A method and related device for reducing a ship shaft frequency electric field signal

Abstract

A ship shaft frequency electric field signal denoising method and related device, comprising: obtaining a clean shaft frequency electric field signal; obtaining a noisy shaft frequency electric field signal with different signal-to-noise ratios through the obtained clean shaft frequency electric field signal; building a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network BRNN; inputting training samples in a data set into the shaft frequency electric field signal denoising network model; optimizing the training samples to obtain a noise prediction value at the current time, and obtaining a mean square error loss function of the noise prediction value and the actual noise at the current time through a loss function; repeatedly training to obtain a trained BRNN denoising network model, and inputting the noisy shaft frequency electric field signal into the trained BRNN denoising network model for denoising. The method can effectively solve the problem that the ship shaft frequency electric field signal is submerged under low signal-to-noise ratio sea conditions, improve the signal-to-noise ratio while retaining the frequency domain characteristics, and help subsequent signal detection.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of ship underwater electric field signal processing, and particularly relates to a ship shaft frequency electric field signal denoising method and a related device. BACKGROUND

[0002] The ship shaft frequency electric field signal has obvious characteristics and a characteristic frequency band, and can be used for target detection. However, complex sea conditions constitute a challenge to the detection of the ship shaft frequency electric field. It has always been a problem for ship underwater electric field detection to detect weak ship shaft frequency electric field signals from complex marine environmental noise. Only by reducing the noise components in the measured signal as much as possible can the shaft frequency electric field signal be restored to the maximum extent. The current shaft frequency electric field signal denoising method is mainly based on traditional signal processing methods, including Fourier transform, wavelet transform, empirical mode decomposition and sparse decomposition. The core of these methods is signal filtering, which finds the noise components and then filters them out. However, the denoising effect of this kind of method is limited, it is difficult to realize the detection of low signal-to-noise ratio signals, and the parameter selection is mainly based on prior knowledge and artificial experience. Therefore, it is of great significance to carry out research on the shaft frequency electric field signal denoising method under low signal-to-noise ratio. SUMMARY

[0003] The purpose of the application is to solve the problem that the shaft frequency electric field signal is submerged under low signal-to-noise ratio sea conditions and is difficult to detect, and to provide a ship shaft frequency electric field signal denoising method and a related device.

[0004] To achieve the above purpose, the application adopts the following technical scheme:

[0005] In a first aspect, the application provides a ship shaft frequency electric field signal denoising method, comprising:

[0006] Collecting the ship shaft frequency electric field signal and obtaining the clean shaft frequency electric field signal after preprocessing;

[0007] Linearly superimposing the obtained clean shaft frequency electric field signal with a section of the long-time noise signal of the measured environment to obtain the noisy shaft frequency electric field signal with different signal-to-noise ratios;

[0008] Building a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network BRNN;

[0009] Taking the noisy shaft frequency electric field signal with different signal-to-noise ratios as the data set for subsequent model training and testing, inputting the training samples in the data set into the shaft frequency electric field signal denoising network model;

[0010] Optimizing the training samples to obtain the noise prediction value at the current time, and obtaining the mean square error loss function of the noise prediction value and the actual noise at the current time through the loss function;

[0011] The trained BRNN denoising network model is obtained through repeated training, and the noisy shaft frequency electric field signal is input into the trained BRNN denoising network model for denoising.

[0012] Optionally, the ship shaft frequency electric field signal is a measured shaft frequency electric field signal of different distances and characteristic frequencies.

[0013] Optionally, the obtained clean shaft frequency electric field signal is linearly superimposed with a section of the measured long-time noise signal of the environment, that is, the long-time noise data is stepwise inserted starting from an arbitrary time point, to obtain a noisy shaft frequency electric field signal with different signal-to-noise ratios, which is used as a data set for subsequent model training and testing.

[0014] Optionally, in the shaft frequency electric field signal denoising network model based on the bidirectional recurrent neural network (BRNN), the BRNN denoising network structure is composed of an RNN layer and an FC full connection layer.

