CN108388907A - Polarization SAR data sorter real time updating method based on various visual angles study - Google Patents
Polarization SAR data sorter real time updating method based on various visual angles study Download PDFInfo
- Publication number
- CN108388907A CN108388907A CN201711482208.1A CN201711482208A CN108388907A CN 108388907 A CN108388907 A CN 108388907A CN 201711482208 A CN201711482208 A CN 201711482208A CN 108388907 A CN108388907 A CN 108388907A
- Authority
- CN
- China
- Prior art keywords
- visual angles
- various visual
- online
- sample
- disaggregated model
- 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
- 230000000007 visual effect Effects 0.000 title claims abstract description 113
- 238000000034 method Methods 0.000 title claims abstract description 78
- 230000010287 polarization Effects 0.000 title claims abstract description 62
- 238000000605 extraction Methods 0.000 claims abstract description 9
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 15
- 238000010168 coupling process Methods 0.000 claims description 12
- 238000005859 coupling reaction Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000002790 cross-validation Methods 0.000 claims description 9
- 230000008878 coupling Effects 0.000 claims description 8
- 241000894007 species Species 0.000 claims description 4
- 241001269238 Data Species 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 17
- 230000000694 effects Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000006165 Knowles reaction Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000003086 colorant Substances 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning 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)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to machine learning algorithms and technical field of image processing, more particularly to a kind of polarization SAR data sorter real time updating method based on various visual angles study, it is intended that solving the problems, such as that grader is unable to real-time update or can only individually update and ignore consistency between visual angle, complementarity is promoted so that influencing nicety of grading, this method includes：S1, the polarimetric SAR image based on t moment, extraction sample polarization characteristic, color characteristic, textural characteristics；S2 passes through online various visual angles disaggregated model, the atural object class label of sample estimates；S3, according to truly species distinguishing label counting loss, the method for solving the online various visual angles disaggregated model closed solutions by method of Lagrange multipliers when loss is more than zero is updated grader；S4 after the polarimetric SAR image for obtaining the t+1 moment, repeats S1 to S3, until whole polarimetric SAR images are disposed.Polarization SAR online data real-time grading may be implemented in the present invention, and online classification error rate is lower.
Description
Technical field
The invention belongs to machine learning algorithms and technical field of image processing, and in particular to it is a kind of based on various visual angles study
Polarization SAR data sorter real time updating method.
Background technology
Polarization SAR is that a kind of advanced microwave remote sensing tool dissipates ground object target under different transmitting-receiving polarization combinations
It penetrates characteristic to measure, the information such as dielectric constant, physical characteristic, geometry and the orientation of target can be obtained.Compared to list
Channel SAR, polarization SAR can obtain more rich terrestrial object information and characteristic of division, be microwave imaging development Main way it
One.Therefore, polarization SAR suffers from wide in fields such as earth resource generaI investigation, environmental disaster control, urban planning, military surveillances
Application prospect.With the development and application of polarization SAR system, the research to the online classification technology of magnanimity polarization SAR data
With important theory value and application value.
Over the last couple of decades, polarization SAR classification problem has obtained extensive concern, and emerged in large numbers include supervision,
A series of semi-supervised and unsupervised sorting techniques, the patent applied such as Xian Electronics Science and Technology University is " based on sparse coding and small
The Classification of Polarimetric SAR Image method of wave self-encoding encoder " (number of patent application：201610407916.8 publication number：CN
106096652 A) " a kind of combination rotational domain polarization zero angle is special for the patent of PLA University of Science and Technology for National Defense's application
The polarization SAR terrain classification method of sign " (number of patent application：201710088598.8 publication number：CN 106909939 A).So
And these existing polarization SAR sorting techniques are all off-line learning algorithms, it is all available when training starts that they, which require data,
, the learning model from training data only when the training is completed can just do classification prediction.In addition, when sample of newly arriving is by mistake point
When, trained grader is no longer updated, or needs to re-start training on entire new data set.Therefore, this kind of
Method environment self-adaption is not strong, and retraining also can be than relatively time-consuming.In addition to this, existing most methods are all single-views point
Class algorithm, that is, only used a type of feature or various features be simply together in series and be combined into a vector characteristics,
And ignore each visual angle characteristic data different attribute and between relationship, affect nicety of grading.
In view of the above-mentioned problems, the present invention proposes a kind of online various visual angles learning method, and for the online of polarization SAR data
In classification task.Different from off-line learning, on-line study is all before capable of efficiently updating grader and will not reusing
Data.For airborne or spaceborne polarization SAR system, data are often large-scale, and continuous according to a continuous sequence
It is acquired.By introducing on-line study, system can incrementally learn a model from data flow, can be high to newly-increased sample
Effect update grader, has compared with strongly-adaptive dynamic environment, has good autgmentability to large-scale data.Therefore, exist
The research of line sorting technique is extremely important to the practical application of polarization SAR.
