CN105139026B - The Feature Extraction Method of spectral domain spatial domain joint related constraint in hyperspectral classification - Google Patents
The Feature Extraction Method of spectral domain spatial domain joint related constraint in hyperspectral classification Download PDFInfo
- Publication number
- CN105139026B CN105139026B CN201510465944.0A CN201510465944A CN105139026B CN 105139026 B CN105139026 B CN 105139026B CN 201510465944 A CN201510465944 A CN 201510465944A CN 105139026 B CN105139026 B CN 105139026B
- Authority
- CN
- China
- Prior art keywords
- sample
- principal component
- constraint
- spectral
- spatial domain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Feature Extraction Methods of spectral domain spatial domain joint related constraint in hyperspectral classification, it is characterized in that, including carrying out dimensionality reduction denoising to original hyperspectral image data using principal component analysis method, and the principal component of original hyperspectral image data is extracted, sample and its label after obtaining principal component analysis;According to the label of the sample after principal component analysis, sample is split into class sample outside sample and class;Building spectral domain correlation constraint is carried out to sample outside sample in class and class respectively;Sample after selecting a principal component analysis is center sample, obtains the neighborhood sample set for corresponding to its set spatial position, seeks the related coefficient of sample in central sample and its neighborhood sample set, constructs spatial domain correlation constraint;The method bound directly using spectral domain correlation constraint and spatial domain correlation constraint constructs spectral domain and spatial domain blended relevance feature vector, obtains the feature vector of the spectral domain spatial domain joint related constraint of high spectrum image.
Description
Technical field
The invention belongs to spectral domain spatial domains in Hyperspectral imagery processing field more particularly to a kind of hyperspectral classification to combine
The Feature Extraction Method of related constraint.
Background technique
Traditional spatial-domain information and spectrum domain information are organically merged and are integrated by high spectrum image, are obtaining scene
While spatial image, the continuous spectrum of all objects in scene is obtained.And in high spectrum image, different objects is to each wave
The spectrum of section has different absorptivity and reflectivity, the spectral signature curve of consecutive variations is shown as, so as to realize foundation
The target that Object Spectra feature is classified and identified.
Classification hyperspectral imagery problem is a great problem of high spectrum image analysis and application technology faced.EO-1 hyperion
The key step of image classification includes two parts: (1) feature extraction and feature selecting;(2) tagsort.Traditional EO-1 hyperion
Characteristic matching classification method needs a large amount of priori knowledge, too high to spectrum characteristic data library dependence, and statistical classification method
Arithmetic speed is slow, and precision is affected by training sample.
Existing feature extraction and classification method are often limited to the defect of high spectrum image itself, show as the steady of algorithm
Qualitative and robustness is insufficient.Meanwhile traditional Feature Extraction Method is based only on spectrum characteristic of field, and ignore space abundant
The information in domain.Studies have shown that the precision of hyperspectral classification can be obviously improved in conjunction with pixel space domain information.
Summary of the invention
In order to make up defect and deficiency of the existing technology, it is empty that the invention proposes spectral domains in a kind of hyperspectral classification
Between domain joint related constraint Feature Extraction Method, this method considers sample light spectral domain correlation and spatial domain correlation, and will
Spectrum domain information is combined with spatial-domain information, is extracted identification feature, is effectively improved the precision of entire hyperspectral classification system.
In order to reach the goals above, technical scheme is as follows:
The Feature Extraction Method of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification, comprising:
Step (1): dimensionality reduction denoising is carried out to original hyperspectral image data using principal component analysis method, and extracts original height
The principal component of spectral image data, sample and its label after obtaining principal component analysis;
Step (2): according to the label of the sample after principal component analysis, sample is split into class sample outside sample and class;
According to related coefficient calculation method, building spectral domain correlation constraint is carried out to sample outside sample in class and class respectively;
Step (3): the sample after selecting a principal component analysis is center sample, obtains corresponding to its set spatial position
Neighborhood sample set, seeks the related coefficient of sample in central sample and its neighborhood sample set, and constructs spatial domain correlation
Constraint;
Step (4): the method bound directly using spectral domain correlation constraint and spatial domain correlation constraint constructs spectrum
Domain and spatial domain blended relevance feature vector, the spy of the final spectral domain spatial domain joint related constraint for obtaining high spectrum image
Levy vector.
