CN109829472A - Semisupervised classification method based on probability neighbour - Google Patents
Semisupervised classification method based on probability neighbour Download PDFInfo
- Publication number
- CN109829472A CN109829472A CN201811598286.2A CN201811598286A CN109829472A CN 109829472 A CN109829472 A CN 109829472A CN 201811598286 A CN201811598286 A CN 201811598286A CN 109829472 A CN109829472 A CN 109829472A
- Authority
- CN
- China
- Prior art keywords
- matrix
- data
- probability
- neighbour
- label
- 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
- 238000000034 method Methods 0.000 title claims abstract description 61
- 239000011159 matrix material Substances 0.000 claims abstract description 138
- 238000012546 transfer Methods 0.000 claims abstract description 25
- 230000001902 propagating effect Effects 0.000 claims description 4
- 238000012804 iterative process Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000001174 ascending effect Effects 0.000 claims description 2
- 230000003190 augmentative effect Effects 0.000 claims description 2
- 230000006854 communication Effects 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000007480 spreading Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This disclosure relates to a kind of semisupervised classification method based on probability neighbour, including prepares data set, data set is pre-processed, probability neighbour's matrix S is constructed, probability transfer matrix P is initialized as to probability neighbour's matrix S, propagate in the enterprising row label of raw data set, update, checking convergence, obtaining classification results.This method solve in the existing semisupervised classification based on figure, the building of similarity graph cannot react the problem of actual conditions, classification inaccuracy well, and the classification method of the disclosure can keep classification results more accurate closer to actual conditions.
Description
Technical field
This disclosure relates to data classification method, in particular, being related to a kind of semisupervised classification method based on probability neighbour.
Background technique
Existing data classification method includes the methods of Supervised classification, semisupervised classification, unsupervised segmentation.Wherein supervise
It needs a large amount of marked data to carry out training pattern in classification method, limits its application scene;Unsupervised segmentation does not need to count
According to classification information, be widely used, but due to lack classification information cause classifying quality bad.It is semi-supervised because only need it is a small amount of
The data of label, procurement cost is low, and can preferably be classified by learning the data distribution of a large amount of Unlabeled data
Effect, thus have a wide range of applications scene.
Semisupervised classification based on figure is an important branch in semisupervised classification, due to taking full advantage of between data
Relationship, often obtain preferable effect, receive extensive attention.In these methods, the relationship between data is embodied in phase
Like in degree figure, and since graph structure and matrix have one-to-one relationship, so similarity graph can be indicated with similarity matrix,
The matrix can be further converted to carry out the probability transfer matrix of label propagation between data simultaneously, and then utilize tape label number
According to classification information carry out label propagation, obtain classification results.
However, similarity graph is often by K- neighbour (KNN) or ε-neighbour in the current semisupervised classification method based on figure
Method construct only used the attributive character of data during structural map, do not use marked data
Classification information, obtained similarity graph cannot react actual conditions well, and classification results are also more inaccurate.
Summary of the invention
In view of the above-mentioned problems, this method solves the present disclosure proposes a kind of semisupervised classification method based on probability neighbour
In the existing semisupervised classification based on figure, the building of similarity graph cannot react actual conditions, classification inaccuracy well
The problem of, the classification method of the disclosure can keep classification results more accurate closer to actual conditions.
Specifically, the present disclosure proposes a kind of semisupervised classification methods based on probability neighbour, comprising:
S100, prepare raw data set, the raw data set includes initial data and its matched label, initial data
Feature described by data attribute, initial data includes marked data and Unlabeled data two parts;
S200, the step S100 data set prepared is pre-processed, obtains the label instruction moment of a vector of marked data
Battle array VL;
S300, the probability neighbour's matrix S for constructing raw data set;
S400, the classification information matrix F and probability transfer matrix P for constructing raw data set, probability transfer matrix P is initial
Turn to probability neighbour's matrix S obtained in the step S300;
S500, based on the probability transfer matrix P after being initialized in step S400, passed in the enterprising row label of raw data set
It broadcasts, obtains new classification information matrix F ';
S600, vector matrix V is indicated using the label of marked dataLNew classification information matrix after propagating label
The marked data matrix F of F 'LIt is updated, prevents the pollution of label information;
New classification information matrix F obtained in S700, checking step S500 ' whether restrain, if new classification information
Matrix F ' it has restrained and has no longer changed, S800 is entered step, otherwise, return step S500;
S800, pass through the new classification information matrix F after convergence ' obtain the classification results of raw data set.