In recent years, it is suggested there are many kinds of on-line study method, such as perceptron (perceptron) algorithm, under online gradient
(online gradient descent, OGD) algorithm and passive attack (passive-aggressive, PA) algorithm etc. drop.Its
In, the distance between the new grader of PA algorithmic minimizings and previous class device, and new grader is minimized simultaneously on current sample
Loss, (referring to bibliography 1：K.Crammer,O.Dekel,J.Keshet,S.Shalev-Shwartz,and
Y.Singer,“Online passive-aggressive algorithms,”J.Mach.Learn.Res.,vol.7,
No.Mar, pp.551-585,2006.), it is widely used due to its preferable effect and lower computation complexity.However,
This method is only applicable to single-view classification problem.Nguyen et al. proposes two visual angle PA algorithms (referring to bibliography 2：
T.Nguyen,K.Chang,and S.Hui,“Two-view online learning,”in Proc.Pacific-Asia
Conf.Knowl.Discov.Data Min.Springer, 2012, pp.74-85.) and adaptive two visual angles PA algorithms, it is denoted as
AdaPA is (referring to bibliography 3：T.Nguyen,K.Chang,and S.Hui,“Adaptive two-view online
learning for math topic classification,”in Proc.Joint European
Conf.Mach.Learn.Knowl.Discov.Databases.Springer, 2012, pp.794-809.), they are deposited respectively
Combine fixed in weight so that importance between visual angle can not be weighed, consistency constraint item can aggravate inconsistency between visual angle and
There is no the problems such as fit term of parameter tradeoff different visual angles, to which its effect is not ideal enough.For these problems, we have proposed
Online two visual angles PA algorithms are (referring to bibliography 4：X.Nie,S.Ding,H.Qiao,B.Zhang,and X.Y.Huang,
“PolSAR data online classification based on multi-view learning,”in
Proc.Int.Conf.Image Process. (ICIP) .IEEE, 2017.), algorithm effect, which has, to be obviously improved.However, above-mentioned
These three methods are only used for two visual angles and two classification problems, for any number of various visual angles and more classification problems and discomfort
With.Wu et al. proposes that a kind of online multi-modal learning distance metric algorithm is used for image zooming-out (referring to bibliography 5：P.Wu,
S.C.Hoi,P.Zhao,C.Miao,and Z.-Y.Liu,“Online multi-modal distance metric
learning with application to image retrieval,”IEEE Trans.Knowl.Data Eng.,
Vol.28, no.2, pp.454-467,2016.), it can be used in any number of various visual angles problem, however, each of which visual angle
Grader be individually it is newer, utilize visual angle between consistency and complementarity relation, to affect nicety of grading.Needle
To the above problem, online classification method proposed by the present invention consider various visual angles in modeling before relationship, be suitable for arbitrary
Two classification at quantity visual angle and more classification problems.
Invention content
In order to solve the above problem in the prior art, real-time update is unable to or can only individually more in order to solve grader
So that influencing the problem of nicety of grading is promoted, the present invention proposes a kind of based on more for consistency, complementarity between newly ignoring visual angle
The polarization SAR data sorter real time updating method of visual angle study, includes the following steps：
Step S1, the polarimetric SAR image based on t moment extract sample polarization characteristicColor characteristicTexture is special
SignThree perspective datas；
Step S2, is based onPass through online various visual angles disaggregated model, the atural object classification of sample estimates
Label
Step S3, according to truly species distinguishing label ytCounting loss lt, by losing threshold comparison judgement sample with setting
Whether by Correct；Classify if sample solves the online various visual angles by misrepresentation, by method of Lagrange multipliers
The method of model closed solutions is updated the grader in the online various visual angles disaggregated model；
Step S4 after the polarimetric SAR image for obtaining the t+1 moment, repeats step S1 to step S3, until whole polarization SARs
Image procossing finishes.
Further, the online various visual angles disaggregated model is the online various visual angles disaggregated model of two classification tasks；
The online various visual angles disaggregated model of two classification task is：
The sample class label of estimation is
Anticipation function isFor weight parameter and satisfaction
Loss function is lt=max { 0,1-ytft, ytFor true sample label；
Wherein,For the weight vectors of t moment grader, m is the number viewpoints for extracting sample, λiFor different visual angles away from
Balance parameters from variation, diCoupling parameter between visual angle, c are a positive punishment parameter, and ξ is slack variable,For
The weight vectors of t+1 moment graders to be asked, ft+1For the anticipation function at t+1 moment,The sample for being i for t moment visual angle.
Further, described " side of the online various visual angles disaggregated model closed solutions to be solved by method of Lagrange multipliers
Method is updated the grader in the online various visual angles disaggregated model ", method is：
Wherein,
Further, the initialization grader weight of the online various visual angles disaggregated modelFor a random niDimension
Column vector, i.e.,I is i-th of visual angle sample of extraction.
