CN105809172B - Effective semi-supervised feature selection approach towards high-resolution remote sensing image - Google Patents
Effective semi-supervised feature selection approach towards high-resolution remote sensing image Download PDFInfo
- Publication number
- CN105809172B CN105809172B CN201610127595.6A CN201610127595A CN105809172B CN 105809172 B CN105809172 B CN 105809172B CN 201610127595 A CN201610127595 A CN 201610127595A CN 105809172 B CN105809172 B CN 105809172B
- Authority
- CN
- China
- Prior art keywords
- feature
- remote sensing
- sample
- sensing image
- resolution remote
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
Effective semi-supervised feature selection approach towards high-resolution remote sensing image belongs to the semi-supervised Feature Selection field of remote sensing images.The present invention is the semi-supervised feature selection approach complex disposal process heavy workload in order to solve the problem of existing high-resolution remote sensing image.Its steps are as follows: acquisition high-resolution remote sensing image data pre-process high-resolution remote sensing image data；Divide sample again, extract feature, obtain sample data, until obtaining characteristic set；Carry out sample class label；Label and selection matrix are obtained to the feature vector of marker samples classification simultaneously, construct the objective function of the probability distribution matrix based on loss function and unmarked sample；Objective function based on foundation is optimized using category probability matrix of the iterative algorithm to the unmarked sample minimum filter depth of selection vector sum, completes the feature selecting of high-resolution remote sensing image.The present invention is used for the feature selecting of high-resolution remote sensing image.
Description
Technical field
The present invention relates to effective semi-supervised feature selection approach towards high-resolution remote sensing image, belong to remote sensing images
Semi-supervised Feature Selection field.
Background technique
Because the characteristics of capable of describing the details of land cover pattern information, high-resolution remote sensing image VHR is in real life
It is widely applied, is especially acted in terms of detection water quality and biomass huge.But its complicated spatial linear and spectrum
Inhomogeneities all proposes stern challenge to common technology pixel-based.Because individual spectral information is not enough to distinguish
The similar grade of spectrum, such as building and road.High-resolution remote sensing image in real life application be often relied on towards
The image analysis (object based image analysis, OBIA) of object.OBIA has rich space letter commonly used in distinguishing
The high-resolution remote sensing image of breath, because excessive feature inevitably leads to the confusion of dimension and reduces main information
Accuracy, so researchers make great efforts to carry out feature differentiation work in a large amount of obtainable features always.
The feature selection approach of existing mainstream is dedicated to carrying out feature selecting with supervision law, it requires the label sample of sufficient amount
This.But since marker samples are not only time-consuming but also both expensive, researcher cannot obtain the sample of sufficient amount sometimes, therefore
It can cause to fit phenomenon.In order to solve this problem, in the past few decades, people start to be dedicated to semi-supervised characteristic
The research of selection method.This method carries out data analysis using unmarked sample, while accuracy can be improved.
Existing semi-supervised method is usually by assuming that sample builds on smooth flow pattern to overcome flag data insufficient
Problem, and improve the accuracy generalized within the scope of large space.These semi-supervised methods are based on flow pattern it is assumed that the inherent letter of research
The possibility geometry of breath is distributed, and improves the accuracy of classification.But flow pattern needs to establish a Laplacian Matrix, matrix
Calculating become excessively difficult with the raising of data bulk.In addition to this, these methods all carry out spy for all grades
Sign selection, increases workload.
Summary of the invention
The invention aims to solve the semi-supervised feature selection approach treatment process of existing high-resolution remote sensing image
The problem of complexity, heavy workload, provides a kind of effective semi-supervised feature selection approach towards high-resolution remote sensing image.
