Rapid reverberation suppression method, system and device based on variational Bayes

Abstract

The invention discloses a rapid reverberation suppression method, system and device based on variational Bayes. The method comprises the following steps: generating a two-dimensional data matrix according to a three-dimensional sonar image sequence; a variational Bayesian algorithm is used to separate low-rank components and sparse components in a data matrix, and a generalized approximate message passing algorithm is used to avoid a matrix inversion step in a variational Bayesian iteration process so as to reduce low-rank sparse decomposition time; and reversely quantizing the two-dimensional sparse matrix into a three-dimensional sparse image sequence. The system comprises a data conversion module, a separation module and a reverse quantity module. The device comprises a memory and a processor used for executing the rapid reverberation suppression method based on variational Bayes. By using the invention, reverberation suppression can be quickly and stably implemented, so that a moving target can be detected. The method can be widely applied to the field of active detection of underwater moving targets.

Description

Technical Field

[0001] This invention relates to the field of active detection of underwater moving targets, and in particular to a fast reverberation suppression method, system and device based on variational Bayes. Background Technology

[0002] The detection, localization, tracking, and identification of small underwater targets at long distances are core technical challenges for active sonar, especially in shallow water environments where reverberation from the seabed, surface, and water body significantly impacts the performance of active sonar and increases the difficulty of detecting small targets. Therefore, reverberation suppression is an essential technical component of active detection.

[0003] Most existing reverberation suppression methods are based on the low-rank sparsity of multi-frame sonar images and use robust principal component analysis to separate low-rank reverberation and sparse targets. These low-rank sparse algorithms are all based on optimization algorithms, and the penalty factor of the sparse term needs to be manually tuned. Especially under low signal-to-mixing conditions, an excessively large sparse penalty factor will cause low-energy targets to not be decomposed in the sparse matrix. Such reverberation suppression algorithms are mostly time-consuming and difficult to meet the real-time requirements of active sonar. Summary of the Invention

[0004] In view of this, in order to solve the technical problem that existing reverberation suppression methods require the selection of penalty factors and the execution of matrix inversion operations, resulting in excessively long computation times, this invention proposes, firstly, a fast reverberation suppression method based on variational Bayesian methods, which specifically includes: First, the original 3D sonar image sequence is vectorized frame by frame in chronological order and then concatenated to obtain a 2D data matrix. Then, the variational Bayesian algorithm is used to separate the low-rank and sparse components in the data matrix, and a generalized approximation message-passing algorithm is used to avoid the matrix inversion step in the variational Bayesian iteration process, thus reducing the time for low-rank sparse decomposition. Next, the 2D sparse matrix is ​​inversely vectorized into a 3D sparse image sequence. Finally, nonlinear superposition is applied to the 3D sparse image sequence to further enhance the target and obtain the trajectory of the moving target. Compared with other reverberation suppression algorithms, this method avoids the selection of penalty factors in optimization algorithms and avoids the time-consuming matrix inversion step, enabling fast and robust reverberation suppression for moving target detection.

[0005] In addition to the above method, the present invention also proposes a fast reverberation suppression system based on variational Bayes, which includes a data conversion module, a separation module, and an inverse vector module.

[0006] This invention also proposes a fast reverberation suppression device based on variational Bayes, comprising: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a fast reverberation suppression method based on variational Bayes as described above.

[0007] Based on the above scheme, this invention provides a fast reverberation suppression method, system, and device based on variational Bayesian methods. By combining variational Bayesian inference with a generalized approximate message passing algorithm to achieve reverberation suppression, it can be better applied to the field of active detection of small underwater moving targets. It avoids the high computational complexity caused by the selection of regularization parameters (penalty factors) and matrix inversion operations in optimization methods. Compared with existing reverberation suppression methods, this invention has lower complexity, especially for high-dimensional data, with shorter computation time. It can suppress more reverberation energy while preserving the target energy, thus improving the robustness of the algorithm and the target detection performance. Attached Figure Description

[0008] Figure 1 This is a flowchart of the steps of a fast reverberation suppression method based on variational Bayes according to the present invention; Figure 2 This is a schematic diagram of a frame from the original sonar image sequence.

