Method for online reverberation cancellation and target detection of active sonar imaging data

Abstract

The application discloses a kind of active sonar imaging data online reverberation elimination and target detection method, comprising: sonar obtains current time data frame;Determine whether current is in initialization phase;With current observation frame and target component, update reverberation base;And the reverberation signal base updated by updated reverberation subspace weight update;Re-estimate target component, this is used as the estimation of detecting target, realize online real-time reverberation elimination and target detection;The application solves the problem of high computational complexity and large resource consumption caused by not considering the time series data characteristics in conventional low-rank-sparse decomposition method.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater acoustic information processing and electronic information technology, and particularly relates to a method for online reverberation elimination and target detection of active sonar imaging data. BACKGROUND

[0002] Ensuring the safety of ports underwater is of great significance to China's coastal areas. In shallow water active detection, typical targets include small underwater vehicles and divers. These targets pose a serious threat to surface ships, ports, wharfs and military facilities due to their good concealment and strong surprise mobility. Active sonar is the main means for detecting underwater intruding targets such as divers and small unmanned vehicles. In general, active sonar works in a noisy or reverberation environment. The reverberation is formed by the scattering and superposition of a large number of ocean scatterers on the active detection signal. In shallow water environment, the bottom reverberation is the main component of the reverberation. Due to the strong scattering signal of the seabed, which is very similar to the target echo signal, strong reverberation reduces the performance of the active sonar detection system, so it is difficult to effectively detect the target through a single frame of sonar data. The low-rank-sparse decomposition method based on multiple imaging frames is an effective means for reverberation suppression and target signal detection. This method decomposes the steady part of the reverberation in a low-rank matrix, and decomposes the target signal and the fluctuation part of the reverberation in a sparse matrix. However, the existing low-rank-sparse decomposition-based sonar imaging reverberation elimination technology mostly uses offline data models. This method stores a large amount of sonar data as multiple frames of images, and needs to use the newly acquired data frame and the stored data frame to perform low-rank-sparse decomposition at each time. This method does not consider the time sequence of sonar image data acquisition, has high complexity, and is not conducive to the low-complexity deployment of sonar target detection algorithms. Therefore, it is of great significance to establish an online real-time low-rank-sparse decomposition-based reverberation suppression algorithm for the acquired data. SUMMARY

[0003] To solve the problems in the prior art, the purpose of the present application is to provide a method for online reverberation elimination and target detection of active sonar imaging data. The present application solves the problems of high computational complexity and large resource consumption caused by the fact that the conventional low-rank-sparse decomposition method does not consider the time sequence of data.

[0004] To achieve the above purpose, the technical scheme adopted by the present application is as follows: a method for online reverberation elimination and target detection of active sonar imaging data, comprising the following steps:

[0005] Step 1: the sonar acquires a data frame at the current time;

[0006] Step 2: Determine whether the current stage is the initialization stage: When in the initialization stage, determine whether a sufficient number of data frames for initialization have been accumulated. If not, continue to acquire data frames. If sufficient data frames have been acquired, calculate the mean of the data frames and remove the mean of the data frames. Consider the data in the initialization stage as reverberation background data and calculate its feature space dimension and basis. When not in the initialization stage, if initialization has just ended and the current frame is the first frame after initialization, subtract the estimated reverberation background data from the data frame to preliminarily estimate the target component. If it is not the first frame after initialization, use the result of the previous moment to preliminarily estimate the target component.

[0007] Step 3: Update the reverberation base using the current observation frame and the target component; and update the reverberation subspace weights using the updated reverberation signal base;

[0008] Step 4: Re-estimate the target component and use it as the estimation of the detected target to achieve online real-time reverberation elimination and target detection; obtain the next data frame and repeat steps 1 to 4.

[0009] As a further improvement of the present invention, the step 1 is specifically as follows:

[0010] In a sonar system using a circular receiving array, after the transmitting element emits a linear frequency modulation signal, the reflected signals at different angles are collected through beamforming and correlated. For each set receiving angle, a vector of length P is formed based on the echo delay. The above process is repeated for a total of Q angles to form a P×Q azimuth-range sonar imaging matrix. If the above operation is continued, a series of azimuth-range sonar imaging matrices at different times are formed, that is, a frame of data. The azimuth-range matrix at a specific time t is recorded as M t ; Concatenate the columns of the matrix to form a PQ×1 sonar data vector m t .

