Magnetic dipole target positioning method and system based on 1D-RCNN

Abstract

The application discloses a magnetic dipole target positioning method and system based on 1D-RCNN, and the method comprises the following steps: acquiring position data, magnetic moment data and magnetic induction intensity data of a magnetic dipole target; adding noise to the magnetic induction intensity data as input data, taking the position data and the magnetic moment data as label data, and randomly dividing them into a training set, a verification set and a test set; constructing a magnetic dipole target positioning model based on a 1D-RCNN network, training, verifying and evaluating the magnetic dipole target positioning model, and then inversely calculating the position and the magnetic moment of the magnetic dipole target according to the trained magnetic dipole target positioning model. The application is applied to the field of magnetic target positioning, can quickly and accurately estimate the position and the magnetic moment of a magnetic target by virtue of the nonlinear inversion capability of the 1D-RCNN network, has higher positioning accuracy even under low signal-to-noise ratio, and solves the problems of low positioning accuracy and poor real-time performance of the magnetic dipole target positioning.

Description

Technical Field

[0001] This invention relates to the field of magnetic target localization technology, specifically a magnetic dipole target localization method and system based on 1D-RCNN. Background Technology

[0002] Magnetic anomalies can be used to detect and locate magnetic objects. Currently, this technology is widely used in areas such as locating unexploded ordnance, monitoring ships in ports, locating objects indoors, and capsule endoscopy. Therefore, researching how to use magnetic anomaly detection technology for target location is of great theoretical value.

[0003] Existing three-component magnetic field localization methods primarily involve equating a magnetic target to a magnetic dipole under far-field conditions, establishing a magnetic dipole model to describe the target's magnetic flux density, and then constructing a set of nonlinear equations for magnetic dipole target localization, which are then solved using algorithms. Currently popular magnetic localization methods can be categorized into numerical optimization algorithms (e.g., the LM algorithm and the Gauss-Newton method) and intelligent optimization algorithms (e.g., the PSO algorithm). Numerical optimization algorithms like the LM algorithm offer high localization accuracy but require an initial solution. If the initial solution differs significantly from the global optimum, the LM algorithm may converge to a local optimum. In practical localization, because the position and magnetic moment of the magnetic target are initially unknown, it is difficult to determine a good initial solution. Numerical optimization algorithms such as the PSO algorithm are global search algorithms, but when the adaptive function is complex or the data contains noise, the PSO algorithm has many local minima. It requires a larger population size and more iterations, which significantly slows down the localization process.

[0004] The above analysis shows that numerical optimization algorithms are too sensitive to initial values, making it difficult to guarantee target localization performance. Intelligent optimization algorithms are also time-consuming and cannot meet real-time requirements. Therefore, exploring high-precision machine learning algorithms that also meet real-time requirements is extremely important for magnetic dipole target localization. Summary of the Invention

[0005] To address the shortcomings of the existing technology, this invention provides a magnetic dipole target localization method and system based on 1D-RCNN, which solves the problem of low positioning accuracy of magnetic dipole targets. It can quickly analyze the target magnetic induction intensity data measured by magnetic sensors, accurately locate the position and magnetic moment of the magnetic target, and meet the real-time requirements. It also has high positioning accuracy even at low signal-to-noise ratios.

[0006] To achieve the above objectives, this invention provides a magnetic dipole target localization method based on 1D-RCNN, comprising the following steps:

[0007] Step 1: Construct a sensor array consisting of multiple magnetic sensors, and based on the sensor array and the magnetic dipole far-field model, acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target.

[0008] Step 2: Add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data. Use the position data and magnetic moment data of the magnetic dipole target as label data. Randomly divide the dataset composed of the input data and the label data into training set, validation set and test set.

[0009] Step 3: Construct a magnetic dipole target localization model based on a 1D-RCNN network, and train, validate and evaluate the magnetic dipole target localization model based on the training set, the validation set and the test set. Then, based on the trained magnetic dipole target localization model, invert the position and magnetic moment of the magnetic dipole target.

