CN105303198A - Remote-sensing image semi-supervision classification method based on customized step-size learning - Google Patents
Remote-sensing image semi-supervision classification method based on customized step-size learning Download PDFInfo
- Publication number
- CN105303198A CN105303198A CN201510788769.9A CN201510788769A CN105303198A CN 105303198 A CN105303198 A CN 105303198A CN 201510788769 A CN201510788769 A CN 201510788769A CN 105303198 A CN105303198 A CN 105303198A
- Authority
- CN
- China
- Prior art keywords
- sample
- classification
- samples
- sensing image
- remote sensing
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
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)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a remote-sensing image semi-supervision classification method based on customized step-size learning. The remote-sensing image semi-supervision classification method comprises the following steps of: pre-processing remote-sensing images, and obtaining marked samples of all kinds of surface features; selecting some or all of unmarked samples, constructing a sparse graph by combining the selected unmarked samples with all of the marked samples, and carrying out initial category calibration on the unmarked samples, so that expanding the number of the marked samples; based on a customized step-size learning algorithm, selecting or rejecting category information with initially-marked samples after the expansion; and selecting a supervision classifier to carry out pixel one-by-one classification on the remote-sensing images. According to the invention, the number of training samples is expanded, and the purpose of reducing falsely-calibrated samples in the training samples is achieved.
Description
Technical field
The present invention relates to a kind of remote sensing image semisupervised classification method from fixed step size study.
Background technology
The feature that remotely-sensed data is efficient with it, economic, synchronous on a large scale, has become one of the important technical of earth resources environmental monitoring, planning and management, and has been widely used in region and even global range.The cardinal principle of classification of remote-sensing images in remote sensing image, shows the features such as different spectrum, texture and space geometry characteristic according to type of ground objects, carries out the identification of terrain object attribute classification.Current classification of remote-sensing images mainly contains two kinds of modes, visual interpretation and computer classification.Wherein computer automatic sorting have efficiently, objective advantage, be the effective ways of extensive, repeated Remote Sensing Data Processing and classification.
As everyone knows, in the computer classification process of remote sensing image, the training sample how obtaining sufficient amount is the key issue of computer image classification.The major way of current acquisition training sample is manually chosen in image by visual interpretation or on-the-spot on-site inspection sampling.But obtain training sample in the mode of visual interpretation and require higher to the specialized capability of decipher expert, and subjective degree being comparatively large, may there is certain error during sample in a large amount of selection; On the other hand, though obtain in the mode of on-the-spot on-site inspection the training sample that training sample can obtain degree of precision, but work efficiency is low, and field study costly, the general selection being only suitable for a small amount of training sample.More than show in the computer classification of remote sensing image, the acquisition of training sample will face the problem of two aspects: (1) or training sample data have certain error or noise; Or a small amount of high precision sample can only be obtained (2).Both of these case is all unfavorable for the classification process of remote sensing image, if this is because sample data contains error or quantity is very few, all will used sample be caused to there is the problem of deviation to the under-represented or representativeness of some ground class.The situation of small sample is only had for above high dimensional data process, remote sensing circle proposes to utilize a large amount of unlabelled sample information in image, adopt the semi-supervised treatment technology in machine learning, supplementing of certain form is carried out to marker samples in shortage, thus reaches the object that the ground class bad to representativeness carries out offset correction.
The conventional semi-supervised technology of remote sensing image process mainly contains following four kinds of modes: based on the method for the method of generation model, self study, mutually study, direct-push and figure.Wherein the process of semi-supervised self study is the process of a kind of sorter recurrence matching, is only originally brought in the sample set marked by the nonstandard random sample of satisfied setting confidence threshold value in each recursive procedure, and participates in next recurrence matching.Method based on figure does not need parameter fitting process, but utilizes certain similarity criterion directly to differentiate thus have the feature of high-efficient simple to the unmarked sample participating in computing.But the semi-supervised learning method of self study or drawing method has respective limitation, such as the accumulation of error is easily caused to propagate based on self study sorter recurrence fit procedure; And the restriction of initial sample information is easily subject to based on the graph model of marker samples similarity.
Summary of the invention
In view of this, the object of the present invention is to provide a kind of remote sensing image semisupervised classification method from fixed step size study, in expansion number of training object simultaneously, decrease the object of by mistake demarcating sample in training sample.
For achieving the above object, the present invention adopts following technical scheme: a kind of remote sensing image semisupervised classification method from fixed step size study, is characterized in that comprising the following steps:
Step S1: pre-service is carried out to remote sensing image, what obtain all kinds of atural object has marker samples;
Step S2: select part or all of unmarked sample, has marker samples to carry out sparse composition, and carries out initial category demarcation to unmarked sample described in associating, thus expands the quantity having marker samples;
Step S3: based on from fixed step size learning algorithm, the classification information after expansion with initial markers sample is selected or rejected;
Step S4: select supervised classifier, carry out classifying by pixel to remote sensing image.
