Optimized iterative shrinkage threshold elastic net regularization generalized inverse beam forming noise source positioning identification method

Abstract

The invention relates to an optimized iterative shrinkage threshold elastic network regularization generalized inverse beam forming noise source positioning and recognition method, and solves the problems that an existing noise source positioning and recognition method cannot achieve engineering application, and underwater target noise source positioning and recognition are poor in precision and low in resolution. The method comprises the following steps: constructing an elastic net regularization generalized inverse beam forming recognition noise source model according to the sparsity of a noise source, and solving the recognition noise source model by using an iterative shrinkage threshold algorithm; constructing a regularization matrix according to an iteration result of the iteration shrinkage threshold algorithm in the previous step; and the regularization matrix obtained in the previous step is used for optimizing elastic net regularization generalized inverse beam forming to obtain a new objective function, the new objective function is solved by using an iterative shrinkage threshold algorithm, and a positioning identification result of the noise source is obtained. According to the method, high-precision positioning of the underwater target noise source can be realized, the noise source identification precision and the spatial resolution are improved, and the method has a good engineering application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater detection, and particularly relates to an underwater noise source localization and identification method based on optimized iterative shrinkage threshold elastic net regularization generalized inverse beamforming. Background Technique

[0002] Acoustic stealth is one of the most basic indicators of underwater targets. Conducting research on underwater target noise source localization and identification, such as "L. Jia, G. J. Zhang, Y. Liu, Z. Y. Bai, Y. N. Geng, Y. D. Wu, J. Zhang, W. D. Zhang, Sonar buoy active detection and localization for underwater targets using high-level sound sources and MEMS hydrophone, Measurement, 241(2025).", to determine the spatial positions of the main noise sources is a prerequisite for noise control. The beamforming technology based on the measurement of a hydrophone array is an efficient method for noise source localization and identification. In noise source localization and identification, the number of sound sources is often sparse relative to the scanned spatial region. Therefore, relevant scholars at home and abroad have carried out research on beamforming based on the sparse problem of sound sources, such as S. Y. Jiang, R. X. Jiang, X. S. Liu, B. X. Gu, Y. W. Chen, Probability-Based Complex-Valued Fast Iterative Shrinkage-Thresholding Algorithm for Deconvolution Beamforming, IEEE Journal of Oceanic Engineering, 49(2024)340-351. etc.

[0003] Among many noise source localization and identification methods, Generalized Inverse Beamforming (GIB) (see S.G. Shi, Y. Gao, D.S. Yang, J. Shi, D.Y. Tian, An Improved Generalized Inverse Beamforming-Noise Source Localization Method Using Acoustic Vector Sensor Arrays, IEEE Sensors Journal, 21(2021)16222-16235.) is widely used in the localization and identification of different types of noise sources because of its fewer iterative inversion processes and its ability to effectively reduce the sidelobe level and improve the sound source identification resolution (see R. Merino-Martínez, S. Luesutthiviboon, R. Zamponi, A.R. Carpio, D. Ragni, P. Sijtsma, M. Snellen, C. Schram, Assessment of the accuracy of microphone array methods for aeroacoustic measurements, Journal of Sound and Vibration, 470(2020).). However, in actual measurements, instrument and equipment errors are inevitable. If noise source localization and identification are directly carried out on it, the results will be biased, and there are also ill-posed problems in the inversion process (see W. Li, S. Zhao, C. Zhou, Y. Qin, H. Zhu, S. Li, Improved fast deconvolution algorithms based on functional beamforming for gas leakage sound source imaging, Measurement, 242(2025)116238.).Regarding the ill-posed problem of generalized inverse beamforming, relevant scholars have carried out various studies. For example, Presezniak et al. in "F. Presezniak, P. A. G. Zavala, G. Steenackers, K. Janssens, J. R. F. Arruda, W. Desmet, P. Guillaume, Acoustic source identification using a Generalized Weighted Inverse Beamforming technique, Mechanical Systems and Signal Processing, 32(2012)349 - 358." proposed a weighted generalized inverse beamforming to improve the source identification accuracy. This method uses a weighted pseudo-inverse method and an optimization process. Compared with generalized inverse beamforming, this method has higher source identification accuracy. Zavala et al. in "P. A. G. Zavala, W. De Roeck, K. Janssens, J. R. F. Arruda, P. Sas, W. Desmet, Generalized inverse beamforming with optimized regularization strategy, Mechanical Systems and Signal Processing, 25(2011)928 - 939." proposed an improved generalized inverse beamforming method with an automatic regularization factor and introduced a virtual target grid to obtain source mapping and intensity estimation. Finally, the performance of generalized inverse beamforming with a fixed regularization factor and regularized generalized inverse beamforming was compared using two simple sound sources. Takao et al. in "T. Suzuki, L1 generalized inverse beam-forming algorithm resolving coherent / incoherent, distributed and multipole sources, Journal of Sound and Vibration, 330(2011)5835 - 5851." proposed to introduce l1-norm regularization constraints to solve generalized inverse beamforming, but this method cannot eliminate the influence of noise interference in the inversion process. The convex optimization toolbox CVX is mainly used during the solution, and this toolbox is not portable when applied in engineering software.

