Orthogonal matching pursuit sound source identification method based on novel atom selection criterion

Abstract

The invention provides an orthogonal matching pursuit sound source identification method based on a novel atom selection criterion, and belongs to the field of sound source identification and localization. A sound source identification model is constructed, and sound pressure data and unknown sound source intensity data are acquired; constructing a transfer function matrix G; obtaining relational data between the sound pressure data and the source intensity data of the unknown sound source, initializing a residual vector as a sound pressure data vector, and indexing a support set as an empty set; measuring the correlation between the transfer function matrix G and the residual vector by using a Tanimoto coefficient matching criterion, screening out a best matching atom, and updating a support set index; corresponding atoms are taken out through index serial numbers in the support set index to form an atom support set matrix, and a sound source intensity vector is solved; updating the residual error vector and the number of iterations, returning to continue execution, stopping iteration when the residual error is smaller than or equal to a set threshold value or the number of iterations k is equal to K, outputting a sound source intensity vector final solution, and performing sound source identification and positioning based on the final solution; and the positioning resolution of sound source identification is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of sound source recognition and positioning, and in particular to an orthogonal matching pursuit sound source recognition method based on a novel atom selection criterion. Background Art

[0002] In recent years, near-field acoustic holography and beamforming technologies have played an important role in acoustic engineering such as noise source identification and positioning, sound field separation, sound field visualization, and speech enhancement. These two technologies are not affected by the shape and size of the sound source, and can easily establish mathematical models and construct transfer matrices. The implementation of near-field acoustic holography and beamforming technologies is mostly based on traditional acoustic arrays such as planar uniform arrays, and data acquisition is based on the Nyquist sampling framework. In order to ensure that the collected signal does not have spectral aliasing, the sampling frequency must be greater than 2 times the highest frequency of the signal when using the above regular array for sound field data acquisition. In order to obtain higher recognition accuracy, the above traditional array requires more microphones, has a larger spatial complexity, increases the measurement cost, and a large amount of measurement data increases the difficulty of storage and transmission. Compressed sensing technology breaks through the Nyquist sampling theorem. It uses the sparsity of the signal in the transform domain, can use a smaller frequency sampling to retain all the information of the original signal, and realizes high-precision reconstruction of the original signal, thereby significantly reducing the number of sampling sensors and reducing the system measurement cost. In the reconstruction of the original sparse signal based on the compressed sensing theory, the orthogonal matching pursuit algorithm (OMP) is the most commonly used reconstruction method. The OMP algorithm has the advantages of small imaging side lobes, few ghost images, and fast calculation speed. However, its shortcomings can be found from the actual application of the OMP algorithm. The OMP algorithm and various compressed sensing greedy reconstruction algorithms are based on the absolute inner product principle to expand the atomic support set. This criterion measures the atomic correlation by measuring the cosine similarity between matrix atoms. When the sound source frequency decreases, the phase of the Green function will change slowly with the distance, which will cause the correlation between atoms to gradually increase and the two columns of atoms to gradually converge; and when the focus surface grid is too densely divided, the spacing between the focus surface grid points corresponding to two adjacent atoms is very small, which will also cause the two columns of atoms to have a high correlation. When the inner product of the atom corresponding to the actual sound source and the residual is less than the inner product value of the adjacent atom and the residual, the absolute inner product matching criterion cannot expand the correct atom to the support set, which will inevitably lead to the failure of sound source identification. In addition, coherence peaks will appear between coherent sound sources with lower frequencies and closer distances, causing the inner product of the atom corresponding to the coherence peak position in the measurement matrix and the signal to even exceed the inner product of the atom corresponding to the actual sound source and the original signal, resulting in the algorithm incorrectly selecting the atom corresponding to the coherence peak position in the first iteration, causing the sound source identification to fail.

