A weighted conflict evidence fusion method based on fuzzy classification

Abstract

The present invention discloses a weighted conflict evidence fusion method based on fuzzy classification, comprising the following steps: first, information of multiple evidences is obtained by a sensor, and the evidences are converted into single subset evidences through an interval probability assignment conversion formula. After the evidences are regarded as vectors, the membership between the evidences is obtained, and then the fuzzy equivalence matrix is ​​obtained through the membership matrix, and then the evidences are fuzzily classified according to the threshold. The mutual support between the same type of evidences is obtained, and the weight of each piece of evidence in the same type of evidence is obtained based on the mutual support, and the evidences are weighted averaged and fused. The weight of each piece of evidence is obtained by combining the relevant support between the evidences with the trust obtained by each piece of evidence after fusion. After the weighted fusion of the evidences, the secondary DS fusion is performed. The present invention has important theoretical significance and application value by fuzzily classifying the evidences and obtaining the weights of the same type of evidence and the different types of evidences using different methods.

Description

Technical Field

[0001] The invention relates to the technical field of target recognition, and in particular to a weighted conflict evidence fusion method based on fuzzy classification. Background Art

[0002] At present, target recognition technology processes and fuses information obtained by multiple sensors, which can effectively avoid the limitations of single sensor decision-making. However, due to external reasons or sensor failures, the collected data may not be very reliable, and even wrong information may be collected, which will lead to wrong decisions. The essence of Dempster-Shafer (DS) evidence theory is a generalization of probability theory, which expands many basic events in probability theory into a power set space of basic events and establishes a basic probability assignment function on this space. Compared with traditional probability theory, DS evidence theory can not only effectively express random uncertainty, but also better express incomplete information and subjective uncertain information. Moreover, compared with Bayesian theory, evidence theory can realize the fusion of evidence without prior knowledge, and after fusion, it can effectively reduce the uncertainty of the system.

[0003] However, the DS evidence theory has limitations when dealing with highly conflicting evidence, and the fusion produces counterintuitive results. In order to solve the problem of highly conflicting evidence, Yager made some improvements to the evidence theory. Considering that the conflict cannot be determined, he assigned the conflict to an unknown proposition. However, his method also has problems in some cases. That is, when a certain evidence assigns a basic probability of 0 to a certain event, the final basic probability assignment result will be 0. Even if other evidence has a higher basic probability assignment for the event, it cannot affect the synthesis result. In order to solve the problem of fusion of conflicting evidence, reference [1] proposed a simple averaging method to eliminate the impact of conflicting evidence on fusion. However, this method only performs a simple average on all evidence and does not consider the mutual influence between evidence. Reference [2] proposed an evidence theory classification fusion algorithm based on Deng entropy. With entropy reduction as the main idea, the evidence is classified and fused. In the decision-making process, the fusion result of the category with the most evidence is used as the overall fusion result to avoid the influence of high-conflict evidence and improve the information validity of the fusion result. However, since only the fusion result of the category with the most evidence is used as the overall fusion result, without considering the fusion results of other categories, some information will be lost, which will have an adverse effect on the accuracy of the result.

[0004] Compared with previous methods, the present invention proposes the concept of mutual support on the basis of fuzzy classification, and adopts different methods to evaluate the weights of evidence belonging to the same category and evidence of different categories respectively. After the weights are obtained, the evidence is weighted averaged to obtain new evidence for self-DS fusion, which further improves the focus of evidence while ensuring that no information is lost. Summary of the invention

[0005] The purpose of the present invention is to provide a weighted conflict evidence fusion method based on fuzzy classification, which can effectively make correct decisions on identification targets.

[0006] The technical solution adopted by the present invention is: a weighted conflict evidence fusion method based on fuzzy classification, which is characterized by:

[0007] Step 1: Obtain the recognition framework of target recognition Θ = {θ 1 ,θ 2 ,θ 3 ,…,θ u}, u is the number of elements in the recognition frame Θ, θ u To identify the type of target. Combining any elements in the recognition framework will produce a power set in is an empty set, there exists θ g and If m(θ g )≥0, then we call θ g To identify the focal element on the frame Θ. The basic probability assignment of the focal element corresponding to the evidence obtained by multiple sensors, and each evidence is regarded as a vector, the i-th evidence can be expressed as:

