropy redundancy, as the criterion function of waveband selection, calculates the affine force coefficient of each initial antibodies wave band respectively, choosing
Select out the higher antibody wave band composition optimum antibody band group merging of affine force coefficient it is cloned and mutation operation.Remove
The partly relatively low antibody wave band of affine force coefficient remaining all antibody wave bands are changed as new initial antibodies wave band
In generation, stop iteration until reaching the iterationses specified, and selective affinity coefficient highest antibody wave band is as optimum wave band
Combination.
The flow chart of the present invention, as shown in figure 1, being divided into three steps, comprises the following steps that：
Step one：The higher antibody wave band of affine force coefficient is chosen.
1）Read in high spectrum image, define antigen and form initial antibodies ripples according to the multiple arrays of the random generation of coding rule
Duan Zuhe, the array that coding rule is randomly generated is made up of multiple randoms number, and each random number represents in high spectrum image
One wave band, respectively in calculating antibody band combination each array affine force coefficient.
High spectrum image is carried out during waveband selection, it is necessary first to select to evaluate the criterion function of band combination, coming with this
Weigh the effect of waveband selection.Tsallis entropy redundancy can preferably embody the correctness of identification and classification, therefore selects
Tsallis entropy redundancy is as criterion function.
High spectrum imageSIn each band imageM n Data available sequence (m n (1),m n (2),…,m n (k)) representing, wherein
1≤n≤N,NWave band number for high spectrum image.This data sequence is normalized：
,.
OrderP={ n (k), thenSCorrelation matrixCIt is represented by：
.
Correlation matrixCDecompose gainedCAnd unit matrixIRepresent cross-correlation matrix and autocorrelation matrix respectively,C’Diagonal
It is all 0.Orderλ n C Represent correlation matrixCEigenvalue, thenλ n C ReflectCMiddle amount of correlated information, wherein 1≤n≤N,λ n C Represent phase
Close matrixC'sNIndividual eigenvalue.Additionally,λ n C Also meet：
,
.
Using above-mentioned relation, correlation matrix can be definedCTsallis entropyH r For：
H r =[ ∑(λ n C /N) r -1]/(1-r),
Wherein,rFor Tsallis entropy coefficient.IfSIn each wave band be completely independent, thenCAll elements be all 0,λ n C It is all 1,H r Take
Obtain maximum（r>1）.IfSIn each wave band data perfectly correlated, thenCAll elements be all,λ n C Middle only one isN（Remaining is
0）,H' obtain minima（r>1）.In order to the quantity of information of band combination can be quantified, therefore define redundancyR r For：
.
As can be seen that working asrWhen=1, this formula cannot directly calculate, and therefore can take its limiting form, i.e. famous perfume (or spice)
Agriculture entropy.The Tsallis entropy redundancy of correlation matrix is represented by：
,.
The Tsallis entropy redundancy of correlation matrix can quantify the quantity of information of band combination, and also has the spy of multi input
Property.Because we enrich and the low band combination of redundancy data volume to be selected, we can be selected or structure using additive method
Build out one with reference to wave bandM, and construct criterion functionE r For：
E r (S)=[∑R r (M,R1,R2,R3)]/R r (S).
By relatively each band combinationE r Size, you can judge the quality of this band combination in terms of comprehensive.
It is first according to coding rule and generates initial antibodies band combination at random, this ensure that the randomness of antibody wave band.
Using the affine force coefficient of Tsallis entropy redundancy calculating antibody wave band and carry out select optimum antibody band combination, can select
Go out preferable antibody wave band and eliminate poor quality antibody wave band.
2）Select Tsallis entropy redundancy as the criterion function of the affine force coefficient of calculating antibody wave band, according to affinity
Coefficient magnitude is selected from initial antibodies band combinationwThe larger antibody wave band composition optimum antibody wave band of individual affine force coefficient
CombinationP, whereinwFor the optimum antibody wave band number being manually set,PIt is to be made up of the larger antibody wave band of multiple affine force coefficients
Optimum antibody band combination.