Further, the online various visual angles disaggregated model is the online various visual angles disaggregated model of more classification tasks；
The online various visual angles disaggregated model of more classification tasks is：
The sample class label of estimation is
Anticipation function isFor weight parameter and satisfaction
Loss function isytFor true sample label,
Wherein,For the weight matrix of t moment grader, m is the number viewpoints for extracting sample, λiFor different visual angles away from
Balance parameters from variation, diCoupling parameter between visual angle, c are a positive punishment parameter, and ξ is slack variable,It is the weight matrix of t+1 moment graders, ft+1For the predicted vector at t+1 moment,Exist for the i-th visual angle
The sample of t moment；F is the Forbenius norms of matrix.
Further, described " side of the online various visual angles disaggregated model closed solutions to be solved by method of Lagrange multipliers
Method is updated the grader in the online various visual angles disaggregated model ", method is：
Wherein,
Further, the initialization grader weight of the online various visual angles disaggregated modelIt is initialized as one at random
K × niMatrix,I is i-th of visual angle sample of extraction.
Further, by cross validation selection parameter, the online various visual angles disaggregated model is selected to estimate to obtain classification mark
Sign error rate minimum one group；
By cross validation select parameter include：
The balance parameters λ that different visual angles distance is deterioratedi, the coupling parameter d between visual anglei, penalty parameter c, weight parameter r1,
r2。
Further, polarization characteristic includes the primitive character extracted in polarization SAR data and its transformation directly from acquisition
With the feature based on polarization decomposing；
Color characteristic includes pseudo color image element, dominant color weight and HSV images and its histogram；
Textural characteristics include local binary patterns histogram, gray level co-occurrence matrixes, Gabor and wavelet conversion coefficient.
Further, the value range of input parameter is：
The balance parameters λ that different visual angles distance is deterioratediIncluding λ1、λ2、λ3；λ1=1, λ2, λ3∈ { 1,1.5 }：
Coupling parameter d between visual angleiIncluding d1、d2、d3；d1=d2=d3{ 0.001,0.01,0.1 }；
Penalty parameter c ∈ { 0.05,0.1,0.15 }；
Weight parameter r1, r2∈ { 0.3,0.4 }, and meet
The present invention compared with prior art, has the following advantages：
(1) classification of polarization SAR online data is realized
On-line study method update step proposed by the present invention based on PA has Analytical Expression, therefore can efficiently update point
Class device is overcome the offline sorting technique of existing polarization SAR and does not do update to grader or need to be instructed again using total data
The problem of practicing grader so that the present invention can realize to classify in real time, and to dynamic environment with very strong adaptivity and
There is very strong autgmentability to large-scale data.
(2) online classification error rate is lower
The present invention is regarded from polarization SAR extracting data polarization, color and textural characteristics, and using them as different
Angle, the present invention takes full advantage of consistency between them and complementarity relation models, and overcomes the prior art and only uses
Certain single features, which causes information not develop fully or be unified into a high dimension vector using certain several feature string, leads to computation complexity
Too high deficiency so that present invention utilizes more complete information, can be lower to the error rate of online classification.
Description of the drawings
Fig. 1 is an embodiment of the present invention
Fig. 2 (a) is by Pauli points of the haplopia polarization SAR data in the extensive and profound in meaning areas Po Fafenhuofen of Germany obtained ESAR
Solve pseudo color image；
Fig. 2 (b) is by the truly species corresponding with Fig. 2 (a) in the extensive and profound in meaning areas Po Fafenhuofen of Germany obtained ESAR
Biao Ji it not scheme；
Fig. 3 is the visual contrast result of whole classification chart after the completion of online two classification；
Fig. 4 is the visual contrast result of whole classification chart after the completion of online more classification.
Specific implementation mode
The preferred embodiment of the present invention described with reference to the accompanying drawings.It will be apparent to a skilled person that this
A little embodiments are used only for explaining the technical principle of the present invention, it is not intended that limit the scope of the invention.
A kind of polarization SAR data sorter real time updating method based on various visual angles study of the present invention, including following step
Suddenly：
Step S1, the polarimetric SAR image based on t moment extract sample polarization characteristicColor characteristicTexture
FeatureThree perspective datas；
Step S2, is based onPass through online various visual angles disaggregated model, the atural object classification of sample estimates
Label
Step S3, according to truly species distinguishing label ytCounting loss lt, by losing threshold comparison judgement sample with setting
Whether by Correct；Classify if sample solves the online various visual angles by misrepresentation, by method of Lagrange multipliers
The method of model closed solutions is updated the grader in the online various visual angles disaggregated model；
Step S4 after the polarimetric SAR image for obtaining the t+1 moment, repeats step S1 to step S3, until whole polarization SARs
Image procossing finishes.