[0010] In one embodiment, step 1, acquiring the position data, magnetic moment data, and magnetic flux density data of the magnetic dipole target, includes:

[0011] Step 1.1: Construct the functional relationship between the position, magnetic moment, and magnetic flux density of the magnetic dipole target;

[0012] Step 1.2: Obtain the position data and magnetic moment data of the magnetic dipole target, and substitute them into the function relationship to obtain the magnetic induction intensity data of the corresponding magnetic dipole target.

[0013] In one embodiment, in step 1.1, the functional relationship is:

[0014]

[0015] Among them, B x B y B z Let be the projection components of the magnetic flux density vector of the magnetic dipole target onto the x, y, and z axes, respectively; μ0 be the free permeability; r be the distance between the magnetic dipole target and the magnetic sensor; (x0, y0, z0) be the position of the magnetic dipole target; and (x, y, z) be the position of the magnetic sensor. x m y m z These are the projection components of the magnetic moment vector of the magnetic dipole target onto the x, y, and z axes, respectively.

[0016] In one embodiment, in step 2, the dataset consisting of the input data and the label data is randomly divided into a training set, a validation set, and a test set, specifically as follows:

[0017] The dataset consisting of the input data and the label data is randomly divided into a training set, a validation set, and a test set in a ratio of 70:15:15.

[0018] In one embodiment, in step 3, the magnetic dipole target localization model includes an initial convolutional layer, three residual blocks, a global average pooling layer, and a linear layer;

[0019] Each residual block has three residual units, and each residual unit has two convolutional layers and two ReLU activation functions.

[0020] In one embodiment, step 3 involves training and validating the magnetic dipole target localization model, specifically as follows:

[0021] Step 3.1: Standardize the standard deviation of the training set, the validation set and the test set respectively, so that the processed data conforms to a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0022] Step 3.2: Train and validate the magnetic dipole target localization model based on the training set and the validation set respectively, and save the trained magnetic dipole target localization model.

[0023] Step 3.3: Input the input data from the test set into the saved magnetic dipole target localization model to obtain the output data of the magnetic dipole target localization model, and evaluate the magnetic dipole target localization model based on the root mean square error between the output data and the label data in the test set.

[0024] In one embodiment, in step 3.2, during the training and verification of the magnetic dipole target localization model, the mean square error between the model output data and the label data is used as the loss function, and an optimizer based on the adaptive moment estimation algorithm is used to optimize the parameters of the magnetic dipole target localization model.

[0025] To achieve the above objectives, the present invention also provides a magnetic dipole target localization system based on 1D-RCNN, which uses the above method to localize magnetic dipole targets. The magnetic dipole target localization system includes:

[0026] The data acquisition module is used to acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target;

[0027] The data processing module is used to add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data, use the position data and magnetic moment data of the magnetic dipole target as label data, and randomly divide the dataset composed of the input data and the label data into training set, validation set and test set.

[0028] The model training module is used to construct a magnetic dipole target localization model based on a 1D-RCNN network, and to train, validate, and evaluate the magnetic dipole target localization model based on the training set, the validation set, and the test set.

[0029] The target inversion module is used to invert the position and magnetic moment of the magnetic dipole target based on the trained magnetic dipole target localization model.

