Underwater DOA estimation method based on residual Fourier global information fusion network

Abstract

The invention discloses an underwater DOA estimation method based on a residual Fourier global information fusion network. The method comprises the following steps: obtaining a target formation of an underwater array sensor; based on the target formation, generating a simulation training data set based on a sparse covariance matrix; constructing a residual Fourier module, and constructing a residual Fourier global information fusion network; based on a preset training strategy, inputting the simulation training data set into a residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network; processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; and determining the direction of arrival of the underwater target based on the estimated spatial spectrum.

Description

Technical Field

[0001] The invention relates to the technical field of marine target direction estimation, and in particular to an underwater DOA estimation method based on a residual Fourier global information fusion network. Background Art

[0002] Target direction estimation (DOA) is a research frontier and hotspot in the field of sonar detection. It plays a vital role in battlefield reconnaissance, underwater navigation, and marine development. DOA estimation mainly collects and processes signals through a multi-sensor array, extracts parameter information representing the target's direction, and estimates the target's incident angle.

[0003] Traditional methods are all based on model-driven methods. When the preset model differs significantly from the actual engineering environment, its DOA estimation performance decays rapidly. At the same time, it is also difficult to make the preset model close to the actual model in actual complex application scenarios. The deep learning method can establish a mapping relationship between the signal received by the hydrophone array and the target direction, so as to better capture the relationship between the signal inside and the target direction. Summary of the invention

[0004] In order to solve the above-mentioned technical problems existing in the prior art, the present invention provides an underwater DOA estimation method, device, equipment and storage medium based on a residual Fourier global information fusion network to solve the problem of insufficient resolution of underwater position estimation in the prior art.

[0005] In order to achieve the above object, the technical solution of the embodiment of the present invention is:

[0006] In a first aspect, an embodiment of the present invention provides an underwater DOA estimation method based on a residual Fourier global information fusion network, the method comprising:

[0007] Obtain target formation of underwater array sensors;

[0008] Based on the target formation, generating a simulation training data set based on a sparse covariance matrix;

[0009] Constructing a residual Fourier module, and based on the residual Fourier module, constructing a residual Fourier global information fusion network;

[0010] Based on a preset training strategy, the simulation training data set is input into the residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network;

[0011] Processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum;

[0012] Based on the estimated spatial spectrum, the incoming wave direction of the underwater target is determined.

[0013] In a second aspect, an embodiment of the present invention provides an underwater DOA estimation device based on a residual Fourier global information fusion network, the device comprising:

[0014] An acquisition module, used to obtain the target formation of the underwater array sensor;

[0015] A generation module, used for generating a simulation training data set based on a sparse covariance matrix based on the target formation;

[0016] A construction module is used to construct a residual Fourier module, and based on the residual Fourier module, a residual Fourier global information fusion network is constructed;

[0017] A training module, used for inputting the simulation training data set into the residual Fourier global information fusion network for training based on a preset training strategy to obtain a trained residual Fourier global information fusion network;

[0018] A processing module, used for processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum;

[0019] The determination module is used to determine the incoming wave direction of the underwater target based on the estimated spatial spectrum.

[0020] In some embodiments, the generating module is further used to generate a receiving signal of the underwater array sensor based on the target formation, x=As+N; wherein A is an array manifold vector; s is a simulated real sound source signal; N=σ 2 I is additive Gaussian white noise, σ 2 is the Gaussian white noise power;

[0021] Based on the received signal, a snapshot covariance matrix of the array is calculated, in,[·] H represents the conjugate transpose of the matrix; R represents the true covariance matrix; ΔR represents the difference between the true covariance matrix R and the snapshot covariance matrix The errors that exist between

[0022] The snapshot covariance matrix is ​​vectorized to obtain a snapshot covariance vector. Where vec(·) represents the matrix vectorization operation; Δy=vec(ΔR), which is the result of vectorization of the covariance matrix error ΔR; η=diag(η1,η2,…,η N ) is the signal power; [·;…;·] means vertically concatenating the matrices or vectors in the brackets. represents the array manifold at the mth angle; e m represents an M×1-dimensional column vector whose mth element is 1 and the rest are 0;

[0023] The snapshot covariance vector is rewritten to obtain a rewritten snapshot covariance vector, in, is the overall error;

[0024] Determining a target pseudospectral formula based on the rewritten snapshot covariance vector;

[0025] Based on the target pseudo-spectral formula, the simulation training data set is calculated.

