Acoustic target substrate extraction method and system based on azimuth history graph

Abstract

The invention relates to the technical field of sonar signal and information processing, in particular to an acoustic target substrate extraction method and system based on an azimuth history graph, and the method comprises the steps: extracting a potential target from an acoustic target azimuth history graph, and constructing an element of the potential target; presetting a plurality of clustering numbers, and clustering the primitives by using a clustering algorithm for each clustering number to obtain a clustering center to which each primitive belongs under each clustering number; for each cluster number, calculating the value of each order derivative of the evaluation function under the cluster number; and selecting the clustering number with the maximum third derivative value from the clustering numbers corresponding to the conditions that the first derivative is less than 0, the second derivative is less than 0 and the third derivative is greater than 0, and taking the clustering center corresponding to the selected clustering number as the acoustic target substrate template. The constructed primitives are clustered, all-order derivatives and related maximum values of evaluation functions under different clustering numbers are calculated, the optimal clustering number and the corresponding clustering center are determined, and a target base template is constructed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of acoustic signal and information processing, and particularly relates to an acoustic target base extraction method and system based on a bearing history graph. BACKGROUND

[0002] The bearing history graph of an acoustic target is used to discover and track the target, which is a common method for underwater acoustic target discovery and bearing tracking. The method for obtaining the bearing history graph of an underwater acoustic target can be represented as the following steps:

[0003] 1. Sonar array received signal interception: intercept a fixed length of sonar array received signal at a certain time interval. This can be represented as:

[0004] s(t) = x(t) * h(t)

[0005] where s(t) represents the sonar array received signal, x(t) represents the target signal, h(t) represents the array impulse response, and t is the time.

[0006] 2. Bearing angle difference calculation: select one element of the sonar array as a reference, and calculate the time difference of the target signal received by each element according to a certain bearing angle interval. This can be represented as:

[0007]

[0008] where Δt j represents the time difference between element j and the reference element, d j represents the distance between element j and the reference element, θ j represents the bearing angle of element j relative to the reference element, θ0 is the observation bearing, and c represents the sound speed.

[0009] 3. Time compensation: according to the time difference of the target signal received by each element, the received signal of each element is time compensated to make the initial phase of the target signal received by each element consistent. This can be represented as:

[0010] s j (t) = s(t - Δt j )

[0011] where s j (t) represents the received signal of element j after time compensation, and s(.) represents the reference signal waveform.

[0012] 4. Synthesis beam forming: the time-compensated received signal is weighted or equally fused to obtain a synthesized beam corresponding to the bearing angle. This can be represented as:

[0013]

[0014] where b(θ j ) represents the synthetic beam at azimuth angle θ j , ω j represents the weight of the array element j, and N represents the number of array elements.

[0015] 5. Bit energy spectrum calculation: Calculate the total energy value of the synthetic beam to obtain the beam energy spectrum with azimuth angle as the horizontal coordinate and energy amplitude as the vertical coordinate, which is called the azimuth energy spectrum of the synthetic beam. This can be expressed as:

[0016] E(θ j ) = |b(θ j )| 2

[0017] where E(θ j ) represents the beam energy at azimuth angle θ j .

[0018] 6. Construction of azimuth history map: Establish a time-azimuth angle space coordinate system to represent the beam energy amplitude at the corresponding time and azimuth with a gray value, and the azimuth history map of the underwater acoustic target can be obtained.

[0019] After obtaining the azimuth history map, the peak value target can be extracted, and then the real / fake underwater acoustic target base template is determined, and then the underwater acoustic target detection is carried out based on the base template and the peak value target. The existing patent document CN117169864A discloses constructing a real underwater acoustic target base template using real underwater acoustic target base elements in historical experience data, and constructing a fake underwater acoustic target base template using fake underwater acoustic target base templates in historical experience data.

[0020] However, when there is no historical experience data or insufficient amount of historical experience data, the target base template cannot be constructed, and thus the underwater acoustic target detection cannot be carried out using the target base template. SUMMARY

[0021] The purpose of the present application is to overcome the problem in the prior art that the target base template cannot be constructed and used for underwater acoustic target detection when there is no historical experience data or insufficient amount of historical experience data, and to provide a target base template extraction method and system based on the azimuth history map of the underwater acoustic target.