Effective semi-supervised feature selection approach of the present invention towards high-resolution remote sensing image, it includes following step
It is rapid:
Step 1: acquisition high-resolution remote sensing image data pre-process high-resolution remote sensing image data；It will be pre-
Treated, and high-resolution remote sensing image data are divided into v+q sample, and wherein v is marker samples quantity, and q is unmarked sample
Quantity, v < < q；To the more a features of each sample extraction m, sample data is obtained；Each feature in sample data is returned
One change processing, obtains corresponding feature vector, and all feature vectors form the characteristic set X after normalization；
Step 2: the corresponding feature vector of v sample randomly selected in characteristic set X carries out sample class mark
Note；
Step 3: label and selection matrix are obtained simultaneously to the feature vector of marker samples classification, building is based on loss letter
The objective function of several and unmarked sample probability distribution matrix；
Step 4: the objective function based on foundation filters selection vector beta and unmarked sample minimum using iterative algorithm deep
Spend yuCategory probability matrix optimize, complete the feature selecting of high-resolution remote sensing image.
Characteristic set X expression formula are as follows:
Wherein xiThe feature vector group that the m feature vector of i-th of sample is formed is corresponded to,For spatial aggregation, i
=1,2,3 ..., v+q.
To the feature vector group x in the v feature vector group extracted in characteristic set XiThe mark of marker samples classification
Note matrix is yi, yijIndicate yiJ-th of element；Make each feature vector group xiIn each feature vector have different samples
This category label, j=1,2,3 ..., m；
For marker samples: if 1 < i < v, works as xiWith j-th of yijY when sample class combinesij=1, otherwise yij=0；
For unmarked sample: if i > v or 0≤yij≤ 1, then for the summation of balanced class probability
C is the sample class total quantity of marker samples in formula.
The sample class of unmarked sample is labeled as 0, for i > v, the initial stage sets yij=0；
Making loss () is loss function, and Ω (W) is regular terms, and λ is regularization coefficient, then objective function are as follows:
Loss function selects minimum quadratic regression, by using limitationStandardization and least square method lose letter
Number, objective function deformation are as follows:
The wherein label matrix of marker samples classificationThe class probability of unmarked sample
Label matrix yu={ yv+1,…,yv+q}T, | | | |FIt is Frobenius norm,
XlFor the eigenvectors matrix of marker samples classification, XuFor the eigenvectors matrix of unmarked sample class, coefficient gamma
<1；
Work as C=2, objective function is further deformed into:
When selection vector beta is fixed, solution category probability matrix, objective function simplifies are as follows:
According to the Laplce L (y of equationu(i, j), τ) it is yuConvex quadratic function:
By equation to yuIt minimizes and derivative is set to 0, then minimum filter depth yuAre as follows:
When category probability matrix is fixed, selection vector beta is solved；The constraint condition of equation is constant, the then La Pu of equation
Lars L (yu(i, j), τ) deformation are as follows:
And then obtain the expression formula of Laplce:
Tr((yl-Xlβ)T(yl-Xlβ))+γTr((yu-Xuβ)T(yu-Xuβ))+λ||β||2,1,
Tr is mark；
The derivative amount that equation is set about parameter is 0, is obtained:
Wherein D is DiiDialogue matrix, DiiFor the element in dialogue matrix；
When selecting vector beta arbitrarily to initialize, Iterative is obtained:
Thus the feature selecting of high-resolution remote sensing image is completed.
Advantages of the present invention: the invention proposes a kind of semi-supervised feature selecting side towards high-resolution remote sensing image
Method --- effectively semi-supervised feature selection approach (efficient semi-supervised feature selection,
ESFS).ESFS method establishes the feature degree based on loss function and probability distribution matrix from semi-supervised method in the present invention
Objective function is measured, an effective algorithm has been used to carry out the class probability of the unmarked target of optimum choice vector sum selectively
Distribution, in the hope of the measurement of each characteristic quantity, wherein the probability of unmarked target is adjustable and suitably infers.Most of half
Supervision Method for Feature Selection relies on image construction to study unmarked target, the method for the present invention semi-supervised feature choosing different from the past
Method is selected, it does not establish the Laplacian Matrix of image, it can realize the linear relationship of point quantity, while it is multiple to reduce calculating
Miscellaneous degree has apparent advantage compared to other feature selection approach when handling mass data, and ESFS method can be adaptive, no
The accuracy rate of selection is improved only, while reducing amount of calculation, facilitates practical application.High-definition remote sensing figure of the present invention
The experimental result of picture is shown, can more effectively be improved than other existing common attribute selection methods using the method for the present invention flat
Equal overall accuracy and Kappa coefficient.
