CN-121978664-A - Quick reverberation suppression method, system and device based on variable decibel leaf

CN121978664ACN 121978664 ACN121978664 ACN 121978664ACN-121978664-A

Abstract

The invention discloses a quick reverberation suppression method, a system and a device based on variable decibels, wherein the method comprises the steps of generating a two-dimensional data matrix according to a three-dimensional sonar image sequence, separating low-rank components and sparse components in the data matrix by utilizing a variable decibels algorithm, avoiding matrix inversion steps in the variable decibels iteration process by using a generalized approximate message transfer algorithm to reduce low-rank sparse decomposition time, and inversely quantizing the two-dimensional sparse matrix into the three-dimensional sparse image sequence. The system comprises a data conversion module, a separation module and a reverse vector module. The apparatus includes a memory and a processor configured to perform the fast reverberation suppression method based on the variational Bayesian. By using the invention, reverberation suppression can be quickly and robustly implemented to detect moving objects. The invention can be widely applied to the field of active detection of underwater moving targets.

Inventors

XU LINGJI
ZENG FANCHANG
LI ZHIXI
LI ZHENGLIN

Assignees

中山大学

Dates

Publication Date: 20260505
Application Date: 20260116

Claims (10)

1. The quick reverberation suppression method based on the variable decibels is characterized by comprising the following steps of: Acquiring an original three-dimensional sonar image sequence and converting into a two-dimensional data matrix; Separating and updating low-rank components and sparse components in the two-dimensional data matrix by combining generalized approximate message transmission and a variable decibel leaf method; and inversely quantizing the sparse components to obtain a three-dimensional sparse image sequence.
2. The method for rapid reverberation suppression based on variable decibels according to claim 1, wherein the signal model of the two-dimensional data matrix is represented as follows: Wherein, the Representing a two-dimensional data matrix of the data, Indicating a low-rank component of the product, The sparse component is represented by a sparse component, Representing noise.
3. The method for fast reverberation suppression based on the variational dbis according to claim 2, wherein the step of combining the generalized approximate message passing and the variational dbis method to separate and update the low rank component and the sparse component in the two-dimensional data matrix specifically comprises the steps of: carrying out low-rank sparse decomposition on the two-dimensional data matrix by a variable decibel leaf method, and separating low-rank components and sparse components; Updating the variance of the low rank components by generalized approximation messaging; updating the precision matrix of the low rank component based on the low rank component and the variance of the low rank component; updating the variance of the sparse component and the sparse component according to the low-rank component, the precision of the noise and the precision of the sparse component; Updating the accuracy of noise according to the low rank component, the sparse component, the variance of the sparse component and the variance of the low rank component; updating the precision of the sparse component according to the sparse component and the variance of the sparse component; And circularly updating until the error of the reconstructed target signal is smaller than a preset error threshold value, and outputting a low-rank component and a sparse component of the current round.
4. A method for fast reverberation suppression based on variational dbis according to claim 3, wherein the step of updating the low rank components and the variance of the low rank components by generalized approximate messaging comprises: and performing eigenvalue decomposition on the precision matrix of the low rank component, and updating the low rank component and the variance of the low rank component by combining the linear step and the nonlinear step of the generalized approximate message transfer algorithm.
5. The quick reverberation suppression method based on the variational Bayesian as set forth in claim 4, wherein the update formula of the precision matrix of the low rank component is as follows: Wherein, the Representation of Is the first of (2) The column vector is used to determine the position of the column, Representing the variance of the low rank component, Representing the diagonalization of the column vectors into operators, Representing the corresponding parameter of the preset function, The value of (2) is the sum of pixel points of each frame of sonar image.
6. The quick reverberation suppression method based on the variational Bayesian according to claim 5, wherein the method comprises the following steps: The update formula of the variance of the sparse component is as follows: Wherein, the The variance of the sparse component is represented, Representing the accuracy of the noise and, Representing the precision of the sparse component; the update formula of the sparse component is as follows: 。
7. The quick reverberation suppression method based on the variational Bayesian according to claim 6, wherein an update formula of the accuracy of the noise is as follows: Wherein, the 、 Representing the corresponding parameter of the preset function, A squaring operator representing the matrix Frobenius norm, The value of (2) is the total frame number of the sonar image, Representing the trace-out operators of the matrix, Representation of Is the first of (2) Line 1 Columns.
8. The quick reverberation suppression method based on the variational Bayesian as set forth in claim 7, wherein the update formula of the precision of the sparse component is as follows: Wherein, the Representing the corresponding preset function parameters.
9. A quick reverberation suppression system based on a variational phylls, comprising: the data conversion module is used for acquiring an original three-dimensional sonar image sequence and converting the original three-dimensional sonar image sequence into a two-dimensional data matrix; the separation module is used for combining generalized approximate message transmission and a variable decibel leaf method, and separating and updating low-rank components and sparse components; And the reverse vector module is used for reversely quantizing the sparse components to obtain a three-dimensional sparse image sequence.
10. A quick reverberation suppression device based on a variational phylls, comprising: At least one processor; At least one memory for storing at least one program; The at least one program, when executed by the at least one processor, causes the at least one processor to implement a variational bayesian-based rapid reverberation suppression method according to any one of claims 1 to 8.