[0003] In the prior art, the Chinese invention patent application with publication number CN116008911A, "An Orthogonal Matching Pursuit Sound Source Identification Method Based on a Novel Atom Matching Criteria", addresses the defect that the Dice coefficient matching criterion cannot evenly measure the correlation between two atomic vectors. By improving the atomic matching criterion of the orthogonal matching pursuit algorithm, the Euclidean distance is used to represent the spatial length of the atomic vector. On the basis of the original Dice coefficient matching criterion, the weighting of the spatial length is appropriately reduced, and the matching accuracy of the atomic support set screening is improved. To a certain extent, it overcomes the problem of incorrect atom selection caused by strong correlation between atoms when the sound source points are too close or the sound source frequency is low. However, after analyzing the sound source identification method through simulation experiments, it was found that Figure 1 As shown in FIG. 1 , when two sound sources are identified, the sound source identification method recognizes two sound sources, but gives a completely wrong position estimate, such as Figure 2 As shown in the figure, when the distance between coaxial adjacent sound sources is expanded to 0.4m, the coherence condition is relatively loose, and the sound source identification method still cannot accurately identify the positions of all sound sources. After analysis, although the Dice coefficient matching criterion increases the measurement of the interatomic space length, the Dice coefficient may not be able to effectively distinguish closely arranged multiple sound sources due to the influence of the union size, giving an incorrect sound source identification position. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an orthogonal matching pursuit sound source identification method based on a new atom selection criterion in view of the deficiencies of the above-mentioned prior art. The method aims to improve the algorithm reconstruction performance in an environment with strong correlation between atoms, and avoid the problem of low sound source identification resolution caused by the enhanced correlation between adjacent atoms in the transfer matrix due to the encryption of the focusing surface grid and the change of the signal phase at low frequency.

[0005] The present invention solves the above technical problems through the following technical solutions: an orthogonal matching pursuit sound source identification method based on a new atom selection criterion, comprising the following steps:

[0006] Step 1: construct a sound source identification model, and obtain sound pressure data and unknown sound source intensity data based on the sound source identification model;

[0007] Step 2: Based on the sound source identification model, the transfer function matrix G is constructed according to the free field Green's function;

[0008] Step 3: Based on the transfer function matrix G, the relationship data between the sound pressure data and the unknown sound source strength data is obtained, and the residual vector is initialized as the sound pressure data vector, and the support set index is an empty set;

[0009] Step 4: Use the Tanimoto coefficient matching criterion to measure the correlation between the transfer function matrix G and the residual vector, select the best matching atom, add the best matching atom to the support set, and update the support set index;

[0010] Step 5: Take out the corresponding atom through the index number in the support set index to form the atomic support set matrix, and solve the sound source intensity vector;

[0011] Step 6, update the residual vector and the number of iterations, return to step 4 to continue, when the residual is less than or equal to the set threshold or the number of iterations k = K, K is the maximum number of iterations, then stop the iteration, output the final solution of the sound source intensity vector, and perform sound source identification and positioning based on the final solution of the sound source intensity vector.

[0012] Preferably, the sound source identification model includes a measurement plane W and a focusing plane T, and step 1 includes:

[0013] 1.1. A rectangular measurement area with a length of Lx and a width of Ly is set in the measurement surface W, and M sensors are arranged in a rectangular array at grid points of x rows and y columns in the rectangular measurement area to form a sensor measurement array for obtaining sound pressure data;

[0014] 1.2. Discretize the rectangular focusing area with a length of Nx and a width of Ny on the focusing surface T into N focusing points evenly distributed in x1 rows and y1 columns. The N focusing points are all potential sound source positions, which are used to obtain the source strength data of unknown sound sources.