Further, the particular content of described step S1 comprises:
Step S11: carry out pre-service to the remote sensing image obtained, described pre-service comprises geometry and radiant correction, image joint and cutting, visual fusion strengthen and image characteristics extracts;
Step S12: in acquisition remote sensing image, every class atural object has marker samples, and has the classification logotype of marker samples described in record, and described remote sensing image comprises c atural object classification, c
i∈ (1,2 ..., c) be the classification logotype of the i-th class sample, X
l=[X
1x
2... X
c] be the set of marker samples, wherein
represent the sample of i classification, wherein, c, i are natural number, n
ibe the number of samples of the i-th classification, x
ijbe a column vector, represent spectral vector or the proper vector of an i-th class atural object jth sample;
Step S13: to marker samples collection X
lwith spectral vector or the proper vector x of unmarked sample
ijbe normalized, make pixel spectral vector two norm value be 1.
Further, the particular content of described step S2 comprises:
Step S21: select part or all of unmarked sample as unmarked sample set, be designated as X
u, wherein in subset, the number of unmarked sample is designated as N
u;
Step S22: combine marker samples collection X
lwith unmarked sample X
u, construct new matrix D ic=[X
1x
u]
b × N, wherein B is the spectral band number of remote sensing image, and N is for there being marker samples collection X
lwith unmarked sample set X
unumber of samples sum;
Step S23: to matrix D ic={y
i| i=1,2 ..., each column element y in N}
i, resolve sparse model
Wherein
be operating as to respective vectors or matrix F-norm square, the restrictive condition be subject to when s.t represents and solves this equation, || α
i||
0≤ K
0represent α
ithe number of middle nonzero element is not more than K
0, α
iirepresent α
iin i-th element, treat all y
i(i=1,2 ..., N) calculate terminate after, the sparse coding α of correspondence
isequential storage is to matrix A=[α
1α
2... α
n];
Step S24: structure similarity matrix W=(| A|+|A
t|)/2, wherein A
trepresent transpose operation to A, | A| represents and to take absolute value computing to matrix A;
Step S25: utilize label pass-algorithm to predict the initial labels of unmarked sample, utilize marker samples to predict the classification information of unmarked sample
Step S26:L
uin each row vector l
icorresponding each unmarked sample z
icategory attribute, and the classification information of each unmarked sample evidence prediction is joined correspondence marker samples collection X
lin subset of all categories, i.e. X'
i=X
i∪ { z
i, thus tentatively obtain markd new samples set X'
l=[X'
1x'
2... X'
c], wherein X
i∪ { z
irepresent sample z
iby set " and " computing joins subset X
iin.
Further, the particular content of described step S3 comprises:
Step S31: add up described new samples set X'
l=[X'
1x'
2... X'
c] in number of all categories, be designated as n'
i, i=1,2 ... c;
Step S32: the Mark-matrix S=[S of initialization Different categories of samples
1s
2... S
c], being wherein labeled as of the i-th classification
wherein n'
i> n
i, s
ij=0,1}, i=1,2 ..., c, j=1,2 ..., n'
iif represent this sample point s
ijselected, being 1, not choosing, is 0, and the method for Mark-matrix S described in initialization is X'
lin from X
lsample labeling be 1, all the other through expand to produce sample labeling be 0;
Step S33: according to described Mark-matrix S, selects mark in every group to correspond to the sample of 1, the supervised classifier training classifier model selected by utilization, and utilizes this model prediction to go out often to organize X'
ithe category attribute of sample in subset, calculates the penalty values after all sample classifications in every group, U=[U
1u
2... U
c], wherein
loss function computing method are: u
ij=1-P
ij, wherein P
ijit is the posterior probability that i-th group of jth sample is classified under train classification models.
Step S34: upgrade samples selection Mark-matrix S, according to the threshold value λ of initial setting up, to often organizing loss function value U
iin be less than position corresponding to described threshold value λ sample labeling be 1, all the other are 0;
Step S35: increase the step-length of threshold value λ: λ=μ * λ, wherein the value of μ be greater than 1 constant;
Step S36: iteration performs step S32 to step S35, and when iterations reaches set point number, or λ is greater than 0.5, then iterative process terminates;
Step S37: according to the S finally often organizing sample labeling
isituation, using all training sample set of sample as last classification being labeled as 1 correspondence, and is recorded as newX=[newX
1newX
2... newX
c], wherein newX
ibe the i-th classification finally have marker samples.
Further, the supervised classifier in described step S33 is maximum likelihood classifier or support vector machine classifier.
Further, in described step S34, the span of λ is (0.01,0.1).
Further, μ=1.1 in described step S35.
Further, the particular content of described step S4 is:
Step S41: select conventional supervised classifier, supervised classifier is herein SVM or k-nearest neighbour classification device, utilizes the new samples newX training classifier model after upgrading;
Step S42: utilize the sorter model trained, the supervised classification by pixel is carried out to raw video data.