[0026] In some embodiments, the residual Fourier module includes at least a main path, a residual branch and a Fourier branch; the residual Fourier module is constructed in the following manner:

[0027] The main path contains two layers of convolution kernels, and the two layers of convolution kernels are activated by a ReLU function to obtain a first output;

[0028] The residual branch performs identity mapping on the input data to obtain a second output;

[0029] The Fourier branch first passes through a layer of inverse Fourier transform, then passes through a 1×1 convolution with a ReLU function activated between the two layers of convolution kernels; and finally passes through a layer of inverse Fourier transform to obtain a third output;

[0030] The first output, the second output and the third output are added to obtain an output of the residual Fourier module.

[0031] In some embodiments, the residual Fourier global information fusion network includes a 25×1 convolution layer and three layers of the residual Fourier module; the main convolution kernels in the residual Fourier module are 15×1, 5×1, and 3×1, respectively, and the number of output channels are 6, 3, and 1, respectively.

[0032] In some embodiments, the apparatus further comprises: an output module, configured to increase the number of input channels from 2 to 12 by the 25×1 convolution layer to obtain local features;

[0033] Reducing the dimension and extracting the global features of the input through the residual Fourier module;

[0034] A spatial spectrum is output based on the local features and the global features.

[0035] In some embodiments, the training module is further used to divide the simulation training data set into a training set and a validation set using a certain ratio;

[0036] The training optimizer is determined to be Adam;

[0037] Use ReduceLROnPlateau method to adaptively adjust the learning rate;

[0038] The residual Fourier global information fusion network is trained using an early stopping training strategy, the number of early stopping rounds is set to 90, and if the validation set loss does not decrease in 90 consecutive rounds, the trained residual Fourier global information fusion network is obtained.

[0039] In some embodiments, the training module is further used to search the estimated spatial spectrum using a preset peak detection method, and determine the vector index corresponding to the maximum peak as the incoming wave direction.

[0040] An embodiment of the present invention provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned underwater DOA estimation method based on the residual Fourier global information fusion network when executing the executable instructions stored in the memory.

[0041] An embodiment of the present invention provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned underwater DOA estimation method based on the residual Fourier global information fusion network.

[0042] The underwater DOA estimation method based on the residual Fourier global information fusion network provided by the present invention obtains the target formation of the underwater array sensor; based on the target formation, generates a simulation training data set based on the sparse covariance matrix; constructs a residual Fourier module, and constructs a residual Fourier global information fusion network; based on a preset training strategy, inputs the simulation training data set into the residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network; processes the data of the target underwater array sensor, and inputs the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; based on the estimated spatial spectrum, determines the incoming wave direction of the underwater target. In this way, the present invention constructs a residual Fourier module by introducing Fourier transform operation in the residual module, which can more comprehensively extract high-frequency and low-frequency components, and then capture the global information of the signal; in the underwater target azimuth estimation, the residual Fourier global information fusion network constructed using the residual Fourier module can more effectively estimate the outgoing wave direction, improve the resolution, and achieve high-resolution DOA estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of an underwater DOA estimation method based on a residual Fourier global information fusion network provided by an embodiment of the present invention;

[0044] Figure 2 It is a flow chart of a high-resolution underwater DOA estimation method based on a residual Fourier global information fusion network provided by an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of a residual Fourier module provided in an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of a residual Fourier global information fusion network provided by an embodiment of the present invention;

[0047] Figure 5 is a peak search schematic diagram provided by an embodiment of the present invention;

[0048] Figure 6 It is an introduction diagram of the SWellEx-96 test provided by an embodiment of the present invention, wherein (a) is a position diagram of the test ship and each array, and (b) is a diagram of the actual geometric shape of the north horizontal array;

[0049] Figure 7 is a spatial spectrum at the 44th minute of the experiment provided by an embodiment of the present invention;

