Multi-sensor information fusion method based on trust index divergence

Abstract

The invention provides a multi-sensor information fusion method based on trust index divergence, which considers the conflict degree between evidences and the uncertainty of the evidences while determining the weight of the evidences, overcomes the problem that the evidence weight determination is mostly based on the conflict degree between the evidences and the characteristics of the evidences are rarely considered in the prior art, and improves the reliability of the evidence weight determination. According to the method, the conflict degree can be reasonably measured and corrected by fully utilizing the existing evidence information, the influence of the conflict evidence can be reduced, and unreasonable results are avoided. Information fusion is carried out on data of a multi-sensor decision-making layer, so that the defects of large calculation amount and subjective factor influence of artificially given weight coefficients in a conventional method can be overcome, the influence of conflict evidences is reduced, and the speed and precision of multi-sensor information fusion are accelerated. The multi-sensor information fusion method provided by the invention is suitable for a target recognition scene of multi-sensor high-conflict information in engineering application, and can effectively solve the fusion problem of the multi-sensor high-conflict information.

Description

Technical Field

[0001] The invention relates to the field of information fusion, in particular to a fusion method of a multi-sensor information fusion algorithm and DS evidence theory. Background Art

[0002] DS evidence theory is a widely used uncertainty modeling and reasoning method. It characterizes uncertainty by defining trust functions and likelihood functions to construct confidence intervals. The combination rules it defines satisfy the commutative law and the associative law. The evidence theory can better handle the uncertainty problems caused by randomness and fuzziness without the need for prior probabilities and conditional probability densities, and effectively solve the problem of multi-sensor information fusion. In recent years, it has been widely used in information fusion, target recognition, fault diagnosis image processing and other fields.

[0003] Multi-sensor information fusion is to automatically analyze, synthesize and optimize the combination of multi-level and multi-angle data information provided by multiple homogeneous or heterogeneous sensors distributed in different locations according to certain criteria to complete the required decision and estimation and obtain the final judgment. Multi-sensor information fusion can reduce the redundancy and contradiction that may exist between multi-sensor information, improve the complementarity of information, reduce its uncertainty, and obtain a consistent interpretation of things or environment. DS evidence theory provides strong theoretical support for the expression and synthesis of uncertain information, and has become an important theory and method in the field of multi-source information fusion. However, in practical applications, due to the interference of the electromagnetic environment and the instability of the sensor itself (sudden failure of very few sensors), the information collected from different sensors has great uncertainty and high conflict. When using DS evidence theory to fuse high-conflict evidence, counter-intuitive results often occur. This defect also limits the application of DS evidence theory.

[0004] At present, in the relevant research on evidence theory, how to more accurately judge the degree of evidence conflict and its measurement method, determine the measurement indicators that characterize the strength of evidence conflict, and how to reasonably use the information of the evidence itself when determining the weight of evidence are important research directions. At present, many research methods are mostly based on the degree of conflict between evidence when determining the weight of evidence, and rarely consider the characteristics of the evidence itself. Although these methods have improved the synthesis results of high-conflict evidence to varying degrees, in general, these methods themselves are not perfect enough and the synthesis effect is not ideal enough. Summary of the invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides a multi-sensor information fusion method based on trust index divergence.

[0006] The steps of the technical solution adopted by the present invention to solve its technical problem are:

[0007] (1) All possible n target single types θ i Form a target type set Θ, expressed as Θ = {θ1,θ2,…θ i ,…,θ n}, i = 1, 2, ... n, define A j is a candidate target type, which contains one or more target single types, i.e. j = 1, 2, ..., M, N sensors detect and identify the same target respectively, and give basic probability values ​​corresponding to the candidate target types respectively. Then the lth sensor is said to provide the candidate target type A j The corresponding basic probability assignment m l (A j ) Listed as the first piece of evidence E l ,Right now

[0008] E l :m l (A1),m l (A2),…,m l (A j ),…,m l (A M ), (l=1,2,…,N), where m l Evidence E l The basic probability assignment function, m l (A j ) is the candidate target type A j The corresponding basic probability assignment is,

