A 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics

Abstract

The application discloses a 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics, and the method comprises the following steps: (1) generating an underwater vehicle wake electromagnetic field calculation data set by a multi-physical field coupling method of hydrodynamics and electromagnetism; (2) extracting time-frequency spectrum characteristics of the wake electromagnetic field by wavelet transform; (3) developing a wake electromagnetic field detection model based on a 2D-CNN deep neural network, and training the model by using the wavelet transformed data, so as to be used for wake electromagnetic field detection. The 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics can effectively improve the detection capability of the wake electromagnetic field, provides a new technical means for the underwater vehicle wake electromagnetic field detection, and has a good application prospect in the field of ocean target electromagnetic field monitoring and early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electromagnetism, in particular to a 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics. BACKGROUND

[0002] When the underwater vehicle is sailing, the wake formed by the water is discharged, which can last for a long time and spread far away. The wake is only limited by the hydrodynamic behavior of the underwater vehicle, and has nothing to do with the construction material itself. The seawater itself contains a large number of charged ions, when the underwater vehicle sails to promote the movement of seawater, the charged ions obtain external excitation. Under the action of the geomagnetic field, the local electric neutrality will be broken, and the induced current will be formed in the whole flow field distribution, thereby exciting the electromagnetic field distribution in space. Wake electromagnetic field detection is a new type of underwater vehicle detection technology and a front development direction, and it is also an important supplement to the existing non-acoustic detection technology. Therefore, it is of great significance and potential application value to explore the wake electromagnetic field detection method.

[0003] Although the electromagnetic field sensing technology has developed rapidly, the advanced electromagnetic field sensor cannot be directly applied to the wake electromagnetic field detection of the underwater vehicle at present. This is because the extremely weak wake electromagnetic field of the underwater vehicle will be overwhelmed by the noise, resulting in difficulty in target detection due to the influence of the marine background electromagnetic field noise. The research on wake electromagnetic field is just starting at home and abroad, and there are few public reports on its detection method. The traditional electromagnetic field detection technology of underwater vehicle mainly relies on simplified models, such as equivalent magnetic dipole or electric dipole model of underwater vehicle, and the difference between the model and noise model is used to extract the weak electromagnetic field signal of underwater vehicle from background noise through algorithm. However, the generation mechanism of the wake electromagnetic field of the underwater vehicle is relatively complex, which needs to be calculated and solved by using the mutual coupling of hydrodynamics and electromagnetism, and there is no effective electromagnetic field simplified model, so the traditional electromagnetic field detection technology of underwater vehicle cannot be applied to the wake electromagnetic field signal detection. SUMMARY

[0004] In order to solve the international problem of wake electromagnetic field detection of underwater vehicle, the present application provides a 2D-CNN (two-dimensional convolutional neural network) wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics.

[0005] The method of the present application uses deep learning method as detection means, speed field simulation as simulation basis, and electromagnetic field simulation as core data source, which greatly improves the amount of data available for model training and improves the detection accuracy.

[0006] More specifically, the present application improves the detection capability of the wake electromagnetic field from three aspects: first, the deep learning model is trained and optimized through a large amount of underwater vehicle wake electromagnetic field calculation data, which solves the problem of constructing a complex wake electromagnetic field detection model; second, the underwater vehicle wake electromagnetic field calculation dataset is generated through the multi-physical field coupling method of hydrodynamics and electromagnetics, which solves the problem of creating deep learning training samples; third, the time-frequency spectrum features of the underwater vehicle wake electromagnetic field are extracted through wavelet transform, which solves the problem of feature representation in deep learning in wake electromagnetic field detection.