[0015] The propagation between each neuron in the RNN layer is linearized using a tanh activation function, and the output layer is linearized using a sigma activation function; the layer mainly includes two parts of forward propagation and backward propagation, and the calculation process is as follows:

[0016] h t =tanh(Ux t +Wh t-1 +b)

[0017] h′ t =tanh(U′x t +W′h t+1 +b′)

[0018] In the formula, x t is the input at the t th time; h t is the forward propagation value at the t th time, h′ t is the backward propagation value at the t th time; U and U′ are state-state weights from the input layer x t to the hidden layer h t ; W and W′ are state-input weights from the previous time hidden layer h t-1 to the current time hidden layer h t ; b and b′ are bias vectors.

[0019] The forward propagation value h t and the backward propagation value h′ t jointly determine the output value o t of the RNN layer, and the activation function is sigma, and the calculation formula is as follows:

[0020] o t =sigma(Vh t +V′h′t +c)

[0021] In the formula: V, V' are hidden layers h t to the weight coefficients of the output layer o t ; c is a bias vector;

[0022] The number of neural units and the output dimension of the FC full connection layer are the same as the output dimension of the RNN layer.

[0023] Optionally, when the training sample in the data set is input to the input layer of the built BRNN denoising network model, the original signal one-dimensional structure is converted into a two-dimensional structure, and the pure noise signal is taken as a label.

[0024] Optionally, the definition of the mean square error loss function is as follows:

[0025]

[0026] In the formula: n is the signal length.

[0027] Optionally, the training sample is forward propagated by using Adam optimization to obtain the noise prediction value at the current time; until the preset accuracy is met or the preset training step number is reached, the training is stopped, and the trained BRNN denoising network model is obtained.

[0028] In a second aspect, the present application provides a ship shaft frequency electric field signal denoising system, comprising:

[0029] A data acquisition module is configured to acquire a ship shaft frequency electric field signal, and obtain a clean shaft frequency electric field signal after preprocessing; linearly superimposes a segment obtained from the clean shaft frequency electric field signal and a long-time noise signal of a measured environment to obtain a noisy shaft frequency electric field signal with different signal-to-noise ratios.

[0030] A model building module is configured to build a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network (BRNN).

[0031] A training module is configured to take the noisy shaft frequency electric field signal with different signal-to-noise ratios as a data set for subsequent model training and testing, input a training sample in the data set into the shaft frequency electric field signal denoising network model, optimize the training sample, obtain a noise prediction value at the current time, and obtain a mean square error loss function of the noise prediction value and an actual noise at the current time through a loss function.

[0032] A denoising output module is configured to repeatedly train the BRNN denoising network model to obtain a trained BRNN denoising network model, and input the noisy shaft frequency electric field signal into the trained BRNN denoising network model for denoising.

[0033] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the ship shaft frequency electric field signal denoising method when executing the computer program.

[0034] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, wherein the computer program implements the steps of the ship shaft frequency electric field signal denoising method when executed by a processor.

[0035] Compared with the prior art, the present application has the following technical effects:

[0036] The ship shaft frequency electric field signal denoising method based on the bidirectional recurrent neural network has the following advantages: the ocean environment electric field noise is additive noise, and the long-term statistical characteristics thereof approximately follow a Gaussian distribution rule; and the shaft frequency electric field characteristics (fundamental frequency, harmonic frequency) of different types of ships at different detection distances are different. Therefore, the BRNN neural network learns the ocean environment electric field noise characteristics having certain rules, and since the BRNN network simultaneously considers the information of the past and the future at the current time, the noise statistical characteristics can be more comprehensively learned, and the noisy shaft frequency electric field signal can be removed to obtain a clean shaft frequency electric field signal. The method reduces the dependence on prior knowledge and artificial experience, can greatly improve the signal-to-noise ratio, and retains the frequency domain characteristics of the shaft frequency electric field signal, and has good adaptability in complex sea conditions with low signal-to-noise ratio, which is helpful for subsequent further signal detection. The method can effectively solve the problem that the ship shaft frequency electric field signal is submerged in low signal-to-noise ratio sea conditions, improve the signal-to-noise ratio while retaining the frequency domain characteristics, and is helpful for subsequent signal detection. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 Flowchart of the ship shaft frequency electric field signal denoising method based on BRNN;

[0038] Figure 2 Ship shaft frequency electric field denoising network model based on BRNN;

[0039] Figure 3 Fig. 3a-c are time domain and frequency domain comparison diagrams of the shaft frequency electric field signal before and after denoising with an initial signal-to-noise ratio of -20.43 dB and a shaft frequency of 6 Hz.

