A method for detecting tail current electromagnetic fields using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform.

Abstract

This invention provides a method for detecting tail-flow electromagnetic fields to solve the international problem of detecting tail-flow electromagnetic fields of underwater vehicles. [Solution] Disclosed is a method for detecting tail current electromagnetic fields using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, the method being performed by a computer, and including: (i) generating a computational dataset of tail current electromagnetic fields of an underwater vessel using a multi-physical-field coupling method of hydrodynamics and electromagnetism; (ii) extracting time-frequency spectral features of the tail current electromagnetic field using wavelet transform; and (iii) developing a tail current electromagnetic field detection model based on a 2D-CNN depth neural network, and training the model using the data after wavelet transform for the detection of tail current electromagnetic fields.

Description

[Technical Field]

[0001] This invention relates to the field of electromagnetism, and more specifically to a method for detecting tail current electromagnetic fields using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform. [Background technology]

[0002] The tail current formed by a submersible moving through water, displacing seawater, can persist for extended periods and travel long distances. The tail current is limited solely by the hydrodynamic behavior of the submersible and is not related to its own manufacturing materials. Seawater itself contains a large amount of charged ions, and when a submersible moves through the water, these charged ions are excited from the outside. Under the influence of the Earth's magnetic field, local electrical neutrality is broken, and an electromagnetic field distribution is excited in space by the formation of induced currents across the entire distribution of the flow field. Tail current electromagnetic field detection is a novel submersible detection technology and a cutting-edge development direction, while also being an important complement to existing non-acoustic detection technologies. Therefore, exploring tail current electromagnetic field detection methods has great significance and potential application value.

[0003] Electromagnetic field sensing technology has advanced significantly, and ultra-low-noise electromagnetic field sensors have been realized both domestically and internationally. However, these advanced electromagnetic field sensors cannot be directly applied to detecting the tail-flow electromagnetic field of underwater vehicles. This is because the very weak tail-flow electromagnetic field of an underwater vehicle is masked by the background electromagnetic field noise of the ocean, making target detection difficult. Research on tail-flow electromagnetic fields is still in its early stages both domestically and internationally, and there are currently very few official reports on detection methods. Conventional electromagnetic field detection techniques for underwater vehicles mainly rely on simplified models. For example, an underwater vehicle is equivalent to a magnetic dipole or electric dipole model, and the difference between this model and a noise model is used to extract the weak electromagnetic field signal of the underwater vehicle from the background noise using an algorithm. However, the generation mechanism of the tail-flow electromagnetic field of an underwater vehicle is relatively complex and needs to be solved by calculation using the interdependence of hydrodynamics and electromagnetism. Since there is no effective simplified electromagnetic field model, conventional electromagnetic field detection techniques for underwater vehicles cannot be applied to detecting the tail-flow electromagnetic field signal. [Overview of the project]

[0004] To address the international problem of detecting the tail-flow electromagnetic field of underwater vehicles, this invention proposes a tail-flow electromagnetic field detection method using a 2D-CNN (two-dimensional convolutional neural network) based on time-frequency spectral features obtained by wavelet transform.

[0005] The present invention uses a deep learning method as the detection means, a velocity field simulation as the basis for the simulation, and an electromagnetic field simulation as the core data source, thereby significantly increasing the amount of data available for model training and improving detection accuracy.

[0006] More specifically, the present invention improves the detection capability of tail-flow electromagnetic fields from three aspects. First, it solves the problem of constructing a complex tail-flow electromagnetic field detection model by training and optimizing a deep learning model with computational data of tail-flow electromagnetic fields of underwater vessels. Second, it solves the problem of creating deep learning training samples by generating computational datasets of tail-flow electromagnetic fields of underwater vessels using a multi-physical-field coupling method of hydrodynamics and electromagnetism. Third, it solves the deep learning feature representation problem in tail-flow electromagnetic field detection by extracting time-frequency spectral features of tail-flow electromagnetic fields of underwater vessels using wavelet transform.

[0007] Specifically, the present invention provides a tail current electromagnetic field detection method using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, the method being performed by a computer and comprising the following steps (1), (2), and (3).

