CN111046926A - Computer vision image classification integrated learning method - Google Patents
Computer vision image classification integrated learning method Download PDFInfo
- Publication number
- CN111046926A CN111046926A CN201911173435.5A CN201911173435A CN111046926A CN 111046926 A CN111046926 A CN 111046926A CN 201911173435 A CN201911173435 A CN 201911173435A CN 111046926 A CN111046926 A CN 111046926A
- Authority
- CN
- China
- Prior art keywords
- decision
- cloud
- cloud model
- concept
- image classification
- 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 24
- 230000006870 function Effects 0.000 claims abstract description 31
- 230000000007 visual effect Effects 0.000 claims abstract description 11
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 7
- 230000010354 integration Effects 0.000 claims abstract description 7
- 238000011156 evaluation Methods 0.000 claims description 29
- 238000007636 ensemble learning method Methods 0.000 claims description 15
- 239000013256 coordination polymer Substances 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 230000003179 granulation Effects 0.000 claims description 6
- 238000005469 granulation Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000019771 cognition Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 239000002245 particle Substances 0.000 claims description 4
- 238000004513 sizing Methods 0.000 claims description 3
- 238000001308 synthesis method Methods 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 238000007635 classification algorithm Methods 0.000 abstract description 2
- 238000005259 measurement Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention particularly relates to a computer vision image classification and integration learning method. The computer visual image classification integrated learning method is characterized in that a group decision model is utilized to determine a loss function on the basis of three decisions of a decision rough set model, and a cloud comprehensive method based on a density center is used for converting a plurality of classification information into a cloud model; the decision rule can be formed only by manually setting the weights of different base classifiers, determining the loss function by using the cloud model and calculating the threshold value by the algorithm. According to the computer visual image classification integrated learning method, the three-branch decision model adopting the cloud comprehensive method is implemented in the computer visual image classification algorithm, when the class of the image is judged, the efficiency and the identification accuracy of the strong classifier are obviously improved, the classification accuracy is improved, and the operation efficiency of a computer on the image identification problem can be greatly improved through distributed type inheritance operation.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a computer vision image classification and integration learning method.
Background
Computer vision is a simulation of biological vision using a computer and associated equipment. Its main task is to obtain three-dimensional information of a corresponding scene by processing captured pictures or videos, as is done every day by humans and many other living beings.
Image classification refers to an image processing method for distinguishing different types of objects from each other according to different characteristics reflected in image information. It uses computer to make quantitative analysis of image, and classifies each picture element or region in the image into one of several categories to replace human visual interpretation.
The three-branch decision is a decision mode based on human cognition, and divides the result into three decision schemes of acceptance, delay and rejection. The three-branch decision theory can be used in the field of computer vision and can be used for processing the problems of uncertainty information and cost sensitivity.
For example, in the image classification problem, an imbalance problem that the number of positive class samples is small and the number of negative class samples is large tends to occur. The determination of the loss function is a key step in the three-branch decision making process, and in the prior art, the most direct and effective method is generally considered to be that the loss function is directly given by a domain expert, but the uncertainty of expert evaluation makes the loss function difficult to quantify, and the expert evaluation is not necessarily relatively objective or persuasive. Particularly in the field of computer vision, a common three-branch decision model cannot be adopted for image classification, and machine learning cannot accept manual intervention after the model is formed, so that a loss function cannot be given by expert judgment.
Aiming at the problems, the invention provides a computer vision image classification and integration learning method.
Disclosure of Invention
In order to make up for the defects of the prior art, the invention provides a simple and efficient computer vision image classification and ensemble learning method.
The invention is realized by the following technical scheme:
a computer vision image classification integrated learning method is characterized in that: determining a loss function by utilizing a group decision model on the basis of three decisions of a decision rough set model, and converting a plurality of classification information into a cloud model by minimizing the weighted distance between the comprehensive evaluation and the basic evaluation in the classification decision and using a density center-based cloud comprehensive method; the decision rule can be formed only by manually setting the weights of different base classifiers, determining the loss function by using the cloud model, providing a new semantic explanation for the loss function, obtaining the most accurate comprehensive evaluation in the computer visual classification group decision, and calculating the threshold value by the algorithm.
