A method, apparatus, and device for reverberation removal in moving target detection

Abstract

This invention discloses a method, apparatus, and device for reverberation removal in moving target detection. The method includes: acquiring multiple frames of target echo data and preprocessing the data; constructing a robust autoencoder based on the preprocessed data; using the constructed robust autoencoder to extract steady-state and dynamic components from the preprocessed data and iteratively optimizing the autoencoder; and using the optimized autoencoder to remove reverberation from the acquired multiple frames of target echo data. The method proposed in this invention can effectively remove reverberation in moving target detection, obtaining a clear trajectory. In artificial sonar detection of moving targets, it can effectively extract the target's trajectory, contributing to improved target detection distance and accuracy in target localization and identification.

Description

Technical Field

[0001] This invention relates to the field of reverberation elimination technology, and in particular to a method, apparatus, and device for reverberation removal in moving target detection. Background Technology

[0002] Active sonar is a commonly used method for underwater target detection and identification. However, target echoes are accompanied by strong reverberation interference at the receiver, making it difficult to distinguish the target from background noise and reverberation, thus degrading the performance of artificial sonar in underwater target detection. Therefore, reducing reverberation interference is key to improving underwater target detection performance.

[0003] Traditional reverberation reduction methods mainly focus on mapping the characteristic subspace, separating the target echo by examining the differences between reverberation and the target echo in Doppler space, discrete wavelet transform domain, and fractional Fourier transform domain. However, these methods perform poorly under low signal-to-mixing ratio conditions. In recent years, low-rank and sparse theories have been introduced into underwater acoustic engineering for reverberation suppression and target detection. For active sonar at a fixed location, the signals received from multiple transmitted pulses have good stationarity, and the multi-pulse target echo can be decomposed into low-rank and sparse matrices. Researchers have proposed reverberation removal methods based on the low-rank and sparse decomposition of matrices. Jia et al. used nonnegative matrix decomposition, Zhang et al. used principal component analysis, and Amailland used robust principal component analysis. Zhu et al. from Northwestern Polytechnical University used alternating direction multipliers to accelerate the matrix decomposition process and improved the robustness of reverberation suppression through sparse compensation. However, the low-rank matrices estimated by the above methods are all based on linear space, which is inaccurate for estimating the nonlinear distribution of reverberation in actual underwater environments.

[0004] References:

[0005] [1]Urick,RJPrinciples of underwater sound,3rd ed[M].PeninsulaPublishing,Newport beach,USA,2013:237–267and 377–400.

[0006] [2]Zhang,Y.,d’Aspremont,and Ghaoui,L.Sparse PCA:Convex relaxations,algorithms and applications[A].In Handbook on Semidefinite,Conic andPolynomial Optimization,M.F.Anjos and J.B.Lasserre,Ed[C].New York:Springer.2012.pp.915-940.

[0007] [3]Amailland,S.,Thomas,J.,Pézerat,C.,and Boucheron,R.Boundary layernoise subtraction in hydrodynamic tunnel using robust principal componentanalysis[J].J.Acoust.Soc.Am.,2018,143(4):2152-2163.

[0008] [4]Zhu,Y.,Duan,R.,Yang,K.,Xue,R.,and Wang,N.Reverberation reductionbased on multi-ping association in a moving target scenario[J].J.Acoust.Soc.Am.2020,148(4):2195-2208.

[0009] [5]Zhu,Y.,Yang,K.,Duan,R.,and Wu,F.Sparse spatial spectral estimationwith heavy sea bottom reverberation in the fractional Fourier domain[J].Applied Acoustics.2020,160:1-12.

[0010] [6]Zhou,C.,and Paffenroth,RCAnomaly Detection with Robust DeepAutoencoders[A].In the 23rd ACM SIGKDD International Conference[C].Canada,2017.pp.666-674. Summary of the Invention

[0011] In view of this, in order to better estimate low-rank reverberation components and separate targets, the present invention aims to propose a reverberation removal method, apparatus, and device for moving target detection. This method utilizes an autoencoder to project high-dimensional data into a low-dimensional space, extracts stable components through nonlinear estimation, and then removes the steady-state components from the overall data to obtain dynamic components. Furthermore, a sparse representation method is used to extract target echoes from the dynamic components. Finally, connected component analysis is used to extract clear target echo trajectories.

