Magnetic target intelligent positioning method and system based on magnetic gradient tensor time sequence

Abstract

The invention relates to the technical field of geomagnetic vector measurement, and particularly provides a magnetic target intelligent positioning method and system based on a magnetic gradient tensor time sequence, and the method comprises the steps: employing a plurality of magnetic sensors to form a cube magnetic sensor array model; constructing magnetic gradient tensor sequence data based on the distance from the original point to the magnetic target represented by the magnetic gradient tensors of any two surfaces in the magnetic sensor array model; and adopting an LSTM neural network positioning model and taking the magnetic gradient tensor sequence data as model input to obtain a magnetic target position. The distance between the magnetic target and the original point is represented by the magnetic gradient tensor, so that the influence of the magnetic vector of the geomagnetic field on the determination of the position of the magnetic target can be reduced. According to the LSTM neural network magnetic target positioning model designed by the invention, key information of different time steps in a sequence is flexibly captured by utilizing a dynamic adjustment memory unit, noise interference is better removed, and effective signals of a magnetic target are highlighted, so that the precision of a positioning task is improved.

Description

Technical Field

[0001] The present invention relates to the field of geomagnetic vector measurement technology, and in particular to a magnetic target intelligent positioning method and system based on a magnetic gradient tensor time series. Background Art

[0002] Magnetic anomaly detection technology is highly sensitive to ferromagnetic materials and boasts high concealment and proactiveness. Magnetic gradient tensor measurement, a key technology for magnetic anomaly detection, is highly resistant to environmental interference. Magnetic anomaly detection methods based on magnetic gradient tensors can provide richer magnetic field information, making them more conducive to locating magnetic targets. Currently, magnetic anomaly detection technology is being used in defense and military industries, geological exploration, biomedicine, underground engineering monitoring, and other fields.

[0003] Nara et al. derived the Euler deconvolution analytical positioning formula for homogeneous functions and proposed using the magnetic gradient tensor and magnetic field vector to determine the position of magnetic targets. Davis et al. introduced the Hilbert transform data processing method based on the Euler deconvolution positioning method. However, because the magnetic field vector generated by a magnetic target is difficult to separate from the geomagnetic field, the Euler deconvolution-based method is significantly affected by geomagnetic field noise, which can cause deviations in the magnetic target positioning results.

[0004] Wiegert et al. proposed a magnetic target positioning method based on the spatial variation of the magnetic gradient tensor invariant, also known as scalar triangulation. Clark et al. studied magnetic target positioning technology using the normalized magnetic source intensity method. By utilizing the rotational invariant of the magnetic gradient tensor, they solved for the position and structural index of the magnetic target. Although these methods minimize the influence of the geomagnetic field in the environment, the mathematical analytical solution is limited by the aspheric errors of the magnetic sensor array structure, resulting in reduced accuracy in magnetic target positioning.

[0005] Many researchers also study magnetic target positioning as a multivariate nonlinear problem. The nonlinear regression algorithm assumes that the relationship between the magnetic field and the magnetic source is nonlinear through the formula for the distribution of magnetic field intensity in space, and uses the nonlinear regression algorithm to fit, iteratively optimize and solve, and continuously reduce the error between the estimated value and the measured value to achieve the positioning of the magnetic source target. Roger Alimi et al. proposed a magnetic target positioning algorithm based on genetic algorithm. Liming Fan et al. proposed a magnetic target positioning algorithm based on particle swarm algorithm. Yaoxin Zheng et al. used the Levenberg-Marquardt (LM) algorithm to achieve magnetic target positioning. However, the positioning model established by the nonlinear regression iterative optimization solution method has high requirements on the initial value of the model setting.

[0006] In recent years, with the development of artificial intelligence (AI), technologies such as fully connected layer networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformer neural networks have achieved promising results in a growing number of applications. This also provides a new approach for magnetic target positioning. Therefore, this paper applies deep learning technology to propose a new positioning method, using an LSTM neural network to construct a model of magnetic anomaly fields and magnetic target locations. Summary of the Invention

[0007] In view of this, the present invention proposes a magnetic target intelligent positioning method and system based on magnetic gradient tensor time series to solve the problems existing in the above-mentioned prior art.