[0030] Compared with the prior art, the present invention has the following beneficial technical effects:

[0031] This invention constructs a magnetic dipole target localization model based on a 1D-RCNN network. This model conforms to the complex relationship between the position of the magnetic dipole and the magnetic moment and magnetic induction intensity. By utilizing the nonlinear inversion capability of the 1D-RCNN network, the magnetic dipole target localization model can quickly and accurately analyze the target magnetic induction intensity data measured by the magnetic sensor, estimate the position and magnetic moment of the magnetic target, and achieve high localization accuracy even at low signal-to-noise ratios. At the same time, it effectively solves the problems of low localization accuracy and poor real-time performance of magnetic dipole targets. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the magnetic dipole target localization method based on 1D-RCNN in Embodiment 1 of the present invention;

[0034] Figure 2 This is a schematic diagram of the magnetic dipole model in Embodiment 1 of the present invention;

[0035] Figure 3 This is a schematic diagram of the magnetic sensor array in Embodiment 1 of the present invention;

[0036] Figure 4 This is a schematic diagram of the 1D-RCNN structure in Embodiment 1 of the present invention;

[0037] Figure 5 This is a schematic diagram of the residual unit in Embodiment 1 of the present invention;

[0038] Figure 6 This is a schematic diagram of the model training and validation loss curves in Embodiment 1 of the present invention;

[0039] Figure 7This is a schematic diagram showing the distribution of predicted and actual values ​​at 30 test data locations in Embodiment 1 of the present invention;

[0040] Figure 8 This is a schematic diagram showing the distribution of predicted and actual magnetic moments of 30 test data points in Embodiment 1 of the present invention;

[0041] Figure 9 This is a structural block diagram of the magnetic dipole target localization device based on 1D-RCNN in Embodiment 2 of the present invention.

[0042] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0044] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of the components in a specific posture (as shown in the attached figure). If the specific posture changes, the directional indicator will also change accordingly. In the description of the present invention, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0045] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0046] Example 1

[0047] like Figure 1 The following is a method for locating magnetic dipole targets based on 1D-RCNN disclosed in this embodiment, including the following steps:

[0048] Step 1: Construct a sensor array consisting of four magnetic sensors, and based on the sensor array and the magnetic dipole far-field model, acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target. When the distance between the magnetic object target and the magnetic sensor exceeds 2.5 times or more the geometric size of the magnetic object target, the magnetic object target can be regarded as a magnetic dipole target.

[0049] Step 2: Add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data. Use the position data and magnetic moment data of the magnetic dipole target as label data. Randomly divide the dataset composed of input data and label data into training set, validation set and test set.

[0050] Step 3: Construct a magnetic dipole target localization model based on a 1D-RCNN network, and train, validate and evaluate the magnetic dipole target localization model based on the training set, validation set and test set. Then, based on the trained magnetic dipole target localization model, invert the position and magnetic moment of the magnetic dipole target.

[0051] In this embodiment, the process of obtaining the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target in step 1 is as follows:

[0052] Step 1.1: Construct a linear array of four magnetic sensors spaced 2 meters apart. Based on the sensor array and the far-field model of the magnetic dipole, construct the functional relationship between the position, magnetic moment, and magnetic induction intensity of the magnetic dipole target. The specific implementation process is as follows:

[0053] In practical applications, four triaxial fluxgate sensors are selected to construct a sensor array. The coordinate positions of each magnetic sensor are known, and the influence of the Earth's magnetic field has been eliminated. The measurement value B of each magnetic sensor is the anomalous field of the magnetic dipole target in the x, y, and z axes in a rectangular coordinate system, that is:

[0054] B = [B x B y B z ]

[0055] The magnetic anomaly measured by the i-th magnetic sensor is described as follows:

[0056]

[0057] The magnetic flux density generated by the magnetic dipole target is:

[0058]

[0059] In a Cartesian coordinate system, expanding the magnetic flux density B along the x, y, and z directions yields the following functional relationship between the position, magnetic moment, and magnetic flux density of the magnetic dipole target:

[0060]

[0061] Where μ0 is the free permeability, with a value of 4π × 10⁻⁶. -7H / m, (x0,y0,z0) is the position of the magnetic dipole target, (x,y,z) is the position of the magnetic sensor, r is the distance between the magnetic dipole target and the magnetic sensor, r is the position vector from the magnetic dipole to the magnetic sensor, m is the magnetic moment vector, B x B y B z Let m be the projection components of the magnetic flux density vector of the magnetic dipole target onto the x, y, and z axes, respectively. x m y m z These are the projection components of the magnetic moment vector of the magnetic dipole target onto the x, y, and z axes, respectively.