[0009] (2) N sensors provide N pieces of evidence; for evidence E l Perform Pignistic probability transformation to obtain the corresponding new evidence E l ′, l = 1, 2, ..., N, and get new evidence E l ′Basic probability assignment function m l ′:

[0010] (3) Calculate any two pieces of evidence E from the N pieces of evidence provided by N sensors u 、E v The trust index divergence BED(E u ,E v ), u=1, 2,...,N, v=1, 2,...,N;

[0011] Calculate the trust index divergence between the N pieces of evidence provided by N sensors and construct an N×N trust divergence matrix DMM;

[0012] Computational Evidence u The average trust index divergence

[0013]

[0014] (4) Determine whether all other evidence is relevant to evidence E u Support Sup u ,

[0015]

[0016] Confirm the evidence E u Credibility u ,

[0017]

[0018] (5) Calculate the evidence E u Credibility Entropy Ed u ,

[0019]

[0020] Computational Evidence u The amount of information I u ,

[0021]

[0022] (6) Confirmation of Evidence E u The weight w u ,

[0023]

[0024] (7) Perform weighted correction on the N pieces of evidence provided by N sensors to obtain the weighted average evidence Its basic probability assignment function is It is expressed as:

[0025]

[0026] Among them, w u Evidence E u The weight of Weighted average evidence Candidate target type A j The basic probability assignment, m u (A j ) is evidence E u Candidate target type A j The basic probability assignment of ;

[0027] (8) Using the DS evidence fusion rule to derive weighted average evidence Perform N-1 synthesis Get the fusion result E;

[0028] The fusion result E is:

[0029] E:m(θ1),m(θ2),…,m(θ i ),…,m(θ n )(i=1,2,…,n)

[0030] The target type with the largest basic probability assignment in the fusion result E is the final result of multi-sensor target recognition.

[0031] The basic probability assignment function m l ′:

[0032]

[0033] in is the Pignistic probability conversion function, m l ′(θ i ) is new evidence E l ′ in which the target single type θ i The basic probability assignment of Pignistic probability conversion function is defined as follows:

[0034] Where |A j | indicates candidate target type A j The number of target types included.

[0035] The trust index divergence BED(E u ,E v )for:

[0036]

[0037] Among them, m u ′(θ i ) is evidence E u After Pignistic probability conversion, the corresponding new evidence E u ′ in which the target single type θ i The basic probability assignment, m v ′(θ i ) is evidence E v After Pignistic probability conversion, the corresponding new evidence E v ′ in which the target single type θ i The basic probability assignment of .

[0038] The trust dispersion matrix DMM is:

[0039]

[0040] The DS evidence fusion rules are as follows:

[0041]

[0042] Among them, m(θ i ) is the target single type θ in the fusion result E i The basic probability assignment, A s ,A t (s=1,2,…,M,t=1,2,…,M) is the weighted average evidence The candidate target type, Weighted average evidence Candidate target type A s ,A t The basic probability assignment of .

[0043] An electronic device comprises one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method as described above.

[0044] A computer-readable storage medium stores program codes, and the program codes can be called by a processor to execute the method as described above.

[0045] The beneficial effect of the present invention is that the present invention considers both the degree of conflict between evidence and the uncertainty of the evidence itself when determining the weight of evidence, overcoming the problem that previous studies mostly determined the weight of evidence based on the degree of conflict between evidence, but rarely considered the characteristics of the evidence itself, and can make full use of existing evidence information to reasonably measure and correct the degree of conflict, which can reduce the impact of conflicting evidence and avoid the generation of unreasonable results. The use of the above method to fuse information on multi-sensor decision-making layer data can overcome the shortcomings of previous methods such as large amount of calculation and the influence of subjective factors of artificially given weight coefficients, reduce the impact of conflicting evidence, and speed up the speed and accuracy of multi-sensor information fusion. The multi-sensor information fusion method proposed in the present invention is suitable for target recognition scenarios with high-conflict information from multiple sensors in engineering applications, and can effectively solve the fusion problem of high-conflict information from multiple sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the multi-sensor information fusion method based on trust index divergence proposed in the present invention. DETAILED DESCRIPTION

[0047] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0048] Taking "aerial target recognition", a function in multi-sensor information fusion, as an example, the specific implementation process of the multi-sensor information fusion method based on evidence similarity is explained, that is, the target type is determined by the proposed method.