ESFS method of the present invention is a kind of semi-supervised feature selection approach, it extends the application range of method,
It can make full use of sample information when to tagsort.Experimental result indicate, this method relative to traditional classical way not only
Accuracy is significantly improved, and can reduce calculating cost when handling magnanimity high-resolution remote sensing image, has more wide
General application prospect.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the invention；
Fig. 2 is initial holy Clement's high-resolution remote sensing image；
Fig. 3 is holy Clement's high-resolution remote sensing image of reference；
Fig. 4 is to use the overall classification accuracy of Bayes and marker samples number for 10 when with spy for holy Clement's image
Levy the change curve of quantity；
Fig. 5 is to use the overall classification accuracy of SVM and marker samples number for 10 when with feature for holy Clement's image
The change curve of quantity；
Fig. 6 is to use the Kappa coefficient of Bayes and marker samples number for 10 when with characteristic for holy Clement's image
The change curve of amount；
Fig. 7 is to use the Kappa coefficient of SVM and marker samples number for 10 when with feature quantity for holy Clement's image
Change curve；
Fig. 8 is to use the overall classification accuracy of Bayes and marker samples number for 30 when with spy for holy Clement's image
Levy the change curve of quantity；
Fig. 9 is to use the overall classification accuracy of SVM and marker samples number for 30 when with feature for holy Clement's image
The change curve of quantity；
Figure 10 is to use the Kappa coefficient of Bayes and marker samples number for 30 when with feature for holy Clement's image
The change curve of quantity；
Figure 11 is to use the Kappa coefficient of SVM and marker samples number for 30 when with characteristic for holy Clement's image
The change curve of amount.
Specific embodiment
Specific embodiment 1: illustrating present embodiment below with reference to Fig. 1 to Figure 11, towards high score described in present embodiment
Effective semi-supervised feature selection approach of resolution remote sensing images, it the following steps are included:
Step 1: acquisition high-resolution remote sensing image data pre-process high-resolution remote sensing image data；It will be pre-
Treated, and high-resolution remote sensing image data are divided into v+q sample, and wherein v is marker samples quantity, and q is unmarked sample
Quantity, v < < q；To the more a features of each sample extraction m, sample data is obtained；Each feature in sample data is returned
One change processing, obtains corresponding feature vector, and all feature vectors form the characteristic set x after normalization；
Step 2: the corresponding feature vector of v sample randomly selected in characteristic set x carries out sample class mark
Note；
Step 3: label and selection matrix are obtained simultaneously to the feature vector of marker samples classification, building is based on loss letter
The objective function of several and unmarked sample probability distribution matrix；
Step 4: the objective function based on foundation filters selection vector beta and unmarked sample minimum using iterative algorithm deep
Spend yuCategory probability matrix optimize, complete the feature selecting of high-resolution remote sensing image.
Characteristic set X expression formula are as follows:
Wherein xiThe feature vector group that the m feature vector of i-th of sample is formed is corresponded to,For spatial aggregation, i=
1,2,3,…,v+q。
To the feature vector group x in the v feature vector group extracted in characteristic set xiThe mark of marker samples classification
Note matrix is yi, yijIndicate yiJ-th of element；Make each feature vector group xiIn each feature vector have different samples
This category label, j=1,2,3 ..., m；
For marker samples: if 1 < i < v, works as xiWith j-th of yijY when sample class combinesij=1, otherwise yij=0；
For unmarked sample: if i > v or 0≤yij≤ 1, then for the summation of balanced class probability
C is the sample class total quantity of marker samples in formula.
