A multi-feature target recognition method

Abstract

The invention discloses a multi-feature target recognition method, belonging to the technical field of target recognition. The method comprises: acquiring feature data of a target to be recognized collected by multiple sensors, wherein the feature data of the target to be recognized comprises sample data to be tested; generating triangular fuzzy numbers according to a target sample data set and the sample data to be tested; generating a neutral intelligence set representation of each target feature under each target type according to the triangular fuzzy number matching of the target sample data set and the sample data to be tested; converting the neutral intelligence set representation of each target feature under each target type into a basic probability assignment; using weighted average to fuse the basic probability assignment of each target feature under each target type to obtain the basic probability assignment of each target type; fusing the basic probability assignment of each target type, and determining the target type to which the target to be recognized belongs according to the fusion result; and outputting the target type to which the target to be recognized belongs. The method can improve the accuracy of target recognition and is used for multi-feature target recognition.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and in particular relates to a multi-feature target recognition method. Background Art

[0002] With the improvement of modern science and technology and production level, automation and intelligence are gradually applied in more and more fields, such as industry, scientific research, medicine, etc. In actual use, the first problem to be solved in the operation of the automation system is the operation object, that is, the application of sensors to achieve target recognition. Therefore, the research on target recognition technology has very important practical significance.

[0003] Target recognition is to analyze the target feature information by using technical processing methods to obtain the qualitative or quantitative properties of the target. Target recognition in the environment can usually be divided into three steps: feature extraction, feature processing, and target classification.

[0004] However, the structure of modern automation systems is complex and is affected by many subjective and objective external factors, which makes the sensor detection information uncertain. In addition, due to the complexity and diversity of modern systems, the recognition accuracy of sensors varies, and the measurement data of different target types are cross-related. Relying on a single sensor to collect target information usually deviates from the actual results, and the accuracy of target recognition is low. Summary of the invention

[0005] In order to solve the problem of low target recognition accuracy in related technologies, the purpose of the present invention is to provide a multi-feature target recognition method that can effectively process the uncertain information detected by the sensor and improve the accuracy of target recognition.

[0006] A multi-feature target recognition method comprises the following steps:

[0007] Step 110: Multiple sensors are placed on m target types F i n target features C j The measured data on the target sample dataset D is generated ij , and measure the samples under n target features to generate the sample data T to be tested j ;

[0008] Step 120, generating a triangular fuzzy number of the target sample data set under each target feature belonging to each target type and a triangular fuzzy number of the sample data to be tested under each target feature;

[0009] Step 130: collect the target sample data into target feature C j In target type F i The triangular fuzzy number under the target feature C jThe triangular fuzzy numbers under the target are matched to generate the neutral set representation of each target feature under each target type;

[0010] Step 140: convert the neutral intelligence set representation of each target feature under each target type into a basic probability assignment;

[0011] Step 150: Use weighted average to fuse the basic probability assignment of each target feature under each target type to obtain the basic probability assignment of each target type;

[0012] Step 160, integrating the basic probability assignments of each target type, and determining the target type to which the target to be identified belongs according to the fusion result;

[0013] Step 170: Output the target type to which the target to be identified belongs.

[0014] Preferably, the method for generating the triangular fuzzy number of the target sample data set under each target feature belonging to each target type is: calculating the target sample data set D ij All the target types F i k samples of the target feature C j The minimum value on average value and maximum value And generate the target sample data set in the target type F i The target feature C j Triangular fuzzy numbers under in, in, The target sample dataset D ij The target type F i k samples of the target feature C j The measured values ​​on , 1≤i≤m, 1≤j≤n, m is the number of target types, n is the number of target features; target sample data set D ij It is the measurement data of n target features under m target types;

[0015] The method for generating the triangular fuzzy number of the sample data to be tested under each target feature is as follows: Calculate the sample data T j In the target feature C j The minimum value on average value and maximum value And generate the sample data to be tested in the target feature C j Triangular fuzzy numbers under Sample data to be tested T j It is the measurement value of the sensor on the sample under n target features.