[0007] Specifically, the present application provides a 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum features, which comprises:

[0008] Step (1) generating a submarine wake electromagnetic field calculation dataset through the multi-physical field coupling method of hydrodynamics and electromagnetics,

[0009] This step (1) comprises:

[0010] (1.1) constructing a geometric model in a dynamic field simulation software and setting a calculation domain of the geometric model;

[0011] (1.2) performing grid division on the set calculation domain;

[0012] (1.3) setting boundary conditions for the geometric model, the boundary conditions at least comprising a velocity inlet, a pressure outlet, a symmetric surface and a wall surface;

[0013] (1.4) solving the velocity field distribution in the calculation domain in the dynamic field simulation software;

[0014] (1.5) based on the obtained velocity field distribution, calculating the wake electric field E and the wake magnetic field B of any point in the calculation domain according to the following formula:

[0015] E(r)=v0(r)×B(r)=v0(r)×(B E (r)+B i (r))≈v0(r)×B E (r)(1)

[0016]

[0017] In formula (1) and formula (2), the position vector r' and the position vector r respectively represent the position of the source point and the position of the point to be measured, wherein J(r') represents the current density at the position r', V represents the integral volume, v0 represents the water flow velocity of the point to be measured obtained according to the velocity field calculation, μ represents the magnetic permeability of seawater, σ represents the electrical conductivity of seawater, and B E represents the geomagnetic field;

[0018] (1.6) modifying the input parameters and the model parameters, repeating the above steps to obtain the wake electric field E and the wake magnetic field B distribution under different input parameters and model parameters, obtaining multiple sets of wake electric field E and wake magnetic field B data to form a data set;

[0019] Step (2) taking the entire calculation domain as a measurement area, selecting a plurality of measurement points from the measurement area, obtaining the data of the electric and magnetic fields of each measurement point under different conditions from the data set, and performing wavelet transform on the obtained electric and magnetic field data to obtain a wavelet-transformed data set;

[0020] Step (3) constructing a 2D-CNN deep neural network, dividing the constructed data set into a training set and a test set, inputting the data in the training set into the 2D-CNN deep neural network, training the 2D-CNN deep neural network and testing it by using the test set, obtaining the trained 2D-CNN deep neural network, and using the obtained trained 2D-CNN deep neural network for wake electromagnetic field detection.

[0021] In a preferred implementation, the dynamic field simulation software is CFD software.

[0022] In another preferred implementation, the calculation domain is a cuboid region, and a SUBOFF model is arranged in the calculation domain to simulate an underwater vehicle.

[0023] In another preferred implementation, the current density J is calculated by the following formula:

[0024] J = σv0x B E .

[0025] wherein σ represents seawater conductivity, v0 represents the flow velocity of the to-be-measured point obtained according to the velocity field, and B E represents the geomagnetic field.

[0026] In another preferred implementation, the wavelet transform is performed by the following formula:

[0027]

[0028] wherein Ψ(t) represents a wavelet base function; Ψ ω,τ (t) represents a sub-wavelet generated by scaling and shifting the wavelet base function; W f (ω, τ) is the result of wavelet transform on the signal f(t).

[0029] In another preferred implementation, the calculation domain uses a local grid partitioning and interpolation method to calculate the wake electromagnetic field.

[0030] In another preferred implementation, the network model comprises a convolution module and a full connector module, the convolution module comprising 7 convolution layers and RELU layers, 3 batch normalization layers, and 4 pooling layers; the full connection classifier module comprising 1 flattening layer, 3 full connection layers, and 2 RELU layers.

[0031] In another aspect, the present application provides a data set acquisition method for training of a wake electromagnetic field detection model,

[0032] The method comprises:

[0033] (1.1) constructing a geometric model in a dynamic field simulation software, and setting a calculation domain of the geometric model;

[0034] (1.2) performing grid division on the calculation domain in the constructed geometric model;

[0035] (1.3) setting boundary conditions for the geometric model, the boundary conditions at least comprising a velocity inlet, a pressure outlet, a symmetric surface, and a wall surface;

[0036] (1.4) solving a velocity field distribution in the calculation domain in the dynamic field simulation software;

[0037] (1.5) based on the obtained velocity field distribution, calculating a wake electric field E and a wake magnetic field B at any point in the calculation domain according to the following formula:

[0038] E(r)=v0(r)×B(r)=v0(r)×(B E (r)+B i (r))≈v0(r)×B E (r)(4)

[0039]

[0040] In formula (4) and formula (5), the position vector r' and the position vector r respectively represent the position of a source point and the position of a point to be measured, wherein J(r') represents a current density at the position r', V represents a volume, v0 represents a velocity obtained according to a velocity field calculation, μ represents a seawater permeability, σ represents a seawater conductivity, and B E represents a geomagnetic field.