[0011] As a further improvement of the present invention, in step 2, after the sonar is turned on, the initialization phase starts from time t=1 and continuously receives data, which can accumulate T c The batch initialization algorithm is used to estimate the reverberation background and parameters based on the frame data.

[0012] As a further improvement of the present invention, a batch initialization algorithm is used to estimate the reverberation background and parameters as follows:

[0013] The sonar data vector m t (t=1,…,T c ) are spliced ​​into a dimension of PQ×T c Matrix In the initialization stage, it is assumed that there is no target in the scene, so the data is column-wise mean-removed, and then singular decomposition is used to determine the signal subspace;

[0014] Construct T c ×T c of size PQ×PQ Perform eigen-decomposition on the smaller dimension square matrix to obtain:

[0015]

[0016] The total of T c groups of eigenvalues and eigenvectors; assuming that the eigenvalues are arranged in descending order, the first B eigenvalues and corresponding eigenvectors that account for the proportion r of the total are obtained (i = 1, …, B); from the eigenvalues and eigenvectors, the i-th group of singular values and singular vectors of Z c are obtained:

[0017]

[0018]

[0019]

[0020] Construct the initialization of the reverberation background component of the sonar:

[0021]

[0022] As a further improvement of the present application, in step 4, online real-time reverberation elimination and target detection are as follows:

[0023] At time t, the sonar observation data vector m t can be expressed as:

[0024] m t = U t w t + g t + n t

[0025] where U t is the subspace in which the reverberation signal is located, and its dimension is PQ×B, where B is determined during initialization; w t is a B×1 subspace coefficient; the PQ×1 vector g t represents the component composed of sparse targets; the PQ×1 vector n t represents the model noise component; online reverberation elimination uses the results U t-1 , w t-1 , g t-1 of the previous moment and the observation data mt updating variable U t , w t and g t ;

[0026] The online reverberation cancellation and target detection is performed by using the following mathematical optimization objective function:

[0027]

[0028] The objective function is divided into three parts, the first part describes the accuracy of the sonar data frame observed at time t using the low-rank-sparse signal model reconstruction, the second part promotes the sparse characteristics of the target component, and the third part promotes the proximity of the reverberation component subspace at time t to that at time t-1; wherein U t ∈Θ indicates that U t should be located in a standard orthogonal subspace, q is a weight value related to the pixel position, and non-negative η and τ are used to control the strength of the second and third terms; after collecting a new data frame at time t, the mean value is estimated using the time smoothing method, and the mean value is removed, and the current sonar data observation frame and the reverberation subspace result estimated at the last time are used to perform online reverberation component cancellation and target signal detection.

[0029] As a further improvement of the present application, the online reverberation component cancellation and target signal detection using the current sonar data observation frame and the reverberation subspace result estimated at the last time specifically includes:

[0030] According to the reverberation background at the last time, i.e. the reverberation subspace basis U t-1 and the coefficient w t-1 , the target signal is preliminarily estimated, i.e.:

[0031]

[0032] can be obtained If the current frame is the first frame after initialization, L c is used instead of U t-1 w t-1 ;

[0033] The reverberation subspace basis U t-1 is updated by a gradient method to obtain the updated value U' t :

[0034]

[0035] Then U' t is orthogonalized and normalized to obtain U t = Θ(U' t );

[0036] The reverberation subspace coefficient w is solved by solving a quadratic optimizationt :

[0037]

[0038] obtained

[0039] According to the update obtained U t , w t Again, the target signal g t :

[0040]

[0041] obtained

[0042] As a further improvement of the application, it also includes:

[0043] Every T e frame of sonar data is collected and online calculation is performed, the initialization process is called again, and the reverberation spatial dimension and basis are recalculated by offline algorithm.

[0044] As a further improvement of the application, it also includes:

[0045] In a continuous time T s , the reverberation subspace basis is not updated, and only the subspace weight adjustment is used to fine-tune the reverberation estimation.