[0016] Preferably, the method for generating the neutrosophic set representation of each target feature under each target type is:

[0017] Use the first formula to generate the sample data T to be tested j In the target feature C j The target type F i The single-valued neutrosophic set represents a ij =<t ij , f ij , g ij >;

[0018] The first formula is:

[0019]

[0020] in Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number The overlapping area, Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number and target feature C j The above belong to target type F p，p≠i The target sample data set triangular fuzzy number The union of the overlapping areas of ;

[0021] According to the sample data T j In the target feature C j The target type F i The single-valued neutrino set representation generates the sample data T to be tested j Neutrosophic set representation of n target features under m target types.

[0022] Preferably, step 140 comprises the following steps:

[0023] Step 141: According to the sample data T to be tested j The neutral set representation of n target features under m target types is normalized using the second formula, which is:

[0024]

[0025] Step 142: Based on the sample data T to be tested jThe normalized processing result of the neutrinodic set representation of n target features under m target types is used to generate the sample data T to be tested according to the third formula j In the target type F i The target feature C j The basic probability assignment is as follows, where the third formula is:

[0026]

[0027]

[0028] Proposition Indicates that the conditions are met All elements of F p，p≠i A collection of Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number and target feature C j The target type F p，p≠i The target sample data set triangular fuzzy number There is no overlap. Proposition Θ\B represents the set of elements generated by removing all elements in proposition B from the elements contained in proposition Θ. The target type is represented by F = {F1, F2, ..., F m}, the target feature is represented by C = {C1, C2, ..., C n}, the identification framework of the system is expressed as Θ = {F1, F2, ..., F m}, its power set Contains 2 N A proposition.

[0029] Preferably, step 150 comprises the following steps:

[0030] Step 151: According to the sample data T to be tested j The neutrinocular set representation of n target features under m target types is used to calculate the sample data T to be tested using the fourth formula j In the target type F i The target feature C j The entropy of the neutrosophic set under the condition is expressed as follows, and the target type F is generated according to the fifth formula i Lower target feature C j The weight of

[0031] The fourth formula is:

[0032] H ij =-[(t ij +g ij)log2(t ij +g ij )+(f ij +g ij )log2(f ij +g ij )];

[0033] The fifth formula is:

[0034]

[0035] Step 152: The sample data to be tested is stored in the target type F i The n basic probability assignments on the n target features are fused using weighted average, and the weights of each target feature in each target type are used to generate the target type F according to the sixth formula i The weighted average basic probability assignment of proposition A is used to fuse the target type F using the seventh formula i The n-1 weighted average basic probability assignments under generate the target type F i The basic probability assignment of; the sixth formula is:

[0036]

[0037] The seventh formula is:

[0038]

[0039]

[0040] Preferably, step 160 comprises the following steps:

[0041] Step 161: According to the basic probability assignments of each target type, the basic probability assignments of target types F1 and F2 are fused using the eighth formula, and the fusion result is fused with the basic probability assignment of target type F3, thereby sequentially fusion of the basic probability assignments of each target type to generate a decision basic probability assignment m; wherein the eighth formula is:

[0042]

[0043] Step 162: Generate target type F according to the ninth formula i The probability of each target type is sorted, and the target type corresponding to the largest probability is determined as the target type to which the target to be identified belongs; the ninth formula is:

[0044] |A| represents the number of elements of proposition A.

[0045] Preferably, the target type to which the target to be identified belongs is output in a display manner or a voice broadcast manner.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The method of the present invention has simple steps, reasonable design, and is easy to implement and use.

[0048] 2. The present invention uses triangular fuzzy numbers to represent target sample information and generates a neutral set representation of the sample to be tested, which can effectively handle the uncertainty of sensor detection information;

[0049] 3. The present invention fuses the measurement information of multiple target features through an information fusion method, thereby improving the accuracy of target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a multi-feature target recognition method shown in this embodiment. DETAILED DESCRIPTION

[0051] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

[0052] The structure of modern automation systems is complex and is affected by many subjective and objective external factors, which makes the sensor detection information uncertain. For this fuzzy phenomenon of diagnostic information, fuzzy mathematics methods need to be used to deal with it. It is necessary to consider multiple factors at the same time and identify the target type from multiple target features based on multi-sensor measurements. At the same time, information fusion technology is used to fuse multi-sensor information to achieve a more accurate and comprehensive understanding of the target. Therefore, this embodiment identifies the target type based on the characteristic information of the sample to be tested detected by the sensor, integrating the neutral intelligence set theory and information fusion technology, which can better handle the uncertainty of sensor information and improve the accuracy of target recognition.