[0015] Preferably, step 2 includes: acquiring sampling point data based on the measurement surface W, acquiring sound source point data based on the focusing surface T, and constructing a transfer function matrix G between the sampling point data and the sound source point data according to the free field Green's function. The formula of the transfer function matrix G is as follows:

[0016]

[0017] Where g(m,n) is the matrix element in the transfer function matrix G, which represents the transfer function between the mth grid point and the nth focus point of the sensor measurement array. The formula of the matrix element g(m,n) is as follows:

[0018]

[0019] Where j is an imaginary unit, c is the speed of sound, and d is mn It represents the distance between the mth grid point and the nth focusing point of the sensor measurement array, f is the sound source frequency, m = 1, 2, 3…M, n = 1, 2, 3…N.

[0020] Preferably, step 3 comprises: based on the transfer function matrix G, using the measurement array to obtain the sound pressure data vector p and the unknown sound source intensity vector q in the sound field generated by the K sound sources, the relationship data is expressed by formula (2):

[0021]

[0022] Among them, r m represents the coordinate vector of the mth measurement point on the measurement surface W, r n represents the coordinate vector of the nth focal point on the focusing plane T, and ξ is the noise vector contained in the measured sound pressure vector data.

[0023] Preferably, step 4 comprises:

[0024] 4.1. Define the Tanimoto coefficient matching criterion as:

[0025]

[0026] 4.2. Use the Tanimoto coefficient matching criterion to select the best matching atom and find the position of the best matching atom in the transfer function matrix G according to formula (3):

[0027] j k =argmax|Tanimoto(G i ,r k-1 )|,j=1,2,3…,N (3)

[0028] 4.3. Update the support set index Γ0: Γ0 = Γ k-1 ∪j k , where Γ k-1 is the support set index of the most matching atom set that has been screened in the previous iteration, j k The support set index of the best matching atom in the transfer function matrix for the current iteration.

[0029] Preferably, the sound source intensity vector in step 5 is expressed by formula (4):

[0030]

[0031] Where inv represents the inverse operation of the matrix. Represents the matrix Request transpose.

[0032] Preferably, the residual vector r in step 6 k for:

[0033]

[0034] The advantages provided by the present invention are:

[0035] 1. The present invention introduces the Tanimoto coefficient matching criterion and considers the similarity of the angle direction between the atoms of the measurement matrix and the single element in the vector at the same time. Compared with the traditional cosine similarity which only uses the cosine value of the angle between two vectors in the vector space as the standard for measuring the similarity of the two vectors, the Tanimoto coefficient matching criterion pays more attention to the comprehensive similarity of the two vectors. By emphasizing the common elements between the sets, it helps to improve the positioning resolution of sound source identification and overcome the problem of incorrect atom selection caused by the enhanced correlation between atoms when the distance between adjacent sound sources is close or the analysis frequency is low.

[0036] 2. Compared with the traditional inner product matching criterion based on cosine similarity, the present invention improves the accuracy of atomic support set screening and matching. The OMP algorithm can find atoms similar to the sound source signal more quickly, thereby accelerating the process of sound source identification and improving the reconstruction performance of the OMP algorithm, thereby achieving accurate screening of atoms, and then using these accurately screened atoms to achieve high-resolution positioning of the sound source, which also expands the application scope of the OMP algorithm.

[0037] 3. The Tanimoto coefficient of the present invention removes the influence of the inner product of two vectors. Removing the autocorrelation term can reduce the influence of the signal's own energy on the correlation calculation, thereby more accurately reflecting the relationship between the two acoustic vectors and improving the robustness of sound source identification. Symmetrical processing of the Tanimoto coefficient can avoid numerical instability problems caused by extreme values ​​of a single acoustic quantity, thereby improving the numerical stability of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a sound source localization effect diagram of the existing sound source identification method when the sound source frequency is 1500Hz;

[0039] Figure 2 The four sound source localization effect diagrams of the existing sound source identification method when the sound source frequency is 1500Hz;

[0040] Figure 3 A flow chart of the orthogonal matching pursuit sound source identification method based on a novel atom selection criterion provided by the present invention;

[0041] Figure 4 Schematic diagram of a sound source identification model composed of a microphone array and a focusing surface in the present invention;

