Iterative optimization method for acoustic-magnetic composite intersection feature confidence

Abstract

The invention discloses an iterative optimization method for acoustic-magnetic composite rendezvous characteristic confidence, which comprises the following steps of: 1, generating an ideal n-dimensional rendezvous characteristic fusion curve Y according to electromagnetic and acoustic composite n-dimensional rendezvous characteristics; 2, giving a confidence initial value; 3, inputting a group of learning data; 4, converting the n-dimensional rendezvous feature learning data S into an n-dimensional rendezvous process identification support degree; step 5, solving a fuzzy measure; 6, calculating a fusion value; 7, calculating an allowable value; 8, solving a fusion result; step 9, skipping to step 6 until fusion of the group of learning data is completed, and obtaining a fusion result; step 10, smoothing the fusion result; step 11, calculating a target function E, judging whether the E is converged to a minimum value, if the E is converged, stopping iteration, and outputting an optimization value Cbest, otherwise, updating the confidence coefficient, and skipping to step 5; 12, skipping to the step 3 until N groups of learning data are all optimized, and outputting a comprehensive optimization result # imgabs0 #

Description

Technical Field

[0001] The present invention relates to the field of multi-physical field characteristic confidence iterative optimization algorithms, and particularly to an iterative optimization method for the confidence of acoustic-magnetic composite intersection characteristics. Background Art

[0002] In the process of detecting and identifying specific targets such as sunken ships underwater, in order to improve the detection and positioning accuracy of the target, multiple physical fields are usually used for composite detection. Due to the different characteristics of seawater and air media, the physical fields currently available for detecting and identifying in the underwater near-field intersection process mainly include electromagnetic fields, sound fields, etc. Due to different detection mechanisms, there are significant differences in the intersection characteristics of electromagnetic fields and sound fields.

[0003] The intersection characteristics of underwater electromagnetic detection mainly include electromagnetic field strength and phase information in different directions, etc., while the intersection characteristics of underwater acoustic detection mainly include the target azimuth calculated from the active acoustic echo signal, and the distance information calculated from the target noise intensity, etc. The intersection characteristics of the above different physical fields can theoretically accurately reflect the entire dynamic intersection process and help the detection system achieve the detection and positioning of the target. However, affected by various adverse factors such as the immaturity of sensor technology, the unreliability of signal processing technology, and the inaccuracy of feature extraction and recognition, there must be differences in the confidence of the actual intersection characteristics of electromagnetic fields and sound fields, and even the confidence of different intersection characteristics of the same physical field will also be different.

[0004] Obviously, the inconsistency of the confidence of multiple intersection characteristics of composite physical fields will bring difficulties to the fusion of multi-physical field information, and unreasonable confidence values will inevitably reduce the accuracy of information fusion. Therefore, how to set reasonable feature confidence is of great significance for improving the accuracy of multi-physical field information fusion. Summary of the Invention

[0005] To solve the problems raised in the above background art, the technical solution adopted by the present invention is as follows:

[0006] An iterative optimization method for the confidence of acoustic-magnetic composite intersection characteristics, comprising the following steps:

[0007] Step 1: Generate an ideal n-dimensional intersection characteristic fusion curve Y according to the electromagnetic and acoustic composite n-dimensional intersection characteristics;

[0008] Step 2: Given the initial confidence value C = {C1, C2,... C n} of the electromagnetic and acoustic composite n-dimensional intersection characteristics;

[0009] Step 3: Input a set of electromagnetic and acoustic composite n-dimensional intersection characteristic learning data S = {S1, S2,... S n}, and the length of each feature learning data is m;

[0010] Step 4: According to the mapping relationship between the electromagnetic and acoustic composite n-dimensional intersection characteristics and the intersection process, convert the n-dimensional intersection characteristic learning data S into the n-dimensional intersection process recognition support X = {X1, X2, … X n};

[0011] Step 5: According to the current confidence level, solve the fuzzy measures of the 2 n -n-1 feature combinations {A i} (i = 1, 2, … 2 n -n-1) of the electromagnetic and acoustic composite n-dimensional intersection characteristics

