An acoustic bearing history plot enhancement method and system

Abstract

This invention belongs to the field of acoustic signal and information processing technology, and particularly relates to a method and system for enhancing acoustic azimuth history maps. The method includes: extracting potential targets by using open peak extraction on each grid of the temporal azimuth history map; for the extracted potential targets, defining a suitable size to construct primitives U(θ). j An adaptive two-dimensional convolution kernel is constructed using an unsupervised method. This kernel is then used to perform two-dimensional convolution on the time-location history map, thereby enhancing the acoustic location history map. The method described in this invention can effectively enhance weak acoustic targets in the time-location history map and reduce environmental noise.

Description

Technical Field

[0001] This invention belongs to the field of acoustic signal and information processing technology, and particularly relates to an acoustic orientation history map enhancement method and system. Background Technology

[0002] In underwater passive detection, time-azimuth history maps (TAHs) serve as a crucial visualization tool for tracking the trajectories of underwater targets. In a THA, the horizontal axis represents azimuth, and the vertical axis represents time. High-intensity regions correspond to strong signals returning from a specific direction at a specific time, indicating the presence and movement of the target, thus enabling target detection and tracking. The goal of underwater passive detection technology is to enhance target detection capabilities, and one of its key pursuits is to expand the detection range. Distant targets are often weak targets in THA. Due to the complexity of the marine environment, weak targets are often obscured by strong targets or drowned out by environmental noise. Therefore, enhancing weak acoustic targets in THA and reducing environmental noise in THA has become a research hotspot. The acquisition method for THA can typically be represented by the following steps:

[0003] 1. Acoustic sensor array received signal interception: At certain time intervals, intercept a fixed length of the acoustic sensor array received signal s(t).

[0004] 2. Calculate the azimuth difference:

[0005]

[0006] Where, Δt j d represents the time of arrival between array element j and the reference array element. j θ represents the distance between array element j and the reference array element. j Let j represent the azimuth angle of array element j relative to the reference array element, and c represent the speed of sound.

[0007] 3. Perform time compensation on the signal:

[0008] s j (t)=s(t-Δt j )

[0009] Among them, s j (t) represents the received signal of array element j after time compensation.

[0010] 4. Calculate the synthesized beamforming:

[0011]

[0012] Where, b(θ) j ) indicates the azimuth angle θ j Synthetic beam at ωj The weight of array element j is represented by N, and the number of array elements is represented by N.

[0013] 5. Calculate the azimuth energy spectrum:

[0014] E(θ j )=|b(θ j )| 2

[0015] Where, E(θ) j ) indicates the azimuth angle θ j Beam energy at that location.

[0016] 6. Construct a time-azimuth history map: Establish a time-azimuth spatial coordinate system, using grayscale values ​​to represent the beam energy amplitude E(θ) at the corresponding <time-azimuth>. j By using ,t), a time-location timeline can be obtained.

[0017] Due to the complexity of the marine environment, weak targets are often obscured by strong targets or drowned out by environmental noise. There is an urgent need for a method to enhance weak acoustic targets in time-azimuth history maps and reduce environmental noise on time-azimuth history maps. Summary of the Invention

[0018] The purpose of this invention is to overcome the shortcomings of the prior art and to propose an acoustic orientation history map enhancement method and system.

[0019] In view of this, the present invention proposes an acoustic orientation history map enhancement method, comprising:

[0020] Step 1: Use open peak extraction on each grid of the time-location history map to obtain potential targets;

[0021] Step 2: For the extracted potential targets, define appropriate sizes to construct primitives U(θ). j Adaptive two-dimensional convolution kernels are constructed using unsupervised methods (t).

[0022] Step 3: Use the constructed adaptive 2D convolution kernel to perform 2D convolution on the temporal azimuth history map to enhance the acoustic azimuth history map.

[0023] Preferably, step 1 includes:

[0024] For the acoustic orientation history diagram (θ) j ,t) grid, if and Then the grid contains a potential target, where θ j-1 θ j and θ j+1 These represent the azimuth angles of array elements j-1, j, and j+1 relative to the reference array element, respectively. and They represent (θ) j-1 ,t),(θ j ,t) and (θ) j+1 The peak value of the grid is denoted by t, where t represents time t.