[0040] Figure 4 Fig. 3a-c are time domain and frequency domain comparison diagrams of the shaft frequency electric field signal before and after denoising with an initial signal-to-noise ratio of -20.43 dB and a shaft frequency of 6 Hz.

[0041] Figure 5 Fig. 3a-c are time domain and frequency domain comparison diagrams of the shaft frequency electric field signal before and after denoising with an initial signal-to-noise ratio of -20.43 dB and a shaft frequency of 6 Hz. DETAILED DESCRIPTION

[0042] The application will be further described in detail below with reference to the accompanying drawings. The embodiments are implemented on the premise of the technical solutions of the application, and detailed implementation modes are given, but the protection scope of the application is not limited to the following embodiments.

[0043] A ship shaft frequency electric field signal denoising method comprises:

[0044] Step one, pre-process the measured shaft frequency electric field signals of different distances and characteristic frequencies to obtain clean shaft frequency electric field signals.

[0045] Step two, linearly superimpose the obtained clean shaft frequency electric field signals with a section of the long-time noise signal of the measured environment, that is, stepwise insert a time point to the long-time noise data to obtain noisy shaft frequency electric field signals of different signal-to-noise ratios as the data set for subsequent model training and testing.

[0046] Step three, build a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network (BRNN) and configure network model parameters. The BRNN denoising network structure is composed of an RNN layer and an FC full connection layer.

[0047] (1) The propagation between each neuron in the RNN layer uses a tanh activation function to linearize, and the output layer uses a sigma activation function to linearize. This layer mainly includes two parts of forward propagation and backward propagation, and the calculation process is as follows:

[0048] h t =tanh(Ux t +Wh t-1 +b)

[0049] h′ t =tanh(U′x t +W′h t+1 +b′)

[0050] In the formula: x t is the input at the t-th moment; h t is the forward propagation value at the t-th moment, h′ t is the backward propagation value at the t-th moment; U and U′ are state-state weights from the input layer x t to the hidden layer h t ; W and W′ are state-input weights from the previous moment hidden layer h t-1 to the current moment hidden layer h t ; b and b′ are bias vectors.

[0051] The forward propagation value h t and the backward propagation value h′ tThe RNN layer jointly determines an output value o t , and the activation function is sigma, whose calculation formula is as follows:

[0052] o t = sigma (Vh t + V'h' t + c)

[0053] In the formula, V and V' are weight coefficients of the hidden layer h t to the output layer o t ; and c is a bias vector.

[0054] (2) The number of neural units and the output dimension of the FC full connection layer are the same as the output dimension of the RNN layer.

[0055] Step four, input the training sample in the data set into the input layer of the constructed BRNN denoising network model. Since the number of sampling points is large, the original one-dimensional structure of the signal is converted into a two-dimensional structure, and the pure noise signal is taken as a label.

[0056] Step five, the training sample is forward propagated by using Adam optimization, and the noise prediction value y i at the current time is obtained.

[0057] Step six, the mean square error (MSE) is used as a loss function, and the MSE_loss value between the actual noise y i at the current time is obtained. The definition of the mean square error loss function is as follows:

[0058]

[0059] In the formula, n is the signal length.

[0060] Step seven, the network parameters are updated through back propagation.

[0061] Step eight, steps five to seven are repeated until the preset accuracy is met or the preset training step number is reached, and the trained BRNN denoising network model is obtained.

[0062] Step nine, input the noisy axis frequency electric field signal into the trained BRNN denoising network model, remove the additive noise in the noisy axis frequency electric field signal, and obtain the denoised axis frequency electric field signal.

[0063] As shown in Figure 1 , the ship axis frequency electric field signal denoising method based on the bidirectional recurrent neural network provided in the embodiment includes the following steps:

[0064] Step 1: Preprocess the measured noisy shaft frequency electric field signals at different distances and characteristic frequencies to obtain clean shaft frequency electric field signals. The ship's speed is 15 knots, water depth is 50 meters, draft is 18 meters, positive lateral distance ranges from 100 meters to 1000 meters, the fundamental frequency range of the shaft frequency signal is 1 Hz to 7 Hz, and the sampling rate is 50 Hz.