[0050] Figure 8 It is a time-direction history diagram provided by an embodiment of the present invention;

[0051] Fig. 9It is a schematic diagram of the composition structure of an underwater DOA estimation device based on a residual Fourier global information fusion network provided by an embodiment of the present invention;

[0052] Fig.10 It is a schematic diagram of the composition structure of an underwater DOA estimation device based on a residual Fourier global information fusion network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0054] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present invention have the same meaning as those commonly understood by those skilled in the art to which the embodiments of the present invention pertain. The terms used in the embodiments of the present invention are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0055] The embodiment of the present invention provides an underwater DOA estimation method based on a residual Fourier global information fusion network, see Figure 1 , Figure 1 FIG. 1 is a flow chart of an underwater DOA estimation method based on a residual Fourier global information fusion network provided by an embodiment of the present invention, which is combined with Figure 1 The steps shown are explained.

[0056] Step S110, obtaining the target formation of the underwater array sensor.

[0057] In some embodiments, the target formation refers to a known formation of underwater array sensors.

[0058] Step S120: generating a simulation training data set based on a sparse covariance matrix based on the target formation.

[0059] In some embodiments, according to the target formation, a series of simulation data containing underwater target signals are generated using an underwater sound wave propagation model and a noise model. By processing these data, a sparse covariance matrix is ​​obtained as an input feature of the neural network. The simulation training data set should cover a variety of possible underwater environments and target characteristics to ensure the generalization ability of the neural network.

[0060] In the present invention, a simulation training data set based on a sparse covariance matrix can be generated according to the known formation of the underwater array sensor.

[0061] Step S130, constructing a residual Fourier module, and based on the residual Fourier module, constructing a residual Fourier global information fusion network.

[0062] In some embodiments, the residual Fourier module is a neural network structure that combines residual learning and Fourier transform, which can efficiently extract frequency domain features from signals. On this basis, a residual Fourier global information fusion network is constructed, which realizes deep feature extraction and global information fusion of input signals by stacking multiple residual Fourier modules.

[0063] In the present invention, an improvement can be made on the basis of the original residual module, a fast Fourier transform branch can be introduced, a residual Fourier module can be constructed, and a residual Fourier global information fusion network can be designed based on the module.

[0064] Step S140: Based on a preset training strategy, the simulation training data set is input into the residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network.

[0065] In some embodiments, a suitable training strategy (such as gradient descent algorithm, learning rate adjustment strategy, etc.) is used to input the simulation training data set into the residual Fourier global information fusion network for training. Through multiple iterations of optimization, the network can accurately extract information related to the target wave direction from the input features.

[0066] In the present invention, based on the simulation training data set and the residual Fourier global information fusion network in the above steps, the residual Fourier global information fusion network can be trained by adaptively adjusting the learning rate, "early stopping" training and other training strategies to obtain a trained residual Fourier global information fusion network.

[0067] Step S150, processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum.

[0068] In some embodiments, the underwater array sensor data actually collected is preprocessed (such as denoising, filtering, etc.) and then input into the trained residual Fourier global information fusion network. The network output is an estimated spatial spectrum, which reflects the intensity distribution of underwater target signals in different directions.

[0069] Step S160: determining the incoming wave direction of the underwater target based on the estimated spatial spectrum.

[0070] In the present invention, based on the trained residual Fourier global information fusion network, the data of the real underwater array sensor is processed to obtain the input data of the network, which is input into the trained network to obtain a one-dimensional spatial spectrum, and then the orientation of the underwater target is estimated.

[0071] The underwater DOA estimation method based on the residual Fourier global information fusion network provided by the present invention obtains the target formation of the underwater array sensor; based on the target formation, generates a simulation training data set based on the sparse covariance matrix; constructs a residual Fourier module, and constructs a residual Fourier global information fusion network; based on a preset training strategy, inputs the simulation training data set into the residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network; processes the data of the target underwater array sensor, and inputs the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; based on the estimated spatial spectrum, determines the incoming wave direction of the underwater target. In this way, the present invention constructs a residual Fourier module by introducing Fourier transform operation in the residual module, which can more comprehensively extract high-frequency and low-frequency components, and then capture the global information of the signal; in the underwater target azimuth estimation, the residual Fourier global information fusion network constructed using the residual Fourier module can more effectively estimate the outgoing wave direction, improve the resolution, and achieve high-resolution DOA estimation.

