CN104200217A - Hyperspectrum classification method based on composite kernel function - Google Patents
Hyperspectrum classification method based on composite kernel function Download PDFInfo
- Publication number
- CN104200217A CN104200217A CN201410386737.1A CN201410386737A CN104200217A CN 104200217 A CN104200217 A CN 104200217A CN 201410386737 A CN201410386737 A CN 201410386737A CN 104200217 A CN104200217 A CN 104200217A
- Authority
- CN
- China
- Prior art keywords
- function
- mapping
- phi
- vector machine
- kernel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 239000002131 composite material Substances 0.000 title abstract description 8
- 230000006870 function Effects 0.000 claims abstract description 63
- 238000012360 testing method Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000012706 support-vector machine Methods 0.000 claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 21
- 238000002790 cross-validation Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims abstract description 8
- 238000010276 construction Methods 0.000 claims abstract description 4
- 150000001875 compounds Chemical class 0.000 claims description 46
- 238000013507 mapping Methods 0.000 claims description 43
- 230000008689 nuclear function Effects 0.000 claims description 43
- 239000000203 mixture Substances 0.000 claims description 17
- 238000001228 spectrum Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract description 11
- 230000008859 change Effects 0.000 abstract description 2
- 238000002474 experimental method Methods 0.000 description 7
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 4
- 235000011613 Pinus brutia Nutrition 0.000 description 4
- 241000018646 Pinus brutia Species 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000013480 data collection Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012729 kappa analysis Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a hyperspectrum classification method based on a composite kernel function. The hyperspectrum classification method comprises the steps of inputting a set of hyperspectrum images in N classes, taking a support vector machine as a base classifier, and meanwhile, randomly selecting S samples from every classes of the hyperspectrum images to form a training set and forming a test set with the left samples, determining the change range of each parameter, next, determining the optimal performance parameters, including a penalty factor and a kernel of the support vector machine by virtue of cross validation for K times, constructing a composite kernel function by use of a composite kernel construction policy and training the support vector machine, and circulating for N times by use of the parameters of a support vector machine decision function obtained in the training process to obtain decision function values of which the test set belongs to every classes and to form a matrix as shown in the specification, and then determining multiple classifier policies, namely finding the maximum values of every columns of the matrix as shown in the specification. The hyperspectrum classification method based on the composite kernel function has the characteristics of better description of distribution features of a data set, relatively high classification accuracy and the like. The time taken by parameter optimization of the hyperspectrum classification method is also relatively short in contrast with a traditional multi-kernel learning method.
Description
Technical field
What the present invention relates to is a kind of sorting technique of high spectrum image, particularly a kind of hyperspectral classification method based on new compound nucleus function (Hyperspectral Image Classification Based on A New Composite Kernel).
Background technology
In recent years, the development of satellite sensor had improved spatial resolution and spectral resolution, had simultaneously also shortened the access time, and then had created condition for the development of hyperspectral classification method.Neural network classifier, k nearest neighbor sorter, Bayes classifier, decision tree classifier and the sorter based on core etc. has been widely used in high spectrum field, and wherein the sorting technique based on kernel function is more and more paid close attention to.Support vector machine (Support Vector Machine, SVM) be the most typical sorting technique based on kernel function, in the time processing limited higher-dimension training sample, still can obtain good classification performance, make it in hyperspectral classification, occupy certain status.In view of the polyphyly of real data and the diversity of expression way, traditional monokaryon has been not enough to practical requirement.Nearly ten years, Multiple Kernel Learning (Multiple Kernel Learning, MKL) technology has obtained paying close attention to and development, mainly improves from four aspects: the 1) improvement of Multiple Kernel Learning strategy; As the simple Multiple Kernel Learning (Simple-MKL) that Rakotomamonjy etc. proposes, it uses gradient descent method to solve Multiple Kernel Learning problem.2) improvement of parameter optimization; The propositions such as Li are not used the parameter optimization mode of any protruding bound term.3) improvement of multiple kernel function combinations; As Xia etc. has utilized the combination of boosting.4) improvement of strategies; As the people such as Gu propose the representative Multiple Kernel Learning of hyperspectral classification.The most common Multiple Kernel Learning method is mixed nucleus function (Mixture Kernels, MKs), and by multiple kernel function linear weighted function summations, the validity of this kernel function has obtained checking and has been widely used in hyperspectral classification field.But it is in training process, and the time that parameter optimization consumes is huge also.