[0004] Regarding the problem of sparse signal recovery of sound sources, relevant scholars have proposed using greedy iterative algorithms (see X.D. Han, G.H. Zhao, X.M. Li, T. Shu, W.X. Yu, Sparse signal reconstruction via expanded subspace pursuit, Journal of Applied Remote Sensing, 13(2019).), orthogonal matching pursuit algorithms (see W.Y. Guo, J.G. Han, H.T. Chen, L. Yu, Z. Wu, An adaptive beamforming algorithm for sound source localisation via hybrid compressive sensing reconstruction, Journal of Vibroengineering, 24(2022)591 - 603.), convex optimization algorithms (see Y.H. Li, X.D. Wang, Z. Wang, A novel super-resolution imaging method based on TDI CCD charge transfer and random exposure, Optics Communications, 426(2018)170 - 181.) and iterative threshold algorithms (see X.Y. Liu, X. Kong, L.S. Qiao, J.L. Zhao, Efficient L 1 / 2 Regularization-Based Reconstruction for Photoacoustic Imaging Using Adaptively Iterative Thresholding Algorithm, Journal of Medical Imaging and Health Informatics, 10(2020)1506 - 1514.), etc. Among them, the greedy iterative algorithm obtains a local optimal solution, and its final calculation result depends on the greedy strategy; the orthogonal matching pursuit algorithm has a large amount of calculation and low decomposition accuracy, and cannot extract noise signals under strong noise backgrounds; the convex optimization algorithm mainly uses the convex optimization toolbox CVX, which is not portable in engineering software, severely restricting its application.

[0005] Daubechies et al. proposed the iterative shrinkage thresholding method in "I. Daubechies, M. Defrise, C. De Mol, An iterative thresholding algorithm for linear inverse problems with a sparsity constraint, Communications on Pure and Applied Mathematics, 57(2004)1413 - 1457." to solve the linear inverse problems that appear in signal and image processing. However, the convergence speed of this method is relatively slow. Therefore, Beck et al. proposed the Fast Iterative Shrinkage Thresholding Algorithm (FISTA) in "K. Bredies, An iterative thresholding - like algorithm for inverse problems with sparsity constraints in Banach space, Journal of Inverse and Ill - Posed Problems, 17(2009)19 - 26." While retaining the simple calculation characteristics of the ISTA algorithm, it greatly improves the global convergence speed of the algorithm. Li et al. applied the FISTA method to beamforming sound source identification in "L. Chen, Y. Xiao, T. Yang, Application of the improved fast iterative shrinkage - thresholding algorithms in sound source localization, Applied Acoustics, 180(2021)108101." This method has higher computational efficiency and faster convergence speed. Shen et al. proposed an improved fast iterative shrinkage thresholding algorithm based on Fourier (FFT - IFISTA) in "L. B. Shen, Z. G. Chu, Y. X. Zhang, Y. Yang, A novel Fourier - based deconvolution algorithm with improved efficiency and convergence, Journal of Low Frequency Noise Vibration and Active Control, 39(2020)866 - 878." This algorithm can select appropriate weighting coefficients to reduce the main lobe and improve the resolution.