The present invention compared with prior art has following beneficial effect:
1, the present invention has universality, is applicable to various types of classification of remote-sensing images process, is also applicable to the semisupervised classification algorithm of various sorter.Such as, for target in hyperspectral remotely sensed image, support vector machine (SVM) sorter etc. can be selected in self study sorter recurrence fit procedure; Multispectral image for middle low resolution then can use maximum likelihood classifier etc.;
2, the present invention is little for initial markers sample requirement amount, and the present invention has good samples selection and rejects mechanism, adopts the quantity of mode Dynamic Recurrent exptended sample from the easier to the more advanced;
3, anti-noise ability of the present invention is strong, and in taxonomy model process of the present invention, while expansion selects nonstandard random sample originally, also dynamically can reject of poor quality, that noise is large sample, therefore, method of the present invention has stronger anti-noise ability;
4, easily extensible application of the present invention, process provides a kind of general not demarcating Method of Sample Selection and reject mechanism, except applying except semisupervised classification, is also applicable to the image processing of image feature feature extraction and dimensionality reduction, target detection etc.
Accompanying drawing explanation
Fig. 1 is the inventive method process flow diagram.
Fig. 2 is that the sample that the present invention is based on sparse patterning process expands process flow diagram.
Fig. 3 the present invention is based on the samples selection and elimination method process flow diagram that learn from fixed step size.
Embodiment
Below in conjunction with drawings and Examples, the present invention will be further described.
As shown in Figure 1, the invention provides a kind of remote sensing image semisupervised classification method from fixed step size study, it is characterized in that comprising the following steps:
Step S1: pre-service is carried out to remote sensing image, what obtain all kinds of atural objects qualified on a small quantity has mark (classification information) sample, and particular content comprises:
Step S11: according to the quality of image in the remote sensing image data source obtained, carry out pre-service to the remote sensing image obtained, described pre-service comprises geometry and radiant correction, image joint and cutting, visual fusion strengthen and feature extraction;
Step S12: according to the data of field factual survey or with visual interpretation mode obtain every class atural object in remote sensing image a little have marker samples, and have the classification logotype of marker samples described in record, described remote sensing image comprises c atural object classification, c
i∈ (1,2 ..., c) be the classification logotype of the i-th class sample, X
l=[X
1x
2... X
c] be the set of marker samples, wherein
represent the sample of i classification, corresponding category label is
Wherein, c, i are natural number, n
ibe the number of samples of the i-th classification, x
ijbe a column vector, represent spectral vector or the proper vector of an i-th class atural object jth sample;
Step S13: to all collection of marker samples X
lwith spectral vector or the proper vector x of unmarked sample
ijbe normalized, each pixel vector is met
namely pixel spectral vector two norm value is 1.
Step S2: select part or all of unmarked sample, marker samples is had to carry out sparse composition (utilizing sparse expression technical construction sample similarity figure) described in associating, and initial category demarcation is carried out to unmarked sample, thus expand the quantity having marker samples, as shown in Figure 2, particular content comprises:
Step S21: according to remote sensing image actual size, selects part or all of unmarked sample as unmarked sample set depending on counting yield situation, is designated as X from remote sensing image
u, wherein in subset, the number of unmarked sample is designated as N
u;
Step S22: combine marker samples collection X
lwith unmarked sample X
u, construct new matrix D ic=[X
1x
u]
b × N, wherein B is spectral band number or the number of features of remote sensing image, and N is for there being marker samples collection X
lwith unmarked sample set X
unumber of samples sum;
Step S23: to Dic={y
i| i=1,2 ..., each column element y in N}
i, resolve sparse model
Wherein
be operating as to respective vectors or matrix F-norm square, the restrictive condition be subject to when s.t represents and solves this equation, || α
i||
0≤ K
0represent α
ithe number of middle nonzero element is not more than K
0, α
iirepresent α
ii-th element in vector, parameter K
0=20, then use Open-Source Tools SPAMS can obtain sparse expression factor alpha
i, treat all y
i(i=1,2 ..., N) calculate terminate after, the sparse coding α of correspondence
isequential storage is to matrix A=[α
1α
2... α
n];
Step S24: structure similarity matrix W=(| A|+|A
t|)/2, wherein A
trepresent transpose operation to A, | A| represents and to take absolute value computing to matrix A; Because W can be expressed as further
the numerical response of often row (or the often going) vector thus in the W matrix figure of corresponding sample in whole Dic connects (similar) situation, and wherein 0 represents and do not connect, and non-zeroly indicates connection;
Step S25: utilize label (classification information) pass-algorithm to predict the initial labels of unmarked sample, utilize marker samples to predict the classification information of unmarked sample
method is as follows: first label information is converted into matrix form L
l, then L
u=(D
uu-W
uu)
-1* W
ul* L
l, wherein D is diagonal matrix, make for carry out accumulation operations by row to W matrix, namely
l
lfor there being the label information (L of marker samples
i=[0 ..., 1 ..., 0] represent that i-th position is 1, all the other are 0), result of calculation
for the initial markers information that unmarked sample is given as calculated afterwards.
