An evidence correction method based on singular value decomposition and information entropy

By employing singular value decomposition and information entropy methods, the inapplicability of traditional Dempster combinatorial rules to highly conflicting evidence is addressed, thereby enhancing the target recognition capability of multi-sensor information fusion systems.

CN114756812BActive Publication Date: 2024-11-19HARBIN ENG UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210199769.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-02
Publication Date
2024-11-19
Estimated Expiration
2042-03-02

AI Technical Summary

Technical Problem

Traditional Dempster combination rules produce counterintuitive results when fusing highly conflicting evidence. The method for determining weighting factors is not comprehensive or objective enough, which affects the target recognition performance of multi-sensor information fusion systems.

Method used

The method employs singular value decomposition and information entropy. By constructing a basic probability assignment matrix, performing matrix transformation and singular value decomposition, calculating the discount correction vector, reconstructing the evidence matrix and measuring information entropy, correcting conflicting evidence, and finally fusing the evidence based on evidence theory.

Benefits of technology

It effectively filters out interfering information, improves the target recognition capability and accuracy of the multi-sensor information fusion system, and reduces the impact of conflicting evidence.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114756812B_ABST
    Figure CN114756812B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of multi-sensor information fusion, and specifically relates to an evidence correction method based on singular value decomposition and information entropy. The present invention utilizes the geometric properties of the basic probability assignment function to transform the problem of differences between evidence into a matrix property problem, and then proposes a reconstruction technology based on singular value decomposition to extract the subspace corresponding to the main direction of the evidence matrix, so as to filter out the interference information in the basic probability assignment matrix; for the evidence matrix with non-unique number of main directions, the information entropy measurement method is adopted to complete the correction of highly conflicting evidence; finally, the corrected evidence information is fused based on evidence theory. The present invention can correct the basic probability assignment with interference according to the maximum direction of the matrix without affecting the consistency data, and weaken its role in the evidence matrix by reducing the modulus of the focus element. The present invention is suitable for scenarios where uncertain information exists in multi-sensor information fusion systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-sensor information fusion, and in particular relates to an evidence correction method based on singular value decomposition and information entropy. Background Art

[0002] Multi-sensor information fusion is an information technology that automatically analyzes and processes the received information of multiple sensors according to certain criteria to achieve decision-making and estimation. Compared with a single sensor system, a multi-sensor system can synergistically exert the complementary characteristics of different sensors, and improve the system's measurement accuracy and sensitivity to the external environment by fusing information and reasonably allocating resources. As the object to be processed in information fusion technology, multi-source information quantitatively represents or qualitatively describes the properties, characteristics and background environmental information of the observed object. In actual multi-sensor systems, due to uncertain factors such as environmental disturbances, noise and transmission errors in the information detection stage, as well as incomplete and inaccurate information descriptions in various steps of the data preprocessing stage and the recognition output stage, the multi-source information to be processed during decision-level fusion often has varying degrees of uncertainty and inaccuracy, and even contradictions between data. These uncertain information seriously affects the effect of multi-sensor information fusion.

[0003] The method of uncertain information processing is based on the four theoretical foundations of probability theory, Dempster-Shafer (DS) evidence theory, fuzzy set theory and rough set theory. Among them, DS evidence theory, as an uncertainty reasoning method with a good mathematical foundation, quantifies the credibility and likelihood of propositions by introducing concepts such as basic probability assignment and trust function, and can effectively model, process and make decisions on incomplete or uncertain problems. Since the traditional Dempster combination rule will produce counterintuitive results when fusing highly conflicting evidence, it is necessary to weight and correct the conflicting fusion evidence. However, the method of determining the weight factor is not comprehensive and objective when combining the internal and external conflicting information provided by the evidence source, which will reduce the target recognition performance of the system.

[0004] Therefore, how to convert data information into weight factors comprehensively and objectively to further improve the sensitivity of evidence theory to interference information and the target recognition ability of information fusion system is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide an evidence correction method based on singular value decomposition and information entropy.