[0025] Preferably, step 2 includes:

[0026] Step 2-1: Apply the directivity function of the conventional beamforming method Differentiate with respect to θ, where N represents the number of array elements, d represents the distance between array elements, and θ represents the azimuth angle; solve for the first positive root using numerical methods, which is the distance D from the main lobe to the side lobe. side The kernel width Δθ is 2D side ;

[0027] Step 2-2: Use clustering to obtain cluster labels and cluster centers;

[0028] Steps 2-3: Using a strategy based on SSE multi-derivatives, for each number of clusters k, execute the clustering algorithm, calculate the total squared error SSE for that value of k, and dynamically adjust the number of clusters;

[0029] For the primitive set E, in each iteration, the i-th primitive e is calculated according to the following formula. i with the j-th cluster center Correlation coefficient between and e i Assign to the cluster center with the highest correlation coefficient;

[0030]

[0031] Where P is the dimension of the convolution kernel. and These represent the p-th dimension values ​​of the primitive and the cluster center, respectively.

[0032] For each cluster j, its cluster center at time t+1 is... for:

[0033]

[0034] Where, S j It is the set of primitive samples assigned to cluster j, |S j | is the number of primitives assigned to cluster j;

[0035] The iteration stops when the change in cluster center is less than the preset threshold ∈, and the final cluster label and cluster center are output.

[0036] Preferably, steps 2-3 include:

[0037] For each cluster number k, the total squared error SSE for that k value is calculated. Using a strategy based on the multiple derivatives of SSE, the first derivative SSE′(k), the second derivative SSE″(k), and the third derivative SSE″′(k) of SSE are calculated respectively. The number of clusters is dynamically adjusted and convergence is determined using the following conditions:

[0038]

[0039] Preferably, step 3 includes:

[0040] Mirror-fill a given BTR matrix B(θ,t) of size P×Q in the time and angle dimensions, with a fill width of w=min(P,Q);

[0041] The clustering yields an adaptive convolution kernel K of size p×q. Adaptive (θ k ,t k Rotate 180°;

[0042] The convolution matrix B is calculated according to the following formula. padded (θ,t):

[0043]

[0044] Where p is the width of the convolution kernel, q is the length of the convolution kernel, and K′ Adaptive (-i,-j) represents the convolution kernel θ after 180° rotation, and t represents the azimuth angle and time, respectively.

[0045] Remove the padding boundaries to restore the original matrix size.

[0046] On the other hand, the present invention provides an acoustic orientation history map enhancement system, comprising:

[0047] The peak extraction module is used to perform open peak extraction on each grid of the time-azimuth history map to obtain potential targets;

[0048] A two-dimensional convolution kernel building module is used to define a suitable size for building primitives U(θ) for the extracted potential targets. j Adaptive two-dimensional convolution kernels are constructed using unsupervised methods (t).

[0049] The enhancement module is used to perform two-dimensional convolution on the time-location history map using the constructed adaptive two-dimensional convolution kernel, thereby enhancing the acoustic location history map.

[0050] Compared with the prior art, the advantages of the present invention are:

[0051] Due to the complexity of ocean noise and the limitations of array aperture, azimuth-time maps (BTR) obtained by traditional beamforming (CBF) typically suffer from low signal-to-noise ratio (SNR) and large main lobe width. To address this issue, this invention proposes an adaptive two-dimensional convolution kernel construction method. This method utilizes an improved k-means clustering algorithm to extract adaptive main lobe visual patterns from historical BTR data as the convolution kernel. Experimental results demonstrate that this method can effectively reduce noise levels in multi-target environments. Attached Figure Description

[0052] Figure 1 It is the original time and location history diagram;

[0053] Figure 2 It is a peak extraction image;

[0054] Figure 3 These are the squared errors and their first, second, and third derivatives for different numbers of clusters;

[0055] Figure 4(a) shows the base of the extracted real target, Figure 4(b) shows the right side lobe, and Figure 4(c) shows the left side lobe.

[0056] Figure 5 This is the result of convolution using real-time two-dimensional convolution kernels;

[0057] Figure 6 This is a comparison chart of single-shot beam response. Detailed Implementation

[0058] This invention introduces an adaptive convolutional kernel construction method for temporal azimuth history maps, which enhances weak acoustic targets and suppresses environmental noise in the temporal azimuth history maps, providing meaningful guidance for subsequent automatic detection and tracking of acoustic targets. It mainly includes the following steps:

[0059] 1. Open peak extraction

[0060] 2. Construct adaptive two-dimensional convolution kernels.

[0061] 3. Use the constructed adaptive 2D convolution kernel to perform 2D convolution on the time-location history map.

[0062] Due to the complexity of the marine environment, weak targets are often obscured by strong targets or drowned out by environmental noise. This invention can effectively enhance weak acoustic targets in time-azimuth history maps and reduce environmental noise on time-azimuth history maps.