Step S26:L
uin each row vector l
icorresponding each unmarked sample z
icategory attribute, its rule determined is i=argmin
il
i, i=1,2 ..., c, joins correspondence marker samples collection X by the classification information of each unmarked sample evidence prediction
lin subset of all categories, i.e. X'
i=X
i∪ { z
i(suppose z here
ibelong to the i-th class), thus tentatively obtain markd new samples set X'
l=[X'
1x'
2... X'
c], wherein X
i∪ { z
irepresent sample z
iby set " and " computing joins subset X
iin.
Step S3: based on from fixed step size learning algorithm, select the classification information after expanding with initial markers sample or reject, as shown in Figure 3, particular content comprises:
Step S31: add up described new samples set X'
l=[X'
1x'
2... X'
c] in number of all categories, be designated as n'
i, i=1,2 ... c, arranges iterations parametric t=0, initial abstraction threshold value λ=0.01, and threshold increase steps μ=1.1;
Step S32: the Mark-matrix S=[S of initialization Different categories of samples
1s
2... S
c], being wherein labeled as of the i-th classification
wherein n'
i> n
i, s
ij=0,1}, i=1,2 ..., c, j=1,2 ..., n'
iif represent this sample point s
ijselected, being 1, not choosing, is 0, and the method for Mark-matrix S described in initialization is X'
lin from X
lsample labeling be 1, all the other through expand to produce sample labeling be 0;
Step S33: according to described Mark-matrix S, mark in every group is selected to correspond to the sample of 1, supervised classifier training classifier model selected by utilization, described supervised classifier is that maximum likelihood classifier or SVM classifier are (if select SVM classifier, then need to utilize logistic regression method that SVM output valve is converted into posterior probability), and utilize this model prediction to go out often to organize X'
ithe category attribute of sample in subset, calculates the penalty values after all sample classifications in every group, U=[U
1u
2... U
c], wherein
loss function computing method are: u
ij=1-P
ij, wherein P
ijit is the posterior probability that i-th group of jth sample is classified under train classification models.
Step S34: upgrade samples selection Mark-matrix S, upgrades selected marker S
i(i=1,2 ..., method c) is the span of the threshold value λ according to initial setting up, λ is 0.01 ~ 0.1, to often organizing loss function value U
iin be less than position corresponding to described threshold value λ sample labeling be 1, all the other are 0;
Step S35: make iterations t=t+1, increases the step-length of threshold value λ: λ=μ * λ;
Step S36: iteration performs step S32 to step S35, and when iterations reaches set point number, or λ is greater than 0.5, then iterative process terminates;
Step S37: according to the S finally often organizing sample labeling
isituation, using all training sample set of sample as last classification being labeled as 1 correspondence, and is recorded as newX=[newX
1newX
2... newX
c], wherein newX
ibe i-th group and final have marker samples.
Step S4: select supervised classifier, carry out by pixel classification remote sensing image, particular content is:
Step S41: select conventional supervised classifier, supervised classifier is herein SVM or k-NN, utilizes the new samples newX training classifier model after upgrading;
Step S42: utilize the sorter model trained, the supervised classification by pixel is carried out to raw video data.
Purposes of the present invention is mainly: when remote sensing image have marker samples less, the sample data that quantity is enough, precision is higher is provided automatically.At the supervised classification of remote sensing image, feature extraction and the aspect such as dimensionality reduction, target detection have vital role, and have use value in teaching and scientific research and production practices.
Generally use together in conjunction with ENVI software from the classification of remote-sensing images of fixed step size study.IDL language is utilized to realize method set forth in the present invention, the remote sensing image semisupervised classification method learnt from fixed step size is made card format, as the Function Extension of ENVI software, also can make separate functional blocks, the sample data file that quantity is enough, precision is higher is provided.
The foregoing is only preferred embodiment of the present invention, all equalizations done according to the present patent application the scope of the claims change and modify, and all should belong to covering scope of the present invention.
Claims (8)
1., from a remote sensing image semisupervised classification method for fixed step size study, it is characterized in that comprising the following steps:
Step S1: pre-service is carried out to remote sensing image, what obtain all kinds of atural object has marker samples;
Step S2: select part or all of unmarked sample, has marker samples to carry out sparse composition, and carries out initial category demarcation to unmarked sample described in associating, thus expands the quantity having marker samples;
Step S3: based on from fixed step size learning algorithm, the classification information after expansion with initial markers sample is selected or rejected;
Step S4: select supervised classifier, carry out classifying by pixel to remote sensing image.