[0022] In one aspect, an acoustic target base extraction method based on an azimuth history map is provided, comprising:

[0023] Step 1: Extract potential targets from the acoustic target azimuth history map and construct base elements of the potential targets, wherein each base element contains at least one peak value;

[0024] Step 2: preset a plurality of cluster numbers, and perform clustering on the constructed primitives using a clustering algorithm for each cluster number to obtain a cluster center to which each primitive belongs under each cluster number;

[0025] Step 3: for each cluster number, calculate the value of each derivative of the evaluation function under the cluster number, wherein the derivatives include first derivative, second derivative and third derivative, and the evaluation function represents a residual square relationship between the primitive and the cluster center to which the primitive belongs;

[0026] Step 4: in the cluster number corresponding to the first derivative less than 0, the second derivative less than 0 and the third derivative greater than 0, select the cluster number with the maximum third derivative value, and take the cluster center corresponding to the selected cluster number as the extracted acoustic target base.

[0027] In an improved acoustic target base extraction method based on the azimuth history graph, step 2 includes:

[0028] Step 2-1: determine an initial cluster center list according to the cluster number, the global similarity and the local optimal similarity of each primitive, and take the initial cluster center list as the current cluster center list;

[0029] Step 2-2: match each primitive with a cluster according to the correlation coefficient between each primitive and each cluster center in the current cluster center list;

[0030] Step 2-3: for each cluster, update the cluster center thereof by using the position of the peak value of all primitives belonging to the cluster to obtain an updated cluster center list;

[0031] Step 2-4: determine whether an iteration termination condition is met, if yes, end; if not, take the updated cluster center list as the current cluster center list, and jump to step 2-2; wherein the iteration termination condition is that the cluster center in the current iteration is not updated or the predetermined iteration number is reached.

[0032] In an improved acoustic target base extraction method based on the azimuth history graph, step 2-1 specifically includes:

[0033] Step 2-1-1: calculate the score of each primitive according to the global similarity and the local optimal similarity of each primitive;

[0034] Step 2-1-2: select a cluster number of primitives from all primitives as the initial cluster center list based on the high and low of the score, and take the initial cluster center list as the current cluster center list.

[0035] In an improved acoustic target basis extraction method based on orientation history map, step 2-1-1 includes: calculating the primitive e using the following formula. i S rating i :

[0036] S i =G i ×L i

[0037] Among them, G i For the basic element e i global similarity, L i For the basic element e i The local optimal similarity.

[0038] In an improved acoustic target basis extraction method based on orientation history maps, the global similarity G i And local optimal similarity L i Calculated using the following formula:

[0039]

[0040] Where i and j are primitive indices, n is the number of primitives, and R ij For the basic element e i and basic element e j The correlation coefficient, e ik It is the basic element e i The value at peak index k, e jk It is the basic element e j The value at peak index k, where m is the number of peaks contained in each primitive, and k is the peak index.

[0041] In an improved acoustic target basis extraction method based on orientation history map, step 2-1-2 includes:

[0042] Select rating S i The top Q primitives are used as the initial list of cluster centers C. (0) The initial cluster center list C (0) As a list of current cluster centers:

[0043]

[0044] Where Q is the number of clusters, q is the cluster index, and 0 indicates the initial clustering. The first cluster center in the current cluster. It is the second cluster center in the current cluster. For the q-th cluster center, Let Q be the Qth cluster center.

[0045] In an improved azimuth history-based acoustic source basis extraction method, the step 2-2 comprises:

[0046] For each basis element e n in the basis set E = {e1, e2, …, en}, the correlation coefficient between it and each cluster center i is calculated, and the basis element e i is assigned to the cluster center with the largest correlation coefficient by using the following formula:

[0047]

[0048] where i is the basis element index, i takes values from 1 to n, n is the number of basis elements, t is the cluster number, q is the cluster index, q takes values from 1 to Q, is the correlation coefficient between the basis element e i and each cluster center , and the calculation formula is as follows:

[0049]

[0050] where is the kth peak value of the basis element sample e i , is the kth peak value of the qth cluster center , t is the cluster number, k is the peak index, takes values from 1 to N, and N is the number of peaks.