[0072] In some embodiments, the above step S120 can be implemented by following the steps S121 to S126:

[0073] Step S121, based on the target formation, generate the receiving signal of the underwater array sensor, x = As + N; wherein A is the array manifold vector; s is the simulated real sound source signal; N = σ 2 I is additive Gaussian white noise, σ 2 is the Gaussian white noise power.

[0074] Step S122, based on the received signal, calculating the snapshot covariance matrix of the array, in,[·] H represents the conjugate transpose of the matrix; R represents the true covariance matrix; ΔR represents the difference between the true covariance matrix R and the snapshot covariance matrix There are errors between.

[0075] Step S123, performing vectorization operation on the snapshot covariance matrix to obtain a snapshot covariance vector. Where vec(·) represents the matrix vectorization operation; Δy=vec(ΔR), which is the result of vectorization of the covariance matrix error ΔR; η=diag(η1,η2,…,η N ) is the signal power; [·;…;·] means vertically concatenating the matrices or vectors in the brackets. represents the array manifold at the mth angle; e m Represents an M×1-dimensional column vector whose mth element is 1 and the rest of the elements are 0.

[0076] Step S124, rewrite the snapshot covariance vector to obtain a rewritten snapshot covariance vector, in, is the overall error.

[0077] Step S125 : determining a target pseudo-spectral formula based on the rewritten snapshot covariance vector.

[0078] Step S126, calculating the simulation training data set based on the target pseudo-spectrum formula.

[0079] In some embodiments, the residual Fourier module includes at least a main path, a residual branch and a Fourier branch; the residual Fourier module is constructed in the following manner:

[0080] First, the main path contains two layers of convolution kernels, and the two layers of convolution kernels are activated by a ReLU function to obtain a first output.

[0081] Secondly, the residual branch performs identity mapping on the input data to obtain a second output.

[0082] Again, the Fourier branch first passes through a layer of inverse Fourier transform, and then passes through a 1×1 convolution with a ReLU function activated between the two layers of convolution kernels; finally, it passes through a layer of inverse Fourier transform to obtain the third output.

[0083] Finally, the first output, the second output and the third output are added to obtain the output of the residual Fourier module.

[0084] In some embodiments, the residual Fourier global information fusion network includes a 25×1 convolution layer and three layers of the residual Fourier module; the main convolution kernels in the residual Fourier module are 15×1, 5×1, and 3×1, respectively, and the number of output channels are 6, 3, and 1, respectively.

[0085] In some embodiments, the method further comprises: increasing the number of input channels from 2 to 12 by the 25×1 convolution layer to obtain local features;

[0086] Reducing the dimension and extracting the global features of the input through the residual Fourier module;

[0087] A spatial spectrum is output based on the local features and the global features.

[0088] In some embodiments, the above step S140 can be implemented by following the steps S141 to S144:

[0089] Step S141, dividing the simulation training data set into a training set and a validation set in a certain ratio.

[0090] Step S142: The training optimizer is determined to be Adam.

[0091] Step S143, using the ReduceLROnPlateau method to adaptively adjust the learning rate.

[0092] Step S144, using the early stopping training strategy to train the residual Fourier global information fusion network, setting the number of early stopping rounds to 90, and the validation set loss does not decrease in 90 consecutive rounds, then the trained residual Fourier global information fusion network is obtained.

[0093] In some embodiments, the above step S160 may be implemented in the following manner: searching the estimated spatial spectrum using a preset peak detection method, and determining the vector index corresponding to the maximum peak as the incoming wave direction.