[0046] The beneficial effects of the application are:

[0047] The application solves the problems of high computational complexity and large resource consumption caused by the fact that the conventional low-rank-sparse decomposition method does not consider the time series data characteristics. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A schematic diagram is constructed for the signal in the embodiment of the application.

[0049] Figure 2 A flowchart of the embodiment of the application. DETAILED DESCRIPTION

[0050] The embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0051] Embodiment

[0052] An active sonar imaging data online reverberation elimination and target detection method, comprising:

[0053] S1: the sonar acquires the current time data frame;

[0054] S2: determine whether the current is in the initialization phase;

[0055] During the initialization phase:

[0056] S2.1: Determine whether a sufficient number of data frames for initialization have been accumulated. If not, proceed to S1 to obtain data frames.

[0057] S2.2: If sufficient data frames have been obtained, calculate the mean of the data frames and remove the mean of the data frames;

[0058] S2.3: Consider the data from the initialization phase as reverberation background data and calculate its feature space dimension and basis;

[0059] When not in the initialization phase:

[0060] S3.1: If initialization has just finished and the current frame is the first frame after initialization, subtract the reverberation background estimated in S2.3 from the data frame to preliminarily estimate the target component;

[0061] S3.2: If it is not the first frame after initialization, use the result of the previous moment to preliminarily estimate the target component;

[0062] S4: Update the reverberation base using the current observation frame and the target component in S3.1 or S3.2;

[0063] S5: Update the reverberation subspace weights using the current observation frame, the target component in S3.1 or S3.2, and the reverberation signal basis updated in S4;

[0064] S6: Using the result in S5, re-estimate the target component and use this as the estimate of the detected target;

[0065] S7: Get the next data frame and repeat the process starting from S1.

[0066] The present embodiment is further described below:

[0067] Figure 1 It is a schematic diagram of signal construction. As shown in 101, the sonar system using a circular receiving array, after the transmitting array emits a linear frequency modulation signal, collects the reflected signals at different angles through beam forming and performs correlation matching. For each set receiving angle, a vector of length P can be formed according to the echo delay. As shown in 102, the above process is performed on Q angles in total, and a P×Q azimuth-range sonar imaging matrix as shown in 103 can be formed. If the above operation is continued, a series of azimuth-range sonar imaging matrices at different times can be formed, that is, a frame of data. The azimuth-range matrix at a specific time t is recorded as M t The columns of the matrix can be concatenated to form a PQ×1 vector m as shown in 104. t .

[0068] Figure 2is the whole process of the embodiment of the present embodiment.

[0069] Initialization phase of the algorithm: after the sonar is started, data is continuously received from t = 1, such as 201, and T c frame data can be accumulated first, and the reverberation background and parameters can be estimated by using a batch algorithm to initialize the algorithm (as shown in 202 and 203). c The value of T is related to the parameters P and Q of the sonar imaging setting and the reverberation dynamic characteristics. Preferably, it is recommended to be no less than 20.

[0070] First, the sonar data vector m t (t = 1, …, T c ) is spliced into a matrix of PQ × T c dimension. In the initialization phase, it can be assumed that there is no target in the scene, so the data can be column-wise mean as 204, and then the signal subspace can be determined by using the conventional singular decomposition, but it is noted that PQ is usually much larger than T c , and the following fast algorithm is used to determine step 205.

[0071] A matrix of T c × T c size is constructed. The eigenvalues of the smaller dimension square matrix are obtained:

[0072]

[0073] There are T c groups of singular values and singular vectors. Not generally, the eigenvalues are arranged in descending order, and the first B eigenvalues and corresponding eigenvectors that occupy the total proportion r (preferably r = 95%) are obtained From these eigenvalues and eigenvectors, the i-th group of singular values and singular vectors of Z c can be obtained as follows:

[0074]

[0075]

[0076]

[0077] From the above quantities, the initialization of the reverberation background component of the sonar can be constructed:

[0078]

[0079] In underwater environment, B is usually small, so the above method can initialize the algorithm quickly with low complexity. Online real-time reverberation cancellation and target detection: At time t, the sonar observation data vector m t can be expressed as:

[0080] m t = U t w t + g t + n t

[0081] where U t is the subspace of reverberation signals with dimension PQx B, where B is determined at initialization; w t is the subspace coefficient of Bx1; g t is the component of sparse targets with PQx1; n t is the model noise component with PQx1. Online reverberation cancellation aims to update the variables U t-1 , w t-1 , g t-1 and m t using the results of the previous time U t , w t , g t and the current observation data m t , so that the reverberation and targets are estimated using less data rather than all the accumulated data.