The process packet of the principal component of original hyperspectral image data is extracted in the step (1) using principal component analysis method
It includes:
Step (1.1): to given high-spectral data sample set, all samples in high-spectral data sample set are sought
Mean value, and to all samples carry out centralization;
Step (1.2): the sample characteristics matrix after building centralization, and calculate the association of the sample characteristics matrix after centralization
Variance matrix;
Step (1.3): carrying out feature decomposition to the covariance matrix that step (1.2) obtains, and obtains one group of descending arrangement
Characteristic value and its corresponding feature vector select the principal component of high-spectral data sample;
Step (1.4): after the completion of principal component selection, according to the corresponding feature vector of characteristic value, principal component mapping square is obtained
Battle array: carrying out principal component mapping to each sample, the sample after obtaining principal component analysis, and then obtains the sample set after principal component analysis
It closes.
The expression formula of the mean value of all samples in high-spectral data sample set is sought in the step (1.1) are as follows:
Wherein,It is the mean value of all samples;XiIt is i-th of sample, i=1,2 ..., N, N are sample numbers.
In the step (1.1) all samples are carried out with the expression formula of centralization are as follows:
Wherein, XiIt is i-th of sample;It is the mean value of all samples;Xi' it is XiSampling feature vectors after centralization, i
=1,2 ..., N, N be sample number.
The covariance matrix of the sample characteristics matrix after centralization is calculated in the step (1.2) are as follows:
C=XT*X (3)
Wherein, X is the sample characteristics matrix after centralization;XTIt is the transposed matrix of X;C∈Rm×mIt is the sample after centralization
The covariance matrix of eigenmatrix;M is characteristic.
The selection method of principal component in the step (1.3) are as follows:
Wherein, λi’It is characterized value, is arranged in descending, i '=1,2 ..., m*;ωi∈RmIt is characterized the corresponding feature of value
Vector j=1,2 ..., m;m*It is the number of the principal component retained, m is the characteristic of sample, and meets m*≤ m, 0.995 meaning
It is the information that retained principal component includes m feature 99.5%.
In the step (1.4) after the completion of principal component selection, principal component mapping matrix U is obtained:
Principal component mapping, sample after obtaining principal component analysis are carried out to each sample, and then after obtaining principal component analysis
Sample set.
Related coefficient calculation method in the step (2) are as follows:
Wherein, x and y indicate vector;R (x, y) ∈ [- 1,1] indicates the related coefficient between vector x and y, two vectors
More related, related coefficient is bigger;| | | | indicate vector field homoemorphism.
The method that sample is divided in the step (2) are as follows:
All samples are split into sample set in class according to sample labelWith sample set outside class
Two parts, divisional mode are as follows:
Wherein,Indicate class in sample set, by with sampleThe identical sample composition of classification,Table
Show the outer sample set of class, by with sampleThe different sample composition of classification, yiIndicate sampleCorresponding label, ykIt indicates
SampleCorresponding label, i, k=1,2 ..., N, N are sample numbers.
Building spectral domain and spatial domain blended relevance feature vector in the step (4) are as follows:
Wherein,It is the spectral domain spatial domain blended relevance feature vector of building, l is spectral domain correlation
The number of bound term;Indicate spectral domain correlation constraint;Representation space domain correlation constraint;α table
Show that central sample corresponds to the sample size of its neighborhood sample set.
The beneficial effects of the present invention are:
(1) the present invention is based on traditional vector correlation coefficient calculation method, structure is simple, and computation complexity is low;
(2) present invention establishes spectral domain correlation constraint, and the selection of bound term is by maximizing in class outside correlation and class
Correlation maximizes the outer distribution of correlation coefficient of class in class, has good adaptivity and robustness;
(3) this method establishes spatial domain correlation constraint, and combines with spectral domain phase relation constraint, and building spectral domain is empty
Between domain blended relevance feature vector, can effectively improve nicety of grading.