[0051] In an improved azimuth history-based acoustic source basis extraction method, the step 2-3 comprises:

[0052] The cluster center of the cluster q is updated by using the following formula:

[0053]

[0054] where S q is the set of basis element samples assigned to the cluster q, |S q | is the number of basis elements assigned to the cluster q, d i represents the set of all peaks of the basis element e i assigned to the cluster q.

[0055] In an improved azimuth history-based acoustic source basis extraction method, the evaluation function comprises the total squared error SSE, and the SSE of each cluster number is calculated by the following formula:

[0056]

[0057] wherein Q is the number of clusters, y i = q represents the qth cluster to which the element e i is assigned, e i denotes the ith element, c q denotes the qth cluster center, ‖e i -c q ‖ represents the Euclidean distance of the element e i to the cluster center c q .

[0058] In another aspect, a system based on the above-mentioned acoustic target basis extraction method based on bearing history graph is provided, comprising: an element construction unit, a clustering unit, an evaluation unit and a basis determination unit;

[0059] The element construction unit is configured to extract potential targets from an acoustic target bearing history graph and construct elements of the potential targets, wherein each element contains at least one peak value;

[0060] The clustering unit is configured to preset a plurality of cluster numbers, and perform clustering on the constructed elements using a clustering algorithm for each cluster number to obtain cluster centers to which each element belongs under each cluster number.

[0061] The evaluation unit is configured to calculate values of derivatives of an evaluation function under each cluster number, wherein the derivatives include first, second and third derivatives, and the evaluation function represents a squared residual relationship between the elements and the cluster centers to which the elements belong.

[0062] The basis determination unit is configured to select a cluster number with the largest third derivative value from the cluster numbers corresponding to the first derivative less than 0, the second derivative less than 0 and the third derivative greater than 0, and take the cluster center corresponding to the selected cluster number as the extracted acoustic target basis.

[0063] Compared with the prior art, the advantages of the present application are:

[0064] In the scheme of the embodiment of the present application, by clustering the constructed elements, calculating the derivatives of the evaluation function under different cluster numbers and the related maximum values, the optimal cluster number and the corresponding cluster center are determined, and the optimal cluster center is used as the extracted acoustic target basis, so that when there is no historical experience data or the amount of historical experience data is insufficient, a target basis template can be constructed, and then the target basis template can be used for underwater acoustic target detection. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A flowchart of the acoustic target basis extraction method based on bearing history graph provided by the embodiment 1 of the present application.

[0066] Figure 2 A raw bearing history plot provided for the embodiment 2 of the present application;

[0067] Figure 3 A peak extraction plot provided for the embodiment 2 of the present application;

[0068] Figure 4 A motif score plot provided for the embodiment 2 of the present application;

[0069] Figure 5 An initial clustering center plot after dimension reduction provided for the embodiment 2 of the present application;

[0070] Figure 6 Sum of squared error under different clustering numbers provided for the embodiment 2 of the present application;

[0071] Figure 7 A first derivative plot of sum of squared error provided for the embodiment 2 of the present application;

[0072] Figure 8 A second derivative plot of sum of squared error provided for the embodiment 2 of the present application;

[0073] Figure 9 A third derivative plot of sum of squared error provided for the embodiment 2 of the present application;

[0074] Figure 10 A extracted motif plot provided for the embodiment 2 of the present application, (a) is a motif plot of a real target, (b), (c) and (d) are motif plots corresponding to false alarms;

[0075] Figure 11 A system plot of the acoustic target motif extraction method based on bearing history plot provided for the embodiment 3 of the present application. DETAILED DESCRIPTION

[0076] The technical solutions provided by the present application are further illustrated below in combination with embodiments.