Compared with prior art, the disclosure has following advantageous effects:
(1) classification method of the disclosure is increased in composition and is calculated between two data points using the classification information of data
Similarity, the similarity graph of building can response data well actual conditions, compare existing semisupervised classification method
Improve the accuracy rate of data classification;
(2) classification method of the disclosure defines probability neighbour's matrix, regards composition problem as a probability ask
Topic, keeps classification results more accurate.
Detailed description of the invention
Fig. 1 shows the flow charts of the semisupervised classification method based on probability neighbour of the disclosure;
The value schematic diagram of probability transfer matrix of Fig. 2 expression based on existing KNN method construct;
Fig. 3 indicates the value schematic diagram of the probability transfer matrix of disclosed method construction;
Fig. 4 indicates the standard classified using the classification method of existing KNN method and the disclosure to a variety of different data collection
True rate contrast schematic diagram.
Specific embodiment
Illustrate the detailed process of the semisupervised classification method based on probability neighbour of the disclosure below in conjunction with attached drawing 1.
In one embodiment, a kind of semisupervised classification method based on probability neighbour is provided, comprising:
S100, prepare raw data set, the raw data set includes initial data and its matched label, initial data
Feature described by data attribute, initial data includes marked data and Unlabeled data two parts;
S200, the step S100 data set prepared is pre-processed, obtains the label instruction moment of a vector of marked data
Battle array VL;
S300, the probability neighbour's matrix S for constructing raw data set;
S400, the classification information matrix F and probability transfer matrix P for constructing raw data set, probability transfer matrix P is initial
Turn to probability neighbour's matrix S obtained in the step S300;
S500, based on the probability transfer matrix P after being initialized in step S400, passed in the enterprising row label of raw data set
It broadcasts, obtains new classification information matrix F ';
S600, vector matrix V is indicated using the label of marked dataLNew classification information matrix after propagating label
The marked data matrix F of F 'LIt is updated, prevents the pollution of label information;
New classification information matrix F obtained in S700, checking step S500 ' whether restrain, if new classification information
Matrix F ' it has restrained and has no longer changed, S800 is entered step, otherwise, return step S500;
S800, pass through the new classification information matrix F after convergence ' obtain the classification results of raw data set.
In this embodiment, the pollution of label information described in step S600 refers to, after step S500 processing, marked data
Label may change, cause marked data label occur mistake, referred to as label information pollute.
In this embodiment, be described in detail the disclosure proposition classification method execution step, including prepare data set,
Data set is pre-processed, probability neighbour's matrix S is constructed, probability transfer matrix P is initialized as probability neighbour's matrix S,
The enterprising row label of raw data set propagates, updates, check convergence, obtains classification results.Disclosed method can be pasted more
The actual conditions of nearly data classification, keep classification results more accurate.
In a preferred embodiment, the step S200 is specifically included:
S201, raw data set is defined as to X ∈ Rn*d, wherein Rn*dThe matrix of n row d column is represented, n is data point
Number, d are the attribute numbers of data point;Defining marked data matrix isUnlabeled data matrix isWherein n1For the number of marked data point, n2For the number of Unlabeled data point, n=n1+n2;
S202, vector matrix is indicated according to the label of the marked data of the corresponding label configurations of marked data:
Vector matrix V ∈ R is indicated according to the corresponding label configurations label of the marked data of raw data setn*c, wherein n be
The number of data, c are the numbers of data category, and the i-th row of label instruction vector matrix V is the label instruction of i-th of data
Vector, if the classification of i-th of data point is j, j-th of element of the row is 1, which is 0;By certain ratio
Example concentrates drawing unit score strong point from initial data, and label is indicated vector matrix V with the label of the data point of selection sequence
In correspondence row form matrixLabel as marked data indicates vector matrix.