Summary of the invention
The object of the present invention is to provide a kind of better distribution characteristics of descriptor data set, and nicety of grading is relatively high, a kind of hyperspectral classification method based on compound nucleus function that the time that parameter optimization consumes is relatively short.
The object of the present invention is achieved like this:
Step 1: input one panel height spectrum picture, classification number is N;
Step 2: taking support vector machine as base sorter, from the each classification of described high spectrum image, choose randomly s sample composition training set simultaneously, residue sample composition test set, determine the variation range of each parameter, then the optimal performance parameter of determining support vector machine in conjunction with K cross validation, comprises penalty factor and nuclear parameter;
Step 3: utilize compound nucleus construction strategy, structure compound nucleus function, trains support vector machine;
Step 4: utilize the parameter of the support vector machine decision function that training process obtains, circulation N time, and then obtain the decision function value that test set belongs to every classification, composition matrix
wherein n
testrepresent the number of test sample book;
Step 5: determine multi-categorizer strategy, find matrix
the maximal value of every row, the prediction label of the corresponding each test sample book of its row sequence number,
i=1 ..., n
test.
Feature of the present invention is:
1, in parameter setting up procedure, the selection of base sorter can be used other sorters based on core to replace support vector machine.
2, the determining of optimal performance parameter in parameter setting up procedure, is the mode that adopts grid search and K cross validation to combine.
3, in training process, the structure of compound nucleus function obtains by Nonlinear Mapping repeatedly, is calculated as follows:
K
1(x,z)=φ
1(x)·φ
1(z)
K
2(x,z)=φ
2[φ
1(x)]·φ
2[φ
1(z)]
...
K
M(x,z)=φ
M[φ
M-1(x)]·φ
M[φ
M-1(z)]
Wherein K
m(x, z) represents the kernel function of sample x and z, φ
mrepresent Nonlinear Mapping function the M time.φ
mcan be Gaussian mapping, polynomial expression mapping or other Nonlinear Mapping, and in the time of M=2, the classification performance based on this compound Kernel function classifier can reach convergence state.
4,, in the time of M=2, the parameter optimization time of compound nucleus function will be much smaller than the parameter optimization time of traditional mixed nucleus function.
Based on These characteristics, a kind of hyperspectral classification method based on new compound nucleus function that the present invention proposes has the better distribution characteristics of descriptor data set, and relatively high of nicety of grading.Meanwhile, the time that its parameter optimization consumes is also relatively short with respect to traditional Multiple Kernel Learning method.
Brief description of the drawings
Fig. 1 is the hyperspectral classification method flow diagram based on a kind of compound nucleus function of the present invention.
Fig. 2 a-Fig. 2 b is high spectrum image Indian Pines, wherein: Fig. 2 a is that gray-scale map, Fig. 2 b are reference features classification.
Fig. 3 a-Fig. 3 b is bimonthly thread form simulated data sets classification boundaries vision figure, wherein: Fig. 3 a is that gaussian kernel function Gauss classification boundaries vision figure, Fig. 3 b are continuous Gaussian mapping compound nucleus function G (G) classification boundaries vision figure.Figure hollow core symbology test sample book, filled symbols represents training sample.
Fig. 4 is for high-spectral data collection Indian Pines different IPs function category performance comparison sheet 1.In form, Linear represents single linear kernel function, Polynomial represents single polynomial kernel function, Gauss represents single gaussian kernel function, MKs represents mixed nucleus function, G (G) represents continuous Gaussian mapping compound nucleus function, G (P) representative polynomial-Gaussian mapping compound nucleus function, P (G) represents Gauss-polynomial expression mapping compound nucleus function, P (P) represents continuous polynomial expression mapping compound nucleus function.
Fig. 5 is that different IPs function parameter is optimized time contrast table 2.
Fig. 6 is the different training sample number table 3 that affects on different IPs function category performance.