[0063] The specific steps are as follows:

[0064] (I) Open Peak Extraction

[0065] 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 specific calculation method for the determination map P is as follows:

[0066]

[0067] (II) Constructing Convolution Kernels

[0068] 1. Adaptive convolution kernel size:

[0069] To construct an adaptive 2D convolutional kernel, we need to calculate the kernel size based on the directivity function of conventional beamforming algorithms in array signal processing, and use an improved K-means algorithm to obtain the visual pattern of the main lobe as the adaptive 2D convolutional kernel. To minimize sidelobe interference and ensure the main lobe is at the visual center of the convolutional kernel, the kernel width is set to twice the distance from the main lobe to the sidelobe. According to conventional beamforming methods, to calculate the distance from the main lobe to the sidelobe, we need to differentiate the directivity function θ of the conventional beamforming method with respect to θ, ignoring the absolute value and constant factor A. The derivative equation is as follows:

[0070]

[0071] The first positive root of the equation, which is the distance D from the main lobe to the side lobe, is obtained by numerical methods. side The kernel width Δθ is defined as:

[0072] Δθ=2D side

[0073] For potential targets extracted from the time-location history map, define an appropriate size to construct the primitive U(θ). j Various convolution kernels are constructed using unsupervised methods.

[0074] 2. Clustering

[0075] (1). Matching: For the primitive set E = {e1, e2, ..., e...} m In each iteration, it is calculated to be related to each cluster center. Correlation coefficient between and the primitive e i Assign it to the cluster center with the highest correlation coefficient.

[0076] The correlation coefficient is calculated using the following formula:

[0077]

[0078] Where P is the dimension of the convolution kernel. and represents the p-th dimension value of the primitive and the cluster center, respectively.

[0079] (2). Update: For each cluster j, its new center is updated as follows:

[0080]

[0081] Where, S j It is the set of primitive samples assigned to cluster j, |S j | is the number of primitives assigned to cluster j.

[0082] (3) Convergence check: The iteration stops when the change in the cluster center is less than the preset threshold ∈.

[0083]

[0084] (4). Output: After the iteration converges, output the final cluster label and cluster center.

[0085] 3. Automatically determine the number of clusters.

[0086] Theoretically, the number of visual templates on the BTR should correspond to the main lobe, left and right side lobes, and various random noise patterns. Regarding the number of noise classifications, a strategy based on the multiple derivatives of SSE can be used. By analyzing the first, second, and third derivatives of SSE, an appropriate number of clusters can be automatically determined to optimize the clustering results. For each cluster number k, the clustering algorithm is executed, and the total squared error SSE for that k value is calculated. The following conditions are used to dynamically adjust the number of clusters and determine convergence:

[0087]

[0088] (III) Two-dimensional convolution

[0089] Given a BTR matrix B(θ,t) of size P×Q and an adaptive convolution kernel K of size p×q obtained through clustering. Adaptive (θ k ,t k Perform two-dimensional convolution operations through the following steps:

[0090] 1. Mirror Fill: Mirror the BTR matrix along the time (vertical axis) and angle (horizontal axis) dimensions, with a fill width of w = min(P,Q).

[0091] 2. Convolution kernel rotation: Rotate K Adaptive (θ k ,t k Rotate 180°.

[0092] 3. Perform two-dimensional convolution operation:

[0093] Calculate the convolution matrix:

[0094]

[0095] 4. Trim: Remove padding boundaries and restore the original matrix size.

[0096] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0097] Example 1

[0098] Embodiment 1 of the present invention proposes an acoustic azimuth history map enhancement method, wherein the temporal azimuth history map to be processed is as follows: Figure 1 As shown. The implementation steps of this invention are as follows:

[0099] (I) Open Peak Extraction

[0100] The presence of a potential target within the (θ,t) grid is determined using logical true and false, where true indicates the presence of a potential target and false indicates the absence of one. The decision map P is extracted, and its specific calculation method is as follows:

[0101]

[0102] The obtained decision spectrum P is as follows Figure 2 As shown, each pixel is labeled as having a potential target (true) or not having a potential target (false).

