Magnetic target state estimation system and device

Abstract

The invention provides a magnetic target state estimation system and device, and relates to the technical field of magnetic target state estimation, and the system mainly comprises a data collection module, a magnetic field model construction module, a deep learning module and a result output module. The data acquisition module comprises a field measurement unit and a data set unit; the magnetic field model building module is used for enabling a magnetic target to be equivalent to a physical model; the physical model is a mixed model of an ellipsoid and a magnetic dipole array; the deep learning module comprises a neural network, the physical model and an optimizer; and the result output module is used for sending out the state prediction result of the magnetic target after the neural network training is completed. According to the scheme, the complex magnetic field data is efficiently fitted through the neural network, and the physical constraint is introduced into the loss function, so that the robustness is improved, and the overfitting problem in simple data fitting is avoided; the calculation time and resources can be greatly reduced; and the solution quality and stability can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic target state estimation, and particularly to a magnetic target state estimation system and device. Background Art

[0002] Magnetic target state estimation refers to the process of modeling a magnetic target as various equivalent magnetic source models such as a magnetic dipole, a rotating ellipsoid, a magnetic dipole array, a mixed model of magnetic dipole and ellipsoid, etc., and performing analytical inversion based on the magnetic signals generated by it in the background geomagnetic field that change with time and space, and finally obtaining its motion state and magnetic parameters.

[0003] Existing magnetic target state estimation methods include analytical or numerical methods such as magnetic gradient tensor analysis method, nonlinear filtering method, inversion method based on parameter optimization, conjugate gradient least squares method, etc. Among them, the conjugate gradient least squares method mainly includes steps such as data acquisition, magnetic field model construction, conjugate gradient least squares method calculation, and result output; in the magnetic field model construction, a magnetic dipole array model is often used. When determining the number of magnetic dipoles, the larger the target scale, the more magnetic dipoles are used: for ship targets, magnetic dipoles are usually evenly distributed on the waterline, and the number is at least 5 to simulate the local non-uniform magnetic field; for small vehicle targets, 2 to 3 magnetic dipoles are often required; for large vehicle targets, 4 magnetic dipoles are usually required.

[0004] Most of these methods are based on simple assumption models. Although in some simple scenarios, these assumption models can obtain accurate results, when facing complex targets (such as irregular shapes, heterogeneous magnetic materials, or multi-magnetic dipole arrays, etc.), these assumption models will have problems of insufficient adaptability, specifically including: not adapting to targets with complex geometric shapes; difficult to model non-uniform magnetic distributions, with large errors in magnetic moment estimation; unable to effectively represent complex magnetic behaviors such as non-linear magnetic moment distributions.

[0005] At the same time, the inversion method usually depends on high-quality magnetic field measurement data, but in practical applications, magnetic field measurement data is often affected by various noises, such as sensor noise, electromagnetic interference, and environmental noise, resulting in various problems, specifically including: noise amplification affecting the solution accuracy; noise causing the optimization algorithm to converge to a local optimal solution, affecting the solution robustness; data cleaning and noise processing will also introduce errors, affecting the reliability of the inversion result.

[0006] Furthermore, these methods usually require repeated numerical calculations. Especially when dealing with large-scale data and complex models, the calculation efficiency becomes a bottleneck, and the specific problems include: huge computational overhead in high-dimensional parameter space; low efficiency of multiple iterative optimizations; unable to utilize hardware acceleration.

[0007] In addition, the inversion method, such as the gradient descent method, usually depends on a suitable initial value. Selecting an inappropriate initial value will cause the inversion process to fall into a local optimal solution and fail to obtain the global optimal solution. Specific problems include: being highly sensitive to the initial guess, and in a high-dimensional parameter space, the sensitivity is even more obvious, which may even lead to inversion failure; when facing high-dimensional problems, it becomes more difficult to select the initial value of the optimization process, thus greatly increasing the uncertainty and difficulty of the solution process; usually relying on manual selection of the initial value, lacking an adaptive adjustment mechanism, it is difficult to directly obtain a reasonable initial magnetic moment, resulting in a decrease in the efficiency of the entire inversion process or even failure. Summary of the Invention

[0008] The purpose of the present invention is to provide a magnetic target state estimation system and device to solve at least one of the above technical problems existing in the prior art.

