CN105279519B - Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning - Google Patents
Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning Download PDFInfo
- Publication number
- CN105279519B CN105279519B CN201510616672.XA CN201510616672A CN105279519B CN 105279519 B CN105279519 B CN 105279519B CN 201510616672 A CN201510616672 A CN 201510616672A CN 105279519 B CN105279519 B CN 105279519B
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- sample set
- grader
- module
- 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
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)
- Image Processing (AREA)
Abstract
The present invention relates to water body remote sensing technology fields, a kind of remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning are provided, when solving to carry out remote sensing image Clean water withdraw using supervised classification method at present, the problem for causing Clean water withdraw precision poor due to the sample size deficiency of artificial selection, the system includes that characteristics of remote sensing image extraction module, dual-view structure module, classifier training module and sort module, classifier training module include initialization module, sample labeling module, sample set update module and iteration control module.Technical solution proposed by the present invention can still ensure the precision of high-resolution remote sensing image Clean water withdraw under Small Sample Size, so as to reduce the workload and complexity of artificial mark training sample.
Description
Technical field
The present invention relates to water body remote sensing technology field, more particularly to a kind of remote sensing shadow based on coorinated training semi-supervised learning
As Clean water withdraw method and system.
Background technology
Traditional remote sensing image Clean water withdraw method includes water body index method, spectrum-photometric method, in addition in image space point
Resolution is higher and wave band it is small numbers of in the case of usually use image classification method.It is more in remote sensing image Clean water withdraw research at present
Using supervised classification method, such as maximum likelihood method, support vector machines, neural network etc., take these methods into
The basic step of row Clean water withdraw is:
(1) feature (based on spectral signature and textural characteristics) composition characteristic vector of the extraction each pixel of image;
(2) some training samples (being labeled as water body or non-water body) are chosen, and using certain supervised learning method to training
Sample set is learnt, and the aim of learning is to reach to minimize empiric risk or structure risk, finally obtain a decision function or
Rule;
(3) it takes above-mentioned decision function or rule to classify the pixel of unknown classification all in image, is judged as
Water body or non-water body, to achieve the purpose that Clean water withdraw.
When background atural object is more complicated in image, due to the presence of jljl different spectrum and same object different images phenomenon, cause
It takes and is often difficult to obtain representative training sample good enough when traditional supervised classification, to influence the essence of Clean water withdraw
Degree.So, in order to improve the precision of Clean water withdraw, enough even infinite number of training samples, Er Qiexun must just be obtained
The quantity for practicing sample is The more the better naturally, therefore a large amount of training sample of acquisition just becomes most important condition.But people
The sample of work mark is limited after all, and the operation complexity for marking training sample is very high, therefore obtains a large amount of marks
The sample of note is unpractical, it is necessary to seek other approach.
Invention content
【Technical problems to be solved】
The remote sensing image Clean water withdraw method based on coorinated training semi-supervised learning that the object of the present invention is to provide a kind of, with
When solving to carry out remote sensing image Clean water withdraw using supervised classification method at present, since the sample size deficiency of artificial selection causes
The poor problem of Clean water withdraw precision.
【Technical solution】
The present invention is achieved by the following technical solutions.