[0053] This embodiment shows a multi-feature target recognition method, such as Figure 1 As shown, the following steps are included:

[0054] Step 110: Multiple sensors are placed on m target types F i n target features C j The measured data on the target sample dataset D is generated ij , and measure the samples under n target features to generate the sample data T to be tested j .

[0055] In actual use, sensors are used to collect m types of targets F1, F2, ..., F i , …, F m Among n target features C1, C2, ..., C j, …, C n The measured data on the target sample dataset D is generated ij , and measure the samples under n target features to generate the sample data T to be tested j , i = 1, 2, ..., m, j = 1, 2, ..., n, the identification framework of the system is expressed as Θ = {F1, F2, ..., F m}, its power set Contains 2 N A proposition.

[0056] Step 120: Generate triangular fuzzy numbers of the target template data set under each target feature belonging to each target type and triangular fuzzy numbers of the sample data to be tested under each target feature.

[0057] Multiple target feature data can more fully reflect the target situation, thereby improving the accuracy of target recognition; secondly, triangular fuzzy numbers are relatively less susceptible to interference and have good stability. Therefore, according to D ij and T j Generate the triangular fuzzy number of the target template data set under each target feature belonging to each target type and the triangular fuzzy number of the sample data to be tested under each target feature.

[0058] The method for generating the triangular fuzzy number of the target sample data set under each target feature belonging to each target type is as follows: Calculate the target sample data set D ij All the target types F i k samples of the target feature C j The minimum value on average value and maximum value And generate the target sample data set in the target type F i The target feature C j Triangular fuzzy numbers under in, in, The target sample dataset D ij The target type F i k samples of the target feature C j The measured values ​​on , 1≤i≤m, 1≤j≤n, m is the number of target types, n is the number of target features; target sample data set D ij It is the measurement data of n target features under m target types.

[0059] The method for generating the triangular fuzzy number of the sample data to be tested under each target feature is as follows: Calculate the sample data T j In the target feature C j The minimum value on average value and maximum value And generate the sample data to be tested in the target feature C j Triangular fuzzy numbers under Sample data to be tested T j It is the measurement value of the sensor on the sample under n target features.

[0060] Step 130: collect the target sample data into target feature C j In target type F i The triangular fuzzy number under the target feature C j The triangular fuzzy numbers under the target type are matched to generate the neutral set representation of each target feature under each target type.

[0061] Among them, the method of generating the neutrosophic set representation of each target feature under each target type is:

[0062] Use the first formula to generate the sample data T to be tested j In the target feature C j The target type F i The single-valued neutrosophic set represents a ij =<t ij , f ij , g ij >;

[0063] The first formula is:

[0064]

[0065] in Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number The overlapping area, Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number and target feature C j The above belong to target type F p，p≠i The target sample data set is triangular fuzzy numbers The union of the overlapping areas.

[0066] According to the sample data T j In the target feature C j The single-valued neutrino set representation of the target type Fi is used to generate the sample data T to be tested. jNeutrosophic set representation of n target features under m target types.

[0067] Step 140: Convert the neutrosophic set representation of each target feature under each target type into a basic probability assignment. Specifically, the following steps are included:

[0068] Step 141: According to the sample data T to be tested j The neutral set representation of n target features under m target types is normalized using the second formula, which is:

[0069]

[0070] Step 142: Based on the sample data T to be tested j The normalized processing result of the neutrinodic set representation of n target features under m target types is used to generate the sample data T to be tested according to the third formula j In the target type F i The target feature C j The basic probability assignment is as follows, where the third formula is:

[0071]

[0072]

[0073] Proposition Indicates that the conditions are met All elements of F p，p≠i A collection of Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set is triangular fuzzy numbers and target feature C j The target type F p，p≠i The target sample data set is triangular fuzzy numbers There is no overlap. Proposition Θ\B represents the set of elements generated by removing all elements in proposition B from the elements contained in proposition Θ. The target type is represented by F = {F1, F2, ..., F m}, the target feature is represented by C = {C1, C2, ..., C n}, the identification framework of the system is expressed as Θ = {F1, F2, ..., F m}, its power set Contains 2 N A proposition.