[0042] Figure 5 A diagram of the measurement array used in the present invention;

[0043] Figure 6a This is the sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 700Hz;

[0044] Figure 6b This is a sound source localization effect diagram of the present invention when the sound source frequency is 700 Hz;

[0045] Figure 7a This is the sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 2000Hz;

[0046] Figure 7b This is a sound source localization effect diagram of the present invention when the sound source frequency is 2000 Hz;

[0047] Figure 8a The sound source localization effect diagram when the grid spacing on the focal plane is 0.01m when positioning using the standard OMP algorithm;

[0048] Figure 8b This is a sound source localization effect diagram of the present invention when the grid spacing on the focusing surface is 0.01m;

[0049] Figure 9a The positioning effect diagram of four similar sound sources with coherent peaks when using the standard OMP algorithm for positioning;

[0050] Figure 9b This is a diagram showing the positioning effect of the present invention on four similar sound sources with coherent peaks;

[0051] Fig.10 This is a sound source localization effect diagram of the sound source identification method of the present invention when the sound source frequency is 1500 Hz;

[0052] Fig.11 This is a diagram showing the localization effects of four sound sources of the sound source identification method of the present invention when the sound source frequency is 1500 Hz. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the technical solution of the present invention is clearly and completely described below in combination with specific embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The orthogonal matching pursuit algorithm is a fast and effective signal reconstruction algorithm in compressed sensing theory. It selects the best atoms in an iterative manner to approximate the original signal. The sound source identification method of the present invention proposes the Tanimoto coefficient matching criterion. The Tanimoto coefficient is often used in chemistry and bioinformatics to quantify the correlation between two sets, such as the similarity between molecular fingerprints and gene expression data. The present invention draws on the application of the matching of biological and chemical molecules to practical physics, and utilizes the superiority of the Tanimoto coefficient combined with the orthogonal matching pursuit algorithm for sound source identification. An orthogonal matching pursuit sound source identification method based on a new atom selection criterion is designed. Through simulation experiments, the sound source identification method of the present invention is compared with the existing sound source identification method. Finally, it is found that integrating the Tanimoto coefficient into the existing OMP algorithm framework can effectively improve the performance of sound source identification.

[0055] like Figure 1 As shown, this embodiment provides an orthogonal matching pursuit sound source identification method based on a new atom selection criterion, comprising the following steps:

[0056] Step 1: construct a sound source identification model, and obtain sound pressure data and unknown sound source intensity data based on the sound source identification model; the sound source identification model includes a measurement surface W and a focusing surface T, and step 1 includes:

[0057] 1.1. A rectangular measurement area with a length of Lx and a width of Ly is set in the measurement surface W. M sensors are arranged in a rectangular array at grid points of x rows and y columns in the rectangular measurement area to form a sensor measurement array for acquiring sound pressure data.

[0058] 1.2. Discretize the rectangular focusing area with a length of Nx and a width of Ny on the focusing surface T into N focusing points evenly distributed in x1 rows and y1 columns. The N focusing points are all potential sound source positions, which are used to obtain the source strength data of unknown sound sources.

[0059] Step 2: Based on the sound source identification model, the transfer function matrix G is constructed according to the free field Green's function;

[0060] The sampling point data is obtained based on the measurement surface W, the sound source point data is obtained based on the focusing surface T, and the transfer function matrix G between the sampling point data and the sound source point data is constructed according to the free field Green's function. The formula of the transfer function matrix G is as follows:

[0061]

[0062] Where g(m,n) is the matrix element in the transfer function matrix G, which represents the transfer function between the mth grid point and the nth focus point of the sensor measurement array. The formula of the matrix element g(m,n) is as follows:

[0063]

[0064] Where j is an imaginary unit, c is the speed of sound, and d is mn It represents the distance between the mth grid point and the nth focusing point of the sensor measurement array, f is the sound source frequency, m = 1, 2, 3…M, n = 1, 2, 3…N.