The remote sensing image Clean water withdraw method based on coorinated training semi-supervised learning that present invention firstly relates to a kind of, including step
Suddenly:
A, the spectral signature and textural characteristics of remote sensing image are extracted, the spectral signature includes at least each wave band of remote sensing image
Data X, water body index NDWI and vegetation index NDVI, the textural characteristics include at least the gray level co-occurrence matrixes of remote sensing image
Angular second moment ASM, the uniformity HOM of the gray level co-occurrence matrixes, the entropy ENT of the gray level co-occurrence matrixes and remote sensing image fractal texture
The fractal dimension FD of model, wherein X=[B1,B2,...,Bn]T, n is wave band number, BiFor the gray value of wave band i images, 1≤i≤
n;
B, the double vision being shown below is built according to the spectral signature of the remote sensing image extracted in step A and textural characteristics
Figure:
C, initial training sample is selected from remote sensing image and initial training sample is labeled to obtain mark sample set
L is closed, is generated at random from remote sensing image residue pixel and does not mark sample set U;
D, maximum iteration is set, never mark in sample set U and take out a sample set U' at random;
E, using the view V of mark sample set L1One grader h of training1, use the view V of mark sample set L2
One grader h of training2, use grader h1Classification is carried out to the data in sample set U' and to the highest p sample of confidence level
Originally it is labeled, uses grader h2To the data in sample set U' carry out classification and to the highest p sample of confidence level into
Rower is noted;
F, the sample newly marked in step E is added in mark sample set L, and random never mark sample set U
Middle selection 2p do not mark sample and add in sample set U';
G, judge whether iterations reach maximum iteration and do not mark whether sample set U is empty, if repeatedly
It is sky that generation number, which reaches maximum iteration or do not mark sample set U, then preserves grader h1With grader h2And it executes next
Step, on the contrary then return to step E progress next iterations;
H, the grader h obtained using step G1With grader h2Classify to remote sensing image, extracts the water of remote sensing image
Body information.
As a preferred embodiment, the step H is specifically included:
Remote sensing image is read, the feature vector of each pixel is extracted;
With grader h1To the view V of pixel1Classify, obtains confidence alpha1;
Use grader h2To the view V of pixel2Classify, obtains confidence alpha2;
If α1>α2, then the pixel is grader h1The classification of prediction, conversely, then the pixel is grader h2The class of prediction
Not.
As another preferred embodiment, the water body indexWherein Green,
NIR is the reflectivity of green light band, near infrared band respectively.
As another preferred embodiment, the water body indexWherein Red, NIR distinguish
It is the reflectivity of red spectral band, near infrared band.
As another preferred embodiment, the sample size of sample set L is marked in the step C and does not mark sample
The ratio of the sample size of this set U is 1:20~1:10.
The remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning that the invention further relates to a kind of, including remote sensing
Image feature extraction module, dual-view structure module, classifier training module and sort module, the classifier training module packet
Initialization module, sample labeling module, sample set update module and iteration control module are included,
The characteristics of remote sensing image extraction module is configured to the spectral signature and textural characteristics of extraction remote sensing image, described
Spectral signature includes at least each wave band data X of remote sensing image, water body index NDWI and vegetation index NDVI, and the textural characteristics are extremely
The angular second moment ASM of gray level co-occurrence matrixes including remote sensing image, the uniformity HOM of the gray level co-occurrence matrixes, the gray scale are total less
The fractal dimension FD, wherein X=[B of the entropy ENT and remote sensing image fractal texture model of raw matrix1,B2,...,Bn]T, n is wave band
Number, BiFor the gray value of wave band i images, 1≤i≤n;
The dual-view structure module is configured to the remote sensing image extracted according to characteristics of remote sensing image extraction module
Spectral signature and textural characteristics build the dual-view being shown below:
The initialization module is configured to:From remote sensing image select initial training sample and to initial training sample into
Rower is noted to obtain mark sample set L, is generated at random from remote sensing image residue pixel and does not mark sample set U, setting is maximum
Iterations never mark in sample set U and take out a sample set U' at random;
The sample labeling module is configured to:Use the view V of mark sample set L1One grader h of training1, make
With the view V of mark sample set L2One grader h of training2, use grader h1Data in sample set U' are divided
Class is simultaneously labeled the highest p sample of confidence level, uses grader h2Classified simultaneously to the data in sample set U'
P sample highest to confidence level is labeled;
The sample set update module is configured to:The sample newly marked in sample labeling module is added to mark sample
In this set L, and selects 2p not mark sample in random never mark sample set U and add in sample set U';
The iteration control module is configured to:Judge whether iterations reach maximum iteration or do not mark sample
Whether set U is empty, if it is sky that iterations, which reach maximum iteration or do not mark sample set U, preserves grader
h1With grader h2And by grader h1With grader h2It is sent to sort module, it is on the contrary then return to sample labeling module and carry out down
An iteration;
The sort module is configured to:Use obtained grader h1With grader h2Classify to remote sensing image, carries
Take the Water-Body Information of remote sensing image.