The computer vision image classification and ensemble learning method integrates the image classification and ensemble learning method into a computer vision algorithm package, and comprises the following steps of:
firstly, determining a state set and a decision action set, and constructing a loss function matrix;
wherein, the state set
Representing that the object belongs to set C or does not belong to set C; decision Action set Action ═ aP,aB,aNThe decision actions for classifying the target object x into a positive domain pos (c), a boundary domain bnd (c) and a negative domain neg (c), respectively;
from the point of view of bayesian decision making different decisions for different states will result in a corresponding decision cost of 6, where λPP，λBP，λNPRespectively representing object acquisitions a in a target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions; lambda [ alpha ]PN，λBNAnd λNNRespectively representing object collections a other than the target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions;
step two, manually setting an image classification base classifier and the weight thereof;
the set of base classifiers is E ═ E1,e2,…,etW ═ W in weight set1,w2,…,wt}TWhere t is the number of base classifiers in the set of base classifiers;
thirdly, converting uncertainty evaluation made by loss functions given by all experts into a cloud model, and marking as:
λk PP(Exk PP,Enk PP,Hek PP)，λk BP(Exk BP,Enk BP,Hek BP)，λk NP(Exk NP,Enk NP,Hek NP)，λk PN(Exk PN,Enk PN,Hek PN)，λk BN(Exk BN,Enk BN,Hek BN)，λk NN(Exk NN,Enk NN,Hek NN) Wherein k is more than or equal to 1 and less than or equal to t, Ex, En and He are three digital characteristics of the cloud model; where Ex (expectation) is the most representative or most typical sample in the concept; en (entropy) qualitative concepts an uncertainty measure; he (super entropy) describes the uncertainty of entropy (En).
Fourthly, in order to reduce the inconsistency of the results of all classifiers, synthesizing the loss functions represented by the cloud model to obtain a comprehensive evaluation cloud model of the loss functions;
fifthly, considering that the expectation of the cloud model reflects the most typical concept extension, substituting the expectation of the comprehensive evaluation cloud model of the loss function into the expectation of the comprehensive evaluation cloud model to calculate each threshold α, gamma of the three decisions;
and sixthly, forming a classification decision rule and giving a judgment standard for judging whether the region image belongs to the class.
In the first step, 6 decision costs or losses satisfy the following inequality:
λPP≤λBP<λNP(1)
λNN≤λBN<λPN(2)
inequality (1) indicates that the penalty for making an acceptance decision for an object x belonging to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making a rejection for it;
inequality (2) indicates that the penalty for making a reject decision for an object x that does not belong to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making an accept decision for it.
In the fourth step, the advantage of the cloud model in the original data granulation is shown in the expression of the sizing concept, and two uncertainties, namely random uncertainty and fuzzy uncertainty, are kept; from the perspective of particle cognition calculation, the group decision is a granulation means, and a plurality of discrete basic evaluations are converted into a comprehensive evaluation;
the cloud integration is a process of integrating a plurality of fine-grained cloud models into a coarse-grained cloud model based on concept connotation, wherein the cloud concept of a low-grained layer is called basic cloud, and the integrated cloud concept is called integrated cloud;
in group decision, the cloud synthesis method is defined as a linear combination form of concept connotation.
In the fourth step, let U be a quantitative theory domain represented by an accurate numerical value, C1And C2Are two basic cloud concepts on the U; if normally distributed N (mu, sigma)2),N1(μ1,σ2 1) And N2(μ2,σ2 2) Are respectively C1And C2The expected, outer and inner envelope curves of (a) correspond to normal distributions with respect to g (N) ═ KL (N | | N)1)+KL(N||N2) Is the optimal solution of C1And C2A comprehensive cloud model M (Ex, En, He) formed based on the symmetrical KL divergence satisfies the following conditions:
in the fourth step, let U be a quantitative discourse field expressed by an accurate numerical value, and C ═ Ci(Exi,Eni,Hei) I 1 … n is n basic cloud concepts on U, and its corresponding weight set Ω ═ { ω }i1 … n }; if N (μ, σ)2) Is the normal distribution of the expected curve corresponding to the cloud model cluster C
Of (2) an optimal solution, Nouter(μouter,σ2 outer) And Ninner(μinner,σ2 inner) Is cloud model cluster C about
s.t. θi∈[-1，]
s.t. θi∈[-1，1]
The cloud model cluster C gives the comprehensive cloud model M (Ex, En, He) formed by symmetrical KL divergences to satisfy:
in the fifth step, the threshold values α and γ of the three decisions are respectively as follows:
in the sixth step, a conditional probability Pr (Cx) is defined as the probability of dividing the object x into a target probability C, where x represents the equivalence class to which the object x belongs; the classification decision rule is as follows:
(P') if Pr (C | [ x ]) is not less than α, the sample belongs to the judgment category, and the sample is a positive sample, namely x belongs to POS (C);
(B') if Pr (Cxx) is less than or equal to α and Pr (Cxx) is greater than or equal to β, making a delay decision to classify the image with a class that is temporarily unrecognizable, i.e., x belongs to BND (C);
(N') if Pr (C | [ x ]) is less than or equal to β, then it is a negative sample (not belonging to the decision category), i.e., x ∈ NEG (C).