[0008] m i =(m i (θ 1 ),m i (θ 2 ),…,m i (θ r ),…,m i (θ u ), where i = 1, 2, ..., n, n is the total number of evidences, u is the number of elements in the recognition frame Θ, r = 1, 2, ..., u, m i (θ r ) is evidence m i θ r Basic probability assignment of focal elements;

[0009] Step 2: When the evidence contains multi-subset focal elements, the multi-subset focal elements of the evidence in step 1 are transformed into single-subset focal elements using the interval probability assignment formula based on the combination of the belief function Bel and the plausibility function Pl; the formula is as follows: Calculate the corrected evidence vector m' i =(m i '(θ 1 ),m i '(θ 2 ),…,m i '(θ r)), where m i '(θ r )=P i (θ r ), assign a value to the basic probability of the modified single-subset focal element. Represents the proposition θ u is a true trust measure, representing the upper limit of the trust level, at this time θ g =θ u , is the singleton focal element; Represents the proposition θ u is a non-false plausibility measure, representing the lower limit of the degree of trust. is a multi-subset focal element, where is an empty set; Represents all singleton focal elements θ r is the accumulation of true trust measures;

[0010] Step 3: Find the membership degree r between n pieces of evidence ij , the larger the value, the higher the correlation between the two pieces of evidence, and the smaller the value, the lower the correlation between the two pieces of evidence; according to the evidence vector m' corrected in step 2 i and m' j Find the membership of the i-th evidence to the j-th evidence. The function formula is: where i, j = 1, 2, ..., n, r ij For evidence i 'For evidence m' j The membership degree, ||m' i || for evidence m i 'It is considered as the two-norm of the vector, e is a natural constant, which is an irrational number approximately equal to 2.718281828459904523536..., exp is an exponential function with the natural number e as the base, m' j T is the corrected evidence vector m' j The transpose of

[0011] Step 4: The membership degree r between the n pieces of evidence obtained in step 3 ij Construct a membership matrix R with n rows and n columns,

[0012] Step 5: According to the properties of the membership matrix R, it is concluded that the R matrix is ​​a fuzzy similarity matrix;

[0013] Step 6: Use the transitive closure method to obtain the fuzzy equivalent matrix from the fuzzy similarity matrix R

[0014] Step 7: According to the set threshold λ, the fuzzy equivalent matrix obtained in step 6 is converted into Change to Boolean equivalent matrix Boolean equivalence matrix Each column element with a value of 1 is classified into one category, and a fuzzy classification at the λ level is obtained;

[0015] Step 8: Perform weighted average fusion of evidence belonging to the same category;

[0016] Step 801: There are l pieces of evidence in the same category. When the similarity between two pieces of evidence is greater, the support between the pieces of evidence is s. ij The higher it is, Where i, j = 1, 2, ..., l, ε is the total number of evidences that need to be calculated for mutual support, and ε = l;

[0017] Step 802: The mutual support s between l pieces of evidence ij , where i,j=1,2,…,l, we can get a mutual support matrix S with l rows and l columns.

[0018] Step 803: Obtain evidence m based on the mutual support matrix S i 'The final support among similar evidences i , Where i = 1, 2, ..., l;

[0019] Step 804: Obtain the final support s of each piece of evidence i Then normalize it Where i = 1, 2, ..., l, and the normalized final support is the weight w of each piece of evidence of the same type i =S i , where i = 1, 2, ..., l, weighted average fusion of similar evidence is performed to obtain new evidence; if the same type of evidence contains only one piece of evidence, the weighted average is still the evidence itself;

[0020] Step 9: Evidence of the same type is weighted and fused to obtain new evidence. Since the number of evidence in each type of evidence is different, the new evidence fused from multiple pieces of evidence should have greater reliability and play a greater leading role than the new evidence fused from a single piece of evidence.

[0021] Step 901: normalize the final support of the new evidence and combine it with the reliability as the weight of each new evidence, and then normalize the weight;

[0022] Step 902: Perform weighted averaging on the new evidence according to the normalized weights to obtain the final evidence;

[0023] Step 10: Perform n-1 DS fusions on the final evidence, and the obtained evidence is the final result.