Below as changing from input parameter selection, at the beginning of parameter in specific implementation process to grader flow more to the present invention
Technical solution is described in detail, as shown in Figure 1, including the following steps：
Step 1：By cross validation selection parameter, it is contemplated that polarization, color and texture feature, therefore be three to regard
The specific range at angle, i.e. m=3, input parameter is as follows：Penalty parameter c ∈ { 0.05,0.1,0.15 }, balance parameters λ1=1, λ2,
λ3∈ { 1,1.5 }, coupling parameter d1=d2=d3∈ { 0.001,0.01,0.1 }, weight parameter r1, r2∈ { 0.3,0.4 }, and it is full
FootHere the range of choice for giving parameter, by way of cross validation can select optimal parameter takes
Value, i.e. one group of parameter of error rate minimum.
Step 2：The weight for initializing grader, to two classification problems, I.e.For a random niThe column vector of dimension；To more classification problems,I.e.
It is initialized as a random K × niMatrix.
Step 3：Obtain polarization SAR covariance data, extraction polarization, color and textural characteristics：Polarization characteristic includes direct
The primitive character that is extracted from the polarization SAR data of acquisition and its transformation and based on the feature of polarization decomposing；Color characteristic includes
Pseudo color image element, dominant color weight and HSV images and its histogram；Textural characteristics include local binary patterns histogram
Figure, gray level co-occurrence matrixes, Gabor and wavelet conversion coefficient etc..Specifically used polarization, color and textural characteristics are shown in the present invention
Table 1, it can be seen that the dimension of this three category feature is n respectively1=45, n2=34 and n3=86, the present invention is regarded them as three
Angular data is used for subsequent classification.
Table 1：For the polarization of online classification, color and textural characteristics
Step 4：Online various visual angles disaggregated model is established, and according to the label of classification function forecast sample.
Specifically, as follows for the online various visual angles disaggregated model of two classification tasks：
s.t.l(w；(xt, yt))≤ξ；Its expression of ξ >=0. is to the loose constraint of two Classification Loss functions, wherein slack variable
ξ must be non-negative.
Wherein,For the weight vectors of t moment grader, m is the number viewpoints for extracting sample, λiDifferent visual angles away from
Balance parameters from variation, diIt is the coupling parameter between visual angle, c is a positive punishment parameter, and ξ is slack variable,For
The weight vectors of t+1 moment graders to be asked, ft+1For the anticipation function at t+1 moment,The sample for being i for t moment visual angle.
Loss function is defined as hinge-loss losses lt=max { 0,1-ytft}.Anticipation function is defined asWherein ri∈ (0,1) is weight parameter and satisfactionThe sample then estimated
Class label
It is as follows for the online various visual angles disaggregated model of more classification tasks：
s.t.lMC(W；(xt, yt))≤ξ；ξ≥0.The loose constraint to multicategory classification loss function is indicated, wherein relaxation becomes
Amount ξ has to be larger than or is equal to 0.
Wherein, λi, di, c is positive parameter,For the weight matrix of t moment grader,When being t+1
Carve the weight matrix of grader, ft+1For the predicted vector at t+1 moment, F is the Forbenius norms of matrix.
Anticipation functionKnown tolMCIt indicates the loss function of multicategory classification, determines
Justice isytFor true sample label,
In most cases, the class label of the sample of estimation
Step 5：According to the corresponding true tag of sample, counting loss ltIf lt=0, indicate that sample is correctly classified,
Grader is not updated；If lt＞ 0 indicates that sample is classified by mistake, needs to make current class device and update.
For two classification problems, l is lostt=max { 0,1-ytft, if lt＞ 0, to two Classified optimizations in previous step
Problem is solved by method of Lagrange multipliers, and following closed solutions can be obtained, be updated to grader with this：
Wherein,
For more classification problems, lossIf lt＞ 0, in previous step
More Classified optimization problems, are solved by method of Lagrange multipliers, and following closed solutions can be obtained, and are carried out more to grader with this
Newly：
Wherein,
Step 6：If also new samples input, step 1 is returned to；If all samples are all disposed, calculate entire
The classification error rate of on-line study process, the error rate for each subclass are by the sample number of mistake point in subclass than subclass gross sample
This number, total false rate are to compare total number of samples by the sample number of mistake point；Final classification chart is drawn, there is phase in polarimetric SAR image
With the pixel of atural object classification, indicated with same color, and then obtain classification chart.
The subdivided step in the case of two classification tasks and more classification tasks is combined to be further described separately below.
In an embodiment of the present invention, include the following steps for the online various visual angles learning algorithm of two classification tasks：
(1) pass through cross validation selection parameter：Penalty parameter c>0, balance parameters λi＞ 0, coupling parameter di＞ 0 and weight
Parameter ri∈ (0,1) and
(2) weight vectors of grader are initialized Wherein, m is visual angle
Number, niIt is the dimension at i-th of visual angle, it is known that
(3) in t moment, polarization SAR data, extraction polarization are receivedColorAnd textural characteristicsMake respectively
For different visual angles, therefore, the sample of reception
(4) anticipation function is calculated
(5) class label of sample estimates
(6) correct label y is receivedt∈ {+1, -1 }；
(7) counting loss lt=max { 0,1-ytft}；
(8) if lt=0, it indicates that sample is correctly classified, grader is not updated, be directly entered next round iteration；Such as
Fruit lt＞ 0 just does following update to current class device：
(9) auxiliary variable is calculated according to following formula：
(10) grader is updated：
(11) if there is new samples arrival, t=t+1 is returned and is executed (3) step, and otherwise algorithm terminates.