2. the remote sensing image semisupervised classification method from fixed step size study according to claim 1, is characterized in that: the particular content of described step S1 comprises:
Step S11: carry out pre-service to the remote sensing image obtained, described pre-service comprises geometry and radiant correction, image joint and cutting, visual fusion strengthen and image characteristics extracts;
Step S12: in acquisition remote sensing image, every class atural object has marker samples, and has the classification logotype of marker samples described in record, and described remote sensing image comprises c atural object classification, c
i∈ (1,2 ..., c) be the classification logotype of the i-th class sample, X
l=[X
1x
2... X
c] be the set of marker samples, wherein
represent the sample of i classification, wherein, c, i are natural number, n
ibe the number of samples of the i-th classification, x
ijbe a column vector, represent spectral vector or the proper vector of an i-th class atural object jth sample;
Step S13: to marker samples collection X
lwith spectral vector or the proper vector x of unmarked sample
ijbe normalized, make pixel spectral vector two norm value be 1.
3. the remote sensing image semisupervised classification method from fixed step size study according to claim 2, is characterized in that: the particular content of described step S2 comprises:
Step S21: select part or all of unmarked sample as unmarked sample set, be designated as X
u, wherein in subset, the number of unmarked sample is designated as N
u;
Step S22: combine marker samples collection X
lwith unmarked sample X
u, construct new matrix D ic=[X
1x
u]
b × N, wherein B is the spectral band number of remote sensing image, and N is for there being marker samples collection X
lwith unmarked sample set X
unumber of samples sum;
Step S23: to matrix D ic={y
i| i=1,2 ..., each column element y in N}
i, resolve sparse model
Wherein
be operating as to respective vectors or matrix F-norm square, the restrictive condition be subject to when s.t represents and solves this equation, || α
i||
0≤ K
0represent α
ithe number of middle nonzero element is not more than K
0, α
iirepresent α
iin i-th element, treat all y
i(i=1,2 ..., N) calculate terminate after, the sparse coding α of correspondence
isequential storage is to matrix A=[α
1α
2... α
n];
Step S24: structure similarity matrix W=(| A|+|A
t|)/2, wherein A
trepresent transpose operation to A, | A| represents and to take absolute value computing to matrix A;
Step S25: utilize label pass-algorithm to predict the initial labels of unmarked sample, utilize marker samples to predict the classification information of unmarked sample
Step S26:L
uin each row vector l
icorresponding each unmarked sample z
icategory attribute, and the classification information of each unmarked sample evidence prediction is joined correspondence marker samples collection X
lin subset of all categories, i.e. X'
i=X
i∪ { z
i, thus tentatively obtain markd new samples set X'
l=[X'
1x'
2... X'
c], wherein X
i∪ { z
irepresent sample z
iby set " and " computing joins subset X
iin.
4. the remote sensing image semisupervised classification method from fixed step size study according to claim 3, is characterized in that: the particular content of described step S3 comprises:
Step S31: add up described new samples set X'
l=[X'
1x'
2... X'
c] in number of all categories, be designated as n'
i, i=1,2 ... c;
Step S32: the Mark-matrix S=[S of initialization Different categories of samples
1s
2... S
c], being wherein labeled as of the i-th classification
wherein n'
i> n
i, s
ij=0,1}, i=1,2 ..., c, j=1,2 ..., n'
iif represent this sample point s
ijselected, being 1, not choosing, is 0, and the method for Mark-matrix S described in initialization is X'
lin from X
lsample labeling be 1, all the other through expand to produce sample labeling be 0;
Step S33: according to described Mark-matrix S, selects mark in every group to correspond to the sample of 1, the supervised classifier training classifier model selected by utilization, and utilizes this model prediction to go out often to organize X'
ithe category attribute of sample in subset, calculates the penalty values after all sample classifications in every group, U=[U
1u
2... U
c], wherein
loss function computing method are: u
ij=1-P
ij, wherein P
ijit is the posterior probability that i-th group of jth sample is classified under train classification models;
Step S34: upgrade samples selection Mark-matrix S, according to the threshold value λ of initial setting up, to often organizing loss function value U
iin be less than position corresponding to described threshold value λ sample labeling be 1, all the other are 0;
Step S35: increase the step-length of threshold value λ: λ=μ * λ, wherein the value of μ be greater than 1 constant;
Step S36: iteration performs step S32 to step S35, and when iterations reaches set point number, or λ is greater than 0.5, then iterative process terminates;
Step S37: according to the S finally often organizing sample labeling
isituation, using all training sample set of sample as last classification being labeled as 1 correspondence, and is recorded as newX=[newX
1newX
2... newX
c], wherein newX
ibe the i-th classification finally have marker samples.
5. the remote sensing image semisupervised classification method from fixed step size study according to claim 4, is characterized in that: the supervised classifier in described step S33 is maximum likelihood classifier or support vector machine classifier.
6. the remote sensing image semisupervised classification method from fixed step size study according to claim 4, is characterized in that: in described step S34, the span of λ is (0.01,0.1).