[0009] In a first aspect, to solve the above technical problems, the present invention provides a magnetic target state estimation system, including a data acquisition module, a magnetic field model construction module, a deep learning module, and a result output module; The data acquisition module includes a field measurement unit and a dataset unit; The field measurement unit is used to measure the on-site magnetic field data of the magnetic target and perform data preprocessing; The dataset unit is used to receive the past magnetic field dataset of the magnetic target and perform dataset preprocessing; The magnetic field model construction module is used to equivalent the magnetic target to a physical model; the physical model is a hybrid model of an ellipsoid and a magnetic dipole array: the ellipsoid is located at the center of the magnetic target, the major axis is equal to the length of the magnetic target, and the minor axis is equal to the width of the magnetic target, which is used to fit the macroscopic magnetic field of the magnetic target; the magnetic dipole array is uniformly distributed on the preset contour line of the magnetic target (such as the waterline of a ship), which is used to simulate the local non-uniform magnetic field of the magnetic target; The deep learning module includes a neural network, the physical model, and an optimizer; The neural network includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the on-site magnetic field data and send it to the hidden layer; The hidden layer includes multiple fully connected layers. Each fully connected layer processes the input data through a non-linear activation function (Tanh), extracts features layer by layer, and learns the implicit non-linear relationship between the magnetic field distribution and the magnetic moment of the magnetic target; in this way, by adjusting the depth and width of the neural network, it is possible to adapt to the magnetic field inversion tasks of magnetic targets of different scales; The output layer is used to output the state prediction result of the magnetic target, including magnetic moment parameters and motion state parameters; The physical model receives the state prediction result, calculates the magnetic field distribution based on the magnetic moment parameters therein, and compares it with the true magnetic field distribution obtained from the past magnetic field dataset to generate a total loss function; The optimizer updates the parameters of the neural network by minimizing the total loss function; The result output module is used to send out the state prediction result of the magnetic target after the neural network training is completed.

[0010] In a feasible implementation manner, the data preprocessing includes signal interception and downsampling, etc.

[0011] In a feasible implementation manner, the dataset preprocessing includes: performing normalization processing on the dataset to reduce the dimensional difference between different features, making the subsequent neural network training more stable and accelerating the convergence speed; performing feature enhancement processing on the dataset to extract spatial features for better capturing of magnetic field changes; expanding the data sample size through dimensionality reduction (PCA) and data augmentation (simulation generation, coordinate rotation, etc.); performing interpolation and alignment processing on sparse measurement points to normalize all samples to the same spatial coordinates or time steps, thereby ensuring a unified input format for the neural network; in this way, the neural network can have stronger anti-noise ability, higher computational efficiency, and better solution robustness.

[0012] In a feasible implementation manner, the magnetic field data sequence of the physical model is specifically expressed as:

[0013] Among them, represents the magnetic field component (magnetic induction intensity) in the direction, represents the magnetic field component in the direction; represents the magnetic moment of the th magnetic dipole; represents the ellipsoidal magnetic moment; represents the number of magnetic dipoles; , and all represent coefficients, and the specific formula includes: When it is the case, then there is: ; Among them, represents the vacuum magnetic permeability; represents the coordinates of the magnetic sensor relative to the th magnetic dipole in the carrier coordinate system, specifically expressed as ; among them, represents the coordinates of the magnetic sensor in the carrier coordinate system, represents the -th magnetic dipole's coordinates in the carrier coordinate system, specifically represented as ; where represents the magnetic dipole spacing, specifically represented as ; where represents the length of the magnetic target; , and respectively represent the coordinates of the -th magnetic dipole in the sensor coordinate system; the carrier coordinate system refers to the coordinate system established with the center of the magnetic target as the origin; the sensor coordinate system refers to the coordinate system established with the center of the sensor as the origin; When , then there is: ; where represents the first simplified parameter, represents the second simplified parameter, represents the third simplified parameter, used to simplify the formula; represents the semi-focal length; represents the width of the magnetic target.