[0008] Step (1): Generate a computational dataset of the tail-flow electromagnetic field of an underwater submersible using a multi-physical-field coupling method of hydrodynamics and electromagnetism. Step (1) is, Steps (1.1) include constructing a geometric model using power field simulation software and setting the computational domain of the geometric model, Steps include (1.2) of dividing the configured computation domain into a mesh, The steps include (1.3) setting boundary conditions for the geometric model, wherein the boundary conditions include at least a velocity inlet, a pressure outlet, a plane of symmetry, and a wall, Step (1.4) involves solving the velocity field distribution within the computational domain using power field simulation software, Based on the obtained velocity field distribution, the tail current electric field E and tail current magnetic field B at any point within the computational domain are calculated according to the following equation.

[0009]

number

[0010]

number

[0011] In equations (1) and (2), position vectors r' and r represent the position of the source and the position of the measured point, respectively, where J(r') represents the current density at position r', V represents the integral volume, v0 represents the water flow velocity at the measured point calculated from the velocity field, μ represents the permeability of seawater, σ represents the conductivity of seawater, and B E This is a step (1.5) representing the Earth's magnetic field, The process includes (1.6) modifying the input and model parameters, repeating the above steps to obtain the distributions of tail current electric field E and tail current magnetic field B under different input and model parameter conditions, and obtaining multiple sets of tail current electric field E and tail current magnetic field B data to construct a dataset.

[0012] Step (2): The entire computation domain is used as the measurement domain. Multiple measurement points are selected from the measurement domain. Data on the time-varying electric and magnetic fields at each measurement point under different conditions is obtained from the dataset. The obtained electric and magnetic field data is subjected to a wavelet transform to obtain the dataset after the wavelet transform.

[0013] Step (3): Construct a 2D-CNN depth neural network, divide the constructed dataset into a training set and a test set, input the data from the training set into the 2D-CNN depth neural network, train the 2D-CNN depth neural network, and test it using the test set to obtain the trained 2D-CNN depth neural network, which is then used for tail current electromagnetic field detection.

[0014] In a preferred embodiment, the power field simulation software is CFD software. In another preferred embodiment, the computational domain is a cuboid region, and a SUBOFF model used to simulate an underwater vehicle is set in the computational domain.

[0015] In another preferred embodiment, the current density J is

Equation

[0016] In another preferred embodiment, the wavelet transform is

Equation

[0017] In another preferred embodiment, the computational domain calculates the wake electromagnetic field using a local mesh division and interpolation method. In another preferred embodiment, the network model includes a convolutional module and a fully-connected classifier module. The convolutional module includes seven convolutional layers, RELU layers, three batch normalization layers, and four pooling layers. The fully-connected classifier module includes one flattening layer, three fully-connected layers, and two RELU layers.

[0018] On the other hand, the present invention provides a method for obtaining a dataset used for training a wake electromagnetic field detection model. The method is Steps (1.1) include constructing a geometric model using power field simulation software and setting the computational domain of the geometric model, In the constructed geometric model, the steps are: (1.2) to mesh the computational domain, The steps include (1.3) setting boundary conditions for the geometric model, wherein the boundary conditions include at least a velocity inlet, a pressure outlet, a plane of symmetry, and a wall, Step (1.4) involves solving the velocity field distribution within the computational domain using power field simulation software, Based on the obtained velocity field distribution, the tail current electric field E and tail current magnetic field B at any point within the computational domain are calculated according to the following equation.

[0019]

number

[0020]

number

[0021] In equations (4) and (5), position vectors r' and r represent the position of the source and the position of the measured point, respectively, where J(r') represents the current density at position r', V represents the volume, v0 represents the velocity calculated from the velocity field, μ represents the permeability of seawater, σ represents the conductivity of seawater, and B E This is a step (1.5) representing the Earth's magnetic field, The process includes (1.6) repeating the above steps to modify the input and model parameters to obtain the distributions of tail current electric field E and tail current magnetic field B under different input and model parameter conditions, obtaining multiple sets of tail current electric field E and tail current magnetic field B data, and constructing a dataset.