[0077] Embodiment 1

[0078] As shown in the figure, it is a flow chart of the acoustic target motif extraction method based on bearing history plot provided for the embodiment 1 of the present application, including the following steps 1 to 4: Figure 1

[0079] Step 1: extract potential targets from the acoustic target bearing history plot and construct motifs of the potential targets.

[0080] This step 1 includes the following steps 1-1 and 1-2:

[0081] Step 1-1: extract potential targets from the acoustic target bearing history plot. ​

[0082] Here, the peak extraction method can be used to extract potential targets from the acoustic target bearing history map.

[0083] Specifically, whether there is a potential target for the (θ, t) grid is distinguished by logic true and false, where true represents a potential target, and false represents no potential target. The decision map P(θ j ,t) is specifically calculated as follows:

[0084]

[0085] Step 1-2: For each potential target, a bearing history map of a set size is intercepted around its corresponding time and bearing, obtaining the element of the potential target, wherein each element contains at least one peak value.

[0086] For the potential targets extracted from the bearing history map, a suitable size can be defined to construct the element U(θ j ,t), and an unsupervised method is used to construct various base templates, which respectively correspond to the real target base template V true and the false target base template V false , i.e. false alarm. The following steps 2 to 4 give the construction method of the base template.

[0087] Step 2: A plurality of clustering numbers are preset, and for each clustering number, a clustering algorithm is used to cluster the constructed elements, obtaining the clustering center to which each element belongs under each clustering number.

[0088] In this step 2, a plurality of clustering numbers can be preset by prior analysis or heuristic method.

[0089] Here, the clustering algorithm such as K-Means clustering, mean shift clustering, noise-based density-based spatial clustering algorithm, etc. can be used to cluster the constructed elements.

[0090] Suppose the preset clustering numbers are 3, 4, 5, 6, 7, 8, 9 and 10, then for the clustering number of 3, the following steps 2-1 to 2-4 are executed, for the clustering number of 4, the following steps 2-1 to 2-4 are executed…for the clustering number of 10, the following steps 2-1 to 2-4 are executed, a total of 8 times. The clustering center to which each element belongs when the clustering number is 3, the clustering center to which each element belongs when the clustering number is 4…the clustering center to which each element belongs when the clustering number is 10 are obtained.

[0091] Specifically, in this step 2, the constructed elements are clustered based on the K-Means clustering algorithm, combined with the actual clustering requirements of the scheme of the application, which can be realized by the following steps 2-1 to 2-4:

[0092] Step 2-1: According to the number of clusters, the global similarity of each element and the local optimal similarity, determine the initial cluster center list as the current cluster center list.

[0093] Here, step 2-1 specifically includes step 2-1-1 and step 2-1-2:

[0094] Step 2-1-1: Calculate the score of each element according to the global similarity and the local optimal similarity of each element;

[0095] Specifically, the score of each element can be calculated by the following first step to third step:

[0096] First step: Calculate the global similarity G of each element by using the following formula (1) i ;

[0097]

[0098] Where i and j are element indices, n is the number of elements, R ij is the correlation coefficient of element e i and element e j , R ij is obtained by the following formula (2):

[0099]

[0100] Where e ik is the value of element e i at peak index k, m is the number of peaks contained in each element, and k is the peak index.

[0101] Second step: Calculate the local optimal similarity L i of each element by using the following formula (3)

[0102]

[0103] Third step: Calculate the score S i of element e i by using the following formula (4)

[0104] S i = G i × L i (4)

[0105] Where G i is the global similarity of element e i , and L i is the local optimal similarity of element e i .

[0106] Step 2-1-2: Based on the scores, select a certain number of primitives from all primitives to form an initial list of cluster centers, and use this initial list of cluster centers as the current list of cluster centers.

[0107] Specifically, a score S can be selected. i The top Q primitives are used as the initial list of cluster centers C. (0) The initial cluster center list C (0) As a list of current cluster centers: Where Q is the number of clusters, q is the cluster index, and 0 indicates the initial clustering. The first cluster center in the current cluster. It is the second cluster center in the current cluster. For the q-th cluster center, Let Q be the Qth cluster center.