Embodiment
For example the present invention is described in more detail below in conjunction with accompanying drawing.
The present invention is a kind of hyperspectral classification method based on new compound nucleus function, comprises input process, parameter setting, training process, assorting process and output procedure five steps.Input process inputs a panel height spectrum picture; Parameter setting is the process of initialization and parameter optimization; Training process is taking support vector machine as base sorter, the process of training base sorter model; Assorting process is the model parameter of utilizing said process to obtain, thereby provides the decision function value process that test set belongs to every classification; Output procedure is to determine multi-categorizer strategy, and provides the process of test sample book prediction label.Provide detailed process below:
Concrete analysis step is as follows:
Step S1: input process.Input a panel height spectrum picture, N represents classification number.
Step S2: parameter setting.From image, in each classification, choose randomly s sample composition training set, residue sample composition test set.Determine the variation range of each parameter, then determine the optimal performance parameter of support vector machine in conjunction with K cross validation.This step is further comprising the steps:
Step S2.1: sample set determine; From N classification of high spectrum image, every class is chosen s sample composition randomly label training sample set D={ (x
1, y
1) ..., (x
n, y
n), wherein x
irepresent i spectral signature that has exemplar, y
irepresent sample x
ilabel, n=16s represents training sample number.Residue sample composition test set
n
testrepresent test sample book number.
Step S2.2: parameter area determine; Adopting polynomial kernel function and gaussian kernel function is that basic kernel function is carried out the synthetic of compound nucleus function.Polynomial kernel is K
p(x, z)=φ
p(x) φ
p(z)=[(xz)+1]
d, its parameter is d, φ
p(x) the polynomial expression mapping function of expression sample x.Gaussian kernel is K
g(x, z)=φ
g(x) φ
g(z)=exp (|| x-z||
2/ σ
2), its Gauss's radius is σ, φ
g(x) the Gaussian mapping kernel function of expression sample x.The penalty factor of support vector machine, the variation range of Gauss's radius sigma and d is respectively: { 2
0, 2
1..., 2
8, { 2
-5, 2
-4..., 2
1and { 2
-2, 2
-1.., 2
4.
S2.3:K cross validation of step; In each parameter combinations situation, training sample set D is divided into K subset, retain successively a subset for test, remaining K-1 subset is used for training svm classifier device model.Cross validation repeats K time, and each subset is verified once, and the mean value of calculating K subseries precision.In the time that mean accuracy reaches maximal value, illustrate that this parameter combinations is optimum.
Step S3: training process.Utilize compound nucleus construction strategy, structure compound nucleus function, trains support vector machine.This step is further comprising the steps:
Step S3.1: the structure of compound nucleus function; Obtain compound nucleus function by Nonlinear Mapping repeatedly, be calculated as follows:
K
1(x,z)=φ
1(x)·φ
1(z)
K
2(x,z)=φ
2[φ
1(x)]·φ
2[φ
1(z)]
...
K
M(x,z)=φ
M[φ
M-1(x)]·φ
M[φ
M-1(z)]
Wherein K
m(x, z) represents the kernel function of sample x and z, φ
mrepresent Nonlinear Mapping function the M time.φ
mcan be Gaussian mapping, polynomial expression mapping or other Nonlinear Mapping, and in the time of M=2, the classification performance based on this compound Kernel function classifier can reach convergence state.In the time of M=2, K
g (G), K
g (P), K
p (G)and K
p (P)be respectively continuous Gaussian mapping compound nucleus function, polynomial expression-Gaussian mapping compound nucleus function, Gauss-polynomial expression mapping compound nucleus function and continuously polynomial expression mapping compound nucleus function.Expression formula is as follows:
K
G(G)(x,z)=φ
G[φ
G(x)]·φ
G[φ
G(z)]
=exp[-||φ
G(x)-φ
G(z)||
2/σ
2 2]
(1)
=exp[-[φ
G(x)·φ
G(x)+φ
G(z)·φ
G(z)-2φ
G(x)·φ
G(z)]/σ
2 2]
=exp[-[K
G(x,x)+K
G(z,z)-2K
G(x,z)]/σ
2 2]
K
G(P)(x,z)=φ
G[φ
P(x)]·φ
G[φ
P(z)]
=exp[-||φ
P(x)-φ
P(z)||
2/σ
2 2]
(2)
=exp[-[φ
P(x)·φ
P(x)+φ
P(z)·φ
P(z)-2φ
P(x)·φ
P(z)]/σ
2 2]
=exp[-[K
P(x,x)+K
P(z,z)-2K
P(x,z)]/σ
2 2]
Wherein x and z represent two pixels, σ
2represent Gauss's radius of Gaussian mapping for the second time, d
2represent the polynomial parameters of polynomial expression mapping for the second time.In attention formula (1) (3), Gauss's radius of Gaussian mapping is σ for the first time
1, similarly, in formula (2) (4), the polynomial parameters of polynomial expression mapping is for the first time d
1.