[0006] An evidence correction method based on singular value decomposition and information entropy includes the following steps:

[0007] Step 1: The target to be identified by the multi-sensor system is θ1, θ2, …, θ n , whose focal element set is F = {θ1, θ2, …, θ n}, the focus element set F = {θ1, θ2, ..., θ n}Basic probability assignment function m i (θ j ) to perform matrix processing and obtain the t×n order preprocessing matrix Y J ; Wherein, i = 1, 2, ..., t, t is the total number of functions; j = 1, 2, ..., n, n is the total number of focal elements;

[0008] Step 2: Calculate the preprocessing matrix Y J Singular value decomposition of

[0009] Where r is the matrix Y J rank, singular value σ i Satisfies σ1≥σ2≥…≥σ r , column vector u i and v i are the i-th left singular vector and the i-th right singular vector respectively;

[0010] Step 3: Calculate the discount correction vector α = [α1, α2, ..., α k ];

[0011] Step 4: According to Calculate the discounted corrected matrix Among them, Θ is the identification frame; y i =[y i (θ1), y i (θ2),…];

[0012] Step 5: Combine the Dempster formula The row vectors of the fusion matrix Y are fused to obtain the fusion result m; where θ∈F={θ1,θ2,…,θ n}, is an empty set;

[0013] Step 6: Finally, the target recognition result of the multi-sensor information fusion system is obtained in represents the calculation so that m(θ j ) takes the maximum value corresponding to θ j .

[0014] Furthermore, the method for obtaining the preprocessing matrix in step 1 is specifically as follows:

[0015] Step 1.1: Calculate the n×n order Jaccard coefficient matrix D = [d1,…,d i , …, d n ] n×n , and normalize its column vector to obtain

[0016] Among them, d i =[d i1 , …, d ij , …, d in ] T , d ij =d ji is the focal element θ i With θ j The ratio of the subset cardinality of the intersection to the subset cardinality of the union, [·] n×n represents the matrix order, [·] T Represents the transpose of a matrix or vector; is the unit column vector, ||·||2 represents the vector 2-norm;

[0017] Step 1.2: Construct a t×n order basic probability assignment matrix M = [m1, ..., m i ,…,m t ] T , and normalize the row vectors of matrix M to obtain

[0018] Among them, m i =[m i (θ1),…,m i (θ n )] T ; is a normalized row vector;

[0019] Step 1.3: Use the normalized Jaccard focal element matrix to preliminarily correct the normalized basic probability assignment matrix to obtain a t×n-order evidence matrix projected onto the unit circle.

[0020] Furthermore, the method for calculating the discount correction vector in step 3 is specifically as follows:

[0021] Step 3.1: If the singular decomposition result satisfies the condition of having a single maximum singular value σ1, that is, σ1≠σ2, then go to step 3.2, otherwise go to step 3.3;

[0022] Step 3.2: Let σ2=…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 And calculate the discount correction vector α = [α1, α2, ..., αt ];

[0023] in,(·) rank1 represents a matrix of rank 1; represents the infinite norm of the matrix, |·| represents taking the absolute value, and max(·) represents taking the maximum value;

[0024] Step 3.3: Let σ s+1 =…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 The matrix Y rank1 Normalizing the row vectors yields Y′=[y′1,…,y′ i ,…,y′ t ] T =[y′ ij (θ j )] t×n , calculate the discount correction vector α = [α1,…,α i , …, α t ];

[0025] Where s is the matrix Y J The number of singular values ​​of is equal to the largest singular value σ1; ||·||1 represents the vector 1 norm; H=[H(y'1),H(y'2),...,H(y' t )];

[0026] The beneficial effects of the present invention are:

[0027] The present invention aims at the problem that the correction method of conflicting evidence is not comprehensive and objective when combining the internal and external conflicting information provided by the evidence source, and provides an evidence correction method based on singular value decomposition and information entropy applicable to a multi-sensor information fusion system. The present invention uses the geometric properties of the basic probability assignment function to transform the problem of differences between evidence into a matrix property problem, and then proposes a reconstruction technology based on singular value decomposition to extract the subspace corresponding to the main direction of the evidence matrix to filter out the interference information in the basic probability assignment matrix; for the evidence matrix with non-unique number of main directions, the information entropy measurement method is adopted to complete the correction of highly conflicting evidence; finally, the corrected evidence information is fused based on evidence theory. The present invention can correct the basic probability assignment with interference according to the maximum direction of the matrix without affecting the consistency data, and weaken its role in the evidence matrix by reducing the modulus of the focus element. The present invention is suitable for scenarios where uncertain information exists in a multi-sensor information fusion system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a principle block diagram of the present invention.