[0049] (1) There are 6 sensors that form a multi-sensor system to observe and identify a certain aerial target. During the identification process, there are three possible single target types, namely θ1: tanker, θ2: AWACS, and θ3: bomber. Then the target type set is Θ = {T: tanker, W: AWACS, R: bomber}. When the 6 sensors observe and identify the aerial target, the basic probability assignments corresponding to the candidate target types are given respectively, that is, the 6 pieces of evidence under the target type set Θ are (where the candidate target types are [A1: T; A2: W; A3: R; A4: T, W; A5: T, R; A6: Θ]):

[0050] E1: m1(T)=0.42, m1(W)=0.28, m1(R)=0.30, m1(T,W)=0, m1(T,R)=0, m1(Θ)=0;

[0051] E2: m2(T)=0.01, m2(W)=0.90, m2(R)=0.08, m2(T,W)=0.01, m2(T,R)=0, m2(Θ)=0;

[0052] E3: m3(T)=0.79, m3(W)=0.16, m3(R)=0.02, m3(T,W)=0, m3(T,R)=0.03, m3(Θ)=0;

[0053] E4: m4(T)=0.52, m4(W)=0, m4(R)=0.18, m4(T,W)=0, m4(T,R)=0, m4(Θ)=0.30;

[0054] E5: m5(T)=0.65, m5(W)=0.09, m5(R)=0.11, m5(T,W)=0.15, m5(T,R)=0, m5(Θ)=0;

[0055] E6: m6(T)=0.75, m6(W)=0.10, m6(R)=0, m6(T,W)=0, m6(T,R)=0, m6(Θ)=0.15.

[0056] (2) Perform Pignistic probability transformation on the six pieces of evidence provided by the six sensors, and the converted six pieces of evidence are:

[0057] E1′: m1′(T)=0.4200, m1′(W)=0.2800, m1′(R)=0.3000;

[0058] E2′: m2′(T)=0.0150, m2′(W)=0.9050, m2′(R)=0.0800;

[0059] E3′: m3′(T)=0.8050, m3′(W)=0.1600, m3′(R)=0.0350;

[0060] E4′: m4′(T)=0.620, m4′(W)=0.1000, m4′(R)=0.2800;

[0061] E5′: m5′(T)=0.7250, m5′(W)=0.1650, m5′(R)=0.1100;

[0062] E6′: m6′(T)=0.8000, m6′(W)=0.1500, m6′(R)=0.0500.

[0063] (3) Calculate the trust index divergence between the six pieces of evidence and obtain a 6×6 Trust Divergence Matrix DMM:

[0064]

[0065] Calculate the average trust index divergence of each evidence:

[0066] BED1=0.2765; BED2=0.7040; BED3=0.2423; BED4=0.2508; BED5=0.2052; BED6=0.2353.

[0067] (4) Obtain the support of each piece of evidence:

[0068] Sup1=3.6162; Sup2=1.4204; Sup3=4.1263; Sup4=3.9866; Sup5=4.8726; Sup6=4.2502.

[0069] Further obtain the credibility of each piece of evidence:

[0070] Cred1=0.1624; Cred2=0.0638; Cred3=0.1853; Cred4=0.1790; Cred5=0.2188; Cred6=0.1908.

[0071] (5) Calculate the credibility entropy of each piece of evidence:

[0072] Ed1=1.5610; Ed2=0.5770; Ed3=1.0039; Ed4=2.3795; Ed5=1.7152; Ed6=1.4751.

[0073] Calculate the information content of the evidence, the results are as follows:

[0074] I1=4.7634; I2=1.7808; I3=2.7288; I4=10.7995; I5=5.5578; I6=4.3716.

[0075] (6) Determine the weight of each piece of evidence:

[0076] w1=0.1439; w2=0.0211; w3=0.0940; w4=0.3596; w5=0.2262; w6=0.1552.

[0077] The 6 pieces of evidence provided by the 6 sensors are weighted and corrected to obtain the weighted average evidence. for:

[0078]

[0079] (7) Weighted average evidence After 5 rounds of synthesis, the final fusion result E is:

[0080] E: m(T)=0.9991, m(W)=0.0002, m(R)=0.0007.