Detailed description of the invention
The Feature Extraction Method flow chart of Fig. 1 spectral domain spatial domain correlation constraint of the present invention;
8. neighbourhood model of spatial domain Fig. 2 of the invention.
Specific embodiment
The invention will be further described with example with reference to the accompanying drawing:
The present invention includes ten kinds of farmings in selected image under the premise of obtaining the high spectrum image in a certain area
Object respectively corresponds ten kinds of classifications;Using the Feature Extraction Method of spectral domain spatial domain correlation constraint proposed by the present invention to height
Spectrum picture is handled, for distinguishing ten kinds of different classes of crops in high spectrum image.
As shown in Figure 1, spectral domain spatial domain combines the Feature Extraction Method of related constraint in hyperspectral classification, comprising:
Step (1): dimensionality reduction denoising is carried out to original hyperspectral image data using principal component analysis method, and extracts original height
The principal component of spectral image data, sample and its label after obtaining principal component analysis;
Step (2): according to the label of the sample after principal component analysis, sample is split into class sample outside sample and class;
According to related coefficient calculation method, building spectral domain correlation constraint is carried out to sample outside sample in class and class respectively;
Step (3): the sample after selecting a principal component analysis is center sample, obtains corresponding to its set spatial position
Neighborhood sample set, seeks the related coefficient of sample in central sample and its neighborhood sample set, and constructs spatial domain correlation
Constraint;
Step (4): the method bound directly using spectral domain correlation constraint and spatial domain correlation constraint constructs spectrum
Domain and spatial domain blended relevance feature vector, the spy of the final spectral domain spatial domain joint related constraint for obtaining high spectrum image
Levy vector.
Further, the mistake of the principal component of original hyperspectral image data is extracted in step (1) using principal component analysis method
Journey includes:
Step (1.1): to given high-spectral data sample set, all samples in high-spectral data sample set are sought
Mean value, and to all samples carry out centralization;
Seek the expression formula of the mean value of all samples in high-spectral data sample set are as follows:
Wherein, giving high-spectral data sample set is X={ Xi}∈Rm;It is the mean value of all samples;XiIt is i-th
Sample, i=1,2 ..., N, N are sample numbers;
All samples are carried out with the expression formula of centralization are as follows:
Wherein, XiIt is i-th of sample;It is the mean value of all samples;Xi' it is XiSampling feature vectors after centralization, i
=1,2 ..., N, N be sample number.
Step (1.2): the sample characteristics matrix after building centralization, and calculate the association of the sample characteristics matrix after centralization
Variance matrix;
Sample characteristics matrix X=[X after constructing centralization1',X'2,...,X'N]T∈RN×m, m is characteristic, and N is sample
Number, subscript T representing matrix transposition.The covariance matrix of sample characteristics matrix after calculating centralization are as follows:
C=XT*X (3)
Wherein, X is the sample characteristics matrix after centralization;XTIt is the transposed matrix of X;C∈Rm×mIt is the sample after centralization
The covariance matrix of eigenmatrix;M is characteristic.
Step (1.3): carrying out feature decomposition to the covariance matrix that step (1.2) obtains, and obtains one group of descending arrangement
Characteristic value and its corresponding feature vector select the principal component of high-spectral data sample;
The selection method of principal component are as follows:
Wherein, λi’It is characterized value, is arranged in descending, i '=1,2 ..., m*;ωi∈RmIt is characterized the corresponding feature of value
Vector j=1,2 ..., m;m*It is the number of the principal component retained, m is the characteristic of sample, and meets m*≤ m, 0.995 meaning
It is the information that retained principal component includes m feature 99.5%.
Step (1.4): after the completion of principal component selection, according to the corresponding feature vector of characteristic value, principal component mapping square is obtained
Battle array: carrying out principal component mapping to each sample, the sample after obtaining principal component analysis, and then obtains the sample set after principal component analysis
It closes.
Further, after the completion of principal component selection, principal component mapping matrix U is obtained:
Principal component mapping is carried out to each sample, the sample after obtaining principal component analysis,
It is i-th of sample after principal component analysis, i=1,2 ..., N, N are sample number, m*It is principal component
Number.And then obtain the sample set after principal component analysis
In this example, the principal component number of extraction is 10, i.e. m*=10.