[0041] (1.6) repeating the above steps to modify input parameters and model parameters, obtaining the wake electric field E and the wake magnetic field B distribution under different input parameters and model parameters, obtaining multiple groups of wake electric field E and wake magnetic field B data, and constituting a data set.

[0042] The present application has the following advantages in realizing wake electromagnetic field detection:

[0043] First: This invention uses a multi-physics coupling method of hydrodynamics and electromagnetism to generate a dataset for calculating the electromagnetic field of the underwater vehicle wake, successfully solving the problem of creating training samples for deep learning.

[0044] First, multiphysics coupling provides a more comprehensive and realistic simulation environment, enabling the generated dataset to more accurately reflect the true characteristics of the wake electromagnetic field of underwater vehicles, thereby improving the training effect of deep learning models. Second, this method can flexibly adjust various parameters and conditions to generate diverse scenarios and samples, improving the wide applicability of wake electromagnetic field training data and the generalization ability of the model. Furthermore, this automated generation technology significantly reduces data acquisition time and cost while ensuring high data quality and consistency, further promoting the development and optimization of wake electromagnetic field detection models.

[0045] Secondly, this invention extracts the time-spectrum features of the wake electromagnetic field of an underwater vehicle using wavelet transform, effectively solving the feature representation problem of deep learning in wake electromagnetic field detection. Its advantages include: wavelet transform can simultaneously analyze the time and frequency information of the signal, thereby capturing subtle changes and local features in the wake electromagnetic field signal, making feature extraction more accurate and richer. Its multi-resolution analysis capability not only improves the deep learning model's ability to perceive and recognize complex wake electromagnetic field signals but also reduces reliance on traditional feature extraction methods. Furthermore, the features generated by wavelet transform enhance the robustness and stability of the model, contributing to improved accuracy and reliability of wake electromagnetic field detection.

[0046] Third, this invention successfully solves the problem of constructing a complex wake electromagnetic field detection model by training and optimizing a deep learning model using a large amount of calculation data of underwater vehicle wake electromagnetic fields.

[0047] This invention employs a novel deep learning model with powerful feature extraction capabilities, enabling it to automatically identify and capture complex patterns and subtle variations in wake electromagnetic fields, reducing reliance on traditional expert knowledge and manual feature engineering. The scalability and flexibility of deep learning methods allow the model to be more easily updated and extended to meet the demands and environmental changes in wake electromagnetic field detection. Attached Figure Description

[0048] Figure 1 The computational domain is set for the SUBOFF model of the underwater vehicle, where (a) is the top view and (b) is the side view;

[0049] Figure 2 This is a schematic diagram showing the thickness relationship of the first layer of mesh cells in the Y region.

[0050] Figure 3 This is a schematic diagram of the high Y+ wall surface treatment method;

[0051] Figure 4 Grid partition results for the SUBOFF model of underwater vehicle;

[0052] Figure 5 Illustration of integration domain: total integration domain (left) and local integration domain (right);

[0053] Figure 6 Illustration of linear interpolation for local integration domain;

[0054] Figure 7 Flowchart for calculation of wake electromagnetic field;

[0055] Figure 8 Illustration of simulation of electromagnetic sensor position swing;

[0056] Figure 9 Comparison results of pure noise electric field data (left) and signal-noise mixed electric field data (right);

[0057] Figure 10 2D-CNN deep neural network structure for wake electromagnetic field detection;

[0058] Figure 11 Comparison of signal detection evaluation results of three algorithms. DETAILED DESCRIPTION

[0059] The present application will be described in detail below in conjunction with the accompanying drawings and embodiments thereof, but the scope of protection of the present application is not limited to the range described in the embodiments.