[0009] Figure 3 Yes Figure 2 A schematic diagram of the VBRPCA reverberation suppression results for the data frame.

[0010] Figure 4 Yes Figure 2 The data frame is a schematic diagram of the VBGAMP reverberation suppression result of the present invention.

[0011] Figure 5 Yes Figure 2 A schematic diagram of the APG reverberation suppression results for the data frames.

[0012] Figure 6 Yes Figure 2 A schematic diagram of the ADMM reverberation suppression results for the data frames.

[0013] Figure 7 This is a comparison of azimuth slices at the target location under different reverberation suppression methods for this frame of data.

[0014] Figure 8 This is a comparison chart showing the summation of the energy of all pixels in each frame of data under different reverberation suppression methods.

[0015] Figure 9 It is the target trajectory obtained by nonlinear superposition after VBGAMP reverberation suppression according to the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] It should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0018] It should be understood that the terms "system," "apparatus," "unit," and / or "module" used in this application are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0019] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "a," and / or "the" are not specifically singular and may include the plural. Generally, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements. An element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes the element.

[0020] In the description of the embodiments of this application, "a plurality of" refers to two or more. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0021] Furthermore, flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Additionally, other operations can be added to these processes, or one or more steps can be removed from them.

[0022] Reference Figure 1 The diagram below illustrates an optional example of the fast reverberation suppression method based on variational Bayes proposed in this invention. This method can be applied to computer devices, and the reverberation suppression method proposed in this embodiment may include, but is not limited to, the following steps: Step S1: Obtain the original three-dimensional sonar image sequence and convert it into a two-dimensional data matrix; Step S2: Combining generalized approximate message passing and variational Bayesian methods, separate and update the low-rank and sparse components in the two-dimensional data matrix; Step S3: Inverse vectorize the sparse components to obtain a three-dimensional sparse image sequence; Step S4: Nonlinearly superimpose the three-dimensional sparse image sequence to obtain the target trajectory matrix.

[0023] In some feasible embodiments, step S1 specifically includes: Original 3D sonar image sequence After processing, a two-dimensional data matrix is ​​obtained, represented as follows: in, , , , This is a matrix vectorization operator.

[0024] In some feasible embodiments, it also includes: Two-dimensional data matrix The model is as follows: in, Represents a two-dimensional data matrix. This indicates the low-rank component, which mainly includes steady-state reverberation; It represents sparse components, mainly including wave reverberation, moving targets, motion interference, etc. Indicates noise.

[0025] In some feasible embodiments, step S2 specifically includes: S2.1. Use the variational Bayesian (VB) method to perform low-rank sparse decomposition to achieve the separation of steady-state reverberation and moving targets.

[0026] Among them, the input data matrix of the overall reverberation suppression algorithm With the parameters of the probability density function Initialize low-rank components sparse components , low-rank components precision matrix sparse components accuracy and noise accuracy .

[0027] S2.2. Use the generalized approximate message passing (GAMP) algorithm to quickly update the low-rank components. and its variance .

[0028] S2.3, Update the precision matrix: ; in, express The nth column vector, This represents the diagonalization operator for column vectors.

[0029] S2.4, Update sparse components variance: ; Representing a two-dimensional data matrix The second dimension is the total number of pixels in each frame of the sonar image; Representing a data matrix The first dimension is the number of frames in the sonar image.

[0030] Update sparse components: ; in This represents the matrix division operator. This represents the matrix dot product operator.

[0031] S2.5, Update the accuracy of noise: ; in The square operator represents the Frobenius norm of a matrix. This represents the trace operator for a matrix. express The Line number List.

[0032] S2.6, Update the precision of sparse components: S2.7, Repeat S2.2-S2.5 until the error of the reconstructed target signal is less than the stopping iteration error. The low-rank component is then output. and sparse components .