The invention has the beneficial effects that: according to the computer visual image classification integrated learning method, the three-branch decision model adopting the cloud comprehensive method is implemented in the computer visual image classification algorithm, when the class of the image is judged, the efficiency and the identification accuracy of the strong classifier are obviously improved, the classification accuracy is improved, and the operation efficiency of a computer on the image identification problem can be greatly improved through distributed type inheritance operation.
Drawings
FIG. 1 is a schematic diagram of a computer vision image classification and ensemble learning method of the present invention.
Detailed Description
In order to make those skilled in the art better understand the technical solution of the present invention, the technical solution in the embodiment of the present invention will be clearly and completely described below with reference to the embodiment of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The computer visual image classification integrated learning method is characterized in that a group decision model is utilized to determine a loss function on the basis of three decisions of a decision rough set model, and a plurality of classification information is converted into a cloud model by minimizing the weighted distance between comprehensive evaluation and basic evaluation in classification decision and using a density center-based cloud comprehensive method; the decision rule can be formed only by manually setting the weights of different base classifiers, determining the loss function by using the cloud model, providing a new semantic explanation for the loss function, obtaining the most accurate comprehensive evaluation in the computer visual classification group decision, and calculating the threshold value by the algorithm.
The computer vision image classification and ensemble learning method integrates the image classification and ensemble learning method into a computer vision algorithm package, and comprises the following steps of:
firstly, determining a state set and a decision action set, and constructing a loss function matrix;
wherein, the state set
Representing that the object belongs to set C or does not belong to set C; decision Action set Action ═ aP,aB,aNThe decision actions for classifying the target object x into a positive domain pos (c), a boundary domain bnd (c) and a negative domain neg (c), respectively;
from the point of view of bayesian decision, making different decisions for different states brings about the corresponding decision cost in 6, as shown in table 1:
TABLE 1 loss function matrix
λPP，λBP，λNPRespectively representing object acquisitions a in a target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions; lambda [ alpha ]PN，λBNAnd λNNRespectively representing object collections a other than the target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions;
the 6 decision costs or losses satisfy the following inequality:
λPP≤λBP<λNP(1)
λNN≤λBN<λPN(2)
inequality (1) indicates that the penalty for making an acceptance decision for an object x belonging to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making a rejection for it;
inequality (2) indicates that the penalty for making a reject decision for an object x that does not belong to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making an accept decision for it.
Step two, manually setting an image classification base classifier and the weight thereof;
the set of base classifiers is E ═ E1,e2,…,etW ═ W in weight set1,w2,…,wt}TWhere t is the number of base classifiers in the set of base classifiers;
thirdly, converting uncertainty evaluation made by loss functions given by all experts into a cloud model, and marking as:
λk PP(Exk PP,Enk PP,Hek PP)，λk BP(Exk BP,Enk BP,Hek BP)，λk NP(Exk NP,Enk NP,Hek NP)，λk PN(Exk PN,Enk PN,Hek PN)，λk BN(Exk BN,Enk BN,Hek BN)，λk NN(Exk NN,Enk NN,Hek NN) Wherein k is more than or equal to 1 and less than or equal to t;
ex, En and He are three digital characteristics of the cloud model; where Ex (expectation) is the most representative or most typical sample in the concept; en (entropy) qualitative concepts an uncertainty measure; he (super entropy) describes the uncertainty of entropy (En).