As a preferred embodiment, the sort module includes:
Remote sensing image read module, is configured to:Remote sensing image is read, the feature vector of each pixel is extracted;
First classification submodule, is configured to:Use grader h1To the view V of pixel1Classify, obtains confidence
Spend α1;
Second classification submodule, is configured to:Use grader h2To the view V of pixel2Classify, obtains confidence
Spend α2;
Classification judgment module, is configured to:If α1>α2, then the pixel is grader h1The classification of prediction, conversely, then should
Pixel is grader h2The classification of prediction.
As another preferred embodiment, the water body indexWherein Green,
NIR is the reflectivity of green light band, near infrared band respectively.
As another preferred embodiment, the vegetation indexWherein Red, NIR distinguish
It is the reflectivity of red spectral band, near infrared band.
As another preferred embodiment, marked in the initialization module sample size of sample set L with not
The ratio for marking the sample size of sample set U is 1:20~1:10.
【Advantageous effect】
Technical solution proposed by the present invention has the advantages that:
Technical scheme of the present invention is continuously increased on the basis of building image feature dual-view by coorinated training mode
The quantity for having mark sample is finally carrying out water body pixel identification to get all very strong two graders of generalization ability
When be subject to the prediction result of that higher grader of confidence level, even in the original situation for having mark sample size less
Under, it is also ever-increasing due to there is mark sample size in training process, can also ensures remote sensing image Clean water withdraw
Precision.Therefore, the present invention can still ensure the precision of high-resolution remote sensing image Clean water withdraw under Small Sample Size, from
And the workload and complexity of artificial mark training sample can be reduced.
Description of the drawings
Fig. 1 is the remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning that the embodiment of the present invention one provides
The structure diagram of system.
Fig. 2 is the remote sensing image Clean water withdraw side based on coorinated training semi-supervised learning that the embodiment of the present invention two provides
The flow chart of method.
Specific implementation mode
It to make the object, technical solutions and advantages of the present invention clearer, below will be to the specific implementation mode of the present invention
Carry out clear, complete description.
Embodiment one
Fig. 1 is the remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning that the embodiment of the present invention one provides
Structure diagram.As shown in Figure 1, the system includes characteristics of remote sensing image extraction module, dual-view structure module, classifier training
Module and sort module, classifier training module include initialization module, sample labeling module, sample set update module and change
For control module.
Characteristics of remote sensing image extraction module is configured to the spectral signature and textural characteristics of extraction remote sensing image, spectral signature
Including each wave band data X of remote sensing image, water body index NDWI and vegetation index NDVI, textural characteristics include at least remote sensing image
The angular second moment ASM of gray level co-occurrence matrixes, the uniformity HOM of the gray level co-occurrence matrixes, the entropy ENT of the gray level co-occurrence matrixes and distant
Feel the fractal dimension FD, wherein X=[B of image fractal texture model1,B2,...,Bn]T, n is wave band number, BiFor wave band i images
Gray value, 1≤i≤n.Water body indexWherein Green, NIR are green light band, near-infrared respectively
The reflectivity of wave band.Vegetation indexWherein Red, NIR are red spectral band, near infrared band respectively
Reflectivity.