[0024] The invention has a simple structure, reasonable design, and clear objectives, and is used to fuzzily classify multiple pieces of evidence. The evidence is classified into corresponding categories, which more intuitively shows the degree of association between the evidence and greatly reduces the degree of confusion between the evidence. The same type of evidence and heterogeneous evidence are processed in different ways, which not only reduces the adverse effect of evidence with too high conflict on the fusion result, but also fully achieves the focusing effect. The multiple self-DS fusions of the weighted summed evidence further enhance the convergence, making the final fusion result more accurate and closer to the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] like Figure 1 As shown, the present invention includes

[0029] Step 1: Obtain the recognition framework of target recognition Θ = {θ 1 ,θ 2 ,θ 3 ,…,θ u}, which is a set of u mutually exclusive elements, θ u To identify the type of target, take the sensor identifying an aerial target as an example. The types of aerial targets are divided into three categories: bombers, civil aircraft, and helicopters. Therefore, the identification framework is determined as: θ = {θ 1 ,θ 2 ,θ 3}, where θ 1 for bombers, θ 2 For civil aircraft, 3 For helicopters. Combining any elements in the identification framework will produce a power set in is an empty set, there exists θ g and If m(θ g )≥0, then we call θ g Assign m to the focal element of the basic probability on the identification frame Θ. Considering each piece of evidence as a vector, the i-th piece of evidence can be expressed as: m i =(m i (θ 1 ),m i (θ 2 ),…,m i (θ r ),…,m i (θ u )), where i = 1, 2, ..., n, n is the total number of evidences, u is the number of elements in the recognition frame Θ, and r = 1, 2, ..., u.

[0030] Step 2: When the evidence contains multiple subset focal elements, the vector is represented as

[0031] m i =(m i (θ 1 ),m i (θ 2 ),…,m i (θ r ),…,m i (θ u ),m i (θ 1 ∪θ 2 ),…,Θ), the interval probability assignment formula based on the combination of the belief function Bel and the plausibility function Pl is adopted:

[0032] Convert the multi-subset focal element of the evidence in step 1 into a single-subset focal element. Calculate the corrected evidence m' i =(m i '(θ 1 ),m i '(θ 2 ),…,m i '(θ r )), where m i '(θ r )=P i (θ r ), assign a value to the basic probability of the modified single-subset focal element. Represents the proposition θ u is a true trust measure, representing the upper limit of the trust level, at this time θ g =θ u , is the singleton focal element; Represents the proposition θu is a non-false plausibility measure, representing the lower limit of the degree of trust. is a multi-subset focal element, where is an empty set; Represents all singleton focal elements θ r The accumulation of true trust measures. According to the interval probability assignment formula, the multi-subset focal element is transformed into a single-subset focal element.

[0033] Step 3: Find the membership degree r between each pair of evidence ij , the larger the value, the higher the correlation between the two pieces of evidence, and the smaller the value, the lower the correlation between the two pieces of evidence. Find the evidence m converted by the interval probability conversion formula in step 2 i 'For evidence m' j The membership degree of is: where i, j = 1, 2, ..., n, r ij For evidence i 'For evidence m' j The membership degree, ||m' i || for evidence m i 'It is considered as the two-norm of the vector, e is a natural constant, which is an irrational number approximately equal to 2.718281828459904523536..., exp is an exponential function with the natural number e as the base, m' j T is the corrected evidence vector m' j The transpose of ij The following conditions are met:

[0034] ①:0≤r ij ≤1;

[0035] ②: r ij =r ji ;

[0036] ③: r ij =1(i=j);

[0037] ④: When evidence m i 'With evidence m' j Totally irrelevant, ij =0;

[0038] From the properties of the two norm, we know that ①② holds true; when the evidence m i 'With evidence m' j When the two pieces of evidence are completely identical, the correlation between them is 1, and ③ is true; when the two pieces of evidence are completely unrelated, the vectors of the evidence are in an orthogonal relationship, and ④ is true. Meeting these conditions allows the function formula to accurately measure the membership between two pieces of evidence, and there will be no error when measuring identical or completely unrelated evidence.