[0022] The present invention aims to realize the detection of tail current electromagnetic fields and has the following advantages. Firstly, the present invention successfully solves the problem of generating a computational dataset of the tail-flow electromagnetic field of an underwater vehicle by employing a multi-physical-field coupling method of hydrodynamics and electromagnetism, and creating deep learning training samples.

[0023] First, multi-physical-field coupling can provide a more comprehensive and realistic simulation environment, allowing the generated dataset to more accurately reflect the actual characteristics of the tail-flow electromagnetic field of an underwater vehicle, thereby improving the training effectiveness of deep learning models. Second, this method can flexibly adjust various parameters and conditions to generate diverse scenarios and samples, improving the broad applicability of tail-flow electromagnetic field training data and the generalization ability of the model. Furthermore, this automated generation technique can significantly reduce the time and cost of data collection while ensuring high quality and consistency of the data, thus further accelerating the development and optimization of tail-flow electromagnetic field detection models.

[0024] Secondly, the present invention effectively solves the problem of feature representation in deep learning for tail-flow electromagnetic field detection by extracting time-frequency spectral features of the tail-flow electromagnetic field of an underwater vehicle using wavelet transform. The advantage is that since wavelet transform can simultaneously analyze the time and frequency information of a signal, it can capture subtle changes and local features in the tail-flow electromagnetic field signal, thereby making feature extraction more accurate and richer. Its multi-resolution analysis capability not only improves the sensing and discrimination ability of deep learning models for complex tail-flow electromagnetic field signals but also reduces reliance on conventional feature extraction methods. Furthermore, the features generated by wavelet transform can enhance the robustness and stability of the model, helping to improve the accuracy and reliability of tail-flow electromagnetic field detection.

[0025] Thirdly, the present invention can successfully solve the problem of constructing a complex tail-flow electromagnetic field detection model by training and optimizing a deep learning model using a large amount of tail-flow electromagnetic field calculation data from underwater vehicles.

[0026] This invention employs a novel deep learning model with powerful feature extraction capabilities, enabling it to automatically identify and capture complex patterns and subtle changes in tailflow electromagnetic fields, thereby reducing reliance on traditional expert knowledge and manual feature engineering. The scalability and flexibility of the deep learning method allow for easier model updates and extensions, enabling adaptation to tailflow electromagnetic field detection needs and environmental changes. [Brief explanation of the drawing]

[0027] [Figure 1] This diagram shows the configuration of the calculation domain for the underwater vehicle SUBOFF model, where (a) is a plan view and (b) is a side view. [Figure 2] This is a schematic diagram showing the thickness relationship of the Y+1st layer mesh cells. [Figure 3] This is a schematic diagram illustrating a high Y+ wall surface treatment method. [Figure 4] This figure shows the mesh division results for the underwater vehicle SUBOFF model. [Figure 5] This is a schematic diagram of the integration region, showing the total integration region (left) and the local integration region (right). [Figure 6] This is a schematic diagram illustrating linear interpolation over a local integration region. [Figure 7] This is a flowchart for calculating the tail current electromagnetic field. [Figure 8] This is a schematic diagram showing the placement of analog electromagnetic sensors. [Figure 9] This figure shows a comparison of electric field data with pure noise (left) and electric field data with signal-noise mixed (right). [Figure 10] This figure shows the structure of a 2D-CNN depth neural network for detecting tail current electromagnetic fields. [Figure 11] This diagram compares the results of signal detection and evaluation using three different algorithms. [Modes for carrying out the invention]

[0028] The present invention will be described in detail below in conjunction with the attached drawings and embodiments thereof, but the scope of protection of the present invention is not limited to the scope described in the embodiments.

[0029] Overall, the tail-flow electromagnetic field detection method of the present invention, based on time-frequency spectral features obtained by wavelet transform, first acquires a large dataset usable for simulation training based on hydrodynamic and electromagnetism simulations. Then, the tail-flow electromagnetic field signal of an underwater vessel is extracted into a time-frequency spectral feature map using wavelet transform, and this is then processed and analyzed by a 2D-CNN. The wavelet transform can effectively capture detailed changes in the signal at different times and frequencies, and the 2D-CNN can perform deep learning and pattern recognition on these time-frequency spectral features, thus enabling accurate detection of the tail-flow electromagnetic field. This method can not only improve the accuracy of feature extraction but also improve the reliability and efficiency of signal recognition, thus providing a new technical means for detecting the tail-flow electromagnetic field of an underwater vessel.