Step S3.2: the training of support vector machine; Support vector machine is trained to the weight vectors of supported vector machine decision function and threshold alpha
*and b
*.
Step S4: assorting process.Utilize the parameter of the support vector machine decision function that training process obtains, circulation N time, and then obtain the decision function value that test set belongs to every classification, composition matrix
wherein n
testrepresent the number of test sample book.
Step S5: output procedure.Determine multi-categorizer strategy, find matrix
the maximal value of every row, the prediction label of the corresponding each test sample book of its row sequence number,
i=1 ..., n
test.
Step S5.1: multi-categorizer strategy determine; Adopt " a pair of remaining " multi-categorizer strategy that many classification problems are changed into multiple two classification problems.
Step S5.2: the calculating of nicety of grading; Calculate respectively overall nicety of grading (the Overall Accuracy that adopts support vector machine in different composite kernel function situation, OA), Kappa coefficient (Kappa Statistic, Kappa), the nicety of grading of average nicety of grading (Average Accuracy, AA) and each classification.
For validity of the present invention is described, spy carries out following experimental demonstration.Experimental data is from bimonthly thread form (Two Moons) simulated data sets and true high-spectral data collection (Indian Pines).
1) Two Moons: this simulated data sets comprises two classifications, contains respectively 96 and 104 pixels, and each pixel is by two character representations.
2) Indian Pines: this high-spectrum remote sensing is from the Indian agricultural in the Indiana, USA northwestward of obtaining for 1992, and it comprises 144 × 144 pixels, 16 classifications, 220 wave bands, because the factors such as noise are removed 20 wave bands wherein.Image is removed the monitoring data that has that comprises 16 class vegetation beyond background.
First parameters more of the present invention are set: first from every class, choose randomly s sample composition D={ (x
1, y
1) ..., (x
n, y
n), residue sample composition test set
the method of selecting ten cross validations (K=10) and grid search to combine is estimated optimum parameter, the penalty factor of support vector machine, and the variation range of Gauss's radius sigma and d is respectively: { 2
0, 2
1..., 2
8, { 2
-5, 2
-4..., 2
1and { 2
-2, 2
-1.., 2
4.For the compound nucleus function proposing, make Nonlinear Mapping number of times M=2, now classification performance convergence.
In first group of experiment, primary study use SVM is base sorter, adopt respectively gaussian kernel (Gauss Kernel) and compound nucleus function (continuous Gaussian Nonlinear Mapping G (G) Composite Kernel), the distribution situation of classification boundaries.As shown in Figure 2, a) b) compound nucleus function G (G) classification boundaries of gaussian kernel function Gauss classification boundaries.During by the more known employing compound nucleus of this group function, the classification boundaries of SVM is the distribution situation of descriptor data set better, and nicety of grading is also relatively high.
The impact of primary study different IPs function on classification performance in second group of experiment.The parameter of this group experiment is set to s=10.Experimental result is as shown in table 1, can be summarized as: 1, for single kernel function, the nicety of grading that the nicety of grading that gaussian kernel function obtains obtains compared with linear kernel (Linear Kernel) and polynomial kernel (Polynomial Kernel) is high; 2, when adopting mixed nucleus function (Mixture Kernels, MKs), i.e. linear weighted function summation mixed nucleus, can effectively improve the nicety of grading of single kernel function; 3,, compared with other kernel form, while adopting compound nucleus function, the nicety of grading of SVM is higher, and G (G) compound nucleus best performance.