[0008] On the one hand, to achieve the above-mentioned purpose, the present invention proposes a magnetic target intelligent positioning method based on magnetic gradient tensor time series, comprising:

[0009] Based on a sensor array structure composed of a cube of several magnetic sensors, a magnetic sensor array model is constructed;

[0010] Constructing magnetic gradient tensor sequence data based on the distance from the origin to the magnetic target represented by the magnetic gradient tensors of any two surfaces in the magnetic sensor array model;

[0011] An LSTM neural network positioning model is constructed, and the magnetic gradient tensor sequence data is used as model input to obtain the magnetic target position.

[0012] Furthermore, in the magnetic sensor array model, a plurality of magnetic sensors are arranged at equal intervals.

[0013] Furthermore, a method for expressing the distance from the origin to the magnetic target based on the magnetic gradient tensor of any two surfaces in the magnetic sensor array model is as follows:

[0014]

[0015] The distance vector between the centers of the +Z plane and the +Y plane is The distance vector from the center of the +Z surface to the origin is ds, dr and ds are both known constants, the magnetic gradient tensor of the +Z surface is G1, and the Y direction component of G1 is G 1Y The magnetic gradient tensor on the +Y plane is G2, and the Z-direction component of G2 is G 2Z .

[0016] Furthermore, the output of the LSTM neural network positioning model is set to 6 channels, and the model output includes the Cartesian coordinate system position and the spherical coordinate system position of the magnetic target. The magnetic target position is obtained based on the redundancy and complementarity between the Cartesian coordinate system and the spherical coordinate system data.

[0017] Furthermore, the LSTM neural network positioning model obtains the Cartesian coordinate system position and the spherical coordinate system position by learning the mapping relationship between the magnetic gradient tensor sequence and the magnetic target position through training.

[0018] Furthermore, the LSTM neural network positioning model uses dynamically adjusted memory units to capture key information at different time steps in the magnetic gradient tensor sequence and remove noise interference.

[0019] On the other hand, to achieve the above-mentioned purpose, the present invention proposes a magnetic target intelligent positioning system based on magnetic gradient tensor time series, comprising a magnetic sensor array, a data processing module, and a magnetic target positioning module;

[0020] The magnetic sensor array is a cube array structure composed of a plurality of magnetic sensors, and magnetic vector data of a magnetic target is collected based on the magnetic sensor array;

[0021] The data processing module is used to convert the magnetic vector data into magnetic gradient tensor data;

[0022] The magnetic target positioning module is used to obtain the magnetic target position based on the magnetic gradient tensor data using an LSTM neural network positioning model.

[0023] Furthermore, in the magnetic sensor array, a plurality of magnetic sensors are arranged at equal intervals.

[0024] Furthermore, the output of the LSTM neural network positioning model is set to 6 channels, and the model output includes the Cartesian coordinate system position and the spherical coordinate system position of the magnetic target. The magnetic target position is obtained based on the redundancy and complementarity between the Cartesian coordinate system and the spherical coordinate system data.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] This method represents the distance of a magnetic target from its origin using a magnetic gradient tensor, reducing the influence of the Earth's magnetic field vector on the determination of the magnetic target's position. The designed LSTM neural network magnetic target positioning model can reduce the magnetic target positioning deviation caused by the inherent errors of a cubic magnetic sensor array. This model uses magnetic gradient tensor sequence data as input and establishes a temporal dependency for all collected magnetic gradient tensor sequence data. This model utilizes dynamically adjusted memory cells to flexibly capture key information at different time steps in the sequence, effectively removing noise interference and highlighting the effective signal of the magnetic target, thereby improving the accuracy of positioning tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0028] Figure 1 This is a flow chart of the positioning method according to the present invention;

[0029] Figure 2 This is a schematic diagram of a cube magnetic sensor array locating a magnetic target in an embodiment of the present invention;

[0030] Figure 3 Schematic diagram of the LSTM neural network magnetic target positioning model in an embodiment of the present invention;

[0031] Figure 4 The relative positioning error distribution diagram of the FCL magnetic target positioning model, where (a) is the x-coordinate error, (b) is the y-coordinate error, (c) is the z-coordinate error, (d) is the radius error, (e) is the elevation angle θ error, and (f) is the direction angle Φ error;

[0032] Figure 5 The relative positioning error distribution diagram of the CNN magnetic target positioning model, where (a) is the x-coordinate error, (b) is the y-coordinate error, (c) is the z-coordinate error, (d) is the radius error, (e) is the elevation angle θ error, and (f) is the direction angle Φ error;