[0006] In summary, relevant scholars have conducted a large number of studies based on the sparse problem of sound sources and the ill-posed problem of the generalized inverse beamforming method. However, there are few studies that regard the elastic net regularization generalized inverse beamforming as similar to the Lasso problem and use the ISTA algorithm to solve it. Moreover, the current noise source localization and identification methods cannot be applied in engineering, and there are still problems such as poor accuracy and low resolution in the localization and identification of underwater target noise sources. Summary of the Invention

[0007] The purpose of the present invention is to solve the problems that the current noise source localization and identification methods cannot be applied in engineering and the localization and identification of underwater target noise sources have poor accuracy and low resolution. An optimized iterative shrinkage threshold elastic net regularization generalized inverse beamforming sound source localization and identification method is proposed. This method regards the elastic net regularization generalized inverse beamforming as a problem similar to Lasso (Least absolute shrinkage and selection operator) and uses the Iterative Shrinkage Thresholding Algorithm (ISTA) to solve the elastic net regularization generalized inverse beamforming, which can realize the localization and identification of underwater target noise sources, improve the localization and identification accuracy, and has high engineering application value.

[0008] To solve the above technical problems, the present invention adopts the following technical solutions:

[0009] An optimized iterative shrinkage threshold elastic net regularization generalized inverse beamforming noise source localization and identification method uses a planar array composed of pressure hydrophones to localize the noise source. The method includes the following steps:

[0010] Step 1: Construct an elastic net regularization generalized inverse beamforming noise source identification model according to the sparsity of the noise source, and use the iterative shrinkage threshold algorithm to solve the noise source identification model;

[0011] Step 2: Construct a regularization matrix according to the iterative result of the iterative shrinkage threshold algorithm in Step 1;

[0012] Step 3: Use the regularization matrix obtained in Step 2 to optimize the elastic net regularization generalized inverse beamforming to obtain a new objective function, and use the iterative shrinkage threshold algorithm to solve the new objective function to obtain the localization and identification result of the noise source.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: By using the optimized iterative shrinkage threshold elastic net regularization generalized inverse beamforming method, the present invention solves the problems of the traditional generalized inverse beamforming, iterative shrinkage threshold elastic net regularization generalized inverse beamforming, and fast iterative shrinkage threshold elastic net regularization generalized inverse beamforming, such as relatively wide main lobes in noise source localization and identification, and poor localization and identification accuracy. By comparing with the noise source localization and identification results of the traditional generalized inverse beamforming, iterative shrinkage threshold elastic net regularization generalized inverse beamforming, fast iterative shrinkage threshold elastic net regularization generalized inverse beamforming, and optimized fast iterative shrinkage threshold elastic net regularization generalized inverse beamforming, it is proved that the method of the present invention can effectively improve the localization and identification accuracy of the noise source, and the improvement of the localization and identification accuracy in the medium and low frequency bands is more obvious. The present invention breaks through the application scenario limitations of the existing methods and improves the localization and identification accuracy of the noise source, and has good engineering application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of the noise source localization method of the present invention;

[0015] Figure 2 is a schematic diagram of the position of the sound source surface and the array during simulation;

[0016] Figure 3 is the sound source localization and identification result at 400 Hz, where Figure (a) is the GIB method, Figure (b) is the ISTA-GIB 12 method, Figure (c) is the FISTA-GIB 12 method, Figure (d) is the OISTA-GIB 12 method, Figure (e) is the OFISTA-GIB 12 method, Figure (f) is the imaging result profile;

[0017] Figure 4 is a schematic diagram of the convergence result, where Figure (a) is the convergence result of OISTA-GIB 12 and Figure (b) is the convergence result of OFISTA-GIB 12 and the convergence result. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following further describes the present invention with reference to the drawings and preferred embodiments.