[0014] In a feasible implementation manner, the motion state parameters include position, velocity, heading, length, width, etc.

[0015] In a feasible implementation manner, the total loss function includes a first loss function ( ) and a second loss function ( ); The first loss function is used to measure the error between the magnetic moment parameters predicted by the neural network and the magnetic field distribution calculated by the physical model. The specific measurement process includes: inputting the magnetic moment parameters predicted by the neural network into the physical model to calculate the magnetic field distribution; comparing the magnetic field distribution with the actual measurement or the true magnetic field distribution to evaluate the error; the error is measured by the mean square error (MSE) or other distance metrics; in this way, physical constraints can be introduced, so that the loss function can reflect the consistency between the magnetic moment of the magnetic target and the actual magnetic field distribution, and further guide the neural network to learn solutions that conform to physical laws; The second loss function is used to calculate the angular error between the magnetic moment parameters predicted by the current neural network and the true magnetic moment direction; in this way, by directly fitting the input features and labels in the past magnetic field dataset, it can be ensured that the neural network can better fit the training data and improve the generalization ability; The weight of the first loss function and the weights of the second loss function , which is used to control the contributions of the two loss functions to the total loss function.

[0016] In a feasible implementation manner, the specific working process of the deep learning module includes: Step a1, forward propagation: Input data is output through a neural network to obtain a state prediction result; Step a2, loss calculation: Based on and , calculate the error between the current state prediction result and the actual data, and combine and to obtain the total loss value; Step a3, backpropagation: Calculate the gradient of the total loss function with respect to the network weights through the backpropagation algorithm; Step a4, parameter update: Based on the gradient, adjust the network weights through an optimizer to minimize the total loss function; Through the above joint optimization method, the neural network can find the best trade-off point between physical consistency and data fitting, achieving the purpose of efficiently and accurately solving the magnetic moment of complex magnetic targets.

[0017] In a feasible implementation manner, the optimizer is an Adam optimizer.

[0018] In a second aspect, based on the same inventive concept, the present application further provides a magnetic target state estimation device, including a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to implement the magnetic target state estimation system as described above. The bus is connected between the functional components to transmit information.

[0019] In a feasible implementation manner, the magnetic target state estimation device further includes a three-axis magnetic sensor for measuring the magnetic field data of the magnetic target, that is, a set of magnetic field components in time or space series.

[0020] Adopting the above technical solutions, the present invention has the following beneficial effects: A magnetic target state estimation system and device provided by the present invention efficiently fits complex magnetic field data through a neural network and introduces physical constraints into the loss function, thereby improving the robustness of the system; the system uses a hybrid model of an ellipsoid and a magnetic dipole array, and at the same time utilizes the non-linear expression ability of the neural network to more accurately characterize the magnetic moment distribution of complex targets; the system can directly incorporate physical laws into the loss function, such as magnetic field equations, etc., so that the inversion process not only depends on data but also follows physical laws, avoiding the defects of traditional optimization algorithms (genetic algorithm, particle swarm optimization, simulated annealing, etc.) and pure data-driven methods based on deep neural networks (convolutional neural network, recurrent neural network, etc.) lacking physical constraints; this guiding mechanism enables the neural network to better resist noise during the training process and avoid the overfitting problem in simple data fitting; This solution is based on a deep learning framework, and the model training process can be accelerated by hardware such as GPUs, greatly improving the computing efficiency. Especially by utilizing the parallel computing characteristics of the neural network, when dealing with large-scale data sets, the training process can be completed quickly; through advanced optimization algorithms (such as Adam, L-BFGS, etc.) and the backpropagation mechanism, this solution can converge to the global optimal solution in fewer iterations, significantly reducing the computing time and resources required in traditional optimization methods (genetic algorithm, particle swarm optimization, simulated annealing, etc.), models based on variational autoencoders (VAEs) and generative adversarial networks (GANs), etc.; The network training process of this solution can gradually approach the global optimal solution, avoiding the limitation on the selection of initial values; through the training process of the deep neural network, the loss function is gradually optimized, which can avoid falling into local optima and gradually approach the global optimal solution. Even if the initial value is not accurately selected, the network can obtain a solution close to the true magnetic moment through iterative learning. At the same time, by adaptively adjusting the weights and biases through training data, it does not depend on preset initial values and can automatically adjust network parameters according to data characteristics, thereby improving the quality and stability of the solution. Description of the Drawings

