DS evidence body trust degree redistribution identification method under high conflict information

Abstract

The invention provides a DS evidence body trust degree redistribution identification method under high conflict information, and relates to the technical field of data processing in identification problems, and the method comprises the following steps: 1, constructing a plurality of evidence bodies according to identification information obtained in a multi-sensor system, calculating a BJS divergence matrix of each evidence body, and obtaining the support degree of each evidence body; 2, calculating the total confidence degree of each evidence body according to the obtained support degree; 3, calculating a weight parameter of a single element set in each evidence body, and calculating to obtain a single element confidence interval; 4, carrying out distribution according to a multi-element subset trust distribution formula, and carrying out trust redistribution on a single-element subset; 5, DS combination is carried out on the redistributed evidence bodies, and an identification result is output. According to a credibility redistribution strategy provided for DS evidence theory application in a high-conflict information environment, the influence of conflict data interference on the identification process can be effectively reduced, and the identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of multi-sensor information fusion, and in particular to a DS evidence body trust redistribution identification method under high conflict information. Background Art

[0002] As an important tool for dealing with uncertain reasoning, DS evidence theory occupies a pivotal position in the field of multi-source information fusion. Its uniqueness lies in the fact that by introducing the basic probability distribution function (BPA), it can flexibly and meticulously characterize the degree of support of evidence for different propositions, and then organically integrate the information of multiple evidence sources to output relatively reasonable decision results in a complex information environment. Taking the intelligent security system as an example, the visual images captured by the camera, the thermal signals perceived by the infrared sensor, the audio information collected by the sound sensor, etc. can all be used as independent evidence. By integrating these evidences with the help of DS evidence theory, it is possible to accurately determine whether there are abnormal targets in the monitoring area.

[0003] However, when placed in a high-conflict information environment, the DS evidence theory reveals obvious shortcomings. The classic DS evidence theory uses the Dempster combination rule to fuse evidence. When the degree of conflict between evidence is low, this rule can produce reasonable results. However, once the degree of conflict between evidence increases, the fusion results are often very different from the actual situation, and may even lead to completely wrong conclusions. This is because the Dempster combination rule simply and crudely distributes all conflicting information to non-conflicting items, and fails to handle conflicting information reasonably and effectively. For example, in a medical diagnosis scenario, for the same disease, if two detection methods give contradictory results, fusing them according to the combination rule of the classic DS evidence theory is likely to mislead doctors into making wrong diagnostic decisions.

[0004] In practical application scenarios such as situation assessment in fault diagnosis of industrial production processes, the complexity and diversity of high-conflict information environments continue to increase, and existing improvement methods are increasingly unable to meet actual needs. Therefore, there is an urgent need for a new identification method based on DS evidence theory and combined with a trust redistribution strategy. By scientifically and rationally redistributing trust, the interference caused by high-conflict information can be effectively resolved, and the identification accuracy and reliability of DS evidence theory in complex high-conflict environments can be greatly improved, thus meeting the growing urgent need for efficient and accurate decision-making in various fields. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention proposes a DS evidence body trust redistribution identification method under high conflict information. Starting from the application scenario of DS evidence theory in a high conflict information environment, a trust redistribution strategy is constructed based on tools such as the basic probability distribution function (BPA) and trust interval in DS evidence theory. By redistributing the trust of the identification information directly obtained by the sensor, the impact of conflicting data when applying DS evidence theory is reduced, which not only enhances the anti-conflict ability of information fusion, but also improves the identification accuracy.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] Step 1: construct multiple evidence bodies according to the identification information obtained from the multi-sensor system, calculate the BJS divergence matrix of each evidence body, and obtain the support of each evidence body;

[0008] Step 2: Calculate the total confidence of each body of evidence based on the obtained support;

[0009] Step 3: Calculate the weight parameter of the single element set in each evidence body, and then calculate the single element confidence interval;

[0010] Step 4: Allocate according to the multi-element subset trust allocation formula, and redistribute the trust of the single-element subset;

[0011] Step 5: Perform DS combination on the redistributed evidence.

[0012] Furthermore, the step 1 includes the following contents:

[0013] For a certain A multi-sensor system with The evidence information obtained in the identification task sampling of the identified object is:

[0014]

[0015] in, , indicating the sensor Identification sampling information;

[0016] Assume that the identification object set is , then the corresponding DS evidence theory application framework is:

[0017]

[0018] The object set of the trust redistribution strategy is:

[0019]

[0020] For simplicity, define the evidence body mapping is the trust value of the corresponding element in the object set. Represents the elements in the evidence body The confidence value in the body of evidence. For example, Corresponding evidence body mapping The evidence body mapping function here is actually the basic probability assignment function (BPA) in DS evidence theory, which satisfies the requirement that the sum of the trust levels of all elements is one.