[0060] In general, the 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics of the present application first obtains a large amount of data set that can be used for simulation training based on hydrodynamics and electromagnetism simulation, then extracts the time-frequency spectrum characteristics of the underwater vehicle wake electromagnetic field signal through wavelet transform, and then processes and analyzes it through 2D-CNN. Wavelet transform effectively captures the detailed changes of the signal at different times and frequencies, and 2D-CNN performs deep learning and pattern recognition on these time-frequency spectrum characteristics, thereby achieving accurate detection of the wake electromagnetic field. This method not only improves the accuracy of feature extraction, but also enhances the reliability and efficiency of signal recognition, providing a new technical means for underwater vehicle wake electromagnetic field detection.

[0061] The method of the present application includes three parts, which are: (1) generating underwater vehicle wake electromagnetic field calculation data set through multi-physical field coupling method of hydrodynamics and electromagnetism, (2) extracting time-frequency spectrum characteristics of wake electromagnetic field through wavelet transform, (3) developing wake electromagnetic field detection model based on 2D-CNN deep neural network, which will be described in detail by the applicant.

[0062] (1) The calculation dataset of the electromagnetic field of the underwater vehicle wake is generated by the multi-physical field coupling method of hydrodynamics and electromagnetism.

[0063] Principle description

[0064] The seawater environment contains a large number of charged particles. Influenced by the Coulomb force, any positively or negatively charged ion will attract several oppositely charged ions around it. When the seawater is in a balanced state, these ions are in an electrically neutral distribution. When an excitation is applied to the seawater (such as the wake formed by the underwater navigation of the underwater vehicle), the charged particles will produce local separation due to their own mass and friction coefficient, generate an electric field between the ion clusters, and form an ion polarization current.

[0065] Taking the ion polarization current as the research object, the force analysis of the ions in the seawater environment is first performed. The ions in the seawater are affected by various forces, and the forces that occupy the main factors are different according to different scenarios. However, in general, the force equation of the ions in the seawater environment can be expressed as follows:

[0066]

[0067] In formula (1), the subscript p represents the p type of ion, and there are seven terms on the right side of the equation from left to right, and their meanings are as follows:

[0068] 1) Electric field force: The electric field force E experienced by the ion with a charge of e p

[0069] 2) Magnetic field force: The Lorentz force experienced by the moving ion in the magnetic field B

[0070] 3) Friction force: The friction force experienced due to the viscosity of the solute, which can be described by the Debye-Huckel theory.

[0071] 4) Relaxation force: The force caused by the disturbance of the ion mobility due to the displacement of the ion atmosphere, which can be described by the Debye-Huckel-Onsanger theory.

[0072]

[0073] 5) Electrophoretic force: The disturbance of the viscous resistance caused by the different speeds of different ions, which is also described by the Debye-Huckel-Onsanger theory.

[0074] 6) Gradient force: The thermodynamic force described by the second law of Fickian diffusion.

[0075] 7) Pressure: The pressure experienced by the ion when moving from a high pressure area to a low pressure area.

[0076] In the formula, V p ​​, v0, p0 represent effective ion volume, fluid velocity and fluid density respectively.

[0077] The polarization current density J can be obtained by analyzing equation (1) as follows:

[0078]

[0079] wherein,

[0080]

[0081] By analyzing each current density in equation (2), it can be found that the first term is derived from the Lorentz term of the geomagnetic field, which can be considered as the ion induction polarization current density, and the second term is derived from the total differential of the disturbance velocity field itself, which is essentially different ion molar mass because of the polarization separation caused by inertial effect to form a polarization current. This term can be considered as the ion inertial polarization current density. The third term is the coupling term of the first two terms, and its order of magnitude is much smaller than the first two terms, so it can be ignored, so equation (3) is obtained:

[0082] J = σv0×B + aω (3)

[0083] In general, the ion polarization current caused by any external excitation disturbance velocity field contains ion induction and ion inertia terms. However, the disturbance velocity field caused by the wake flow of the underwater vehicle when it is sailing underwater has a low frequency, and the ion inertia term has a very weak contribution to the polarization current density, so the ion induction polarization current will occupy the main part.