Fourthly, in order to reduce the inconsistency of the results of all classifiers, synthesizing the loss functions represented by the cloud model to obtain a comprehensive evaluation cloud model of the loss functions;
the advantages of the cloud model on the original data granulation are shown in the expression of the sizing concept, and meanwhile, two uncertainties, namely random uncertainty and fuzzy uncertainty are reserved; from the perspective of particle cognition calculation, the group decision is a granulation means, and a plurality of discrete basic evaluations are converted into a comprehensive evaluation; the cloud integration is a process of integrating a plurality of fine-grained cloud models into a coarse-grained cloud model based on concept connotation, wherein the cloud concept of a low-grained layer is called basic cloud, and the integrated cloud concept is called integrated cloud; in group decision, the cloud synthesis method is defined as a linear combination form of concept connotation.
Let U be a quantitative theory domain represented by an exact numerical value, C1And C2Are two basic cloud concepts on the U; if normally distributed N (mu, sigma)2),N1(μ1,σ2 1) And N2(μ2,σ2 2) Are respectively C1And C2The expected, outer and inner envelope curves of (a) correspond to normal distributions with respect to g (N) ═ KL (N | | N)1)+KL(N||N2) Is the optimal solution of C1And C2A comprehensive cloud model M (Ex, En, He) formed based on the symmetrical KL divergence satisfies the following conditions:
let U be the quantitative discourse field expressed as an exact numerical value, C ═ Ci(Exi,Eni,Hei) I 1 … n is n basic cloud concepts on U, and its corresponding weight set Ω ═ { ω }i1 … n }; if N (μ, σ)2) Is the normal distribution of the expected curve corresponding to the cloud model cluster C
Of (2) an optimal solution, Nouter(μouter,σ2 outer) And Ninner(μinner,σ2 inner) Is cloud model cluster C about
The optimal solution is that the cloud model cluster C is given symmetrical KL powderAnd the formed comprehensive cloud model M (Ex, En, He) satisfies the following conditions:
and fifthly, considering that the expectation of the cloud model reflects the most typical concept extension, substituting the expectation of the comprehensive evaluation cloud model of the loss function into the expectation of the cloud model to obtain the threshold values α and gamma of each item of the three decisions, wherein the calculation formulas are as follows:
sixthly, forming a classification decision rule and giving a judgment standard for judging whether the region image belongs to the class;
defining a conditional probability Pr (Cx) as the probability of dividing the object x into a target probability C, where x represents the equivalence class to which the object x belongs; the loss caused by the different decision actions made by the belonging equivalence class [ x ] on the object x is represented as follows:
make a least lossy decision on object x:
R(aP|[x])≤R(aB|[x]) And R (a)P|[x])≤R(aN|[x]) Then make aAccepting a decision, i.e. x ∈ POS (C);
R(aB|[x])≤R(aP|[x]) And R (a)B|[x])≤R(aN|[x]) Then a delay decision is made, i.e. x ∈ bnd (c);
R(aN|[x])≤R(aP|[x]) And R (a)N|[x])≤R(aB|[x]) A rejection decision is made, i.e. x e neg (c).
Due to the fact that
The inequality can be converted into:
if Pr (Cx) is equal to or more than α and Pr (Cx) is equal to or more than gamma, making an acceptance decision, i.e. x belongs to POS (C);
if Pr (Cx) is less than or equal to α and Pr (Cx) is greater than or equal to β, then a delay decision is made, i.e., x belongs to BND (C);
if Pr (Cx) is less than or equal to β and Pr (Cx) is less than or equal to γ, then a rejection decision is made, i.e., x ∈ NEG (C).
The classification decision rule is as follows:
(P') if Pr (C | [ x ]) is not less than α, the sample belongs to the judgment category, and the sample is a positive sample, namely x belongs to POS (C);
(B') if Pr (Cxx) is less than or equal to α and Pr (Cxx) is greater than or equal to β, making a delay decision to classify the image with a class that is temporarily unrecognizable, i.e., x belongs to BND (C);
(N') if Pr (C | [ x ]) is less than or equal to β, then it is a negative sample (not belonging to the decision category), i.e., x ∈ NEG (C).