Step 2：Optimum antibody band combination is cloned and is made a variation.
1）To optimum antibody band combinationPInwIndividual antibody wave band is cloned, and generates interim antibody band combination.Gram
Grand propagationgenTimes, obtainw×genIndividual antibody wave band forms interim antibody band combination, whereingenTake { 2,3 }.Clone operations
It is that selected antibody wave band is replicated, the probability of clone operations is affine to antibody wave band, and force coefficient is directly proportional, antibody ripple
Duan Qinhe force coefficient is bigger, and the antibody wave band obtaining after clone is more, for exampleaAntibody wave band sum after the clone of individual antibody wave band
WithAIt is expressed as follows：,
Wherein,kFor cloning coefficient,bFor antibody wave band according to the sequence number after affine force coefficient descending,βFor antibody wave band parent
And force coefficient, round is rounding operation.
Because the purpose of this process is to select optimum antibody wave band and it is cloned, antagonist wave band is not therefore needed to enter
Row selects, and after setting the clone of each antibody wave band, number is identical.
2）High frequency closedown operation, mutation probability and the affine force coefficient of antibody wave band are carried out to the antibody wave band obtaining after clone
It is inversely proportional to, thezThe mutation probability of individual antibody wave bandV z （zSpan arrives for 0w×gen）For：,
Wherein,αFor the coefficient of variation set in advance,β z It iszThe affine force coefficient of individual antibody wave band,β max It is all antibody ripples
The maximum of affine force coefficient in section.
As can be seen that the affine force coefficient of antibody wave band is bigger, mutation probability is less.According to the mutation probability calculating, with
Machine ground changes the numerical value on some positions in the antibody wave band after cloning.Mutation operation is random 2 points of variations, randomly selects
Select the numerical value on the internal diverse location of two antibody wave bands and go forward side by side the exchange of line number value so that antibody wave band changes, conversion
Affine force coefficient is inversely proportional to probability with antibody wave band, has so both reached the purpose of random variation, also ensure that antibody wave band parent
The little setting of the antibody wave band mutation probability big with force coefficient.High frequency closedown process makes the affine force coefficient of antibody wave band in office
Constantly become big in portion, thereby serve to the purpose of microcosmic optimizing regulation and control, and the appearance to optimal solution plays a positive role.
3）Recalculate the affine force coefficient of antibody wave band after mutation process, and big according to the affine force coefficient of antibody wave band
Little, select from antibody band combinationwIndividual antibody wave band the is affine larger optimum antibody wave band composition optimum antibody ripple of force coefficient
Duan ZuheQ,WhereinwIn step onewFor same variable.
Step 3：Constantly iteration calculates, and until reaching given number of iterations iteration stopping, selects antibody wave band parent
Maximum antibody wave band is as optimal bands composite with force coefficient.
With the optimum antibody band combination obtaining in step 2QReplace the antibody wave band that affine force coefficient is relatively low,And return
Step one the 2nd）Stepping row iteration calculates.Iterationses are relevant with the purpose of waveband selection, and iterationses are more, and Selection effect is got over
Good, but selection time is longer, the selection of concrete iterationses is determined by experimental data, and the range of choice of iterationses is 50 ~ 500
Between.
After iterative process terminates, the antibody wave band affine force coefficient highest occurring during selecting whole waveband selection resists
Bulk wave section is as high spectrum image waveband selection result.
The present invention compared with prior art has the advantage that：
The present invention makes improvement using Tsallis entropy redundancy to clonal selection algorithm, has reached to high spectrum image band group
Close and carry out the purpose evaluated of high accuracy, select that nicety of grading is higher, can comprehensively reflect the ripple of high spectrum image information
Duan Zuhe.In traditional hyperspectral image band selection method based on Immune Clone Selection, evaluation criterion is typically chosen existing standard
Then function, such as Shannon entropy, optimum index, Jeffreys-Matusita distance, spectral modeling drawing etc., but these criterion functions
All there are some problems makes the nicety of grading of selection result not high, and the Tsallis entropy redundancy being constructed by Tsallis entropy
Degree both can quantify the quantity of information that band combination is comprised, and can embody the redundancy size within combination, therefore can conduct
A kind of high accuracy criterion function and be chosen.Further, since it has multi input characteristic, also for the affine force coefficient of antibody wave band
Calculating provides conveniently.The present invention can improve the nicety of grading of selection result, can accelerate antibody wave band in iterative process again
The calculating of affine force coefficient, is one kind preferably band selection method.