[0030] The method of the present invention comprises three parts: (i) generating a computational dataset of the tail-flow electromagnetic field of an underwater vessel using a multi-physical-field coupling method of hydrodynamics and electromagnetism; (ii) extracting time-frequency spectral features of the tail-flow electromagnetic field using wavelet transform; and (iii) developing a tail-flow electromagnetic field detection model based on a 2D-CNN depth neural network. The applicant will now describe each of these three parts in detail.

[0031] (1) Generate a computational dataset of the tail-flow electromagnetic field of an underwater vehicle using a multi-physical-field coupling method of hydrodynamics and electromagnetism.

[0032] (Explanation of the principle) The seawater environment contains a large number of charged particles. Due to the influence of Coulomb forces, several ions with the opposite charge are attracted to any positively or negatively charged ion. When seawater is in a state of balance, these ions exhibit an electrically neutral distribution. When seawater is excited (for example, by a tail current formed by the movement of an underwater vehicle), these charged particles separate locally due to their own mass and coefficient of friction, generating an electric field between the ion microclusters and forming an ionic polarization current.

[0033] To study ionic polarization currents, we first perform a force analysis on ions in a seawater environment. Ions in seawater are affected by various forces, and the dominant forces differ depending on the scenario. However, generally, the force equation for ions in a seawater environment can be expressed as follows.

[0034]

number

[0035] In equation (6), the subscript p represents p types of ions, and there are a total of seven terms on the right side of the equation from left to right, with the following meanings: Electric field force: The amount of electric charge is e p The electric field force E acting on the ions. Magnetic field force: The Lorentz force exerted on a moving ion within a magnetic field B. Frictional force: This is the frictional force acting on a viscous solute, and this force can be described by the Debye-Huckel theory.

[0036] Relaxation force: This force is caused by perturbations in ion mobility due to displacement of the ionic atmosphere, and can be described by the Debye-Huckel-Onsanger theory. Electrophoretic force is a perturbation of viscous resistance caused by the movement of different ions at different speeds, and can similarly be described by the Debye-Hückel-Onsanger theory.

[0037] Gradient force: A thermodynamic force described by Fieck's second law of diffusion. Pressure: The pressure that ions experience when moving from a high-pressure region to a low-pressure region. In the formula, V p v0 and ρ0 represent the effective ion volume, fluid velocity, and fluid density, respectively. Analyzing equation (6), the polarization current density J is as follows:

[0038]

number

number

[0039] Analyzing the current density of each term in equation (7), the first term originates from the Lorentz term of the Earth's magnetic field and can be considered as the ion-induced polarization current density. The second term originates from the total differential of the perturbation velocity field itself, and essentially, different ions have different molar masses, and due to inertial effects, polarization separation occurs, forming a polarization current. This term can be considered as the ion inertial polarization current density. The third term is a combined term of the previous two terms, and an order analysis of it shows that it is much smaller than the previous two terms, so it can be ignored, and thus the following equation is obtained.

[0040]

number

[0041] Generally, ionic polarization currents due to any externally excited perturbation velocity field include both an ionic induction term and an ionic inertia term. However, the frequency of the perturbation velocity field caused by the tail current of an underwater vehicle during underwater navigation is relatively low, and the contribution of the ionic inertia term to the polarization current density is very weak. Therefore, the ionic induction current accounts for the majority of the current. By introducing a system of source-based Maxwell's equations and assuming seawater as a homogeneous medium, the following equation is obtained.

