CN110009017A - A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character - Google Patents
A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character Download PDFInfo
- Publication number
- CN110009017A CN110009017A CN201910226726.XA CN201910226726A CN110009017A CN 110009017 A CN110009017 A CN 110009017A CN 201910226726 A CN201910226726 A CN 201910226726A CN 110009017 A CN110009017 A CN 110009017A
- Authority
- CN
- China
- Prior art keywords
- visual angle
- model
- angle characteristic
- category label
- data
- 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.)
- Pending
Links
- 230000000007 visual effect Effects 0.000 title claims abstract description 111
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000002372 labelling Methods 0.000 title claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000012360 testing method Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 26
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000013145 classification model Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 230000004069 differentiation Effects 0.000 abstract description 3
- 238000010801 machine learning Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 abstract description 2
- 241001269238 Data Species 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000003550 marker Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 102000008297 Nuclear Matrix-Associated Proteins Human genes 0.000 description 1
- 108010035916 Nuclear Matrix-Associated Proteins Proteins 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000299 nuclear matrix Anatomy 0.000 description 1
- 239000012466 permeate Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
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)
Abstract
The present invention relates to the Multi-label learning technologies in machine learning field, are related to a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character, comprising: S1, obtain training data, establish category label matrix；Each visual angle characteristic data are mapped to the linear model of category label matrix after S2, building category label；S3, on the basis of linear model, establish each visual angle characteristic contribution degree model；S4, using canonical item constraint visual angle characteristic contribution degree model, keep each visual angle characteristic data with uniformity in prediction result；S5, the similitude that corresponding model coefficient is marked using popular canonical constraint related category；S6, label prediction, give a test sample t, test sample t are brought into and obtains fusion forecasting value in step S1-S5.The technical solution of the application realizes effective use multi-source information, learns different characteristic in each visual angle and preferably carries out Multi-label learning task to the differentiation performance of category label.
Description
Technical field
The present invention relates to the Multi-label learning technology in machine learning field, it is related to for the view in multi-angle of view Multi-label learning
The study of angle generic character and sorting technique, in particular to a kind of multi-angle of view multiple labeling classification side based on the study of visual angle generic character
Method.
Background technique
Under big data environment, the semanteme and knowledge of data carry out table often through the content information at multiple modalities or visual angle
It reaches, and each data sample may belong to multiple semantic markers simultaneously.For example, a document can in text categorization task
It can include the numerous types of data such as text, image, video and hyperlink, and at the same time comprising multiple semantic topics, such as " engineering
Practise ", " data mining " and " Multi-label learning " etc..Multi-angle of view Multi-label learning is the important of data mining and machine learning field
Research direction.For multi-angle of view Multi-label learning task, the complementarity and consistency of multi-angle of view data are made full use of, is excavated each
The discriminating power that perspective data indicates proposes effective fusion and indicates mechanism, promotes the Shandong of multi-angle of view Multi-label learning model
Stick and accuracy are particularly important.
Multi-angle of view Multi-label learning research at present is broadly divided into front-end convergence and rear end fusion.It front-end convergence or will regard more
Angular data indicates the data expression that permeates, and then resettles classifier, such as by matrix decomposition, multi-angle of view data are all reflected
Being mapped to same sub-spaces indicates, and constrains subspace and indicate orthogonal and low-rank, and the mode finally by matrix fill-in is not to marking
Numeration is according to being predicted.But this method sub-spaces indicate that study independently of prediction process, does not use mark information effectively.
Or the merging features at all visual angles are constituted into character representation together and carry out classifier study, during this
By being L to classifier coefficient2,1Norm constraint carries out F norm constraint by the coefficient to each visual angle to select feature
To optimize the importance at each visual angle.But the feature of this method selection will be shared by all labels, not account for each visual angle
Judgement index of the middle single feature to classification.
Or utilize the structural information of each multi-angle of view data and one son of semantic structure information learning of category label
Space representation, and subspace is indicated that study and SVM support vector machine classifier train this two step to optimize simultaneously.It saves
The structural information and semantic structure information of multiple perspective datas, but this mode space complexity is higher, is not suitable for extensive
Data.
Or multi-angle of view data, which are mapped to same sub-spaces, by Non-negative Matrix Factorization indicates, then constructs one again
Linear classifier.This method does not account for the contribution degree of single visual angle characteristic data.