[0039] Step 4: Calculate the membership degree r between n pieces of evidence from step 3 ij , construct a membership matrix R with n rows and n columns, and the diagonal elements of the matrix are all 1, that is, Where i = 1, 2, ..., n, so

[0040] Step 5: Determine the properties of the membership matrix R. If a matrix is ​​a symmetric matrix, then the matrix is ​​symmetric. If the values ​​of the main diagonal elements of the matrix are all 1, then the matrix is ​​reflexive. Based on these two properties, we can determine whether a matrix is ​​a fuzzy similarity matrix. If the matrix satisfies the above two properties at the same time, it is called a fuzzy similarity matrix. ii =1, where i = 1, 2, ..., n, so the matrix satisfies the reflexivity, R T =R, so the matrix satisfies the symmetry. Since the matrix R satisfies these two properties at the same time, the matrix R is a fuzzy similarity matrix.

[0041] Step 6: Use the transfer closure method to obtain the fuzzy equivalence matrix R obtained in step 5

[0042] Step 601: When finding the fuzzy equivalent matrix of n pieces of evidence, the maximum number of transitive closure operations required is σ=log 10 n 2 +1, for example, if we need to find the fuzzy equivalent matrix of 5 pieces of evidence, σ = log 10 5 2 +1, which is approximately equal to 3, so the most complicated case is R→R 2 →R 4 →R 8 , the maximum number of times R 8 , and perform three transitive closure operations.

[0043] Step 602: The transitive closure operation of the fuzzy similarity matrix is ​​as follows:

[0044] ο is the definition of transitive closure,

[0045] e=max(min(a,a),min(b,c)), f=max(min(a,b),min(b,d)),

[0046] g=max(min(c,a),min(d,c)),h=max(min(c,b),min(d,d)), until we get N and K are any positive integers and N<K. In this case, the fuzzy equivalent matrix of the fuzzy similarity matrix R is Fuzzy Equivalence Matrix Satisfies reflexivity, symmetry and transitivity.

[0047] Step 7: Find the fuzzy equivalent matrix Then, according to the set threshold λ (the threshold λ represents the judgment limit of fuzzy classification), the fuzzy equivalence matrix is ​​converted into a Boolean equivalence matrix through the threshold λ Find the element value r that is not less than λ in the fuzzy equivalence matrix ij '≥λ, then let r ij '=1, the new matrix obtained after changing the matrix element value is the Boolean equivalent matrix Then, according to the elements in each column of the matrix, the elements with a value of 1 are divided into the same category, and the fuzzy classification at the λ level can be obtained.

[0048] Step 8: After the evidence is fuzzy classified, the evidence classified into the same category is fused by weighted average within one category.

[0049] Step 801: l pieces of evidence are classified into the same category, and there is mutual support s between the pieces of evidence ij , when two evidence vectors are more similar, the support between them is higher. The mutual support of intra-class evidence is:

[0050] Where i,j=1,2,…,l, ε is the total number of evidences that need to be calculated for mutual support, then ε=l, and mutual support satisfies the following conditions:

[0051] ①: 0≤s ij ≤1;

[0052] ②:s ij =1(i=j);

[0053] ③:s ij =s ji

[0054] The proof is as follows: When i=j, s ij =exp{0}=1, ② holds true; when i and j are swapped, the exponential coefficient remains unchanged, ③ holds true. When the similarity between the evidence is smaller, s ij The smaller it is, but it is always not less than 0, and ① is established. Meeting the conditions can accurately measure the mutual support between evidences and always ensure that the mutual support is within the interval [0,1].

[0055] Step 802: The mutual support s between l pieces of evidence ij Construct a mutual support matrix S with l rows and l columns, Each element in the S matrix represents the mutual support between the corresponding evidence.

[0056] Step 803: The mutual support matrix S obtained in step 802 is used to add up the support of each piece of evidence received from other pieces of evidence to obtain the final support s of this piece of evidence in this category. i , Where i=1,2,…,l.

[0057] Step 804: The final support s of each piece of evidence is obtained i Normalization is performed, and the final support of the evidence after normalization is expressed as S i express, Where i = 1, 2, ..., l. In the same type of evidence, the normalized final support of each piece of evidence is the weight w of each piece of evidence in the same type of evidence. i =S i , where i = 1, 2, ..., l. Perform weighted averaging based on the weights to obtain new evidence m": where i = 1, 2, ..., l, θ r It is the focal element of the evidence. For an evidence vector that occupies a category alone, it still remains itself after weighted averaging within the category.