[0094] like Figure 2 As shown, the present invention discloses a high-resolution underwater DOA estimation method based on a residual Fourier global information fusion network, which mainly includes the following four steps, which will be further explained in conjunction with specific embodiments and implementation methods. The DOA estimation method includes the following steps:

[0095] Step S1: Setting the angle scanning range The number of incoming wave signals is 2, and the angle difference between the two incoming wave signals is Δθ=98, 99, 100, ..., 233. At each Δθ, the angles of the two incoming wave signals θ=[θ1, θ2] are scanned from [0, 1, ..., 360] at the same time, with a step size of 1. The specific scanning method is: in the angle scanning range Internal fixation θ1, calculation θ2 = θ1 + Δθ (need to ensure ), we get 26588 pairs of angle samples, so there are 26588 training data in total. At the same time, we construct a 361-dimensional vector as a label, which is set to 1 at the angle of the incoming signal and 0 at the rest.

[0096] Since the array is placed on the seabed, the pitch angle Direction vector at pitch angle Direction angle θ i The following expression is The signal-to-noise ratio is randomly selected from SNR = [-10dB, 0dB], and the signal frequency is f c=79Hz, the number of snapshots is Nsnap=256, the number of array elements is M=27, and the given real array geometry p i Construct the array manifold, the formula is as follows:

[0097]

[0098] Among them, the wave number Where c is the underwater sound speed, which is roughly taken as 1500m / s.

[0099] The snapshot simulation signal s is generated by random numbers, and the signal is noised at a given signal-to-noise ratio to obtain the simulated array observation signal:

[0100] x=As+N,

[0101] Where N = σ 2 I, represents the Gaussian white noise power, P s is the signal power.

[0102] Calculate the snapshot covariance matrix. At the same time, it should be noted that the snapshot covariance matrix needs to be normalized in actual application:

[0103]

[0104] The snapshot covariance matrix is ​​vectorized to obtain the snapshot covariance vector:

[0105]

[0106] Where vec(·) represents the matrix vectorization operation; Δy=vec(ΔR), which is the result of vectorization of the covariance matrix error ΔR; η=diag(η1,η2,…,η N ) is the signal power; [·;…;·] means vertically concatenating the matrices or vectors in the brackets; represents the array manifold at the mth angle; e m Represents an M×1-dimensional column vector whose mth element is 1 and the rest of the elements are 0.

[0107] Rewrite the snapshot covariance vector:

[0108]

[0109] in, is the overall error.

[0110] Calculate the pseudo-spectrum as input data to train the neural network:

[0111]

[0112] Step S2: Construct the residual Fourier module, such as Figure 3 As shown. This module has a main path and two branches, which are called residual branch and Fourier branch. In the main path, there are two layers of one-dimensional convolution with large convolution kernels, and the two layers of convolution are activated by ReLU function. The residual branch directly performs identity mapping on the input. In the Fourier branch, it first passes through a layer of FFT transformation, then passes through two layers of 1×1 convolution with ReLU activation in the middle, and finally passes through a layer of inverse FFT transformation. Finally, the outputs of these three paths are added together to get the output of the module. The constructed residual Fourier module is used to build a residual Fourier global information fusion network. The network is built by a 25×1 convolution layer and three residual Fourier modules. First, a convolution layer is used to increase the dimension of the input with 2 channels to 12 channels to obtain local features. Then, the residual Fourier module is used to gradually reduce the dimension and extract the global features of the input. The convolution kernels of the main path in the three-layer residual Fourier module are 15×1, 5×1, and 3×1, respectively, and the number of output channels are 6, 3, and 1, respectively. Finally, the network uses global and local features to output the spatial spectrum, such as Figure 4 shown.

[0113] Step S3: Divide the training data set into a training set and a validation set in a ratio of 8:2. The MSE loss function is selected during training. The MSE calculation expression for a batch is as follows:

[0114]

[0115] Among them, B represents the number of samples in a batch, η i represents the real spatial spectrum under the i-th sampling, η i Represents the predicted spatial spectrum under the i-th sampling; during the training process, one epoch usually contains multiple batches, and the MSE calculation expression under one epoch is as follows:

[0116]

[0117] Among them, E represents the number of batches contained in a round, Represents the MSE for the e-th batch.