[0065] Step two involves linearly superimposing the obtained clean axis-frequency electric field signal with a segment of long-term noise signal from the measured environment. This is achieved by arbitrarily selecting a time point and progressively inserting the signal into the long-term noise data to obtain noisy axis-frequency electric field signals with different signal-to-noise ratios. This dataset serves as the training and testing dataset for the subsequent model. The dataset contains noisy axis-frequency electric field data and pure noise data, with the noise data serving as the label. It includes 3200 training samples and 800 test samples. To ensure data reliability, there is no overlap or duplication between the training and test sets.

[0066] Step 3: Construct a noise reduction network model for axis frequency electric field signals based on a bidirectional recurrent neural network (BRNN), such as... Figure 2 As shown, the network model parameters were configured, with the training steps set to 100 epochs, the learning rate set to 0.001, and each training batch containing 128 samples. The BRNN denoising network structure consists of two parts: an RNN layer and a fully connected (FC) layer.

[0067] (1) RNN Layer: The RNN layer adopts a bidirectional RNN neural network structure, which can not only extract the information relationship between a certain time point and previous time points, but also obtain the relationship with information about future time points. The RNN input layer dimension is 400, that is, the sequence length is 400, the hidden layer dimension is also 400, and the output layer has 1 layer. The propagation between each neuron uses the tanh activation function to linearize, and the output layer uses the sigma activation function to linearize. This RNN layer mainly includes two parts: forward propagation and backward propagation. Its calculation process is as follows:

[0068] h t =tanh(Ux t +Wh t-1 +b)

[0069] h′ t =tanh(U′x t +W′h t+1 +b′)

[0070] In the formula: x t h is the input at time t; t Let h′ be the forward propagation value at time t. t U is the backpropagation value at time t; U and U′ are the input layer values ​​x. t To the hidden layer ht State-to-state weights of the previous time step; W, W' are the hidden layer h t-1 to the current time step hidden layer h t State-to-input weights; b, b' are bias vectors.

[0071] Forward propagation value h t and backward propagation value h' t together determine the output value o t of the RNN layer, and the activation function is sigma, whose calculation formula is as follows:

[0072] o t = sigma(Vh t + V'h' t + c)

[0073] In the formula: V, V' are weight coefficients of the hidden layer h t to the output layer o t ; c is a bias vector.

[0074] (2) FC full connection layer: with 400 neural units, and the output dimension is the same as the output layer dimension of the RNN layer, which is also 400.

[0075] Step four, input the training samples in the data set into the input layer of the constructed BRNN denoising network model. Since the number of sampling points is large, the original one-dimensional structure of the signal is converted into a two-dimensional structure, and the pure noise signal is taken as a label.

[0076] Step five, the Adam optimization algorithm is used to perform forward propagation on the training samples to obtain the noise prediction value of the current time step.

[0077] Step six, the mean square error (MSE) is used as a loss function to obtain the MSE_loss value between the actual noise y i of the current time step. The definition of the mean square error loss function is as follows:

[0078]

[0079] In the formula: n is the signal length.

[0080] Step seven, update the network parameters through back propagation.

[0081] Step eight, repeat steps five to seven until the preset accuracy is met or the preset training step number is reached, and stop training to obtain the trained BRNN denoising network model. The model training process is carried out in a supervised learning manner in a labeled environment.

[0082] Step nine, input the noisy shaft frequency electric field signal into the trained BRNN denoising network model, eliminate the additive noise in the noisy shaft frequency electric field signal, and obtain the denoised shaft frequency electric field signal, as shown in Figure 3 It can be seen that under this signal-to-noise ratio condition, the shaft frequency signal is submerged in noise and cannot be directly distinguished on the time domain graph (a) and the frequency domain graph (b).

[0083] Figure 4 The time-frequency contrast result after denoising. It can be seen that under the extremely low signal-to-noise ratio condition of-20.43dB, the shaft frequency signal is completely submerged in noise, and the shaft frequency signal cannot be detected on the time domain. While effectively reducing the noise by the BRNN denoising network, the signal-to-noise ratio of the signal is improved to-4.87dB, which has a good denoising effect.