[0062] Step 1.2: Obtain the position data and magnetic moment data of the magnetic dipole target, and substitute them into the functional relationship to obtain the magnetic flux density data of the corresponding magnetic dipole target. The specific implementation process is as follows:

[0063] The sampling range on the x, y, and z axes of the rectangular coordinate system is set to 1-10.1m, with a sampling interval of 1.3m, and the magnetic moment three components are m. x m y m z The sampling range is set to 100-880 A·m. 2 The sampling interval is 130 A·m 2 A total of 175,616 (8×8×8×7×7×7) points were generated. Given the position (x, y, z) and magnetic moment (m) of the magnetic dipole target... x ,m y ,m z By doing so, the three components B of the magnetic flux density of the corresponding magnetic dipole target can be calculated. x B y B z Substituting the position and magnetic moment (6 parameters) of the magnetic dipole target into the functional relationship between the position, magnetic moment, and magnetic flux density of the magnetic dipole target yields 12 magnetic flux density data points. Therefore, each point generates a magnetic flux density (12 elements) and the position and magnetic moment parameters (6 parameters) of the magnetic target, which are considered as a data pair.

[0064] In this embodiment, the specific implementation process of randomly dividing the dataset composed of input data and label data into training set, validation set and test set in step 2 is as follows:

[0065] Step 2.1, transfer the magnetic field strength data (B) x1 B y1 B z1 B x4 B y4 B z4 Add 0.1 nT of noise and use it as input data;

[0066] Step 2.2: Combine the position data and magnetic moment data (x, y, z, m) of the magnetic dipole. x m y m z ) as tag data;

[0067] Step 2.3: Randomly divide the dataset consisting of input data and label data into training set, validation set and test set in a ratio of 70:15:15.

[0068] In this embodiment, the magnetic dipole target localization model includes an initial convolutional (Conv) layer, three residual blocks, a global average pooling layer, and a linear layer. Each residual block has three residual units, each residual unit has two convolutional layers and two ReLU activation functions, and the number of channels in each residual block is 16, 32, and 64, respectively.

[0069] In practice, the process of training, validating, and evaluating the magnetic dipole target localization model based on the training set, validation set, and test set is as follows:

[0070] Step 3.1: Standardize the standard deviation of the training set, validation set, and test set respectively, so that the processed data conforms to a standard normal distribution with a mean of 0 and a standard deviation of 1. Specifically:

[0071]

[0072] Where, x * denoted as σ, where σ is the data after standard deviation standardization, x is the data before standard deviation standardization, μ is the mean of all sample data, and σ is the standard deviation of all sample data.

[0073] Step 3.2: Train and validate the magnetic dipole target localization model based on the training set and validation set respectively, and save the trained magnetic dipole target localization model. During the training and validation of the magnetic dipole target localization model, the mean square error between the model output data and the label data is used as the loss function, and an optimizer based on the adaptive moment estimation algorithm is used to optimize the parameters of the magnetic dipole target localization model.

[0074] Step 3.3: Input the input data from the test set into the saved magnetic dipole target localization model to obtain the output data of the magnetic dipole target localization model. Then, evaluate the magnetic dipole target localization model based on the root mean square error between the output data and the label data in the test set.

[0075]

[0076] Where RMES is the root mean square error, N is the number of test data samples, and x iThese are the predicted parameter values. These are the actual parameter values.

[0077] The magnetic dipole target localization method in this embodiment will be further explained below with specific examples.