[0103] (ii) Constructing adaptive convolution kernels

[0104] 1. Adaptive convolution kernel size:

[0105] To construct an adaptive 2D convolutional kernel, we need to calculate the kernel size based on the directivity function of conventional beamforming algorithms in array signal processing, and use an improved K-means algorithm to obtain the visual pattern of the main lobe as the adaptive 2D convolutional kernel. To minimize sidelobe interference and ensure the main lobe is at the visual center of the convolutional kernel, the kernel width is set to twice the distance from the main lobe to the sidelobe. According to conventional beamforming methods, to calculate the distance from the main lobe to the sidelobe, we need to differentiate the directivity function θ of the conventional beamforming method with respect to θ, ignoring the absolute value and constant factor A. The derivative equation is as follows:

[0106]

[0107] The first positive root of the equation, which is the distance D from the main lobe to the side lobe, is obtained by numerical methods. side The kernel width Δθ is defined as:

[0108] Δθ=2D side

[0109] In one embodiment, the signal bandwidth used is 20-380Hz, and the receiving array is a uniform linear array consisting of 256 elements spaced 1.5m apart, used for passive detection. We selected the center frequency of this broadband signal when f = 200Hz, d = 1.5m, and N = 256, because the angular resolution of the time-azimuth history plot used is 0.5°, and the distance D from the main lobe to the side lobe is... side Defined as the position of the first side lobe maximum, due to the formula

[0110]

[0111] There is no analytical solution, but the distance D from the main lobe to the side lobe can be obtained through numerical simulation. side ≈2.3°. Therefore, the width of the adaptive convolution kernel should be 9 pixels. For the potential targets extracted from the temporal orientation map, primitive U(θ) is constructed according to a size of (9,9). j ,t).

[0112] 2. Clustering

[0113] (1). Matching: For the primitive set E = {e1, e2, ..., e...} m In each iteration, it is calculated to be related to each cluster center. Correlation coefficient between and the primitive e i Assign it to the cluster center with the highest correlation coefficient.

[0114] The correlation coefficient is calculated using the following formula:

[0115]

[0116] Where P is the dimension of the convolution kernel. and represents the p-th dimension value of the primitive and the cluster center, respectively.

[0117] (2). Update: For each cluster j, its new center is updated as follows:

[0118]

[0119] Where, S j It is the set of primitive samples assigned to cluster j, |S j | is the number of primitives assigned to cluster j.

[0120] (3) Convergence check: The iteration stops when the change in the cluster center is less than the preset threshold ∈.

[0121]

[0122] (4). Output: After the iteration converges, output the final cluster label and cluster center.

[0123] 3. Automatically determine the number of clusters

[0124] Calculate SSE: Whenever the cluster centers are updated, calculate the total squared error SSE and its derivative for the current number of clusters k.

[0125] Determine the optimal value of k: If SSE′(k)<0 and SSE″(k)<0, continue with the current number of clusters; if SSE″′(k)>0 and is the maximum value, reduce the number of clusters.

[0126] Choose k that satisfies the following condition as the cluster number:

[0127] 1. SSE ′ (k)<0

[0128] 2.SSE ″ (k)<0

[0129] 3. SSE″′(k) > 0 and is the largest among these k values.

[0130] Assuming the minimum number of clusters is 3, the resulting graphs of the squared error and its first, second, and third derivatives are as follows: Figure 3 As shown, except when the number of clusters is 1, the third derivative is the largest when the number of clusters is 3, and both the first and second derivatives are negative. Therefore, the adaptively selected number of clusters is 3.

[0131] The basis constructed using clustering algorithms is shown in Figure 4(a), which is the basis of the real target, namely the extracted adaptive two-dimensional convolution kernel. Figure 4(b) shows the right side lobe, and Figure 4(c) shows the left side lobe.

[0132] (III) Two-dimensional convolution

[0133] Given a BTR matrix B(θ,t) of size 600×361 and an adaptive convolution kernel K of size 9×9 obtained through clustering. Adaptive (θ k ,t k Perform two-dimensional convolution operations through the following steps:

[0134] 1. Mirror Fill: Mirror fill the BTR matrix in the time (vertical axis) and angle (horizontal axis) dimensions, with a fill width of w = min(P,Q) = 361;

[0135] 2. Convolution kernel rotation: Rotate K Adaptive (θ k ,t k Rotate 180°;

[0136] 3. Perform two-dimensional convolution operation:

[0137] Calculate the convolution matrix:

[0138]

[0139] 4. Trim: Remove padding boundaries and restore the original matrix size.

[0140] Test results:

[0141] The result of convolution using a real-time 2D convolution kernel is shown in the image below. Figure 5 As shown in the figure, the comparison results of the single-shot data for the 305th shot are as follows. Figure 6 As shown, compared with the original beam response, the main lobe width obtained by using our proposed two-dimensional deconvolution method is narrower, the background noise is almost completely suppressed, and it can enhance weak targets.