The sample class of unmarked sample is labeled as 0, for i > v, the initial stage sets yij=0；
Making loss () is loss function, and Ω (W) is regular terms, and λ is regularization coefficient, then objective function are as follows:
Loss function selects minimum quadratic regression, by using limitationStandardization and least square method lose letter
Number, objective function deformation are as follows:
The wherein label matrix of marker samples classificationThe class probability of unmarked sample
Label matrix yu={ yv+1,…,yv+q}T, | | | |FIt is Frobenius norm,
XlFor the eigenvectors matrix of marker samples classification, XuFor the eigenvectors matrix of unmarked sample class, coefficient gamma
<1；
Work as C=2, objective function is further deformed into:
When selection vector beta is fixed, solution category probability matrix, objective function simplifies are as follows:
According to the Laplce L (y of equationu(i, j), τ) it is yuConvex quadratic function:
By equation to yuIt minimizes and derivative is set to 0, then minimum filter depth yuAre as follows:
When category probability matrix is fixed, selection vector beta is solved；The constraint condition of equation is constant, the then La Pu of equation
Lars L (yu(i, j), τ) deformation are as follows:
And then obtain the expression formula of Laplce:
Tr((yl-Xlβ)T(yl-Xlβ))+γTr((yu-Xuβ)T(yu-Xuβ))+λ||β||2,1,
Tr is mark；
The derivative amount that equation is set about parameter is 0, is obtained:
Wherein D is DiiDialogue matrix, DiiFor the element in dialogue matrix；
When selecting vector beta arbitrarily to initialize, Iterative is obtained:
Thus the feature selecting of high-resolution remote sensing image is completed.
In step 2, it is labeled as 1 by setting corresponding sample class, is set as 0 for remaining, it can be by c classification
The problem of be converted to two-dimensional problem.
In step 3, most start, is 0 by unlabelled sample labeling.To any i > v, the initial stage sets yij=0.It obtains
Obtain general objectives function.
Simplified style to objective function, which is that constant is simplified by its first term, to be obtained.
Characteristic vector pickup is carried out to each sample in step 1, described eigenvector is as shown in following table:
Carrying out pretreatment to remote sensing image data includes being split to it, extracting feature and normalization etc..To each spy
Sign vector normalized, which refers to, normalizes to [0,1] section for each feature.
The method of the present invention is from semi-supervised learning method, by the optimization key variables of selectivity, select vector sum without
The class probability of label object is distributed, and is solved to target equation.ESFS method can be with the feature of measurement data, and to it
It is selected.
Effect of the invention is verified using specific example below:
The performance in high-resolution remote sensing image processing is being solved the problems, such as in order to verify ESFS method proposed by the invention,
By it and Laplce's score (Laplacian Score, LS), mcLogisticC algorithm, SPEC algorithm, TRCFS algorithm,
S2FS2R algorithm carries out experiment comparison.Testing data set used is two panel height resolution remote sensing images, the sage of respectively Fig. 2 and Fig. 3
The Worldview of Clement, to the two images by eCognition Developer using multi-scale division technology into
Row segmentation.Partitioning parameters include compactness, smoothness, shape, color and scale parameter, are separately arranged as 0.5,0.5,0.1,
0.9 and 50.Both common supervised classification methods using SVM and Bayes respectively, are respectively 10 and 30 feelings in marker samples number
It is tested under condition, experimentation, which is repeated as many times, to carry out, and all dimensions for calculating every kind of algorithm acquired results are averaged overall accuracy
With Kappa coefficient, experimental result is shown in Fig. 4 to Figure 11.
With reference to feature quantity above, by two groups of parameters shown in picture each method average overall accuracy and
Kappa coefficient.When feature quantity improves, ESFS method occupies some superiority on average overall accuracy and Kappa coefficient.Work as spy
When levying quantity greater than 20, the average overall accuracy and Kappa coefficient of ESFS method, which remain unchanged, even to be reduced.ESFS method is in institute
The characteristic of selection effect when within 100 is significantly better than other algorithms.It furthermore being averaged in ESFS method as seen from the figure
Overall accuracy than in other methods it is best it is taller go out close in the Kappa coefficient ratio other methods of 25%, ESFS method most
Good is taller out close to 30%.With the continuous increase of characteristic, the result of all algorithms tends to be identical.Illustrate in feature
When number is high, all kinds of algorithm effects are similar, but when characteristic is limited, the ESFS algorithm effect that we are proposed is considerably better.