[0029] Figure 2 A confidence distribution table for basic probability assignments in an embodiment of the present invention.

[0030] Figure 3 It is a basic probability assignment table of the discounted and corrected evidence matrix Y in an embodiment of the present invention.

[0031] Figure 4 This is a table of evidence fusion effects in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention is further described below in conjunction with the accompanying drawings.

[0033] The present invention constructs the evidence matrix and performs singular value decomposition on the basic probability assignment function to obtain the matrix subspace information; reconstructs the evidence matrix according to the main direction information to filter out the interference information; calculates the discount correction coefficient based on the information entropy to correct the residual interference; and finally fuses the corrected evidence information based on the evidence theory to obtain the decision result. The specific implementation steps are:

[0034] (1) The target to be identified by the multi-sensor system is θ1, θ2, …, θ n , whose focal element set is F = {θ1, θ2, ..., θ n}, the focus element set F = {θ1, θ2, ..., θ n}Basic probability assignment function m i (θ j ) to perform matrix processing and obtain the t×n order preprocessing matrix Y J , where i = 1, 2, ..., t, t is the total number of functions, j = 1, 2, ..., n, n is the total number of focal elements. The specific steps are as follows:

[0035] ① Calculate the n×n order Jaccard coefficient matrix D = [d1,…,d i , …, d n ] n×n , where d i =[d i1 , …, d ij , …, d in ] T , d ij =d ji is the focal element θ i With θ j The ratio of the subset cardinality of the intersection to the subset cardinality of the union, [·] n×n represents the matrix order, [·] Trepresents the transpose of a matrix or vector. And normalize its column vector to get in is the unit column vector, ||·||2 represents the vector 2-norm.

[0036] ②Construct a t×n order basic probability assignment matrix M=[m1,…,m i ,…,m t ] T , where m i =[m i (θ1),…,m i (θ n )] T . And normalize the row vector of matrix M to get in is a normalized row vector.

[0037] ③Use the normalized Jaccard focal element matrix to preliminarily correct the normalized basic probability assignment matrix and obtain the t×n-order evidence matrix projected onto the unit circle.

[0038] (2) Calculate the preprocessing matrix Y J Singular value decomposition of Where r is the matrix Y J rank, singular value σ i Satisfies σ1≥σ2≥…≥σ r , column vector u i and v i are the i-th left singular vector and the i-th right singular vector respectively.

[0039] (3) Calculate the discount correction vector α = [α1, α2, ..., α k ].

[0040] ①If the singular decomposition result satisfies the condition of having a single maximum singular value σ1, that is, σ1≠σ2, then go to ②, otherwise go to ③.

[0041] ②Let σ2=…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 represents a matrix of rank 1. And calculate the discount correction vector α = [α1, α2, …, α r ],in represents the infinite norm of the matrix, |·| represents taking the absolute value, and max(·) represents taking the maximum value.

[0042] ③Let σ s+1 =…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 Where s is the matrix Y J The number of singular values ​​of is equal to the largest singular value σ1. rank1 Normalizing the row vectors yields Y′=[y′1,…,y′ i ,…,y′ t ]T=[y′ ij (θ j )] t×n ,in ||·||1 represents the vector 1 norm. Calculate the discount correction vector α = [α1,…,α i , …, α t ],in H=[H(y′1),H(y′2),…,H(y′ t )],

[0043] (4) According to the formula Where Θ is the identification frame, and the discounted corrected matrix is ​​calculated y i =[y i (θ1), y i (θ2),…].

[0044] (5) According to Dempster combination formula The row vectors of the fusion matrix Y are fused to obtain the fusion result m, where θ∈F={θ1,θ2,…,θ n}, Is an empty set.

[0045] (6) Obtain target recognition results of the multi-sensor information fusion system in represents the calculation so that m(θ j ) takes the maximum value corresponding to θ j .

[0046] The following is a specific example 1 with reference to specific parameters:

[0047] The embodiment of the present application proposes an evidence correction method based on singular value decomposition and information entropy according to the geometric properties of the basic probability assignment function. The method applied for addresses the problem that the Dempster combination rule in evidence theory is not applicable in high-conflict evidence scenarios. It uses a method based on singular value decomposition to reconstruct the evidence matrix to filter out interference information and combines information entropy to perform discount correction processing on the evidence information. The method applied for can effectively correct conflicting evidence and improve the processing capability of the combination rule and the recognition capability of the system.