[0042]

Number

Number

Number

Number

[0043] Here, "∇×" represents the rotation operator, "∇·" represents the divergence operator, and ∂ / ∂t represents the partial derivative with respect to time. H is the magnetic field strength of the wake, D is the electric displacement vector, E is the electric field strength of the wake, J is the ion polarization current density vector in seawater, and μ represents the magnetic permeability of seawater. From the current density formula of Equation (8), the magnetic induction intensity B can be regarded as the sum of two parts, namely the geomagnetic field B E and the wake magnetic field B i That is,

Number

Number

Number

[0044] The problem of interest in studying tailflow electromagnetic fields is a low-frequency problem, and since the phase difference of the electromagnetic field at different locations within the range of interest is almost zero, the time-harmonic term can be ignored when finding the solution, and based on equation (15), the relation for tailflow electromagnetic fields can be obtained as follows.

[0045]

number

number

[0046] Given the physicochemical parameters of seawater and the geomagnetic field region, in equations (16) and (17), only the tail current velocity field v0 is an unknown parameter. By dividing the velocity field into discrete mesh cells, we can complete the process of obtaining a discrete solution for the tail current electromagnetic field.

[0047] Therefore, solving the tail current electromagnetic field in this invention can be divided into two steps. The first step is to solve the tail current velocity field v0 using CFD fluid simulation software and employing the k-ε turbulence model (this step is completed using general-purpose simulation software), and the second step is to input the obtained tail current velocity field v0 as a parameter and obtain the tail current electric field E and tail current magnetic field B using numerical calculations.

[0048] The specific process for solving the tail current electromagnetic field and obtaining the dataset is as follows: (1) Fluid simulation using CFD software: This includes steps such as geometric modeling, meshing, setting simulation material properties, setting boundary and initial conditions, and solving using a solver. The important steps are as follows:

[0049] 1. Geometric modeling and setting of the computational domain. As shown in Figure 1, the computational domain of the present invention is a rectangular parallelepiped region, the underwater vehicle employs a SUBOFF model, the diving depth is h, the height of the air layer is 20D-h (where D is the diameter of the SUBOFF model), the height of the seawater layer is 20D+h, the width of the computational domain is 2L+2wavelength (where L is the length of the SUBOFF model), and the length of the computational domain is 7L+2wavelength. In Figure 1(a), the computational domain is divided into a solution-finding region and an attenuation region. The attenuation region plays a role in preventing wave reflection at the upstream, downstream, and lateral boundaries of the computational domain during the simulation. The width of the attenuation region is 1.wavelength, and this length represents the maximum wavelength λ of the waves generated by the SUBOFF. This is determined by the dispersion relation of linear surface waves under deep water conditions and can be calculated by the relationship between wavelength λ and the SUBOFF model velocity U. Note that the SUBOFF model is not simulated as a model that moves forward at velocity U, but rather as a uniform inflow of velocity U.

[0050] 2. Mesh division From the perspective of CFD fluid simulation calculations, meshing is a crucial step in determining whether the simulation can converge, whether the calculation accuracy meets the requirements, and whether the calculation time is sufficiently short. For the SUBOFF model, the mesh type selected by this invention is a hexahedral mesh, and the thickness X of the first layer of meshing near the wall surface is very important in order to effectively simulate turbulent flow near the outer wall surface of the SUBOFF model. To calculate this, the target Y of the problem is... + We need to know the value, Y + The value is a dimensionless distance related to the selection of the turbulence model, and represents the length from the center of the first layer mesh cell to the wall (as shown in Figure 2). As shown in Figure 3, Y + When processing a wall surface using a standard, high Y + A wall treatment approach is employed, which involves dividing the wall surface using a relatively coarse mesh and using a standard wall function to handle the viscous region near the wall surface, with a value between 30 and 500.

[0051] Y+ The mesh was divided with a value of =100, and the division result is shown in Figure 4. In this case, the body mesh near the wall surface of the underwater vehicle and the mesh on the wall surface are finely divided, while the mesh size of the remaining parts is relatively large, resulting in a good balance between calculation accuracy and calculation speed.

[0052] 3. Setting boundary conditions and initial conditions The boundary conditions are set as follows: Velocity inlet: The inlet end face, the upper part of the air layer, and the bottom of the seawater layer are set as boundary conditions for the velocity inlet, corresponding to the SUBOFF speed. Pressure outlet: The outlet end face is set as the boundary condition for the pressure outlet. Symmetry plane: By setting the left and right sides of the calculation domain in the SUBOFF motion direction as boundary conditions for the symmetry plane, the flow field is considered to exhibit a symmetric form. Wall: The SUBOFF model itself is set as the boundary condition for a non-slip wall. The initial setting parameters include the cruising speed, underwater depth, and scale size of the SUBOFF model.