[0012] According to one aspect of the present invention, a reverberation removal method for moving target detection is provided, comprising: acquiring multiple frames of target echo data and preprocessing the multiple frames of target echo data; constructing a robust autoencoder based on the preprocessed multiple frames of target echo data; using the constructed robust autoencoder to extract steady-state components and dynamic components of the preprocessed multiple frames of target echo data and iteratively optimizing the constructed robust autoencoder; and removing reverberation corresponding to the acquired multiple frames of target echo data according to the iteratively optimized robust autoencoder.

[0013] According to another aspect of the present invention, a reverberation removal device for moving target detection is provided, comprising: an acquisition module, a construction module, a training module, and a prediction module; the acquisition module is configured to acquire multi-frame target echo data and preprocess the multi-frame target echo data; the construction module is configured to construct a robust autoencoder based on the preprocessed multi-frame target echo data; the training module is configured to iteratively optimize the constructed robust autoencoder by extracting steady-state and dynamic components of the preprocessed multi-frame target echo data; and the prediction module is configured to remove reverberation corresponding to the acquired multi-frame target echo data based on the iteratively optimized robust autoencoder.

[0014] According to another aspect of the present invention, a reverberation removal device for moving target detection is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the reverberation removal method for moving target detection as described in any of the preceding claims.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the reverberation removal method for moving target detection as described in any of the preceding claims.

[0016] It can be observed that, to address the issue of inaccurate reverberation estimation in nonlinear distributions within actual underwater environments, the above scheme first uses a robust autoencoder to nonlinearly estimate the background reverberation, taking advantage of its low rank and target echo sparsity. After reverberation removal, target echo components are estimated using a regularized adaptive estimation based on threshold sparsity compensation to extract the target trajectory. Finally, connected component analysis is used to further highlight the target echo trajectory based on the correlation of target positions across multiple time frames. The method proposed in this invention can effectively remove reverberation and obtain a clear trajectory in moving target detection. When using artificial sonar to detect moving targets, it can effectively extract the target trajectory, contributing to improved accuracy in target detection range, positioning, and identification. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the reverberation removal method for moving target detection according to the present invention;

[0019] Figure 2 This is a true target echo intensity map of an embodiment of the reverberation removal method in moving target detection of the present invention;

[0020] Figure 3 This is a mean normalization result diagram of an embodiment of the reverberation removal method in moving target detection of the present invention;

[0021] Figure 4 This is an image showing the adaptive background reverberation component estimation result of an embodiment of the reverberation removal method in moving target detection of the present invention;

[0022] Figure 5 This is an image showing the adaptive target echo component estimation result of an embodiment of the reverberation removal method in moving target detection of the present invention;

[0023] Figure 6 This is a diagram showing the echo component binarization result of an embodiment of the reverberation removal method in moving target detection of the present invention;

[0024] Figure 7 This is a target motion trajectory diagram after connected component analysis processing, according to an embodiment of the reverberation removal method in moving target detection of the present invention. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention provides a reverberation removal method for moving target detection. It utilizes an autoencoder to project high-dimensional data into a low-dimensional space, extracting stable components through nonlinear estimation. Then, the steady-state components are removed from the overall data to obtain dynamic components. Furthermore, a sparse representation method is used to extract target echoes from the dynamic components. Finally, connected component analysis is employed to extract clear target echo trajectories.

[0027] Please see Figure 1 , Figure 1 This is a schematic flowchart of an embodiment of the reverberation removal method for moving target detection according to the present invention. It should be noted that if substantially the same result is obtained, the method of the present invention is not necessarily identical. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown, the method includes the following steps:

[0028] S100: Acquire multi-frame target echo data and preprocess the multi-frame target echo data.

[0029] Please see Figure 2 , Figure 3 In this embodiment, a column vector is defined to represent the acquired multi-frame target echo data, and the multi-frame target echo data represented by the vector is normalized by mean, and the normalized multi-frame target echo data is labeled.

[0030] In this embodiment, the target echo data of multiple frames is generally selected from more than 500 frames, preferably 800 frames. It can be understood that when 800 frames of echo data are selected, the processing efficiency is the highest.

[0031] Specifically,

[0032] S101. Define a column vector. Let M represent the echo of the received i-th pulse, where M represents the dimension in the distance direction, x... i This is called the i-th frame of data. Then, multi-frame data can be represented as... X = [x1, x2, ... x N Please see ]. Figure 2 Multiple frames of target echo data can be used to generate an echo intensity map through vector representation.