[0084] The active Maxwell equation set is introduced, and the seawater is assumed to be a homogeneous medium, and equation (4) is obtained:

[0085]

[0086]

[0087] wherein denotes the curl operator, denotes the divergence operator, denotes the partial derivative with respect to time. H is the wake magnetic field strength, D is the electric displacement vector, E is the wake electric field strength, J is the ion polarization current density vector in seawater, and μ represents the seawater permeability. From the current density expression of equation (3), the magnetic induction strength B can be considered as the sum of the geomagnetic field B E and the wake magnetic field B i , that is:

[0088] J = σv0×(B E +B i ) (8)

[0089] Since B E >>Bi and the seawater does not contain displacement current, and the following constitutive relation is combined:

[0090] B = μH (9)

[0091] Then the Maxwell-Ampere loop formula in formula (7) can be changed to:

[0092]

[0093] The problem of the electromagnetic field of the wake is a low-frequency problem, and the phase difference of the electromagnetic field at different positions in the concerned range is almost zero, so the harmonic term can be ignored when solving, and according to formula (10), the relationship formula of the electromagnetic field of the wake can be obtained as:

[0094] E(r) = v0(r) x B(r) = v0(r) x (B E (r) + B i (r)) ≈ v0(r) x B E (r) (11)

[0095]

[0096] In formula (11) and formula (12), the position vector r' and the position vector r represent the position of the source point and the position of the to-be-measured point respectively.

[0097] When the seawater physical and chemical parameters and the geomagnetic region are given, only the wake velocity field v0 in formula (12) and formula (13) is an unknown parameter, and the discrete grid unit of the velocity field is divided, so that the discrete solution of the electromagnetic field of the wake can be completed.

[0098] Therefore, the solving of the electromagnetic field of the wake of the application is divided into two steps: the first step is to use the CFD fluid simulation software to solve the wake velocity field v0 by using the k-ε turbulence model (this step is completed by using general simulation software); the second step is to obtain the wake velocity field v0 as a parameter input, and the wake electric field E and the wake magnetic field B are obtained by numerical calculation.

[0099] The specific process of solving the electromagnetic field of the wake and obtaining the data set is as follows:

[0100] (1) Use CFD software to simulate fluid: including steps of geometric modeling, mesh division, simulation material attribute setting, setting boundary conditions and initial conditions, solver solving and the like, and the key steps are as follows:

[0101] 1. Geometric modeling and calculation domain setting

[0102] As shown in the attached Figure 1As shown, the calculation domain of the present application is a cuboid region, the underwater vehicle adopts SUBOFF model, the diving depth is h, the air layer height is 20D-h (D is the diameter of the SUBOFF model), the seawater layer height is 20D+h, the width of the calculation domain is 2L+2wavelength (L is the length of the SUBOFF model), and the length of the calculation domain is 7L+2wavelength. Figure 1 In (a), the calculation domain is divided into a solution domain and a damping zone, and the damping zone is used to prevent wave reflection at the upstream, downstream and side boundaries of the calculation domain in the simulation process. The width of the damping zone is 1.wavelength, and this length represents the maximum wavelength λ of the SUBOFF generated wave, which is determined by the dispersion relationship of linear surface waves in deep water conditions and can be calculated by the relationship between the wavelength λ and the SUBOFF model velocity U. It should be noted that the SUBOFF model is simulated in a uniform flow with a velocity of U, rather than the model itself moving forward at a velocity of U.

[0103] 2. Grid division

[0104] From the perspective of CFD fluid simulation calculation, grid division is a key step for whether the simulation calculation can converge, whether the calculation precision meets the requirements, and whether the calculation time is sufficiently short. For the SUBOFF model, the grid type selected by the present application is hexahedral grid, and in order to effectively simulate the turbulent flow near the outer wall surface of the SUBOFF model, the thickness X of the first layer of grid division near the wall surface is very important. In order to calculate it, the target Y + value of the problem needs to be known. + The Y + value is a dimensionless distance related to the selection of the turbulent flow model, which represents the length of the center of the first layer of grid cells from the wall surface (as shown in the accompanying drawings). When the Y + value is used for wall treatment, high Y + wall treatment is adopted, which uses relatively coarse grid to divide the wall surface, and uses standard wall function to process the viscous region near the wall surface, and the value is 30-500, as shown in the accompanying drawings. Figure 3

[0105] When Y ω,τ =100 is taken for grid division, the division result is as shown in the accompanying drawings. At this time, the body grid and the wall surface grid near the wall surface of the underwater vehicle are fine, and the grid size of the remaining part is relatively large, and a good balance between calculation precision and calculation speed is achieved. Figure 4

[0106] 3. Boundary condition setting and initial condition setting

[0107] The boundary condition setting is as follows:

[0108] ​​1) Velocity Inlet: The inflow end face, the top of the air layer and the bottom of the seawater layer are set as velocity inlet boundary conditions, corresponding to the SUBOFF speed.