[0053] 4. Solving using a solver For the SUBOFF model, a two-phase medium of air-seawater is present. In this invention, VOF multiphase flow calculation is adopted, the k-ε turbulence model is adopted as the turbulence model, gravity, VOF waves, and unit mass restoration are selected as additional models, and the steady-state solution method is used to perform simulations with CFD software to determine the velocity field v0 under the current simulation conditions.

[0054] As shown in Figure 5, when calculating the solution for the tailflow magnetic field, the mesh size of the tailflow velocity field to be calculated is assumed to be N × N × M (where N represents the mesh size of the horizontal plane and M represents the mesh size of the depth plane), and one local mesh of size n × n × n is selected from among them.

[0055] According to the definition of the integral used to calculate the tail-flow magnetic field for any given intermediate point P, a discrete sum should be performed within the entire integration domain. However, due to the decay characteristics, only one local region is selected around the location of point P to perform the discrete sum and obtain an approximate value of the tail-flow magnetic field. For each point within the integration range (each discrete point within the selected local region), the result v0 of the velocity field is substituted into equation (17) to obtain the magnetic field at the corresponding point (multiple iterative cycles are possible considering boundary effects), and the electric field at the corresponding point is calculated using equation (16).

[0056] Selectively, as shown in Figure 6, in addition to the above operations, the calculation accuracy of the tail current electric field can be further improved by interpolating the current density for the nearest mesh to the midpoint P. As shown in Figure 7, a flowchart for calculating the tail current induced electromagnetic field based on the above calculation concept is shown. By changing the input conditions or model parameters of the simulation process and repeating the above process, electromagnetic field data for different cases can be obtained.

[0057] (ii) Extraction of time-frequency spectral features of tailflow electromagnetic fields using wavelet transform Several measurement points are selected from the entire measurement area, and data (or change functions) of the electric and magnetic fields at each measurement point that change over time are obtained. The resulting electromagnetic field data is then subjected to a wavelet transform. Here, the electromagnetic field dataset can be further expanded by selecting the locations of different measurement points.

[0058] The wavelet transform overcomes the drawback of the short-time Fourier transform, which cannot simultaneously improve both temporal and frequency resolution while retaining its advantages. It can match an input signal by representing the signal using wavelet basis functions and then scaling and translating the wavelet basis functions, and can be expressed as follows:

[0059]

number

[0060] Here, Ψ(t) represents a wavelet basis function also called a mother wavelet, and the present invention adopts the Morse wavelet basis function, Ψ ω,τ (t) represents a subwavelet generated by scaling and translating the wavelet basis function, W f (ω, τ) is the result of wavelet transforming the signal f(t). For a time-sampled signal, it contains time-domain and frequency-domain information. This invention uses the wavelet method to extract time-frequency information of a tailing electromagnetic field.

[0061] As shown in Figure 8, N analog electromagnetic sensors are positioned in front of the underwater vehicle, all on one side, with two adjacent sensors spaced 10 m apart. Taking the electric field signal at measurement point S1 shown in Figure 8 as an example, the sampling rate is 3 Hz and the sampling time is 146 s. Gaussian white noise is superimposed on the data from S1, and the signal-to-noise ratio is -5 dB. As shown in Figure 9, a wavelet transform is performed, and the time domain and wavelet transform results of the signal-to-noise mixed electric field data and the pure noise electric field data are compared. When the signal-to-noise ratio is -5 dB, there is basically no difference in the peak values ​​and waveforms of the pure noise electric field data and the signal-to-noise mixed electric field data, indicating that the target signal can no longer be identified at this point. However, the frequency point characteristics and duration of the target signal can be extracted from the time-frequency spectrogram of the wavelet transform.