Claims (4)
1. a kind of effective semi-supervised feature selection approach towards high-resolution remote sensing image, it the following steps are included:
Step 1: acquisition high-resolution remote sensing image data pre-process high-resolution remote sensing image data；It will pretreatment
High-resolution remote sensing image data afterwards are divided into v+q sample, and wherein v is marker samples quantity, and q is unmarked sample number
Amount, v < < q；To m feature of each sample extraction, sample data is obtained；Each feature in sample data is normalized
Processing, obtains corresponding feature vector, and all feature vectors form the characteristic set X after normalization；
Step 2: the corresponding feature vector of v sample randomly selected in characteristic set X carries out sample class label；
Step 3: obtaining label and selection matrix to the feature vector of marker samples classification simultaneously, building based on loss function and
The objective function of the probability distribution matrix of unmarked sample；
Step 4: the objective function based on foundation filters depth y to selection vector beta and unmarked sample minimum using iterative algorithmu's
Category probability matrix optimizes, and completes the feature selecting of high-resolution remote sensing image；
It is characterized in that, the sample class of unmarked sample is labeled as 0, for i > v, the initial stage sets yij=0, yijIt indicates
yiJ-th of element；
Making loss () is loss function, and Ω (W) is regular terms, and λ is regularization coefficient, then objective function are as follows:
Loss function selects minimum quadratic regression, by using limitationStandardization and least square method loss function,
Objective function deformation are as follows:
The wherein label matrix of marker samples classificationThe mark of the class probability of unmarked sample
Remember matrix yu={ yv+1,…,yv+q}T, | | | |FIt is Frobenius norm,
XlFor the eigenvectors matrix of marker samples classification, XuFor the eigenvectors matrix of unmarked sample class, coefficient gamma < 1；
Work as c=2, objective function is further deformed into:
2. effective semi-supervised feature selection approach according to claim 1 towards high-resolution remote sensing image, feature
It is, characteristic set X expression formula are as follows:
Wherein xiThe feature vector group that the m feature vector of i-th of sample is formed is corresponded to,For spatial aggregation, i=1,2,
3,…,v+q。
3. effective semi-supervised feature selection approach according to claim 2 towards high-resolution remote sensing image, feature
It is,
To the feature vector group x in the v feature vector group extracted in characteristic set XiThe label matrix of marker samples classification
For yi, yijIndicate yiJ-th of element；Make each feature vector group xiIn each feature vector have different sample class
Label, j=1,2,3 ..., m；
For marker samples: if 1 < i < v, works as xiWith j-th of yijY when sample class combinesij=1, otherwise yij=0；
For unmarked sample: if i > v or 0≤yij≤ 1, then for the summation of balanced class probability
C is the sample class total quantity of marker samples in formula.
4. effective semi-supervised feature selection approach according to claim 1 towards high-resolution remote sensing image, feature
It is,
When selection vector beta is fixed, solution category probability matrix, objective function simplifies are as follows:
According to the Laplce L (y of equationu(i, j), τ) it is yuConvex quadratic function:
By equation to yuIt minimizes and derivative is set to 0, then minimum filter depth yuAre as follows:
When category probability matrix is fixed, selection vector beta is solved；The constraint condition of equation is constant, the then Laplce L of equation
(yu(i, j), τ) deformation are as follows:
And then obtain the expression formula of Laplce:
Tr((yl-Xlβ)T(yl-Xlβ))+γTr((yu-Xuβ)T(yu-Xuβ))+λ||β||2,1,
Tr is mark；
The derivative amount that equation is set about parameter is 0, is obtained:
Wherein D is DiiDialogue matrix, DiiFor the element in dialogue matrix；
When selecting vector beta arbitrarily to initialize, Iterative is obtained:
Thus the feature selecting of high-resolution remote sensing image is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610127595.6A CN105809172B (en) | 2016-03-07 | 2016-03-07 | Effective semi-supervised feature selection approach towards high-resolution remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610127595.6A CN105809172B (en) | 2016-03-07 | 2016-03-07 | Effective semi-supervised feature selection approach towards high-resolution remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105809172A CN105809172A (en) | 2016-07-27 |
| CN105809172B true CN105809172B (en) | 2019-07-02 |
Family
ID=56466819