[0033] Figure 6 The relative positioning error distribution diagram of the Transformer magnetic target positioning model, where (a) is the x-coordinate error, (b) is the y-coordinate error, (c) is the z-coordinate error, (d) is the radius error, (e) is the elevation angle θ error, and (f) is the direction angle Φ error;

[0034] Figure 7 The relative positioning error distribution diagram of the LSTM magnetic target positioning model of the present invention, where (a) is the x-coordinate error, (b) is the y-coordinate error, (c) is the z-coordinate error, (d) is the radius error, (e) is the elevation angle θ error, and (f) is the direction angle Φ error;

[0035] Figure 8 This is a distance error change trend chart of the 10,000 points with the largest positioning distance errors according to the LSTM neural network magnetic target positioning model described in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0037] This embodiment aims to solve the problem of reduced magnetic target positioning accuracy caused by the inherent errors of the geomagnetic field and the cube sensor array structure. In order to reduce the error caused by the geomagnetic field on magnetic target positioning, a method is proposed that only uses the magnetic gradient tensor to represent the distance relationship between the magnetic target and the origin. In order to reduce the impact of the asphericity error of the cube magnetic sensor array model on magnetic target positioning, a neural network positioning method based on magnetic gradient tensor sequence data is proposed, such as Figure 1 As shown, including:

[0038] Create a cube magnetic sensor array model:

[0039] like Figure 2 As shown, 8 magnetic sensors are placed on the sensor bracket. The distance between any two adjacent magnetic sensors is l, so the placement of these 8 sensors forms a cube. The front of the cube is regarded as the +X surface, the back is regarded as the -X surface, the right side is regarded as the +Y surface, the left side is regarded as the -Y surface, the bottom is regarded as the +Z surface, and the top is regarded as the -Z surface. The magnetic gradient tensor of any two surfaces can be used to represent the distance from the origin to the magnetic target. In this way, the influence of the geomagnetic field on the determination of the distance of the magnetic target from the origin can be effectively reduced. In this embodiment, the +Z surface and the +Y surface are used to deduce the distance of the magnetic target from the origin. Let the distance vector from the center of the +Z surface to the magnetic target be The magnetic induction intensity collected at the center of the +Z surface is B1, and the distance vector from the center of the +Y surface to the magnetic target is The magnetic induction intensity collected at the center of the +Y plane is B2, and the distance between the centers of the +Z plane and the +Y plane is The coordinates of the +Z surface center to the origin are The distance vector from the coordinate origin to the magnetic target is

[0040] The formula for magnetic dipole distance is:

[0041]

[0042] in, is the relative position vector from the target to the observation point in the Cartesian coordinate system, is the magnetic moment vector of the magnetic dipole, and μ0 is the magnetic permeability of vacuum.

[0043]

[0044] B S1 、B S2 、B S3 、B S4 、B S5 、B S6 、B S7 、B S8 are the magnetic field vectors collected by magnetic sensors 1 to 8 respectively. The magnetic field vector at the center of the +Z plane is The magnetic field vector at the center of the +Y plane is

[0045] According to formula (1), we can get formula (7) (8):

[0046]

[0047] Substituting formula (3) into formula (9) yields:

[0048]

[0049] Substituting formula (10) into formula (2) yields:

[0050]

[0051] It is (Bxy Byy Bzy) in G1, so Belong to G 1， That is G 1Y .

[0052] It is (Bxz Byz Bzz) in G2, so Belongs to G2, that is, G 1Z .

[0053] so,

[0054] and are all known constant vectors, so we know The value of is mainly affected by the magnetic gradient tensor value of each surface in the cube magnetic sensor array. Expressed as G, it can reduce the influence of the magnetic vector of the Earth's magnetic field on the determination of the distance between the magnetic target and the origin.

[0055] LSTM neural network positioning model:

[0056] This example records all magnetic vector information fed back when a magnetic target is at the same position over a continuous period of time. This method also records the magnetic vector information of the magnetic target at different measurement points. The magnetic vector data from the eight magnetic sensors in a cube-shaped magnetic sensor array is converted into magnetic gradient tensor data for the six faces (G1, G2, G3, G4, G5, and G6) of the cube-shaped magnetic sensor array and used as model input. The position of the magnetic target in the Cartesian coordinate system (x, y, z) and the spherical coordinate system (r, θ, Φ) are used as model outputs. These are then input into the LSTM neural network model for training.