[0021] Calculate the BJS divergence between each evidence body and construct the BJS divergence matrix:

[0022]

[0023] Among them, the calculation formula of BJS divergence is:

[0024]

[0025] According to the calculation formula of BJS divergence, divergence is a characteristic quantity that increases monotonically with the increase of conflict between evidence bodies, and:

[0026] The support degree of the evidence body is measured based on the conflict degree between the evidence body and other evidence bodies, and the evidence body is calculated. Average Divergence:

[0027]

[0028] Calculate support:

[0029]

[0030] Furthermore, the step 2 includes the following contents:

[0031] Defining overall confidence is the sum of the confidence intervals of all single elements, that is: ;

[0032] The support of the evidence obtained based on the average divergence calculation is a reflection of the degree of trust in the identification information obtained by each sensor. According to the support of the evidence, the degree to which each evidence is changed in the trust redistribution strategy is calculated, and the total confidence of each evidence is calculated. :

[0033]

[0034] The total confidence and the number of elements in the identification object set , Number of sensors and the degree of support of the body of evidence About. and Under certain circumstances, with Increase, It decreases, indicating that the data trust level of this piece of evidence is higher in the redistribution strategy.

[0035] Furthermore, the step 3 includes the following contents:

[0036] According to the total confidence of each evidence body calculated in step 2, the trust redistribution operation can be performed on each evidence body. According to the trust redistribution object set defined in step 1, the following definitions are made for simplicity:

[0037] definition is the trust value of each single element set in the evidence body, that is: ;

[0038] definition For missing elements only A multi-element collection, namely: ;

[0039] definition is the trust value of each multi-element set in the evidence body, that is: ;

[0040] In the calculation redistribution strategy, a single element The weights of and their assigned confidence intervals:

[0041]

[0042]

[0043] Through relevant calculations, the total confidence is distributed according to the weights based on the information of the original evidence body, so that the evidence body with high support retains the original information, and the evidence body with low support gradually averages the distribution of trust values.

[0044] Furthermore, the step 4 includes the following contents:

[0045] According to the total confidence and confidence interval of a single element calculated in steps 2 and 3, Redistribute the trust value:

[0046]

[0047] According to the allocation strategy of multi-element subsets, the sum of the trust values ​​of single-element subsets can be calculated as:

[0048]

[0049] Then the remaining trust values ​​are calculated according to the single element trust values ​​of the original evidence To distribute by rights:

[0050]

[0051] Complete the original evidence Redistribution of trust.

[0052] Furthermore, the step 5 includes the following contents:

[0053] Dempster combination is performed on each piece of evidence after trust redistribution. The combination rule can be expressed as , defined as follows:

[0054]

[0055] in,

[0056] By performing Dempster combination on all evidence bodies, a fused evidence body can be obtained, and the one with the largest trust value is taken as the identification result.

[0057] Compared with the prior art, the present invention adopts the above technical solution, and has the following beneficial effects:

[0058] 1. Starting from the identification problem in a high-conflict information environment, the present invention develops a trust redistribution strategy to address the problems existing in the DS evidence theory. Different from the existing DS evidence theory improvement scheme, this strategy adopts a trust readjustment of all evidence bodies, so that the impact of conflicting data is reduced rather than not adopted at all, that is, there is no need to strictly judge the quality of sensors or adopt weights, which reduces the difficulty of applying the identification method. Therefore, the present invention has strong practical application significance.

[0059] 2. The identification method developed by the present invention based on DS evidence theory and trust redistribution strategy reduces the impact of the conflict environment by reducing the trust distribution difference between high-conflict evidence, which not only enhances the identification process's resistance to information conflicts, but also improves the accuracy of the identification results. Experiments have shown that the identification success rate of the present invention is more accurate than other DS evidence theory improvement schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the method of the present invention;

[0061] Figure 2 This is a comparison chart of the recognition accuracy of three recognition schemes under different information conflict environments. DETAILED DESCRIPTION

[0062] The present invention proposes a DS evidence body trust redistribution identification method under high conflict information. When implemented, the following steps should be followed:

[0063] (1) For a A multi-sensor system with In the sampling of identification tasks for individual identification objects, the different information obtained needs to be processed, and the size of the evidence set and the information of the identification object need to be checked and unified.