Online various visual angles learning algorithm in an embodiment of the present invention for more classification tasks includes the following steps：
(1) pass through cross validation selection parameter：c>0, λi＞ 0, di＞ 0 and ri∈ (0,1) andWherein m
It is the number at visual angle；
(2) weight vectors of grader are initialized Wherein K is classification
Number, niIt is the dimension at i-th of visual angle, it is known that
(3) in t moment, polarization SAR data, extraction polarization are receivedColorAnd textural characteristicsMake respectively
For different visual angles, therefore, the sample of reception
(4) anticipation function is calculatedKnown to
(5) class label of sample estimates
(6) correct label y is receivedt∈ Y=[1,2 ..., K }；
(7) it calculates
(8) counting loss
(9) if lt=0, it indicates that sample is correctly classified, grader is not updated, be directly entered next round iteration, it is no
Then if lt＞ 0 just does following update to current class device：
(10) auxiliary variable is calculated according to following formula：
(11) grader is updated：
(12) if there is new samples arrival, t=t+1 is returned and is executed (3) step, and otherwise algorithm terminates.
It is described further with reference to Fig. 2-4 pairs of effects of the invention：
Experimental data and condition：
The present invention does test experiments using true polarization SAR data, is the extensive and profound in meaning amber of Germany obtained by E-SAR sensors
The haplopia L-band data in the areas Fa Fenhuofen, can download to from European Space board web.The Pauli of the data decomposes pseudo-colours
Shown in image such as Fig. 2 (a), size 1300*1200, Fig. 2 (b) are its corresponding true atural object classification chart, color lump therein
City, forest land, highway, farmland, other scenes are indicated respectively.
In emulation experiment, the software that uses：MATLAB R2015b, processor：Intel (R) Core (TM) i7-6700HQ,
Memory：20.0GB operating system：64 Windows10.
Experiment content and interpretation of result：
Two classification and more classification tasks are considered in experiment respectively, they correspond to City scenarios extraction and terrain classification respectively
Problem.It is special in polarization, color, texture and combination thereof respectively with PA more preferably to assess the effect of method proposed by the present invention
Comparative result in sign is denoted as PA_Pol, PA_Col, PA_Tex, PA_Cat respectively.In addition, also with propose in bibliography 3
The result of the OMDML methods proposed in AdaPA methods and bibliography 5 compares, and notices that AdaPA methods are only applicable to two
Classification problem, so in more classification tasks not compared with it.In order to preferably compare these methods, the parameter that they include is logical
Cross validation is crossed to choose, the selection range setting of parameter is as follows：The attack parameter c ∈ [0.05,0.15] of PA methods；AdaPA
The coupling parameter d ∈ { 0.001,0.01,0.1 } of method, weight parameter r ∈ (0,1), penalty parameter c ∈ [0.01,0.15]；
The penalty parameter c ∈ [0.01,0.15] of OMDML methods, discount parameter beta ∈ [0.8,1]；The method of the present invention, λ1=1, λ2, λ3
∈ { 1,1.5 }, d1=d2=d3∈ { 0.001,0.01,0.1 }, r1, r2∈ { 0.3,0.4 }, c ∈ { 0.05,0.1,0.15 }.
Fig. 3 illustrates the visual contrast of whole classification chart after the completion of online two classification as a result, Fig. 3 (a) is urban settings carries
The true category label figure taken, non-downtown area is marked as white；Fig. 3 (b)-(h) is PA_Pol, PA_Col, PA_ respectively
The classification results figure of Tex, PA_Cat, AdaPA, OMDML and the present invention.Table 2 gives point of these methods in the case of two classification
Class error rate comparing result.From figure 3, it can be seen that the classification results of PA_Cat will be significantly better than PA_Pol, PA_Col and PA_
Tex's as a result, because PA_Cat in used more characteristic informations, this point can also be confirmed from table 2.In addition, root
It is 43.95% He respectively to be divided into the ratio in non-city by mistake according to urban area in the result of AdaPA and OMDML known to table 2
51.08%, to be significantly higher than mistake point rates 29.72% of the PA_Cat to city, which can also find out from Fig. 3 (e)-(g),
There are many places to be divided into white area by mistake in (f) and (g).From table 2 and Fig. 3 it could be assumed that, side proposed by the present invention
Method has obtained minimum positive sample (i.e. city) classification error rate 22.89% and minimum general classification error rate 7.05%.