[0109] 2) Pressure Outlet: The outflow end face is set as a pressure outlet boundary condition.

[0110] 3) Symmetry: The left and right sides of the calculation domain in the direction of the SUBOFF movement are set as symmetry boundary conditions, and it is considered that the flow field is in a symmetrical form.

[0111] 4) Wall: The SUBOFF model itself is set as a no-slip wall boundary condition.

[0112] The setting parameters of the initial conditions include the speed, underwater depth and size of the SUBOFF model.

[0113] 4, Solver solution

[0114] For the SUBOFF model, there are air-seawater two-phase media, the present application adopts VOF multiphase flow calculation, the turbulence model adopts k-ε turbulence model, the additional model selects gravity, VOF wave and unit mass repair, and the steady-state solution is simulated in the CFD software to obtain the velocity field v0 under the current simulation condition.

[0115] When the wake magnetic field is solved and calculated, it is assumed that the grid size of the wake velocity field to be calculated is N x N x M (N represents the horizontal plane grid size, and M represents the depth plane grid size), and a local grid of n x n x n size is selected in it, as shown in the accompanying Figure 5

[0116] For the position of any intermediate point P, according to the wake magnetic field calculation integral definition, it should be discretized and summed in the full integral domain, but due to the attenuation characteristics, only a local region around the position of the P point is selected for discrete summation to obtain the approximate value of the wake magnetic field. For each point in the integral range (each discrete point in the selected local region), the result v0 of the velocity field is brought into formula (12) to obtain the magnetic field of the corresponding point (considering the boundary effect, it can be iterated for several times), and the electric field of the corresponding point is calculated by using formula (11).

[0117] Optionally, in addition to the above operation, the current density of the grid closest to the intermediate point P is interpolated to further improve the calculation accuracy of the wake electric field, as shown in the accompanying Figure 6 Figure 7 Based on the above calculation idea, a flow chart for calculating the wake induced electromagnetic field is given, as shown in the accompanying

[0118] ​​The input condition or model parameter of the simulation process is changed, and the above process is repeated to obtain electromagnetic field data under different conditions.

[0119] (II) Extracting time-frequency spectrum features of the wake electromagnetic field by wavelet transform

[0120] From the entire measurement area, select several measurement points to obtain the data (or change function) of the electric and magnetic fields changing with time at each measurement point, and perform wavelet transform on the obtained electromagnetic field data. Here, different measurement point positions can further expand the electromagnetic field data set.

[0121] The wavelet transform retains the advantages of the short-time Fourier transform while overcoming the defect that the time resolution and frequency resolution cannot be improved simultaneously. It uses a wavelet basis function to represent the signal, and by scaling and translating the wavelet basis function, it can match the input signal, which can be expressed as follows:

[0122]

[0123] where Ψ(t) represents the wavelet basis function, also known as the mother wavelet, and the Morse wavelet basis function is used in the present application; Ψ ω,τ (t) represents a sub-wavelet generated by scaling and translating the wavelet basis function; W f (ω,τ) is the result of wavelet transform of the signal f(t). For a time-sampled signal, it contains time-domain information and frequency-domain information. The present application uses wavelet method to extract the time-frequency information of the wake electromagnetic field.

[0124] As shown in the accompanying Figure 8 , N simulated electromagnetic sensors are placed in front of the underwater vehicle and all on one side, with a spacing of 10m between adjacent sensors.