[0019] The overall flowchart of the present invention is as Figure 1 shown, Figure 1 where the red arrow is the process step of the present invention, and the blue arrow is the process step of the traditional iterative shrinkage threshold regularization generalized inverse beamforming noise source localization method, and the q iDenotes the final result solved by the noise source localization method of traditional iterative shrinkage threshold regularization generalized inverse beamforming. Here, \(i\) is the number of iterations, and \(i < M\); \(q\) i' Denotes the final result solved by the present invention. Here, \(i'\) is the number of iterations, and \(i' < M\). The present invention specifically includes the following steps:

[0020] Step 1: Construct an identification noise source model and solve the size of the noise source. In this step, an elastic net regularization generalized inverse beamforming identification noise source model is constructed according to the sparsity of the noise source, and the iterative shrinkage threshold algorithm is used to solve the identification noise source model.

[0021] Assume that there is a planar array composed of \(M\) element pressure hydrophones in the acoustic field space, and the pressure vector \(P(r)\) composed of the received acoustic field information. The cross-spectral matrix \(R\) of the signals received by the planar array (p) Can be expressed as \(R\) (p) = \(P(r)P\) H (r). Performing eigenvalue decomposition on the cross-spectral matrix \(R\) (p) Can obtain:

[0022]

[0023] In the formula, \(U\) is a unitary matrix containing orthogonal eigenvectors; \(\Lambda\) is a diagonal matrix containing eigenvalues; \(H\) is the conjugate transpose; \(\sigma\) m Is the \(m\)-th eigenvalue, and \(u\) m Represents the \(m\)-th column vector in the unitary matrix \(U\), and its corresponding eigenvalue is \(\sigma\) m . Further, according to the \(m\)-th column vector and its corresponding eigenvalue \(\sigma\) m The expression of the \(m\)-th order eigenmode vector \(p\) m Can be obtained as:

[0024]

[0025] During the process of obtaining the mode vector, threshold truncation filtering is used to eliminate smaller eigenvalues. Usually, the truncation threshold is taken as 0.1% - 10% of the maximum eigenvalue \(\sigma\) max , for example, taking 5%. When the \(i\)-th eigenvalue satisfies \(\sigma\) i ≥ 0.05\(\sigma\) max , equation (2) is transformed into:

[0026]

[0027] Among them, \(\sigma\) i Is the truncated eigenvalue, that is, the eigenvalue corresponding to \(u\) i , and \(p\) i Is the \(i\)-th order eigenmode vector, and \(u\) i Is the \(i\)-th column vector in the unitary matrix \(U\).

[0028] The idea of the GIB method is to use the eigenmodes to reconstruct the sound energy distribution information over the entire sound source scanning domain. Therefore, the following acoustic transfer equation can be established:

[0029] p m =G green q m (4)

[0030] where q m is the sound source amplitude vector of dimension N×1, which describes the sound energy distribution at N divided grid points in the scanning domain under the m-th eigenmode; G green is the free-field Green's function transfer matrix from the grid points to each array element, and its dimension is M×N. By inverting Equation (4), the sound source amplitude vector q m can be obtained.

[0031] Usually, the number of array elements is much smaller than the number of scanning points on the sound source surface. Therefore, Equation (4) is usually ill-conditioned and cannot be solved directly by inversion. To ensure the accuracy and stability of the solution, it is solved by combining the L2 norm constraint. Therefore, the expression of the identified noise source model constructed in this step is as follows:

[0032]

[0033] Combining Equations (2)-(5) gives:

[0034]

[0035] where q i is the sound source amplitude vector of the i-th order, i < M; η is the L2 norm regularization parameter, L is the regularization matrix, and η and L work together to reduce the ill-conditioning of Equation (6). Usually, for the convenience of calculation, the regularization matrix L is set to the identity matrix I. At this time, Equation (6) is the solution form of the L2 norm-constrained generalized inverse beamforming. Since the accuracy of the sound field reconstructed by this method is low and the number of sound sources is sparse compared to the sound source surface, Equation (6) can be converted into the solution form of Equation (7):

[0036]

[0037] where δ is the constraint parameter, whose selection is affected by the signal-to-noise ratio, frequency, and test distance, and has a great influence on the result of Equation (7). Therefore, Equation (7) is converted into a regularization optimization model:

[0038]

[0039] Among them, λ is the L1-norm regularization parameter, and its selection is related to the signal-to-noise ratio. When the signal-to-noise ratio is large, the value of λ is small; when the signal-to-noise ratio is small, the value of λ is large. Equation (8) contains both the L1-norm regularization term and the L2-norm regularization term. This equation obtains the sparse solution of the sound source intensity through the L1-norm and uses the L2-norm to ensure the robustness of the solution process.