[0012] Step 6: Use the fusion function T to calculate the fusion value T i of the j-th (j = 1, 2, … m) bit support data of {A j}({A i});

[0013] Step 7: Use the admissible function R to calculate the admissible value R i of the j-th (j = 1, 2, … m) bit support data of {A j}({A i});

[0014] Step 8: According to the admissible value R j}({A i}) and the fuzzy measure M({A i}) of {A i}, solve the fusion result W j of the j-th (j = 1, 2, … m) bit n-dimensional intersection characteristic learning data(S);

[0015] Step 9: Jump to Step 6 until j = m, that is, complete the fusion of this group of learning data to obtain the fusion result W(S);

[0016] Step 10: Smooth the fusion result W(S);

[0017] Step 11: Calculate the objective function E, and judge whether E converges to the minimum value. If it has converged, stop the iteration and output the optimization value C best of the confidence level of this group of learning data, otherwise update to obtain the new confidence level C' = {C' n} of the electromagnetic and acoustic composite n-dimensional intersection characteristics, and jump to Step 5;

[0018] Step 12: Jump to Step 3 to start the optimization of a new group of learning data until all N groups of learning data are optimized, and output the comprehensive optimization result of the confidence level of the electromagnetic and acoustic composite n-dimensional intersection characteristics

[0019] Further, in step one, when generating the ideal n-dimensional intersection feature fusion curve Y, the formula is as follows:

[0020]

[0021] Where: l represents the intersection point; d represents the action area threshold value, which is given by the system.

[0022] Further, in step four, when converting the n-dimensional intersection feature learning data S into the n-dimensional intersection process recognition support degree, the formula is as follows:

[0023]

[0024] Where, X i,j is the recognition support degree of the j-th data point of the i-th dimension feature, S i,j represents the j-th data point of the i-th dimension feature, l j represents the j-th intersection point.

[0025] Further, in step five, when solving the fuzzy measure M({A i}), the formula is as follows:

[0026]

[0027] Where: Assume k is the number of feature types of the feature combination A i ; λ is the fuzzy measure coefficient, which can be obtained by solving the equation M(S) = 1, equivalent to solving the equation:

[0028]

[0029] Further, in step six, the fusion function T is calculated by the following formula:

[0030]

[0031] Further, in step seven, the tolerance function R is calculated by the following formula:

[0032] R j (A i ) = min(R1 (2) (d), R2 (2) (d), … R q (2) (d))

[0033]

[0034] Where, R q (2)(d) is the feature combination A i is the allowable value of any two-dimensional feature combination in d is the difference in the recognition support degree of the intersection process of two-dimensional feature quantities |X i -X j |; a1 and a2 are preset parameters.

[0035] Furthermore, in step eight, the fusion result W of the n-dimensional intersection feature learning data is solved by the following formula j (S):

[0036]

[0037] Furthermore, in step ten, when performing smoothing processing, cubic smoothing spline fitting is adopted, and the smoothing parameter is taken as 0.3.

[0038] Furthermore, in step eleven, the objective function E is calculated by the following formula

[0039]

[0040] where r is the number of iterative optimization times; W i is the fusion result of the intersection feature learning data in the i-th iterative optimization; is the smoothing result of W i .

[0041] Furthermore, in step twelve, the comprehensive optimization result is the mean value of the n-dimensional intersection feature confidence C best obtained for each group of learning data after all N groups of learning data have been optimized.

[0042] Compared with the prior art, the beneficial effects of the present invention are

[0043] The iterative optimization method for the confidence of the acoustic-magnetic composite intersection feature provided by the present invention uses the prior data of multi-physical field features, and completes the iterative optimization of the confidence of the electromagnetic and acoustic composite detection intersection features through machine learning methods, which can effectively solve the problem caused by the inconsistency of the confidence of multiple intersection features in the composite physical field and improve the accuracy of information fusion. Description of the Drawings

[0044] Figure 1a and Figure 1b are the setting diagrams for collecting intersection feature quantities in the specific embodiment of the iterative optimization method for the confidence of the acoustic-magnetic composite intersection feature provided by the present invention;

[0045] Figure 2a and Figure 2b are the schematic diagrams of the intersection process in the specific embodiment;

[0046] Figure 3 Schematic diagram of the variation of each objective function with the number of iterations in a specific embodiment;

[0047] Figure 4 Schematic diagram of the variation of the optimized value of each feature confidence with the number of iterations in a specific embodiment. Detailed implementation manners

[0048] To make the technical means, creative features, achieved objectives and effects of the present invention easy to understand, the following further elaborates how the present invention is implemented in conjunction with the accompanying drawings and specific implementation manners.