[0033] In this embodiment, an active sonar is a device used to detect underwater targets and determine their motion parameters such as distance, bearing, speed, and heading. It consists of a sonar emitting a detection signal. This signal travels through the water, encounters obstacles or targets along its path, and is reflected back to the transmitting point where it is received. Since target information is stored in the echo reflected back from the target, the target's parameters can be determined based on the received echo signal. This sonar can accurately determine target distance and can detect stationary targets. Based on this, it can be known that active sonar detection has a certain directional characteristic in a specific direction, which can be used to characterize the direction and distance of obstacles. Therefore, in this embodiment, the column vector is defined using the distance of the echo as its dimension, which can better represent multi-frame target echo data. Similarly, the column vector dimension for multi-frame target echo data can be selected from other parameters; this embodiment does not limit this.

[0034] S102. Mean normalization of multi-frame target echo data: Normalize the mean of the matrix X of the received N frames of data by calculating the mean of each row in X. Then, after normalizing the mean of each frame of data, we get... The matrix after mean normalization of X is denoted as Xmean. ′ =[x ′ 1,x ′ 2,…x ′ N Please see ]. Figure 3 Multiple frames of target echo data can be used to generate an echo intensity map through vector representation. After further mean normalization, the result map shown in the figure is obtained.

[0035] In this embodiment, the multi-frame target echo data is large in volume and exhibits significant numerical differences between the presence and absence of obstacles, with substantial differences in magnitude across multiple dimensions. Therefore, this embodiment employs mean normalization to perform uniform preprocessing on the multi-frame target echo data, facilitating subsequent comparison and calculation.

[0036] S200. Construct a robust autoencoder based on the preprocessed multi-frame target echo data;

[0037] In this embodiment, the autoencoder is an unsupervised method for data dimensionality compression and data feature representation. In most instances where autoencoders are mentioned, the compression and decompression functions are implemented using neural networks. The robust autoencoder constructed in this embodiment projects high-dimensional, pre-processed multi-frame target echo data into a low-dimensional space, extracts the steady-state components of the pre-processed multi-frame target echo data through nonlinear estimation, and then removes the steady-state components from the overall data to obtain the dynamic components.

[0038] In this embodiment, the steady-state component refers to the background reverberation component that needs to be removed from the multi-frame target echo data; the dynamic component refers to the target echo component in the multi-frame target echo data.

[0039] S300. The robust autoencoder is used to extract the steady-state and dynamic components of the preprocessed multi-frame target echo data and to iteratively optimize the robust autoencoder.

[0040] In this embodiment, a robust autoencoder is used to extract the background reverberation matrix and target matrix from the preprocessed multi-frame target echo data. The components in the background reverberation matrix and target matrix represent the steady-state component and dynamic component of the i-th frame, respectively, and are initialized to zero matrices. The preprocessed multi-frame target echo data is then assigned the background reverberation matrix, and the parameters of the robust autoencoder are randomly initialized. The dynamic component is fixed, and the robust autoencoder is used to perform nonlinear estimation of the steady-state component by minimizing... To calculate the steady-state components; and through the minimized The calculated steady-state components are used to calculate the dynamic components; and the minimized components are fixed. The calculated steady-state components are optimized using a proximal operator; and the convergence coefficient is calculated and iterated by optimizing the convergence coefficient until the robust autoencoder achieves convergence.

[0041] In this embodiment, based on the low rank of the background reverberation and the sparsity of the target echo, a robust autoencoder is used to nonlinearly estimate the background reverberation. After removing the reverberation, the target echo component is estimated by regularization adaptive estimation based on threshold sparsity compensation, and the target trajectory is extracted.

[0042] Specifically,

[0043] S301. Extract background reverberation and target components using a robust autoencoder. Divide X′ into two parts L=[l1,l2,…,l N ] and S = [s1,s2,…,s N ], and initialize it to a zero matrix, where l i and s iThese represent the steady-state component and the dynamic component of the i-th frame, respectively;

[0044] In this embodiment, the steady-state component refers to the background reverberation component that needs to be removed from the multi-frame target echo data; the dynamic component refers to the target echo component in the multi-frame target echo data.

[0045] In this embodiment, X′ represents preprocessed multi-frame target echo data.

[0046] In this embodiment, background reverberation and target components can be understood as data features of multi-frame target echo data, while an autoencoder is a data compression and feature extraction representation method based on unsupervised learning. That is, the autoencoder can be used to extract data features (i.e., background reverberation and target components) of multi-frame target echo data.