[0033] In some feasible embodiments, step S2.2 specifically includes: Step S2.2.1: Adjust the precision matrix Perform eigenvalue decomposition Invert the eigenvalues ​​and copy them. The matrix is ​​obtained .in, This represents the operator for extracting diagonal elements of a matrix to form a column vector. Let... And perform generalized approximate message passing algorithm signal initialization. , as well as .

[0034] Step S2.2.2: The GAMP algorithm outputs linear steps. , ; Step S2.2.3: The CAMP algorithm outputs a nonlinear step. , ; Step S2.2.4: Input linear step of CAMP algorithm, , ; Step S2.2.5: Input nonlinear step for the CAMP algorithm. , .

[0035] in Represents the eigenvalue matrix. Represents the eigenvector matrix, Represents the error matrix. Represents the measurement matrix. This represents the variance of the linear output signal. This represents the mean of the linear output signal. This represents the mean of the nonlinear output signal. This represents the variance of the nonlinear output signal. This represents the variance of the linear input signal. This represents the mean of a linear input signal. This represents the variance of the nonlinear input signal (i.e., the variance of the low-rank component). This represents the mean of the nonlinear input signal (i.e., the low-rank component).

[0036] In some feasible embodiments, step S3, will Inverse vectorization yields a 3D sparse image sequence. .

[0037] In some feasible embodiments, step S4 specifically includes: Three-dimensional sparse image sequences Nonlinear superposition is used to further enhance the target and obtain the trajectory of the moving target. The formula for nonlinear superposition is as follows: in, Representing the image matrix lie in Pixel value at; The mean of the sparse image sequence; The order of the nonlinear accumulation of the sparse image sequence is given. The entire reverberation suppression algorithm based on low-rank sparse decomposition outputs a sparse image sequence. and its nonlinear superposition result .

[0038] The invention provides a method for rapid reverberation suppression of sonar image sequences containing small underwater moving targets, enabling robust target detection and trajectory formation. To verify the effectiveness and robustness of the invention's variational Bayesian generalized approximate message passing (VBGAMP) method, real experimental data was processed and analyzed in the embodiments, and compared with existing reverberation suppression methods such as variational Bayesian robust principal component analysis (VBRPCA), accelerated proximal gradient (APG), and alternating direction method of multipliers (ADMM).

[0039] Figure 2 This is a frame from the original sonar image sequence, with the target circled in red. It can be seen that the target's energy is much lower than the surrounding reverberation energy, classifying it as a weak target at long range. Figures 3 to 6 To Figure 2 The results obtained by applying four reverberation suppression methods to the data frames show that although the four methods can separate the target, they have different degrees of suppression of background reverberation. Among them, the VBGAMP method of this invention has the best reverberation suppression performance. Figure 7 This is a comparison of the azimuth slices of the target location under different reverberation suppression methods for this frame of data. This image more intuitively demonstrates that the VBGAMP method of this invention can better suppress background reverberation while preserving the target energy. Figure 8 This is a comparison chart of the summation of the energy of all pixels in each frame of data under different reverberation suppression methods. It can be seen that the background energy of the data processed by the VBGAMP method of this invention is the lowest. Figure 9The target trajectory map is obtained by nonlinear superposition after VBGAMP reverberation suppression using the method of this invention, demonstrating the robustness of the method for detecting weak targets at long distances. Furthermore, the data dimension for reverberation suppression is 342×359917, which is considered high-dimensional computation. A comparison of the time consumption of the four methods is as follows: ADMM method is 546 s, APG method is 404 s, VBRPCA method is 459 s, and the VBGAMP method of this invention is 201 s. This shows that the method of this invention has good practical application value.

[0040] A fast reverberation suppression system based on variational Bayes, comprising: The data conversion module is used to perform step S1; The separation module is used to execute step S2; The inverse vector module is used to execute step S3; The trajectory matrix generation module is used to execute step S4.