[0108] Step 2-2: Match each primitive with a cluster based on the correlation coefficient between each primitive and each cluster center in the current cluster center list.

[0109] For the primitive set E = {e1, e2, ..., e} n Each primitive e in} i Calculate its relationship with each cluster center Correlation coefficient between And use the following formula (5) to convert the basic element e i Assigned to the cluster center with the highest correlation coefficient

[0110]

[0111] Where i is the primitive index, ranging from 1 to n, n is the number of primitives, t is the number of clusters, and q is the cluster index, ranging from 1 to Q. For the basic element e i With each cluster center The correlation coefficient between them can be calculated using the following formula (6):

[0112]

[0113] in, It is the primitive sample e i The kth peak, It is the q-th cluster center The kth peak, t is the number of clusters, k is the peak index, which takes values ​​from 1 to N, and N is the number of peaks.

[0114] Steps 2-3: For each cluster, update its cluster center using the positions of the peaks of all primitives belonging to that cluster, and obtain the updated list of cluster centers;

[0115] Update the cluster centers of cluster q using the following formula (7).

[0116]

[0117] Among them, S q It is the set of primitive samples assigned to cluster q, |S q | is the number of primitives assigned to cluster q, and d i The primitive e represents the element assigned to cluster q. i The set of all peaks.

[0118] Formula (7) updates the cluster center by calculating the average position of all peaks belonging to the cluster, making the points within the cluster more compact.

[0119] Step 2-4: Determine if the iteration termination condition is met. If yes, end the iteration; otherwise, use the updated cluster center list as the current cluster center list and jump to step 2-2. The iteration termination condition is: the cluster centers in this iteration are no longer updated or the predetermined number of iterations is reached.

[0120] The cluster centers in this iteration are no longer updated, which means that the following formula (8) is satisfied:

[0121]

[0122] Step 3: For each number of clusters, calculate the values ​​of the derivatives of the evaluation function for that number of clusters. The derivatives include the first, second, and third derivatives. The evaluation function expresses the squared residual relationship between the primitive and its cluster center.

[0123] Evaluation functions that can express the residual square relationship between primitives and their respective cluster centers include: total squared error (SSE), mean squared error (MME), and root mean squared error (RMSE).

[0124] SSE measures the sum of squared distances from all data points to their respective cluster centers. A smaller SSE indicates better clustering. Q represents the current number of clusters, i.e., the number of cluster centers.

[0125] When the evaluation function uses SSE, the SSE for each cluster number is calculated using the following formula (9):

[0126]

[0127] Where Q is the number of clusters, y i =q represents the primitive e i Assigned to the q-th cluster, e i Let c represent the i-th primitive. q Let q be the cluster center, and ||e|| i -c q ‖ represents the primitive e i To cluster center c q Euclidean distance.

[0128] Here, the first derivative SSE'(Q) represents the rate of change of SSE with respect to Q, the second derivative SSE""Q" represents the rate of change of the rate of change of SSE, and the third derivative SSE"'(Q) represents the rate of change of the rate of change of SSE.

[0129] A negative first derivative (SSE'(Q)<0) indicates that SSE decreases as the number of clusters Q increases, meaning that increasing the number of clusters can better fit the data. A negative second derivative (SSE'(Q)<0) indicates that the rate of SSE reduction is accelerating, meaning that the improvement in clustering performance is faster as Q increases. A positive and maximum third derivative indicates that at this Q value, the rate of SSE reduction begins to slow down, and the rate of slowing down reaches a relative peak. This means that although increasing the number of clusters can still reduce SSE, this reduction effect begins to weaken significantly. After this point, increasing the number of clusters may not bring significant benefits, and may even lead to a deterioration in model performance due to overfitting.

[0130] Step 4: Among the number of clusters that satisfy the condition that the first derivative is less than 0, the second derivative is less than 0, and the third derivative is greater than 0, select the number of clusters with the largest third derivative value, and use the cluster center corresponding to the selected number of clusters as the extracted acoustic target basis.