The computer visual image classification and integration learning method gives out semantic interpretation of a comprehensive cloud method and geometric interpretation of a drift measurement space of a concept: in the conceptual drift measurement space, the weighted sum of the distances to each basic evaluation is evaluated comprehensively, and if the weight is regarded as the mass of the particle, the comprehensive cloud model is the density center of all basic cloud models.
The above description describes a computer vision image classification and ensemble learning method in the embodiment of the present invention in detail. While the present invention has been described with reference to specific examples, which are provided to assist in understanding the core concepts of the present invention, it is intended that all other embodiments that can be obtained by those skilled in the art without departing from the spirit of the present invention shall fall within the scope of the present invention.
Claims (8)
1. A computer vision image classification integrated learning method is characterized in that: determining a loss function by utilizing a group decision model on the basis of three decisions of a decision rough set model, and converting a plurality of classification information into a cloud model by minimizing the weighted distance between the comprehensive evaluation and the basic evaluation in the classification decision and using a density center-based cloud comprehensive method; the decision rule can be formed only by manually setting the weights of different base classifiers, determining the loss function by using the cloud model, providing a new semantic explanation for the loss function, obtaining the most accurate comprehensive evaluation in the computer visual classification group decision, and calculating the threshold value by the algorithm.
2. The computer vision image classification ensemble learning method of claim 1, wherein integrating the image classification ensemble learning method into a computer vision algorithm package, comprises the steps of:
firstly, determining a state set and a decision action set, and constructing a loss function matrix;
wherein, the state set
Representing that the object belongs to set C or does not belong to set C; decision Action set Action ═ aP,aB,aNThe decision actions for classifying the target object x into a positive domain pos (c), a boundary domain bnd (c) and a negative domain neg (c), respectively;
from the point of view of bayesian decision making different decisions for different states will result in a corresponding decision cost of 6, where λPP，λBP，λNPRespectively representing object acquisitions a in a target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions; lambda [ alpha ]PN，λBNAnd λNNRespectively representing object collections a other than the target concept CP，aB，aNThe decision cost or loss resulting from the three decision actions;
step two, manually setting an image classification base classifier and the weight thereof;
the set of base classifiers is E ═ E1,e2,…,etW ═ W in weight set1,w2,…,wt}TWhere t is the number of base classifiers in the set of base classifiers;
thirdly, converting uncertainty evaluation made by loss functions given by all experts into a cloud model, and marking as:
λk PP(Exk PP,Enk PP,Hek PP)，λk BP(Exk BP,Enk BP,Hek BP)，λk NP(Exk NP,Enk NP,Hek NP)，λk PN(Exk PN,Enk PN,Hek PN)，λk BN(Exk BN,Enk BN,Hek BN)，λk NN(Exk NN,Enk NN,Hek NN) Wherein k is more than or equal to 1 and less than or equal to t, Ex, En and He are three digital characteristics of the cloud model; where Ex is the most representative or most typical sample in the concept; en characterizes a concept of uncertainty measure; he describes the uncertainty in entropy (En);
fourthly, in order to reduce the inconsistency of the results of all classifiers, synthesizing the loss functions represented by the cloud model to obtain a comprehensive evaluation cloud model of the loss functions;
fifthly, considering that the expectation of the cloud model reflects the most typical concept extension, substituting the expectation of the comprehensive evaluation cloud model of the loss function into the expectation of the comprehensive evaluation cloud model to calculate each threshold α, gamma of the three decisions;
and sixthly, forming a classification decision rule and giving a judgment standard for judging whether the region image belongs to the class.
3. The computer vision image classification ensemble learning method of claim 2, wherein: in the first step, 6 decision costs or losses satisfy the following inequality:
λPP≤λBP<λNP(1)
λNN≤λBN<λPN(2)
inequality (1) indicates that the penalty for making an acceptance decision for an object x belonging to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making a rejection for it;
inequality (2) indicates that the penalty for making a reject decision for an object x that does not belong to the target concept C is less than or equal to the penalty for making a delay decision for it, and both of these penalties are less than the penalty for making an accept decision for it.