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic diagram showing the working principle of a magnetic target state estimation system provided by an embodiment of the present invention; Figure 2 It is a schematic diagram showing the working principle of a deep learning module provided by an embodiment of the present invention; Figure 3 A diagram for explaining the coordinate system provided by an embodiment of the present invention; Figure 4 A comparison diagram of the x-direction magnetic moment provided by an embodiment of the present invention; Figure 5 A comparison diagram of the y-direction magnetic moment provided by an embodiment of the present invention; Figure 6 A comparison diagram of the z-direction magnetic moment provided by an embodiment of the present invention. Detailed implementation manners

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0025] In the description of the present invention, it should be noted that unless otherwise clearly specified and defined, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0026] The present invention will be further explained and described below in combination with specific implementation manners.

[0027] It should also be noted that the following specific embodiments or specific implementation manners are a series of optimized setting manners listed by the present invention to further explain the specific invention content, and these setting manners can be combined with each other or used in association with each other.

[0028] Embodiment 1: As Figure 1As shown in the figure, this embodiment provides a magnetic target state estimation system, including a data acquisition module, a magnetic field model construction module, a deep learning module, and a result output module; The data acquisition module includes a field measurement unit and a data set unit; The field measurement unit is used to measure the on-site magnetic field data of the magnetic target and perform data preprocessing; the magnetic target is a ship; The data set unit is used to receive the past magnetic field data set of the magnetic target and perform data set preprocessing; The magnetic field model construction module is used to equivalent the magnetic target to a physical model; the physical model is a hybrid model of an ellipsoid and a magnetic dipole array: the ellipsoid is located at the center of the magnetic target, the major axis is equal to the length of the magnetic target, and the minor axis is equal to the width of the magnetic target, which is used to fit the macroscopic magnetic field of the magnetic target; the magnetic dipole array is evenly distributed on the waterline of the ship and is used to simulate the local non-uniform magnetic field of the magnetic target; The deep learning module, as Figure 2 shown, includes a neural network, the physical model, and an optimizer; The neural network includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the on-site magnetic field data and send it to the hidden layer; The hidden layer includes multiple fully connected layers. Each fully connected layer processes the input data through the Tanh function, extracts features layer by layer, and learns the implicit non-linear relationship between the magnetic field distribution and the magnetic moment of the magnetic target; in this way, by adjusting the depth and width of the neural network, it is possible to adapt to the magnetic field inversion tasks of magnetic targets of different scales; The output layer is used to output the state prediction result of the magnetic target, including magnetic moment parameters and motion state parameters; The physical model receives the state prediction result, calculates the magnetic field distribution according to the magnetic moment parameters therein, and compares it with the true magnetic field distribution obtained from the past magnetic field data set to generate a total loss function; The optimizer updates the parameters of the neural network by minimizing the total loss function; The result output module is used to send out the state prediction result of the magnetic target after the neural network training is completed.

[0029] Furthermore, the data preprocessing includes conventional methods such as signal interception and downsampling.

[0030] Furthermore, the preprocessing of the dataset includes: normalizing the dataset to reduce the dimensional difference between different features, making the subsequent neural network training more stable and accelerating the convergence speed; performing feature enhancement processing on the dataset to extract spatial features, such as calculating the magnetic field gradient (first derivative, second derivative, which can be calculated through the automatic differentiation mechanism of Python), frequency domain components (extracting frequency components by performing Fourier transform on the magnetic field signal), etc., in order to better capture the magnetic field changes; expanding the data sample size through dimensionality reduction (PCA) and data augmentation (simulation generation, coordinate rotation, etc.); performing interpolation and alignment processing on sparse measurement points to normalize all samples to the same spatial coordinates or time steps, thereby ensuring the unified input format of the neural network; in this way, the neural network can have stronger anti-noise ability, higher computational efficiency, and better solution robustness.