Dual-view structure module is configured to the spectrum of the remote sensing image extracted according to characteristics of remote sensing image extraction module
Feature and textural characteristics build the dual-view being shown below:
Initialization module is configured to:Initial training sample is selected from remote sensing image and to initial training sample into rower
Note obtains mark sample set L, is generated at random from remote sensing image residue pixel and does not mark sample set U, and greatest iteration is arranged
Number never marks in sample set U and takes out a sample set U' at random.Specifically, sample class is water body and non-water body
Two classes, label are { -1 ,+1 }, mark sample set L={ x1,x2,...,xl, sample set U={ x are not markedl+1,...,
xl+u, the sample size of mark sample set L and the ratio for not marking the sample size of sample set U are 1:15, i.e. l:U=1:
15。
Sample labeling module is configured to:Use the view V of mark sample set L1One grader h of training1, use mark
Note the view V of sample set L2One grader h of training2, use grader h1Classified simultaneously to the data in sample set U'
P sample highest to confidence level is labeled, and uses grader h2To the data in sample set U' carry out classification and it is opposed
The highest p sample of reliability is labeled.It is to be appreciated that the present invention is not intended to limit the value of p.
Sample set update module is configured to:The sample newly marked in sample labeling module is added to mark sample set
It closes in L, and selects 2p not mark sample in random never mark sample set U and add in sample set U'.
Iteration control module is configured to:Judge whether iterations reach maximum iteration or do not mark sample set
Whether U is empty, if it is sky that iterations, which reach maximum iteration or do not mark sample set U, preserves grader h1With
Grader h2And by grader h1With grader h2It is sent to sort module, it is on the contrary then return to sample labeling module and carry out next time
Iteration.
Sort module is configured to:Use obtained grader h1With grader h2Classify to remote sensing image, extraction is distant
Feel the Water-Body Information of image.Specifically, sort module includes remote sensing image read module, the first classification submodule, the second classification
Submodule and classification judgment module.
Remote sensing image read module is configured to:Remote sensing image is read, the feature vector of each pixel is extracted;
First classification submodule is configured to:Use grader h1To the view V of pixel1Classify, obtains confidence alpha1;
Second classification submodule is configured to:Use grader h2To the view V of pixel2Classify, obtains confidence alpha2;
Classification judgment module is configured to:If α1>α2, then the pixel is grader h1The classification of prediction, conversely, the then picture
Member is grader h2The classification of prediction.
Embodiment two
Fig. 2 is the remote sensing image Clean water withdraw method provided by Embodiment 2 of the present invention based on coorinated training semi-supervised learning
Flow chart.As shown in Fig. 2, the method comprising the steps of S1 to step S7, is separately below described in detail above-mentioned steps.
Step S1:Characteristics of remote sensing image extracts.
In step S1, the spectral signature and textural characteristics of remote sensing image are extracted, spectral signature includes each wave band of remote sensing image
Data X, water body index NDWI and vegetation index NDVI, textural characteristics include the angular second moment of the gray level co-occurrence matrixes of remote sensing image
Point of ASM, the uniformity HOM of the gray level co-occurrence matrixes, the entropy ENT of the gray level co-occurrence matrixes and remote sensing image fractal texture model
Shape dimension FD, wherein X=[B1,B2,...,Bn]T, n is wave band number, BiFor the gray value of wave band i images, 1≤i≤n.Water body refers to
NumberWherein Green, NIR are the reflectivity of green light band, near infrared band respectively.Vegetation indexWherein Red, NIR are the reflectivity of red spectral band, near infrared band respectively.
Step S2:Dual-view is built.
In step S2, according to the spectral signature of the remote sensing image extracted in step S1 and textural characteristics structure such as following formula institute
The dual-view shown:
Step S3:Classifier training initializes.
In step S3, initial training sample is selected from remote sensing image and is labeled initial training sample to be marked
Sample set L is generated from remote sensing image residue pixel and is not marked sample set U at random, and maximum iteration is arranged, never marks
A sample set U' is taken out at random in note sample set U.Specifically, sample class is water body and two class of non-water body, corresponding
Label is { -1 ,+1 }, mark sample set L={ x1,x2,...,xl, sample set U={ x are not markedl+1,...,xl+u, mark
The sample size of note sample set L and the ratio for not marking the sample size of sample set U are 1:15, i.e. l:U=1:15.
Step S4:Sample marks.