Table 2：Classification error rate comparing result in the case of two classification
| Method | PA_Pol | PA_Col | PA_Tex | PA_Cat | AdaPA | OMDML | This method |
| City | 0.4169 | 0.5449 | 0.4442 | 0.2972 | 0.4395 | 0.5108 | 0.2289 |
| Non- city | 0.0777 | 0.1039 | 0.0847 | 0.0571 | 0.0346 | 0.0199 | 0.0396 |
| Total false rate | 0.1333 | 0.1758 | 0.1435 | 0.0965 | 0.1007 | 0.1012 | 0.0705 |
Fig. 4 is the visual contrast of whole classification chart after the completion of more classifying online as a result, its corresponding true class label figure
See Fig. 2 (b).Fig. 4 (a)-(f) is PA_Pol, PA_Col, PA_Tex, the classification results of PA_Cat, OMDML and the present invention respectively
Figure.Table 3 gives the classification error rate comparing result of these methods in the case of more classification.It can be seen that compared to color and line
Feature is managed, polarization characteristic provides better discriminant information, because of the gross errors rate ratio PA_Col and PA_Tex of PA_Pol
Low 10%, in addition, the pixel that PA_Col and PA_Tex has more than half in highway and farmland region is divided by mistake.And PA_Cat
Result to be significantly better than PA_Pol, PA_Col and PA_Tex, and PA_Cat improves the identification of borderline region, such as Fig. 4
(d) shown in.The overall classification accuracy ratio PA_Cat of OMDML improves 6%, and according to Fig. 4 (e), and the boundary of different zones becomes
It must be more clear.According to the visual results in the numerical result and Fig. 4 in table 3, it can be seen that method proposed by the present invention makes
Most sample is correctly classified, and compared with other methods, has reached minimum gross errors rate.
Table 3：Classification error rate comparison in the case of more classification
Those skilled in the art should be able to recognize that, side described in conjunction with the examples disclosed in the embodiments of the present disclosure
Method step, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate electronic hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is executed with electronic hardware or software mode actually, depends on the specific application and design constraint of technical solution.
Those skilled in the art can use different methods to achieve the described function each specific application, but this reality
Now it should not be considered as beyond the scope of the present invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can use hardware, processor to execute
The combination of software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only memory
(ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field
In any other form of storage medium well known to interior.
Term " comprising " or any other like term are intended to cover non-exclusive inclusion, so that including a system
Process, method, article or equipment/device of row element includes not only those elements, but also includes being not explicitly listed
Other elements, or further include the intrinsic element of these process, method, article or equipment/devices.
So far, it has been combined preferred embodiment shown in the drawings and describes technical scheme of the present invention, still, this field
Technical staff is it is easily understood that protection scope of the present invention is expressly not limited to these specific implementation modes.Without departing from this
Under the premise of the principle of invention, those skilled in the art can make the relevant technologies feature equivalent change or replacement, these
Technical solution after change or replacement is fallen within protection scope of the present invention.
Claims (10)
1. a kind of polarization SAR data sorter real time updating method based on various visual angles study, which is characterized in that including following step
Suddenly：
Step S1, the polarimetric SAR image based on t moment extract sample polarization characteristicColor characteristicTextural characteristicsThree perspective datas；
Step S2, is based onPass through online various visual angles disaggregated model, the atural object class label of sample estimates
Step S3, according to truly species distinguishing label ytCounting loss lt, by whether losing threshold comparison judgement sample with setting
By Correct；If sample solves the online various visual angles disaggregated model by misrepresentation, by method of Lagrange multipliers
The method of closed solutions is updated the grader in the online various visual angles disaggregated model；
Step S4 after the polarimetric SAR image for obtaining the t+1 moment, repeats step S1 to step S3, until whole polarimetric SAR images
It is disposed.
2. the polarization SAR data sorter real time updating method according to claim 1 based on various visual angles study, feature
It is, the online various visual angles disaggregated model is the online various visual angles disaggregated model of two classification tasks；
The online various visual angles disaggregated model of two classification task is：
The sample class label of estimation is
Anticipation function isri∈ (0,1) is weight parameter and satisfaction
Loss function is lt=max { 0,1-ytft, ytFor true sample label；
Wherein,For the weight vectors of t moment grader, m is the number viewpoints for extracting sample, λiBecome for different visual angles distance
The balance parameters of difference, diCoupling parameter between visual angle, c are a positive punishment parameter, and ξ is slack variable,To wait asking
T+1 moment graders weight vectors, ft+1For the anticipation function at t+1 moment,The sample for being i for t moment visual angle.
3. the polarization SAR data sorter real time updating method according to claim 2 based on various visual angles study, feature
It is, " the method that the online various visual angles disaggregated model closed solutions are solved by method of Lagrange multipliers, to described online
Grader in various visual angles disaggregated model is updated ", method is：
Wherein,
4. the polarization SAR data sorter real time updating method according to claim 3 based on various visual angles study, feature
It is, the initialization grader weight of the online various visual angles disaggregated modelFor a random niThe column vector of dimension,I is i-th of visual angle sample of extraction.