[0064] After completing the preprocessing of the adopted data, the evidence body information is obtained as follows:

[0065]

[0066] in, , indicating that the sensor after preprocessing Identification sampling information;

[0067] Assume that the identification object set is , then the corresponding DS evidence theory application framework is:

[0068]

[0069] The object set of the trust redistribution strategy is:

[0070]

[0071] For simplicity, define the evidence body mapping is the trust value of the corresponding element in the object set. Represents the elements in the evidence body The confidence value in the body of evidence. For example, Corresponding evidence body mapping

[0072] Then calculate the BJS divergence between each evidence body and construct the BJS divergence matrix:

[0073]

[0074] Among them, the calculation formula of BJS divergence is:

[0075]

[0076] According to the calculation formula of BJS divergence, divergence is a characteristic quantity that increases monotonically with the increase of conflict between evidence bodies, and:

[0077] The support degree of the evidence body is measured based on the conflict degree between the evidence body and other evidence bodies, and the evidence body is calculated. Average Divergence:

[0078]

[0079] Calculate support:

[0080]

[0081] Consider the problem of identifying target types through multiple sensors. The type of object to be identified is a fixed-wing drone, and the evidence is provided by three different sensors. The initial evidence is initially processed by decision probability conversion to obtain BBAs of uniform size, such as The identification framework consists of three objects, namely the fixed-wing drone that we want to identify in this simulation experiment, and the unmanned helicopter and quad-rotor drone used as interference factors, namely the identification framework. {Fixed-wing drone, unmanned helicopter, quad-rotor drone}. To define in the recognition framework On An independent BBA.

[0082] Table 1 BBAs based on multi-sensor target recognition

[0083]

[0084] (2) Define the total confidence is the sum of the confidence intervals of all single elements, that is: ;

[0085] The support of the evidence body obtained based on the average divergence calculation , is a reflection of the degree of trust in the identification information obtained by each sensor. According to the degree of support for the evidence body, the degree to which each evidence body is changed in the trust redistribution strategy is calculated, and the total confidence of each evidence body is calculated. :

[0086]

[0087] The total confidence and the number of elements in the identification object set , Number of sensors and the degree of support of the body of evidence About. and Under certain circumstances, with Increase, It decreases, indicating that the data trust level of this piece of evidence is higher in the redistribution strategy.

[0088] Taking α=0.1 as an example, we calculate the Scatter matrix, and calculate the support of each evidence body , the process is as follows:

[0089] The BJS divergence matrix is ​​as follows:

[0090]

[0091] The average BJS divergence vector is:

[0092]

[0093] The support vector is:

[0094]

[0095] (3) Based on the total confidence of each evidence body calculated in step 2, the trust redistribution operation can be performed on each evidence body. According to the trust redistribution object set defined in step 1, the following definitions are made for simplicity:

[0096] definition is the trust value of each single element set in the evidence body, that is: ;

[0097] definition For missing elements only A multi-element collection, namely: ;

[0098] definition is the trust value of each multi-element set in the evidence body, that is: ;

[0099] In the calculation redistribution strategy, a single element The weights of and their assigned confidence intervals:

[0100]

[0101]

[0102] Through relevant calculations, the total confidence is distributed according to the weights based on the information of the original evidence body, so that the evidence body with high support retains the original information, and the evidence body with low support gradually averages the distribution of trust values.

[0103] The trust redistribution strategy requires the trust distribution object to be changed from the original identification object set , redistribute to the collection The calculation result of this step reflects the tendency of distributing trust for different evidence bodies according to the global support of the evidence body and the original trust distribution.

[0104] Calculate the confidence of each body of evidence for the case :

[0105] Confidence vector:

[0106] Compute single-element confidence intervals for each body of evidence for a case:

[0107] Single-element confidence interval matrix:

[0108]

[0109] (4) According to the total confidence and confidence interval of a single element calculated in steps 2 and 3, Redistribute the trust value:

[0110]

[0111] After completing the trust distribution of the multi-element subsets, the sum of the trust values ​​of the single-element subsets can be calculated for:

[0112]

[0113] Then the remaining trust values ​​are calculated according to the single element trust values ​​of the original evidence To distribute by rights:

[0114]

[0115] Complete the original evidence The existing identification information is transformed from the original evidence body information matrix:

[0116]

[0117] Expands to:

[0118]

[0119] The trust levels of the cases are assigned to multiple element subsets, and the results are shown in the following table:

[0120] Table 2 Trust distribution of the evidence body to the multi-element subset after redistribution

[0121]

[0122] The trust degree is assigned to the single element subset of the case, and the assignment results are shown in the following table:

[0123] Table 3 Trust distribution of the evidence body to the single element subset after redistribution

[0124]

[0125] (5) Perform Dempster combination on each piece of evidence after trust redistribution. The combination rule can be expressed as , defined as follows:

[0126]

[0127] in,

[0128] By performing Dempster combination on all evidence bodies, a fused evidence body can be obtained.