[0082] The embodiment proposes to use the following mathematical optimization objective function for online sonar reverberation cancellation and target detection:

[0083]

[0084] The objective function is divided into three parts, the first part describes the accuracy of the low-rank-sparse signal model used to reconstruct the observed sonar data frame at time t, the second part promotes the sparsity of the target component, and the third part promotes the proximity of the reverberation component subspace at time t to that at time t-1. Where U t ∈ Θ indicates that U t should be located in a standard orthogonal subspace, q is a weight value related to the pixel position, and non-negative η and τ are used to control the strength of the second and third terms. After collecting a new data frame at time t, the mean is estimated using the time smoothing method, and the mean is removed (206). The following steps can be used to perform online reverberation component cancellation and target signal detection using the current sonar data observation frame and the reverberation subspace result estimated at the previous time.

[0085] First step (208, 209): Estimate the target signal (209) based on the reverberation background at the previous time, i.e. the reverberation subspace basis U t-1 and the coefficient w t-1 , i.e.

[0086]

[0087] The updated value U' can be obtained If the current frame is the first frame after initialization, the L c is replaced by U t-1 w t-1 (208).

[0088] Second step (210): update the reverberation subspace basis U t-1 by gradient method t :

[0089]

[0090] Then U' is orthogonalized and normalized to obtain U t : t = Θ (U' ) t ).

[0091] Third step (211): solve the reverberation subspace coefficient w t by solving the quadratic optimization

[0092]

[0093] The updated value U' can be obtained

[0094] Fourth step (212): according to the updated U t , w t , update the estimated target signal g t again:

[0095]

[0096] The updated value U' can be obtained

[0097] To further reduce the complexity of the algorithm and increase the stability of the algorithm, the following methods can be used to adjust the above-mentioned methods. Including:

[0098] Method one: after collecting T e frame of sonar data and performing online calculation, the initialization process is called again, and the reverberation space dimension and basis are recalculated by offline algorithm to avoid the deficiency of long-time running of online algorithm on reverberation subspace tracking, and increase the stability of the system. That is, the judgment of 202 module is replaced by "whether it is necessary to initialize at T e time".

[0099] Method two: considering that the sonar reverberation background is a slow-changing process, a continuous time T s can be considered.The inner reverberation subspace basis is not updated through step 210, but is only adjusted through 211 subspace weight adjustment, and the reverberation estimation is fine-tuned, so that the overall complexity of the target detection system is reduced. That is, a judgment "whether T s time" is added between the 209 and 210 modules. If yes, the 209 module is still linked to the 210 module. If no, the 209 module is directly connected to the 211 module.