[0047] S302, Order L D =X′, L D =[l D1 ,l D2 ,…,l DN At the same time, the parameters of the autoencoder are randomly initialized;

[0048] In this embodiment, the processed data X′ is assigned to a low-rank reverberation matrix as the training input for the robust autoencoder.

[0049] S303, Fixed S i , let l i =l Di Using a robust autoencoder for l i Perform nonlinear estimation by minimizing To calculate l i .

[0050] In this embodiment, the dynamic component (i.e., the target echo component in the multi-frame target echo data) is fixed, and the steady-state component (i.e., the background reverberation component that needs to be removed from the multi-frame target echo data) is nonlinearly estimated using a robust autoencoder. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The steady-state component obtained after nonlinear estimation processing (i.e., the background reverberation component that needs to be removed from the multi-frame target echo data).

[0051] S304, via s i =x′ i -l i Calculate the dynamic components, with l fixed. i Using proximal operators To optimize s i For s i Each element s in ij Compensation is performed using the segmentation factor λ.

[0052]

[0053] λ is adaptively selected based on the mean intensity and standard deviation of the dynamic component S, and is defined as:

[0054]

[0055]

[0056] In the formula, s ji Let be the element in row i and column j of S.

[0057] In this embodiment, the target echo component is estimated through regularized adaptive estimation based on threshold sparsity compensation. This can be understood as using an adaptive threshold (shrinkage factor) to determine and compensate for the sparsity of the target component, increasing its sparsity and making the estimated target component more accurate. Dynamic components contain a lot of clutter, so optimization is needed in this step. The purpose of optimization is to set conditions to extract the target component more accurately; otherwise, there may be too much clutter. Please refer to [link to relevant documentation]. Figure 5 The result of calculating the dynamic component (i.e., the target echo component in multi-frame target echo data) by using nonlinear estimation of the steady-state component.

[0058] S305. Calculate the convergence coefficient. and Check whether the constraint condition that the sum of L and S is close to X′ satisfies convergence;

[0059] S306. Determine whether c1 and c2 are less than the error factor ε. If c1 < ε or c2 < ε, then complete the decomposition and return the steady-state component L, the noise component G = XLS, and the target component D = S; otherwise, let L D =XS, proceed to step S303, until L and S are estimated.

[0060] In this embodiment, based on the low-rank characteristics of background reverberation and the sparsity of target echoes, a robust autoencoder is used to nonlinearly estimate the background reverberation. After removing the reverberation, the target echo components are estimated using a regularized adaptive estimation based on threshold sparsity compensation, and the target trajectory is extracted. This technical solution can effectively remove reverberation in moving target detection, obtaining a clear trajectory. When using artificial sonar to detect moving targets, it can effectively extract the target's trajectory, contributing to improvements in target detection range and the accuracy of target localization and identification.

[0061] S400. Based on the iteratively optimized robust autoencoder, the reverberation of the acquired multi-frame target echo data is removed.

[0062] In this embodiment, the target echo is extracted from the multi-frame target echo data according to the iteratively optimized robust autoencoder; and the horizontal and vertical pixel length thresholds in the connected components are set. If the horizontal and vertical lengths are less than the thresholds, they are considered as reverberation or interference and the connected components are deleted. Connected components that meet the threshold conditions are retained, and the target trajectory is extracted.

[0063] Specifically,

[0064] S401. The extracted target component D is binarized with a threshold of 0. Components greater than 0 are assigned a value of 1, and components less than 0 are assigned a value of 0.

[0065] In this embodiment, please refer to Figure 6 Binarization reduces the range of color values ​​from 256 to 2, which does improve computation speed, but also results in the loss of more information. Therefore, the specific processing method should be chosen based on the specific circumstances. In this embodiment, binarizing the target component D allows for a more intuitive view of the target echo trajectory and facilitates subsequent connected component analysis.

[0066] S402. Set the horizontal and vertical pixel length thresholds in the connected components. Components with horizontal and vertical lengths less than the thresholds are considered reverberation or interference and are deleted. Components that meet the threshold conditions are retained, and the target trajectory is extracted.

[0067] In this embodiment, connected component analysis is used to further highlight the trajectory of the target echo based on the correlation of the target location across multiple time frames. Please refer to [link to relevant documentation]. Figure 7 By using connected components, the intensity of the target echo trajectory in the graph can be enhanced.