[0049] In a specific embodiment, the present invention provides an iterative optimization method for the confidence of acoustic-magnetic composite intersection features, including the following steps:

[0050] Step 1: Generate an ideal n-dimensional intersection feature fusion curve Y according to the electromagnetic and acoustic composite n-dimensional intersection characteristics.

[0051] The electromagnetic and acoustic composite n-dimensional intersection characteristics include electromagnetic field intersection feature quantities: the electromagnetic field intensity S1 and phase information S2 of the z-axis received by an electromagnetic sensor, and the electromagnetic field intensity S3 of the x-axis; and also include acoustic field intersection feature quantities: the sound pressure amplitude S4 of the target noise and the target azimuth interval S5.

[0052] Referring to Figure 1a as shown, Figure 1a is a side view. The detection system can be installed on an underwater vehicle. The electromagnetic field intersection feature quantities are obtained in the following manner: a rod-shaped spiral coil electromagnetic transmitting antenna is used, and its axis is perpendicular to the longitudinal axis of the detection system. X, Y, and Z-axis electromagnetic receiving coils are used, where the x-axis direction is the same as the longitudinal axis direction of the detection system, and the z-axis is the vertical direction. The electromagnetic field intensity S1 and phase information S2 of the z-axis, and the electromagnetic field intensity S3 of the x-axis are extracted respectively.

[0053] Referring to Figure 1b as shown, Figure 1b is a top view. The acoustic field intersection feature quantities are obtained in the following manner: a low-frequency underwater acoustic receiving transducer is used to collect the target radiation noise during the intersection process, and the sound pressure amplitude S4 is extracted. A medium-frequency active sonar is used to collect the target echo signal, and the target azimuth interval S5 is calculated and generated.

[0054] Referring to Figure 2a and Figure 2b as shown, in this embodiment, the intersection distance h between the detection system and the typical target is taken as 5 m. The intersection distance refers to the shortest distance between the detection system and the typical target during the intersection process. The intersection path considers a straight line interval range of -8 m to +8 m with the vertical projection point of the typical target on the intersection path as the origin, where the approach process is negative and the departure process is positive, and the intersection angle is taken as 60°.

[0055] On the intersection path, the system collects signals and calculates feature quantities once every 0.025 m, and the data length of each feature quantity is m = 650.

[0056] When generating the ideal n-dimensional intersection feature fusion curve Y, the formula is as follows:

[0057]

[0058] Where: l represents the intersection point; d represents the action area threshold value, which is given by the system, and d = 2 can be taken.

[0059] Step 2: Given the initial confidence values C = {C1, C2,... C n} of the electromagnetic and acoustic composite n-dimensional intersection features.

[0060] In this embodiment, the initial confidence values C = {C1, C2,... C n}, and C1 = C2 = C3 = C4 = C5 = 0.85 are taken respectively.

[0061] Step 3: Input a set of electromagnetic and acoustic composite n-dimensional intersection feature learning data S = {S1, S2,... S n}, and the length of each feature learning data is m.

[0062] Step 4: According to the mapping relationship between the electromagnetic and acoustic composite n-dimensional intersection features and the intersection process, convert the n-dimensional intersection feature learning data S into the n-dimensional intersection process recognition support X = {X1, X2,... X n}.

[0063] When converting the n-dimensional intersection feature learning data S into the n-dimensional intersection process recognition support, the formula is as follows:

[0064]

[0065] Where, X i,j is the recognition support of the j-th data point of the i-th dimension feature, S i,j represents the j-th data point of the i-th dimension feature, and l j represents the j-th intersection point.