Brief description
Fig. 1 is the flow chart of the present invention.
Fig. 2 is the overall classification accuracy comparison diagram of present invention selection result in the case of selecting different number wave bands.
Fig. 3 is the average classification accuracy comparison figure of present invention selection result in the case of selecting different number wave bands.
Specific embodiment
Specific embodiment with reference to embodiment and the brief description present invention：Will based on Tsallis entropy redundancy and
The EO-1 hyperion band selection method of Immune Clone Selection applies in high spectrum image waveband selection.
Provide the description of hyperspectral image data first：
Experimental subject is Indian Pines hyperspectral image data, and wave-length coverage is 400nm-2500nm, comprises 224 ripples
Section, size is 145 × 145pixel.Eliminate some wave bands of Atmospheric Absorption interference in data set, leave 220 wave bands and make
For experimental subject, view data is designated asIMG (145×145,220).
Select in the present embodimentwSpan be 5 ~ 30, from 220 dimension wave bands select the wave band containing important information
Carry out dimensionality reduction, select, from the wave band that one group of quantity is 220, the band combination that quantity is 5 ~ 30, make antibody wave band affine
Force coefficient reaches maximum, and in search procedure, the wave band number of band-limiting combination is 5 ~ 30.Take in the present embodimentr=3.
Execution step one：Input hyperspectral image dataIMG (145×145,220)With corresponding label.Define antigen and generate
30 initial antibodies wave bands form initial antibodies band combination, and each antibody wave band is made up of 5 ~ 30 randoms number, and each is random
The span of number is 1 ~ 220.Calculate the affinity coefficient magnitude of each antibody wave band using Tsallis entropy redundancy, takewFor
10, select the larger optimum antibody wave band composition of the affine force coefficient of 10 antibody wave bands according to affinity coefficient magnitude optimal
Antibody band combinationP.
Execution step two：10 optimum antibody wave bands in step one are cloned, that is, carries out duplication interim to generate
Antibody band combination, takes clone's multiple to be 2, and symbiosis becomes 20 clone's new antibodies wave bands.Clone's new antibodies wave band is carried out
Mutation operation, takes the coefficient of variationαFor 1.3, variation method adopts random 2 points of variations, the selection of mutation probability and antibody wave band
Affinity coefficient magnitude is relevant, and antibody wave band is affine, and force coefficient is bigger, and mutation probability is less.After variation, after recalculating variation
The affine force coefficient of antibody wave band, and select the affine force coefficient of 10 antibody wave bands according to antibody wave band affinity coefficient magnitude
Larger optimum antibody wave band composition optimum antibody band combinationQ.
Execution step three：Use optimum antibody band combinationQReplace initial antibodies band combination in 20 affine force coefficients relatively
Low antibody wave band returning to is iterated in step one calculating, and when iterationses reached for 100 generation, stops iteration and choosing
The antibody wave band affine force coefficient highest antibody wave band occurring during selecting whole waveband selection is as high spectrum image wave band
Selection result.
The present embodiment conclusion：Waveband selection the results are shown in Table 1.Present invention selection result in the case of selecting different number wave bands
Overall classification accuracy comparison diagram see Fig. 1 and Fig. 2 with average classification precision result.Using based on Tsallis entropy redundancy and gram
The EO-1 hyperion band selection method of grand selection has carried out 7 waveband selections to high spectrum image, selects 5,10,15,20 respectively,
25,30 wave bands.It can be seen that in the case of selecting 5 wave bands, the nicety of grading of band combination relatively low it is impossible to generation well
The fundamental characteristics of table high spectrum image.In the case of selecting 10 ~ 25 wave bands, the nicety of grading of band combination is higher, and selects
Select result to be evenly distributed in all wave bands, the fundamental characteristics of high spectrum image can be represented well, Selection effect is preferable.?