4. The computer vision image classification ensemble learning method of claim 3, wherein: in the fourth step, the advantage of the cloud model in the original data granulation is shown in the expression of the sizing concept, and two uncertainties, namely random uncertainty and fuzzy uncertainty, are kept; from the perspective of particle cognition calculation, the group decision is a granulation means, and a plurality of discrete basic evaluations are converted into a comprehensive evaluation;
the cloud integration is a process of integrating a plurality of fine-grained cloud models into a coarse-grained cloud model based on concept connotation, wherein the cloud concept of a low-grained layer is called basic cloud, and the integrated cloud concept is called integrated cloud;
in group decision, the cloud synthesis method is defined as a linear combination form of concept connotation.
5. The computer vision image classification ensemble learning method of claim 4, wherein: in the fourth step, let U be a quantitative theory domain represented by an accurate numerical value, C1And C2Is two bases on UA cloud concept; if normally distributed N (mu, sigma)2)，N1(μ1,σ2 1) And N2(μ2,σ2 2) Are respectively C1And C2The expected, outer and inner envelope curves of (a) correspond to normal distributions with respect to g (N) ═ KL (N | | N)1)+KL(N||N2) Is the optimal solution of C1And C2A comprehensive cloud model M (Ex, En, He) formed based on the symmetrical KL divergence satisfies the following conditions:
6. the computer vision image classification ensemble learning method of claim 4, wherein: in the fourth step, let U be a quantitative discourse field expressed by an accurate numerical value, and C ═ Ci(Exi,Eni,Hei) I 1 … n is n basic cloud concepts on U, and its corresponding weight set Ω ═ { ω }i1 … n }; if N (μ, σ)2) Is the normal distribution of the expected curve corresponding to the cloud model cluster C
Of (2) an optimal solution, Nouter(μouter,σ2 outer) And Ninner(μinner,σ2 inner) Is cloud model cluster C about
The cloud model cluster C gives the comprehensive cloud model M (Ex, En, He) formed by symmetrical KL divergences to satisfy:
7. the computer vision image classification ensemble learning method of claim 5 or 6, wherein in the fifth step, the thresholds α and γ of the three-branch decision are respectively as follows:
8. the computer vision image classification ensemble learning method of claim 2, wherein: in the sixth step, a conditional probability Pr (Cx) is defined as the probability of dividing the object x into a target probability C, where x represents the equivalence class to which the object x belongs; the classification decision rule is as follows:
(P') if Pr (C | [ x ]) is not less than α, the sample belongs to the judgment category, and the sample is a positive sample, namely x belongs to POS (C);
(B') if Pr (Cxx) is less than or equal to α and Pr (Cxx) is greater than or equal to β, making a delay decision to classify the image with a class that is temporarily unrecognizable, i.e., x belongs to BND (C);
(N') if Pr (C | [ x ]) is less than or equal to β, it does not belong to the decision category, and it is a negative sample, i.e., x ∈ NEG (C).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911173435.5A CN111046926B (en) | 2019-11-26 | 2019-11-26 | Computer vision image classification integrated learning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911173435.5A CN111046926B (en) | 2019-11-26 | 2019-11-26 | Computer vision image classification integrated learning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111046926A true CN111046926A (en) | 2020-04-21 |
| CN111046926B CN111046926B (en) | 2023-09-19 |
Family
ID=70233494
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911173435.5A Active CN111046926B (en) | 2019-11-26 | 2019-11-26 | Computer vision image classification integrated learning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111046926B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111768381A (en) * | 2020-06-29 | 2020-10-13 | 北京百度网讯科技有限公司 | Part defect detection method and device and electronic equipment |
| CN112668871A (en) * | 2020-12-25 | 2021-04-16 | 中国人民解放军63892部队 | Dynamic assignment method for expert weight in multi-round group decision |
| CN112766415A (en) * | 2021-02-09 | 2021-05-07 | 第四范式（北京）技术有限公司 | Method, device and system for explaining artificial intelligence model |