[0048] In order to more clearly illustrate the method, the present embodiment of the present invention uses simulation experiments to illustrate the process and demonstrate the effect, but does not limit the scope of the present embodiment of the present invention. The experimental conditions are: the recognition framework of the target to be identified by the multi-sensor system is θ = {θ1, θ2, θ3}, the total number of focal elements of the focal element set F generated by the system detection is n = 3, the total number of basic probability assignment functions is t = 4, and the basic probability assignment function m i (θ j ) is as follows: Figure 2 shown.

[0049] Figure 1 : is a principle block diagram of the method of the present invention, the method comprising:

[0050] S110. Basic probability assignment function m i (θ j )Matrix preprocessing:

[0051] ① Calculate the n×n order Jaccard coefficient matrix D = [d1,…,d i , …, d n ] n×n , where d i =[d i1 , …, d ij , …, d in ] T , d ij =d ji is the focal element θ i With θ j The ratio of the subset cardinality of the intersection to the subset cardinality of the union, [·] n×n represents the matrix order, [·] T represents the transpose of a matrix or vector. And normalize its column vector to get in is the unit column vector, ||·||2 represents the vector 2-norm.

[0052] ②Construct a t×n order basic probability assignment matrix M=[m1,…,m i ,…,m t ] T , where m i =[m i (θ1),…,m i (θ n )] T . And normalize the row vector of matrix M to get in is a normalized row vector.

[0053] ③Use the normalized Jaccard focal element matrix to preliminarily correct the normalized basic probability assignment matrix and obtain the t×n-order evidence matrix projected onto the unit circle.

[0054] S120. Preprocessing matrix reconstruction and discount coefficient calculation, the specific steps are as follows:

[0055] ① Calculate the preprocessing matrix Y J Singular value decomposition of Where r is the matrix Y J rank, singular value σ i Satisfies σ1≥σ2≥…≥σ r , column vector u i and v i are the i-th left singular vector and the i-th right singular vector respectively.

[0056] ②If the singular decomposition result satisfies the condition of having a single maximum singular value σ1, that is, σ1≠σ2, then go to ③, otherwise go to ④.

[0057] ③Let σ2=…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 (·) rank1 represents a matrix of rank 1. And calculate the discount correction vector α = [α1, α2, …, α t ],in represents the infinite norm of the matrix, |·| represents taking the absolute value, and max(·) represents taking the maximum value.

[0058] ④Let σ s+1 =…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 Where s is the matrix Y J The number of singular values ​​of is equal to the largest singular value σ1. rank1 Normalizing the row vectors yields Y′=[y′1,…,y′ i , ..., y′ t ] T =[y′ ij (θ j )] t×n ,in ||·||1 represents the vector 1 norm. Calculate the discount correction vector α = [α1,…,α i , …, α t ],in H=[H(y′1),H(y′2),…,H(y′ t )],

[0059] S130. Evidence matrix discount correction and target identification. The specific steps are as follows:

[0060] ①According to Where Θ is the identification frame, and the discounted corrected matrix is ​​calculated y i =[y i (θ1), y i (θ2),…].

[0061] ②According to Dempster combination formula The row vectors of the fusion matrix Y are fused to obtain the fusion result m, where θ∈F={θ1,θ2,…,θ n}, Is an empty set.

[0062] ③ Obtain the target recognition results of the multi-sensor information fusion system in represents the calculation so that m(θ j ) takes the maximum value corresponding to θ j .

[0063] The evidence matrix Y after discount correction is shown in Table 2, and the evidence matrix after fusion is shown in Table 3. The comparison algorithm is the target recognition result of directly using the traditional Dempster combination rule in evidence theory. It can be found that the evidence corrected by the proposed method of the present invention satisfies the rank of 1 except for the column where the recognition framework is located, that is, its row vectors are proportional to each other; and the conflicting evidence m2 in the matrix Y obtained after correction is successfully corrected and weakened, which can effectively avoid the influence of the conflicting evidence m2 and obtain the correct target recognition result.

[0064] In summary, the method of this implementation case can filter out the influence of interference information according to the spatial distribution characteristics of the basic probability assignment, and adopt information entropy discount processing for matrices with non-unique spatial distribution. The present invention can concentrate the information of the basic probability function in the subspace corresponding to the maximum singular value, reduce the influence of interference information on the fusion result, and at the same time increase the system's fusion convergence speed and target recognition accuracy by correcting and weakening the conflicting evidence.