Select 30 wave bands in the case of, the nicety of grading of band combination is also higher, but with select 10 ~ 25 wave bands in the case of carry out
Contrast, nicety of grading does not improve too much, and increased the waveband selection time, therefore selects 10 ~ 25 wave bands to be optimum selection.
The present invention can select that nicety of grading is higher and band combination that can more fully reflect high spectrum image information.
Table 1 present invention selection result in the case of selecting different number wave bands compares
| Select wave band number | High spectrum image waveband selection result |
| 5 | 21,38,86,103,131 |
| 10 | 5,19,23,24,26,47,90,112,114,130 |
| 15 | 12,26,51,54,68,79,101,109,118,121,138,150,167,192,211 |
| 20 | 6,20,31,43,45,52,84,94,96,99,103,116,127,133,143,152,167,182, 199,215 |
| 25 | 4,23,37,39,49,65,67,76,87,95,102,104,129,133,143,150,153,154, 157,165,172,191,202,211,219 |
| 30 | 7,20,25,33,37,43,62,69,72,76,79,90,95,98,100,106,108,115,116, 120,132,134,153,161,170,182,185,189,191,202 |
Claims (4)
1. a kind of EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection is it is characterised in that it includes following step
Suddenly：
Step one：Read in high spectrum image, define antigen and form initial antibodies according to the multiple arrays of the random generation of coding rule
Band combination, respectively in calculating antibody band combination each array affine force coefficient；Select Tsallis entropy redundancy conduct
The criterion function of calculating antibody wave band is affine force coefficient, selects from initial antibodies band combination according to affinity coefficient magnitude
Optimum antibody band combinationP；
Step 2：To optimum antibody band combinationPIn antibody wave band clonal expansiongenTimes；To the antibody ripple obtaining after clone
Duan Jinhang high frequency closedown operates, and affine force coefficient is inversely proportional to mutation probability with antibody wave band；Select from antibody band combinationwIndividual
Antibody wave band is affine, and the larger optimum antibody wave band of force coefficient forms optimum antibody band combinationQ；
Step 3：Constantly iteration calculates, and until reaching given number of iterations iteration stopping, selects antibody wave band affinity
The maximum antibody wave band of coefficient is as optimal bands composite.
2. the EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection according to claim 1, its feature exists
In described step one it is：
1）Read in high spectrum image, define antigen and form initial antibodies band group according to the multiple arrays of the random generation of coding rule
Close, the array that coding rule is randomly generated is made up of multiple randoms number, and each random number represents one of high spectrum image
Wave band, respectively in calculating antibody band combination each array affine force coefficient；
High spectrum imageSIn each band imageM n Data available sequence (m n (1),m n (2),…,m n (k)) representing, wherein 1≤n
≤N,NWave band number for high spectrum image；This data sequence is normalized：
,；
OrderP={ n (k), thenSCorrelation matrixCIt is represented by：；
Correlation matrixCDecompose gainedCAnd unit matrixIRepresent cross-correlation matrix and autocorrelation matrix respectively,C’Diagonal be all
0；Orderλ n C Represent correlation matrixCEigenvalue, thenλ n C ReflectCMiddle amount of correlated information, wherein 1≤n≤N,λ n C Represent Correlation Moment
Battle arrayC'sNIndividual eigenvalue；Additionally,λ n C Also meet：,；
Using above-mentioned relation, correlation matrix can be definedCTsallis entropyH r For：H r =[ ∑(λ n C /N) r -1]/(1-r),
Wherein,rFor Tsallis entropy coefficient；IfSIn each wave band be completely independent, thenCAll elements be all 0,λ n C It is all 1,H r Take
Obtain maximum（r>1）；IfSIn each wave band data perfectly correlated, thenCAll elements be all,λ n C Middle only one isN（Remaining is
0）,H' obtain minima（r>1）；In order to the quantity of information of band combination can be quantified, therefore define redundancyR r For：；
As can be seen that working asrWhen=1, this formula cannot directly calculate, and therefore can take its limiting form, i.e. famous Shannon entropy；Phase
The Tsallis entropy redundancy closing matrix is represented by：,；
The Tsallis entropy redundancy of correlation matrix can quantify the quantity of information of band combination, and also has the characteristic of multi input；Cause
Enrich and the low band combination of redundancy for our data volumes to be selected, we can be selected using additive method or construct one
Individual reference wave bandM, and construct criterion functionE r For：E r (S)=[∑R r (M,R1,R2,R3)]/R r (S)；
2）Select Tsallis entropy redundancy as the criterion function of the affine force coefficient of calculating antibody wave band, according to affine force coefficient
Size is selected from initial antibodies band combinationwThe larger antibody wave band composition optimum antibody band combination of individual affine force coefficientP, whereinwFor the optimum antibody wave band number being manually set,PIt is to be made up of the larger antibody wave band of multiple affine force coefficients
Good antibody band combination.