[0058] Step 9: Perform weighted fusion again on the fused new evidence, because the number of evidences in each class before fusion is different. Compared with the case where a class contains only one evidence vector, the more evidence vectors there are in the class, the more reliable the new evidence after fusion should be, and it should play a greater leading role. Initially, there are n pieces of evidence, and the evidence is divided into p classes. If a class contains q pieces of evidence, then the reliability of the new evidence obtained after fusion of this type of evidence is Where i=1,2…,p.

[0059] Step 901: The evidence is finally divided into p categories, so there will be p new pieces of evidence after fusion. It is necessary to consider not only the support of each piece of evidence by other evidence, but also the reliability of the evidence itself. The support and reliability of the evidence are combined to obtain the weight of each piece of evidence in the fused evidence. Where i=1,2…,p,S i For evidence i "The final support after normalization, T i For evidence i "Reliability. Find the weight of evidence W i After that, the weights are normalized Where i=1,2…,p.

[0060] Step 902: According to the obtained weight w i 'Weighted average of the p pieces of fused evidence is used to obtain new evidence and the final evidence m' is obtained: θr Jiao Yuan is the evidence.

[0061] Step 10: Until the final evidence m”’ is obtained, perform n-1 DS fusions on this evidence to enhance the focusing ability. The evidence after DS fusion is the final result. The DS fusion rules are as follows:

[0062] is an empty set, m(A) is the basic probability assignment of the focal element to A, in K represents the degree of conflict between two pieces of evidence. The larger K is, the greater the degree of conflict between the two pieces of evidence is.

[0063] The following is an example to illustrate:

[0064] Step 1: Now there are 5 different sensors to identify a certain aerial target. Assume that the recognition framework is Θ = {θ 1 :Bomber,θ 2 :Civil aircraft,θ 3 : helicopter}, the information obtained by the five sensors on the observed targets is converted into basic probability assignments as shown below.

[0065] Evidence 1 :m 1 (θ 1 )=0.41,m 1 (θ 2 )=0.29,m 1 (θ 3 )=0.3;

[0066] Evidence 2 :m 2 (θ 2 )=0.9,m 2 (θ 3 )=0.1;

[0067] Evidence 3 :m 3 (θ 1 )=0.58,m 3 (θ 2 )=0.07,m 3 (θ 1 ,θ 3 )=0.35;

[0068] Evidence 4 :m 4 (θ 1 )=0.55,m 4 (θ 2 )=0.1,m 4 (θ1 ,θ 3 )=0.35;

[0069] Evidence 5 :m 5 (θ 1 )=0.6,m 5 (θ 2 )=0.1,m 5 (θ 1 ,θ 3 )=0.3;

[0070] These evidences are represented by vectors, namely m 1 =(0.41,0.29,0.3,0), m 2 =(0,0.9,0.1,0), m 3 =(0.58,0.07,0,0.35), m 4 =(0.55,0.1,0,0.35), m 5 =(0.6,0.1,0,0.3).

[0071] Step 2: The vector obtained by correcting the evidence through the interval probability conversion formula is: m' 1 =(0.41,0.29,0.3), m' 2 =(0,0.9,0.1), m' 3 =(0.8547,0.0974,0.0479),m' 4 =(0.813,0.1391,0.0479), m' 5 =(0.8532,0.1341,0.0307).

[0072] Step 3: Find the membership degree r between each pair of evidence ij , Where i, j = 1, 2, 3, 4, 5. 11 =1, r 12 =0.23, r 13 =0.45, r 14 =0.49, r 15 =0.45, r 22 =1, r 23 =0.03, r 24 =0.06, r 25 =0.05, r 33 =1, r 34 =0.99, r 35 =0.99, r 44 =1, r 45 =0.99, r 55 =1.

[0073] Step 4: Find the membership degree r of each piece of evidence with other pieces of evidence ij Then, a membership matrix R is constructed from these memberships.

[0074] Step 5: According to the properties of the membership matrix R, that is, r ii =1, where i = 1, 2, 3, 4, 5, satisfies reflexivity, and R T =R, satisfies the symmetry, and judges that the membership matrix R is a fuzzy similarity matrix.

[0075] Step 6: Use the transitive closure method to obtain the fuzzy equivalent matrix from the fuzzy similarity matrix R

[0076] Step 601: n=5, σ=log 10 n 2 +1, σ≈3, so we only need to use the transitive closure formula three times at most to get the fuzzy equivalent matrix R→R 2 →R 4 →R 8 .