[0078] like Figure 2 As shown, let the center coordinates of the magnetic target be Q(x0,y0,z0), its magnetic moment vector be m, the coordinates of the magnetic sensor be P(x,y,z), and the radius vector from the center point of the magnetic target to the sensor be r, with a distance of r. From the magnetic dipole model, the expression for the magnetic induction intensity B at point P is:

[0079]

[0080] Expanding equation (1) in a spatial rectangular coordinate system yields equation (2), where (B) x B y B z ) and (m x ,m y ,m z These represent the projection components of the magnetic flux density vector and the magnetic moment vector onto each coordinate axis, respectively.

[0081]

[0082] According to equation (2), only three equations can be established for a triaxial magnetic sensor, while the number of unknowns in the model is six. Therefore, at least two magnetic sensors are needed to perform measurements, forming six equations to solve the equation system and obtain the target's position and magnetic moment parameters. Assuming the number of magnetic sensors is N (N≥2), then equation (2) will become a highly nonlinear, multi-extremum overdetermined equation system consisting of 3N equations. Let B... x1 B y1 B z1 ,···,B xN B yN B zN These represent the actual magnetic flux density values ​​measured by the measurement array, while the theoretical values ​​calculated using the magnetic dipole model are... Based on the least squares criterion, by minimizing the squared error between the actual measured value and the theoretical calculated value, the following magnetic anomaly localization problem is obtained:

[0083]

[0084] The x that makes equation (3) hold is the optimal estimate of the required magnetic target parameter.

[0085] Sensor arrays such as Figure 3As shown, this considers four magnetic sensors, and the magnetic sensor array is a linear array with a spacing of 2m. The magnetic dipole target localization model based on the 1D-RCNN network is as follows. Figure 5 As shown, it consists of an initial convolutional (Conv) layer, three residual blocks (each residual block has three residual units), a global average pooling layer, and a linear layer. The number of channels in each residual block is 16, 32, and 64, respectively. The optimizer is an adaptive moment estimator (Adam), and the loss function is defined as mean squared error (MSE Loss). Figure 4 Let H(x) be a residual unit. The core of a residual neural network is the shortcut connection within the residual unit, which skips one or more layers, directly adding the input x of the previous unit to the output of the stacked layers. Thus, the output of this unit changes from F(x) to H(x), where H(x) = F(x) + x. Furthermore, the optimization objective of the residual neural network changes from the desired mapping F(x) to the residual mapping F(x) = H(x) - x. Experiments show that the residual unit H(x) using a deep residual network is easier to optimize than F(x). This shortcut connection neither adds extra parameters nor increases computational complexity, further improving the training speed and accuracy gain of the model.

[0086] Magnetic induction intensity data is acquired within the tracking space above the magnetic sensor. The sampling range on the x, y, and z axes is 1-10.1 m, with a sampling interval of 1.3 m. The three components of the magnetic moment are m. x m y m z The sampling range was 100-880, with a sampling interval of 130. A total of 175,616 (8×8×8×7×7×7) points were generated. At each point, magnetic induction intensity (12 elements) and magnet pose (6 elements) were generated as a data pair. 0.1 nT noise was added to the input dataset of the above 175,616 samples, and 70% (122,931 pairs) were randomly selected for training, 15% (26,343 pairs) for validation, and 15% (26,342 pairs) for testing.

[0087] Standardize the data using the standard deviation function:

[0088]

[0089] To evaluate positioning accuracy, the root mean square error of each parameter is defined as:

[0090]

[0091] The parameters of 1D-RCNN are set as follows: The network structure consists of an initial convolutional (Conv) layer, three residual blocks (each residual block has three residual units, each residual unit has two convolutional layers and two ReLU activation functions), a global average pooling layer, and a linear layer. The number of channels in each residual block is 16, 32, and 64, respectively. The network is trained for 100 epochs; the batch size is 64; the initial learning rate is set to 0.001, decreasing by a factor of 0.95 each epoch; the optimizer is the Adaptive Moment Estimator (Adam); the loss function is defined as Mean Squared Error (MSELoss). The training and validation loss graphs of 1D-RCNN are shown in Figure 6.