[0131] This embodiment 1 presents a method for acoustic target substrate extraction. The following embodiment 2 will further illustrate this method. Figure 2 The orientation history diagram shown illustrates the process of acoustic target substrate extraction in detail.

[0132] Example 2

[0133] Assuming the location history diagram being processed is as follows Figure 2 As shown.

[0134] (I) Extracting potential targets

[0135] The presence of a potential target in the (θ,t) grid is distinguished by logical true and false, where true indicates the presence of a potential target and false indicates the absence of a potential target. The decision map P is extracted, and the specific calculation method is as follows:

[0136]

[0137] The obtained decision spectrum P is as follows Figure 3 As shown, each pixel is labeled as having a potential target (true) or not having a potential target (false).

[0138] (II) Extracting the substrate

[0139] For the potential targets extracted from the orientation history map, construct primitive U(θ) according to the size (7,7). j A real target basis V can be constructed using unsupervised learning methods. true And three other types of false target base templates V false This is a false alarm. Figure 3 12,303 primitives were extracted from the determination graph P.

[0140] 1. Initial Steps: Setting up and Optimizing Initial Cluster Centers

[0141] Here, the initial number of clusters is set to Q = 30.

[0142] Calculate global similarity G i :

[0143]

[0144] Among them, R ij Representing the primitive e i With the basic element e j The correlation coefficient between them.

[0145] Among them, the correlation coefficient R ij The calculation formula is:

[0146]

[0147] Among them, e ik It is the basic element e i The value at peak k, where m is the number of peaks contained in each primitive.

[0148] Calculate the local optimal similarity L i :

[0149]

[0150] Select initial cluster centers and calculate score S. i :

[0151] S i =G i ×L i

[0152] Select rating S i The top 30 primitives are used as the initial cluster centers.

[0153]

[0154] exist Figure 3 12303 primitives were extracted from the decision graph P, and the score graph for each primitive is shown below. Figure 4 As shown, the primitives whose scores belong to the top 30 have been placed in... Figure 3 The values ​​are marked with a "+" sign. To more intuitively observe the top 30 initial cluster centers with the highest scores, principal component analysis was used to reduce the dimensionality of the original data. The resulting primitive distribution diagram after dimensionality reduction is shown below. Figure 5 As shown, the initial cluster centers are in Figure 5 The text has been marked with a "+" sign.

[0155] 2. Clustering Iterative Process:

[0156] Matching: For the primitive set E = {e1, e2, ..., e} n In each iteration, it is calculated to be related to each cluster center. Correlation coefficient between Let the basic element e i Assign it to the cluster center with the highest correlation coefficient.

[0157]

[0158] The correlation coefficient is calculated using the following formula:

[0159]

[0160] in, It is the primitive sample e i The kth peak, Cluster center The kth peak.

[0161] Update cluster center:

[0162]

[0163] Among them, S q It is the set of primitive samples assigned to cluster q, |S q | is the number of primitives assigned to cluster q.

[0164] 3. Dynamically adjust the number of clusters

[0165] Calculate SSE: Whenever the number of clusters is updated, calculate the SSE and its derivative for the current number of clusters Q.

[0166] Choose the number of clusters Q that simultaneously satisfies the following three conditions:

[0167] 1. SSE ′ (Q)<0

[0168] 2. SSE″(Q)<0

[0169] 3. SSE″′(Q) > 0 and is the largest among these Q values.

[0170] Based on the physical meaning, the minimum number of clusters is 3, namely the main lobe, left side lobe, and right side lobe. Therefore, we set the minimum number of clusters to 3. The resulting graphs of the squared error and its first, second, and third derivatives are shown below. Figure 6 , Figure 7 , Figure 8 and Figure 9 As shown. It should be noted that... Figure 6 , Figure 7 , Figure 8 and Figure 9 For ease of observation, the values ​​of the squared errors and their first, second, and third derivatives for clusters with numbers of 1 and 2 are also plotted. Figure 9 As can be seen, except when the number of clusters is 1, the third derivative is the largest when the number of clusters is 4, and both the first and second derivatives are negative. Therefore, the adaptively selected number of clusters is 4.