[0062] (3) Development of a tail current electromagnetic field detection model based on a 2D-CNN depth neural network The structure of the 2D-CNN depth neural network designed by the present invention is as shown in Figure 10, and includes two modules: a convolutional layer and a fully connected classifier layer. The convolutional module includes seven convolutional layers, a RELU layer, three batch normalization layers, four pooling layers, etc., and the fully connected classifier module includes one flattening layer, three fully connected layers, and two RELU layers, employing a dropout process.

[0063] The target data generation method involves selecting data from several electromagnetic field sensors, setting the signal-to-noise ratio to -10 to 0 dB, and using a step size of 1 dB. The total length of the time-domain signal is 440, where the length of the target signal is 195. To make signal detection more realistic and simultaneously increase the diversity of features in the dataset, the target signal is made to appear randomly throughout the signal.

[0064] Next, the model is trained using the training set. The training set is configured to have a total of 200 data points for a given electromagnetic sensor and signal-to-noise ratio, with target signal data and noise data each accounting for half of the data. For example, if the signal-to-noise ratio is -5dB, the S1 electric field sensor has a total of 200 data points, of which 100 have the target signal and the remaining 100 do not. Therefore, the training set has a total of 22,000 time-domain data points. These are then transformed using wavelet transform to obtain 22,000 two-dimensional time-frequency spectrograms. Similarly, the test set has 4,400 time-frequency spectrograms.

[0065] The training loss function used is the cross-entropy function, the optimization method is Adam, the learning rate is lr=0.0001, the batch size is batchsize=44, and the training rounds are epoch=20. We will use a trained model to test electromagnetic field tailflow detection.

[0066] The depth neural network proposed in this invention was compared with other neural networks, including FCN (Fully Connected Neural Network) and 1D-CNN (One-Dimensional Convolutional Neural Network). The results are shown in Figure 11. It was found that all three methods issued alarm signals during the time period of tail-flow electromagnetic field signals, indicating that tail-flow electromagnetic field signals were detected. However, during some pure noise periods, the latter two algorithms exhibited false alarm phenomena. This indicates that the detection performance of the depth neural network proposed in this invention is superior to that of FCN neural networks and 1D-CNN neural networks.

[0067] Unlike 1D-CNNs, 2D-CNNs target images, while 1D-CNNs target curves. The structure of the present invention is a 2D-CNN, and unlike the results of the FCN model, a typical 2D-CNN model includes convolutional layers, pooling layers, and fully connected layers. However, the FCN removes the fully connected layers and global average pooling layers of the 2D-CNN and introduces transposed convolutional layers.

[0068] In short, the tail current electromagnetic field detection method using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, as proposed in this invention, can effectively improve the detection capability of tail current electromagnetic fields, providing a new technical means for detecting tail current electromagnetic fields of underwater vehicles, and is expected to be applied in the fields of electromagnetic field monitoring and early warning of marine targets.

[0069] The principles of the present invention have been described in detail above, along with preferred embodiments of the present invention. However, those skilled in the art will understand that the above embodiments are merely illustrative descriptions of the present invention and do not limit the scope of the invention. Details in the embodiments do not limit the scope of the invention, and obvious modifications such as equivalent transformations and simple substitutions based on the technical solutions of the present invention are all included within the scope of protection of the present invention without departing from the spirit and scope of the invention.