[0041] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0042] A fast reverberation suppression device based on variational Bayes: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a fast reverberation suppression method based on variational Bayes as described above.

[0043] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0044] A storage medium storing processor-executable instructions, which, when executed by a processor, are used to implement a variational Bayes-based fast reverberation suppression method as described above.

[0045] The content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0046] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A fast reverberation suppression method based on variational Bayes, characterized in that, Includes the following steps: Acquire the original three-dimensional sonar image sequence and convert it into a two-dimensional data matrix; By combining generalized approximate message passing and variational Bayesian methods, the low-rank and sparse components in the two-dimensional data matrix are separated and updated. The sparse components are inversely vectorized to obtain a three-dimensional sparse image sequence.

2. The fast reverberation suppression method based on variational Bayes as described in claim 1, characterized in that, The signal model representation of a two-dimensional data matrix is ​​as follows: in, Represents a two-dimensional data matrix. Indicates a low-rank component. Indicates sparse components, Indicates noise.

3. The fast reverberation suppression method based on variational Bayes as described in claim 2, characterized in that, The step of separating and updating the low-rank and sparse components in the two-dimensional data matrix by combining generalized approximate message passing and variational Bayesian methods specifically includes: The two-dimensional data matrix is ​​decomposed into low-rank sparse components by variational Bayesian method, separating the low-rank components and sparse components. Update the low-rank components and their variances using generalized approximate message passing. The accuracy matrix of the low-rank component is updated based on the low-rank component and its variance. Update the variance of the sparse component and the sparse component based on the accuracy of the low-rank component, the noise, and the sparse component. The accuracy of the noise is updated based on the low-rank component, the sparse component, the variance of the sparse component, and the variance of the low-rank component. Update the precision of the sparse components based on the sparse components and their variance; The update process continues until the error of the reconstructed target signal is less than a preset error threshold, at which point the low-rank and sparse components of the current round are output.

4. The fast reverberation suppression method based on variational Bayes as described in claim 3, characterized in that, The step of updating the low-rank components and their variances through generalized approximate message passing specifically includes: The precision matrix of the low-rank components is decomposed into eigenvalues, and the low-rank components and their variances are updated by combining the linear and nonlinear steps of the generalized approximate message passing algorithm.

5. The fast reverberation suppression method based on variational Bayes as described in claim 4, characterized in that, The update formula for the precision matrix of the low-rank component is as follows: in, express The Column vector, This represents the variance of the low-rank components. This represents the diagonalization operator for column vectors. This indicates the corresponding preset function parameters. The value is the total number of pixels in each frame of the sonar image.

6. The fast reverberation suppression method based on variational Bayes as described in claim 5, characterized in that: The update formula for the variance of sparse components is as follows: in, Represents the variance of the sparse component. The precision of noise representation, The precision of the sparse component; The update formula for sparse components is as follows: 。 7. The fast reverberation suppression method based on variational Bayes as described in claim 6, characterized in that, The formula for updating the accuracy of the noise is as follows: in, , This indicates the corresponding preset function parameters. The square operator represents the Frobenius norm of a matrix. The value is the total number of frames in the sonar image. This represents the trace operator for a matrix. express The Line number List.

8. The fast reverberation suppression method based on variational Bayes as described in claim 7, characterized in that, The formula for updating the precision of the sparse component is as follows: in, This indicates the corresponding preset function parameters.

9. A fast reverberation suppression system based on variational Bayes, characterized in that, include: The data conversion module is used to acquire the original three-dimensional sonar image sequence and convert it into a two-dimensional data matrix; The separation module is used to combine generalized approximate message passing and variational Bayesian methods to separate and update low-rank and sparse components. The inverse vector module is used to inverse vectorize the sparse components to obtain a three-dimensional sparse image sequence.

10. A fast reverberation suppression device based on variational Bayes, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a fast reverberation suppression method based on variational Bayes as described in any one of claims 1-8.

Images