[0065] Step 3: Based on the transfer function matrix G, obtain the relationship data between the sound pressure data and the unknown sound source strength data, and initialize the residual vector r0=p, the support set index in Is an empty set.

[0066] The relationship data constructed by using the transfer function matrix G in the sound field generated by K sound sources and the sound pressure data vector p and the unknown sound source intensity vector q is expressed by formula (2):

[0067]

[0068] Among them, r m represents the coordinate vector of the mth measurement point on the measurement surface W, r n represents the coordinate vector of the nth focal point on the focusing plane T, and ξ is the noise vector contained in the measured sound pressure vector data.

[0069] Step 4: Use the Tanimoto coefficient matching criterion to measure the correlation between the transfer function matrix G and the residual vector, select the best matching atom, and add the best matching atom to the support set to update the support set index.

[0070] 4.1. Define the Tanimoto coefficient matching criterion as:

[0071]

[0072] 4.2. Use the Tanimoto coefficient matching criterion to select the best matching atom and find the position of the best matching atom in the transfer function matrix G according to formula (3):

[0073] j k =argmax|Tanimoto(G i ,r k-1 )|,j=1,2,3…,N (3)

[0074] 4.3. Update the support set index Γ0: Γ0 = Γ k-1 ∪j k , where Γ k-1 is the support set index of the most matching atom set that has been screened in the previous iteration, j kThe support set index of the best matching atom in the transfer function matrix for the current iteration.

[0075] Step 5: Take out the corresponding atom through the index number in the support set index to form the atomic support set matrix, and solve the sound source intensity vector;

[0076] The sound source intensity vector is expressed by formula (4):

[0077]

[0078] Where inv represents the inverse operation of the matrix. Represents the matrix Request transpose.

[0079] Step 6, update the residual vector and the number of iterations, return to step 4 to continue, when the residual is less than or equal to the set threshold or the number of iterations k = K, K is the maximum number of iterations, then stop the iteration, output the final solution of the sound source intensity vector, and perform sound source identification and positioning based on the final solution of the sound source intensity vector.

[0080] Residual vector r k It is expressed by formula (5):

[0081]

[0082] While following the traditional sound source localization method, the present invention improves the atom matching criterion of the orthogonal matching pursuit algorithm to achieve accurate sound source localization. The present invention introduces the Tanimoto coefficient matching criterion, and simultaneously considers the similarity of the angle direction between the atoms of the measurement matrix and the single element in the vector. Compared with the traditional cosine similarity, which only uses the cosine value of the angle between two vectors in the vector space as the standard for measuring the similarity of the two vectors, the Tanimoto coefficient matching criterion pays more attention to the comprehensive similarity of the two vectors. By emphasizing the common elements between the sets, it helps to improve the positioning resolution of the sound source identification, and can overcome the problem of atomic selection errors caused by the enhanced correlation between atoms when the adjacent sound sources are close or the analysis frequency is low. Compared with the traditional inner product matching criterion based on cosine similarity, the present invention improves the atomic support set screening matching accuracy, and the OMP algorithm can find atoms similar to the sound source signal faster, thereby accelerating the process of sound source identification, improving the reconstruction performance of the OMP algorithm, thereby achieving the precise screening effect of atoms, and then using these accurately screened atoms to achieve high-resolution positioning of the sound source, and also expanding the application scope of the OMP algorithm.

[0083] The present invention adopts the Tanimoto coefficient matching criterion, and the Tanimoto coefficient removes the influence of the inner product of two vectors, which is the cross-correlation term in the numerator. In general acoustic signal processing, removing the autocorrelation term can reduce the influence of the signal's own energy on the correlation calculation, thereby more accurately reflecting the relationship between the two acoustic vectors and improving the robustness of sound source identification. In addition, the Tanimoto coefficient is defined based on a specific acoustic correlation measure, which is more suitable for processing the characteristics of acoustic signals and takes into account the phase information and frequency characteristics of the acoustic signal. The Tanimoto coefficient focuses more on the similarity in the angle direction, which is conducive to the picking of the correlation of the acoustic vector. The numerator and denominator of the Tanimoto coefficient are both summed, which achieves normalization to a certain extent compared to the Dice coefficient.