The impact of primary study different IPs function on the parameter optimization time in the 3rd group of experiment.This group experiment all adopts cross validation, i.e. K=10 ten times.Experimental result is as table 2, known in the time of Nonlinear Mapping number of times M=2, more single core, and the time that compound nucleus function parameter is optimized is longer, meanwhile, the time that its optimization time will consume much smaller than mixed nucleus function.
The impact of the different training sample numbers of primary study on different IPs function performance in the 4th group of experiment.This group experiment makes s, and in set, { in 5,10,50,100}, change, experimental result is as shown in table 3, is summarized as: 1,, along with s increases, the classification capacity of different IPs function has obviously improved, but the difference between different IPs function diminishes; 2,, under Small Sample Size, more can embody the advantage of kernel function; 3, each kernel function, it is not constant effective, its validity is relevant with the actual distribution situation of data set.If the performance that kernel function can obtain under certain conditions, just illustrate that so this kernel function is effective.Based on this, the compound nucleus function in the present invention is effective kernel function.
Claims (3)
1. a kind of hyperspectral classification method based on compound nucleus function, is characterized in that:
Step 1: input one panel height spectrum picture, classification number is N;
Step 2: taking support vector machine as base sorter, from the each classification of described high spectrum image, choose randomly s sample composition training set simultaneously, residue sample composition test set, determine the variation range of each parameter, then the optimal performance parameter of determining support vector machine in conjunction with K cross validation, comprises penalty factor and nuclear parameter;
Step 3: utilize compound nucleus construction strategy, structure compound nucleus function, trains support vector machine;
Step 4: utilize the parameter of the support vector machine decision function that training process obtains, circulation N time, and then obtain the decision function value that test set belongs to every classification, composition matrix
wherein n
testrepresent the number of test sample book;
Step 5: determine multi-categorizer strategy, find matrix
the maximal value of every row, the prediction label of the corresponding each test sample book of its row sequence number,
2. a kind of hyperspectral classification method based on compound nucleus function according to claim 1, is characterized in that: described step 2 specifically comprises:
Step 2.1: every class is chosen s sample composition randomly from N classification of high spectrum image label training sample set D={ (x
1, y
1) ..., (x
n, y
n), wherein x
irepresent i spectral signature that has exemplar, y
irepresent sample x
ilabel, n=16s represents training sample number, residue sample composition test set
n
testrepresent test sample book number;
Step 2.2: adopting polynomial kernel function and gaussian kernel function is that basic kernel function is carried out the synthetic of compound nucleus function, and polynomial kernel is K
p(x, z)=φ
p(x) φ
p(z)=[(xz)+1]
d, its parameter is d, φ
p(x) the polynomial expression mapping function of expression sample x; Gaussian kernel is K
g(x, z)=φ
g(x) φ
g(z)=exp (|| x-z||
2/ σ
2), its Gauss's radius is σ, φ
g(x) the Gaussian mapping kernel function of expression sample x; The penalty factor of support vector machine, the variation range of Gauss's radius sigma and d is respectively: { 2
0, 2
1..., 2
8, { 2
-5, 2
-4..., 2
1and { 2
-2, 2
-1.., 2
4;
Step 2.3: in each parameter combinations situation, training sample set D is divided into K subset, retain successively a subset for test, remaining K-1 subset is used for training svm classifier device model, cross validation repeats K time, and each subset is verified once, and the mean value of calculating K subseries precision, in the time that mean accuracy reaches maximal value, this parameter combinations is optimum.
3. a kind of hyperspectral classification method based on compound nucleus function according to claim 1 and 2, is characterized in that: described step 3 specifically comprises:
Step 3.1: obtain compound nucleus function by Nonlinear Mapping repeatedly, be calculated as follows:
K
1(x,z)=φ
1(x)·φ
1(z)
K
2(x,z)=φ
2[φ
1(x)]·φ
2[φ
1(z)]
...