Description

Quick reverberation suppression method, system and device based on variable decibel leaf Technical Field The invention relates to the field of active detection of underwater moving targets, in particular to a quick reverberation suppression method, system and device based on variable decibels. Background Detection, positioning, tracking and identification of underwater remote small targets are core technical problems of active sonar, particularly in shallow water environments, reverberation from the seabed, the sea surface and water bodies seriously affects the working performance of the active sonar, and the detection difficulty of the small targets is increased. Thus, reverberation suppression is an essential technical component in active detection. The existing reverberation suppression method is mostly based on low-rank sparsity of multi-frame sonar images, a robust principal component analysis technology is used for realizing separation of low-rank reverberation and sparse targets, the low-rank sparse algorithms are all realized based on an optimization algorithm, and a penalty factor of a sparse term needs to be manually adjusted, particularly, under the condition of low signal-to-noise ratio, the low-energy targets cannot be decomposed in a sparse matrix due to the fact that the sparse penalty factor is too large, and most of the reverberation suppression algorithms are time-consuming and difficult to meet instantaneity of active sonar. Disclosure of Invention In view of this, in order to solve the technical problem that the existing reverberation suppression method needs to select a penalty factor and needs to perform matrix inversion operation, which results in too long calculation time, in a first aspect, the present invention provides a fast reverberation suppression method based on a variational Bayesian, which specifically includes: The method comprises the steps of vectorizing each frame of images according to a time sequence of an original three-dimensional sonar image sequence, then spelling together to obtain a two-dimensional data matrix, separating low-rank components and sparse components in the data matrix by using a variable dB leaf algorithm, avoiding matrix inversion steps in a variable dB leaf iteration process by using a generalized approximate message transfer algorithm to reduce low-rank sparse decomposition time, inversely quantifying the two-dimensional sparse matrix into a three-dimensional sparse image sequence, and finally using nonlinear superposition to further enhance a target to obtain a moving target track. Compared with other reverberation suppression algorithms, the method avoids the selection of punishment factors in an optimization algorithm, avoids a time-consuming matrix inversion step, and can rapidly and robustly implement reverberation suppression so as to detect a moving target. The invention also provides a quick reverberation suppression system based on the variable decibels, which comprises a data conversion module, a separation module and a reverse vector module. The invention also provides a quick reverberation suppression device based on the variable dB leaf, which comprises: At least one processor; At least one memory for storing at least one program; the at least one program, when executed by the at least one processor, causes the at least one processor to implement a fast reverberation suppression method based on a variational Bayesian as described above. Based on the scheme, the invention provides a quick reverberation suppression method, a system and a device based on the variable decibel leaf, which can be better applied to the field of underwater moving small target active detection by combining a generalized approximate message transmission algorithm through the variable decibel leaf inference method, avoid the selection of regularization parameters (penalty factors) in an optimization method and the high computational complexity caused by matrix inversion operation, and have lower complexity compared with the traditional reverberation suppression method, particularly have shorter calculation time for high-dimensional data, can suppress more reverberation energy while keeping target energy, and improve algorithm robustness and target detection performance. Drawings FIG. 1 is a flow chart showing the steps of a method for fast reverberation suppression based on a variable decibel leaf according to the present invention; Fig. 2 is a schematic diagram of a frame of image result in an original sonar image sequence. Fig. 3 is a graph illustrating the result of VBRPCA reverberations suppression on the data frame of fig. 2. Fig. 4 is a schematic diagram of reverberation suppression results of the method VBGAMP of the present invention performed on the data frame of fig. 2. Fig. 5 is a schematic diagram of APG reverberation suppression results for the data frame of fig. 2. Fig. 6 is a schematic diagram of ADMM reverberation suppression results for the da