Further, the related coefficient calculation method in step (2) are as follows:
Wherein, x and y indicate vector;R (x, y) ∈ [- 1,1] indicates the related coefficient between vector x and y, two vectors
More related, related coefficient is bigger;| | | | indicate vector field homoemorphism.
Further, method sample divided in step (2) are as follows:
All samples are split into sample set in class according to sample labelWith sample set outside class
Two parts, divisional mode are as follows:
Wherein,Indicate class in sample set, by with sampleThe identical sample composition of classification,Table
Show the outer sample set of class, by with sampleThe different sample composition of classification, yiIndicate sampleCorresponding label, ykIt indicates
SampleCorresponding label, i, k=1,2 ..., N, N are sample numbers.
The outer related coefficient of class calculates in class, according to any two vector correlation coefficient calculation method in step (2.1), meter
Calculate sampleWith sample set in classThe related coefficient of middle sample selects maximum l1It is a
Related coefficient is denoted as Rin(i), i=1,2 ..., l1;Calculate sampleWith sample set outside classThe related coefficient of middle sample selects maximum l2A related coefficient, is denoted as Rout(i), i=1,2 ..., l2;
It further, is sampleConstruct spectral domain correlation constraint
Wherein, l=l1+l2Indicate the number of spectral domain correlation constraint item.
Further, the method for step (3) building spatial domain correlation constraint includes:
8. field correlation constraint Neigh () of definition space part of the present invention, with sampleIt is corresponding centered on sample
In high spectrum imageSpatial position 8. neighborhoods sample set Xspatial={ Xk, k=1,2 ..., 8, such as Fig. 2 institute
Show, according to any two vector correlation coefficient calculation method, calculates sampleWith 8. neighborhood sample set XspatialInterior sample
Related coefficient is denoted asFor sampleConstruct spatial domain correlation constraint
Wherein, Xk∈Xspatial, indicate k-th of sample in 8. neighborhood sample sets, k=1,2 ..., 8.
The method of step (4) building spectral domain spatial domain blended relevance feature vector are as follows:
To sampleIn conjunction with spectral domain correlation constraint in step (2)And step
Suddenly spatial domain correlation constraint in (3)The present invention constructs spectral domain spatial domain using the method bound directly
Blended relevance feature vector:
Wherein,It is the spectral domain spatial domain blended relevance feature vector of building, l is spectral domain correlation
The number of bound term;Indicate spectral domain correlation constraint;Representation space domain correlation constraint;α table
Show that central sample corresponds to the sample size of its neighborhood sample set.Table 1 is the nicety of grading comparison of five kinds of methods of the present embodiment
As a result.
The nicety of grading comparing result of 1 five kinds of methods of table
Wherein, Ori-RF is primitive character+random forest grader;Ori-SVM is primitive character+support vector cassification
Device;CCSVM is Feature Extraction Method+support vector machine classifier of the invention;CCRF be Feature Extraction Method of the invention+with
Machine forest classified device;CSRF is the specified tree random forest classification method of Feature Extraction Method+class of the invention.
Wherein, Kappa coefficient is an important factor of interpretive classification method, and the range of the value is 0~1, is worth bigger, table
Show that classifying quality is better.It can be seen that by the global classification precision and Kappa coefficient of table 1, hyperspectral classification spectral domain of the invention is empty
Between the Feature Extraction Method of domain correlation constraint make full use of spectral information and spatial information, solve the problems, such as classification hyperspectral imagery
Effectively improve the nicety of grading classifying quality of high spectrum image.