[0066] Step 5: According to the current confidence, solve the fuzzy measures of 2 n -n-1 feature combinations {A i} (i = 1, 2,... 2 n -n-1) of the electromagnetic and acoustic composite n-dimensional intersection features

[0067] When solving the fuzzy measure M({A i}), the formula is as follows:

[0068]

[0069] Among them: Assume that k is the number of feature types of the feature combination A i ; λ is the fuzzy measure coefficient, which can be obtained by solving the equation M(S) = 1, equivalent to solving the equation:

[0070]

[0071] Step Six: Use the fusion function T to calculate the fusion value F i of the j-th (j = 1, 2,..., m) bit support degree data of {A j} i ).

[0072] The fusion function T is calculated by the following formula:

[0073]

[0074] Step Seven: Use the tolerance function R to calculate the tolerance value R i of the j-th (j = 1, 2,..., m) bit support degree data of {A j} i ).

[0075] The tolerance function R is calculated by the following formula:

[0076] R j (A i ) = min(R1 (2) (d), R2 (2) (d),... R q (2) (d))

[0077]

[0078] Among them, R q (2) (d) is the tolerance value of any 2D feature combination in the feature combination A i ; d is the recognition support degree difference |X - X i - X j | of the 2D feature quantity intersection process; a1 can take 0.2, and a2 can take 0.6.

[0079] Step Eight: According to the tolerance value R j ({A i}) and the fuzzy measure M({A i}) of {A i}, solve the fusion result W of the j-th (j = 1, 2,..., m) bit n-dimensional intersection feature learning dataj (S).

[0080] Solve for the fusion result \(W\) of the \(n\)-dimensional intersection feature learning data through the following formula j (S):

[0081]

[0082] Step Nine: Jump to Step Six until \(j = m\), that is, complete the fusion of this group of learning data to obtain the fusion result \(W(S)\).

[0083] Step Ten: Smooth \(W(S)\).

[0084] When performing smoothing, use cubic smoothing spline fitting, and the smoothing parameter can be taken as 0.3.

[0085] Step Eleven: Calculate the objective function \(E\), and determine whether \(E\) converges to the minimum value. If it has converged, stop the iteration and output the optimized value \(C\) of the confidence level of this group of learning data best , otherwise update to obtain the new confidence level \(C'\) of the electromagnetic and acoustic composite \(n\)-dimensional intersection feature, and jump to Step Five.

[0086] Calculate the objective function \(E\) through the following formula

[0087]

[0088] where \(r\) is the number of iterative optimization times; \(W\) i is the fusion result of the intersection feature learning data in the \(i\)-th iterative optimization; is the smoothing result of \(W\) i .

[0089] Refer to Figure 3 shown, which is a schematic diagram of the change of each objective function \(E\) with the number of iterations in this embodiment.

[0090] Refer to Figure 4 shown, which is a schematic diagram of the change of the optimized value of each feature confidence level \(C\) best with the number of iterations in this embodiment.

[0091] Step Twelve: Jump to Step Three and start optimizing a new group of learning data until all \(N\) groups of learning data are optimized, and output the comprehensive optimized result of the confidence level of the electromagnetic and acoustic composite \(n\)-dimensional intersection feature

[0092] The said comprehensive optimized result is the mean value of the \(n\)-dimensional intersection feature confidence level \(C\) best obtained for each group of learning data after all \(N\) groups of learning data are optimized.