[0125] As shown in the accompanying Figure 8 , the electric field signal of S1 measurement point is taken as an example, the sampling rate is 3Hz, and the sampling time is 146s. The data of S1 is superimposed with Gaussian white noise, and the signal-to-noise ratio is -5dB. 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, as shown in the accompanying Figure 9 . It can be found that when the signal-to-noise ratio is -5dB, the peak-to-peak value and waveform of the pure noise electric field data and the signal-to-noise mixed electric field data are basically the same, and at this time the target signal cannot be distinguished. But the time-frequency spectrum diagram of the wavelet transform can extract the frequency point features and duration of the target signal.

[0126] (III) Developing a wake electromagnetic field detection model based on 2D-CNN deep neural network The 2D-CNN deep neural network structure designed by the present application is as shown in the accompanying Figure 10As shown, two modules including a convolutional layer and a full connection classifier layer are contained, wherein the convolutional module contains 7 convolutional layers and RELU layers, 3 batch normalization layers, 4 pooling layers, etc., the full connection classifier module contains 1 flattening layer, 3 full connection layers, 2 RELU layers, and adopts a Dropout processing.

[0127] The target data is generated in a manner that data of a plurality of electromagnetic field sensors is selected, a signal-to-noise ratio is set to -10-0 dB with a step of 1 dB, and the total length of the time domain signal is 440, of which the length of the target signal is 195. In order to make the signal detection more in line with the actual situation and increase the diversity of the characteristics of the data set, the target signal is randomly generated in the whole signal.

[0128] Then, the model is trained by using the training set. In the training set setting, for a given electromagnetic sensor and signal-to-noise ratio, there are a total of 200 data, and the target signal data and the noise data each account for half. For example, for the signal-to-noise ratio of -5 dB, the data of the S1 electric field sensor has a total of 200, of which 100 have target signals, and the other 100 do not have target signals. Therefore, the training set data has a total of 22000 time domain data. Wavelet transform is performed on the 22000 time domain data to obtain 22000 two-dimensional time-frequency spectrograms. Similarly, the test set has 4400 time-frequency spectrograms.

[0129] The loss function adopts a cross-entropy function during training, the optimization method adopts Adam, the learning rate lr is 0.0001, the batch size batchsize is 44, and the training round epoch is 20.

[0130] The trained model is used for electromagnetic field wake detection test.

[0131] The deep neural network proposed in the application is compared with other neural networks, including FCN (full connection neural network) and 1D-CNN (one-dimensional convolutional neural network), and the results are shown in the accompanying Figure 11 As shown, it can be found that the three methods all give alarm indications in the wake electromagnetic field signal time period, indicating that the wake electromagnetic field signal is detected. However, in some pure noise time periods, the latter two algorithms appear false alarm. This shows that the deep neural network proposed in the application is superior to the FCN neural network and the 1D-CNN neural network in detection performance.

[0132] Unlike 1D-CNN, 2D-CNN is aimed at images, while 1D-CNN is aimed at curves; the structure of the application is 2D-CNN, and the results of the FCN model are different. A typical 2D-CNN model includes a convolutional layer, a pooling layer and a full connection layer, while the FCN removes the full connection layer and the global average pooling layer in the 2D-CNN, and introduces a transposed convolutional layer.

[0133] In summary, the 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics can effectively improve the detection capability of the wake electromagnetic field, provides a new technical means for the wake electromagnetic field detection of the underwater vehicle, and has a good application prospect in the field of ocean target electromagnetic field monitoring and early warning.

[0134] Although the principles of the present application have been described in detail above with reference to the preferred embodiments thereof, it is to be understood that the above-described embodiments are merely illustrative of the present application and are not intended to limit the scope of the present application. The details in the embodiments do not constitute a limitation on the scope of the present application, and any equivalent changes, simple replacements, etc. based on the technical solutions of the present application, without departing from the spirit and scope of the present application, fall within the protection scope of the present application.