7. the remote sensing image semisupervised classification method from fixed step size study according to claim 4, is characterized in that: μ=1.1 in described step S35.
8. the remote sensing image semisupervised classification method from fixed step size study according to claim 4, is characterized in that: the particular content of described step S4 is:
Step S41: select conventional supervised classifier, supervised classifier is herein SVM or k-nearest neighbour classification device, utilizes the new samples newX training classifier model after upgrading;
Step S42: utilize the sorter model trained, the supervised classification by pixel is carried out to raw video data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510788769.9A CN105303198B (en) | 2015-11-17 | 2015-11-17 | A kind of remote sensing image semisupervised classification method learnt from fixed step size |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510788769.9A CN105303198B (en) | 2015-11-17 | 2015-11-17 | A kind of remote sensing image semisupervised classification method learnt from fixed step size |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105303198A true CN105303198A (en) | 2016-02-03 |
| CN105303198B CN105303198B (en) | 2018-08-17 |
Family
ID=55200440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510788769.9A Expired - Fee Related CN105303198B (en) | 2015-11-17 | 2015-11-17 | A kind of remote sensing image semisupervised classification method learnt from fixed step size |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105303198B (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250836A (en) * | 2016-07-26 | 2016-12-21 | 山东师范大学 | Two benches facial image sorting technique under a kind of condition of small sample and system |
| CN107194373A (en) * | 2017-06-16 | 2017-09-22 | 河海大学 | A kind of target in hyperspectral remotely sensed image feature extraction and classifying method |
| CN107203755A (en) * | 2017-05-31 | 2017-09-26 | 中国科学院遥感与数字地球研究所 | It is a kind of to increase new methods, devices and systems automatically for remote sensing images time series marker samples |
| CN107239741A (en) * | 2017-05-10 | 2017-10-10 | 杭州电子科技大学 | A kind of single sample face recognition method based on sparse reconstruct |
| CN107657282A (en) * | 2017-09-29 | 2018-02-02 | 中国石油大学(华东) | Peptide identification from step-length learning method |
| CN107909120A (en) * | 2017-12-28 | 2018-04-13 | 南京理工大学 | Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method |
| CN107944479A (en) * | 2017-11-16 | 2018-04-20 | 哈尔滨工业大学 | Disease forecasting method for establishing model and device based on semi-supervised learning |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| WO2018072663A1 (en) * | 2016-10-18 | 2018-04-26 | 腾讯科技（深圳）有限公司 | Data processing method and device, classifier training method and system, and storage medium |
| CN108333140A (en) * | 2018-02-01 | 2018-07-27 | 福州大学 | Characteristic spectrum index and detection method for rigid bamboo poison moth hazard detection |
| CN108388907A (en) * | 2017-12-29 | 2018-08-10 | 中国科学院自动化研究所 | Polarization SAR data sorter real time updating method based on various visual angles study |
| CN108509970A (en) * | 2017-02-27 | 2018-09-07 | 意法半导体股份有限公司 | Learning method, corresponding system, equipment and computer program product |
| CN109146847A (en) * | 2018-07-18 | 2019-01-04 | 浙江大学 | A kind of wafer figure batch quantity analysis method based on semi-supervised learning |
| CN109447111A (en) * | 2018-09-20 | 2019-03-08 | 杭州师范大学 | A kind of remote sensing supervised classification method based on subclass training sample |
| CN109726641A (en) * | 2019-01-24 | 2019-05-07 | 常州大学 | A kind of remote sensing image cyclic sort method based on training sample Automatic Optimal |
| CN109902390A (en) * | 2018-12-13 | 2019-06-18 | 中国石油大学(华东) | A kind of Favorable Reservoir development area prediction technique expanded based on small sample |
| CN110096997A (en) * | 2019-04-29 | 2019-08-06 | 宁波大学 | A kind of littoral zone sophisticated category method of integrated multi- source Remote Sensing Data data |
| CN110232317A (en) * | 2019-05-05 | 2019-09-13 | 五邑大学 | Hyperspectral image classification method based on super-pixel segmentation and two phase classification strategy |
| CN110298211A (en) * | 2018-03-21 | 2019-10-01 | 北京大学 | A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image |
| CN110796262A (en) * | 2019-09-26 | 2020-02-14 | 北京淇瑀信息科技有限公司 | Test data optimization method and device of machine learning model and electronic equipment |
| CN110852175A (en) * | 2019-10-16 | 2020-02-28 | 华东师范大学 | Depth network hyperspectral remote sensing classification method based on mixed limited Boltzmann machine |
| CN110874465A (en) * | 2018-08-31 | 2020-03-10 | 浙江大学 | Mobile equipment entity identification method and device based on semi-supervised learning algorithm |
| CN111222539A (en) * | 2019-11-22 | 2020-06-02 | 国际竹藤中心 | Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image |