[0031] Furthermore, the magnetic field data sequence of the physical model is specifically expressed as:

[0032] Among them, represents the magnetic field component in the direction (magnetic induction intensity), represents the magnetic field component in the direction; represents the th magnetic dipole moment; represents the magnetic moment of the ellipsoid; represents the number of magnetic dipoles; , and all represent coefficients, and the specific formulas include: When , then there is: ; Among them, represents the vacuum permeability; represents the coordinates of the magnetic sensor relative to the th magnetic dipole in the carrier coordinate system, specifically expressed as ; among them, represents the coordinates of the magnetic sensor in the carrier coordinate system, represents the th magnetic dipole in the carrier coordinate system, specifically expressed as ; among them, represents the magnetic dipole spacing, specifically expressed as ; among them, represents the length of the magnetic target; , and respectively represent the coordinates of the th magnetic dipole in the sensor coordinate system; the carrier coordinate system refers to the coordinate system established with the center of the magnetic target as the origin; the sensor coordinate system refers to the coordinate system established with the center of the sensor as the origin, as shown in Figure 3 ; When , then there is: ; Among them, represents the first simplified parameter; represents the second simplified parameter; represents the third simplified parameter; represents the semi-focal length; represents the width of the magnetic target.

[0033] Furthermore, the motion state parameters include position, velocity, heading, length (ship length), and width (ship width), etc.

[0034] Furthermore, the total loss function includes a first loss function ( ) and a second loss function ( ); The first loss function is used to measure the error between the magnetic moment parameters predicted by the neural network and the magnetic field distribution calculated by the physical model. The specific measurement process includes: inputting the magnetic moment parameters predicted by the neural network into the physical model to calculate the magnetic field distribution; comparing the magnetic field distribution with the actual measurement or the true magnetic field distribution to evaluate the error; the error is measured by the mean square error (MSE) or other distance metrics; in this way, physical constraints can be introduced, so that the loss function can reflect the consistency between the magnetic moment of the magnetic target and the actual magnetic field distribution, and further guide the neural network to learn solutions that conform to physical laws; the specific formula of the first loss function can be: ; Among them, represents the predicted value of the th magnetic dipole's magnetic moment in the direction; represents the predicted value of the th magnetic dipole's magnetic moment in the direction; represents the predicted value of the th magnetic dipole's magnetic moment in the direction; represents the predicted value of the Predicted value of magnetic field component; denote Predicted value of magnetic field component; Figure 2 in denotes the transformation matrix between the vehicle coordinate system and the sensor coordinate system; The second loss function is used to calculate the angular error (such as cosine similarity or angular difference) between the magnetic moment parameters predicted by the current neural network and the true magnetic moment direction; in this way, by directly fitting the input features and labels in the past magnetic field dataset, it can be ensured that the neural network can better fit the training data and improve the generalization ability; The weight of the first loss function and the weight of the second loss function are used to control the contributions of the two loss functions to the total loss function.

[0035] Furthermore, as Figure 2 shown, the specific working process of the deep learning module includes: Step a1, forward propagation: Pass the input data through the neural network to output the state prediction result; Step a2, loss calculation: Based on and , calculate the error between the current state prediction result and the actual data, and combine and to obtain the total loss value (weighted summation, that is * + * ); Step a3, backpropagation: Calculate the gradient of the total loss function with respect to the network weights through the backpropagation algorithm; Step a4, parameter update: Based on the gradient, adjust the network weights through the optimizer to minimize the total loss function; Through the above joint optimization method, the neural network can find the best trade-off point between physical consistency and data fitting, achieving the purpose of efficiently and accurately solving the magnetic moment of complex magnetic targets.