[0068] It can be observed that in this embodiment, to address the problem of inaccurate reverberation estimation due to nonlinear distribution in actual underwater environments, a robust autoencoder is first used to nonlinearly estimate the background reverberation based on its low rank and the sparsity of the target echo. After removing the reverberation, the target echo components are estimated using a regularized adaptive estimation based on threshold sparsity compensation, and the target trajectory is extracted. Finally, connected component analysis is used to further highlight the target echo trajectory based on the correlation of target positions across multiple time frames. The method proposed in this invention can effectively remove reverberation and obtain a clear trajectory in moving target detection. When using artificial sonar to detect moving targets, it can effectively extract the target trajectory, contributing to the improvement of target detection range and the accuracy of target positioning and identification.

[0069] This invention also provides a reverberation removal device for moving target detection, which utilizes an autoencoder to project high-dimensional data into a low-dimensional space, extracts stable components through nonlinear estimation, and then removes the steady-state components from the overall data to obtain dynamic components. Furthermore, a sparse representation method is used to extract target echoes from the dynamic components. Finally, connected component analysis is used to extract clear target echo trajectories.

[0070] In this embodiment, the reverberation removal device for moving target detection includes:

[0071] The module includes an acquisition module, a construction module, a training module, and a prediction module.

[0072] The acquisition module is used to acquire multiple frames of target echo data and preprocess the multiple frames of target echo data.

[0073] The construction module is used to construct a robust autoencoder based on the preprocessed multi-frame target echo data;

[0074] The training module is used to extract the steady-state and dynamic components of the preprocessed multi-frame target echo data using the constructed robust autoencoder to iteratively optimize the constructed robust autoencoder.

[0075] The prediction module is used to remove reverberation from the acquired multi-frame target echo data based on the iteratively optimized robust autoencoder.

[0076] Optionally, the acquisition module is specifically used for:

[0077] Define a column vector to represent the acquired multi-frame target echo data, perform mean normalization on the multi-frame target echo data represented by the vector, and label the mean normalized multi-frame target echo data.

[0078] Optionally, this training module is specifically used for:

[0079] A robust autoencoder is used to extract the background reverberation matrix and target matrix from the preprocessed multi-frame target echo data. The components in the background reverberation matrix and target matrix represent the steady-state component and dynamic component of the i-th frame, respectively, and are initialized to zero matrices. The target echo matrix L to be optimized for the multi-frame target echo is set. D The parameters of the robust autoencoder are randomly initialized, the dynamic components are fixed, and the steady-state components are nonlinearly estimated using the robust autoencoder. This is achieved by minimizing... To calculate the steady-state components; and through the minimized The calculated steady-state components are used to calculate the dynamic components; and the minimized components are fixed. The calculated steady-state components are optimized using a proximal operator; and the convergence coefficient is calculated and iterated by optimizing the convergence coefficient until the robust autoencoder achieves convergence.

[0080] Optionally, the prediction module is specifically used for:

[0081] The target echo is extracted from the multi-frame target echo data according to the iteratively optimized robust autoencoder; and the horizontal and vertical pixel length thresholds in the connected components are set. If the horizontal and vertical lengths are less than the thresholds, they are considered as reverberation or interference and the connected components are deleted. Connected components that meet the threshold conditions are retained and the target trajectory is extracted.

[0082] Each unit module of the reverberation removal device in the moving target detection can execute the corresponding steps in the above method embodiment. Therefore, the details of each unit module will not be elaborated here. Please refer to the description of the corresponding steps above for details.

[0083] The present invention also provides a reverberation removal device for moving target detection, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described reverberation removal method for moving target detection.

[0084] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0085] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 52 can be used to store data used by the processor during operation.

[0086] The present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the above-described method embodiments.

[0087] It can be observed that, to address the issue of inaccurate reverberation estimation in nonlinear distributions within actual underwater environments, the above scheme first uses a robust autoencoder to nonlinearly estimate the background reverberation, taking advantage of its low rank and target echo sparsity. After reverberation removal, target echo components are estimated using a regularized adaptive estimation based on threshold sparsity compensation to extract the target trajectory. Finally, connected component analysis is used to further highlight the target echo trajectory based on the correlation of target positions across multiple time frames. The method proposed in this invention can effectively remove reverberation and obtain a clear trajectory in moving target detection. When using artificial sonar to detect moving targets, it can effectively extract the target trajectory, contributing to improved accuracy in target detection range, positioning, and identification.

[0088] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection of apparatuses or units, and may be electrical, mechanical, or other forms.