In step S4, the view V of mark sample set L is used1One grader h of training1, use mark sample set L's
View V2One grader h of training2, use grader h1Classification is carried out to the data in sample set U' and to confidence level highest
P sample be labeled, use grader h2Classification is carried out to the data in sample set U' and to confidence level highest p
Sample is labeled.It is to be appreciated that the present invention is not intended to limit the value of p.
Step S5:Sample set updates.
In step S5, the sample newly marked in step S4 is added in mark sample set L, and never mark sample at random
It selects 2p not mark sample in this set U to add in sample set U'.
Step S6:Judge whether to meet stopping criterion for iteration, grader h is preserved if meeting stopping criterion for iteration1With
Grader h2And execute step S7, on the contrary then return to step S4 progress next iterations.
In step S6, judge whether iterations reach maximum iteration and do not mark whether sample set U is sky,
If it is sky that iterations, which reach maximum iteration or do not mark sample set U, grader h is preserved1With grader h2And it holds
Row next step, on the contrary then return to step S4 progress next iterations.
Step S7:Classify to remote sensing image, extracts the Water-Body Information of remote sensing image.
In step S7, the grader h that is obtained using step S61With grader h2Classify to remote sensing image, extracts remote sensing
The Water-Body Information of image.Step S7 is specifically included:Remote sensing image is read, the feature vector of each pixel is extracted;With grader h1
To the view V of pixel1Classify, obtains confidence alpha1;Use grader h2To the view V of pixel2Classify, is set
Reliability α2;If α1>α2, then the pixel is grader h1The classification of prediction, conversely, then the pixel is grader h2The classification of prediction.
As can be seen from the above embodiments, the embodiment of the present invention passes through association on the basis of building image feature dual-view
The quantity of mark sample is continuously increased with training method, to get all very strong two graders of generalization ability, finally
It is subject to the prediction result of that higher grader of confidence level when carrying out the identification of water body pixel, even has mark original
In the case that sample size is less, it is also ever-increasing due to there is mark sample size in training process, can also protects
Demonstrate,prove the precision of remote sensing image Clean water withdraw.Therefore, the present invention can still ensure high-definition remote sensing shadow under Small Sample Size
As the precision of Clean water withdraw, so as to reduce the workload and complexity of artificial mark training sample.
It is to be appreciated that the embodiment of foregoing description is a part of the embodiment of the present invention, rather than whole embodiments, also not
It is limitation of the present invention.Based on the embodiment of the present invention, those of ordinary skill in the art are not making the creative labor premise
Lower obtained every other embodiment, belongs to protection scope of the present invention.
Claims (10)
1. a kind of remote sensing image Clean water withdraw method based on coorinated training semi-supervised learning, it is characterised in that including step:
A, the spectral signature and textural characteristics of remote sensing image are extracted, the spectral signature includes at least each wave band data of remote sensing image
X, water body index NDWI and vegetation index NDVI, the textural characteristics include at least the angle two of the gray level co-occurrence matrixes of remote sensing image
Rank square ASM, the uniformity HOM of the gray level co-occurrence matrixes, the entropy ENT of the gray level co-occurrence matrixes and remote sensing image fractal texture model
Fractal dimension FD, wherein X=[B1,B2,...,Bn]T, n is wave band number, BiFor the gray value of wave band i images, 1≤i≤n;
B, the dual-view being shown below is built according to the spectral signature of the remote sensing image extracted in step A and textural characteristics:
V1、V2It is the comprehensive spy of spectral signature and textural characteristics composition respectively
Sign vector;
C, initial training sample is selected from remote sensing image and initial training sample is labeled to obtain mark sample set L,
It is generated at random from remote sensing image residue pixel and does not mark sample set U;
D, maximum iteration is set, never mark in sample set U and take out a sample set U' at random;
E, contain V using mark sample set L1One grader h of view training1, contain V using mark sample set L2's
One grader h of view training2, use grader h1Classification is carried out to the data in sample set U' and to the highest p of confidence level
A sample is labeled, and uses grader h2Classification is carried out to the data in sample set U' and to the highest p sample of confidence level
Originally it is labeled;
F, the sample newly marked in step E is added in mark sample set L, and is selected in random never mark sample set U
It selects 2p and does not mark sample and add in sample set U';
G, judge whether iterations reach maximum iteration and do not mark whether sample set U is empty, if iteration time
It is sky that number, which reaches maximum iteration or do not mark sample set U, then preserves grader h1With grader h2And execute next step
Suddenly, on the contrary then return to step E progress next iterations;
H, the grader h obtained using step G1With grader h2Classify to remote sensing image, extracts the water body letter of remote sensing image
Breath.