| CN112906666A (en) * | 2021-04-07 | 2021-06-04 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
| CN115205689A (en) * | 2022-09-14 | 2022-10-18 | 北京数慧时空信息技术有限公司 | Improved unsupervised remote sensing image anomaly detection method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
| CN103593676A (en) * | 2013-11-29 | 2014-02-19 | 重庆大学 | High-spectral remote-sensing image classification method based on semi-supervision sparse discriminant embedding |
-
2015
- 2015-11-17 CN CN201510788769.9A patent/CN105303198B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
| CN103593676A (en) * | 2013-11-29 | 2014-02-19 | 重庆大学 | High-spectral remote-sensing image classification method based on semi-supervision sparse discriminant embedding |
Non-Patent Citations (2)
| Title |
|---|
| 宋文等: ""基于半监督KFCM及邻域信息的遥感图像分类算法"", 《计算机工程与应用》 * |
| 祁亨年等: ""基于多类支持向量机的遥感图像分类及其半监督式改进策略"", 《复旦学报（自然科学版）》 * |
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250836A (en) * | 2016-07-26 | 2016-12-21 | 山东师范大学 | Two benches facial image sorting technique under a kind of condition of small sample and system |
| WO2018072663A1 (en) * | 2016-10-18 | 2018-04-26 | 腾讯科技（深圳）有限公司 | Data processing method and device, classifier training method and system, and storage medium |
| US11860976B2 (en) | 2016-10-18 | 2024-01-02 | Tencent Technology (Shenzhen) Company Limited | Data processing method and device, classifier training method and system, and storage medium |
| US11960988B2 (en) | 2017-02-27 | 2024-04-16 | Stmicroelectronics S.R.L. | Learning method, corresponding system, device and computer program product to update classifier model parameters of a classification device |
| CN108509970A (en) * | 2017-02-27 | 2018-09-07 | 意法半导体股份有限公司 | Learning method, corresponding system, equipment and computer program product |
| CN108509970B (en) * | 2017-02-27 | 2023-08-25 | 意法半导体股份有限公司 | Learning method, corresponding system, device and computer program product |
| CN107239741A (en) * | 2017-05-10 | 2017-10-10 | 杭州电子科技大学 | A kind of single sample face recognition method based on sparse reconstruct |
| CN107239741B (en) * | 2017-05-10 | 2020-09-08 | 厦门瞳景智能科技有限公司 | Single-sample face recognition method based on sparse reconstruction |
| CN107203755B (en) * | 2017-05-31 | 2021-08-03 | 中国科学院遥感与数字地球研究所 | Method, device and system for automatically adding new time sequence mark samples of remote sensing images |
| CN107203755A (en) * | 2017-05-31 | 2017-09-26 | 中国科学院遥感与数字地球研究所 | It is a kind of to increase new methods, devices and systems automatically for remote sensing images time series marker samples |
| CN107194373A (en) * | 2017-06-16 | 2017-09-22 | 河海大学 | A kind of target in hyperspectral remotely sensed image feature extraction and classifying method |
| CN107657282A (en) * | 2017-09-29 | 2018-02-02 | 中国石油大学(华东) | Peptide identification from step-length learning method |
| CN107944479A (en) * | 2017-11-16 | 2018-04-20 | 哈尔滨工业大学 | Disease forecasting method for establishing model and device based on semi-supervised learning |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| CN107909120A (en) * | 2017-12-28 | 2018-04-13 | 南京理工大学 | Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method |
| CN108388907A (en) * | 2017-12-29 | 2018-08-10 | 中国科学院自动化研究所 | Polarization SAR data sorter real time updating method based on various visual angles study |
| CN108388907B (en) * | 2017-12-29 | 2021-03-02 | 中国科学院自动化研究所 | Real-time updating method of polarized SAR data classifier based on multi-view learning |
| CN108333140A (en) * | 2018-02-01 | 2018-07-27 | 福州大学 | Characteristic spectrum index and detection method for rigid bamboo poison moth hazard detection |
| CN108333140B (en) * | 2018-02-01 | 2020-10-09 | 福州大学 | Characteristic spectral index for detecting damage of bamboo moths and detection method |
| CN110298211A (en) * | 2018-03-21 | 2019-10-01 | 北京大学 | A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image |
| CN109146847B (en) * | 2018-07-18 | 2022-04-05 | 浙江大学 | Wafer map batch analysis method based on semi-supervised learning |
| CN109146847A (en) * | 2018-07-18 | 2019-01-04 | 浙江大学 | A kind of wafer figure batch quantity analysis method based on semi-supervised learning |
| CN110874465A (en) * | 2018-08-31 | 2020-03-10 | 浙江大学 | Mobile equipment entity identification method and device based on semi-supervised learning algorithm |
| CN109447111B (en) * | 2018-09-20 | 2021-12-03 | 杭州师范大学 | Remote sensing supervision classification method based on subclass training samples |
| CN109447111A (en) * | 2018-09-20 | 2019-03-08 | 杭州师范大学 | A kind of remote sensing supervised classification method based on subclass training sample |
| CN109902390B (en) * | 2018-12-13 | 2023-10-24 | 中国石油大学(华东) | Method for predicting development area of favorable reservoir based on small sample expansion |
| CN109902390A (en) * | 2018-12-13 | 2019-06-18 | 中国石油大学(华东) | A kind of Favorable Reservoir development area prediction technique expanded based on small sample |