[0118] Adam is selected as the training optimizer; the learning rate is adaptively adjusted using the ReduceLROnPlateau method; the residual Fourier global information fusion network constructed is trained using the "early stopping" training strategy, and the number of "early stopping" rounds is set to 90, indicating that if the training does not reach the preset maximum number of rounds, and the validation set loss does not decrease in 90 consecutive rounds, the neural network can be considered to have been trained, and the training is stopped. At the same time, the model corresponding to the minimum loss of the validation set is selected as the optimal model for use. The output of the residual Fourier global information fusion network is the estimated spatial spectrum.

[0119] Step S4: Use the 65-minute data of the north horizontal array in the S59 event as the network input, and the sampling frequency fs = 3267.8 Hz. When processing the data, set the FFT points to 4096, the overlap rate to 50%, and the frequency point to fc = 79 Hz. When the number of snapshots is Nsnap = 10, construct the covariance matrix and calculate the input pseudo-spectrum of the neural network:

[0120]

[0121] The pseudo-spectral data is input into the trained network to obtain the estimated one-dimensional spatial spectrum, such as Figure 7 The one-dimensional spatial spectrum estimated by the network within 65 minutes is spliced ​​by row (time dimension) to obtain the time orientation history diagram, as shown in Figure 8 For the one-dimensional spatial spectrum in each time dimension, the angle of the incoming wave signal in each time dimension is obtained by setting the number of incoming wave signals in the peak search method.

[0122] Fig. 9 : is a schematic diagram of the composition structure of an underwater DOA estimation device based on a residual Fourier global information fusion network provided by an embodiment of the present invention, such as Fig. 9 As shown, an underwater DOA estimation device 900 based on a residual Fourier global information fusion network includes: an acquisition module 901, used to obtain a target formation of an underwater array sensor; a generation module 902, used to generate a simulation training data set based on a sparse covariance matrix based on the target formation; a construction module 903, used to construct a residual Fourier module, and based on the residual Fourier module, construct a residual Fourier global information fusion network; a training module 904, used to input the simulation training data set into the residual Fourier global information fusion network for training based on a preset training strategy, and obtain a trained residual Fourier global information fusion network; a processing module 905, used to process the data of the target underwater array sensor, and input the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; a determination module 906, used to determine the incoming wave direction of the underwater target based on the estimated spatial spectrum.

[0123] In some embodiments, the generating module is further used to generate a receiving signal of the underwater array sensor based on the target formation, x=AS+N; wherein A is the array manifold vector; s is the simulated real sound source signal; N=σ 2 I is additive Gaussian white noise, σ 2 is the Gaussian white noise power; based on the received signal, the snapshot covariance matrix of the array is calculated, in,[·] H represents the conjugate transpose of the matrix; R represents the true covariance matrix; ΔR represents the difference between the true covariance matrix R and the snapshot covariance matrix The snapshot covariance matrix is ​​vectorized to obtain a snapshot covariance vector, Where vec(·) represents the matrix vectorization operation; Δy=vec(ΔR), which is the result of vectorization of the covariance matrix error ΔR; η=diag(η1,η2,…,η N ) is the signal power; [·;…;·] means vertically concatenating the matrices or vectors in the brackets. represents the array manifold at the mth angle; e m represents an M×1-dimensional column vector whose mth element is 1 and the remaining elements are 0; rewriting the snapshot covariance vector to obtain a rewritten snapshot covariance vector, in, is the overall error; based on the rewritten snapshot covariance vector, determining the target pseudo-spectral formula; based on the target pseudo-spectral formula, calculating the simulation training data set.

[0124] In some embodiments, the residual Fourier module includes at least a main path, a residual branch and a Fourier branch; the residual Fourier module is constructed in the following manner: the main path contains two layers of convolution kernels, and the two layers of convolution kernels are activated by a ReLU function to obtain a first output; the residual branch performs identity mapping on the input data to obtain a second output; the Fourier branch first passes through a layer of inverse Fourier transform, and then passes through a 1×1 convolution with a ReLU function activated between the two layers of convolution kernels; finally, a layer of inverse Fourier transform is passed through to obtain a third output; the first output, the second output and the third output are added to obtain the output of the residual Fourier module.