[0093] In summary, the iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature provided by the present invention uses the prior data of multi-physical field features and completes the iterative optimization of the confidence level of the electromagnetic and acoustic composite detection intersection features through machine learning methods. It can effectively solve the problems caused by the inconsistency of the confidence levels of various intersection features in the composite physical field and improve the accuracy of information fusion.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An iterative optimization method for the confidence level of acoustic-magnetic composite intersection features, characterized in that Including the following steps: Step 1: Generate an ideal n-dimensional intersection characteristic fusion curve Y according to the electromagnetic and acoustic composite n-dimensional intersection characteristics; Step 2: Given the initial confidence values C = {C1, C2, ..., Cn} of the electromagnetic and acoustic composite n-dimensional intersection features n}; Step 3: Input a set of electromagnetic and acoustic composite n-dimensional intersection feature learning data S = {S1, S2, LS n}, and the length of each feature learning data is m; Step 4: According to the mapping relationship between the electromagnetic and acoustic composite n-dimensional intersection characteristics and the intersection process, convert the n-dimensional intersection characteristic learning data S into the n-dimensional intersection process recognition support degree X = {X1, X2, LX n}; Step Five: Solve the fuzzy measures of the 2 n -n-1 feature combinations {A i}(i = 1, 2, L2 n -n-1) of the electromagnetic and acoustic composite n-dimensional intersection features Step 6: Use the fusion function T to calculate the fusion value T i of the support degree data of the j-th (j = 1, 2,..., m) bit of {A j ({A i}); Step 7: Using the tolerance function R, calculate the tolerance value R of the support degree data of the j-th (j = 1, 2, … m) bit of {A i}; j ({A i}); Step Eight: According to the tolerance value R j ({A i}) and the fuzzy measure M({A i}) of {A i}, solve the fusion result W j (S) of the j-th (j = 1, 2, … m) n-dimensional intersection feature learning data; Step 9: Jump to Step 6 until j = m, that is, complete the fusion of this set of learning data to obtain the fusion result W(S); Step 10: Smooth the fusion result W(S); Step Eleven: Calculate the objective function E, and determine whether E converges to the minimum value. If it has converged, stop the iteration and output the optimized value C of the confidence level of this set of learning data best , otherwise, update to obtain the new confidence level C' of the electromagnetic and acoustic composite n-dimensional intersection feature, C' = {C'1, C'2, LC' n}, and jump to Step Five; Step Twelve: Jump to Step Three and start optimizing a new set of learning data until all N sets of learning data have been optimized, and output the comprehensive optimization result of the confidence level of the electromagnetic and acoustic composite n-dimensional intersection features 2. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 1, characterized in that In Step 1, when generating the ideal n-dimensional intersection characteristic fusion curve Y, the formula is as follows: Where: l represents the intersection point; d represents the action area threshold value, which is given by the system.

3. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 2, characterized in that, In Step 4, when converting the n-dimensional intersection characteristic learning data S into the n-dimensional intersection process recognition support degree, the formula is as follows: Among them, X i,j is the recognition support degree of the j-th data point of the i-th dimensional feature, S i,j represents the j-th data point of the i-th dimensional feature, l j represents the j-th intersection point.

4. The iterative optimization method for the confidence of the acoustic-magnetic composite intersection feature according to claim 3, wherein In step five, when solving the fuzzy measure M({A i}), the formula is as follows: Where: Assume that k is the number of feature types of the feature combination A i ; λ is the fuzzy measure coefficient, which can be obtained by solving the equation M(S)=1, which is equivalent to solving the equation:

5. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 4, wherein In Step 6, the fusion function T is calculated by the following formula:

6. The iterative optimization method for the confidence of the acoustic-magnetic composite intersection feature according to claim 5, characterized in that, In Step 7, the tolerance function R is calculated by the following formula: R j (A i ) = min(R1 (2) (d), R2 (2) (d), … R q (2) (d)) Among them, R q (2) (d) is the allowable value of any two-dimensional feature combination in feature combination A i ; d is the difference in recognition support degree |X -X i -X j | in the intersection process of two-dimensional feature quantities; a1 and a2 are preset parameters.

7. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 6, wherein In step eight, the fusion result W of the n-dimensional intersection feature learning data is solved by the following formula j (S):

8. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 7, characterized in that, In Step 10, when performing smoothing processing, cubic smoothing spline fitting is adopted, and the smoothing parameter is taken as 0.

3.

9. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 8, wherein In Step 11, the objective function E is calculated by the following formula: Among them, r is the number of iterative optimization times; W i is the fusion result of the intersection feature learning data in the i-th iterative optimization; is the smoothing result of W i .

10. The iterative optimization method for the confidence level of the acoustic-magnetic composite intersection feature according to claim 9, wherein In step twelve, the comprehensive optimization result is the mean value of the n-dimensional intersection feature confidence C best obtained for each set of learning data after all N sets of learning data have been fully optimized.

Images