[0084] Performance comparison of different algorithms

[0085] In order to further illustrate the effectiveness of the BRNN shaft frequency electric field signal denoising method, the method and the traditional signal processing based denoising algorithm are respectively subjected to 100 times of Monte Carlo test under low signal-to-noise ratio conditions of-13.88dB and-22.43dB. Figure 5 The time-frequency graphs after denoising by different methods under the initial signal-to-noise ratio of-13.88dB are shown. Figure 5 From the comparison results of the time-frequency graphs after denoising by different methods, it can be seen that the time-frequency graphs after denoising by different algorithms have no obvious difference. In the figure, the 6Hz frequency is displayed as a bright bar at about 150s-200s time. However, the traditional signal processing method appears many noise points in other frequency bands which do not exist originally. The denoising effect of the BRNN denoising algorithm is better than that of the traditional signal processing method, which not only eliminates the noise, but also restores the original signal and retains the frequency domain characteristics of the signal.

[0086] In order to facilitate comparison and analysis, Table 1 lists the average signal-to-noise ratios after denoising by different algorithms. From Table 1, it can be seen that the BRNN denoising algorithm not only realizes denoising under low signal-to-noise ratio and extremely low signal-to-noise ratio, but also has better denoising effect than the traditional signal processing denoising method, and has the largest signal-to-noise ratio improvement amplitude, which is helpful for further signal recognition and lays a foundation for subsequent shaft frequency electric field signal detection.

[0087] Table 1

[0088]

[0089] In another embodiment of the present application, a ship shaft frequency electric field signal denoising system is provided, which can be used to realize the above-mentioned ship shaft frequency electric field signal denoising method. Specifically, the system comprises:

[0090] The data acquisition module is used for acquiring a ship shaft frequency electric field signal, and obtaining a clean shaft frequency electric field signal after preprocessing; a clean shaft frequency electric field signal obtained is linearly superimposed with a section of a long-time noise signal of a measured environment, and a noisy shaft frequency electric field signal with different signal-to-noise ratios is obtained;

[0091] The model building module is used for building a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network (BRNN).

[0092] The training module is used for taking the noisy shaft frequency electric field signal with different signal-to-noise ratios as a data set for subsequent model training and testing, inputting a training sample in the data set into the shaft frequency electric field signal denoising network model, optimizing the training sample, obtaining a noise prediction value at a current time, and obtaining a mean square error loss function of the noise prediction value and an actual noise at the current time through a loss function.

[0093] The denoising output module is used for repeatedly training to obtain a trained BRNN denoising network model, inputting the noisy shaft frequency electric field signal into the trained BRNN denoising network model for denoising.

[0094] The division of the modules in the embodiments of the application is illustrative, and is only a logical function division; another division mode can be used in actual implementation; in addition, the function modules in each embodiment of the application can be integrated in one processor, or can be physically separated, or two or more modules can be integrated in one module.

[0095] In still another embodiment of the present application, a computer device is provided, which comprises a processor and a memory, the memory being configured to store a computer program, the computer program comprising program instructions, and the processor being configured to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are particularly suitable for loading and executing one or more instructions in the computer storage medium to implement a corresponding method flow or a corresponding function; the processor in the embodiments of the present application can be used for the operation of the ship shaft frequency electric field signal denoising method.

[0096] In still another embodiment of the present application, the present application further provides a storage medium, specifically a computer readable storage medium (Memory), which is a memory device in the computer device, and is configured to store programs and data. It can be understood that the computer readable storage medium herein can include the built-in storage medium in the computer device, and of course can also include the expansion storage medium supported by the computer device. The computer readable storage medium provides a storage space, and the storage space stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and the instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium herein can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory. One or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the ship shaft frequency electric field signal denoising method in the above embodiments.