2. the remote sensing image Clean water withdraw method according to claim 1 based on coorinated training semi-supervised learning, feature
It is that the step H is specifically included:
Remote sensing image is read, the feature vector of each pixel is extracted;
With grader h1V is contained to pixel1View classify, obtain confidence alpha1;
Use grader h2V is contained to pixel2View classify, obtain confidence alpha2;
If α1> α2, then the pixel is grader h1The classification of prediction, conversely, then the pixel is grader h2The classification of prediction.
3. the remote sensing image Clean water withdraw method according to claim 1 or 2 based on coorinated training semi-supervised learning, special
Sign is the water body indexWherein Green, NIR are green light band, near infrared band respectively
Reflectivity.
4. the remote sensing image Clean water withdraw method according to claim 1 or 2 based on coorinated training semi-supervised learning, special
Sign is the vegetation indexWherein Red, NIR are the reflection of red spectral band, near infrared band respectively
Rate.
5. the remote sensing image Clean water withdraw method according to claim 1 or 2 based on coorinated training semi-supervised learning, special
Sign is that the ratio for the sample size for marking the sample size of sample set L in the step C and not marking sample set U is 1:
20~1:10.
6. a kind of remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning, it is characterised in that including remote sensing image
Characteristic extracting module, dual-view structure module, classifier training module and sort module, the classifier training module include just
Beginningization module, sample labeling module, sample set update module and iteration control module,
The characteristics of remote sensing image extraction module is configured to the spectral signature and textural characteristics of extraction remote sensing image, the spectrum
Feature includes at least each wave band data X of remote sensing image, water body index NDWI and vegetation index NDVI, the textural characteristics at least wrap
Include the angular second moment ASM of the gray level co-occurrence matrixes of remote sensing image, the uniformity HOM of the gray level co-occurrence matrixes, the gray scale symbiosis square
The fractal dimension FD, wherein X=[B of the entropy ENT and remote sensing image fractal texture model of battle array1,B2,...,Bn]T, n is wave band number, Bi
For the gray value of wave band i images, 1≤i≤n;
The dual-view structure module is configured to the spectrum of the remote sensing image extracted according to characteristics of remote sensing image extraction module
Feature and textural characteristics build the dual-view being shown below:
V1、V2It is the comprehensive spy of spectral signature and textural characteristics composition respectively
Sign vector;
The initialization module is configured to:Initial training sample is selected from remote sensing image and to initial training sample into rower
Note obtains mark sample set L, is generated at random from remote sensing image residue pixel and does not mark sample set U, and greatest iteration is arranged
Number never marks in sample set U and takes out a sample set U' at random;
The sample labeling module is configured to:Contain V using mark sample set L1One grader h of view training1, make
Contain V with mark sample set L2One grader h of view training2, use grader h1To the data in sample set U' into
Row is classified and is labeled to the highest p sample of confidence level, and grader h is used2Data in sample set U' are divided
Class is simultaneously labeled the highest p sample of confidence level;
The sample set update module is configured to:The sample newly marked in sample labeling module is added to mark sample set
It closes in L, and selects 2p not mark sample in random never mark sample set U and add in sample set U';
The iteration control module is configured to:Judge whether iterations reach maximum iteration or do not mark sample set
Whether U is empty, if it is sky that iterations, which reach maximum iteration or do not mark sample set U, preserves grader h1With
Grader h2And by grader h1With grader h2It is sent to sort module, it is on the contrary then return to sample labeling module and carry out next time
Iteration;
The sort module is configured to:Use obtained grader h1With grader h2Classify to remote sensing image, extraction is distant
Feel the Water-Body Information of image.