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (9)
1. the Feature Extraction Method of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification characterized by comprising
Step (1): dimensionality reduction denoising is carried out to original hyperspectral image data using principal component analysis method, and extracts original EO-1 hyperion
The principal component of image data, sample and its label after obtaining principal component analysis;
Step (2): according to the label of the sample after principal component analysis, sample is split into class sample outside sample and class;Foundation
Related coefficient calculation method carries out building spectral domain correlation constraint to sample outside sample in class and class respectively;
Step (3): the sample after selecting a principal component analysis is center sample, obtains the neighborhood for corresponding to its set spatial position
Sample set, seeks the related coefficient of sample in central sample and its neighborhood sample set, and constructs spatial domain correlation constraint;
Step (4): the method bound directly using spectral domain correlation constraint and spatial domain correlation constraint, building spectral domain and
Spatial domain blended relevance feature vector, the feature of the final spectral domain spatial domain joint related constraint for obtaining high spectrum image to
Amount;
The method bound directly in the step (4) using spectral domain correlation constraint and spatial domain correlation constraint constructs light
Spectral domain and spatial domain blended relevance feature vector are as follows:
Wherein,It is the spectral domain spatial domain blended relevance feature vector of building, l is spectral domain correlation constraint
The number of item;Indicate spectral domain correlation constraint;Representation space domain correlation constraint;In α expression
Heart sample corresponds to the sample size of its neighborhood sample set, i=1, and 2 ..., N, N are sample numbers,It is after principal component analysis
I sample.
2. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as described in claim 1
Method, which is characterized in that extract the mistake of the principal component of original hyperspectral image data in the step (1) using principal component analysis method
Journey, comprising:
Step (1.1): to given high-spectral data sample set, the equal of all samples in high-spectral data sample set is sought
Value, and centralization is carried out to all samples;
Step (1.2): the sample characteristics matrix after building centralization, and calculate the covariance of the sample characteristics matrix after centralization
Matrix;
Step (1.3): feature decomposition is carried out to the covariance matrix that step (1.2) obtains, obtains the feature of one group of descending arrangement
Value and its corresponding feature vector select the principal component of high-spectral data sample;
Step (1.4): after the completion of principal component selection, according to the corresponding feature vector of characteristic value, principal component mapping matrix is obtained: right
Each sample carries out principal component mapping, the sample after obtaining principal component analysis, and then obtains the sample set after principal component analysis.
3. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as claimed in claim 2
Method, which is characterized in that the expression formula of the mean value of all samples in high-spectral data sample set is sought in the step (1.1)
Are as follows:
Wherein,It is the mean value of all samples;XiIt is i-th of sample.
4. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as claimed in claim 2
Method, which is characterized in that in the step (1.1) all samples are carried out with the expression formula of centralization are as follows:
Wherein, XiIt is i-th of sample;It is the mean value of all samples;X′iIt is XiSampling feature vectors after centralization.
5. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as claimed in claim 2
Method, which is characterized in that the covariance matrix of the sample characteristics matrix after centralization is calculated in the step (1.2) are as follows:
C=XT*X (3)
Wherein, X is the sample characteristics matrix after centralization;XTIt is the transposed matrix of X;C∈Rm×mIt is the sample characteristics after centralization
The covariance matrix of matrix;M is characteristic.
6. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as claimed in claim 2
Method, which is characterized in that the selection method of principal component in the step (1.3) are as follows:
Wherein, λi'It is characterized value, is arranged in descending, i'=1,2 ..., m*;J=1,2 ..., m;m*It is the principal component retained
Number, m is the characteristic of sample, and meets m*≤ m, 0.995 meaning are that retained principal component includes m feature
99.5% information.
7. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as claimed in claim 2
Method, which is characterized in that in the step (1.4) after the completion of principal component selection, obtain principal component mapping matrix U:
Principal component mapping is carried out to each sample, the sample after obtaining principal component analysis, and then obtain the sample after principal component analysis
Set, m*It is the number of the principal component retained, m is the characteristic of sample.
8. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as described in claim 1
Method, which is characterized in that the related coefficient calculation method in the step (2) are as follows:
Wherein, x and y indicate vector;R (x, y) ∈ [- 1,1] indicates that the related coefficient between vector x and y, two vectors get over phase
It closes, related coefficient is bigger;| | | | indicate vector field homoemorphism.