[0100] The above-described embodiments only express specific implementation manners of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for online reverberation elimination and target detection of active sonar imaging data, characterized in that: The following steps are involved: Step 1: Sonar obtains the current time data frame; Step 2: Determine whether the system is currently in the initialization phase. If the system is in the initialization phase, determine whether a sufficient number of data frames have been accumulated for initialization. If not, continue to acquire data frames. If sufficient data frames have been acquired, calculate the mean of the data frames and remove the mean of the data frames. The data in the initialization phase is regarded as the reverberation background data, and its feature space dimension and basis are calculated. When the data is not in the initialization phase, such as when the initialization has just ended and the current frame is the first frame after initialization, the estimated reverberation background data is subtracted from the data frame to preliminarily estimate the target component. If it is not the first frame after initialization, use the result of the previous moment to preliminarily estimate the target component; Step 3: Update the reverberation base using the current observation frame and the target component; and updating the reverberation subspace weights using the updated reverberation signal basis; Step 4: Re-estimate the target component and use it as the estimation of the detected target to achieve online real-time reverberation elimination and target detection; obtain the next data frame and repeat steps 1 to 4; In step 4, online real-time reverberation elimination and target detection are as follows: At time t, the sonar observation data vector m t It can be expressed as: m t =U t w t +g t +n t Among them U t is the subspace where the reverberation signal is located, and its dimension is PQ×B, where B is determined by initialization; w t is the B×1 subspace coefficient; PQ×1 vector g t Represents the components of the sparse target; PQ×1 vector n t Represents the model noise component; online reverberation elimination uses the result U of the previous moment t-1 、w t-1 、g t-1 And the observation data m at this moment t Update variable U t 、w t and g t ; The following mathematical optimization objective function is used for online sonar reverberation elimination and target detection: The objective function is divided into three parts. The first part characterizes the accuracy of reconstructing the observed sonar data frame using the low-rank-sparse signal model at time t. The second part promotes the sparse characteristics of the target component. The third part promotes the closeness of the reverberation component subspace at time t to that at time t-1. Among them U t ∈Θ indicates that U t It should be located in a standard orthogonal subspace, q is a weighted value related to the pixel position, and non-negative η and τ are used to control the strength of the second and third terms; After collecting a new data frame at time t, the time smoothing method is used to estimate the mean and perform de-meaning. The reverberation component elimination and target signal detection are performed online using the current sonar data observation frame and the reverberation subspace results estimated at the previous moment.

2. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 1, characterized in that: The step 1 is specifically as follows: In a sonar system using a circular receiving array, after the transmitting element emits a linear frequency modulation signal, the reflected signals at different angles are collected through beamforming and correlated. For each set receiving angle, a vector of length P is formed based on the echo delay. The above process is repeated for a total of Q angles to form a P×Q azimuth-range sonar imaging matrix. If the above operation is continued, a series of azimuth-range sonar imaging matrices at different times are formed, that is, a frame of data. The azimuth-range matrix at a specific time t is recorded as m t ; Concatenate the columns of the matrix to form a PQ×1 sonar data vector m t .

3. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 2, characterized in that: In step 2, the initialization phase starts from t=1 after the sonar is turned on and continuously receives data, which can accumulate T c The batch initialization algorithm is used to estimate the reverberation background and parameters based on the frame data.

4. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 3, characterized in that: The batch initialization algorithm is used to estimate the reverberation background and parameters as follows: The sonar data vector m t , t=1,…,T c , spliced ​​into a dimension of PQ×T c Matrix In the initialization phase, it is assumed that there is no target in the scene, so the data is meaned by column, and then the signal subspace is determined by singular decomposition; Build T c ×T c Matrix of size Perform eigendecomposition on the smaller dimensional matrix to obtain: Total T c Group eigenvalues ​​and eigenvectors; assuming that the eigenvalues ​​are sorted from large to small, obtain the first B eigenvalues ​​and corresponding eigenvectors that occupy a proportion r of the total Depend on Eigenvalues ​​and eigenvectors, obtain Z c The i-th group of singular values ​​and singular vectors: Construct the initialization of the reverberation background component of the sonar:

5. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 1, characterized in that: The reverberation component elimination and target signal detection are performed online using the current sonar data observation frame and the reverberation subspace result estimated at the previous moment. Specifically, the following steps are performed: According to the reverberation background of the previous moment, that is, the reverberation subspace basis U t-1 and coefficient w t-1 Preliminary estimate of the target signal, namely: Available If the current frame is the first frame after initialization, use L c Alternative U t-1 w t-1 ; Update the reverberation subspace basis U by gradient method t-1 Get the updated value U′t: Then U′ t Orthogonalize and normalize to obtain U t =Θ(U′ t ); Solve the reverberation subspace coefficient w by solving the quadratic optimization t : available According to the updated U t 、w t Estimate the target signal g again t : available 6. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 5, characterized in that: Also includes: Each collection T e After the frame sonar data is collected and calculated online, the initialization process is called again and the reverberation space dimensions and floor are recalculated using the offline algorithm.

7. The method for online reverberation elimination and target detection of active sonar imaging data according to claim 5, characterized in that: Also includes: In a continuous period of time T s The reverberation subspace basis is not updated, and the reverberation estimation is fine-tuned only by adjusting the subspace weights.

Images