3. the EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection according to claim 1, its feature exists
In described step 2 it is：
1）To optimum antibody band combinationPInwIndividual antibody wave band is cloned, and generates interim antibody band combination；Clone increases
GrowgenTimes, obtainw×genIndividual antibody wave band forms interim antibody band combination, whereingenTake { 2,3 }；Clone operations are right
Selected antibody wave band is replicated, and the probability of clone operations is affine to antibody wave band, and force coefficient is directly proportional, antibody wave band parent
Bigger with force coefficient, the antibody wave band obtaining after clone is more, for exampleaAfter the clone of individual antibody wave band, antibody wave band sum is usedA
It is expressed as follows：, wherein,kFor cloning coefficient,bArrange according to affine force coefficient descending for antibody wave band
Sequence number after row,βFor the affine force coefficient of antibody wave band, round is rounding operation；
Because the purpose of this process is to select optimum antibody wave band and it is cloned, antagonist wave band is not therefore needed to be selected
Select, after setting the clone of each antibody wave band, number is identical；
2）High frequency closedown operation is carried out to the antibody wave band obtaining after clone, mutation probability is affine with antibody wave band, and force coefficient becomes anti-
Than, thezThe mutation probability of individual antibody wave bandV z （zSpan arrives for 0w×gen）For：,
Wherein,αFor the coefficient of variation set in advance,β z It iszThe affine force coefficient of individual antibody wave band,β max It is all antibody wave bands
In affine force coefficient maximum；
3）Recalculate the affine force coefficient of antibody wave band after mutation process, and according to antibody wave band affinity coefficient magnitude, from
Select in antibody band combinationwIndividual antibody wave band the is affine larger optimum antibody wave band composition optimum antibody band combination of force coefficientQ,WhereinwIn step onewFor same variable.