[0125] In some embodiments, the residual Fourier global information fusion network includes a 25×1 convolution layer and three layers of the residual Fourier module; the main convolution kernels in the residual Fourier module are 15×1, 5×1, and 3×1, respectively, and the number of output channels are 6, 3, and 1, respectively.

[0126] In some embodiments, the device also includes: an output module, used to increase the number of input channels from 2 to 12 by the 25×1 convolutional layer to obtain local features; reduce the dimension and extract the global features of the input through the residual Fourier module; and output a spatial spectrum based on the local features and the global features.

[0127] In some embodiments, the training module is also used to divide the simulation training data set into a training set and a validation set in a certain ratio; the training optimizer is determined to be Adam; the learning rate is adaptively adjusted using the ReduceLROnPlat eau method; the residual Fourier global information fusion network is trained using an early stopping training strategy, the number of early stopping rounds is set to 90, and the validation set loss does not decrease in 90 consecutive rounds, then the trained residual Fourier global information fusion network is obtained.

[0128] In some embodiments, the training module is further used to search the estimated spatial spectrum using a preset peak detection method, and determine the vector index corresponding to the maximum peak as the incoming wave direction.

[0129] In some embodiments, it should be noted that the description of the device of the embodiment of the present invention is similar to the description of the method embodiment described above, and has similar beneficial effects as the same method embodiment, so it will not be repeated. For technical details not disclosed in the embodiment of the device, please refer to the description of the method embodiment of the present invention for understanding.

[0130] It should be noted that, in the embodiment of the present invention, if the above-mentioned underwater DOA estimation method based on the residual Fourier global information fusion network is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for a terminal to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), disk or optical disk and other media that can store program codes. In this way, the embodiment of the present invention is not limited to any specific combination of hardware and software.

[0131] Correspondingly, an embodiment of the present invention provides an underwater DOA estimation device based on a residual Fourier global information fusion network. Fig.10 : is a schematic diagram of the composition structure of an underwater DOA estimation device based on a residual Fourier global information fusion network provided by an embodiment of the present invention, such as Fig.10 As shown, the underwater DOA estimation device 1000 based on the residual Fourier global information fusion network includes at least: a processor 1001 and a computer-readable storage medium 1002 configured to store executable instructions, wherein the processor 1001 generally controls the overall operation of the underwater DOA estimation device based on the residual Fourier global information fusion network. The computer-readable storage medium 1002 is configured to store instructions and applications executable by the processor 1001, and can also cache data to be processed or processed by each module in the processor 1001 and the underwater DOA estimation device 1000 based on the residual Fourier global information fusion network, which can be implemented by flash memory (FLASH) or random access memory (RAM, Random Access Memory).

[0132] An embodiment of the present invention provides a storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will be caused to execute the method provided by the embodiment of the present invention, for example, Figure 1 The method shown.

[0133] In some embodiments, the storage medium can be a computer-readable storage medium, for example, a ferroelectric random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a compact disk read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above memories.

[0134] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0135] As an example, executable instructions may, but need not necessarily, correspond to a file in a file system, may be stored as part of a file storing other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions). As an example, executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed at multiple locations and interconnected by a communication network.

[0136] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

[0137] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0138] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, an element defined by the statement "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed.

[0139] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An underwater DOA estimation method based on residual Fourier global information fusion network, characterized in that: The method comprises: Obtain target formation of underwater array sensors; Based on the target formation, generating a simulation training data set based on a sparse covariance matrix; Constructing a residual Fourier module, and based on the residual Fourier module, constructing a residual Fourier global information fusion network; Based on a preset training strategy, the simulation training data set is input into the residual Fourier global information fusion network for training to obtain a trained residual Fourier global information fusion network; Processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; Based on the estimated spatial spectrum, the incoming wave direction of the underwater target is determined.