[0077] Step 602:

[0078]

[0079]

[0080] R 2 It is the fuzzy equivalent matrix of R

[0081] Step 7: Find the fuzzy equivalent matrix Finally, a threshold λ is set. This threshold represents the judgment limit of the correlation between two pieces of evidence. When λ=1, the same evidence can be classified into one category. Therefore, in order to ensure the rigor of classification, a larger λ value is set as much as possible. The element value that is not less than λ is r ij '≥λ changes to r ij '=1, the matrix after changing the element value is called a Boolean equivalent matrix. ij '=1 elements are grouped into one category, and we can get equivalent classification at the λ level. Here, λ=0.8, the Boolean equivalent matrix Therefore, m' 3 , m' 4 , m' 5 Classified as one, m' 1 For a class, m'2 For another category.

[0082] Step 8: After the evidence is classified, the evidence in the same category is weighted averaged. 3 , m' 4 , m' 5 Weighted fusion, m' 1 and m' 2 After weighted fusion, it still remains the same.

[0083] Step 801: There is mutual support s between the three pieces of evidence belonging to the same category. ij , when two pieces of evidence are more similar, the support between them is higher. The mutual support formula of intra-class evidence is: where i, j = 3, 4, 5, ε = 3, s 34 =0.96, s 35 =0.97, s 45 =0.98.

[0084] Step 802: Calculate the mutual support s between each piece of evidence and other pieces of evidence ij Then, construct the mutual support matrix

[0085] Step 803: The support degree of each piece of evidence received by other pieces of evidence is accumulated through the mutual support matrix S to obtain the final support degree of the evidence. where i = 3, 4, 5, s 3 =1+0.96+0.97=2.93,s 4 =0.96+1+0.98=2.94,s 5 =0.97+0.98+1=2.95.

[0086] Step 804: After obtaining the final support of each piece of evidence, normalize it. Where i = 3, 4, 5,

[0087] At this time, these evidences are of the same type, and weighted fusion of the evidences is performed directly, with the weights being w i =S i , where i = 3, 4, 5, w 3 =0.332, w 4 =0.333,w 5 =0.335.

[0088] m″ 3 (θ 1 )=w 3 ×m′ 3 (θ1 )+w 4 ×m' 4 (θ 1 )+w 5 ×m' 5 (θ 1 )=0.834,

[0089] m″ 3 (θ 2 )=w 3 ×m' 3 (θ 2 )+w 4 ×m' 4 (θ 2 )+w 5 ×m' 5 (θ 2 )=0.126,

[0090] m″ 3 (θ 3 )=w 3 ×m' 3 (θ 3 )+w 4 ×m' 4 (θ 3 )+w 5 ×m' 5 (θ 3 )=0.040.

[0091] m' 1 and m' 2 It occupies a category alone, so after weighted average, it is still itself, m″ 1 =m' 1 ,m″ 2 =m' 2

[0092] Step 9: Find m″ 1 ,m″ 2 ,m″ 3 The mutual support between Mutual support matrix between evidence After normalizing the mutual support, the normalized final support of each piece of evidence is S 1 =0.39, S 2 =0.28, S 3 =0.33. There are 5 pieces of evidence before classification. The evidence is divided into 3 categories, and the number of pieces in each category is 1, 1, and 3 respectively. Therefore, the reliability of each piece of new evidence after fusion is calculated as

[0093] Step 901: Combine support and reliability to obtain weight

[0094] Normalize it to get the final weight Where i = 1, 2, 3, w 1 '=0.30,w' 2 =0.21, w' 3 =0.49.

[0095] Step 902: After knowing the weight of each new evidence, perform weighted fusion on the new evidence, m"'(θ 1 )=w 1 'm 1 ” (θ 1 )+w' 2 m″ 2 (θ 1 )+w' 3 m″ 3 (θ 1 ), m"'(θ 2 )=w 1 'm 1 ” (θ 2 )+w' 2 m″ 2 (θ 2 )+w' 3 m″ 3 (θ 2 ), m"'(θ 3 )=w 1 'm 1 ” (θ 3 )+w' 2 m″ 2 (θ 3 )+w' 3 m″ 3 (θ 3 ), so m"'(θ 1 )=0.53,m″'(θ 2 )=0.34,m″'(θ 3 )=0.13.