[0171] Test results: The dimensionality reduction results of the initial center selection of the optimized clustering algorithm are as follows: Figure 5 As shown in the figure. This initialization selection method exhibits high stability and avoids the local optimum problem that may be caused by random initialization. The result of adaptively determining the number of clusters is shown in the figure. Figure 9 As shown, the number of clusters is adaptively selected based on a series of quantization strategies. The extracted four basis classes are as follows: Figure 10 As shown, (a) is the basis map V corresponding to the real target. true (b), (c) and (d) are the basis maps V corresponding to false alarms. false .

[0172] Example 3

[0173] Embodiment 3 of the present invention provides a system based on the above-described acoustic target basis extraction method based on azimuth history maps, such as... Figure 11 As shown, it includes: primitive building unit 110, clustering unit 111, evaluation unit 112 and basis determination unit 113;

[0174] The primitive construction unit 110 is used to extract potential targets from the acoustic target orientation history map and construct primitives of potential targets, wherein each primitive contains at least one peak.

[0175] The clustering unit 111 is used to preset multiple cluster numbers, and for each cluster number, the constructed primitives are clustered using a clustering algorithm to obtain the cluster center of each primitive under each cluster number.

[0176] The evaluation unit 112 is used to calculate the values ​​of the derivatives of the evaluation function for each number of clusters, wherein each derivative includes the first derivative, the second derivative, and the third derivative. The evaluation function expresses the residual square relationship between the primitive and its corresponding cluster center.

[0177] The basis determination unit 113 is used to select the number of clusters with the largest third derivative value from the number of clusters that satisfy the condition that the first derivative is less than 0, the second derivative is less than 0, and the third derivative is greater than 0, and use the cluster center corresponding to the selected number of clusters as the extracted acoustic target basis.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting acoustic target basis based on bearing history plot, comprising: Step 1: extracting potential targets from an acoustic target bearing history plot and constructing basis elements of the potential targets, wherein each basis element contains at least one peak; Step 2: presetting a plurality of cluster numbers, and for each cluster number, using a clustering algorithm to cluster the constructed basis elements to obtain cluster centers to which each basis element belongs under each cluster number; Step 3: for each cluster number, calculating values of derivatives of an evaluation function under the cluster number, wherein the derivatives include first derivative, second derivative and third derivative, and the evaluation function represents a squared residual relationship between a basis element and a cluster center to which the basis element belongs; Step 4: in the cluster numbers corresponding to the conditions that the first derivative is less than 0, the second derivative is less than 0 and the third derivative is greater than 0, selecting the cluster number with the maximum third derivative value, and taking the cluster center corresponding to the selected cluster number as an extracted acoustic target basis.

2. The bearing history plot based acoustic source basis extraction method of claim 1, wherein, In Step 2, for each cluster number, using a clustering algorithm to cluster the constructed basis elements, comprising: Step 2-1: determining an initial cluster center list according to the cluster number, global similarity and local optimal similarity of each basis element, and taking the initial cluster center list as a current cluster center list; Step 2-2: matching a cluster for each basis element according to a correlation coefficient between the basis element and each cluster center in the current cluster center list; Step 2-3: for each cluster, updating its cluster center by using positions of peaks of all basis elements belonging to the cluster to obtain an updated cluster center list; Step 2-4: judging whether an iteration termination condition is met, if yes, ending, and if no, taking the updated cluster center list as the current cluster center list and jumping to Step 2-2; wherein the iteration termination condition is that the cluster centers in the current iteration are no longer updated or a predetermined iteration number is reached.

3. The bearing history plot based acoustic source basis extraction method of claim 2, wherein, Step 2-1 specifically comprises: Step 2-1-1: calculating a score of each basis element according to the global similarity and the local optimal similarity of the basis element; Step 2-1-2: selecting cluster number basis elements from all basis elements based on the high and low of the scores to obtain an initial cluster center list, and taking the initial cluster center list as the current cluster center list.