| CN112884024A (en) * | 2021-02-01 | 2021-06-01 | 中南大学 | Froth flotation working condition identification method based on group decision integrated learning |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104239785A (en) * | 2014-09-30 | 2014-12-24 | 中国人民解放军国防科学技术大学 | Intrusion detection data classification method based on cloud model |
| CN105844927A (en) * | 2016-04-06 | 2016-08-10 | 深圳榕亨实业集团有限公司 | Novel control system and novel control method for sensing and controlling road intersection group signals |
| JP2016212523A (en) * | 2015-04-30 | 2016-12-15 | キヤノン株式会社 | Information processing apparatus and information processing method |
| CN106250913A (en) * | 2016-07-21 | 2016-12-21 | 江苏大学 | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis |
| US20170011294A1 (en) * | 2015-07-06 | 2017-01-12 | Texas Instruments Incorporated | Efficient Decision Tree Traversal in an Adaptive Boosting (ADABOOST) Classifier |
| CN108154172A (en) * | 2017-12-21 | 2018-06-12 | 重庆邮电大学 | Image-recognizing method based on three decisions |
| CN108596329A (en) * | 2018-05-11 | 2018-09-28 | 北方民族大学 | Threedimensional model sorting technique based on end-to-end Deep integrating learning network |
| US10140553B1 (en) * | 2018-03-08 | 2018-11-27 | Capital One Services, Llc | Machine learning artificial intelligence system for identifying vehicles |
| US20190042894A1 (en) * | 2018-09-28 | 2019-02-07 | Intel Corporation | Embedding human labeler influences in machine learning interfaces in computing environments |
| CN109816022A (en) * | 2019-01-29 | 2019-05-28 | 重庆市地理信息中心 | A kind of image-recognizing method based on three decisions and CNN |
| CN109934203A (en) * | 2019-03-25 | 2019-06-25 | 南京大学 | A kind of cost-sensitive increment type face identification method based on comentropy selection |
| CN110008925A (en) * | 2019-04-15 | 2019-07-12 | 中国医学科学院皮肤病医院 | A kind of skin automatic testing method based on integrated study |
| CN110135587A (en) * | 2019-05-10 | 2019-08-16 | 山西大学 | It is hesitated based on section and obscures the Multiple Attribute Group Decision of more granularity decision rough sets |
| CN110210550A (en) * | 2019-05-28 | 2019-09-06 | 东南大学 | Image fine granularity recognition methods based on integrated study strategy |
-
2019
- 2019-11-26 CN CN201911173435.5A patent/CN111046926B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104239785A (en) * | 2014-09-30 | 2014-12-24 | 中国人民解放军国防科学技术大学 | Intrusion detection data classification method based on cloud model |
| JP2016212523A (en) * | 2015-04-30 | 2016-12-15 | キヤノン株式会社 | Information processing apparatus and information processing method |
| US20170011294A1 (en) * | 2015-07-06 | 2017-01-12 | Texas Instruments Incorporated | Efficient Decision Tree Traversal in an Adaptive Boosting (ADABOOST) Classifier |
| CN105844927A (en) * | 2016-04-06 | 2016-08-10 | 深圳榕亨实业集团有限公司 | Novel control system and novel control method for sensing and controlling road intersection group signals |
| CN106250913A (en) * | 2016-07-21 | 2016-12-21 | 江苏大学 | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis |
| CN108154172A (en) * | 2017-12-21 | 2018-06-12 | 重庆邮电大学 | Image-recognizing method based on three decisions |
| US10140553B1 (en) * | 2018-03-08 | 2018-11-27 | Capital One Services, Llc | Machine learning artificial intelligence system for identifying vehicles |
| CN108596329A (en) * | 2018-05-11 | 2018-09-28 | 北方民族大学 | Threedimensional model sorting technique based on end-to-end Deep integrating learning network |
| US20190042894A1 (en) * | 2018-09-28 | 2019-02-07 | Intel Corporation | Embedding human labeler influences in machine learning interfaces in computing environments |
| CN109816022A (en) * | 2019-01-29 | 2019-05-28 | 重庆市地理信息中心 | A kind of image-recognizing method based on three decisions and CNN |
| CN109934203A (en) * | 2019-03-25 | 2019-06-25 | 南京大学 | A kind of cost-sensitive increment type face identification method based on comentropy selection |
| CN110008925A (en) * | 2019-04-15 | 2019-07-12 | 中国医学科学院皮肤病医院 | A kind of skin automatic testing method based on integrated study |
| CN110135587A (en) * | 2019-05-10 | 2019-08-16 | 山西大学 | It is hesitated based on section and obscures the Multiple Attribute Group Decision of more granularity decision rough sets |
| CN110210550A (en) * | 2019-05-28 | 2019-09-06 | 东南大学 | Image fine granularity recognition methods based on integrated study strategy |
Non-Patent Citations (3)
| Title |
|---|
| 刘久兵;顾萍萍;周献中;黄兵;李华雄;: "基于优化模型的直觉模糊三支群决策方法", no. 05 * |