Or a k neighbour figure is constructed to each perspective data first, then pass through a linear combination for all visual angles
On k neighbour's figure be fused into a new k neighbour figure, this fused k neighbour is finally schemed to the feature as all data
It indicates, and learns a linear classifier on this basis.Save the k Near-neighbor Structure information of multiple perspective datas, space
Complexity is higher, is not suitable for large-scale data.
Rear end fusion mainly first learns a model to each perspective data, then retells the prediction that multiple models obtain
As a result it is merged, such as a kind of direct-push multi-angle of view Multi-label learning method, the intersection information of training sample is broadcast to and is not marked
Remember sample, k neighbour figure is established to each visual angle, constrains similar sample with similar marker.The algorithm complexity is number of samples
5 powers, be difficult to apply to extensive multi-angle of view multiple labeling data.For another example pass through Boosting (promotion) and Bagging (dress
Bag) strategy, several SVM support vector machine classifiers are learnt to each perspective data, finally merge on all visual angles own
The result of classifier is predicted.Also have in the training stage, a SVM classifier is learnt to each perspective data, is then led to
SVM is crossed to predict training sample, it is comprehensive according to the prediction result of training sample and the mark information of its neighbour's sample
Learn the blending weight at each visual angle out.In test phase, by the close of the prediction result of each SVM classifier and test sample
The mark information of adjacent sample obtains final predictive marker according to fusion weight.Latter two method is although it is contemplated that single visual angle is special
The contribution degree of data is levied, but does not account in single perspective data each feature to the contribution degree of category label.
In conclusion multiple labeling data are by multiple visual angles or the information representation of mode, dimension is usually higher, each visual angle number
According to physical significance and statistical property it is different, the data at each visual angle indicate there is different discriminating powers, and single visual angle
The different characteristic of data is also different to the judgement index of each classification.Existing method is although it is contemplated that the tributes of single visual angle characteristic data
Degree of offering, but due to there are noise and redundancy feature etc., may cause the visual angle contribution degree directly learnt in multi-angle of view data
Weight inaccuracy.And existing method does not all account in single perspective data each feature to the contribution degree of category label.Cause
How this, efficiently use multi-source information, learns different characteristic in each visual angle and preferably carries out to the differentiation performance of category label
Multi-label learning task is extremely crucial.
Summary of the invention
The present invention provides a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character, realizes effective use
Multi-source information learns different characteristic in each visual angle and preferably carries out Multi-label learning task to the differentiation performance of category label.
To realize the above-mentioned technical purpose, the technical solution that the application uses for, it is a kind of based on visual angle generic character study
Multi-angle of view multiple labeling classification method, includes the following steps:
S1, training data is obtained, and training data is subjected to category label, establish category label matrix；
Each visual angle characteristic data are mapped to the linear model of category label matrix after S2, building category label, using as point
Class device；
S3, on the basis of linear model, according to the contribution degree of visual angle characteristic data, establish each visual angle characteristic contribution degree mould
Type；
S4, using canonical item constraint visual angle characteristic contribution degree model, have each visual angle characteristic data in prediction result
There is consistency；
S5, the influence using the correlation between category label to the model coefficient of linear model: if two category labels
Correlation is stronger, then their corresponding model coefficients can be very close on the contrary then more dissimilar；Associated class is constrained using popular canonical
The similitude of corresponding model coefficient is not marked；
S6, label prediction, give a test sample t, test sample t are brought into and obtains fusion forecasting value in step S1-S5.
As the improved technical solution of the present invention, obtaining training data includes that setting training data has m kind character representationM is any positive integer, and v kind visual angle characteristic is expressed as a real number matrix Xv,Wherein, n is indicated
Number of samples, dvIndicate Characteristic Number,Indicate real number field.
It include setting Y ∈ { 0,1 } by training data progress category label as the improved technical solution of the present inventionn×qIt is
The category label matrix of training data, q indicate total category label number, wherein YiiThe member of the i-th row j column in representing matrix Y
Element, TijI-th of sample of=1 expression belongs to j-th of category label, otherwise Yij=0, i are 1 to the positive integer between n, j 1
To the positive integer between q.
As the improved technical solution of the present invention, step S2 includes establishing to indicate X based on any v kind visual angle characteristicvIt learns
Practise the linear classification model f for being mapped to category label matrix Yv(Xv, Wv)=XvWv, and to model parameter
It is L1Canonical constrains to learn the generic character in each visual angle characteristic data expression, obtains minimizing target formula:
In formula one,For m model parameter to be solved, λ1For nonnegative curvature coefficient, codomain is { 10-5, 10-4, 10-3, 10-2, 10-1, 100, 101}。
As the improved technical solution of the present invention, establishing each visual angle characteristic contribution degree model in step S3 includes, definition view
The contribution degree of corner characteristics data integrates as θ=[θ1, θ2..., θm], define the contribution degree θ of each visual angle characteristic datav>=0, v 1
To the positive integer between m, value is bigger, and the contribution degree for illustrating that v kind visual angle characteristic data indicate is bigger, on the contrary then smaller, and
Constraining the sum of total contribution degree of all visual angle characteristic data is 1；Minimum target based on each visual angle characteristic contribution degree model is public
Formula
Wherein,It is model parameter to be solved, λ with θ2For nonnegative curvature coefficient, codomain is { 104, 105,
106}。
It include: constraint any two visual angle characteristic data sorter in step S4 as the improved technical solution of the present invention
Output similitude,
Wherein,It is model parameter to be solved, λ with θ3For nonnegative curvature coefficient, codomain is { 100, 101,
102, 103}；
It enablesWherein | | it is ABS function, θiFor the contribution degree of i-th of visual angle characteristic data, θvFor
The contribution degree of v-th of visual angle characteristic data；
S=1/m is smoothing factor, prevents θiWith θvDenominator is zero when equal；
For regular terms, for constrain any two viewpoint classification device output it is similar
Property；XiIt is that i-th kind of visual angle characteristic indicates, WiIt is that x is indicated based on i-th kind of visual angle characteristiciThe model coefficient of study.
It include: corresponding using popular canonical constraint mark of correlation as the improved technical solution of the present invention, in step S5
The similitude of model coefficient,
Wherein,It is model parameter to be solved, λ with θ4For nonnegative curvature coefficient, codomain is { 103, 104,
105, 106}；Tr () representing matrix trace norm,For category label correlation matrixLaplce
Matrix, category label correlation matrix P are obtained by the column count cosine similarity in category label matrix Y.
As the improved technical solution of the present invention, step S6 label prediction: the m visual angle of a given test sample t is special
Sign indicates Indicate that the v kind visual angle characteristic of test sample t indicates vector.
The m model coefficient obtained according to training stage formula fiveTest data t is obtained a pre-
Measured value
According to the contribution degree weight θ=[θ for the m perspective data that formula five obtains1, θ2..., θm] and formula six in obtain
As a result, calculate test sample t fusion forecasting value
According to the fusion forecasting value of test sample t obtained in formula seven, and the threshold value threshold of setting, calculates and survey
The final output label vector yt ∈ { 0,1 } of sample sheet1×q, wherein [] is indicator function, returned when the condition in bracket meets
1 is returned, otherwise returns to 0,
yt(l)=[ft(l) > threshold], 1≤l≤q formula eight.
Beneficial effect
Multi-angle of view data are usually mapped to a sub-spaces and indicated by conventional multi-view Multi-label learning algorithm, are not accounted for
The contribution that single feature classifies to multiple labeling into the expression of each visual angle characteristic, while some methods are used using fusion visual angle
K neighbour figure or nuclear matrix, space complexity are higher.Compared to conventional method, the application indicates to learn to each visual angle characteristic
One low-dimensional generic character indicates, obtains the low-dimensional character representation for having judgement index in each visual angle expression to classification, can reduce
Redundancy and noise characteristic influence the performance that multiple labeling is classified, and reduce model complexity, and memory space needed for saving model can
To be used for extensive multi-angle of view Multi-label learning.On the other hand, which indicates that study and visual angle are contributed for visual angle generic character
Degree is unified to a frame with Consistency Learning, so that low-dimensional character representation learns and visual angle sharing degree study mutually guidance, it is total
With promotion, the low-dimensional character representation for having strong judgement index and each visual angle characteristic number of suitable multi-angle of view Multi-label learning are finally obtained
According to contribution degree weight, to more effectively instruct multi-angle of view multiple labeling classification task.
Detailed description of the invention
The multi-angle of view multiple labeling classification method schematic diagram that Fig. 1 is learnt based on visual angle generic character.
Specific embodiment
To keep purpose and the technical solution of the embodiment of the present invention clearer, below in conjunction with the attached of the embodiment of the present invention
Figure, is clearly and completely described the technical solution of the embodiment of the present invention.Obviously, described embodiment is of the invention
A part of the embodiment, instead of all the embodiments.Based on described the embodiment of the present invention, those of ordinary skill in the art
Every other embodiment obtained, shall fall within the protection scope of the present invention under the premise of being not necessarily to creative work.
As shown in Figure 1, a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character, including walk as follows
It is rapid:
S1, training data is obtained, and training data is subjected to category label, establish category label matrix；
Each visual angle characteristic data are mapped to the linear model of category label matrix after S2, building category label, using as point
Class device；
S3, on the basis of linear model, according to the contribution degree of visual angle characteristic data, establish each visual angle characteristic contribution degree mould
Type；
S4, using canonical item constraint visual angle characteristic contribution degree model, have each visual angle characteristic data in prediction result
There is consistency；
S5, the influence using the correlation between category label to the model coefficient of linear model: if two category labels
Correlation is stronger, then their corresponding model coefficients can be very close on the contrary then more dissimilar；Associated class is constrained using popular canonical
The similitude of corresponding model coefficient is not marked；
S6, label prediction, give a test sample t, test sample t are brought into and obtains fusion forecasting value in step S1-S5.
Specifically, a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character, it is assumed that training data has m
Kind character representationV kind visual angle characteristic is expressed as a real number matrixWherein indicate that n indicates sample
Number, dvIndicate Characteristic Number,Indicate real number field.Y ∈ { 0,1 }n×qIt is the category label matrix of training data, q indicates total
Category label number, wherein YijThe element of the i-th row j column in representing matrix Y, YijI-th of sample of=1 expression belongs to j-th
Category label, otherwise Yij=0, i are 1 to the positive integer between n, and j is 1 to the positive integer between q.Based on visual angle generic character
The multi-angle of view multiple labeling sorting algorithm of habit mainly includes model construction and training and label two stages of prediction.
(1) model construction and training:
(1-1) we use linear model as classifier, indicate X based on any v kind visual angle characteristicvStudy one is reflected
It is mapped to the linear classification model f of category label matrix Yv(Xv, Wv)=XvWv, and to model parameterIt is L1Canonical
It constrains to learn the generic character in each perspective data expression, thus minimizes mesh shown in our available formula (1)
Mark formula:
WhereinFor m model parameter to be solved, λ1For nonnegative curvature coefficient, codomain is { 10-5, 10-4,
10-3, 10-2, 10-1, 100, 101}
The judgement index that (1-2) is indicated in view of each perspective data is different, we learn the tribute of each visual angle characteristic data
Degree of offering θv>=0 (v is 1 to the positive integer between m), value is bigger, and the contribution degree for illustrating that v kind visual angle characteristic indicates is bigger, instead
It is then smaller, and the sum of total contribution degree for constraining all perspective datas be 1.Enable θ=[θ1, θ2..., θm], available formula
(2) target formula is minimized shown in:
WhereinIt is model parameter to be solved, λ with θ1Meaning and setting be same as above.λ2For nonnegative curvature coefficient,
Codomain is { 104, 105, 106}。
(1-3) has centainly in prediction result although the contribution degree of each visual angle characteristic data is different
Consistency, so their model exports fv(Xv, Wv) should be as consistent as possible.And the contribution degree for working as two models is closer,
Then their output is also more similar, on the contrary then differ bigger.It enablesWherein | | it is ABS function, θiIt is
The contribution degree of i visual angle characteristic data, θvFor the contribution degree of v-th of visual angle characteristic data.S=1/m is smoothing factor, prevents θi
With θvDenominator is zero when equal.Therefore, we increase a regular terms To constrain any two
The similitude of the output of a viewpoint classification device.To obtain minimizing target formula shown in formula (3):
WhereinIt is model parameter to be solved, λ with θ1And λ2Meaning and setting be same as above.λ3For nonnegative curvature system
Number, codomain are { 100, 101, 102, 103}。
(1-4) modeling utilizes the correlation between label, if two category label correlations are stronger, their corresponding moulds
Type coefficient can be very close on the contrary then more dissimilar.In the model coefficient that v-th of perspective data learnsIn, benefit
With the similitude of the corresponding model coefficient of popular canonical constraint mark of correlation, obtain minimizing target formula shown in formula (4):
WhereinIt is model parameter to be solved, λ with θ1, λ2And λ3Meaning and setting be same as above.λ4For non-negative right
Weight coefficient, codomain are { 103, 104, 105, 106}.Tr () representing matrix trace norm,For category label correlation
MatrixLaplacian Matrix, category label correlation matrix P passes through the column count cosine in category label matrix Y
Similarity obtains.
(2) label prediction:
The m visual angle characteristic of (2-1) given test sample t indicates
The m model coefficient that (2-2) is obtained according to the training stage (1-4)Test data t is obtained
A predicted value
The contribution degree weight θ=[θ for the m perspective data that (2-3) basis (1-4) obtains1, θ2..., θm], and (2-2)
Obtained in as a result, calculate test sample t fusion forecasting value
The fusion forecasting value of (2-4) test sample t according to obtained in (2-3), and the threshold value threshold of setting,
Calculate the final output label vector y of test samplet∈ { 0,1 }1×q, wherein [] is indicator function, when the condition in bracket is full
1 is returned when sufficient, otherwise returns to 0.
yt(l)=[ft(l) > threshold], 1≤l≤q.
The above method effectively solves to learn the low-dimensional generic character that single visual angle characteristic indicates to indicate, removal noise characteristic and
Redundancy feature；Learn the contribution degree weight of single visual angle characteristic data.
To sum up, the application indicates that study low-dimensional generic character indicates to each visual angle characteristic；It has the technical effect that obtain each
There is the low-dimensional character representation of judgement index in the expression of visual angle to classification, remove redundancy and noise characteristic, reduce model complexity, saves
Memory space needed for storage model can be used for extensive multi-angle of view Multi-label learning.
Visual angle generic character is indicated that study and visual angle contribution degree are combined with consistency by the application；Having the technical effect that makes
It obtains the study of low-dimensional character representation and sharing degree study in visual angle is mutually instructed, collectively promote, finally obtain suitable multi-angle of view multiple labeling
The low-dimensional character representation for having strong judgement index of study and the contribution degree weight of each visual angle characteristic data.
The above is only embodiments of the present invention, and the description thereof is more specific and detailed, and but it cannot be understood as right
The limitation of the invention patent range.It should be pointed out that for those of ordinary skill in the art, not departing from the present invention
Under the premise of design, various modifications and improvements can be made, these are all belonged to the scope of protection of the present invention.
Claims (8)
1. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character, which comprises the steps of:
S1, training data is obtained, and training data is subjected to category label, establish category label matrix；
Each visual angle characteristic data are mapped to the linear model of category label matrix after S2, building category label, using as classifier；
S3, on the basis of linear model, according to the contribution degree of visual angle characteristic data, establish each visual angle characteristic contribution degree model；
S4, using canonical item constraint visual angle characteristic contribution degree model, so that each visual angle characteristic data is had one in prediction result
Cause property；
S5, the influence using the correlation between category label to the model coefficient of linear model: if two category label correlations
Property is stronger, then their corresponding model coefficients can be very close on the contrary then more dissimilar；Related category mark is constrained using popular canonical
Remember the similitude of corresponding model coefficient；
S6, label prediction, give a test sample t, test sample t are brought into and obtains fusion forecasting value in step S1-S5.
2. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is that obtaining training data includes that setting training data has m kind character representationM is any positive integer, v kind view
Corner characteristics are expressed as a real number matrix Xv,Wherein, n indicates number of samples, dvIndicate Characteristic Number,It indicates
Real number field.
3. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is, includes setting Y ∈ { 0,1 } by training data progress category labeln×qIt is the category label matrix of training data, q table
Show total category label number, wherein YijThe element of the i-th row j column in representing matrix Y, YijI-th of sample category of=1 expression
In j-th of category label, otherwise Yij=0, i are 1 to the positive integer between n, and j is 1 to the positive integer between q.
4. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is that step S2 includes establishing to indicate X based on any v kind visual angle characteristicvStudy one is mapped to category label matrix Y's
Linear classification model fv(Xv, Wv)=XvWv, and to model parameterIt is L1Canonical constraint is special to learn each visual angle
The generic character in data expression is levied, obtains minimizing target formula:
In formula one,For m model parameter to be solved, λ1For nonnegative curvature coefficient, codomain is { 10-5, 10-4, 10-3, 10-2, 10-1, 100, 101}。
5. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is that each visual angle characteristic contribution degree model is established in step S3 includes, define visual angle characteristic data contribution degree integrate as θ=
[θ1, θ2..., θm], define the contribution degree θ of each visual angle characteristic datav>=0, v are 1 to the positive integer between m, and value is bigger,
The contribution degree for illustrating that v kind visual angle characteristic data indicate is bigger, on the contrary then smaller, and constrains total tribute of all visual angle characteristic data
The sum of degree of offering is 1；Minimum target formula based on each visual angle characteristic contribution degree model
Wherein,It is model parameter to be solved, λ with θ2For nonnegative curvature coefficient, codomain is { 104, 105, 106}。
6. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is, includes: the similitude for constraining the output of any two visual angle characteristic data sorter in step S4,
Wherein,It is model parameter to be solved, λ with θ3For nonnegative curvature coefficient, codomain is { 100, 101, 102,
103}；
It enablesWherein | | it is ABS function, θiFor the contribution degree of i-th of visual angle characteristic data, θvIt is v-th
The contribution degree of visual angle characteristic data；
S=1/m is smoothing factor, prevents θiWith θvDenominator is zero when equal；
For regular terms, for constrain any two viewpoint classification device output similitude；
XiIt is that i-th kind of visual angle characteristic indicates, WiIt is that X is indicated based on i-th kind of visual angle characteristiciThe model coefficient of study.
7. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is, includes: the similitude that the corresponding model coefficient of mark of correlation is constrained using popular canonical in step S5,
Wherein,It is model parameter to be solved, λ with θ4For nonnegative curvature coefficient, codomain is { 103, 104, 105,
106}；Tr () representing matrix trace norm,For category label correlation matrixLaplacian Matrix,
Category label correlation matrix P is obtained by the column count cosine similarity in category label matrix Y.
8. a kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character according to claim 1, special
Sign is that step S6 label prediction: the m visual angle characteristic of a given test sample t indicates It indicates
The v kind visual angle characteristic of test sample t indicates vector；
The m model coefficient obtained according to training stage formula fiveA predicted value is obtained to test data t
According to the contribution degree weight θ=[θ for the m perspective data that formula five obtains1, θ2..., θm] and formula six obtained in tie
Fruit calculates the fusion forecasting value of test sample t
According to the fusion forecasting value of test sample t obtained in formula seven, and the threshold value threshold of setting, test specimens are calculated
This final output label vector yt∈ { 0,1 }1×q, wherein [] is indicator function, 1 is returned when the condition in bracket meets,
Otherwise 0 is returned,
yt(l)=[ft(l) > threshold], 1≤l≤q.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910226726.XA CN110009017A (en) | 2019-03-25 | 2019-03-25 | A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910226726.XA CN110009017A (en) | 2019-03-25 | 2019-03-25 | A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110009017A true CN110009017A (en) | 2019-07-12 |
Family
ID=67167899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910226726.XA Pending CN110009017A (en) | 2019-03-25 | 2019-03-25 | A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110009017A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110378739A (en) * | 2019-07-23 | 2019-10-25 | 中国联合网络通信集团有限公司 | A kind of data traffic matching process and device |
| CN110619367A (en) * | 2019-09-20 | 2019-12-27 | 哈尔滨理工大学 | Joint low-rank constraint cross-view-angle discrimination subspace learning method and device |
| CN111723241A (en) * | 2020-05-08 | 2020-09-29 | 天津大学 | Short video automatic labeling method based on feature and multi-label enhanced representation |
| WO2021022571A1 (en) * | 2019-08-05 | 2021-02-11 | 南京智谷人工智能研究院有限公司 | Interactive modeling-based multi-label distance metric learning method |
| CN113380413A (en) * | 2021-06-24 | 2021-09-10 | 首都医科大学附属北京天坛医院 | Method and device for constructing invalid re-circulation (FR) prediction model |
| CN117237748A (en) * | 2023-11-14 | 2023-12-15 | 南京信息工程大学 | Picture identification method and device based on multi-view contrast confidence |
-
2019
- 2019-03-25 CN CN201910226726.XA patent/CN110009017A/en active Pending
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110378739A (en) * | 2019-07-23 | 2019-10-25 | 中国联合网络通信集团有限公司 | A kind of data traffic matching process and device |
| CN110378739B (en) * | 2019-07-23 | 2022-03-29 | 中国联合网络通信集团有限公司 | Data traffic matching method and device |
| WO2021022571A1 (en) * | 2019-08-05 | 2021-02-11 | 南京智谷人工智能研究院有限公司 | Interactive modeling-based multi-label distance metric learning method |
| CN110619367A (en) * | 2019-09-20 | 2019-12-27 | 哈尔滨理工大学 | Joint low-rank constraint cross-view-angle discrimination subspace learning method and device |
| CN111723241A (en) * | 2020-05-08 | 2020-09-29 | 天津大学 | Short video automatic labeling method based on feature and multi-label enhanced representation |
| CN111723241B (en) * | 2020-05-08 | 2023-11-03 | 天津大学 | Short video automatic labeling method based on feature and multi-label enhancement representation |
| CN113380413A (en) * | 2021-06-24 | 2021-09-10 | 首都医科大学附属北京天坛医院 | Method and device for constructing invalid re-circulation (FR) prediction model |
| CN117237748A (en) * | 2023-11-14 | 2023-12-15 | 南京信息工程大学 | Picture identification method and device based on multi-view contrast confidence |
| CN117237748B (en) * | 2023-11-14 | 2024-02-23 | 南京信息工程大学 | Picture identification method and device based on multi-view contrast confidence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110009017A (en) | A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character | |
| CN106407352B (en) | Traffic image search method based on deep learning | |
| CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
| CN104573669B (en) | Image object detection method | |
| CN104881689B (en) | A kind of multi-tag Active Learning sorting technique and system | |
| CN103927394B (en) | A kind of multi-tag Active Learning sorting technique and system based on SVM | |
| CN106227851A (en) | Based on the image search method searched for by depth of seam division that degree of depth convolutional neural networks is end-to-end | |
| CN108734210A (en) | A kind of method for checking object based on cross-module state multi-scale feature fusion | |
| CN106951498A (en) | Text clustering method | |
| CN108510559A (en) | It is a kind of based on have supervision various visual angles discretization multimedia binary-coding method | |
| CN106202256A (en) | Propagate based on semanteme and mix the Web graph of multi-instance learning as search method | |
| CN105005794A (en) | Image pixel semantic annotation method with combination of multi-granularity context information | |
| CN105574545B (en) | The semantic cutting method of street environment image various visual angles and device | |
| CN105938565A (en) | Multi-layer classifier and Internet image aided training-based color image emotion classification method | |
| CN108647595A (en) | Vehicle recognition methods again based on more attribute depth characteristics | |
| CN107563407A (en) | A kind of character representation learning system of the multi-modal big data in network-oriented space | |
| CN110175657A (en) | A kind of image multi-tag labeling method, device, equipment and readable storage medium storing program for executing | |
| CN115830402A (en) | Fine-grained image recognition classification model training method, device and equipment | |
| CN111859909B (en) | Semantic scene consistency recognition reading robot | |
| CN102945372B (en) | Classifying method based on multi-label constraint support vector machine | |
| CN111582506A (en) | Multi-label learning method based on global and local label relation | |
| CN107480194A (en) | The construction method and system of the multi-modal automatic learning model of the representation of knowledge | |
| CN114547307A (en) | Text vector model training method, text matching method, device and equipment | |
| CN106326451A (en) | Method for judging webpage sensing information block based on visual feature extraction | |
| CN106910252B (en) | A kind of online mask method of threedimensional model based on semantic space projective transformation and system |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190712 |
|
| RJ01 | Rejection of invention patent application after publication |