[0074] Step 150: Use weighted average to fuse the basic probability assignment of each target feature under each target type to obtain the basic probability assignment of each target type. Specifically, the following steps are included:

[0075] Step 151: According to the sample data T to be tested j The neutrinocular set representation of n target features under m target types is used to calculate the sample data T to be tested using the fourth formula j In the target type F i The target feature C j The entropy of the neutrosophic set under the condition is expressed as follows, and the target type F is generated according to the fifth formula i Lower target feature C j The fourth formula is H ij =-[(t ij +g ij )log2(t ij +g ij )+(f ij +g ij )log2(f ij +g ij )];

[0076] The fifth formula is:

[0077]

[0078] Step 152: The sample data to be tested is stored in the target type F i The n basic probability assignments on the n target features are fused using weighted average, and the weights of each target feature in each target type are used to generate the target type F according to the sixth formula i The weighted average basic probability assignment of proposition A is used to fuse the target type F using the seventh formula i The n-1 weighted average basic probability assignments under generate the target type F i The basic probability assignment of; the sixth formula is:

[0079]

[0080] The seventh formula is:

[0081]

[0082]

[0083] Step 160: Fuse the basic probability assignments of each target type, and determine the target type to which the target to be identified belongs according to the fusion result. Specifically, the following steps are included:

[0084] Step 161: According to the basic probability assignments of each target type, the basic probability assignments of target types F1 and F2 are fused using the eighth formula, and the fusion result is fused with the basic probability assignment of target type F3, thereby sequentially fusion of the basic probability assignments of each target type to generate a decision basic probability assignment m; wherein the eighth formula is:

[0085]

[0086] Step 162: Generate target type F according to the ninth formula i The probability of each target type is sorted, and the target type corresponding to the largest probability is determined as the target type to which the target to be identified belongs; the ninth formula is:

[0087] |A| represents the number of elements of proposition A.

[0088] Step 170: Output the target type to which the target to be identified belongs.

[0089] The target type to which the target to be identified belongs can be outputted through display mode, voice broadcast mode, etc.

[0090] The multi-feature target recognition method and device provided by the present application obtain feature data of the target to be recognized collected by multiple sensors; generate triangular fuzzy numbers based on the target sample data set and the sample data to be tested; generate a neutral intelligence set representation of each target feature under each target type based on the triangular fuzzy number matching of the target sample data set and the sample data to be tested; convert the neutral intelligence set representation of each target feature under each target type into a basic probability assignment; use weighted average to fuse the basic probability assignment of each target feature under each target type to obtain the basic probability assignment of each target type; finally, fuse the basic probability assignment of each target type, and determine the target type to which the target to be recognized belongs based on the fusion result; then output the target type to which the target to be recognized belongs. Compared with the existing technology, the method and device can effectively process the uncertain information detected by the sensor and improve the accuracy of target recognition.

Claims

1. A multi-feature target recognition method, characterized in that The following steps are involved: Step 110: Multiple sensors are placed on m target types F i n target features C j The measured data on the target sample dataset D is generated ij , and measure the samples under n target features to generate the sample data T to be tested j ; Step 120, generating a triangular fuzzy number of the target sample data set under each target feature belonging to each target type and a triangular fuzzy number of the sample data to be tested under each target feature; Step 130: collect the target sample data into target feature C j In target type F i The triangular fuzzy number under the target feature C j The triangular fuzzy numbers under are matched to generate the neutral set representation of each target feature under each target type. The method is: Use the first formula to generate the sample data T to be tested j In the target feature C j The target type F i The single-valued neutrosophic set represents a ij = <t ij , f ij , g ij > The first formula is: in Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number The overlapping area, Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number and target feature C j The above belong to target type F p，p≠i The target sample data set triangular fuzzy number The union of the overlapping areas of ; According to the sample data T j In the target feature C j The target type F i The single-valued neutrino set representation generates the sample data T to be tested j Neutrosophic set representation of n target features under m target types; Step 140: convert the neutral intelligence set representation of each target feature under each target type into a basic probability assignment; Step 150: Use weighted average to fuse the basic probability assignment of each target feature under each target type to obtain the basic probability assignment of each target type; Step 160, integrating the basic probability assignments of each target type, and determining the target type to which the target to be identified belongs according to the fusion result; Step 170: Output the target type to which the target to be identified belongs.