| 姚富光;钟先信;周靖超;: "粒计算:一种大数据融合智能建模新方法", 南京理工大学学报, no. 04 * |
| 骆公志;梅焘;: "监督机制多粒度决策粗糙集模型及应用", no. 18 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111768381A (en) * | 2020-06-29 | 2020-10-13 | 北京百度网讯科技有限公司 | Part defect detection method and device and electronic equipment |
| CN112668871A (en) * | 2020-12-25 | 2021-04-16 | 中国人民解放军63892部队 | Dynamic assignment method for expert weight in multi-round group decision |
| CN112668871B (en) * | 2020-12-25 | 2023-10-13 | 中国人民解放军63892部队 | Method for dynamically distributing special weights in multi-round group decision |
| CN112884024A (en) * | 2021-02-01 | 2021-06-01 | 中南大学 | Froth flotation working condition identification method based on group decision integrated learning |
| CN112884024B (en) * | 2021-02-01 | 2022-05-13 | 中南大学 | Froth flotation working condition identification method based on group decision integrated learning |
| CN112766415A (en) * | 2021-02-09 | 2021-05-07 | 第四范式（北京）技术有限公司 | Method, device and system for explaining artificial intelligence model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111046926B (en) | 2023-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111046926A (en) | Computer vision image classification integrated learning method | |
| CN111181939B (en) | Network intrusion detection method and device based on ensemble learning | |
| JP5567448B2 (en) | Image area dividing apparatus, image area dividing method, and image area dividing program | |
| CN110852288B (en) | Cell image classification method based on two-stage convolutional neural network | |
| CN114283469B (en) | Improved YOLOv4-tiny target detection method and system | |
| CN109934278B (en) | High-dimensionality feature selection method for information gain mixed neighborhood rough set | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN111539451B (en) | Sample data optimization method, device, equipment and storage medium | |
| CN112560755A (en) | Target detection method for identifying urban exposed garbage | |
| WO2019167784A1 (en) | Position specifying device, position specifying method, and computer program | |
| CN112149758A (en) | Hyperspectral open set classification method based on Euclidean distance and deep learning | |
| CN111860359B (en) | Point cloud classification method based on improved random forest algorithm | |
| CN114612721A (en) | Image classification method based on multilevel adaptive feature fusion type increment learning | |
| CN106570514A (en) | Automobile wheel hub classification method based on word bag model and support vector machine | |
| CN111461002B (en) | Sample processing method for thermal imaging pedestrian detection | |
| CN113065395A (en) | Radar target new class detection method based on generation countermeasure network | |
| CN116704208B (en) | Local interpretable method based on characteristic relation | |
| CN113657394A (en) | Remote sensing image segmentation method based on Markov random field and evidence theory | |
| JPH09171552A (en) | Picture recognizing device | |
| CN115294406B (en) | Method and system for multi-modal interpretable classification based on attributes | |
| CN115810105A (en) | Panorama segmentation method, device, equipment and storage medium | |
| CN114943290B (en) | Biological intrusion recognition method based on multi-source data fusion analysis | |
| CN114328174A (en) | Multi-view software defect prediction method and system based on counterstudy | |
| CN115409776A (en) | Power transmission line hardware fitting graph convolution detection method and system based on orientation reasoning | |
| Kiran et al. | Ground truth energies for hierarchies of segmentations |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230823 Address after: Building S02, 1036 Gaoxin Langchao Road, Jinan, Shandong 250100 Applicant after: Shandong Inspur Scientific Research Institute Co.,Ltd. Address before: North 6th floor, S05 building, Langchao Science Park, 1036 Langchao Road, hi tech Zone, Jinan City, Shandong Province, 250100 Applicant before: SHANDONG INSPUR ARTIFICIAL INTELLIGENCE RESEARCH INSTITUTE Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20200421 Assignee: Shandong Inspur Digital Business Technology Co.,Ltd. Assignor: Shandong Inspur Scientific Research Institute Co.,Ltd. Contract record no.: X2023980053547 Denomination of invention: A Computer Vision Image Classification Ensemble Learning Method Granted publication date: 20230919 License type: Exclusive License Record date: 20231226 |