[0092] The root mean square errors of each parameter in the test data for 26342 are shown in Table 1 below:

[0093] Table 11 shows the root mean square error of each parameter in the D-RCNN test data.

[0094]

[0095] Thirty pairs of data were randomly selected from the test data, and their distributions of true position and magnetic moment versus predicted position and magnetic moment are shown in the following figure. Figure 7 , Figure 8 As shown.

[0096] Additionally, a new set of data is randomly generated within the positioning area, with a sampling range of 1.1-9.5m on the x, y, and z axes, a sampling interval of 2.8m, and a magnetic moment three-component m. x m y m z The sampling range is 110-710 A·m. 2 The sampling interval is 300 A·m 2 A total of 1728 (4×4×4×3×3×3) new test data were generated.

[0097] Using the previously trained weight data, the root mean square error of each parameter is shown in Table 2:

[0098] Table 21 shows the root mean square error of each parameter in the new D-RCNN test data.

[0099]

[0100] The average time required to run 1D-RCNN once is calculated to be 0.0024 s.

[0101] The stability of the model was tested by adding different noise levels to the dataset. The root mean square errors of each parameter of the test data at noise levels of 0.1nT, 0.5nT, and 1nT are shown in Table 3 below.

[0102] Table 31 shows the root mean square error of each parameter of D-RCNN under different standard noise deviations.

[0103]

[0104] As shown in Table 3, as the standard noise deviation increases, the root mean square error of each parameter in the model's test data gradually increases. However, even when the noise standard deviation is 1nT, the root mean square error of each parameter obtained by inversion is very small, indicating that 1D-RCNN has a certain stability and high positioning accuracy even under low signal-to-noise ratio conditions.

[0105] Table 4 shows the localization performance of the BP neural network, the ensemble algorithm XGBoost, and 1D-RCNN. As can be seen from Table 4, 1D-RCNN has the smallest root mean square error for each parameter obtained from the test data, indicating that it has the highest localization accuracy.

[0106] Table 4. Root mean square error of parameters for different neural networks

[0107]

[0108] The magnetic dipole target localization method based on 1D-RCNN in this embodiment can achieve high localization accuracy within the localization area and meet real-time requirements. It can effectively improve the localization accuracy of magnetic dipole targets even under low signal-to-noise ratio conditions, and has better localization performance compared with other machine learning algorithms.

[0109] Example 2

[0110] Based on the magnetic dipole target localization method based on 1D-RCNN in Embodiment 1, this embodiment discloses a magnetic dipole target localization device based on 1D-RCNN. (See reference...) Figure 9 The magnetic dipole target localization device includes a data acquisition module, a data processing module, a model training module, and a target inversion module. This magnetic dipole target localization device is used to execute some or all of the steps of the magnetic dipole target localization method in Embodiment 1, thereby achieving the localization of the magnetic dipole target. Specifically:

[0111] The data acquisition module is used to acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target;

[0112] The data processing module is used to add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data, use the position data and magnetic moment data of the magnetic dipole target as label data, and randomly divide the dataset composed of the input data and the label data into training set, validation set and test set.

[0113] The model training module is used to construct a magnetic dipole target localization model based on a 1D-RCNN network, and to train, validate, and evaluate the magnetic dipole target localization model based on the training set, the validation set, and the test set.

[0114] The target inversion module is used to invert the position and magnetic moment of the magnetic dipole target based on the trained magnetic dipole target localization model.

[0115] In this embodiment, the specific working process and working principle of the data acquisition module, data processing module, model training module and target inversion module are the same as those in Embodiment 1, so they will not be described again in this embodiment.