[0040] When the regularization matrix L = I, Equation (8) is the elastic net regularization generalized inverse beamforming, which is similar to the Lasso problem, where g(q i ) = λ||q i ||1 is a continuous convex function with unknown smoothness, is a smooth convex function; therefore, Equation (8) can be transformed into a second-order cone programming problem, and then methods such as the interior point method can be used to solve it. However, in large-scale problems, due to the too large data dimension, and the algorithm complexity of the interior point method is O(N 3 ), where N is the dimension, resulting in very time-consuming solution, so the ISTA algorithm or the FISTA algorithm is used to solve it.

[0041] In each iteration of the ISTA algorithm, q i is updated through a shrinkage threshold operation, and its specific iteration formula is as follows:

[0042]

[0043] Among them, is the i-th order sound source amplitude vector at the (k + 1)-th iteration; is the i-th order sound source amplitude vector at the k-th iteration; t (k) > 0 is the step size at the k-th iteration of the ISTA algorithm, and generally it is required that k is the number of iterations; soft λt (·) is the shrinkage operator, and its formula is:

[0044]

[0045] Among them, || is the absolute value; () + means to take and the maximum value of 0, α = λt (k) ; sgn is the sign function, indicating that if is greater than 0, then sgn returns 1; if is equal to 0, then sgn returns 0; if is less than 0, then sgn returns -1.

[0046] To address the problem of the relatively slow convergence rate of the iterative shrinkage threshold algorithm, Beck et al. proposed the fast iterative shrinkage threshold method, which preserves the simple calculation characteristics of the ISTA algorithm while greatly improving the global convergence rate of the algorithm. The initialization of the FISTA algorithm from the calculation result of the k-th iteration to the calculation result of the (k + 1)-th iteration is as follows:

[0047]

[0048] where P + represents the Euclidean projection in the non-negative quadrant; is the i-th order sound source amplitude vector at the (k + 1)-th iteration, which is the expression in the FISTA algorithm; is the gradient function; L is the Lipschitz constant; t (k+1) is the step size at the (k + 1)-th iteration of the FISTA algorithm. Combining Equation (11) with Equation (9) gives:

[0049]

[0050] Step 2: Construct the regularization matrix. Since the penalty intensity of the identity matrix for all grid points in Equation (8) is the same, the accuracy of the solution of the generalized inverse beamforming for the sparse sound source surface is very limited. To improve the accuracy of its solution and the spatial resolution, a regularization matrix with a large penalty intensity for the non-sound source region is selected. The regularization matrix is constructed using the result of the previous iteration, that is, when solving the calculation result at the k-th iteration, a regularization matrix constructed from the calculation result at the (k - 1)-th iteration is added to increase the penalty intensity for the non-sound source region. The constructed regularization matrix is:

[0051]

[0052] where is the absolute value of the i-th order sound source amplitude vector calculated at the (k - 1)-th iteration, is the infinity norm of the i-th order sound source amplitude vector calculated at the (k - 1)-th iteration, and diag represents the operation of diagonalizing the calculation result, that is, turning into a diagonal matrix; where the initial input is constructed from the noise source localization and identification result obtained by the ISTA method for optimizing the elastic net regularization generalized inverse beamforming (ISTA-GIB 12 ), that is where is ISTA-GIB 12The absolute value of the sound source amplitude vector obtained by superimposing the solution results of the method for i times is ISTA-GIB 12 The infinity norm of the sound source amplitude vector obtained by superimposing the solution results of the method for i times