[0097] Those skilled in the art will appreciate that embodiments of the application can be devised for a method, a system, or a computer program product. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0098] The present application is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0099] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0101] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing the technical solutions of the present application, but not for limiting it. Although the present application is described in detail with reference to the above embodiments, those skilled in the field should understand that the specific embodiments of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A method for denoising a ship shaft frequency electric field signal, characterized in that, The method comprises the following steps: Collecting a ship shaft frequency electric field signal, and obtaining a clean shaft frequency electric field signal after pretreatment; Linearly superimposing the obtained clean shaft frequency electric field signal on a section of a long-time noise signal actually measured in an environment to obtain a shaft frequency electric field signal with different signal-to-noise ratios; Building a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network BRNN; Taking the shaft frequency electric field signal with different signal-to-noise ratios as a data set for subsequent model training and testing, inputting training samples in the data set into the shaft frequency electric field signal denoising network model; Optimizing the training samples, obtaining a noise prediction value at a current time, and obtaining a mean square error loss function of the noise prediction value and actual noise at the current time through a loss function; Repeating the training to obtain a trained BRNN denoising network model, and inputting the noise shaft frequency electric field signal into the trained BRNN denoising network model for denoising; In the shaft frequency electric field signal denoising network model based on the bidirectional recurrent neural network BRNN, the BRNN denoising network structure is composed of an RNN layer and an FC full connection layer. The propagation between each neuron in the RNN layer uses a tanh activation function to linearize, and the output layer uses a sigma activation function to linearize; the layer mainly includes two parts of forward propagation and backward propagation, and the calculation process is as follows: where: is the input at time step ; is the forward propagated value at time step , is the backward propagated value at time step ; , is the state-to-state weight from the input layer to the hidden layer ; , is the state-to-input weight from the hidden layer at the previous time step to the hidden layer at the current time step ; , is the bias vector; forward-propagated values and backward-propagated values together determine the output values of the RNN layer with an activation function sigma, which is calculated as follows: In the formula: , is a weight coefficient of the hidden layer to the output layer ; is a bias vector; The number of neural units and the output dimension of the FC full connection layer are the same as the output dimension of the RNN layer.

2. The method according to claim 1, wherein, The ship shaft frequency electric field signal is a measured shaft frequency electric field signal with different distances and characteristic frequencies.

3. The method according to claim 1, wherein, Linearly superimposing the obtained clean shaft frequency electric field signal on a section of a long-time noise signal actually measured in an environment is: step-by-step insertion starting from an arbitrary time point into the long-time noise data to obtain a shaft frequency electric field signal with different signal-to-noise ratios as a data set for subsequent model training and testing.

4. The method of claim 1, wherein, When inputting the training samples in the data set into the input layer of the built BRNN denoising network model, the original one-dimensional signal structure is converted into a two-dimensional structure, and the pure noise signal is taken as a label.

5. The method of claim 1, wherein, The definition of the mean square error loss function is as follows: In the formula: is the signal length.

6. The method of claim 1, wherein, The training samples are forward propagated by using Adam optimization to obtain a noise prediction value at a current time; the training is stopped until a preset accuracy is met or a preset training step number is reached, and a trained BRNN denoising network model is obtained.

7. A system for denoising a ship shaft frequency electric field signal, the system comprising: The method comprises the following steps: A data acquisition module is used to collect a ship shaft frequency electric field signal, and obtain a clean shaft frequency electric field signal after pretreatment; Linearly superimposing the obtained clean shaft frequency electric field signal on a section of a long-time noise signal actually measured in an environment to obtain a shaft frequency electric field signal with different signal-to-noise ratios; A model building module is used to build a shaft frequency electric field signal denoising network model based on a bidirectional recurrent neural network BRNN; A training module is used to take the shaft frequency electric field signal with different signal-to-noise ratios as a data set for subsequent model training and testing, input the training samples in the data set into the shaft frequency electric field signal denoising network model, optimize the training samples, obtain a noise prediction value at a current time, and obtain a mean square error loss function of the noise prediction value and actual noise at the current time through a loss function; The noise reduction output module is configured to input the noisy shaft-frequency electric field signal into the trained BRNN noise reduction network model for noise reduction. In the shaft-frequency electric field signal noise reduction network model based on the bidirectional recurrent neural network (BRNN), the BRNN noise reduction network structure is composed of an RNN layer and an FC full connection layer. The propagation between each neuron in the RNN layer is linearized by using a tanh activation function, and the output layer is linearized by using a sigma activation function; the layer mainly includes two parts of forward propagation and backward propagation, and the calculation process is as follows: where: is the input at time step ; is the forward propagated value at time step , is the backward propagated value at time step ; , is the state-to-state weight from the input layer to the hidden layer ; , is the state-to-input weight from the hidden layer at the previous time step to the hidden layer at the current time step ; , is the bias vector; forward-propagated values and backward-propagated values jointly determine the output values of the RNN layer with an activation function sigma, which is computed as follows: In the formula: , is a weight coefficient of the hidden layer to the output layer ; is a bias vector; The number of neural units and the output dimension of the FC full connection layer are the same as the output dimension of the RNN layer.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the ship shaft-frequency electric field signal noise reduction method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the steps of the ship shaft-frequency electric field signal noise reduction method according to any one of claims 1 to 6.

Images