[0142] Example 2

[0143] Embodiment 2 of the present invention provides an acoustic orientation history map enhancement system, implemented based on the method of Embodiment 1, the system comprising:

[0144] The peak extraction module is used to perform open peak extraction on each grid of the time-azimuth history map to obtain potential targets;

[0145] A two-dimensional convolution kernel building module is used to define a suitable size for building primitives U(θ) for the extracted potential targets. j Adaptive two-dimensional convolution kernels are constructed using unsupervised methods (t).

[0146] The enhancement module is used to perform two-dimensional convolution on the time-location history map using the constructed adaptive two-dimensional convolution kernel, thereby enhancing the acoustic location history map.

[0147] It is worth noting that in the embodiments of the above system, the modules included are divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional module are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0148] 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 enhancing acoustic orientation history maps, comprising: Step 1: Use open peak extraction on each grid of the time-location history map to obtain potential targets; Step 2: For the extracted potential targets, define appropriate sizes to construct primitives U(θ). j Adaptive two-dimensional convolution kernels are constructed using unsupervised methods (t). Step 3: Use the constructed adaptive 2D convolution kernel to perform 2D convolution on the temporal azimuth history map to enhance the acoustic azimuth history map.

2. The acoustic orientation history map enhancement method according to claim 1, characterized in that, Step 1 includes: For the acoustic orientation history diagram (θ) j ,t) grid, if and Then the grid contains a potential target, where θ j-1 θ j and θ j+1 These represent the azimuth angles of array elements j-1, j, and j+1 relative to the reference array element, respectively. and They represent (θ) j-1 ,t),(θ j ,t) and (θ) j+1 The peak value of the grid is denoted by t, where t represents time t.

3. The acoustic orientation history map enhancement method according to claim 1, characterized in that, Step 2 includes: Step 2-1: Apply the directivity function of the conventional beamforming method Differentiate with respect to θ, where N represents the number of array elements, d represents the distance between array elements, and θ represents the azimuth angle; solve for the first positive root using numerical methods, which is the distance D from the main lobe to the side lobe. side The kernel width Δθ is 2D side ; Step 2-2: Use clustering to obtain cluster labels and cluster centers; Steps 2-3: Using a strategy based on SSE multi-derivatives, for each number of clusters k, execute the clustering algorithm, calculate the total squared error SSE for that value of k, and dynamically adjust the number of clusters; For the primitive set E, in each iteration, the i-th primitive e is calculated according to the following formula. i with the j-th cluster center Correlation coefficient between and e i Assign to the cluster center with the highest correlation coefficient; Where P is the dimension of the convolution kernel. and These represent the p-th dimension values ​​of the primitive and the cluster center, respectively. For each cluster j, its cluster center at time t+1 is... for: Where, S j It is the set of primitive samples assigned to cluster j, |S j | is the number of primitives assigned to cluster j; The iteration stops when the change in cluster center is less than the preset threshold ∈, and the final cluster label and cluster center are output.

4. The acoustic orientation history map enhancement method according to claim 3, characterized in that, Steps 2-3 include: For each cluster number k, the total squared error SSE for that k value is calculated. Using a strategy based on the multiple derivatives of SSE, the first derivative SSE′(k), the second derivative SSE″(k), and the third derivative SSE″′(k) of SSE are calculated respectively. The number of clusters is dynamically adjusted and convergence is determined using the following conditions:

5. The acoustic orientation history map enhancement method according to claim 1, characterized in that, Step 3 includes: Mirror-fill a given BTR matrix B(θ,t) of size P×Q in the time and angle dimensions, with a fill width of w=min(P,Q); The clustering yields an adaptive convolution kernel K of size p×q. Adaptive (θ k ,t k Rotate 180°; The convolution matrix B is calculated according to the following formula. padded (θ,t): Where p is the width of the convolution kernel, q is the length of the convolution kernel, and K′ Adaptive (-i,-j) represents the convolution kernel θ after 180° rotation, and t represents the azimuth angle and time, respectively. Remove the padding boundaries to restore the original matrix size.

6. An acoustic orientation history map enhancement system, characterized in that, include: The peak extraction module is used to perform open peak extraction on each grid of the time-azimuth history map to obtain potential targets; A two-dimensional convolution kernel building module is used to define a suitable size for building primitives U(θ) for the extracted potential targets. j Adaptive two-dimensional convolution kernels are constructed using unsupervised methods (t). and The enhancement module is used to perform two-dimensional convolution on the time-location history map using the constructed adaptive two-dimensional convolution kernel, thereby enhancing the acoustic location history map.

Images