[0057] like Figure 3 As shown in the figure, the LSTM neural network can establish temporal dependencies between all collected data. By dynamically adjusting memory units, key information at different time steps in the sequence can be flexibly captured, effectively removing noise interference and highlighting the effective signal of the magnetic target, thereby improving the accuracy of the positioning task. The model output is set to six channels: Cartesian coordinate position (x, y, z) and spherical coordinate position (r, θ, Φ). These outputs jointly determine the magnetic target position and improve the robustness of the positioning task.

[0058] The Cartesian coordinate system (x, y, z) provides the specific position of a magnetic target in three-dimensional space, directly reflecting its offset and distribution in all directions. The spherical coordinate system (r, θ, Φ) better describes the distance and direction relationship between the magnetic target and the origin. Although a certain mathematical conversion relationship exists between the Cartesian and spherical coordinate systems, the x, y, z components of the Cartesian coordinate system and the r, θ, Φ components of the spherical coordinate system each have unique data distributions and characteristic patterns. Therefore, the Cartesian coordinate system position and the spherical coordinate system position are used together to determine the magnetic target position.

[0059] This example tests the model performance and positioning error of the FCL, CNN, Transformer, and LSTM models, including the following steps:

[0060] Through field experiments, a total of 19,419,600 data points were collected. The data was divided into a training set and a test set at a ratio of 4:1, resulting in a training set of 1,559,680 data points and a test set of 3,883,920 data points. Mean absolute error (MAE) and root mean square error (RMSE) were used to evaluate the positioning performance of each measurement point. Table 1 shows the MAE and RMSE values for the FCL, CNN, Transformer, and LSTM models in the magnetic target positioning task.

[0061] Table 1

[0062] Model MAE RMSE FCL 0.0356 0.0547 CNN 0.0163 0.0248 Transformer 0.0665 0.1107 LSTM 0.002 0.0036

[0063] The MAE value of the FCL model is 0.0356, and the RMSE value is 0.0547. The MAE and RMSE values of this model in each dimension are relatively high, indicating that it cannot effectively capture the complex characteristics and time series dependencies of magnetic target signals.

[0064] The CNN model achieved a MAE of 0.0163 and an RMSE of 0.0248. The CNN model performed better than the FCL model, indicating that it can extract local features and improve positioning accuracy to a certain extent. However, the improvement in overall positioning accuracy is limited, and it still cannot fully capture long-term dependencies in the time series.

[0065] The Transformer model achieved a MAE of 0.0665 and an RMSE of 0.1107. The Transformer model performed averagely on this task. This suggests that while it can effectively capture global dependencies in magnetic target signals, its ability to capture local features in time series data with small data volumes, simple relationships, and uncomplicated features still needs improvement.

[0066] The LSTM model achieved a MAE of 0.002 and an RMSE of 0.0036. This model is particularly suitable for time series data with small data volumes, simple relationships, and simple features. It has better data sequence sensitivity and can effectively capture the long-term dependencies and dynamic change characteristics of magnetic target signals.

[0067] The MAE and RMSE values of the LSTM magnetic target positioning model are smaller than those of other models, indicating that the model has a better magnetic target positioning effect.

[0068] Figure 4 The positioning error of the FCL magnetic target positioning model is shown. Figure 4It can be concluded that in the Cartesian coordinate system, the mean (mu) of the distance error distribution of the FCL magnetic target positioning model on the x-axis is 0.00465, the standard deviation (sigma) is 0.3039, and the distance error range is -0.3cm to 0.2cm. The mean (mu) of the distance error distribution on the y-axis is -0.02828, the standard deviation (sigma) is 0.05473, and the distance error range is -0.5cm to 0.2cm. The mean (mu) of the distance error distribution on the z-axis is 0.06121, the standard deviation (sigma) is 0.07068, and the distance error range is -0.5cm to 0.6cm. In the spherical coordinate system, the mean (mu) of the r coordinate distance error distribution is 0.03416, the standard deviation (sigma) is 0.03996, and the distance error range is -0.3 cm to 0.3 cm. The mean (mu) of the θ coordinate angle error distribution is 0.02088, the standard deviation (sigma) is 0.02517, and the distance error range is -0.2 to 0.15. The mean (mu) of the Φ coordinate angle error distribution is 0.00549, the standard deviation (sigma) is 0.02515, and the distance error range is -0.2 to 0.15.