9. the feature extraction side of spectral domain spatial domain joint related constraint in a kind of hyperspectral classification as described in claim 1
Method, which is characterized in that the method that sample is divided in the step (2) are as follows:
All samples are split into sample set in class according to sample labelWith sample set outside classTwo
Point, divisional mode is as follows:
Wherein,Indicate class in sample set, by with sampleThe identical sample composition of classification,Indicate class
Outer sample set, by with sampleThe different sample composition of classification, yiIndicate sampleCorresponding label, ykIndicate sampleCorresponding label, k=1,2 ..., N.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510465944.0A CN105139026B (en) | 2015-08-03 | 2015-08-03 | The Feature Extraction Method of spectral domain spatial domain joint related constraint in hyperspectral classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510465944.0A CN105139026B (en) | 2015-08-03 | 2015-08-03 | The Feature Extraction Method of spectral domain spatial domain joint related constraint in hyperspectral classification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105139026A CN105139026A (en) | 2015-12-09 |
CN105139026B true CN105139026B (en) | 2019-02-12 |
Family
ID=54724369
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510465944.0A Active CN105139026B (en) | 2015-08-03 | 2015-08-03 | The Feature Extraction Method of spectral domain spatial domain joint related constraint in hyperspectral classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105139026B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109271874B (en) * | 2018-08-23 | 2022-02-11 | 广东工业大学 | Hyperspectral image feature extraction method fusing spatial and spectral information |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463247A (en) * | 2014-12-09 | 2015-03-25 | 山东大学 | Extracting method of optical spectrum vector cross-correlation features in hyper-spectral image classification |
-
2015
- 2015-08-03 CN CN201510465944.0A patent/CN105139026B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463247A (en) * | 2014-12-09 | 2015-03-25 | 山东大学 | Extracting method of optical spectrum vector cross-correlation features in hyper-spectral image classification |
Non-Patent Citations (2)
Title |
---|
A spectral-spatial hyperspectral data classification approach using random forest with label constraints;Yuemei Ren等;《IEEE》;20141231;第344-347页 |
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》.2012,第27-56页. |
Also Published As
Publication number | Publication date |
---|---|
CN105139026A (en) | 2015-12-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yuan et al. | Remote sensing image segmentation by combining spectral and texture features | |
CN107145836B (en) | Hyperspectral image classification method based on stacked boundary identification self-encoder | |
CN109993072B (en) | Low-resolution pedestrian re-identification system and method based on super-resolution image generation | |
CN108460391B (en) | Hyperspectral image unsupervised feature extraction method based on generation countermeasure network | |
CN107590515B (en) | Hyperspectral image classification method of self-encoder based on entropy rate superpixel segmentation | |
CN109034224B (en) | Hyperspectral classification method based on double branch network | |
CN111753828B (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
CN111476170A (en) | Remote sensing image semantic segmentation method combining deep learning and random forest | |
CN107292339A (en) | The unmanned plane low altitude remote sensing image high score Geomorphological Classification method of feature based fusion | |
CN110084159A (en) | Hyperspectral image classification method based on the multistage empty spectrum information CNN of joint | |
CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
CN106203483B (en) | A kind of zero sample image classification method based on semantic related multi-modal mapping method | |
CN103208011B (en) | Based on average drifting and the hyperspectral image space-spectral domain classification method organizing sparse coding | |
CN109389080A (en) | Hyperspectral image classification method based on semi-supervised WGAN-GP | |
CN108154094B (en) | Hyperspectral image unsupervised waveband selection method based on subinterval division | |
CN108182449A (en) | A kind of hyperspectral image classification method | |
CN106339674A (en) | Hyperspectral image classification method based on edge preservation and graph cut model | |
CN108427913A (en) | The Hyperspectral Image Classification method of combined spectral, space and hierarchy information | |
CN105184314B (en) | Wrapper formula EO-1 hyperion band selection methods based on pixel cluster | |
CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
CN108734199A (en) | High spectrum image robust classification method based on segmentation depth characteristic and low-rank representation | |
CN109034213B (en) | Hyperspectral image classification method and system based on correlation entropy principle | |
Fang et al. | Multiattention joint convolution feature representation with lightweight transformer for hyperspectral image classification | |
CN111222545A (en) | Image classification method based on linear programming incremental learning | |
CN114266961A (en) | Method for integrating, learning and classifying marsh vegetation stacks by integrating hyperspectral and multiband fully-polarized SAR images |
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 |