2. A multi-feature target recognition method according to claim 1, characterized in that The method for generating the triangular fuzzy number of the target sample data set under each target feature belonging to each target type is as follows: Calculate the target sample data set D ij All the target types F i k samples of the target feature C j The minimum value on average value and maximum value And generate the target sample data set in the target type F i The target feature C j Triangular fuzzy numbers under in, in, The target sample dataset D ij The target type F i k samples of the target feature C j The measured values ​​on , 1≤i≤m, 1≤j≤n, m is the number of target types, n is the number of target features; target sample data set D ij It is the measurement data of n target features under m target types; The method for generating the triangular fuzzy number of the sample data to be tested under each target feature is as follows: Calculate the sample data T j In the target feature C j The minimum value on average value and maximum value And generate the sample data to be tested in the target feature C j Triangular fuzzy numbers under Sample data to be tested T j It is the measurement value of the sensor on the sample under n target features.

3. A multi-feature target recognition method according to claim 1, characterized in that Step 140 includes the following steps: Step 141: According to the sample data T to be tested j The neutral set representation of n target features under m target types is normalized using the second formula, which is: Step 142: Based on the sample data T to be tested j The normalized processing result of the neutrinodic set representation of n target features under m target types is used to generate the sample data T to be tested according to the third formula j In the target type F i The target feature C j The basic probability assignment is as follows, where the third formula is: Proposition Indicates that the conditions are met All elements of F p，p≠i A collection of Represents the target feature C j The triangular fuzzy number of the sample data to be tested on With the target feature C j The target type F i The target sample data set triangular fuzzy number and target feature C j The target type F p，p≠i The target sample data set triangular fuzzy number There is no overlap. Proposition Θ\B represents the set of elements generated by removing all elements in proposition B from the elements contained in proposition Θ. The target type is represented by F = {F1, F2, ..., F m }, the target feature is represented by C = {C1, C2, ..., C n }, the identification framework of the system is expressed as Θ = {F1, F2, ..., F m }, its power set Contains 2 N A proposition.

4. A multi-feature target recognition method according to claim 3, characterized in that Step 150 includes the following steps: Step 151: According to the sample data T to be tested j The neutrinocular set representation of n target features under m target types is used to calculate the sample data T to be tested using the fourth formula j In the target type F i The target feature C j The entropy of the neutrosophic set under the condition is expressed as follows, and the target type F is generated according to the fifth formula i Lower target feature C j The weight of The fourth formula is: H ij =-[(t ij +g ij )log2(t ij +g ij )+(f ij +g ij )log2(f ij +g ij )]; The fifth formula is: Step 152: The sample data to be tested is stored in the target type F i The n basic probability assignments on the n target features are fused using weighted average, and the weights of each target feature in each target type are used to generate the target type F according to the sixth formula i The weighted average basic probability assignment of proposition A is used to fuse the target type F using the seventh formula i The n-1 weighted average basic probability assignments under generate the target type F i The basic probability assignment of; the sixth formula is: The seventh formula is:

5. A multi-feature target recognition method according to claim 4, characterized in that Step 160 includes the following steps: Step 161: According to the basic probability assignments of each target type, the basic probability assignments of target types F1 and F2 are fused using the eighth formula, and the fusion result is fused with the basic probability assignment of target type F3, thereby sequentially fusion of the basic probability assignments of each target type to generate a decision basic probability assignment m; wherein the eighth formula is: Step 162: Generate target type F according to the ninth formula i The probability of each target type is sorted, and the target type corresponding to the largest probability is determined as the target type to which the target to be identified belongs; the ninth formula is: |A| represents the number of elements of proposition A.

6. A multi-feature target recognition method according to claim 1, characterized in that The target type of the target to be identified is output in a display mode or a voice broadcast mode.

Images