[0084] The symmetric processing of the Tanimoto coefficients does not destroy the phase information of the original acoustic vector signal, which is very important for sound source identification based on compressed sensing. The phase difference can provide key information in sound source identification. The symmetric processing of the Tanimoto coefficients can avoid numerical instability problems caused by extreme values ​​of a single acoustic quantity, thereby improving the numerical stability of the algorithm. Compared with the Dice coefficient denominator, which only sums the squares of each and then takes the square root, the symmetry of the Tanimoto coefficient is more obvious and complete. This symmetric processing helps to improve the accuracy and reliability of acoustic correlation measurement because it can better capture the relationship between signals.

[0085] By comparison Fig.10 and Figure 1 , Fig.11 and Figure 2 It can be found that when the signal-to-noise ratio is 20dB, the Dice coefficient may have a greater impact on the accuracy of sound source recognition when the signal-to-noise ratio is low, resulting in a decrease in recognition accuracy. When calculating the Tanimoto coefficient, since the overall information of the vector is taken into account, the influence of noise is effectively suppressed to a certain extent, thereby improving the performance of the sound source recognition system in a low signal-to-noise ratio environment and having better robustness to noise and interference.

[0086] In addition, the present invention also has good medium and low frequency reconstruction performance. In medium and low frequency environments, as the coherence between atoms in the measurement matrix increases, the spatial resolution of the acoustic array decreases, and the traditional OMP algorithm can no longer implement accurate positioning, and it is easy to generate false sound sources and artifacts, while the present invention can still accurately reconstruct the location information of the real sound source. At the same time, the present invention has a wider analysis band and has a greater spatial resolution advantage at low frequencies.

[0087] Simulation experiment 1

[0088] It is verified that compared with the standard OMP algorithm positioning method, the present invention has better mid- and low-frequency reconstruction performance for sparse sound source signals.

[0089] Simulation process: Assume that two monopole point sound sources of equal intensity are located on the focusing plane T, the size of the focusing plane T is 1m×1m, and a total of 21×21 grid points are evenly divided into 441 focusing positions. In order to avoid the two sound sources forming a strong energy coherence peak in the middle position during low-frequency simulation and affecting the simulation results when the single variable is frequency, the coordinate position distribution of the monopole sound source in space is set to (-0.25, 0, 0.2) m, (0.25, 0, 0.2) m. The measurement array is located on the grid points of the measurement surface W. The measurement surface is divided into 21×21 grid points, and the 49 measurement array elements used in the measurement are distributed on the grid points, such as Figure 5 As shown. The distance between the measurement plane W and the focusing plane T is 0.2m. In order to simulate the noise in the actual measurement process, 20dB Gaussian white noise is added to the acoustic signal measured by the array. In order to quantitatively measure the reconstruction effect of the standard OMP algorithm and the improved OMP algorithm, the error definition method is given and represented by the following formula:

[0090]

[0091] Where: err is the source strength reconstruction error, q bul is the source intensity reconstruction vector, q sou is the standard source strength position vector, and ||·||2 is the vector 2-norm.

[0092] Figure 6a This is the sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 700Hz. Figure 6b This is a sound source localization effect diagram of the present invention when the sound source frequency is 700Hz. In the image of the sound source identification result, "*" indicates the actual location of the sound source point. When the sound source frequency is 700Hz, the atoms corresponding to the sound source position and the adjacent atoms have strong coherence, and the traditional OMP algorithm selects the wrong atom, and the sound source identification error is 1.168, causing the sound source identification to fail. However, the OMP algorithm improved by the present invention can still accurately locate the position of the sound source, and the sound source identification error is 0.022. It can be seen that the sound source identification performance of the method of the present invention is better than that of the standard OMP algorithm at medium and low frequencies.