[0096] Step 10: The fused evidence is subjected to 4 DS fusions, m(θ 1 )=0.9991,m(θ 2 )=0.0009,m(θ 3 )=0.

[0097] When using DS combination, when a certain evidence assigns a basic probability of 0 to a certain event, the final basic probability assignment result will be 0. Even if other evidence assigns a higher probability to the event or new evidence shows that the event has a higher probability, it will not affect the synthesis result. Reference [1] Murphy uses the method of weighted average of evidence. Although it can effectively identify the target and make the correct choice, it does not take into account the correlation between the evidences. It simply averages the evidence and fuses the evidences of different conflicting degrees with the same weight, which makes the fusion process unreasonable. Although references [3-5] effectively identify the target, their accuracy is not as good as that of the method of the present invention. The comparison results are shown in Table 1.

[0098] Table 1: Results and comparison of different combination rules

[0099]

[0100]

[0101] [1]Murphy C K.Combining belief functions when evidence conflicts[J].Decision support systems,2000,29(1):1-9.

[0102] [2] Jiang Wen, Zhang Yu, Xie Chunhe. Classification and fusion method of multi-sensor collaborative detection evidence theory[J]. Navigation Positioning & Timing, 2019, 6(05): 32-37.

[0103] [3]Deng Y, Shi WK, Zhu ZF, et al. Combining belief functions based on distance of evidence[J]. Decision Support Systems, 2004, 38(3): 489-493.

[0104] [4]Zhang Z, Liu T, Chen D, et al. Novel algorithm for identifying and fusing conflicting data in wireless sensor networks [J]. Sensors, 2014, 14(6): 9562-9581.

[0105] [5] Yuan KJ, Xiao FY, Fei LG, et al.Conflict management based on belieffunction entropy in sensor fusion[J].Springerplus,2016,5(1):638..https: / / doi.org / 10.1186 / s40064-016-2205-6

[0106] In summary, the method proposed in the patent of the present invention has a simple structure, a clear process, a reasonable design, and is easy to operate during implementation and use. The present invention performs fuzzy classification on multiple complex evidences, which greatly reduces the confusion of the evidence, reduces the complexity of the target information, and makes the prediction results more accurate and more realistic. The present invention scientifically explains the differences between evidences, which is logical. The present invention uses the degree of membership between evidences as the statistical characteristics of the known evidence classification, which is the premise for fuzzy classification, provides a basis for fuzzy classification, and reflects the actual situation between the evidences. The present invention effectively suppresses the influence of evidence with greater conflicts on the fusion results, so that the results have a better focusing effect.

[0107] In the description of the present invention, it should be noted that directional words, such as the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like, indicating directions and positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and cannot be understood as limiting the specific protection scope of the present invention.