[0129] Perform DS combination on the cases to obtain the final identification data:

[0130] Table 4. UAV target type recognition results

[0131]

[0132] The trust in the obtained evidence is the final identification data after integrating all sensor information. Its trust distribution is more accurate than the original evidence distribution. The one with the largest trust value is taken as the identification result.

[0133] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying trust redistribution of DS evidence under high conflict information, characterized in that: The following steps are involved: Step 1: construct multiple evidence bodies according to the identification information obtained from the multi-sensor system, calculate the BJS divergence matrix of each evidence body, and obtain the support of each evidence body; Step 2: Calculate the total confidence of each body of evidence based on the obtained support; Step 3: Calculate the weight parameter of the single element set in each evidence body, and then calculate the single element confidence interval; Step 4: Allocate according to the multi-element subset trust allocation formula, and redistribute the trust of the single-element subset; Step 5: Perform DS combination on the redistributed evidence.

2. According to the method for redistributing the trust of DS evidence under high conflict information in claim 1, it is characterized in that: Step 1 includes the following: For a certain A multi-sensor system with The evidence information obtained in the identification task sampling of the identified object is: in, , indicating the sensor Identification sampling information; Assume that the identification object set is , then the corresponding DS evidence theory application framework is: The object set of the trust redistribution strategy is: For simplicity, define the evidence body mapping is the trust value of the corresponding element in the object set, for example Represents the elements in the evidence body The confidence value in the body of evidence, for example, Corresponding evidence body mapping Then calculate the BJS divergence between each evidence body and construct the BJS divergence matrix: Among them, BJS divergence is a representation of the conflict relationship between two evidence bodies, and its specific calculation formula is as follows: According to the calculation formula of BJS divergence, divergence is a characteristic quantity that increases monotonically with the increase of conflict between evidence bodies, and: The support degree of the evidence body is measured based on the conflict degree between the evidence body and other evidence bodies, and the evidence body is calculated. Average divergence : Calculate support : 。 3. According to the method for redistributing the trust of DS evidence under high conflict information in claim 1, it is characterized in that: Step 2 includes the following: Defining overall confidence is the sum of the confidence intervals of all single elements, that is: ; The support of the evidence body obtained based on the average divergence calculation is a reflection of the degree of trust in the identification information obtained by each sensor. According to the degree of support of the evidence body, the degree to which each evidence body is changed in the trust redistribution strategy is calculated, and the total confidence of each evidence body is calculated. : The total confidence and the number of elements in the identification object set , Number of sensors and the degree of support of the body of evidence about, in and Under certain circumstances, with Increase, It decreases, indicating that the data trust level of this piece of evidence is higher in the redistribution strategy.

4. According to claim 1, a DS evidence trust redistribution identification method under high conflict information is characterized in that: Step 3 includes the following: According to the total confidence of each evidence body calculated in step 2, the trust redistribution operation can be performed on each evidence body. If no specific statement is made later, it is all for the evidence body. To perform the operation, according to the set of objects for trust redistribution defined in step 1, the following definitions are made for simplicity: definition is the trust value of each single element set in the evidence body, that is: ; definition For missing elements only A multi-element collection, namely: ; definition is the trust value of each multi-element set in the evidence body, that is: ; In the calculation redistribution strategy, a single element The weights of and their assigned confidence intervals: Through relevant calculations, the total confidence is weighted according to the information of the original evidence body, so that the evidence body with high support retains the original information, and the evidence body with low support gradually averages the distribution of trust values.

5. According to claim 1, a DS evidence trust redistribution identification method under high conflict information is characterized in that: Step 4 includes the following: According to the total confidence and confidence interval of a single element calculated in steps 2 and 3, Redistribute the trust value: According to the allocation strategy of multi-element subsets, the sum of the trust values ​​of single-element subsets can be calculated as: Then the remaining trust values ​​are calculated according to the single element trust values ​​of the original evidence To distribute by rights: Complete the original evidence Redistribution of trust.

6. According to claim 1, a DS evidence trust redistribution identification method under high conflict information is characterized in that: Step 5 includes the following: Dempster combination is performed on each piece of evidence after trust redistribution. The combination rule can be expressed as , defined as follows: in, By performing Dempster combination on all evidence bodies, a fused evidence body can be obtained, and the one with the largest trust value is taken as the identification result for output.

Images