In this embodiment, it pre-processed, obtain the mark of marked data to how carrying out data set in step S200
Label instruction vector matrix VLIt is illustrated.Selection percentage in the step S202 can according to the actual situation, from original
Beginning data and the data of middle selection 5%-10% are as marked data configuration VL, remaining to be used as Unlabeled data.In this public affairs
It opens in subsequent experiment, is chosen with 10% ratio to be tested.
In a preferred embodiment, the step S300 is specifically included:
The augmented matrix A ∈ R of S301, definition about datan*(d+c), d is data dimension, the number for indicating data attribute, c
Indicate the number of data category, n is the number of data point, A=[X, V];
S302, D ∈ R is definedn*nAs the similarity matrix between data point, number is calculated using Euclidean distance as hygrometer
Similitude between strong point i and jIt enablesIt is arranged as the i-th row jth in matrix D
Specific value is arranged matrix D by row ascending order, xiIndicate the i-th row of raw data set X, xjIndicate the jth of raw data set X
Row;
S303, the probability neighbour's matrix for defining raw data set are S ∈ Rn*n, S is the matrix of n*n, the i-th row in s-matrix
The value S of j columnI, jA possibility that data point i and data point j is as probability neighbour is represented, definition k is neighbour's number, according to K- neighbour
(KNN) method obtains K- neighbour k before choosing matrix D by row, and probability neighbour, tool are then constructed on the basis of K- neighbour
Body are as follows: define p (i, k) becomes the probability of probability neighbour between data point i and its k-th of neighbour, by formulaIt obtains;
Pass through formula againAssignment is carried out, S (i, k) represents i-th of number in s-matrix
Element unassignable in s-matrix is assigned a value of 0 by the value of k-th of neighbour at strong point, probability neighbour's matrix S after obtaining assignment;
WhereinThe specific value that jth row (k+1) arranges in matrix D is represented,It is the specific of the i-th row jth column in matrix D
Numerical value,It is the specific value that the i-th row the 1st arranges in matrix D, n is the number of data point, and X indicates raw data set.
In this embodiment, the specific method to the probability neighbour's matrix S for how constructing raw data set in step S300
Carried out expansion explanation, wherein in composition increase calculated using the classification information of data it is similar between two data points
Degree, and probability neighbour's matrix is defined, regard composition problem as a probability problem to solve, so that classification method is more
Closing to reality situation keeps classification results more accurate.
In this embodiment, good effect can be obtained by the local characteristics of data, it is described can be by step S303
The value of middle k neighbour number is selected as 5~20.
In a preferred embodiment, the step S400 is specifically included:
Construct the classification information matrix F of raw data set: F(0)=[FL, 0], F(0)Represent the F matrix of initialization:
It constructs the probability transfer matrix P of raw data set: probability transfer matrix P is initialized as obtaining in the step S300
Probability neighbour matrix S after the assignment arrived, has: P(0)=S;Wherein, P ∈ Rn*nIt is probability transfer matrix, P(0)Represent initialization
P matrix, PI, jThe numerical value for indicating the i-th row jth column in P matrix indicates that i-th of data point will be with probability PI, jIt is broadcast to j-th of number
Strong point.
In a preferred embodiment, the step S500 is specifically included:
Pass through formula F '(t+1)=P*F '(t)Label propagation is carried out, wherein subscript t indicates the number of iterations, F(t)It indicates the t times
F ' the matrix that iterative process obtains, F '(t+1)Indicate the F ' matrix that the t+1 times iterative process obtains, above formula is indicated from the t times to the
T+1 label communication process, wherein as t=0, F '(t)Indicate original classification information matrix F.
In this embodiment, the detailed process propagated label is illustrated, on the basis of original classification information matrix F
Upper join probability transfer matrix carries out successive ignition, mark to unlabelled data according to the label of marked data
Label are propagated, and are classified to it.