[0053] Step 3: Construct a new objective function. Use the regularization matrix obtained in Step 2 to optimize the elastic net regularization generalized inverse beamforming to obtain a new objective function, and use the iterative shrinkage threshold algorithm to solve the new objective function to obtain the relative magnitudes of the noise sources on the sound source surface, thereby obtaining the localization and identification results of the noise sources

[0054] Substitute Equation (13) into Equation (8), and use Equation (9) to solve to obtain the following equation

[0055]

[0056] Equation (14) is the iterative formula for solving the elastic net regularization generalized inverse beamforming (OISTA-GIB 12 ) by the optimized iterative shrinkage threshold algorithm. Perform iterative calculations according to Equation (14), and superimpose the solution results of the i-th time to finally obtain the localization result of the noise source

[0057] Substitute Equation (13) into Equation (8), and use Equation (11) to solve to obtain the following equation

[0058]

[0059] Equation (15) is the optimized fast iterative shrinkage threshold algorithm for solving the elastic net regularization generalized inverse beamforming (OFISTA-GIB 12 ). Perform iterative calculations according to Equation (15), and superimpose the solution results of the i-th time, and the localization result of the noise source can also be finally obtained

[0060] To verify the performance of the underwater noise source localization and identification method of the optimized iterative shrinkage threshold elastic net regularization generalized inverse beamforming proposed by the present invention, the present invention uses the following simulation conditions to compare the noise source localization and identification effects of different methods. The positions of the sound source surface and the array are as Figure 2As shown in the figure, there are two non-coherent single-frequency sound sources with equal intensity on the sound source surface. The sound source positions are (x1, z1) = (-3, 0) m and (x2, z2) = (3, 0) m respectively. The signal-to-noise ratio is 0 dB. The sound source surface is a plane with a length of 12 m and a width of 12 m. Its scanning surface is discretized into 61×61 scanning points; the array is a 49×43 planar array, the element spacing is 0.15 m, and the sound source surface is 10 m away from the planar array. For the convenience of comparing and analyzing the results of each method, the acoustic imaging results are normalized to 0 dB, and its dynamic display range is 15 dB. The following respectively use Generalized Inverse Beamforming (GIB), ISTA method to optimize Elastic Net Regularized Generalized Inverse Beamforming (ISTA-GIB 12 ), FISTA method to optimize Elastic Net Regularized Generalized Inverse Beamforming (FISTA-GIB 12 ), ISTA method with iterative regularization matrix to optimize Elastic Net Regularized Generalized Inverse Beamforming (OISTA-GIB 12 ) and FISTA method with iterative regularization matrix to optimize Elastic Net Regularized Generalized Inverse Beamforming for noise source localization and identification (OFISTA-GIB 12 ).

[0061] Figure 3 Figure shows the acoustic imaging results of the 400 Hz sound source obtained by using different noise source localization and identification methods. Among them, Figures (a)-(e) are the localization and identification results of GIB, ISTA-GIB 12 , FISTA-GIB 12 , OISTA-GIB 12 and OFISTA-GIB 12 respectively, and Figure (f) is the cross-sectional view of the imaging result. The blue '×' and '·' in the figure are the true positions of the sound sources. When the frequency is 400 Hz, comparing GIB with ISTA-GIB 12 , FISTA-GIB 12 , OISTA-GIB 12 , OFISTA-GIB 12 methods and combining with the cross-sectional results, it can be seen that GIB cannot localize and identify the two noise sources when the acoustic array distance is 10 m, and the other methods can all localize and identify; comparing ISTA-GIB 12 , FISTA-GIB 12 , OISTA-GIB 12 and OFISTA-GIB 12 and combining with their cross-sectional views, it can be seen that the main lobe of the OISTA-GIB 12 method is the narrowest. Among them, since OFISTA-GIB 12 relative to OISTA-GIB 12It has a faster convergence speed and fewer iterations, and its penalty intensity for non-source regions is weaker than that of OISTA-GIB 12 , so its main lobe is wider than that of OISTA-GIB 12 method, and the convergence processes of both OISTA-GIB 12 and OFISTA-GIB 12 are as shown in Figure 4 . According to Figure 4 , it can be seen that OISTA-GIB 12 reaches the iteration termination condition after 666 iterations, and OFISTA-GIB 12 reaches the iteration termination condition after 235 iterations, verifying that the OFISTA-GIB 12 method has a faster convergence speed, and the obtained convergence speed is OISTA-GIB 12 > OFISTA-GIB 12 > ISTA-GIB 12 > FISTA-GIB 12 > GIB. The method of the present invention can effectively improve the positioning and recognition accuracy of noise sources.