5. the polarization SAR data sorter real time updating method according to claim 1 based on various visual angles study, feature
It is, the online various visual angles disaggregated model is the online various visual angles disaggregated model of more classification tasks；
The online various visual angles disaggregated model of more classification tasks is：
The sample class label of estimation is
Anticipation function isri∈ (0,1) is weight parameter and satisfaction
Loss function isytFor true sample label,
Wherein,For the weight matrix of t moment grader, m is the number viewpoints for extracting sample, λiBecome for different visual angles distance
The balance parameters of difference, diCoupling parameter between visual angle, c are a positive punishment parameter, and ξ is slack variable,It is the weight matrix of t+1 moment graders, ft+1For the anticipation function at t+1 moment,Exist for the i-th visual angle
The sample of t moment；F is the Forbenius norms of matrix.
6. the polarization SAR data sorter real time updating method according to claim 5 based on various visual angles study, feature
It is, " the method that the online various visual angles disaggregated model closed solutions are solved by method of Lagrange multipliers, to described online
Grader in various visual angles disaggregated model is updated ", method is：
Wherein,
7. the polarization SAR data sorter real time updating method according to claim 6 based on various visual angles study, feature
It is, the initialization grader weight of the online various visual angles disaggregated modelIt is initialized as a random K × niSquare
Battle array,I is i-th of visual angle sample of extraction.
8. according to polarization SAR data sorter real-time update side of the claim 1-7 any one of them based on various visual angles study
Method, which is characterized in that by cross validation selection parameter, the online various visual angles disaggregated model is selected to estimate to obtain class label mistake
Accidentally one group of rate minimum；
By cross validation select parameter include：
The balance parameters λ that different visual angles distance is deterioratedi, the coupling parameter d between visual anglei, penalty parameter c, weight parameter r1, r2。
9. the polarization SAR data sorter real time updating method according to claim 8 based on various visual angles study, feature
It is,
Polarization characteristic includes the primitive character extracted in polarization SAR data directly from acquisition and its transformation and is based on polarization decomposing
Feature；
Color characteristic includes pseudo color image element, dominant color weight and HSV images and its histogram；
Textural characteristics include local binary patterns histogram, gray level co-occurrence matrixes, Gabor and wavelet conversion coefficient.
10. the polarization SAR data sorter real time updating method according to claim 8 based on various visual angles study, special
Sign is that the value range of input parameter is：
The balance parameters λ that different visual angles distance is deterioratediIncluding λ1、λ2、λ3；λ1=1, λ2, λ3∈ { 1,1.5 }：
Coupling parameter d between visual angleiIncluding d1、d2、d3；d1=d2=d3∈ { 0.001,0.01,0.1 }；
Penalty parameter c ∈ { 0.05,0.01,0.15 }；
Weight parameter r1, r2∈ { 0.3,0.4 }, and meet
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711482208.1A CN108388907B (en) | 2017-12-29 | 2017-12-29 | Real-time updating method of polarized SAR data classifier based on multi-view learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711482208.1A CN108388907B (en) | 2017-12-29 | 2017-12-29 | Real-time updating method of polarized SAR data classifier based on multi-view learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108388907A true CN108388907A (en) | 2018-08-10 |
| CN108388907B CN108388907B (en) | 2021-03-02 |
Family
ID=63076666
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711482208.1A Active CN108388907B (en) | 2017-12-29 | 2017-12-29 | Real-time updating method of polarized SAR data classifier based on multi-view learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108388907B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222793A (en) * | 2019-06-20 | 2019-09-10 | 中国科学院自动化研究所 | Online semisupervised classification method and system based on multi-angle of view Active Learning |
| CN110222778A (en) * | 2019-06-11 | 2019-09-10 | 中国科学院自动化研究所 | Online multi-angle of view classification method, system, device based on depth forest |
| CN110414623A (en) * | 2019-08-06 | 2019-11-05 | 广东工业大学 | Disaggregated model construction method and device based on multi-angle of view study |
| CN110766084A (en) * | 2019-10-28 | 2020-02-07 | 北京理工大学 | Small sample SAR target identification method based on CAE and HL-CNN |
| CN111949794A (en) * | 2020-08-14 | 2020-11-17 | 扬州大学 | Online active machine learning method for text multi-classification task |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514456A (en) * | 2013-06-30 | 2014-01-15 | 安科智慧城市技术（中国）有限公司 | Image classification method and device based on compressed sensing multi-core learning |
| CN104463193A (en) * | 2014-11-04 | 2015-03-25 | 西安电子科技大学 | Polarization SAR image classifying method based on depth sparsity ICA |
| US20150316646A1 (en) * | 2014-05-01 | 2015-11-05 | Utah State University | Synthetic aperture radar target modeling |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
| CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