4. The bearing history plot based acoustic source basis extraction method of claim 3, wherein, The step 2-1-1 comprises calculating the score S of the primitive e by using the following formula i : S = 1 - (1 - e) / (1 - e i : S i = G i x L i where G i is the global similarity of the primitive e i and L i is the local optimal similarity of the primitive e i .

5. The bearing history plot based acoustic source basis extraction method of claim 3, wherein, the global similarity G i and the local optimal similarity L i by the following formula: where i and j are the basis indices, n is the number of bases, R ij is the correlation coefficient of the bases e i and e j , e ik is the value of the basis e i at the peak index k, e jk is the value of the basis e j at the peak index k, and m is the number of peaks contained in each basis and k is the peak index.

6. The bearing history plot based acoustic source basis extraction method of claim 4, wherein, The Step 2-1-2 comprises: Selecting scores S i The top Q motifs as the initial cluster center list C (0) , the initial cluster center list C (0) as the current cluster center list: wherein Q is the number of clusters, q is the cluster index, 0 represents the initial cluster, is the first cluster center in the current cluster, is the second cluster center in the current cluster, is the qth cluster center, is the Qth cluster center.

7. The bearing history plot based acoustic source basis extraction method of claim 2, wherein, The Step 2-2 comprises: For each element e n in the set E = {e1, e2,..., e i , the correlation coefficient between it and each cluster center is calculated and the element e i is assigned to the cluster center with the largest correlation coefficient using the following formula Wherein, i is the index of the base element, i is 1 to n, n is the number of base elements, t is the number of clusters, q is the index of the cluster, q is 1 to Q, The base element e i The correlation coefficient between each cluster center The correlation coefficient between each cluster center wherein, is the kth peak value of the element sample e i is the kth peak value of the qth cluster center t is the number of clustering, k is the peak index, and N is the number of peaks.​ 8. The bearing history plot based acoustic source basis extraction method of claim 7, wherein, The Step 2-3 comprises: The cluster center of the cluster q is updated using the following formula where S q is the set of motif samples assigned to cluster q, |S q | is the number of motifs assigned to cluster q, d i denotes the set of all peaks of motif e i assigned to cluster q.

9. The bearing history plot based acoustic source basis extraction method of claim 1, wherein, The evaluation function comprises a total squared error SSE, and the SSE under each cluster number is calculated by the following formula: where Q is the number of clusters, y i = q represents the q-th cluster to which the element e i is assigned, e i represents the i-th element, c q represents the q-th cluster center, ‖e i -c q ‖ represents the Euclidean distance of the element e i to the cluster center c q .

10. A system for acoustic target basis extraction based on the bearing history plot based method of claim 1, characterized by, The method comprises: A basis element constructing unit, a clustering unit, an evaluation unit and a basis determining unit; The basis element constructing unit is configured to extract potential targets from an acoustic target bearing history plot and construct basis elements of the potential targets, wherein each basis element contains at least one peak; The clustering unit is configured to preset a plurality of cluster numbers, and for each cluster number, use a clustering algorithm to cluster the constructed basis elements to obtain cluster centers to which each basis element belongs under each cluster number; The basis determining unit is configured to for each cluster number, calculate values of derivatives of an evaluation function under the cluster number, wherein the derivatives include first derivative, second derivative and third derivative, and the evaluation function represents a squared residual relationship between a basis element and a cluster center to which the basis element belongs; in the cluster numbers corresponding to the conditions that the first derivative is less than 0, the second derivative is less than 0 and the third derivative is greater than 0, select the cluster number with the maximum third derivative value, and take the cluster center corresponding to the selected cluster number as an extracted acoustic target basis. The evaluation unit is configured to calculate, for each cluster number, a value of each derivative of an evaluation function at the cluster number, wherein the derivatives include a first derivative, a second derivative, and a third derivative, and the evaluation function represents a squared residual relationship between a basis element and a cluster center to which the basis element belongs. The basis determination unit is configured to select, from among the cluster numbers corresponding to the conditions that the first derivative is less than 0, the second derivative is less than 0, and the third derivative is greater than 0, a cluster number with the largest value of the third derivative, and take the cluster center corresponding to the selected cluster number as the extracted acoustic target basis.

Images