[0062] The present invention first proposes to use the ISTA algorithm to solve the elastic net regularized generalized inverse beamforming. Since the penalty intensity of the identity matrix for all sound source surface scanning points in this method is the same, the positioning and recognition accuracy of noise sources is limited. In order to improve the accuracy and spatial resolution of sound source recognition, an iterative regularization matrix is proposed to replace the identity matrix, and the ISTA algorithm is used to solve the elastic net regularized generalized inverse beamforming optimized by the iterative regularization matrix. Finally, through the simulation data processing, it is verified that the optimized iterative shrinkage threshold elastic net regularized generalized inverse beamforming method has the highest noise source positioning and recognition accuracy and resolution.

[0063] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as these technical feature combinations do not conflict, they should be considered as the scope described in this specification.

[0064] The above-described embodiments only represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent should be subject to the appended claims.

Claims

1. An optimized iterative shrinkage threshold elastic net regularized generalized inverse beamforming noise source location and identification method, characterized in that: The noise source is located by using a plane array composed of acoustic pressure hydrophones, and the method comprises the following steps: Step 1: According to the sparsity of the noise source, an elastic net regularized generalized inverse beamforming model for identifying the noise source is constructed, and the iterative shrinkage threshold algorithm is used to solve the model for identifying the noise source; Step 2: construct a regularization matrix according to the iterative results of the iterative shrinkage threshold algorithm in step 1; Step 3: The regularized matrix obtained in step 2 is used to optimize the elastic net regularized generalized inverse beamforming to obtain a new objective function, and the iterative shrinkage threshold algorithm is used to solve the new objective function to obtain the positioning and identification results of the noise source.

2. The noise source location and identification method according to claim 1 is characterized in that: The expression for identifying the noise source model in step 1 is: Among them, q i is the amplitude vector of the i-th order sound source, u i represents the i-th column vector in the unitary matrix U, and its corresponding eigenvalue is σ i ,G green is the free field Green’s function transfer matrix from the grid point to each array element, L is the regularization matrix, η is the L2 norm regularization parameter, and λ is the L1 norm regularization parameter.

3. The noise source location and identification method according to claim 2 is characterized in that: The iterative formula for solving the objective function using the iterative shrinkage threshold algorithm is as follows: in, is the i-th order sound source amplitude vector calculated at the k+1th iteration; is the i-th order sound source amplitude vector calculated at the k-th iteration; t (k) >0 is the step size at the kth iteration; k is the number of iterations; soft λt (·) is the contraction operator, and its formula is: Among them, || is the absolute value; () + Indicates taking The maximum value among the sum of 0, α=λ·t (k) ; sgn is the sign function.

4. The optimized iterative shrinkage threshold elastic net regularized generalized inverse beamforming noise source location and identification method according to claim 3, characterized in that: The regularization matrix is: in, is the absolute value of the amplitude vector of the i-th order sound source calculated at the k-1th iteration, is the infinite norm of the i-th order sound source amplitude vector calculated at the k-1th iteration, and diag() means Operation to transform into a diagonal matrix.

5. The optimized iterative shrinkage threshold elastic net regularized generalized inverse beamforming noise source location and identification method according to claim 4, characterized in that: When solving the new objective function in step 3, when solving the k+1th calculation result When adding the result of the kth calculation The constructed regularization matrix increases the penalty intensity for non-sound source areas. The iterative formula for solving the new objective function using the iterative shrinkage threshold algorithm is as follows:

Images