[0065] In the process of correcting uncertain information in a multi-sensor information fusion system, the method of the present application weakens the influence of conflicting information in evidence fusion by reducing the modulus of the focal element, thereby improving the applicability of the traditional Dempster combination rule and the target recognition capability of the fusion system.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An evidence correction method based on singular value decomposition and information entropy, characterized in that: The following steps are involved: Step 1: The target to be identified by the multi-sensor system is θ1, θ2, …, θ n , whose focal element set is F = {θ1,θ2,…,θ n }, the focus element set F = {θ1,θ2,…,θ n }Basic probability assignment function m i (θ j ) to perform matrix processing and obtain the t×n order preprocessing matrix Y J ; Where i = 1, 2, ..., t, t is the total number of functions; j = 1, 2, ..., n, n is the total number of focal elements; Step 2: Calculate the preprocessing matrix Y J Singular value decomposition of Where r is the matrix Y J rank, singular value σ i Satisfies σ1≥σ2≥…≥σ r , column vector u i and v i are the i-th left singular vector and the i-th right singular vector respectively; Step 3: Calculate the discount correction vector α = [α1, α2, …, α k ]; Step 4: According to Calculate the discounted corrected matrix Among them, Θ is the identification frame; y i =[y i (θ1),y i (θ2),…]; Step 5: Combine the Dempster formula The row vectors of the fusion matrix Y are fused to obtain the fusion result m; where θ∈F={θ1,θ2,…,θ n }, is an empty set; Step 6: Finally, the target recognition result of the multi-sensor information fusion system is obtained in represents the calculation so that m(θ j ) when the corresponding θ takes the maximum value j .

2. The evidence correction method based on singular value decomposition and information entropy according to claim 1 is characterized in that: The method of obtaining the preprocessing matrix in step 1 is specifically as follows: Step 1.1: Calculate the n×n order Jaccard coefficient matrix D = [d1,…,d i ,…,d n ] n×n , and normalize its column vector to obtain Among them, d i =[d i1 ,…,d ij ,…,d in ] T , d ij =d ji is the focal element θ i With θ j The ratio of the subset cardinality of the intersection to the subset cardinality of the union, [·] n×n represents the matrix order, [·] T Represents the transpose of a matrix or vector; is the unit column vector, ||·||2 represents the vector 2-norm; Step 1.2: Construct a t×n order basic probability assignment matrix M = [m1,…,m i ,…,m t ] T , and normalize the row vectors of matrix M to obtain Among them, m i =[m i (θ1),…,m i (θ n )] T ; is a normalized row vector; Step 1.3: Use the normalized Jaccard focal element matrix to preliminarily correct the normalized basic probability assignment matrix to obtain a t×n-order evidence matrix projected onto the unit circle.

3. The evidence correction method based on singular value decomposition and information entropy according to claim 1 is characterized in that: The method for calculating the discount correction vector in step 3 is specifically as follows: Step 3.1: If the singular decomposition result satisfies the condition of having a single maximum singular value σ1, that is, σ1≠σ2, then go to step 3.2, otherwise go to step 3.3; Step 3.2: Let σ2=…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 And calculate the discount correction vector α = [α1, α2, …, α t ]; in,(·) rank1 represents a matrix of rank 1; represents the infinite norm of the matrix, |·| represents taking the absolute value, and max(·) represents taking the maximum value; Step 3.3: Let σ s+1 =…=σ r =0Reconstruction matrix Y J , and obtain a matrix of rank 1 The matrix Y rank1 Normalizing the row vectors yields Y'=[y'1,…,y' i ,…,y' t ] T =[y' ij (θ j )] t×n , calculate the discount correction vector α = [α1,…,α i ,…,α t ]; Where s is the matrix Y J The number of singular values ​​of is equal to the largest singular value σ1; ||·||1 represents the vector 1 norm; H=[H(y'1),H(y'2),…,H(y' t )];

Citation Information

Patent Citations

  • Method for mechanical fault diagnosis based on information entropies and evidence theory

    CN105373700A

  • Credit determination method, device and equipment

    CN113689114A

  • Method of operating a plant and a control system for controlling the same

    US20040083013A1

  • High-speed high-accuracy matrix singular value decomposition method, program, and device

    US20090028455A1