[0089] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0090] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0091] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A reverberation removal method in moving target detection, characterized in that, The method comprises the following steps: acquiring a plurality of target echo data, and preprocessing the plurality of target echo data; constructing a robust autoencoder based on the preprocessed plurality of target echo data; extracting steady-state components and dynamic components of the preprocessed plurality of target echo data by using the constructed robust autoencoder, and iteratively optimizing the constructed robust autoencoder; removing reverberation corresponding to the plurality of target echo data from the acquired plurality of target echo data according to the iteratively optimized robust autoencoder; the acquiring a plurality of target echo data, and preprocessing the plurality of target echo data comprises: defining a column vector to vectorize the acquired plurality of target echo data, and performing mean normalization on the vectorized plurality of target echo data, and marking the mean normalized plurality of target echo data; the extracting steady-state components and dynamic components of the preprocessed plurality of target echo data by using the constructed robust autoencoder, and iteratively optimizing the constructed robust autoencoder comprises: extracting a background reverberation matrix and a target matrix in the preprocessed plurality of target echo data by using the robust autoencoder, the components in the background reverberation matrix and the target matrix representing the steady-state components and the dynamic components of the i-th frame respectively, and initializing the background reverberation matrix and the target matrix as zero matrices; assigning the preprocessed plurality of target echo data to the background reverberation matrix, and randomly initializing the parameters of the robust autoencoder, fixing the dynamic components, performing nonlinear estimation on the steady-state components by using the robust autoencoder, calculating the steady-state components through the nonlinear estimation, calculating the dynamic components through the steady-state components calculated through the nonlinear estimation, fixing the steady-state components calculated through the nonlinear estimation, optimizing the dynamic components by using a proximal operator, calculating a convergence coefficient, and optimizing the convergence coefficient through iteration until the robust autoencoder meets the convergence; the removing reverberation corresponding to the plurality of target echo data from the acquired plurality of target echo data according to the iteratively optimized robust autoencoder comprises: extracting target echoes corresponding to the plurality of target echo data from the plurality of target echo data according to the iteratively optimized robust autoencoder; and setting a pixel length threshold in the horizontal direction and the vertical direction of a connected domain, regarding the horizontal length and the vertical length less than the threshold as reverberation or interference, deleting the connected domain, and retaining the connected domain meeting the threshold condition to extract a target trajectory.

2. A reverberation removing device in a moving target detection, characterized by, The method comprises the following steps: an acquisition module, a construction module, a training module, and a prediction module; the acquisition module is configured to acquire a plurality of target echo data, and preprocess the plurality of target echo data; the construction module is configured to construct a robust autoencoder based on the preprocessed plurality of target echo data; the training module is configured to extract steady-state components and dynamic components of the preprocessed plurality of target echo data by using the constructed robust autoencoder, and iteratively optimize the constructed robust autoencoder; The prediction module is configured to remove reverberation corresponding to the collected multi-frame target echo data according to the iteratively optimized robust autoencoder. The acquisition module is specifically configured to: define a column vector to vectorize the acquired multi-frame target echo data, perform mean normalization on the vectorized multi-frame target echo data, and mark the mean normalized multi-frame target echo data; The training module is specifically configured to: extract a background reverberation matrix and a target matrix from the preprocessed multi-frame target echo data using a robust autoencoder, wherein the components in the background reverberation matrix and the target matrix represent the steady-state component and the dynamic component of the i-th frame, respectively, and initialize them as zero matrices; assign the preprocessed multi-frame target echo data to the background reverberation matrix, randomly initialize the parameters of the robust autoencoder, fix the dynamic component, perform nonlinear estimation on the steady-state component using the robust autoencoder, and calculate the steady-state component through nonlinear estimation; calculate the dynamic component through the steady-state component calculated through nonlinear estimation; fix the steady-state component calculated through nonlinear estimation, optimize the dynamic component using a proximal operator, calculate a convergence coefficient, and optimize the convergence coefficient through iteration until the robust autoencoder satisfies the convergence; The prediction module is specifically configured to: extract target echoes corresponding to the multi-frame target echo data according to the iteratively optimized robust autoencoder, set a pixel length threshold in the horizontal and vertical directions of the connected domain, consider the horizontal length and the vertical length less than the threshold as reverberation or interference, delete the connected domain, and retain the connected domain meeting the threshold condition to extract the target trajectory.

3. A reverberation removing device in a moving target detection, characterized by, It includes: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the reverberation removal method in the moving target detection according to claim 1.

4. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to implement the reverberation removal method in the moving target detection according to claim 1. The computer program is executed by the processor to implement the reverberation removal method in the moving target detection according to claim 1.

Images