In a preferred embodiment, the step S600 is specifically included:
FL=VL, by marked data F after each iteration label propagationLAgain it is assigned a value of VL, FLIndicate classification information
The marked data matrix of matrix F, VLIndicate the label instruction vector matrix of marked data.
In this embodiment it is that in order to prevent by the way that after the processing of step S500, the label of marked data is contaminated, because
This needs to carry out assignment again.
In a preferred embodiment, in the step S800:
New classification information matrix F ' every data line after convergence indicates the classification of the data, if FijValue is 1, then
The classification of i-th of data point is j, thus from F ' the corresponding classification of each data can be obtained in matrix, obtain raw data set
Classification results.
It is the detailed description to the semisupervised classification method based on probability neighbour of the disclosure above, according to this method logarithm
According to classifying, classification results can be kept more accurate closer to actual conditions.
Experiment:
In order to verify the semisupervised classification method based on probability neighbour of disclosure proposition compared to the existing classification based on KNN
The advantages of method, has carried out experiment and has compared verifying.
In first experiment, using bimonthly type data set, bimonthly type data set is conventionally known data set type,
Data point number can be chosen according to actual needs, and the bimonthly type data point number chosen in this experiment is 400.In this experiment
Choose data set 10% is used as marked data configuration VL, remaining to be used as Unlabeled data.
Fig. 2 embodies the value of the probability transfer matrix based on existing KNN method construct, and Fig. 3 embodies the method for the present invention
The value of the probability transfer matrix of construction.Fig. 2 and Fig. 3 is the probability transfer matrix that two methods generate on same data set
Value, figure midpoint gray scale represents the probability value for becoming neighbours between different nodes, as can be seen that the two methods from result
The ratio that nonzero element accounts in the probability transfer matrix of generation all very littles, therefore corresponding graph structure is all than sparse.No
Be with place: in Fig. 2, the corresponding gray scale of nonzero element is identical, therefore in KNN method, Neighbor Points label having the same
Probability of spreading is not inconsistent with actual conditions;And in Fig. 3, the corresponding gray scale of nonzero element is different, the label probability of spreading of Neighbor Points
Difference more tallies with the actual situation, and expresses the correlation between data.
In second experiment, for existing a variety of different data sets (as shown in Fig. 4 abscissa), using KNN+ label
The method for propagating (KNN+LP) carries out data classification and method PNN+ label of the invention propagates the method progress of (PNN+LP)
Data classification illustrates the accuracy of two methods classification results on different data sets with Fig. 4 comparison, can be with from Fig. 4
It was found that being better than the prior art using the classification results that the classification method (PNN+LP) of the disclosure obtains, the accuracy rate of classification is more
It is high.
To sum up, the method that the disclosure proposes solves in the existing semisupervised classification based on figure, the building of similarity graph
The problem of actual conditions, classification inaccuracy cannot be reacted well, and compared to existing technologies, the classification method of the disclosure can be more
Closing to reality situation keeps classification results more accurate.
Although embodiment of the present invention is described in conjunction with attached drawing above, the invention is not limited to above-mentioned
Specific embodiments and applications field, above-mentioned specific embodiment are only schematical, directiveness, rather than restricted
's.Those skilled in the art are under the enlightenment of this specification and in the range for not departing from the claims in the present invention and being protected
In the case where, a variety of forms can also be made, these belong to the column of protection of the invention.