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610127595.6A Expired - Fee Related CN105809172B (en) | 2016-03-07 | 2016-03-07 | Effective semi-supervised feature selection approach towards high-resolution remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105809172B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111898710B (en) * | 2020-07-15 | 2023-09-29 | 中国人民解放军火箭军工程大学 | Feature selection method and system of graph |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680169A (en) * | 2015-03-18 | 2015-06-03 | 哈尔滨工业大学 | Semi-supervised diagnostic characteristic selecting method aiming at thematic information extraction of high-spatial resolution remote sensing image |
| CN105320963A (en) * | 2015-10-21 | 2016-02-10 | 哈尔滨工业大学 | High resolution remote sensing image oriented large scale semi-supervised feature selection method |
-
2016
- 2016-03-07 CN CN201610127595.6A patent/CN105809172B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680169A (en) * | 2015-03-18 | 2015-06-03 | 哈尔滨工业大学 | Semi-supervised diagnostic characteristic selecting method aiming at thematic information extraction of high-spatial resolution remote sensing image |
| CN105320963A (en) * | 2015-10-21 | 2016-02-10 | 哈尔滨工业大学 | High resolution remote sensing image oriented large scale semi-supervised feature selection method |
Non-Patent Citations (2)
| Title |
|---|
| A Convex Formulation for Semi-Supervised Multi-Label Feature Selection;X. Chang等;《Twenty-Eighth AAAI Conference on Artificial Intelligence》;20141231;第1-7页 |
| 基于空间覆盖的半监督特征选择方法;陈红等;《计算机工程与应用》;20101231;第130-132页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105809172A (en) | 2016-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108491849B (en) | Hyperspectral image classification method based on three-dimensional dense connection convolution neural network | |
| Ranzato et al. | Automatic recognition of biological particles in microscopic images | |
| CN107480620B (en) | Remote sensing image automatic target identification method based on heterogeneous feature fusion | |
| CN108830312B (en) | Integrated learning method based on sample adaptive expansion | |
| Ranjan et al. | Hyperspectral image classification: A k-means clustering based approach | |
| CN106203483B (en) | A kind of zero sample image classification method based on semantic related multi-modal mapping method | |
| Lv et al. | Novel adaptive region spectral-spatial features for land cover classification with high spatial resolution remotely sensed imagery | |
| CN105913081B (en) | SAR image classification method based on improved PCAnet | |
| CN105718942B (en) | High spectrum image imbalance classification method based on average drifting and over-sampling | |
| CN111401426B (en) | Small sample hyperspectral image classification method based on pseudo label learning | |
| CN108664986B (en) | Based on lpNorm regularized multi-task learning image classification method and system | |
| CN109409438B (en) | Remote sensing image classification method based on IFCM clustering and variational inference | |
| CN103955709B (en) | Weighted synthetic kernel and triple markov field (TMF) based polarimetric synthetic aperture radar (SAR) image classification method | |
| CN105069478A (en) | Hyperspectral remote sensing surface feature classification method based on superpixel-tensor sparse coding | |
| CN105160351B (en) | Semi-supervised hyperspectral classification method based on anchor point sparse graph | |
| Chen et al. | Agricultural remote sensing image cultivated land extraction technology based on deep learning | |
| Zhang et al. | Object-oriented subspace analysis for airborne hyperspectral remote sensing imagery | |
| CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
| CN105320963B (en) | The semi-supervised feature selection approach of large scale towards high score remote sensing images | |
| Tirandaz et al. | Unsupervised texture-based SAR image segmentation using spectral regression and Gabor filter bank | |
| CN105894035B (en) | SAR image classification method based on SAR-SIFT and DBN | |
| CN107464247A (en) | One kind is based on G0Stochastic gradient variation Bayes's SAR image segmentation method of distribution | |
| CN105809172B (en) | Effective semi-supervised feature selection approach towards high-resolution remote sensing image | |
| He et al. | Wood species identification based on an ensemble of deep convolution neural networks | |
| WO2024082374A1 (en) | Few-shot radar target recognition method based on hierarchical meta transfer |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Yuan Dechun Inventor after: Chen Xi Inventor after: Zhang Ye Inventor before: Chen Xi Inventor before: Zhang Ye Inventor before: Song Lin |
|
| COR | Change of bibliographic data | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190702 Termination date: 20200307 |