7. the remote sensing image Clean water withdraw system according to claim 6 based on coorinated training semi-supervised learning, feature
It is that the sort module includes:
Remote sensing image read module, is configured to:Remote sensing image is read, the feature vector of each pixel is extracted;
First classification submodule, is configured to:Use grader h1V is contained to pixel1View classify, obtain confidence
Spend α1;
Second classification submodule, is configured to:Use grader h2V is contained to pixel2View classify, obtain confidence
Spend α2;
Classification judgment module, is configured to:If α1> α2, then the pixel is grader h1The classification of prediction, conversely, the then picture
Member is grader h2The classification of prediction.
8. the remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning described according to claim 6 or 7, special
Sign is the water body indexWherein Green, NIR are green light band, near infrared band respectively
Reflectivity.
9. the remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning described according to claim 6 or 7, special
Sign is the vegetation indexWherein Red, NIR are the reflection of red spectral band, near infrared band respectively
Rate.
10. the remote sensing image Clean water withdraw system based on coorinated training semi-supervised learning described according to claim 6 or 7,
It is characterized in that marking the sample size of sample set L and the sample size for not marking sample set U in the initialization module
Ratio is 1:20~1:10.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510616672.XA CN105279519B (en) | 2015-09-24 | 2015-09-24 | Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510616672.XA CN105279519B (en) | 2015-09-24 | 2015-09-24 | Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105279519A CN105279519A (en) | 2016-01-27 |
CN105279519B true CN105279519B (en) | 2018-09-21 |
Family
ID=55148500
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510616672.XA Active CN105279519B (en) | 2015-09-24 | 2015-09-24 | Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105279519B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107229910A (en) * | 2017-05-18 | 2017-10-03 | 北京环境特性研究所 | A kind of remote sensing images icing lake detection method and its system |
CN107451616A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | Multi-spectral remote sensing image terrain classification method based on the semi-supervised transfer learning of depth |
US11120337B2 (en) * | 2017-10-20 | 2021-09-14 | Huawei Technologies Co., Ltd. | Self-training method and system for semi-supervised learning with generative adversarial networks |
CN107644235A (en) * | 2017-10-24 | 2018-01-30 | 广西师范大学 | Image automatic annotation method based on semi-supervised learning |
CN108229425A (en) * | 2018-01-29 | 2018-06-29 | 浙江大学 | A kind of identifying water boy method based on high-resolution remote sensing image |
CN108805208B (en) * | 2018-06-13 | 2021-12-31 | 哈尔滨工业大学 | Collaborative training method based on consistency judgment of label-free samples |
CN112131912A (en) * | 2019-06-25 | 2020-12-25 | 北京尚水信息技术股份有限公司 | Remote sensing image underlying surface extraction method |
CN110633758A (en) * | 2019-09-20 | 2019-12-31 | 四川长虹电器股份有限公司 | Method for detecting and locating cancer region aiming at small sample or sample unbalance |
CN111144487B (en) * | 2019-12-27 | 2023-09-26 | 二十一世纪空间技术应用股份有限公司 | Method for establishing and updating remote sensing image sample library |
CN111259366B (en) * | 2020-01-22 | 2021-06-18 | 支付宝(杭州)信息技术有限公司 | Verification code recognizer training method and device based on self-supervision learning |
CN112598685A (en) * | 2021-01-14 | 2021-04-02 | 西安中科星图空间数据技术有限公司 | Automatic water body identification method and device based on multi-scale segmentation |
CN112906666A (en) * | 2021-04-07 | 2021-06-04 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102054274A (en) * | 2010-12-01 | 2011-05-11 | 南京大学 | Method for full automatic extraction of water remote sensing information in coastal zone |
CN102855494A (en) * | 2012-08-20 | 2013-01-02 | 中国测绘科学研究院 | Method and device for extracting water body of satellite remote sensing image |