Claims (8)
1. a kind of semisupervised classification method based on probability neighbour, comprising:
S100, prepare raw data set, the raw data set includes initial data and its matched label, the spy of initial data
Sign is described by data attribute, and initial data includes marked data and Unlabeled data two parts;
S200, the step S100 data set prepared is pre-processed, obtains the label instruction vector matrix V of marked dataL;
S300, the probability neighbour's matrix S for constructing raw data set;
S400, the classification information matrix F and probability transfer matrix P for constructing raw data set, probability transfer matrix P is initialized as
Probability neighbour's matrix S obtained in the step S300;
S500, it is obtained based on the probability transfer matrix P after being initialized in step S400 in the enterprising row label propagation of raw data set
To new classification information matrix F ';
S600, vector matrix V is indicated using the label of marked dataLNew classification information matrix F after propagating label '
Marked data matrix FLIt is updated, prevents the pollution of label information;
New classification information matrix F obtained in S700, checking step S500 ' whether restrain, if new classification information matrix
F ', which has restrained, no longer to be changed, and S800 is entered step, otherwise, return step S500;
S800, pass through the new classification information matrix F after convergence ' obtain the classification results of raw data set.
2. the step S200 is specifically included according to the method described in claim 1, preferred:
S201, raw data set is defined as to X ∈ Rn*d, wherein Rn*dThe matrix of n row d column is represented, n is of data point
Number, d is the attribute number of data point;Defining marked data matrix isUnlabeled data matrix isWherein n1For the number of marked data point, n2For the number of Unlabeled data point, n=n1+n2;
S202, vector matrix is indicated according to the label of the marked data of the corresponding label configurations of marked data:
Vector matrix V ∈ R is indicated according to the corresponding label configurations label of the marked data of raw data setn*c, wherein n is data
Number, c is the number of data category, the i-th row of label instruction vector matrix V be the label of i-th of data point indicate to
Amount, if the classification of i-th of data point is j, j-th of element of the row is 1, which is 0;According to original number
The sequence of marked data in, selection label indicate that the correspondence row in vector V constitutes matrixAs having marked
The label for the evidence that counts indicates vector matrix.
3. according to the method described in claim 2, the step S300 is specifically included:
The augmented matrix A ∈ R of S301, definition about datan*(d+c), d is data dimension, the number for indicating data attribute, and c is indicated
The number of data category, n are the number of data point, A=[X, V];
S302, D ∈ R is definedn*nAs the similarity matrix between data point, data point i is calculated using Euclidean distance as hygrometer
Similitude between jIt enablesIt is arranged as the i-th row jth in matrix D specific
Numerical value is arranged matrix D by row ascending order, xiIndicate the i-th row of raw data set X, xjIndicate the jth row of raw data set X;
S303, the probability neighbour's matrix for defining raw data set are S ∈ Rn*n, S is the matrix of n*n, and the i-th row jth arranges in s-matrix
Value Si,jA possibility that data point i and data point j is as probability neighbour is represented, definition k is neighbour's number, according to K- neighbour
(KNN) method obtains K- neighbour k before choosing matrix D by row, and probability neighbour, tool are then constructed on the basis of K- neighbour
Body are as follows: define p (i, k) becomes the probability of probability neighbour between data point i and its k-th of neighbour, by formulaIt obtains;
Pass through formula againAssignment is carried out, S (i, k) represents i-th of data point in s-matrix
K-th of neighbour value, element unassignable in s-matrix is assigned a value of 0, probability neighbour's matrix S after obtaining assignment;WhereinThe specific value that the i-th row (k+1) arranges in matrix D is represented,It is the specific number that the i-th row jth arranges in matrix D
Value,It is the specific value that the i-th row the 1st arranges in matrix D, n is the number of data point, and X indicates raw data set.
4. according to the method described in claim 3, the step S400 is specifically included:
Construct the classification information matrix F of raw data set: F(0)=[FL, 0], F(0)Represent the F matrix of initialization:
It constructs the probability transfer matrix P of raw data set: probability transfer matrix P is initialized as obtained in the step S300
Probability neighbour matrix S after assignment, has: P(0)=S;Wherein, P ∈ Rn*nIt is probability transfer matrix, P(0)Represent the P square of initialization
Battle array, PI, jThe numerical value for indicating the i-th row jth column in P matrix indicates that i-th of data point will be with probability PI, jIt is broadcast to j-th of data
Point.