| CN109726641A (en) * | 2019-01-24 | 2019-05-07 | 常州大学 | A kind of remote sensing image cyclic sort method based on training sample Automatic Optimal |
| CN110096997A (en) * | 2019-04-29 | 2019-08-06 | 宁波大学 | A kind of littoral zone sophisticated category method of integrated multi- source Remote Sensing Data data |
| CN110232317A (en) * | 2019-05-05 | 2019-09-13 | 五邑大学 | Hyperspectral image classification method based on super-pixel segmentation and two phase classification strategy |
| CN110232317B (en) * | 2019-05-05 | 2023-01-03 | 五邑大学 | Hyper-spectral image classification method, device and medium based on super-pixel segmentation and two-stage classification strategy |
| CN110796262B (en) * | 2019-09-26 | 2023-09-29 | 北京淇瑀信息科技有限公司 | Test data optimization method and device of machine learning model and electronic equipment |
| CN110796262A (en) * | 2019-09-26 | 2020-02-14 | 北京淇瑀信息科技有限公司 | Test data optimization method and device of machine learning model and electronic equipment |
| CN110852175A (en) * | 2019-10-16 | 2020-02-28 | 华东师范大学 | Depth network hyperspectral remote sensing classification method based on mixed limited Boltzmann machine |
| CN111222539B (en) * | 2019-11-22 | 2021-07-30 | 国际竹藤中心 | Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image |
| CN111222539A (en) * | 2019-11-22 | 2020-06-02 | 国际竹藤中心 | Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image |
| WO2022214039A1 (en) * | 2021-04-07 | 2022-10-13 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
| CN112906666A (en) * | 2021-04-07 | 2021-06-04 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
| CN115205689A (en) * | 2022-09-14 | 2022-10-18 | 北京数慧时空信息技术有限公司 | Improved unsupervised remote sensing image anomaly detection method |
| CN115205689B (en) * | 2022-09-14 | 2022-11-18 | 北京数慧时空信息技术有限公司 | Improved unsupervised remote sensing image anomaly detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105303198B (en) | 2018-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105303198A (en) | Remote-sensing image semi-supervision classification method based on customized step-size learning | |
| Deng et al. | Active transfer learning network: A unified deep joint spectral–spatial feature learning model for hyperspectral image classification | |
| Ye et al. | A lightweight model of VGG-16 for remote sensing image classification | |
| Gong et al. | A spectral and spatial attention network for change detection in hyperspectral images | |
| Tarawneh et al. | Invoice classification using deep features and machine learning techniques | |
| Cheng et al. | Semi-supervised learning and feature evaluation for RGB-D object recognition | |
| Xia et al. | Exploiting deep features for remote sensing image retrieval: A systematic investigation | |
| Zhou et al. | Image retrieval based on effective feature extraction and diffusion process | |
| CN105184298A (en) | Image classification method through fast and locality-constrained low-rank coding process | |
| CN113657425A (en) | Multi-label image classification method based on multi-scale and cross-modal attention mechanism | |
| WO2023019698A1 (en) | Hyperspectral image classification method based on rich context network | |
| CN109492589A (en) | The recognition of face working method and intelligent chip merged by binary features with joint stepped construction | |
| Bai et al. | Coordinate CNNs and LSTMs to categorize scene images with multi-views and multi-levels of abstraction | |
| Wang et al. | CLARE: A joint approach to label classification and tag recommendation | |
| Liao et al. | Class-wise graph embedding-based active learning for hyperspectral image classification | |
| Schäfer et al. | Classifying Land Cover from Satellite Images Using Time Series Analytics. | |
| CN114780767A (en) | Large-scale image retrieval method and system based on deep convolutional neural network | |
| Yang et al. | Unsupervised images segmentation via incremental dictionary learning based sparse representation | |
| CN114492634B (en) | Fine granularity equipment picture classification and identification method and system | |
| CN107633264A (en) | Linear common recognition integrated fusion sorting technique based on empty spectrum multiple features limit study | |
| Liu et al. | A novel deep framework for change detection of multi-source heterogeneous images | |
| Jiang et al. | Automated detection of multitype landforms on mars using a light-weight deep learning-based detector | |
| CN117132804B (en) | Hyperspectral image classification method based on causal cross-domain small sample learning | |
| Sahbi et al. | Active learning for interactive satellite image change detection | |
| Zhou et al. | Wavelet domain multi-view active learning for hyperspectral image analysis |
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: 20180817 Termination date: 20211117 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |