Method and device for constructing deconvolution complex-valued beam map based on Bayesian compressed sensing

Abstract

The present invention discloses a method and device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, which comprises the following steps: (1) obtaining a reflection signal received by a sonar system, and using a beamforming algorithm to construct a complex-valued beam map according to the reflection signal; (2) constructing a complex-valued deconvolution problem based on the complex-valued beam map, and using an extended complex-valued Bayesian compressed sensing algorithm to search for candidate target points in the complex-valued beam map under a priori probability conditions; (3) checking and reconstructing the index number vector and control parameter vector corresponding to the candidate target point to obtain a screening result of the candidate target point; (4) using the reconstructed index number vector and control parameter vector as a basis, using a complex-valued Bayesian compressed sensing algorithm to search the screening result of the candidate target point again to obtain a target point in the complex-valued beam map; and (5) constructing a deconvolution complex-valued beam map based on the target point in the complex-valued beam map. The method can obtain a high-quality deconvolution complex-valued beam map.

Description

Technical Field

[0001] The present invention relates to the technical field of phased array sonar systems, and in particular to a method and device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing. Background Art

[0002] Beamforming is widely used in sonar, radar and other fields, with low computational complexity and good robustness. Beamforming processes the data of the sampling array and outputs a beam pattern within the visible range. There are multiple peaks in the beam pattern, where the peak beam corresponding to the target direction is called the main lobe, and the peak beams in the remaining directions are called side lobes. The main lobe has a certain width, and multiple targets within this range are difficult to distinguish from the beam pattern, while strong side lobes are easy to confuse weak targets.

[0003] The main lobe width and side lobe peak of the beam pattern are important indicators for measuring beamforming performance. For beamforming, the main lobe width depends on the array size and signal frequency. To reduce the main lobe width, it is necessary to increase the signal frequency or array size. At the same time, in order to avoid the appearance of grating lobes, the array spacing of the array elements decreases as the signal frequency increases. Therefore, reducing the main lobe width requires higher hardware costs and computational burdens. The side lobe peak is usually related to array weighting. There are already many weighted methods to reduce the side lobe peak, such as Chebyshev weighting, Taylor weighting, etc., but the cost is to increase the main lobe width.

[0004] In order to reduce the main lobe width and side lobe peak while avoiding increasing hardware costs, many deconvolution methods have emerged in recent years. They describe the beam pattern from another perspective, considering it as the result of a point target under the action of a point spread function. This process is called convolution. Therefore, from the perspective of image processing, the use of deconvolution methods to remove the influence of the spread function in the beam pattern effectively improves the resolution of the beam pattern, which is equivalent to reducing the main lobe width and side lobe peak. Existing deconvolution methods, such as DAMAS2, FFT-NNLS, RL, CLEAN, etc., all perform deconvolution on the beam intensity, which is effective for incoherent targets, but not for coherent targets. Under the assumption of incoherent targets, the beam intensity is equal to the superposition of the intensities of all targets under the action of the point spread function, and the interference between targets is ignored. For coherent targets, the beam intensity is also affected by the cross terms of multiple targets, and the closer the targets are, the higher the cross term intensity.

[0005] In a narrowband sonar system, the narrowband signal generated by the transmitter is reflected by the target, and the echo signals have the same frequency and a fixed phase difference, so they are coherent. In this case, the deconvolution method for intensity does not perform well, and the complex-valued beam results need to be deconvolved. The complex-valued deconvolution problem is difficult to solve with existing methods. Considering the sparsity of targets in actual usage scenarios, the complex-valued deconvolution problem can be transformed into the complex-valued compressed sensing inverse problem.

[0006] For solving the inverse problem of complex-valued compressed sensing, existing methods are based on complex value decomposition, which converts the problem into the real number domain for solution, which will reduce sparsity and increase the computational burden. In addition, in the complex-valued deconvolution problem, the observation matrix corresponding to the inverse problem of complex-valued compressed sensing is a matrix composed of point spread function vectors, which is a rank-loss matrix and easily causes the failure of solving the inverse compressed sensing algorithm. Summary of the invention

[0007] In view of the above, the purpose of the present invention is to provide a method and device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, which can reduce the computational consumption of computing equipment and improve computing efficiency while obtaining high-quality deconvolution complex-valued beam maps.

[0008] To achieve the above-mentioned object of the invention, an embodiment of the present invention provides a method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, comprising the following steps:

[0009] Step 1, obtaining the reflected signal received by the sonar system, and using a beamforming algorithm to construct a complex-valued beam diagram according to the reflected signal;

[0010] Step 2: construct a complex-valued deconvolution problem based on the complex-valued beam map, and use an extended complex-valued Bayesian compressed sensing algorithm to search for candidate target points in the complex-valued beam map under the condition of assigning a priori probability;

[0011] Step 3, checking and reconstructing the index number vector and control parameter vector corresponding to the candidate target point to obtain the screening result of the candidate target point;

[0012] Step 4: Based on the reconstructed index number vector and control parameter vector, the screening results of the candidate target points are searched again using the complex-valued Bayesian compressed sensing algorithm to obtain the target point in the complex-valued beam map;

[0013] Step 5: construct a deconvolution complex-valued beam map based on the target point in the complex-valued beam map.

[0014] To achieve the above-mentioned purpose of the invention, an embodiment of the present invention also provides a device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing.

[0015] Compared with the prior art, the present invention has the following beneficial technical effects:

[0016] The method and device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing provided by the present invention use the complex-valued output of beamforming as the deconvolution object, avoiding the model mismatch caused by the cross-influence of coherent targets in the beam intensity, thereby effectively improving the accuracy of deconvolution and reducing the main lobe width and the side lobe peak; using a complex-valued Bayesian compressed sensing algorithm to process complex value problems, compared with other Bayesian methods limited to real values, it is possible to improve the estimation accuracy, that is, to obtain a high-quality deconvolution complex-valued beam map, while reducing the computational consumption and improving the computational efficiency; using an extended complex-valued Bayesian compressed sensing algorithm to overcome the rank loss problem existing in the complex-valued point spread function matrix, thereby improving the accuracy of the calculation results.

[0017] Based on the above technical effects, the method and device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing provided by the present invention are suitable for coherent target scenes represented by active narrow-band sonar, which improves the resolution and detection performance of the beamforming results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic diagram of a process of constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing provided in an embodiment;

[0020] Figure 2 A schematic diagram of the beamforming result under the coherent target condition in step 1 provided in the embodiment;

[0021] Figure 3 A schematic diagram of a process of using an extended complex-valued Bayesian compressed sensing algorithm to search for candidate target points in a complex-valued beam map in step 2 provided in an embodiment;

[0022] Figure 4 The deconvolution beam pattern obtained in step 3 provided in the embodiment;

[0023] Figure 5 A schematic diagram of a process of re-searching the screening results of candidate target points using a complex-valued Bayesian compressed sensing algorithm in step 4 provided in the embodiment;

[0024] Figure 6 The deconvolution complex-valued beam pattern obtained in step 5 provided in the embodiment. DETAILED DESCRIPTION

[0025] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0026] In the embodiment, the transducer array in the active narrowband sonar system is a uniform linear array with 100 array elements arranged at half a wavelength, and the transmitting transducer generates a narrowband pulse signal of 53us, 375kHz; assuming that there are seven targets located in the far field and at the same distance from the center of the receiving array, their backscattered signals will be coherently superimposed and sampled by the receiving array, and the azimuth and reflected signal strength of the target are shown in Table 1:

[0027] Table 1

[0028] Serial number Target direction Reflected signal strength Serial number Target direction Reflected signal strength 1 -12.8° 0dB 5 0.9° 0dB 2 -12.6° -3dB 6 12.4° -10dB 3 -0.7° -20dB 7 12.6° -14dB 4 0.7° -6dB

[0029] Assuming the signal-to-noise ratio is 10 dB, the response signal vector of the receiving array is represented by x, and beamforming uses uniform weighting, that is, w is a column vector of all 1s.

[0030] Figure 1 The following is a flow chart of a method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing provided in an embodiment. Figure 1 As shown, the method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing provided in the embodiment includes the following steps:

[0031] Step 1: Obtain the reflected signal received by the sonar system, and use a beamforming algorithm to construct a complex-valued beam diagram based on the reflected signal.

[0032] In the embodiment, assuming that the field of view is ±90°, the beamforming algorithm constructs 513 beams in different directions, the rotation vector of each beam is composed of 100 complex coefficients, the size of the rotation vector matrix A is 513×100, and the complex-valued beam diagram constructed according to the reflected signal using the beamforming algorithm is:

[0033] B=A(w·x)

[0034] Wherein, B represents a complex-valued beam vector, i.e., the corresponding complex-valued beam diagram, A represents a matrix composed of rotation vectors, w represents a weighting coefficient, x represents a received reflected signal, and the symbol · represents element-wise multiplication. Figure 2 It represents the normalized beam pattern corresponding to the complex-valued beam vector B. It can be seen that only three targets are distinguished in the complex-valued beam pattern.

[0035] Step 2: construct a complex-valued deconvolution problem based on the complex-valued beam map, and use the extended complex-valued Bayesian compressed sensing algorithm to search for candidate target points in the complex-valued beam map under the condition of assigning prior probability.

[0036] In an embodiment, a complex-valued deconvolution problem is constructed based on a complex-valued beam graph, including:

[0037] The complex-valued beam vector B is regarded as the product of the target vector and the complex-valued point spread function matrix, that is, the complex-valued beam vector corresponding to the complex-valued beam diagram is also expressed as:

[0038] B=Φt+e

[0039] Among them, Φ represents the complex-valued point spread function matrix of size M×M, t represents the potential target complex-valued vector corresponding to the target point, with size M×1, and e represents the complex-valued Gaussian noise vector of size M×1. The complex-valued point spread function matrix Φ can be calculated as follows:

[0040]

[0041] Where Φ(m,n) represents the (m,n)th element, i represents the imaginary unit, N represents the number of elements in the receiving array, d represents the array spacing, λ represents the signal wavelength, and u represents the m,n It represents the sine difference between the m and n directions, and sinc represents the sinc function.

[0042] The number of non-zero elements in the potential target complex-valued vector t is much smaller than the vector length. The complex-valued deconvolution problem constructed based on the complex-valued beam vector is:

[0043] min||t||0,subject to||B-Φt||2<e

[0044] In this complex-valued deconvolution problem, B = A(w·x).

[0045] Based on the complex-valued deconvolution problem description, a two-step Bayesian compressed sensing algorithm is used to solve the problem description and obtain the positions and values ​​of non-zero elements in t, thereby constructing a deconvolution beam map.

[0046] In the embodiment, the constructed complex-valued deconvolution problem is incorporated into the Bayesian framework, and given a priori probability condition:

[0047] p(t)=Πi p(t i |α i )p(t i+1 |α i )

[0048] Among them, α i is the i-th element in the control vector of size (M-1)×1, where i represents a natural number from 1 to M-1, and p(t i |α i ) and p(t i+1 |α i ) represent the i-th element t in the potential target complex-valued vector t i and the i+1th element t i+1 About α i The prior distribution of , Π represents the cumulative multiplication, and p(t) represents the overall prior distribution of the potential target complex-valued vector t. In this prior probability condition, it is assumed that each element in the potential target complex-valued vector t obeys a zero-mean Gaussian distribution as follows:

[0049]

[0050]

[0051] in, represents a Gaussian distribution, represents the real part operation, Indicates the operation of taking the imaginary part, the superscript - 1 Indicates the countdown, express The mean is 0 and the variance is Similarly, assume that the complex-valued Gaussian noise vector e has the following prior probability conditions:

[0052]

[0053] Wherein, the subscript j represents a natural number from 1 to M, and β represents a control parameter.

[0054] In the embodiment, the value of β depends on the situation, and the value range may be 0.01 to 0.001.

[0055] Under this prior probability condition, the extended complex-valued Bayesian compressed sensing algorithm is executed to preliminarily infer the posterior probability distribution of each element in the target complex-valued vector t, so as to search for candidate target points in the complex-valued beam map, as shown in Figure 3 As shown, the following steps are included:

[0056] Step 2-1, initialize update parameters:

[0057]

[0058] Where m represents a natural number from 1 to M-1, which is used to index the candidate target in the potential target complex-valued vector t, S m Represents the sparse coefficient matrix of the mth candidate target, with a size of 2×2, Q m represents the quality coefficient vector of the mth candidate target, with a size of 2 × 1. The superscript H represents the conjugate transpose, Ψ m From the adjacent columns Φ in the complex-valued point spread function matrix Φ m and Φ m+1 Composition means:

[0059] Ψ m =[Φ m ,Φ m+1 ]

[0060] Construct the expected vector μ and variance matrix Σ of the posterior probability of the potential target complex-valued vector t, where μ only contains the elements in t whose expected values ​​of the posterior probability are not 0, and their corresponding index values ​​are stored in the index number vector v, and the variance matrix Σ only contains the posterior variances of the corresponding elements in μ;

[0061] Construct a control parameter vector Λ to store the variance control parameter α corresponding to the index value in v i , construct the matrix It is used to store the column vectors in the complex-valued point spread function matrix Φ corresponding to the index value in v. In the initialization stage, the expected value of the posterior probability of all elements in t is 0, that is, represents an empty vector or an empty matrix;

[0062] Step 2-2, calculate the possible likelihood growth of all candidate target points based on the parameters:

[0063] s m =mean(diag(S m ))

[0064]

[0065] ml m =(q m -s m ) / s m +log(s m / q m )

[0066] Among them, mean means the mean operation, diag means taking the elements on the main diagonal of the matrix, represents the square of the two-norm, s m Indicates S m The result of the mean of the main diagonal elements, q m Indicates Q mThe result of the square of the second norm, ml m represents the possible likelihood growth of the mth candidate target point,

[0067] Select the maximum value from all ml and record the index number, represented by idx, and put it into the index number vector, that is, v (new) =[v (old) ,idx]

[0068] Calculate the q corresponding to the index extracted from the index number vector idx 、s idx To calculate the control parameter α idx , and put the control parameters into the control parameter vector Λ:

[0069]

[0070] Λ (new) =[Λ (old) ,α idx ]

[0071] Among them, the superscripts new and old represent the variable values ​​before and after the update respectively;

[0072] Step 2-3, based on all candidate target points, update the expected vector, variance and related parameters of the posterior probability:

[0073]

[0074]

[0075]

[0076] Among them, Σ idx =(α idx I2+S idx ) -1 μ idx =Σ idx Q idx I represents the identity matrix;

[0077] Step 2-4, calculate the iteration stop condition cr1. When the iteration stop condition is less than the first threshold, the iteration ends and the posterior probability distribution of each element in the target complex-valued vector t is preliminarily obtained. Otherwise, return to step 2-2 to continue the calculation.

[0078] The iteration stop condition is calculated using the following formula:

[0079]

[0080]

[0081] cr1=(L(new) -L (old) ) / L (old) .

[0082] In the embodiment, the first threshold value is determined according to the situation, and the value range may be 0.05 to 0.01.

[0083] Step 3: Check and reconstruct the index number vector and control parameter vector corresponding to the candidate target point to obtain the screening result of the candidate target point.

[0084] In the embodiment, the process of checking and reconstructing the index number vector and the control parameter vector corresponding to the candidate target point includes:

[0085] First, a new index number vector v' and a new control parameter vector Λ' are constructed according to the control parameter vector Λ and the index number vector v corresponding to the candidate target point:

[0086] v'=[v(1),v(1)+1,v(2),v(2)+1,...,v(k),v(k)+1]

[0087] Λ'=[Λ(1),Λ(1)+1,Λ(2),Λ(2)+1,...,Λ(k),Λ(k)+1]

[0088] Wherein, k is the number of elements of vector v and vector Λ.

[0089] In the experimental example, when some of the estimated parameters are as follows:

[0090] μ=[-5.76+79.67i,28.98+47.98i,…,-8.67+12.99i]

[0091]

[0092] v=[200,260,312]Λ=[0.0065,0.001,0.0006]

[0093] The new index number vector and new control parameter vector constructed are:

[0094] v'=[200,201,260,261,312,313]

[0095] Λ'=[0.0065,0.0065,0.001,0.001,0.0006,0.0006]

[0096] Then, check whether there are duplicate elements in the new index number vector v'. If there are no duplicate elements, there is no need to merge and delete them. When there are duplicate elements, perform the following operations on each group of duplicate elements:

[0097] μ(r1)=μ(r1)+μ(r2)

[0098]

[0099] Σ(r1,:)=Σ(r1,:)+Σ(r2,:)

[0100] Σ(:,r1)=Σ(:,r1)+Σ(:,r2)

[0101] Among them, r1 and r2 represent the order in which the repeated elements appear in v', Σ(r2,:) represents the r2th row in Σ, and Σ(:,r2) represents the r2th column in Σ. express Column r2 in the middle;

[0102] Finally, delete μ(r2), Λ'(r2), Σ(r2,:), Σ(:,r2), and The elements of the position, where express The r2th column in , the screening results of candidate target points are obtained.

[0103] The complex valued beam diagram corresponding to the result after the inspection and adjustment is as follows Figure 4 As shown, targets 1, 2, 6, and 7 have been estimated, but targets 3, 4, and 5 still need to be adjusted because they are close to each other.

[0104] Step 4: Based on the reconstructed index number vector and control parameter vector, the screening results of the candidate target points are searched again using the complex-valued Bayesian compressed sensing algorithm to obtain the target points in the complex-valued beam map.

[0105] like Figure 5 As shown, in the embodiment, the process of re-searching the screening results of the candidate target points using the complex-valued Bayesian compressed sensing algorithm includes:

[0106] Step 4-1, construct new update parameters:

[0107]

[0108] Among them, the index range of m is expanded to natural numbers from 1 to M. express The first row and first column element of express The second row and second column element of express The first element of express The second element of and Respectively represent the sparse coefficient and quality coefficient of the updated m-th candidate target. In the subsequent steps, and Transformed to scalar, indicated by non-bold font.

[0109] Step 4-2, calculate the possible likelihood growth of all candidate target points based on the new update parameters and the reconstructed index number vector:

[0110]

[0111] Where the subscript j represents the jth element in a vector or the jth column in a matrix, and:

[0112]

[0113]

[0114] The maximum value is selected from all ml and is represented by ML. The index number corresponding to ML is represented by idx.

[0115] Step 4-3, calculate the following iteration stop condition cr2, when the iteration stop condition cr2 reaches the second threshold, the iteration ends, and the posterior probability distribution of each element in the target complex-valued vector t is obtained, that is, the target point in the corresponding complex-valued beam map, otherwise execute step 4-4;

[0116] cr2=(ML (new) -ML (old) ) / ML (old)

[0117] In the embodiment, the first threshold value is determined according to the situation and the value range may be 10 -4 ~10 -6 .

[0118] Step 4-4, parameter update: If θ idx <0 execute update strategy (a); if θ idx > 0 and idx∈v' executes update strategy (b); if θ idx >0 and Execute the update strategy (c); after executing the corresponding parameter update strategy, return to step 4-2;

[0119] Strategy (a):

[0120]

[0121]

[0122] μ (new) =μ (old) -μ j Σ j / Σ jj

[0123]

[0124]

[0125] Among them, j represents the element position of idx in v', Σ jj Represents Σ (old) The element in row j and column j in the (new) The jth element in Σ (new) The element in the jth row and jth column;

[0126] Strategy (b):

[0127] k=(Σ jj +(θ idx -α j ) -1 ) -1

[0128]

[0129] μ (new) =μ (old) -kμ j Σ j

[0130]

[0131]

[0132]

[0133] Strategy (c):

[0134]

[0135] Σ idx =(α idx +S idx ) -1

[0136] μ idx =Σ idx Q idx

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143] v' (new) =[v' (old) ,idx]

[0144] Λ' (new) =[Λ' (old) ,θ idx ]

[0145] Step 5: construct a deconvolution complex-valued beam map based on the target point in the complex-valued beam map.

[0146] In an embodiment, constructing a deconvolution complex-valued beam map based on a target point in the complex-valued beam map includes:

[0147] Step 5-1, construct a zero vector Dv of the same size as the complex-valued beam vector;

[0148] Step 2: Perform the following operation on each element in the expected vector μ of the posterior probability of the potential target complex-valued vector t corresponding to the target point in the complex-valued beam map:

[0149] Dv(v'(i))=μ(i)

[0150] Among them, i represents the index of each element in μ, v'(i) represents the i-th element in v', which specifies the element position of μ(i) in Dv. In this step, the i-th element in μ is placed in the v'(i)-th position in Dv one by one. The filled element Dv is the posterior probability expectation value of the potential target complex-valued vector t, that is, the constructed deconvolution complex-valued beam map.

[0151] In the experimental example, some estimated parameters are as follows:

[0152] μ=[-3.48+79.44i,27.83+49.52i,…,-0.05-0.01i]

[0153]

[0154] v'=[200,201,260,261,312,313,254,…,210]

[0155] Λ=[1e-6,1e-6,1e-6,…,,0.003,0.077,0.596]

[0156] Construct a zero vector Dv of length 513, and perform the following operations on each element in μ:

[0157] Dv(v'(i))=u(i)

[0158] The deconvolution complex valued beam pattern vector Dv is obtained; Figure 6 As shown in the figure, the directions and intensities of the seven targets have been accurately estimated, and the sidelobe intensity is suppressed below -37dB. Figure 2 The beamforming result is processed by the complex-valued beam image deconvolution method based on Bayesian compressed sensing, and the beam pattern has higher spatial resolution and detection performance.

[0159] An embodiment also provides a device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing are implemented.

[0160] In practical applications, the computer memory can be a volatile memory at the near end, such as RAM, or a non-volatile memory, such as ROM, FLASH, floppy disk, mechanical hard disk, etc., or a remote storage cloud. The computer processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), that is, the steps of the method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing can be implemented by these processors.

[0161] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, characterized in that: The following steps are involved: Step 1, obtaining the reflected signal received by the sonar system, and using a beamforming algorithm to construct a complex-valued beam diagram according to the reflected signal; Step 2: construct a complex-valued deconvolution problem based on the complex-valued beam map, and use an extended complex-valued Bayesian compressed sensing algorithm to search for candidate target points in the complex-valued beam map under the condition of assigning a priori probability; Step 3, checking and reconstructing the index number vector and control parameter vector corresponding to the candidate target point to obtain the screening result of the candidate target point; Step 4: Based on the reconstructed index number vector and control parameter vector, the screening results of the candidate target points are searched again using the complex-valued Bayesian compressed sensing algorithm to obtain the target point in the complex-valued beam map; Step 5: construct a deconvolution complex-valued beam map based on the target point in the complex-valued beam map.

2. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 1, characterized in that: In step 1, a beamforming algorithm is used to construct a complex-valued beam pattern according to the reflected signal: B=A(w·x) Among them, B represents the complex-valued beam vector of size M×1, that is, the corresponding complex-valued beam pattern, A represents the M×N matrix composed of rotation vectors, w represents the weighting coefficient vector of size N×1, x represents the received reflection signal of size N×1, the symbol · represents element-by-element multiplication, N is the number of receiving array elements of the sonar system, and M is the number of observation directions of the complex-valued beam pattern.

3. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 2, characterized in that: In step 2, constructing a complex-valued deconvolution problem based on the complex-valued beam graph includes: The complex-valued beam vector corresponding to the complex-valued beam diagram is also expressed as: B=Φt+e Where Φ represents the complex-valued point spread function matrix of size M×M, t represents the potential target complex-valued vector corresponding to the target point of size M×1, and e represents the complex-valued Gaussian noise vector of size M×1; The number of non-zero elements in the potential target complex-valued vector t is much smaller than the vector length. The complex-valued deconvolution problem constructed based on the complex-valued beam vector is: min||t||0,subject to||B-Φt||2<e In this complex-valued deconvolution problem, B = A(w·x).

4. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 2, characterized in that: In step 2, the constructed complex-valued deconvolution problem is incorporated into the Bayesian framework and given the following prior probability conditions: p(t)=Π i p(t i |a i )p(t i+1 |a i ) Among them, α i is the i-th element in the control vector of size (M-1)×1, where i represents a natural number from 1 to M-1, and p(t i |α i ) and p(t i+1 |α i ) represent the i-th element t in the potential target complex-valued vector t i and the i+1th element t i+1 About α i The prior distribution of , Π represents the cumulative multiplication, and p(t) represents the overall prior distribution of the potential target complex-valued vector t; In this prior probability condition, it is assumed that each element in the potential target complex-valued vector t obeys a zero-mean Gaussian distribution as follows: in, represents a Gaussian distribution, represents the real part operation, Indicates the operation of taking the imaginary part, the superscript -1 Indicates the countdown, express The mean is 0 and the variance is Gaussian distribution, similarly, assuming that the complex-valued Gaussian noise vector e has the following prior probability conditions: Wherein, the subscript j represents a natural number from 1 to M, and β represents a control parameter.

5. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 4, characterized in that: In step 2, under the prior probability condition, the extended complex-valued Bayesian compressed sensing algorithm is executed to preliminarily infer the posterior probability distribution of each element in the target complex-valued vector t, so as to search for candidate target points in the complex-valued beam map, including: Step 2-1, initialize update parameters: Where m represents a natural number from 1 to M-1, which is used to index the candidate target in the potential target complex-valued vector t, S m Represents the sparse coefficient matrix of the mth candidate target, with a size of 2×2, Q m represents the quality coefficient vector of the mth candidate target, with a size of 2×1, and the superscript H represents the conjugate transpose, Ψ m From the adjacent columns Φ in the complex-valued point spread function matrix Φ m and Φ m+1 Composition means: P m =[Φ m ,F m+1 ] Construct the expected vector μ and variance matrix Σ of the posterior probability of the potential target complex-valued vector t, where μ only contains the elements in t whose expected values ​​of the posterior probability are not 0, and their corresponding index values ​​are stored in the index number vector v, and the variance matrix Σ only contains the posterior variances of the corresponding elements in μ; Construct a control parameter vector Λ to store the variance control parameter α corresponding to the index value in v i , construct the matrix It is used to store the column vectors in the complex-valued point spread function matrix Φ corresponding to the index value in v. In the initialization stage, the expected value of the posterior probability of all elements in t is 0, that is, represents an empty vector or an empty matrix; Step 2-2, calculate the possible likelihood growth of all candidate target points based on the parameters: Among them, mean means the mean operation, diag means taking the elements on the main diagonal of the matrix, represents the square of the two-norm, s m Indicates S m The result of the mean of the main diagonal elements, q m Indicates Q m The result of the square of the second norm, ml m represents the possible likelihood growth of the mth candidate target point, Select the maximum value from all ml and record the index number, represented by idx, and put it into the index number vector, that is, v (new) =[v (old) ,idx] Calculate the q corresponding to the index extracted from the index number vector idx 、s idx To calculate the control parameter α idx , and put the control parameters into the control parameter vector Λ: Among them, the superscripts new and old represent the variable values ​​before and after the update respectively; Step 2-3, based on all candidate target points, update the expected vector, variance and related parameters of the posterior probability: Among them, Σ idx =(α idx I2+S idx ) -1 μ idx =Σ idx Q idx I represents the identity matrix; Step 2-4, calculate the iteration stop condition cr1. When the iteration stop condition is less than the first threshold, the iteration ends and the posterior probability distribution of each element in the target complex-valued vector t is preliminarily obtained. Otherwise, return to step 2-2 to continue the calculation. The iteration stop condition is calculated using the following formula:

6. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 5, characterized in that: Step 3 includes: First, a new index number vector v' and a new control parameter vector Λ' are constructed according to the control parameter vector Λ and the index number vector v corresponding to the candidate target point: v'=[v(1),v(1)+1,v(2),v(2)+1,...,v(k),v(k)+1] Λ'=[Λ(1),Λ(1)+1,Λ(2),Λ(2)+1,...,Λ(k),Λ(k)+1] Where k is the number of elements in vector v and vector Λ; Then, check whether there are repeated elements in the new index number vector v'. If there are repeated elements, perform the following operations on each group of repeated elements: μ(r1)=μ(r1)+μ(r2) Σ(r1,:)=Σ(r1,:)+Σ(r2,:) Σ(:,r1)=Σ(:,r1)+Σ(:,r2) Among them, r1 and r2 represent the order in which the repeated elements appear in v', Σ(r2,:) represents the r2th row in Σ, and Σ(:,r2) represents the r2th column in Σ. express Column r2 in the middle; Finally, delete μ(r2), Λ'(r2), Σ(r2,:), Σ(:,r2), and The elements of the position, where express The r2th column in , the screening results of candidate target points are obtained.

7. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 5, characterized in that: Step 4 includes: Step 4-1, construct new update parameters: Among them, the index range of m is expanded to natural numbers from 1 to M. express The first row and first column element of express The second row and second column element of express The first element of express The second element of and Respectively represent the sparse coefficient and quality coefficient of the updated m-th candidate target. In the subsequent steps, and Transformed into scalars, indicated by non-bold font; Step 4-2, calculate the possible likelihood growth of all candidate target points based on the new update parameters and the reconstructed index number vector: Where the subscript j represents the jth element in a vector or the jth column in a matrix, and: The maximum value is selected from all ml and is represented by ML. The index number corresponding to ML is represented by idx. Step 4-3, calculate the following iteration stop condition cr2, when the iteration stop condition cr2 reaches the second threshold, the iteration ends, and the posterior probability distribution of each element in the target complex-valued vector t is obtained, that is, the target point in the corresponding complex-valued beam map, otherwise execute step 4-4; cr2=(ML (new) -ML (old) ) / ML (old) Step 4-4, parameter update: If θ idx <0 execute update strategy (a); if θ idx > 0 and idx∈v' executes update strategy (b); if θ idx >0 and Execute the update strategy (c); after executing the corresponding parameter update strategy, return to step 4-2; Strategy (a): Among them, j represents the element position of idx in v', Σ jj Represents Σ (old) The element in row j and column j in the (new) The jth element in Σ (new) The element in the jth row and jth column; Strategy (b): Strategy (c):

8. The method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing according to claim 1, characterized in that: Step 5 includes: Step 5-1, construct a zero vector Dv of the same size as the complex-valued beam vector; Step 2: Perform the following operation on each element in the expected vector μ of the posterior probability of the potential target complex-valued vector t corresponding to the target point in the complex-valued beam map: Dv(v'(i))=μ(i) Among them, i represents the index of each element in μ, v'(i) represents the i-th element in v', which specifies the element position of μ(i) in Dv. In this step, the i-th element in μ is placed in the v'(i)-th position in Dv one by one. The filled element Dv is the posterior probability expectation value of the potential target complex-valued vector t, that is, the constructed deconvolution complex-valued beam map.

9. A device for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the method for constructing a deconvolution complex-valued beam map based on Bayesian compressed sensing as described in any one of claims 1 to 8.

Images