5. according to the method described in claim 4, the step S500 is specifically included:
Pass through formula F '(t+1)=P*F '(t)Label propagation is carried out, wherein subscript t indicates the number of iterations, F(t)Indicate the t times iteration
F ' the matrix that process obtains, F '(t+1)Indicate the F ' matrix that the t+1 times iterative process obtains, above formula is indicated from the t times to t+1
Secondary label communication process, wherein as t=0, F '(t)Indicate original classification information matrix F.
6. according to the method described in claim 5, the step S600 is specifically included:
FL=VL, by marked data F after each iteration label propagationLAgain it is assigned a value of VL, FLIndicate classification information matrix F
Marked data matrix, VLIndicate the label instruction vector matrix of marked data.
7. according to the method described in claim 6, in the step S800:
New classification information matrix F ' every data line after convergence indicates the classification of the data, if FijValue is 1, then i-th
The classification of a data point is j, thus from F ' the corresponding classification of each data can be obtained in matrix, obtain the classification of raw data set
As a result.
8. according to the method described in claim 3, the value of k neighbour's number is selected as 5~20 in the step S303.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811598286.2A CN109829472B (en) | 2018-12-24 | 2018-12-24 | Semi-supervised classification method based on probability nearest neighbor |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811598286.2A CN109829472B (en) | 2018-12-24 | 2018-12-24 | Semi-supervised classification method based on probability nearest neighbor |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109829472A true CN109829472A (en) | 2019-05-31 |
CN109829472B CN109829472B (en) | 2024-05-14 |
Family
ID=66861165
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811598286.2A Active CN109829472B (en) | 2018-12-24 | 2018-12-24 | Semi-supervised classification method based on probability nearest neighbor |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109829472B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046914A (en) * | 2019-11-20 | 2020-04-21 | 陕西师范大学 | Semi-supervised classification method based on dynamic composition |
CN111488923A (en) * | 2020-04-03 | 2020-08-04 | 陕西师范大学 | Enhanced anchor point image semi-supervised classification method |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080147574A1 (en) * | 2006-12-14 | 2008-06-19 | Xerox Corporation | Active learning methods for evolving a classifier |
CN104463202A (en) * | 2014-11-28 | 2015-03-25 | 苏州大学 | Multi-class image semi-supervised classifying method and system |
CN104794489A (en) * | 2015-04-23 | 2015-07-22 | 苏州大学 | Deep label prediction based inducing type image classification method and system |
CN105608690A (en) * | 2015-12-05 | 2016-05-25 | 陕西师范大学 | Graph theory and semi supervised learning combination-based image segmentation method |
CN105608471A (en) * | 2015-12-28 | 2016-05-25 | 苏州大学 | Robust transductive label estimation and data classification method and system |
CN105931253A (en) * | 2016-05-16 | 2016-09-07 | 陕西师范大学 | Image segmentation method combined with semi-supervised learning |
CN106596900A (en) * | 2016-12-13 | 2017-04-26 | 贵州电网有限责任公司电力科学研究院 | Transformer fault diagnosis method based on improved semi-supervised classification of graph |
CN108009571A (en) * | 2017-11-16 | 2018-05-08 | 苏州大学 | A kind of semi-supervised data classification method of new direct-push and system |
-
2018
- 2018-12-24 CN CN201811598286.2A patent/CN109829472B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080147574A1 (en) * | 2006-12-14 | 2008-06-19 | Xerox Corporation | Active learning methods for evolving a classifier |
CN104463202A (en) * | 2014-11-28 | 2015-03-25 | 苏州大学 | Multi-class image semi-supervised classifying method and system |
CN104794489A (en) * | 2015-04-23 | 2015-07-22 | 苏州大学 | Deep label prediction based inducing type image classification method and system |
CN105608690A (en) * | 2015-12-05 | 2016-05-25 | 陕西师范大学 | Graph theory and semi supervised learning combination-based image segmentation method |
CN105608471A (en) * | 2015-12-28 | 2016-05-25 | 苏州大学 | Robust transductive label estimation and data classification method and system |