Claims

1. A 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics, characterized in that, The method comprises: Step (1) generating a submarine wake electromagnetic field calculation dataset by a multi-physical field coupling method of hydrodynamics and electromagnetism, This step (1) comprises: (1.1) constructing a geometric model in a dynamic field simulation software, and setting a calculation domain of the geometric model; (1.2) performing mesh division on the set calculation domain; (1.3) setting boundary conditions for the geometric model, wherein the boundary conditions at least comprise a velocity inlet, a pressure outlet, a symmetry surface and a wall surface; (1.4) solving a velocity field distribution in the calculation domain in the dynamic field simulation software; (1.5) based on the obtained velocity field distribution, calculating an electric field and a magnetic field of any point in the calculation domain according to the following formula: E(r) = v0(r) x B(r) = v0(r) x (B E (r) + B i (r)) ≈ v0(r) x B E (r)(1) In formula (1) and formula (2), the position vector r' and the position vector r respectively represent the position of the source point and the position of the point to be measured, wherein J(r') represents the current density at the position r', V represents the integral volume, v0 represents the water flow velocity of the point to be measured obtained according to the velocity field, μ represents the seawater permeability, σ represents the seawater conductivity, B E represents the geomagnetic field; (1.6) modifying input parameters and model parameters, repeating steps (1.3)-(1.5) to obtain electric field and magnetic field distributions under different input parameters and model parameters, obtaining multiple groups of electric field and magnetic field data to form a dataset; Step (2) taking the entire calculation domain as a measurement area, selecting a plurality of measurement points from the measurement area, obtaining electric and magnetic field data of each measurement point under different conditions from the dataset, performing wavelet transform on the obtained electric and magnetic field data to obtain a wavelet-transformed dataset; Step (3) constructing a 2D-CNN deep neural network, dividing the constructed dataset into a training set and a test set, inputting data in the training set into the 2D-CNN deep neural network, training the 2D-CNN deep neural network and testing it by using the test set, obtaining the trained 2D-CNN deep neural network, and using the obtained trained 2D-CNN deep neural network for wake electromagnetic field detection.

2. The 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics according to claim 1, characterized in that The dynamic field simulation software is a CFD software.

3. The 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics according to claim 1, characterized in that The calculation domain is a cuboid region, and a SUBOFF model is arranged in the calculation domain to simulate an underwater vehicle.

4. The 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics according to claim 1, characterized in that The current density J is calculated by the following formula: J = σv0 x B E ; where σ represents the seawater conductivity, v0represents the water flow velocity of the point to be measured obtained according to the velocity field, B E represents the geomagnetic field.

5. The 2D-CNN network wake electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics according to claim 1, characterized in that The calculation domain uses a local grid partitioning and interpolation method to calculate the wake electromagnetic field.

6. The wavelet transform time-frequency spectrum feature based 2D-CNN network wake electromagnetic field detection method according to claim 1, characterized in that, The 2D-CNN deep neural network comprises a convolution module and a full connector module, the convolution module comprises 7 convolution layers and RELU layers, 3 batch normalization layers and 4 pooling layers, and the full connection classifier module comprises 1 flattening layer, 3 full connection layers and 2 RELU layers.

7. A dataset acquisition method for training of a wake electromagnetic field detection model, characterized in that The method comprises: (1.1) constructing a geometric model in a dynamic field simulation software, and setting a calculation domain of the geometric model; (1.2) In the constructed geometry model, mesh the computational domain; (1.3) Set boundary conditions for the geometry model, the boundary conditions at least include velocity inlet, pressure outlet, symmetry plane and wall surface; (1.4) Solve the velocity field distribution in the computational domain in the dynamic field simulation software; (1.5) Based on the obtained velocity field distribution, calculate the electric field and magnetic field at any point in the computational domain according to the following formula: E(r) = v0(r) x B(r) = v0(r) x (B E (r) + B i (r)) ≈ v0(r) x B E (r) (4) In formula (4) and formula (5), the position vector r' and the position vector r respectively represent the position of the source point and the position of the point to be measured, wherein J(r') represents the current density at the position r', V represents the volume, v0 represents the velocity obtained according to the velocity field, μ represents the magnetic permeability of seawater, σ represents the electrical conductivity of seawater, B E represents the geomagnetic field; (1.6) Repeat the above steps (1.3)-(1.5) to modify the input parameters and model parameters, obtain the electric field and magnetic field distribution under different input parameters and model parameters, obtain multiple sets of electric field and magnetic field data, and constitute a data set.

Images