K
M(x,z)=φ
M[φ
M-1(x)]·φ
M[φ
M-1(z)]
Wherein K
m(x, z) represents the kernel function of sample x and z, φ
mrepresent Nonlinear Mapping function the M time; φ
mgaussian mapping, polynomial expression mapping or other Nonlinear Mapping, and in the time of M=2, the classification performance based on this compound Kernel function classifier reaches convergence state; In the time of M=2, K
g (G), K
g (P), K
p (G)and K
p (P)be respectively continuous Gaussian mapping compound nucleus function, polynomial expression-Gaussian mapping compound nucleus function, Gauss-polynomial expression mapping compound nucleus function and continuously polynomial expression mapping compound nucleus function; Expression formula is as follows:
K
G(G)(x,z)=φ
G[φ
G(x)]·φ
G[φ
G(z)]=exp[-||φ
G(x)-φ
G(z)||
2/σ
2 2] (1)=exp[-[φ
G(x)·φ
G(x)+φ
G(z)·φ
G(z)-2φ
G(x)·φ
G(z)]/σ
2 2]=exp[-[K
G(x,x)+K
G(z,z)-2K
G(x,z)]/σ
2 2]
K
G(P)(x,z)=φ
G[φ
P(x)]·φ
G[φ
P(z)]=exp[-||φ
P(x)-φ
P(z)||
2/σ
2 2] (2)=exp[-[φ
P(x)·φ
P(x)+φ
P(z)·φ
P(z)-2φ
P(x)·φ
P(z)]/σ
2 2]=exp[-[K
P(x,x)+K
P(z,z)-2K
P(x,z)]/σ
2 2]
Wherein x and z represent two pixels, σ
2represent Gauss's radius of Gaussian mapping for the second time, d
2represent the polynomial parameters of polynomial expression mapping for the second time; In formula (1) (3), Gauss's radius of Gaussian mapping is σ for the first time
1, in formula (2) (4), the polynomial parameters of polynomial expression mapping is for the first time d
1;
Step 3.2: support vector machine is trained to the weight vectors of supported vector machine decision function and threshold alpha
*and b
*.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410386737.1A CN104200217B (en) | 2014-08-07 | 2014-08-07 | Hyperspectrum classification method based on composite kernel function |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410386737.1A CN104200217B (en) | 2014-08-07 | 2014-08-07 | Hyperspectrum classification method based on composite kernel function |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104200217A true CN104200217A (en) | 2014-12-10 |
CN104200217B CN104200217B (en) | 2017-05-24 |
Family
ID=52085507
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410386737.1A Expired - Fee Related CN104200217B (en) | 2014-08-07 | 2014-08-07 | Hyperspectrum classification method based on composite kernel function |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104200217B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104992178A (en) * | 2015-06-16 | 2015-10-21 | 中国石油化工股份有限公司 | Tight sandstone fluid type identification method based on support vector machine simulation cross plot |
CN107220677A (en) * | 2017-06-28 | 2017-09-29 | 深圳源广安智能科技有限公司 | The accurate hyperspectral image data categorizing system of one kind classification |
CN107563324A (en) * | 2017-08-30 | 2018-01-09 | 广东工业大学 | A kind of hyperspectral image classification method and device of the learning machine that transfinited based on core basis |
CN107633268A (en) * | 2017-09-22 | 2018-01-26 | 重庆交通大学 | A kind of high spectrum image combination core minimal noise separation transform characteristics extracting method |
CN107976432A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | A kind of heat resisting steel aging level measurement method based on support vector machines |
CN109596543A (en) * | 2018-11-25 | 2019-04-09 | 西安建筑科技大学 | The spectral reflectance recovery method of particle group optimizing multicore support vector regression |
CN109697465A (en) * | 2018-12-18 | 2019-04-30 | 辽宁工程技术大学 | Hyperspectral Image Classification method based on the Fisher discriminant analysis of Wavelet Kernel part |
CN109901204A (en) * | 2019-03-27 | 2019-06-18 | 北京航空航天大学 | A kind of GBAS integrity performance estimating method based on pseudorange error distributed model |
CN109961093A (en) * | 2019-03-07 | 2019-07-02 | 北京工业大学 | A kind of image classification method based on many intelligence integrated studies |
CN110163101A (en) * | 2019-04-17 | 2019-08-23 | 湖南省中医药研究院 | The difference of Chinese medicine seed and grade quick discrimination method |