CN103034863A (en) * | 2012-12-24 | 2013-04-10 | 重庆市勘测院 | Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction |
CN103559500A (en) * | 2013-10-15 | 2014-02-05 | 北京航空航天大学 | Multispectral remote sensing image land feature classification method based on spectrum and textural features |
CN103793907A (en) * | 2013-12-12 | 2014-05-14 | 中国科学院深圳先进技术研究院 | Water body information extracting method and device |
-
2015
- 2015-09-24 CN CN201510616672.XA patent/CN105279519B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102054274A (en) * | 2010-12-01 | 2011-05-11 | 南京大学 | Method for full automatic extraction of water remote sensing information in coastal zone |
CN102855494A (en) * | 2012-08-20 | 2013-01-02 | 中国测绘科学研究院 | Method and device for extracting water body of satellite remote sensing image |
CN103034863A (en) * | 2012-12-24 | 2013-04-10 | 重庆市勘测院 | Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction |
CN103559500A (en) * | 2013-10-15 | 2014-02-05 | 北京航空航天大学 | Multispectral remote sensing image land feature classification method based on spectrum and textural features |
CN103793907A (en) * | 2013-12-12 | 2014-05-14 | 中国科学院深圳先进技术研究院 | Water body information extracting method and device |
Non-Patent Citations (2)
Title |
---|
"组合标记的多视图半监督协同分类算法";于重重 等;《计算机应用》;20131101;第33卷(第11期);第3090-3093页 * |
REGION-BASED SATELLITE IMAGE CLASSIFICATION:METHOD AND VALIDATION;X.Gigandet 等;《Image Processing,2005. ICIP.2005. IEEE International Conference on》;20050327;正文第1-4页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105279519A (en) | 2016-01-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105279519B (en) | Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning | |
Oh et al. | Approaching the computational color constancy as a classification problem through deep learning | |
CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
CN105975931B (en) | A kind of convolutional neural networks face identification method based on multiple dimensioned pond | |
CN108932500B (en) | A kind of dynamic gesture identification method and system based on deep neural network | |
CN106997597B (en) | It is a kind of based on have supervision conspicuousness detection method for tracking target | |
CN105512661B (en) | A kind of Remote Image Classification based on multi-modal Fusion Features | |
CN104318570B (en) | A kind of self adaptation camouflage painting method for designing based on background | |
CN105184309B (en) | Classification of Polarimetric SAR Image based on CNN and SVM | |
CN104809481B (en) | A kind of natural scene Method for text detection based on adaptive Color-based clustering | |
Varga et al. | Fully automatic image colorization based on Convolutional Neural Network | |
CN108010049A (en) | Split the method in human hand region in stop-motion animation using full convolutional neural networks | |
CN106874905B (en) | A method of the natural scene text detection based on self study Color-based clustering | |
CN107909015A (en) | Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion | |
CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
CN109657612B (en) | Quality sorting system based on facial image features and application method thereof | |
CN107844740A (en) | A kind of offline handwriting, printing Chinese character recognition methods and system | |
CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
CN106296695A (en) | Adaptive threshold natural target image based on significance segmentation extraction algorithm | |
CN107146229B (en) | Polyp of colon image partition method based on cellular Automation Model | |
CN107944459A (en) | A kind of RGB D object identification methods | |
CN107358176A (en) | Sorting technique based on high score remote sensing image area information and convolutional neural networks | |
CN108399625A (en) | A kind of SAR image orientation generation method generating confrontation network based on depth convolution | |
CN103984963B (en) | Method for classifying high-resolution remote sensing image scenes | |
CN109753864A (en) | A kind of face identification method based on caffe deep learning frame |
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 |