[0036] Furthermore, the optimizer is Adam.

[0037] Exemplarily, taking a certain ship as a magnetic target, through this solution, the on-site magnetic field data with noise of this magnetic target is collected in real time as the input of this system, and the magnetic moment (predicted value) in the state prediction result of this magnetic target is compared with the true magnetic moment (true value): The line graph comparison of the magnetic moment in the x direction, as Figure 4As shown; it can be seen from this figure that the predicted value line and the true value line basically coincide, with only slight errors at the 3rd sample and the 12th sample; Line graph comparison of the y-direction magnetic moment, as Figure 5 shown; it can be seen from this figure that the predicted value line and the true value line basically coincide, with only slight errors at the 4th sample and the 7th sample; Line graph comparison of the z-direction magnetic moment, as Figure 6 shown; it can be seen from this figure that the predicted value line and the true value line basically coincide, with only a slight error at the 15th sample; After calculation, the overall average error is only 0.3%; thus, it can be seen that the state estimation effect of this system is excellent.

[0038] Embodiment 2: This embodiment provides a magnetic target state estimation device, including a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to implement the magnetic target state estimation system as described above. The bus is connected between each functional component for transmitting information.

[0039] Furthermore, the magnetic target state estimation device further includes a three-axis magnetic sensor for measuring the magnetic field data of the magnetic target, that is, a set of magnetic field components in a time or space sequence.

[0040] In another implementation manner of this solution, it can be implemented in the form of an integrated device, and this device can include corresponding modules that execute each or several steps in the above various embodiments. The module can be one or more hardware modules specifically configured to execute the corresponding steps, or implemented by a processor configured to execute the corresponding steps, or stored in a computer-readable medium for implementation by the processor, or implemented through a certain combination.

[0041] The processor executes the various methods and processes described above. For example, the method implementation in this solution can be implemented as a software program, which is tangibly included in a machine-readable medium, such as a memory. In some embodiments, part or all of the software program can be loaded and / or installed via the memory and / or the communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps in the method described above can be executed. Alternatively, in other embodiments, the processor can be configured to execute one of the above methods in any other appropriate manner (for example, by means of firmware).

[0042] The device can be implemented using a bus architecture. The bus architecture can include any number of interconnected buses and bridges, depending on the specific application of the hardware and the overall design constraints. The bus connects together various circuits of one or more processors, memories, and / or hardware modules. The bus can also connect various other circuits such as peripheral devices, voltage regulators, power management circuits, external antennas, etc.

[0043] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Component (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A magnetic target state estimation system, characterized in that: It includes data acquisition module, magnetic field model building module, deep learning module and result output module; The data acquisition module includes a field measurement unit and a data set unit; The field measurement unit is used to measure the field magnetic field data of the magnetic target and perform data preprocessing; The data set unit is used to receive past magnetic field data sets of magnetic targets and perform data set preprocessing; The magnetic field model building module is used to convert the magnetic target into a physical model; the physical model is a hybrid model of an ellipsoid and a magnetic dipole array: the ellipsoid is located at the center of the magnetic target, the major axis is equal to the length of the magnetic target, and the minor axis is equal to the width of the magnetic target, which is used to fit the macroscopic magnetic field of the magnetic target; the magnetic dipole array is evenly distributed on the preset contour line of the magnetic target, which is used to simulate the local inhomogeneous magnetic field of the magnetic target; The deep learning module includes a neural network, the physical model and an optimizer; The neural network includes an input layer, a hidden layer and an output layer; The input layer is used to receive the field magnetic field data and send it to the hidden layer; The hidden layer includes a plurality of fully connected layers, each of which processes input data through a nonlinear activation function, extracts features layer by layer, and learns the implicit nonlinear relationship between the magnetic field distribution and the magnetic moment of the magnetic target; The output layer is used to output the state prediction results of the magnetic target, including magnetic moment parameters and motion state parameters; The physical model receives the state prediction result, calculates the magnetic field distribution according to the magnetic moment parameters therein, and compares it with the real magnetic field distribution obtained according to the past magnetic field data set to generate a total loss function; The optimizer updates the parameters of the neural network by minimizing the total loss function; The result output module is used to send out the state prediction result of the magnetic target after the neural network training is completed.