CN110717528A (en) * | 2019-09-25 | 2020-01-21 | 中国石油大学(华东) | Support vector machine-based sedimentary microfacies identification method using conventional logging information |
CN113205143A (en) * | 2021-05-10 | 2021-08-03 | 郑州轻工业大学 | Multi-scale superpixel hyperspectral remote sensing image classification method based on space-spectrum coupling characteristics |
CN113779236A (en) * | 2021-08-11 | 2021-12-10 | 齐维维 | Method and device for problem classification based on artificial intelligence |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100142805A1 (en) * | 2008-12-05 | 2010-06-10 | Tandent Vision Science, Inc. | Constraint generation for use in image segregation |
CN101807254A (en) * | 2009-02-13 | 2010-08-18 | 烟台海岸带可持续发展研究所 | Implementation method for data characteristic-oriented synthetic kernel support vector machine |
CN101976361A (en) * | 2010-11-23 | 2011-02-16 | 中国矿业大学 | Multi-kernel support vector machine classification method for remote sensing images |
-
2014
- 2014-08-07 CN CN201410386737.1A patent/CN104200217B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100142805A1 (en) * | 2008-12-05 | 2010-06-10 | Tandent Vision Science, Inc. | Constraint generation for use in image segregation |
CN101807254A (en) * | 2009-02-13 | 2010-08-18 | 烟台海岸带可持续发展研究所 | Implementation method for data characteristic-oriented synthetic kernel support vector machine |
CN101976361A (en) * | 2010-11-23 | 2011-02-16 | 中国矿业大学 | Multi-kernel support vector machine classification method for remote sensing images |
Non-Patent Citations (4)
Title |
---|
张朝福,等: ""基于复合核函数KPCA的红外人脸识别"", 《计算机仿真》 * |
曲昆鹏: ""基于支持向量机的杂草识别研究"", 《中国优秀硕士学位论文全文数据库(信息科技辑)》 * |
谭熊,等: ""高光谱影像的多核SVM分类"", 《仪器仪表学》 * |
高恒振: ""高光谱遥感图像分类技术研究"", 《中国博士学位论文全文数据库(信息科技辑)》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104992178A (en) * | 2015-06-16 | 2015-10-21 | 中国石油化工股份有限公司 | Tight sandstone fluid type identification method based on support vector machine simulation cross plot |
CN107220677A (en) * | 2017-06-28 | 2017-09-29 | 深圳源广安智能科技有限公司 | The accurate hyperspectral image data categorizing system of one kind classification |
CN107563324A (en) * | 2017-08-30 | 2018-01-09 | 广东工业大学 | A kind of hyperspectral image classification method and device of the learning machine that transfinited based on core basis |
CN107633268A (en) * | 2017-09-22 | 2018-01-26 | 重庆交通大学 | A kind of high spectrum image combination core minimal noise separation transform characteristics extracting method |
CN107976432B (en) * | 2017-10-16 | 2020-12-22 | 华南理工大学 | Heat-resistant steel aging grade measuring method based on support vector machine |
CN107976432A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | A kind of heat resisting steel aging level measurement method based on support vector machines |
CN109596543A (en) * | 2018-11-25 | 2019-04-09 | 西安建筑科技大学 | The spectral reflectance recovery method of particle group optimizing multicore support vector regression |
CN109596543B (en) * | 2018-11-25 | 2021-09-10 | 西安建筑科技大学 | Spectral reflectivity reconstruction method for particle swarm optimization multi-core support vector regression |
CN109697465A (en) * | 2018-12-18 | 2019-04-30 | 辽宁工程技术大学 | Hyperspectral Image Classification method based on the Fisher discriminant analysis of Wavelet Kernel part |
CN109697465B (en) * | 2018-12-18 | 2022-11-25 | 辽宁工程技术大学 | Hyperspectral image classification method based on wavelet kernel local Fisher discriminant analysis |
CN109961093A (en) * | 2019-03-07 | 2019-07-02 | 北京工业大学 | A kind of image classification method based on many intelligence integrated studies |
CN109961093B (en) * | 2019-03-07 | 2021-10-15 | 北京工业大学 | Image classification method based on crowd-sourcing integrated learning |
CN109901204B (en) * | 2019-03-27 | 2020-12-04 | 北京航空航天大学 | GBAS integrity performance evaluation method based on pseudo-range error distribution model |
CN109901204A (en) * | 2019-03-27 | 2019-06-18 | 北京航空航天大学 | A kind of GBAS integrity performance estimating method based on pseudorange error distributed model |