[0093] In order to compare the sound source identification performance of the standard OMP algorithm and the improved OMP algorithm at high frequencies, the sound source frequency was changed to 2000 Hz and the spatial positions of the two monopole sound sources were changed to (-0.1, 0, 0.2) m and (0.1, 0, 0.2) m, while keeping other parameters unchanged. Figure 7a The sound source localization effect diagram of the standard OMP algorithm, the sound source identification error is 0.018; Figure 7bThis is a sound source localization effect diagram of the present invention when the sound source frequency is 2000 Hz, and the sound source recognition error is 0.015; it can be seen that the sound source recognition effect of the present invention at high frequencies is slightly better than that of the standard OMP algorithm.

[0094] Simulation experiment 2

[0095] It is verified that the sound source localization resolution of the method of the present invention is higher than that of the standard OMP algorithm.

[0096] Simulation process: The denser the focus plane is divided, the higher the resolution of sound source identification is in theory. In order to compare the sound source localization resolution of the method of the present invention and the standard OMP algorithm, the focus plane is divided into a uniform grid of 101*101, and the spacing between adjacent grids is 0.01m. The sound source frequency is set to 2000Hz and the coordinate positions of the two monopole sound sources are (-0.05, 0, 0.2)m and (0.05, 0, 0.2)m respectively. The rest of the simulation conditions are the same as those of simulation experiment 1. Figure 8a This is the sound source localization effect diagram when using the standard OMP algorithm.

[0097] Figure 8b The effect diagram of sound source localization after adopting the present invention. When the spacing between adjacent grids on the focal plane is 0.01m, the spacing between the focal points is reduced, resulting in enhanced coherence between atoms in each column of the corresponding transfer matrix. The standard OMP algorithm selects atoms at the weak coherence peak and generates a false sound source, resulting in incorrect sound source identification, with a sound source identification error of 1.352. However, the OMP algorithm improved by the present invention can still accurately locate the position of the sound source, with a sound source identification error of 0.031. It can be seen that the method of the present invention is significantly superior to the standard OMP algorithm in sound source localization resolution.

[0098] Simulation experiment 3

[0099] It is verified that when four sound sources that are close to each other and have coherent peaks are identified, the method of the present invention can effectively identify the four sound sources, and the sound source localization resolution is higher than that of the standard OMP algorithm.

[0100] Simulation process: Four monopole sound sources of equal intensity are placed on the focusing plane, with the coordinate positions of the sound sources being (-0.1, 0, 0.2) m, (0.1, 0, 0.2) m, (0, -0.1, 0.2) m, (0, 0.1, 0.2) m, and the frequency being 1500 Hz. The rest of the simulation settings are the same as those of simulation experiment 1. Figure 9a This is the sound source positioning effect diagram using the standard OMP algorithm. Figure 9bThe figure shows the positioning effect of four similar sound sources with coherent peaks after using the present invention. When there are four coherent sound sources, the standard OMP algorithm can only identify the false sound source corresponding to the coherent peak position, and it is difficult to distinguish the exact position of the four sound sources. In addition, there are sound source artifacts, and the sound source identification error is 1.436, which makes the sound source identification invalid. However, the present invention still accurately identifies the positions of four adjacent sound sources, and the sound source identification error is only 0.045. The sound source identification accuracy is significantly better than the standard OMP algorithm.