[0081] In the final fusion result E, the basic probability of the single target type T is the largest, which is 0.9991. From the synthesis results, we can know that when using multiple sensors to observe aerial targets, a multi-sensor information fusion method based on trust index divergence is used to identify aerial targets, and the real target is identified as T: tanker, and the basic probability of target T is 0.9991.

Claims

1. A multi-sensor information fusion method based on trust index divergence, characterized in that The steps include: (1) All possible n target single types θ i Form a target type set Θ, expressed as Θ = {θ1,θ2,…θ i ,…,θ n }, i = 1, 2, ... n, define A j is a candidate target type, which contains one or more target single types, i.e. N sensors detect and identify the same target respectively, and give basic probability values ​​corresponding to the candidate target types respectively. Then the lth sensor is said to provide candidate target type A. j The corresponding basic probability assignment m l (A j ) Listed as the first piece of evidence E l , that is, E l :m l (A1),m l (A2),…,m l (A j ),…,m l (A M ), (l=1,2,…,N), where m l Evidence E l The basic probability assignment function, m l (A j ) is the candidate target type A j The corresponding basic probability assignment is, (2) N sensors provide N pieces of evidence; for evidence E l Perform Pignistic probability transformation to obtain the corresponding new evidence E l ′, l = 1, 2, ..., N, and get new evidence E l ′Basic probability assignment function m l ′: (3) Calculate any two pieces of evidence E from the N pieces of evidence provided by N sensors u 、E v The trust index divergence BED(E u ,E v ), u=1, 2,...,N, v=1, 2,...,N; Calculate the trust index divergence between the N pieces of evidence provided by N sensors and construct an N×N trust divergence matrix DMM; Computational Evidence u The average trust index divergence (4) Determine whether all other evidence is relevant to evidence E u Support Sup u , Confirm the evidence E u Credibility u , (5) Calculate the evidence E u Credibility EntropyEd u , Computational Evidence u The amount of information I u , (6) Confirmation of Evidence E u The weight w u , (7) Perform weighted correction on the N pieces of evidence provided by N sensors to obtain the weighted average evidence Its basic probability assignment function is It is expressed as: Among them, w u Evidence E u The weight of Weighted average evidence Candidate target type A j The basic probability assignment, m u (A j ) is evidence E u Candidate target type A j The basic probability assignment of ; (8) Using the DS evidence fusion rule to derive weighted average evidence Perform N-1 synthesis Get the fusion result E; The fusion result E is: E:m(θ1),m(θ2),…,m(θ i ),…,m(θ n (i=1,2,…,n) The target type with the largest basic probability assignment in the fusion result E is the final result of multi-sensor target recognition.

2. The multi-sensor information fusion method based on trust index divergence according to claim 1 is characterized in that: The basic probability assignment function m l ′: in is the Pignistic probability conversion function, m l ′(θ i ) is new evidence E l ′ in which the target single type θ i The basic probability assignment of Pignistic probability conversion function is defined as follows: Where |A j | indicates candidate target type A j The number of target single types included.

3. The multi-sensor information fusion method based on trust index divergence according to claim 2 is characterized in that: The trust index divergence BED(E u ,E v )for: Among them, m u ′(θ i ) is evidence E u After Pignistic probability conversion, the corresponding new evidence E u ′ in which the target single type θ i The basic probability assignment, m v ′(θ i ) is evidence E v After Pignistic probability conversion, the corresponding new evidence E v ′ in which the target single type θ i The basic probability assignment of .

4. The multi-sensor information fusion method based on trust index divergence according to claim 3 is characterized in that: The trust dispersion matrix DMM is:

5. The multi-sensor information fusion method based on trust index divergence according to claim 6 is characterized in that: The DS evidence fusion rules are as follows: Among them, m(θ i ) is the target single type θ in the fusion result E i The basic probability assignment, A s ,A t (s=1,2,…,M,t=1,2,…,M) is the weighted average evidence The candidate target type, Weighted average evidence Candidate target type A s ,A t The basic probability assignment of .

6. An electronic device, characterized in that: include: one or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, and the program codes can be called by a processor to execute the method according to any one of claims 1 to 5.

Images