Claims

1. A method for detecting tail current electromagnetic fields using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, which is performed by a computer, Steps include constructing a geometric model and defining the computational domain of the geometric model (1.1), Step 1.2 involves meshing the configured computation domain, The steps include (1.3) setting boundary conditions for the geometric model, wherein the boundary conditions include at least a velocity inlet, a pressure outlet, a plane of symmetry, and a wall, Step 1.4: Solving the velocity field distribution within the computational domain, Based on the obtained velocity field distribution, the tail current electric field E and tail current magnetic field B at any point in the calculation domain are calculated according to the following equation: [Math 21] [Number 22] In equations (1) and (2), the position vectors r' and r represent the position of the source and the position of the measured point, respectively, where J(r') represents the current density at position r', V represents the integral volume, and v 0 μ represents the water flow velocity at the measured point obtained from the velocity field, μ represents the permeability of seawater, σ represents the conductivity of seawater, and B E This is a step (1.5) representing the Earth's magnetic field, Step (1) generates a computational dataset of the tail electromagnetic field of an underwater submersible by a multi-physical-field coupling method of hydrodynamics and electromagnetism, which includes step (1.6) modifying the input parameters and model parameters, repeating the above steps to obtain the distribution of tail current electric field E and tail current magnetic field B under different input parameter and model parameter conditions, obtaining multiple sets of tail current electric field E and tail current magnetic field B data, and constructing a dataset. Step (2) involves using the entire computational domain as the measurement area, selecting multiple measurement points from the measurement area, obtaining time-varying data of the electric and magnetic fields at each measurement point under different conditions from the dataset, performing a wavelet transform on the obtained electric and magnetic field data, and obtaining a dataset after the wavelet transform. A method for detecting tail-flow electromagnetic fields using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, comprising the steps of: (3) constructing a 2D-CNN depth neural network; dividing the constructed dataset into a training set and a test set; inputting the data in the training set into the 2D-CNN depth neural network; training the 2D-CNN depth neural network; and testing it using the test set to obtain the trained 2D-CNN depth neural network; and using the obtained trained 2D-CNN depth neural network for tail-flow electromagnetic field detection.

2. The tail current electromagnetic field detection method by a 2D-CNN network based on time-frequency spectral features by wavelet transform according to claim 1, characterized in that steps (1.1) and (1.4) are performed with power field simulation software, the power field simulation software being CFD software.

3. The tail current electromagnetic field detection method using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, characterized in that the computation domain is a rectangular parallelepiped region, and a SUBOFF model for simulating an underwater vehicle is set in the computation domain.

4. Current density J is, [Number 23] Calculated by, Here, σ represents the conductivity of seawater, and v 0 This represents the water flow velocity at the measured point, calculated from the velocity field, and B E The method for detecting a tail current electromagnetic field using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform according to claim 1, characterized in that it represents the Earth's magnetic field.

5. Wavelet transform is, [Number 24] We will adopt this approach. Here, Ψ(t) represents a wavelet basis function, and Ψ ω,τ (t) represents a subwavelet generated by scaling and translating the wavelet basis function, W f The method for detecting a tail current electromagnetic field using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, characterized in that (ω, τ) is the result of wavelet transforming the signal f(t), as described in claim 1.

6. The method for detecting a tailflow electromagnetic field using a 2D-CNN network based on time-frequency spectral features obtained by wavelet transform, as described in claim 1, characterized in that the computational domain calculates the tailflow electromagnetic field using local mesh division and interpolation methods.

7. The method for detecting tail current electromagnetic fields using a 2D-CNN network based on time-frequency spectral features by wavelet transform, as described in claim 1, wherein the network model includes a convolutional module and a fully coupled module, the convolutional module includes seven convolutional layers and RELU layers, three batch normalization layers, and four pooling layers, and the fully coupled classifier module includes one flattening layer, three fully coupled layers, and two RELU layers.

8. Steps include constructing a geometric model and defining the computational domain of the geometric model (1.1), In the constructed geometric model, the steps include (1.2) meshing the computational domain, The steps include (1.3) setting boundary conditions for the geometric model, wherein the boundary conditions include at least a velocity inlet, a pressure outlet, a plane of symmetry, and a wall, Step 1.4: Solving the velocity field distribution within the computational domain, Based on the obtained velocity field distribution, the tail current electric field E and tail current magnetic field B at any point in the calculation domain are calculated according to the following equation: [Number 25] [Number 26] In equations (4) and (5), the position vectors r' and r represent the position of the source and the position of the measured point, respectively, where J(r') represents the current density at position r', V represents the volume, and v 0 μ represents the velocity calculated from the velocity field, μ represents the permeability of seawater, σ represents the conductivity of seawater, and B represents the velocity calculated from the velocity field. E This is a step (1.5) representing the Earth's magnetic field, A method for obtaining a dataset used to train a tail current electromagnetic field detection model, comprising the steps of: (1.6) repeating the above steps to modify the input parameters and model parameters to obtain the distribution of tail current electric field E and tail current magnetic field B under different input parameter and model parameter conditions, thereby obtaining multiple sets of tail current electric field E and tail current magnetic field B data and constructing a dataset.

Images