[0069] Figure 5 The positioning error of the CNN magnetic target positioning model is shown. Figure 5 It can be concluded that in the Cartesian coordinate system, the mean (mu) of the x-axis distance error distribution of the CNN magnetic target positioning model is 0.00183, the standard deviation (sigma) is 0.01316, and the distance error range is -0.08cm to 0.1cm. The mean (mu) of the y-axis distance error distribution is -0.00592, the standard deviation (sigma) is 0.01759, and the distance error range is -0.2cm to 0.2cm. The mean (mu) of the z-axis distance error distribution is -0.01368, the standard deviation (sigma) is 0.04032, and the distance error range is -0.4cm to 0.4cm. In the spherical coordinate system, the mean (mu) of the r-axis distance error distribution is -0.01675, the standard deviation (sigma) is 0.0246, and the distance error range is -0.2cm to 0.2cm. The mean (mu) of the θ coordinate angle error distribution is 0.00413, the standard deviation (sigma) is 0.01524, and the range error range is -0.2 to 0.2. The mean (mu) of the Φ coordinate angle error distribution is 0.00706, the standard deviation (sigma) is 0.01258, and the range error range is -0.06 to 0.06.

[0070] Figure 6 The positioning error of the Transformer magnetic target positioning model is shown. Figure 6It can be seen that in the Cartesian coordinate system, the Transformer magnetic target positioning model has an x-axis distance error distribution with a mean (mu) of -0.03022 and a standard deviation (sigma) of 0.05981, with a range of -0.2cm to 0.15cm. The y-axis distance error distribution has a mean (mu) of -0.03179 and a standard deviation (sigma) of 0.07295, with a range of -0.3cm to 0.2cm. The z-axis distance error distribution has a mean (mu) of 0.0273 and a standard deviation (sigma) of 0.21247, with a range of -1cm to 2cm. In the spherical coordinate system, the mean (mu) of the r coordinate distance error distribution is -0.00204, the standard deviation (sigma) is 0.07753, and the distance error range is -0.4cm to 0.3cm. The mean (mu) of the θ coordinate angle error distribution is -0.0416, the standard deviation (sigma) is 0.07717, and the distance error range is -0.6 to 0.5. The mean (mu) of the Φ coordinate angle error distribution is -0.00998, the standard deviation (sigma) is 0.0552, and the distance error range is -0.3 to 0.1.

[0071] Figure 7 The positioning error of the LSTM magnetic target positioning model of the present invention is shown. Figure 7 It can be seen that in the Cartesian coordinate system, the LSTM magnetic target positioning model has an x-axis distance error distribution with a mean (mu) of 7.80369E-5 and a standard deviation (sigma) of 0.00319, with a range of -0.04cm to 0.04cm. The y-axis distance error distribution has a mean (mu) of -4.46244E-4 and a standard deviation (sigma) of 0.0036, with a range of -0.06cm to 0.05cm. The z-axis distance error distribution has a mean (mu) of -4.46244E-4 and a standard deviation (sigma) of 0.0043, with a range of -0.1cm to 0.07cm. In the spherical coordinate system, the mean (mu) of the r coordinate distance error distribution is -1.74892E-4, the standard deviation (sigma) is 0.00283, and the distance error range is -0.04cm to 0.03cm. The mean (mu) of the θ coordinate angle error distribution is 1.34798E-4, the standard deviation (sigma) is 0.00199, and the distance error range is -0.03 to 0.03. The mean (mu) of the Φ coordinate angle error distribution is -6.47934E-4, the standard deviation (sigma) is 0.00338, and the distance error range is -0.05 to 0.04.

[0072] Comprehensive analysis shows that the mean, standard deviation and error distribution of each component of the LSTM magnetic target positioning model are basically the smallest, indicating that the model makes full use of the time series characteristics of the magnetic target signal in the magnetic target positioning task and has a good detection effect in the magnetic target spatial position positioning task.

[0073] This embodiment also provides a magnetic gradient tensor measurement system, which consists of a data acquisition module, a data processing module, a data transmission module, a power supply module, and a host computer. The data acquisition module includes 8 magnetometers and a bracket made of acrylic material. The 8 magnetometers are fixed on a cube acrylic bracket, and the distance between two adjacent magnetometers is 30 cm. The data processing module is composed of 6 NI acquisition cards. The 24-channel magnetic vector data generated by the 8 magnetometers can be converted into analog-to-digital data. The data transmission module uses Ethernet transmission to realize communication between the data processing module and the host computer. The power supply module uses an outdoor power supply to power the system. The host computer is made with the LabVIEW programming language to control the switch of the magnetic signal acquisition command.