CN110163101A (en) * | 2019-04-17 | 2019-08-23 | 湖南省中医药研究院 | The difference of Chinese medicine seed and grade quick discrimination method |
CN110163101B (en) * | 2019-04-17 | 2022-09-23 | 湖南省中医药研究院 | Method for rapidly distinguishing seeds of traditional Chinese medicinal materials and rapidly judging grades of seeds |
CN110717528A (en) * | 2019-09-25 | 2020-01-21 | 中国石油大学(华东) | Support vector machine-based sedimentary microfacies identification method using conventional logging information |
CN113205143A (en) * | 2021-05-10 | 2021-08-03 | 郑州轻工业大学 | Multi-scale superpixel hyperspectral remote sensing image classification method based on space-spectrum coupling characteristics |
CN113779236A (en) * | 2021-08-11 | 2021-12-10 | 齐维维 | Method and device for problem classification based on artificial intelligence |
Also Published As
Publication number | Publication date |
---|---|
CN104200217B (en) | 2017-05-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104200217A (en) | Hyperspectrum classification method based on composite kernel function | |
Kim et al. | Star-galaxy classification using deep convolutional neural networks | |
Luus et al. | Multiview deep learning for land-use classification | |
CN113486851B (en) | Hyperspectral image classification method based on double-branch spectrum multi-scale attention network | |
CN104732240B (en) | A kind of Hyperspectral imaging band selection method using neural network sensitivity analysis | |
CN107563428A (en) | Classification of Polarimetric SAR Image method based on generation confrontation network | |
CN108764173A (en) | The hyperspectral image classification method of confrontation network is generated based on multiclass | |
CN105069478B (en) | High-spectrum remote-sensing terrain classification method based on super-pixel tensor sparse coding | |
CN107292343A (en) | A kind of Classification of hyperspectral remote sensing image method based on six layers of convolutional neural networks and spectral space information consolidation | |
CN109766858A (en) | Three-dimensional convolution neural network hyperspectral image classification method combined with bilateral filtering | |
CN106203523A (en) | The classification hyperspectral imagery of the semi-supervised algorithm fusion of decision tree is promoted based on gradient | |
CN107292336A (en) | A kind of Classification of Polarimetric SAR Image method based on DCGAN | |
Dong et al. | Band attention convolutional networks for hyperspectral image classification | |
CN106874688A (en) | Intelligent lead compound based on convolutional neural networks finds method | |
CN103632168A (en) | Classifier integration method for machine learning | |
CN104021396A (en) | Hyperspectral remote sensing data classification method based on ensemble learning | |
CN102938072A (en) | Dimension reducing and sorting method of hyperspectral imagery based on blocking low rank tensor analysis | |
CN103336968A (en) | Hyperspectral data dimension reduction method based on tensor distance patch calibration | |
CN104392251A (en) | Hyperspectral image classification method based on semi-supervised dictionary learning | |
CN107145836A (en) | Hyperspectral image classification method based on stack boundary discrimination self-encoding encoder | |
CN103049760B (en) | Based on the rarefaction representation target identification method of image block and position weighting | |
CN103440493A (en) | Hyperspectral image blur classification method and device based on related vector machine | |
CN111353531A (en) | Hyperspectral image classification method based on singular value decomposition and spatial spectral domain attention mechanism | |
CN105913092A (en) | Figure regular hyperspectral image band selection method based on subspace learning | |
CN104268579A (en) | Hyperspectral remote sensing image classifying method based on hierarchy ensemble learning |
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: 20170524 |