4. the EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection according to claim 1, its feature exists
In described step 3 it is：
With the optimum antibody band combination obtaining in step 2QReplace the antibody wave band that affine force coefficient is relatively low,And return to step
One the 2nd）Stepping row iteration calculates；Iterationses are relevant with the purpose of waveband selection, and iterationses are more, and Selection effect is better,
But selection time is longer, the selection of concrete iterationses is determined by experimental data, the range of choice of iterationses 50 ~ 500 it
Between；After iterative process terminates, select the antibody wave band affine force coefficient highest antibody ripple of appearance during whole waveband selection
Duan Zuowei high spectrum image waveband selection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610902077.7A CN106485709A (en) | 2016-10-17 | 2016-10-17 | EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610902077.7A CN106485709A (en) | 2016-10-17 | 2016-10-17 | EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106485709A true CN106485709A (en) | 2017-03-08 |
Family
ID=58270139
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610902077.7A Pending CN106485709A (en) | 2016-10-17 | 2016-10-17 | EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106485709A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437098A (en) * | 2017-07-31 | 2017-12-05 | 湖北工业大学 | A kind of hyperspectral image band selection method based on the improved binary ant colony algorithm of differential evolution |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101826160A (en) * | 2010-03-31 | 2010-09-08 | 北京航空航天大学 | Hyperspectral image classification method based on immune evolutionary strategy |
| CN101853392A (en) * | 2010-04-21 | 2010-10-06 | 河海大学 | Remote sensing hyperspectral image band selection method based on conditional mutual information |
| CN104318515A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Hyper-spectral image wave band dimension descending method based on NNIA evolutionary algorithm |
-
2016
- 2016-10-17 CN CN201610902077.7A patent/CN106485709A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101826160A (en) * | 2010-03-31 | 2010-09-08 | 北京航空航天大学 | Hyperspectral image classification method based on immune evolutionary strategy |
| CN101853392A (en) * | 2010-04-21 | 2010-10-06 | 河海大学 | Remote sensing hyperspectral image band selection method based on conditional mutual information |
| CN104318515A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Hyper-spectral image wave band dimension descending method based on NNIA evolutionary algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 张淼 等: "基于Tsallis熵冗余度的超谱特征选择算法性能评价", 《光电子*激光》 * |
| 杨三美: "基于克隆选择算法的高光谱图像波段选择", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437098A (en) * | 2017-07-31 | 2017-12-05 | 湖北工业大学 | A kind of hyperspectral image band selection method based on the improved binary ant colony algorithm of differential evolution |
| CN107437098B (en) * | 2017-07-31 | 2019-11-26 | 湖北工业大学 | A kind of hyperspectral image band selection method based on the improved binary ant colony algorithm of differential evolution |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huang et al. | Dsanet: Dual self-attention network for multivariate time series forecasting | |
| Dainese et al. | Report on the Physics at the HL-LHC, and Perspectives for the HE-LHC | |
| Azzi et al. | Standard Model physics at the HL-LHC and HE-LHC | |
| Aad et al. | Comprehensive measurements of t-channel single top-quark production cross sections at s= 7 TeV with the ATLAS detector | |
| Baldi et al. | Improved energy reconstruction in NOvA with regression convolutional neural networks | |
| Li et al. | Software defect prediction based on ensemble learning | |
| CN112052755A (en) | Semantic convolution hyperspectral image classification method based on multi-path attention mechanism | |
| CN107315765A (en) | A kind of method of the concentrated-distributed proximity search of extensive picture | |
| Józefowicz et al. | Potential for optimizing the Higgs boson CP measurement in H→ τ τ decays at the LHC including machine learning techniques | |
| CN100354864C (en) | A method of feature selection based on mixed mutual information in data mining | |
| CN104504121A (en) | Video retrieval method and device | |
| Heldmann et al. | An improved tau-Identification for the ATLAS experiment | |
| Deliot et al. | Top quark physics at the Tevatron using tt¯ events | |
| CN104318515A (en) | Hyper-spectral image wave band dimension descending method based on NNIA evolutionary algorithm | |
| Wei et al. | Efficient local histogram searching via bitmap indexing | |
| CN109800815B (en) | Training method, wheat recognition method and training system based on random forest model | |
| Irles et al. | Experimental prospects for precision observables in $ e^{-} e^{+}\rightarrow q\bar {q} $ with $ q= b, c $ processes at the ILC operating at 250 and 500 GeV of center of mass | |
| Khan et al. | A comparative study on solar power forecasting using ensemble learning | |
| CN106485709A (en) | EO-1 hyperion band selection method based on entropy redundancy and Immune Clone Selection | |
| CN105528623B (en) | A kind of imaging spectrum sparse representation method based on atural object category classification redundant dictionary | |
| CN104616027A (en) | Non-adjacent graph structure sparse face recognizing method | |
| CN106101732A (en) | The vector quantization scheme of Fast Compression EO-1 hyperion signal | |
| CN111553338A (en) | Hyperspectral feature selection method based on simulated annealing algorithm | |
| Dobbs et al. | Les Houches' Physics at Tev Colliders 2003'QCD/SM Working Group: Summary Report | |
| Turkar | Polarimetric SAR image classification by using artificial neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170308 |