CN105931253A (en) * | 2016-05-16 | 2016-09-07 | 陕西师范大学 | Image segmentation method combined with semi-supervised learning |
CN106596900A (en) * | 2016-12-13 | 2017-04-26 | 贵州电网有限责任公司电力科学研究院 | Transformer fault diagnosis method based on improved semi-supervised classification of graph |
CN108009571A (en) * | 2017-11-16 | 2018-05-08 | 苏州大学 | A kind of semi-supervised data classification method of new direct-push and system |
Non-Patent Citations (4)
Title |
---|
刘桂锋;汪满容;刘海军: "基于概率超图半监督学习的专利文本分类方法研究", 情报杂志, no. 009, 31 December 2016 (2016-12-31) * |
王小攀;马丽;刘福江;: "一种基于线性邻域传播的加权K近邻算法", 计算机工程, no. 07, 15 July 2013 (2013-07-15) * |
田勋;汪西莉: "自适应近邻的极小极大标签传播", 小型微型计算机系统, no. 011, 31 December 2017 (2017-12-31) * |
黎隽男;吕佳;: "基于近邻密度和半监督KNN的集成自训练方法", 计算机工程与应用, no. 20, 23 November 2017 (2017-11-23) * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046914A (en) * | 2019-11-20 | 2020-04-21 | 陕西师范大学 | Semi-supervised classification method based on dynamic composition |
CN111046914B (en) * | 2019-11-20 | 2023-10-27 | 陕西师范大学 | Semi-supervised classification method based on dynamic composition |
CN111488923A (en) * | 2020-04-03 | 2020-08-04 | 陕西师范大学 | Enhanced anchor point image semi-supervised classification method |
CN111488923B (en) * | 2020-04-03 | 2023-02-07 | 陕西师范大学 | Enhanced anchor point image semi-supervised classification method |
Also Published As
Publication number | Publication date |
---|---|
CN109829472B (en) | 2024-05-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110472065B (en) | Cross-language knowledge graph entity alignment method based on GCN twin network | |
CN107132516B (en) | A kind of Radar range profile's target identification method based on depth confidence network | |
CN110135459B (en) | Zero sample classification method based on double-triple depth measurement learning network | |
CN110443367A (en) | A kind of method of strength neural network model robust performance | |
CN105701502A (en) | Image automatic marking method based on Monte Carlo data balance | |
CN110132263A (en) | A kind of method for recognising star map based on expression study | |
CN112487805B (en) | Small sample Web service classification method based on meta-learning framework | |
CN103778441B (en) | A kind of sequence Aircraft Target Recognition based on DSmT and HMM | |
CN105701509B (en) | A kind of image classification method based on across classification migration Active Learning | |
CN110232445B (en) | Cultural relic authenticity identification method based on knowledge distillation | |
CN110082738B (en) | Radar target identification method based on Gaussian mixture and tensor recurrent neural network | |
CN109829472A (en) | Semisupervised classification method based on probability neighbour | |
CN108898244B (en) | Digital signage position recommendation method coupled with multi-source elements | |
CN102867192B (en) | A kind of Scene Semantics moving method propagated based on supervision geodesic line | |
CN104598920A (en) | Scene classification method based on Gist characteristics and extreme learning machine | |
CN113469119B (en) | Cervical cell image classification method based on visual converter and image convolution network | |
CN109190053A (en) | One kind being based on point of interest importance and the authoritative point of interest recommended method of user | |
CN112685504A (en) | Production process-oriented distributed migration chart learning method | |
CN107704531A (en) | The automatic merging method of road network intersection node in a kind of OSM map datums | |
CN111832637A (en) | Distributed deep learning classification method based on alternative direction multiplier method ADMM | |
CN106021402A (en) | Multi-modal multi-class Boosting frame construction method and device for cross-modal retrieval | |
CN109165675A (en) | Image classification method based on periodically part connection convolutional neural networks | |
CN109815986A (en) | The semisupervised classification method of fusion part and global characteristics | |
CN105701510B (en) | A kind of zero lens image classification method based on Share Model space learning | |
CN112529415A (en) | Article scoring method based on combined multi-receptive-field-map neural network |
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 |