2. The method according to claim 1, characterized in that The step of generating a simulation training data set based on a sparse covariance matrix based on the target formation includes: Based on the target formation, the receiving signal of the underwater array sensor is generated, x=As+N; wherein A is the array manifold vector; s is the simulated real sound source signal; N=σ 2 I is additive Gaussian white noise, σ 2 is the Gaussian white noise power; Based on the received signal, a snapshot covariance matrix of the array is calculated, in,[·] H represents the conjugate transpose of the matrix; R represents the true covariance matrix; ΔR represents the difference between the true covariance matrix R and the snapshot covariance matrix The errors that exist between The snapshot covariance matrix is ​​vectorized to obtain a snapshot covariance vector. Wherein, vec(·) represents the matrix vectorization operation; Δy=vec(ΔR), which is the result obtained by vectorizing the covariance matrix error ΔR; η=diag(η1, η2, ..., η N ) is the signal power; [·;…;·] means vertically concatenating the matrices or vectors in the brackets. represents the array manifold at the mth angle; e m represents an M×1-dimensional column vector whose mth element is 1 and the rest are 0; The snapshot covariance vector is rewritten to obtain a rewritten snapshot covariance vector, in, is the overall error; Determining a target pseudospectral formula based on the rewritten snapshot covariance vector; Based on the target pseudo-spectral formula, the simulation training data set is calculated.

3. The method according to claim 1, characterized in that: The residual Fourier module at least includes a main path, a residual branch and a Fourier branch; The residual Fourier module is constructed in the following way: The main path contains two layers of convolution kernels, and the two layers of convolution kernels are activated by a ReLU function to obtain a first output; The residual branch performs identity mapping on the input data to obtain a second output; The Fourier branch first passes through a layer of inverse Fourier transform, then passes through a 1×1 convolution with a ReLU function activated between the two layers of convolution kernels; and finally passes through a layer of inverse Fourier transform to obtain a third output; The first output, the second output and the third output are added to obtain an output of the residual Fourier module.

4. The method according to claim 3, characterized in that The residual Fourier global information fusion network includes a 25×1 convolution layer and three residual Fourier modules; the main convolution kernels in the residual Fourier modules are 15×1, 5×1, and 3×1, respectively, and the number of output channels are 6, 3, and 1, respectively.

5. The method according to claim 4, characterized in that The method further comprises: The 25×1 convolutional layer increases the number of input channels from 2 to 12 to obtain local features; Reducing the dimension and extracting the global features of the input through the residual Fourier module; A spatial spectrum is output based on the local features and the global features.

6. The method according to claim 1, characterized in that Based on the preset training strategy, the simulation training data set is input into the residual Fourier global information fusion network for training to obtain the trained residual Fourier global information fusion network, including: Dividing the simulation training data set into a training set and a validation set using a certain ratio; The training optimizer is determined to be Adam; Use ReduceLROnPlateau method to adaptively adjust the learning rate; The residual Fourier global information fusion network is trained using an early stopping training strategy, the number of early stopping rounds is set to 90, and if the validation set loss does not decrease in 90 consecutive rounds, the trained residual Fourier global information fusion network is obtained.

7. The method according to claim 1, characterized in that The step of determining the incoming wave direction of the underwater target based on the estimated spatial spectrum comprises: The estimated spatial spectrum is searched using a preset peak detection method, and a vector index corresponding to a maximum peak is determined as the incoming wave direction.

8. An underwater DOA estimation device based on residual Fourier global information fusion network, characterized in that: The device comprises: An acquisition module, used to obtain the target formation of the underwater array sensor; A generation module, used for generating a simulation training data set based on a sparse covariance matrix based on the target formation; A construction module is used to construct a residual Fourier module, and based on the residual Fourier module, a residual Fourier global information fusion network is constructed; A training module, used for inputting the simulation training data set into the residual Fourier global information fusion network for training based on a preset training strategy to obtain a trained residual Fourier global information fusion network; A processing module, used for processing the data of the target underwater array sensor, and inputting the processed data into the trained residual Fourier global information fusion network to obtain an estimated spatial spectrum; The determination module is used to determine the incoming wave direction of the underwater target based on the estimated spatial spectrum.

9. An electronic device, characterized in that: include: A memory for storing executable instructions; A processor, used to implement the underwater DOA estimation method based on residual Fourier global information fusion network as described in any one of claims 1 to 7 when executing the executable instructions stored in the memory.

10. A computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the underwater DOA estimation method based on the residual Fourier global information fusion network as described in any one of claims 1 to 7.

Images