[0101] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An orthogonal matching pursuit sound source identification method based on a new atom selection criterion, characterized by: The following steps are involved: Step 1: construct a sound source identification model, and obtain sound pressure data and unknown sound source intensity data based on the sound source identification model; Step 2: Based on the sound source identification model, the transfer function matrix G is constructed according to the free field Green's function; Step 3: Based on the transfer function matrix G, the relationship data between the sound pressure data and the unknown sound source strength data is obtained, and the residual vector is initialized as the sound pressure data vector, and the support set index is an empty set; Step 4: Use the Tanimoto coefficient matching criterion to measure the correlation between the transfer function matrix G and the residual vector, select the best matching atom, add the best matching atom to the support set, and update the support set index; Step 5: Take out the corresponding atom through the index number in the support set index to form the atomic support set matrix, and solve the sound source intensity vector; Step 6, update the residual vector and the number of iterations, return to step 4 to continue, when the residual is less than or equal to the set threshold or the number of iterations k = K, K is the maximum number of iterations, then stop the iteration, output the final solution of the sound source intensity vector, and perform sound source identification and positioning based on the final solution of the sound source intensity vector.

2. The orthogonal matching pursuit sound source identification method based on a novel atom selection criterion according to claim 1 is characterized in that: The sound source identification model includes a measurement plane W and a focusing plane T, and step 1 includes: 1.

1. A rectangular measurement area with a length of Lx and a width of Ly is set in the measurement surface W, and M sensors are arranged in a rectangular array at grid points of x rows and y columns in the rectangular measurement area to form a sensor measurement array for obtaining sound pressure data; 1.

2. Discretize the rectangular focusing area with a length of Nx and a width of Ny on the focusing surface T into N focusing points evenly distributed in x1 rows and y1 columns. The N focusing points are all potential sound source positions, which are used to obtain the source strength data of unknown sound sources.

3. The orthogonal matching pursuit sound source identification method based on the novel atom selection criterion according to claim 2 is characterized in that: The step 2 includes: acquiring sampling point data based on the measurement surface W, acquiring sound source point data based on the focusing surface T, and constructing a transfer function matrix G between the sampling point data and the sound source point data according to the free field Green's function. The formula of the transfer function matrix G is as follows: Where g(m,n) is the matrix element in the transfer function matrix G, which represents the transfer function between the mth grid point and the nth focus point of the sensor measurement array. The formula of the matrix element g(m,n) is as follows: Where j is an imaginary unit, c is the speed of sound, and d is mn It represents the distance between the mth grid point and the nth focusing point of the sensor measurement array, f is the sound source frequency, m = 1, 2, 3…M, n = 1, 2, 3…N.

4. The orthogonal matching pursuit sound source identification method based on a novel atom selection criterion according to claim 1 is characterized in that: The step 3 comprises: based on the transfer function matrix G, using the measurement array to obtain the relationship data between the sound pressure data vector p and the unknown sound source intensity vector q in the sound field generated by the K sound sources, and the relationship data is expressed by formula (2): Among them, r m represents the coordinate vector of the mth measurement point on the measurement surface W, r n represents the coordinate vector of the nth focal point on the focusing plane T, and ξ is the noise vector contained in the measured sound pressure vector data.

5. The orthogonal matching pursuit sound source identification method based on a novel atom selection criterion according to claim 1 is characterized in that: The step 4 comprises: 4.

1. Define the Tanimoto coefficient matching criterion as: 4.

2. Use the Tanimoto coefficient matching criterion to select the best matching atom and find the position of the best matching atom in the transfer function matrix G according to formula (3): j k 1argmax|Tanimoto(G i ,r k-1 )|,jJ1,2,3...,N (3) 4.

3. Update the support set index Γ0: Γ0 = Γ k-1 ∪j k , where Γ k-1 is the support set index of the most matching atom set that has been screened in the previous iteration, j k The support set index of the best matching atom in the transfer function matrix for the current iteration.

6. The orthogonal matching pursuit sound source identification method based on a novel atom selection criterion according to claim 1 is characterized in that: The sound source intensity vector in step 5 is expressed by formula (4): Where inv represents the inverse operation of the matrix. Represents the matrix Request transpose.

7. The orthogonal matching pursuit sound source identification method based on a novel atom selection criterion according to claim 1 is characterized in that: The residual vector r in step 6 k for:

Images