[0074] The target area was measured as a semi-cylinder with a radius of 200 cm and a height of 57 cm. Magnetic targets were placed sequentially at four distances: 50 cm, 100 cm, 150 cm, and 200 cm. Magnetic targets were placed at five orientations: 0 degrees, 45 degrees, 90 degrees, 145 degrees, and 180 degrees. They were also placed at five heights: 27.8 cm, 35.1 cm, 42.4 cm, 49.7 cm, and 57 cm. The magnetic targets were placed at the 100 measurement points described above and allowed to rest. The magnetic gradient tensor measurement system was then used to perform sequential measurements, with a total acquisition time of 30 seconds per measurement point. A total of 19,419,600 data points were obtained. The sequence data was allocated to the training and test sets in a 4:1 ratio, resulting in 1,559,680 training and 3,883,920 test data points.

[0075] Figure 8 The following is a trend chart of the range error for the 10,000 points with the largest range errors. The dashed line in the figure marks the maximum range value, which is located near index 10,000 and has a range of approximately 0.10 cm. The actual maximum range error is 0.099 cm, indicating that the model has good magnetic target positioning performance.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A magnetic target intelligent positioning method based on magnetic gradient tensor time series, characterized in that: include: Based on a sensor array structure composed of several magnetic sensors forming a cube, a magnetic sensor array model is constructed; Constructing magnetic gradient tensor sequence data based on the distance from the origin to the magnetic target represented by the magnetic gradient tensors of any two surfaces in the magnetic sensor array model; An LSTM neural network positioning model is constructed, and the magnetic gradient tensor sequence data is used as model input to obtain the magnetic target position.

2. The magnetic target intelligent positioning method based on magnetic gradient tensor time series according to claim 1 is characterized in that: In the magnetic sensor array model, a plurality of magnetic sensors are arranged at equal intervals.

3. The magnetic target intelligent positioning method based on magnetic gradient tensor time series according to claim 1 is characterized in that: The method for expressing the distance from the origin to the magnetic target based on the magnetic gradient tensor of any two surfaces in the magnetic sensor array model is as follows: Among them, the distance vector between the centers of the +Z plane and the +Y plane is The distance vector from the center of the +Z surface to the origin is and are all known constants; the magnetic gradient tensor on the +Z plane is G1, and the Y-direction component of G1 is G 1Y ; The magnetic gradient tensor on the +Y plane is G2, and the Z direction component of G2 is G 2Z .

4. The magnetic target intelligent positioning method based on magnetic gradient tensor time series according to claim 1 is characterized in that: The output of the LSTM neural network positioning model is set to 6 channels, and the model output includes the Cartesian coordinate system position and the spherical coordinate system position of the magnetic target. The magnetic target position is obtained based on the redundancy and complementarity between the Cartesian coordinate system and the spherical coordinate system data.

5. The magnetic target intelligent positioning method based on magnetic gradient tensor time series according to claim 4 is characterized in that: The LSTM neural network positioning model learns the mapping relationship between the magnetic gradient tensor sequence and the magnetic target position through training, thereby obtaining the Cartesian coordinate system position and the spherical coordinate system position.

6. The magnetic target intelligent positioning method based on magnetic gradient tensor time series according to claim 1 is characterized in that: The LSTM neural network positioning model uses dynamically adjusted memory units to capture key information at different time steps in the magnetic gradient tensor sequence and remove noise interference.

7. A magnetic target intelligent positioning system based on magnetic gradient tensor time series according to any one of claims 1 to 6, characterized in that: It includes a magnetic sensor array, a data processing module, and a magnetic target positioning module; The magnetic sensor array is a cube array structure composed of a plurality of magnetic sensors, and magnetic vector data of a magnetic target is collected based on the magnetic sensor array; The data processing module is used to convert the magnetic vector data into magnetic gradient tensor data; The magnetic target positioning module is used to obtain the magnetic target position based on the magnetic gradient tensor data using an LSTM neural network positioning model.

8. The magnetic target intelligent positioning system according to claim 7, characterized in that: In the magnetic sensor array, a plurality of magnetic sensors are arranged at equal intervals.

9. The magnetic target intelligent positioning system according to claim 7, characterized in that: The output of the LSTM neural network positioning model is set to 6 channels, and the model output includes the Cartesian coordinate system position and the spherical coordinate system position of the magnetic target. The magnetic target position is obtained based on the redundancy and complementarity between the Cartesian coordinate system and the spherical coordinate system data.

Images