[0108] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged in appropriate circumstances, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0109] Note that the above are only preferred embodiments of the present invention and the principles of the application technology. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention is described in more detail through the above embodiments, the present invention is not limited to the specific embodiments described herein, and may include more other effective embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A weighted conflict evidence fusion method based on fuzzy classification, characterized by: The steps include: Step 1: Obtain the recognition framework of target recognition Θ = {θ1,θ2,θ3,…,θ u }, u is the number of elements in the recognition frame Θ, θ u To identify the type of target, this is a set consisting of u mutually exclusive elements, and there exists θ k and If m(θ k )≥0, then we call θ k Assign m to the focal element of the basic probability on the identification frame Θ; obtain the basic probability assignment of the focal element of the evidence corresponding to the measurement information of multiple sensors, and regard each evidence as a vector. The i-th evidence can be expressed as: i =(m i (θ1),m i (θ2),…,m i (θ r ),…,m i (θ u )), where i = 1, 2, ..., n, n is the total number of evidences, u is the number of elements in the recognition frame Θ, r = 1, 2, ..., u; m i (θ r ) is evidence m i θ r Basic probability assignment of focal elements; Step 2: When the evidence contains multi-subset focal elements, the multi-subset focal elements of the evidence in step 1 are transformed into single-subset focal elements using the interval probability assignment formula based on the combination of the belief function Bel and the plausibility function Pl; the formula is as follows: Calculate the corrected evidence vector m' i =(m' i (θ1),m' i (θ2),…,m' i (θ r )), where m' i (θ r )=P i (θ r ), assigns the basic probability of the modified single subset focal element, Represents the proposition θ u is a true trust measure, representing the upper limit of the trust level, at this time θ k =θ u , is the singleton focal element; Represents the proposition θ u is a non-false plausibility measure, representing the lower limit of the degree of trust. is a multi-subset focal element; Represents all singleton focal elements θ r is the accumulation of true trust measures; Step 3: Find the membership degree r between n pieces of evidence ij , the larger the value, the higher the correlation between the two pieces of evidence, and the smaller the value, the lower the correlation between the two pieces of evidence; according to the evidence vector m' corrected in step 2 i and m' j Find the membership of the i-th evidence to the j-th evidence. The function formula is: where i,j=1,2,…,n,m' j T is the corrected evidence vector m' j The transpose of Step 4: The membership degree r between the n pieces of evidence obtained in step 3 ij Construct a membership matrix R with n rows and n columns, Step 5: From the properties of matrix R, it can be concluded that matrix R is a fuzzy similarity matrix; Step 6: Use the transitive closure method to obtain the fuzzy equivalent matrix R from the fuzzy similarity matrix R; The transitive closure method is: is the definition of transitive closure, e = max(min(a,a),min(b,c)), f=max(min(a,b),min(b,d)), g=max(min(c,a),min(d,c)), h=max(min(c,b),min(d,d)), until we get N and K are any positive integers and N<K. In this case, the fuzzy equivalent matrix of the fuzzy similarity matrix R is Step 7: According to the set threshold λ, the fuzzy equivalent matrix will be obtained Change to Boolean equivalent matrix Boolean equivalence matrix Each column element with a value of 1 is classified into one category, and a fuzzy classification at the λ level is obtained; Step 8: Perform weighted average fusion of evidence belonging to the same category; Step 801: There are l pieces of evidence in the same category. When the similarity between two pieces of evidence is greater, the support between the pieces of evidence is s. ij The higher it is, Where i, j = 1, 2, ..., l, ε is the total number of evidences that need to be calculated for mutual support, and ε = l; Step 802: The mutual support s between l pieces of evidence ij Construct a mutual support matrix S with l rows and l columns, Step 803: Obtain evidence m' from the mutual support matrix S i The final support s in the same category of evidence i , Where i = 1, 2, ..., l; Step 804: The final support of each piece of evidence s i Normalize Where i = 1, 2, ..., l, and the normalized final support is the weight w of each piece of evidence of the same type i =S i , where i = 1, 2, ..., l, weighted average fusion of similar evidence is performed to obtain new evidence; if the same type of evidence contains only one piece of evidence, the weighted average is still the evidence itself; Step 9: The same type of evidence in step 8 is weighted and fused to obtain new evidence. Since the number of evidences in each type of evidence is different, the new evidence fused from multiple pieces of evidence is more reliable than the new evidence fused from a single piece of evidence and plays a greater leading role. Step 901: normalize the final support of the new evidence and combine it with the reliability as the weight of each new evidence, and then normalize the weight; Step 902: Perform weighted averaging of the new evidence using the obtained weights to obtain final evidence; Step 10: Perform n-1 DS fusions on the final evidence, and the obtained evidence is the final result.

2. The weighted conflict evidence fusion method based on fuzzy classification according to claim 1 is characterized by: The DS fusion in step 10 adopts the Dempster combination rule: where θ r and θ u To identify the focal element on the frame Θ, it represents the degree of conflict between the two pieces of evidence. The larger K is, the greater the degree of conflict between the two pieces of evidence is.

3. The weighted conflict evidence fusion method based on fuzzy classification according to claim 1 is characterized by: The step 901 is specifically as follows: if the evidence is finally divided into p categories, then there will be p new pieces of evidence after fusion. It is necessary to consider not only the support degree of each piece of evidence received from other pieces of evidence, but also the reliability of the evidence itself. The support degree and reliability of the evidence are combined to obtain the weight of each piece of evidence in the fused evidence. S i For evidence i "The final support after normalization, T i For evidence i "Reliability; find the weight of evidence W i After that, the weights are normalized 4. The weighted conflict evidence fusion method based on fuzzy classification according to claim 1 is characterized by: The step 902 obtains the final evidence m'': θ r Jiao Yuan is the evidence.

Images