[0116] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for localizing magnetic dipole targets based on 1D-RCNN, characterized in that, Includes the following steps: Step 1: Construct a sensor array consisting of multiple magnetic sensors, and based on the sensor array and the magnetic dipole far-field model, acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target. Step 2: Add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data. Use the position data and magnetic moment data of the magnetic dipole target as label data. Randomly divide the dataset composed of the input data and the label data into training set, validation set and test set. Step 3: Construct a magnetic dipole target localization model based on a 1D-RCNN network, and train, validate and evaluate the magnetic dipole target localization model based on the training set, the validation set and the test set. Then, based on the trained magnetic dipole target localization model, invert the position and magnetic moment of the magnetic dipole target. In step 3, the magnetic dipole target localization model includes an initial convolutional layer, three residual blocks, a global average pooling layer, and a linear layer. Each residual block has three residual units, and each residual unit has two convolutional layers and two ReLU activation functions; The magnetic dipole target localization model is trained and validated, specifically as follows: Step 3.1: Standardize the standard deviation of the training set, the validation set and the test set respectively, so that the processed data conforms to a standard normal distribution with a mean of 0 and a standard deviation of 1. Step 3.2: Train and validate the magnetic dipole target localization model based on the training set and the validation set respectively, and save the trained magnetic dipole target localization model. Step 3.3: Input the input data from the test set into the saved magnetic dipole target localization model to obtain the output data of the magnetic dipole target localization model, and evaluate the magnetic dipole target localization model based on the root mean square error between the output data and the label data in the test set. In step 3.2, during the training and verification of the magnetic dipole target localization model, the mean square error between the model output data and the label data is used as the loss function, and an optimizer based on the adaptive moment estimation algorithm is used to optimize the parameters of the magnetic dipole target localization model.

2. The magnetic dipole target localization method based on 1D-RCNN according to claim 1, characterized in that, Step 1, acquiring the position data, magnetic moment data, and magnetic flux density data of the magnetic dipole target, includes: Step 1.1: Construct the functional relationship between the position, magnetic moment, and magnetic flux density of the magnetic dipole target; Step 1.2: Obtain the position data and magnetic moment data of the magnetic dipole target, and substitute them into the function relationship to obtain the magnetic induction intensity data of the corresponding magnetic dipole target.

3. The magnetic dipole target localization method based on 1D-RCNN according to claim 2, characterized in that, In step 1.1, the functional relationship is as follows: in, B x , B y , B z The magnetic flux density vectors of the magnetic dipole target are respectively in x , y , z Projected components on the axis, μ 0 is the permeability of free space. r The distance between the magnetic dipole target and the magnetic sensor is ( x 0, y 0, z 0) represents the position of the magnetic dipole target. x , y , z () represents the position of the magnetic sensor. m x , m y , m z The magnetic moment vectors of the magnetic dipole target are respectively in x , y , z Projected components on the axis.

4. The magnetic dipole target localization method based on 1D-RCNN according to claim 1, 2, or 3, characterized in that, In step 2, the dataset consisting of the input data and the label data is randomly divided into a training set, a validation set, and a test set, specifically as follows: The dataset consisting of the input data and the label data is randomly divided into a training set, a validation set, and a test set in a ratio of 70:15:

15.

5. A magnetic dipole target localization system based on 1D-RCNN, characterized in that, The magnetic dipole target is located using the method described in any one of claims 1 to 4, wherein the magnetic dipole target location system comprises: The data acquisition module is used to acquire the position data, magnetic moment data, and magnetic induction intensity data of the magnetic dipole target; The data processing module is used to add noise to the magnetic induction intensity data of the magnetic dipole target and use it as input data, use the position data and magnetic moment data of the magnetic dipole target as label data, and randomly divide the dataset composed of the input data and the label data into training set, validation set and test set. The model training module is used to construct a magnetic dipole target localization model based on a 1D-RCNN network, and to train, validate, and evaluate the magnetic dipole target localization model based on the training set, the validation set, and the test set. The target inversion module is used to invert the position and magnetic moment of the magnetic dipole target based on the trained magnetic dipole target localization model.

Images