2. The system according to claim 1, characterized in that The data preprocessing includes signal interception and down sampling.

3. The system according to claim 1, characterized in that The data set preprocessing includes: standardizing the data set to reduce the dimensional differences between different features; performing feature enhancement processing on the data set to extract spatial features; expanding the data sample size through dimensionality reduction and data enhancement; interpolating and aligning sparse measurement points to normalize all samples to the same spatial coordinates or time steps.

4. The system according to claim 1, characterized in that The magnetic field data sequence of the physical model is specifically expressed as: in, express The magnetic field component, express The magnetic field component, express Magnetic field component; Indicates The magnetic moment of a magnetic dipole; represents the ellipsoidal magnetic moment; represents the number of magnetic dipoles; , and All represent coefficients, and the specific formulas include: when When , we have: ； in, represents the vacuum permeability; Indicates the magnetic sensor relative to the first The coordinates of the magnetic dipoles are expressed as ;in, represents the coordinates of the magnetic sensor in the carrier coordinate system, Indicates The coordinates of a magnetic dipole in the carrier coordinate system are specifically expressed as ;in, represents the magnetic dipole spacing, which is specifically expressed as ;in, Indicates the magnetic target length; , and Respectively represent The coordinates of a magnetic dipole in the sensor coordinate system; the carrier coordinate system refers to a coordinate system established with the center of the magnetic target as the origin; the sensor coordinate system refers to a coordinate system established with the center of the sensor as the origin; when When , we have: ； in, represents the first simplified parameter; represents the second simplified parameter; represents the third simplified parameter; represents the half focal length; Indicates the magnetic target width.

5. The system according to claim 1, characterized in that The motion state parameters include position, speed, heading, length and width.

6. The system according to claim 1, characterized in that The total loss function includes a first loss function and a second loss function; The first loss function is used to measure the error between the magnetic moment parameters predicted by the neural network and the magnetic field distribution calculated by the physical model. The specific measurement process includes: inputting the magnetic moment parameters predicted by the neural network into the physical model to calculate the magnetic field distribution; comparing the magnetic field distribution with the actual measurement or real magnetic field distribution to evaluate the error; the error is measured by the mean square error; the specific formula is: ； in, Indicates A magnetic dipole Prediction of magnetic moment; Indicates A magnetic dipole Prediction of magnetic moment; Indicates A magnetic dipole Prediction of magnetic moment; Indicates the heading angle; express Predicted value of magnetic field component; express Predicted value of magnetic field component; express Predicted value of magnetic field component; The second loss function is used to calculate the angle error between the magnetic moment parameter predicted by the current neural network and the true magnetic moment direction; The weight of the first loss function and the weights of the second loss function , which is used to control the contribution of the two loss functions to the total loss function.

7. The system according to claim 6, characterized in that The specific workflow of the deep learning module includes: Step a1, forward propagation: input data, pass it through the neural network, and output the state prediction result; Step a2, loss calculation: Based on the first loss function and the second loss function, calculate the error between the current state prediction result and the actual data, and combine and , find the total loss value; Step a3, back propagation: Calculate the gradient of the total loss function with respect to the network weights through the back propagation algorithm; Step a4, parameter update: Based on the gradient, the network weights are adjusted through the optimizer to minimize the total loss function.

8. The system according to claim 6, characterized in that The optimizer is the Adam optimizer.

9. A magnetic target state estimation device, characterized in that: It includes a processor, a memory and a bus, wherein the memory stores instructions and data read by the processor, the processor is used to call the instructions and data in the memory to implement the system as described in any one of claims 1-8, and the bus connects the functional components for transmitting information.

10. The device according to claim 9, characterized in that Also included is a three-axis magnetic sensor for measuring magnetic field data from magnetic targets.

Images