| US20170193337A1 (en) * | 2015-12-31 | 2017-07-06 | Dropbox, Inc. | Generating and utilizing normalized scores for classifying digital objects |
| CN107133649A (en) * | 2017-05-09 | 2017-09-05 | 西安电子科技大学 | Increment type polarization SAR terrain classification method based on a domain distance |
-
2017
- 2017-12-29 CN CN201711482208.1A patent/CN108388907B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514456A (en) * | 2013-06-30 | 2014-01-15 | 安科智慧城市技术（中国）有限公司 | Image classification method and device based on compressed sensing multi-core learning |
| US20150316646A1 (en) * | 2014-05-01 | 2015-11-05 | Utah State University | Synthetic aperture radar target modeling |
| CN104463193A (en) * | 2014-11-04 | 2015-03-25 | 西安电子科技大学 | Polarization SAR image classifying method based on depth sparsity ICA |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
| CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
| US20170193337A1 (en) * | 2015-12-31 | 2017-07-06 | Dropbox, Inc. | Generating and utilizing normalized scores for classifying digital objects |
| CN107133649A (en) * | 2017-05-09 | 2017-09-05 | 西安电子科技大学 | Increment type polarization SAR terrain classification method based on a domain distance |
Non-Patent Citations (2)
| Title |
|---|
| XIANGLI NIE等: "POLSAR DATA ONLINE CLASSIFICATION BASED ON MULTI-VIEW LEARNING", 《IEEE SIGNAL PROCESSING SOCIETY》 * |
| 杨慧: "基于极化SAR技术的舰船检测与识别算法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222778A (en) * | 2019-06-11 | 2019-09-10 | 中国科学院自动化研究所 | Online multi-angle of view classification method, system, device based on depth forest |
| CN110222778B (en) * | 2019-06-11 | 2021-04-13 | 中国科学院自动化研究所 | Online multi-view classification method, system and device based on deep forest |
| CN110222793A (en) * | 2019-06-20 | 2019-09-10 | 中国科学院自动化研究所 | Online semisupervised classification method and system based on multi-angle of view Active Learning |
| CN110222793B (en) * | 2019-06-20 | 2021-06-22 | 中国科学院自动化研究所 | Online semi-supervised classification method and system based on multi-view active learning |
| CN110414623A (en) * | 2019-08-06 | 2019-11-05 | 广东工业大学 | Disaggregated model construction method and device based on multi-angle of view study |
| CN110414623B (en) * | 2019-08-06 | 2022-06-24 | 广东工业大学 | Classification model construction method and device based on multi-view learning |
| CN110766084A (en) * | 2019-10-28 | 2020-02-07 | 北京理工大学 | Small sample SAR target identification method based on CAE and HL-CNN |
| CN110766084B (en) * | 2019-10-28 | 2021-06-15 | 北京理工大学 | Small sample SAR target identification method based on CAE and HL-CNN |
| CN111949794A (en) * | 2020-08-14 | 2020-11-17 | 扬州大学 | Online active machine learning method for text multi-classification task |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108388907B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108388907A (en) | Polarization SAR data sorter real time updating method based on various visual angles study | |
| Chen et al. | An automated approach for updating land cover maps based on integrated change detection and classification methods | |
| Zhou et al. | Wheat ears counting in field conditions based on multi-feature optimization and TWSVM | |
| He et al. | Learning and incorporating top-down cues in image segmentation | |
| CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
| CN112052754B (en) | Polarization SAR image ground object classification method based on self-supervision characterization learning | |
| Zhao et al. | Automatic recognition of loess landforms using Random Forest method | |
| CN104915676A (en) | Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method | |
| CN106503035A (en) | A kind of data processing method of knowledge mapping and device | |
| CN104408483B (en) | SAR texture image classification methods based on deep neural network | |
| CN107358142A (en) | Polarimetric SAR Image semisupervised classification method based on random forest composition | |
| CN112132014B (en) | Target re-identification method and system based on non-supervised pyramid similarity learning | |
| CN105069796B (en) | SAR image segmentation method based on small echo both scatternets | |
| Li et al. | Cauchy graph embedding optimization for built-up areas detection from high-resolution remote sensing images | |
| CN102945553B (en) | Remote sensing image partition method based on automatic difference clustering algorithm | |
| CN105243139A (en) | Deep learning based three-dimensional model retrieval method and retrieval device thereof | |
| CN113223042B (en) | Intelligent acquisition method and equipment for remote sensing image deep learning sample | |
| CN104751175B (en) | SAR image multiclass mark scene classification method based on Incremental support vector machine | |
| CN103745233B (en) | The hyperspectral image classification method migrated based on spatial information | |
| CN108537121A (en) | The adaptive remote sensing scene classification method of environment parament and image information fusion | |
| CN113761259A (en) | Image processing method and device and computer equipment | |
| Li et al. | A novel image template matching based on particle filtering optimization | |
| CN109359525A (en) | The Classification of Polarimetric SAR Image method of differentiation spectral clustering based on sparse low-rank | |
| CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
| CN103456017B (en) | Image partition method based on the semi-supervised weight Kernel fuzzy clustering of subset |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |