A multi-physical field-based moving target detection method, device and medium

Abstract

The application discloses a kind of based on multi-physical field's moving target detection method, device and medium, it is related to target detection technical field.The method comprises: before detector and moving target intersection, obtain the magnetic field signal and electrostatic field signal that detector detects moving target under multiple physical fields;Multiple physical fields include magnetic field and electrostatic field;Magnetic field signal and electrostatic field signal are fused, obtain fusion signal, and according to fusion signal, predict the remaining sampling point number before detector reaches zero-crossing point position from current position;According to the sampling rate of detector and the remaining sampling point number, determine the time gap before intersection between detector and moving target at current time;According to the motion parameter of detector and time gap, determine the distance between detector and zero-crossing point position at current time.The method breaks through the static limitation of fixed threshold method, and significantly improves the detection accuracy of moving target.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, in particular to a moving target detection method, device and medium based on multiple physical fields. BACKGROUND

[0002] In the fields of resource exploration, industrial detection and automatic driving, accurate detection of moving targets is a crucial technology.

[0003] Currently, real-time target detection mainly relies on single physical field detection methods, among which electromagnetic wave detection technology and electrostatic field detection technology are mainly used. However, in the moving target detection scene, single magnetic field or electrostatic field detection is easily disturbed by environmental noise, target material and motion state; and the existing distance determination methods generally have limitations: the limitations of fixed threshold method, which roughly corresponds to the distance by presetting the sensor signal threshold, without considering dynamic factors such as target motion speed, approaching angle and environmental electromagnetic interference, resulting in large distance estimation error. SUMMARY

[0004] Therefore, it is necessary to provide a moving target detection method, device and medium based on multiple physical fields, which breaks through the static limitations of the fixed threshold method and significantly improves the detection accuracy of moving targets.

[0005] The present application adopts the following technical solutions:

[0006] The present application provides a moving target detection method based on multiple physical fields, comprising:

[0007] Before the detector intersects with the moving target, the magnetic field signal and the electrostatic field signal of the detector detecting the moving target under multiple physical fields are obtained; the multiple physical fields include magnetic field and electrostatic field;

[0008] The magnetic field signal and the electrostatic field signal are fused to obtain a fused signal, and the number of remaining sampling points of the detector from the current position to the zero-crossing position is predicted according to the fused signal; the zero-crossing position is the position of the detector when the distance between the detector and the moving target is the shortest at a future time;

[0009] According to the sampling rate of the detector and the number of remaining sampling points, the time gap before the detector intersects with the moving target at the current time is determined;

[0010] According to the motion parameters of the detector and the time gap, the distance between the detector and the zero-crossing position at the current time is determined.

[0011] Optionally, the detector and the moving target generate a signal waveform showing a rising trend first and then a falling trend in the intersection motion process, and the position where the detector and the moving target intersect is the intersection point, i.e. the zero-crossing position.

[0012] Optionally, according to the fusion signal, the remaining sampling point number of the probe from the current position to the zero-crossing position is predicted, comprising:

[0013] Performing empirical mode decomposition on the fusion signal to extract a low-frequency topographic feature component;

[0014] Inputting the low-frequency topographic feature component into a pre-trained distance waveform time sequence mapping model to obtain the remaining sampling point number of the probe from the current position to the zero-crossing position; the calculation mode of the distance waveform time sequence mapping model is a matrix multiplication mode or a polynomial calculation mode.

[0015] Optionally, the training process of the distance waveform time sequence mapping model comprises:

[0016] Obtaining a prior data set under multiple modes; the prior data set is the rising data information of the sample probe before intersecting with the sample moving target;

[0017] Taking the remaining sampling point number before reaching the zero-crossing position as the output label, training the pruned long short-term memory network through the prior data set, and testing the trained network until the validation error of the trained network is within a preset range;

[0018] Extracting the parameter matrix of the network with the validation error within the preset range, and converting the network calculation process into a matrix multiplication or a polynomial calculation mode according to the parameter matrix to obtain the distance waveform time sequence mapping model.

[0019] Optionally, obtaining the prior data set under multiple modes comprises:

[0020] Establishing a magnetic field and electrostatic field dual-mode detection model in a digital space;

[0021] Constructing an intersection scene of the sample probe and the sample moving target, including a spatial coordinate system, an initial position and a motion trajectory, and a parameter combination of different speed ranges, different detection axial directions and different intersection angles;

[0022] According to the orthogonal test design method, the magnetic field and electrostatic field dual-mode detection model is simulated with multiple parameter combinations, the sample magnetic field signal and the sample electrostatic field signal in the intersection process of the sample probe and the sample moving target are recorded, and the corresponding distance information, time information and label information of the remaining sampling point number before reaching the zero-crossing position are marked for each sampling point;

[0023] For any kind of parameter combination, the sample magnetic field signal and the sample electrostatic field signal are preprocessed respectively, and the preprocessed sample magnetic field signal and sample electrostatic field signal are fused to obtain a sample fusion signal;

[0024] The sample fusion signal is subjected to modal decomposition to obtain a sample low-frequency topographic feature component;

[0025] The sample low-frequency topographic feature component and the corresponding number of remaining sampling points before reaching the zero-crossing point are determined as a priori data set.

[0026] Optionally, the magnetic field detection model is based on a magnetic dipole model, considering the magnetic moment size and direction of the target object, and a mathematical model of the magnetic field strength changing with distance is established; and the electrostatic field detection model is based on a charge distribution model, considering the charge amount and distribution characteristics of the target object, and a mathematical model of the electric field strength changing with distance is established.

[0027] Optionally, according to the sampling rate of the detector and the number of remaining sampling points, the time gap before the intersection between the detector and the moving target at the current time is determined, comprising:

[0028] The product of the sampling rate of the detector and the number of remaining sampling points is determined as the time gap before the intersection between the detector and the moving target at the current time.

[0029] Optionally, the motion parameter includes velocity; according to the motion parameter of the detector and the time gap, the distance between the detector and the zero-crossing point position at the current time is determined, comprising:

[0030] The product of the velocity of the detector at the current time and the time gap is determined as the distance between the detector and the zero-crossing point position at the current time.

[0031] The present application provides a kind of based on the motion target detection device of multi-physical field, comprising:

[0032] The acquisition module is used to obtain the magnetic field signal and electrostatic field signal detected by the detector to the moving target under the multi-physical field before the intersection between the detector and the moving target;Multi-physical field includes magnetic field and electrostatic field;

[0033] The prediction module is used to fuse the magnetic field signal and electrostatic field signal, obtain fusion signal, and predict the number of remaining sampling points before the detector reaches the zero-crossing point position from the current position according to the fusion signal;Zero-crossing point position is the position of the detector when the shortest distance between the detector and the moving target at future time;

[0034] The calculation module is used to determine the time gap before the intersection between the detector and the moving target at the current time according to the sampling rate of the detector and the number of remaining sampling points;According to the motion parameter of the detector and the time gap, the distance between the detector and the zero-crossing point position at the current time is determined.

[0035] The present application provides a kind of computer readable storage medium, the storage medium stores computer program, the computer program is executed when processor realizes the above-mentioned motion target detection method based on multi-physical field.

[0036] The application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned multi-physical field based moving target detection method when executing the program.

[0037] The above-mentioned at least one technical scheme adopted by the application can achieve the following beneficial effects:

[0038] In the application, the environmental noise is inhibited by the complementary characteristics of the magnetic field disturbance signal and the electrostatic field vector change data in combination with the double physical field cooperative mechanism, the recognition degree of the signal characteristics in a complex environment is improved, and the noise interference of a single physical field is reduced; based on the regularity timing characteristics of the magnetic field disturbance and the electrostatic field vector change in the intersection movement process of the target and the detector, the evolution trend that the signal waveform rises first and then falls, and the intersection point being a zero-crossing point, the signal evolution law before the non-intersection stage is used to realize the dynamic prediction of the zero-crossing point timing; finally, the distance is inversely calculated according to the zero-crossing point timing and the target motion parameters, the method breaks through the static limitation of the fixed threshold method, the detection precision and the response speed of the moving target are significantly improved through dynamic prediction, and an efficient solution is provided for real-time detection and interception of the moving target. BRIEF DESCRIPTION OF DRAWINGS

[0039] The accompanying drawings, which are included to provide a further understanding of the application and constitute a part of this application, illustrate certain illustrative embodiments of the application and together with the description serve to explain the application. In the drawings:

[0040] Figure 1 A flowchart of a multi-physical field based moving target detection method provided by the application is shown in the figure;

[0041] Figure 2 A schematic diagram of a zero-crossing point phenomenon in the intersection process of a detector and a moving target in physical field detection provided by the application is shown in the figure, wherein (a) is a schematic diagram of a zero-crossing point phenomenon in the intersection process of a detector and a moving target in magnetic field detection, and (b) is a schematic diagram of a zero-crossing point phenomenon in the intersection process of a detector and a moving target in electrostatic field detection;

[0042] Figure 3 A schematic diagram of a signal extracted from a main trend component after modal decomposition provided by the application is shown in the figure, wherein (a) is a schematic diagram of an original signal before decomposition, and (b) is a schematic diagram of a main trend component extracted by modal decomposition;

[0043] Figure 4 A modeling schematic diagram of a sample detector and a sample moving target in parallel intersection in a digital space provided by the application is shown in the figure;

[0044] Figure 5A modeling schematic diagram of a digital space sample detector intersecting a 45-degree sample motion target is provided for the present application.

[0045] Figure 6 A schematic diagram of a magnetic field detection motion target zero-crossing intersection trend is provided for the present application.

[0046] Figure 7 A training process schematic diagram of a distance waveform timing mapping model is provided for the present application.

[0047] Figure 8 A flowchart of another motion target detection method based on multi-physical field is provided for the present application.

[0048] Figure 9 A flowchart of another motion target detection method based on multi-physical field is provided for the present application.

[0049] Figure 10 A structural diagram of a motion target detection system based on multi-physical field is provided for the present application.

[0050] Figure 11 A computer device schematic diagram for implementing a motion target detection method based on multi-physical field is provided for the present application. DETAILED DESCRIPTION

[0051] To make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in combination with specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] Magnetic field detection technology principle: magnetic field detection is based on the disturbance mechanism of target objects to the background magnetic field (such as the geomagnetic field). When the target containing ferromagnetic material (such as submarine, mineral resources) enters the detection area, its own magnetic field will change the local space magnetic force line distribution, forming a measurable magnetic anomaly signal. Through high-sensitivity magnetic sensors (such as Hall sensors, TMR sensors) to capture the magnetic field strength, gradient and spatial distribution characteristics, combined with inversion algorithm can analyze the position, shape and motion state of the target. The current technology development trend focuses on improving sensitivity and anti-interference ability.

[0053] The principle of electrostatic field detection technology: electrostatic field detection relies on the disturbance effect of the surface charge distribution of the target object on the local electric field. Its physical basis is Coulomb's law and Gauss's theorem: static electric charges form an electrostatic field in space, and when a charged target (such as a high-speed moving frictionally electrified device) or a conductor approaches, it will cause potential line distortion or field strength vector change. By monitoring the electrostatic field distribution characteristics through the electric field sensor, the charge amount, geometric structure and spatial position of the target can be inferred.

[0054] The defects of the current magnetic field and electrostatic field composite detection technology in target distance determination application mainly reflect the static limitation of the fixed threshold method and the lack of dynamic prediction ability. The existing technology mainly relies on preset signal threshold for distance determination, and does not fully consider dynamic factors such as target motion speed, approaching angle and environmental electromagnetic interference, resulting in large distance estimation error and easy to cause response delay or misjudgment. At the same time, the traditional method only relies on the real-time signal amplitude, and cannot use the time sequence characteristics (such as waveform evolution trend) of the magnetic field disturbance and the electrostatic field vector change in the motion process for forward-looking inference, which is difficult to meet the real-time demand of high-speed moving target detection. In addition, single physical field detection is easily disturbed by environmental noise, target material difference and motion state. Magnetic field and electrostatic field composite detection is a non-contact target detection technology that combines the disturbance effect of magnetic field and the vector distribution characteristics of electrostatic field. The core is to realize real-time distance determination between the detector and the moving target through multi-physical field collaborative perception. Composite detection can significantly improve the robustness of target recognition through multi-physical field information fusion (such as synchronous acquisition of magnetic field anomaly and electrostatic field vector change). However, the existing composite detection technology mainly uses simple threshold superposition or data fusion, and cannot effectively solve the feature complementarity problem between magnetic field and electrostatic field, resulting in low signal recognition degree and weak anti-interference ability in complex environment.

[0055] Based on this, the application provides a kind of motion target detection method, device and medium based on multi-physical field. The method establishes distance waveform time sequence mapping model by excavating the regularity signal waveform (first rise and then fall, intersection point is zero crossing point) generated when target and detector intersect, and uses distance waveform time sequence mapping model to learn magnetic field and electrostatic field data characteristics in non-intersection stage, realizes dynamic prediction of zero crossing point position, so as to inversely solve distance according to zero crossing point time sequence and motion parameters. The method aims to break through the static limitation of fixed threshold method, improve the dynamic response accuracy and speed through time sequence prediction, and at the same time, it inhibits environmental noise by fusing double physical field collaborative mechanism, providing a high-precision and high-robustness solution for real-time detection of moving target.

[0056] The technical solutions provided by the embodiments of the application will be described in detail below with reference to the drawings.

[0057] Figure 1 For the motion target detection method based on multi-physical field in the application, the flowchart specifically includes the following steps:

[0058] S101, before the detector intersects with the moving target, acquires the magnetic field signal and electrostatic field signal of the detector detecting the moving target under multi-physics field; the multi-physics field includes magnetic field and electrostatic field.

[0059] During the intersection of the detector and the moving target, a signal waveform is generated that first rises and then falls. The point where the detector and the moving target intersect is called the intersection point, or zero-crossing point. This invention utilizes the characteristic that a zero-crossing point is generated when the detector and the moving target intersect during magnetic field and electrostatic field detection. Combined with a distance waveform time-series mapping model, the zero-crossing point is predicted by using the information of the rising segment of the signal before intersection. This allows for the prediction of the time interval between the current position and the zero-crossing point. Furthermore, by combining the motion parameters that the detector can acquire, the distance information to the target is calculated, thereby achieving distance determination.

[0060] like Figure 2 As shown, Figure 2 This is a schematic diagram of the zero-crossing phenomenon during the intersection of a detector and a moving target under physical field detection. Figure 2 Figure (a) shows a schematic diagram of the zero-crossing phenomenon during the intersection of the detector and the moving target under magnetic field detection. Figure 2 Figure (b) in the figure is a schematic diagram of the zero-crossing phenomenon during the process of the detector and the moving target intersecting under electrostatic field detection.

[0061] Figure 2 The described scenario is that the probe is flying relative to the target, with the probe positioned below the target. Figure 2 Figure (a) in the diagram describes the magnetic field detection pattern during the convergence process:

[0062] 1. The total field is the mean square value of each axis of X, Y, and Z.

[0063] 2. The X-axis curve represents the X-axis detection signal during the flight of the detector and the target. During the target intersection process, the X-axis gets closer and closer to the target, and the signal rises until it is directly below the target, which is considered as the two intersecting. As the intersection is completed, the signal changes direction and gradually moves away, showing a downward trend.

[0064] 3. Since the Y-axis is parallel to the target and is located in the negative half-axis (the detector is below the target), it continuously approaches the target before they intersect, and its modulus increases as the distance decreases. After the intersection, the two move away from each other, and the model gradually weakens.

[0065] 4. The Z-axis is similar to the X-axis, but the difference lies in the detection axis and sensitivity, resulting in a trend of amplitude and signal first decreasing and then increasing.

[0066] Figure 2 Figure (b) in the figure describes the electrostatic field detection law during the intersection process, which is consistent with the magnetic field detection law, and will not be repeated here.

[0067] Therefore, before the detector intersects with the moving target, signals are collected synchronously by magnetic field sensors and electrostatic field sensors to obtain magnetic field signals and electrostatic field signals respectively. This process ensures that the magnetic field and electrostatic field signals are precisely synchronized in time.

[0068] S102 fuses the magnetic field signal and the electrostatic field signal to obtain a fused signal, and predicts the number of remaining sampling points before the detector reaches the zero-crossing position from the current position based on the fused signal; the zero-crossing position is the position of the detector when the distance between the detector and the moving target is the shortest in the future.

[0069] Optionally, the magnetic field signal and the electrostatic field signal are preprocessed separately, and then the preprocessed magnetic field signal and electrostatic field signal are input into a pre-trained signal fusion model to obtain the fused signal output by the signal fusion model. The preprocessing includes filtering, denoising and normalization; among them, preliminary filtering is performed by hardware circuitry, and a bandpass filter is used to eliminate low-frequency drift and high-frequency noise.

[0070] In one embodiment, predicting the number of remaining sampling points of the detector before it reaches the zero-crossing position from the current position based on the fused signal includes: performing empirical mode decomposition on the fused signal to extract low-frequency topographic feature components; inputting the low-frequency topographic feature components into a pre-trained distance waveform time-series mapping model to obtain the number of remaining sampling points of the detector before it reaches the zero-crossing position from the current position; the calculation mode of the distance waveform time-series mapping model is matrix multiplication mode or polynomial calculation mode.

[0071] Empirical Mode Decomposition (EMD) is used to decompose the fused signal into multiple Intrinsic Mode Function (IME) components. First, the decomposed signals are classified according to time scale (frequency). The low-frequency signal containing signal morphological features is selected as the pre-signal for trend prediction. Then, 25% of the total signal length is taken as the sampling window length, and the backward differential rate of change is calculated for the data within the window. ,when When the value exceeds a certain set threshold (this threshold is negatively correlated with the prediction sensitivity; if the sensitivity is high, a lower value can be used, meaning that a slight upward trend is considered to indicate the presence of a target and its imminent intersection, but this can easily lead to false triggers and requires specific analysis based on the application scenario), the data within the window is taken as the input signal for trend prediction, i.e., the low-frequency morphological feature component, used to predict the number of remaining sampling points before reaching the zero-crossing position. For example... Figure 3 As shown, Figure 3 This is a schematic diagram of the signal from which the main trend components were extracted after mode decomposition. Figure 3 Figure (a) in the diagram is a schematic diagram of the original signal before decomposition.Figure 3 Figure (b) in the diagram is a schematic diagram of the main trend components extracted by mode decomposition.

[0072] In one embodiment, the training process of the distance waveform time-series mapping model includes the following steps:

[0073] S201, acquire prior datasets under multiple modes; the prior datasets are the rising data information before the sample detector and the sample moving target intersect.

[0074] Acquire prior datasets under multiple modes, including: establishing a dual-mode detection model of magnetic and electrostatic fields in digital space; constructing an intersection scenario between the sample detector and the moving sample target, including spatial coordinate system, initial position, and trajectory, as well as parameter combinations for different speed ranges, different detection axes, and different intersection angles; performing multi-parameter combination simulations on the dual-mode detection model based on magnetic and electrostatic fields according to orthogonal experimental design method, recording the sample magnetic field signal and sample electrostatic field signal during the intersection process of the sample detector and the moving sample target, and labeling each sampling point with corresponding distance information, time information, and the number of remaining sampling points before reaching the zero-crossing position; preprocessing the sample magnetic field signal and sample electrostatic field signal for any parameter combination, and fusing the preprocessed sample magnetic field signal and sample electrostatic field signal to obtain the sample fused signal; performing mode decomposition on the sample fused signal to obtain the sample low-frequency morphological feature components; and determining the sample low-frequency morphological feature components and the corresponding number of remaining sampling points before reaching the zero-crossing position as the prior dataset.

[0075] Specifically, using finite element simulation software such as COMSOL Multiphysics and ANSYS, and with the physical parameters (size, material properties, etc.) of the moving target and the performance parameters (sensitivity, frequency response, etc.) of the detector clearly defined, the magnetic and electrostatic fields of different moving targets during the detection process are modeled and simulated. Surrogate models are extracted and imported into systems engineering simulation environments such as MATLAB / Simulink. In the calculation process, the magnetic field detection model is based on a magnetic dipole model, considering the magnitude and direction of the magnetic moment of the target object, and establishing a mathematical model of the magnetic field strength changing with distance; the electrostatic field detection model is based on a charge distribution model, considering the charge quantity and distribution characteristics of the target object, and establishing a mathematical model of the electric field strength changing with distance.

[0076] Secondly, construct the intersection scenario between the sample detector and the moving sample target, including the spatial coordinate system, initial position and trajectory, and set different combinations of parameters such as different speed ranges (e.g., 500m / s-1500m / s), different detection axes (0°-360°), and different intersection angles (0°-90°).

[0077] like Figure 4 andFigure 5 As shown, Figure 4 This is a schematic diagram illustrating the parallel intersection of a sample detector and a moving sample target in digital space. Figure 5 A schematic diagram of the modeling of the sample detector and the moving sample target intersecting at a 45° angle in digital space.

[0078] Next, parametric simulation was performed. Following orthogonal experimental design, a systematic multi-parameter combined simulation was conducted, recording the sample magnetic field and electrostatic field signals during the intersection of the sample detector and the moving sample target. Each sampling point was labeled with corresponding distance information, time information, and zero-crossing position information, such as... Figure 6 As shown, Figure 6 This diagram illustrates the convergence trend of moving targets crossing zero points in magnetic field detection, laying the data foundation for subsequent supervised learning.

[0079] Finally, the sample magnetic field signal and the sample electrostatic field signal are preprocessed (filtered, denoised, and normalized) respectively. The preprocessed sample magnetic field signal and the sample electrostatic field signal are fused to obtain the sample fused signal. The sample fused signal is then subjected to mode decomposition to obtain the sample low-frequency morphological feature components. The sample low-frequency morphological feature components and the corresponding zero-crossing position indices are determined as the prior dataset.

[0080] S202 uses the number of remaining sampling points before reaching the zero point as the output label, trains the pruned Long Short-Term Memory network using the prior dataset, and tests the trained network until the validation error of the trained network is within the preset range.

[0081] The prior dataset was divided into training, validation, and test sets in a 7:2:1 ratio and stored in a structured format, including sample data information, label information, and parameter configurations, forming a zero-crossing target signal prediction training dataset. This dataset was then imported into the network training platform for later use. Figure 7 As shown, Figure 7 This is a schematic diagram of the training process of the distance waveform time-series mapping model.

[0082] The goal of this step is to train a distance waveform time-series mapping model using the zero-crossing target signal prediction training dataset generated previously.

[0083] First, a basic network framework is selected. Considering the limitation of network speed caused by high-speed intersection in actual use, a small-scale parametric network model is selected. The pruned Long Short-Term Memory (LSTM) network model is used as the basic framework. The network structure includes at least an input layer, a hidden layer, a Dropout layer, and an output layer. The number of neurons in the hidden layer is no less than 50. The hidden layer uses the ReLU activation function, and the output layer uses the linear activation function.

[0084] Figure 7 The paper also presents the internal structure of one memory cell of an LSTM, with the current input being... The short-term memory of the previous moment is (Previous output), the long-term memory of the previous moment is (Cellular state), processing of three "gates":

[0085] Forgot the door: View and The decision was made to start from long-term memory. Which information should be discarded? Output a number between 0 and 1. 0 represents "completely forget", and 1 represents "completely retain".

[0086] Input Gate: Simultaneously, view and This determines which new information to store in long-term memory. It consists of two parts: a sigmoid layer that decides which values ​​to update, resulting in... A tanh layer creates a new candidate value vector. These are the new contents that may be added.

[0087] Output gate: based on and ,calculate And decide the next short-term memory. What should it be, that controls the flow from current to long-term memory? How much information is read from the file and output as... .

[0088] Update cell state (long-term memory):

[0089] This is the most crucial step in LSTM: old long-term memories Multiply by the output of the forget gate to forget what you decided to forget, add the product of the input gate's output and the new candidate value, and store the new information to be updated. This gives you the new, updated long-term memory. .

[0090] Output (Short Memory): , to create new long-term memories The output is scaled using the tanh function and then multiplied by the output of the output gate to obtain the final output for this time step. It will also serve as short-term memory input for the next time step.

[0091] Secondly, network data preparation and training optimization are performed. A segment of trend signal before the zero-crossing point is extracted as input, with a length of 100-500 sampling points. The number of remaining sampling points before reaching the zero-crossing point is used as the output label. During training, the training data can be expanded by adding noise, changing the time scale, etc. Mean Squared Error (MSE) is used as the loss function, and the Adam optimizer is used. The initial learning rate is set to 0.001. The training loss and validation loss are monitored in real time, and an early stopping mechanism is set to prevent overfitting until the network training ends.

[0092] Next, the network performance is evaluated using metrics such as prediction accuracy, mean absolute error, and root mean square error. The network's generalization ability is evaluated using the K-fold cross-validation method. The network performance is tested using small-scale experimental equivalent data until the optimization validation error is within the preset range (acceptable range).

[0093] S203, extract the parameter matrix of the network whose verification error is within the preset range, and transform the network calculation process into matrix multiplication or polynomial calculation mode based on the parameter matrix to obtain the distance waveform time-series mapping model.

[0094] The parameter matrix of the network is extracted, and the network calculation process is transformed into matrix multiplication or polynomial calculation mode. The calculation process is further optimized and redundant parameters are simplified. Thus, the construction of the distance waveform time series mapping model is completed.

[0095] In this way, during the actual moving target detection process, the low-frequency topographic feature components obtained after multi-physics field detection are acquired, and the low-frequency topographic feature components are input into the trained distance waveform time-series mapping model. The low-frequency topographic feature components are then used to perform operations with the network's parameter matrix to obtain the position number that the detector needs to collect from the current position to the intersection point.

[0096] S103, based on the detector's sampling rate and the number of remaining sampling points, determine the time gap before the detector and the moving target intersect at the current moment; based on the detector's motion parameters and the time gap, determine the distance between the detector and the zero-crossing position at the current moment.

[0097] The product of the detector's sampling rate and the number of remaining sampling points can be used to determine the time gap before the detector and the moving target intersect at the current moment.

[0098] Optionally, the remaining number of sampling points can be converted into standard time points based on the detector's sampling rate, and the time gap between the current time point and the standard time point can be calculated.

[0099] The detector's motion parameters, including its current velocity and attitude information, are obtained through internal sensors (accelerometers, gyroscopes, etc.). These kinematic parameters are then used to invert the distance between the detector and its zero-crossing position at the current moment. Taking the simplest linear motion model as an example, the detector's velocity at the current moment... With time gap The product of these two factors is used to determine the distance between the detector and the zero-crossing position at the current moment. .

[0100] It should be noted that for variable acceleration motion or attitude correction, a compensation coefficient can be introduced for correction to obtain the distance parameter within a prediction period.

[0101] Throughout the convergence process, the distances measured multiple times are weighted and averaged to improve accuracy. When the distance requirement between the detector and the moving target is met, the response maneuver can be completed to achieve the detection requirements.

[0102] In one embodiment, such as Figure 8 As shown, Figure 8 A flowchart illustrating a moving target detection method based on multiphysics provided by this invention specifically includes:

[0103] In the target detection modeling simulation and training set generation module, the digital domain model is simulated (COMSOL, Modelica, Simulink, etc.) under different intersection speeds, different intersection angles, different sensor axes, and different intersection directions using physical information empirical formulas to obtain magnetic field and electrostatic field signals. Then, the dataset can be expanded using passive noise interference and active noise interference.

[0104] Based on the above data, a zero-crossing target signal prediction training dataset is constructed. The zero-crossing target signal prediction training dataset is preprocessed, and the network is trained based on the preprocessed data. The parameter matrix is ​​extracted to obtain the distance waveform time-series mapping model.

[0105] Thus, before the detector intersects with the moving target, magnetic field and electrostatic field signals are acquired through the detector's sensor array (multi-mode and multi-axial). These signals are then denoised, fused, and subjected to empirical mode decomposition to extract low-frequency topographic feature components. These low-frequency feature components are then input into a range waveform time-series mapping model. Within this model, the position index of the zero-crossing point is output through calculations with the network's parameter matrix. In the zero-crossing-based distance calculation output, the time gap between the zero-crossing point is predicted using the remaining sampling points from which the detector reaches the zero-crossing point, and the vector distance information between the detector and the zero-crossing point is calculated based on the motion parameters input externally to the detector.

[0106] Based on a combined magnetic field and electrostatic field detection model, a dual-mode target detection model (which can be a systems engineering model or a finite element model) is established in digital space. Using existing physical information and empirical formulas combined with simulation to deduce the intersection process, prior information on the target intersection under different velocities, detection axes, and intersection angles is obtained, acquiring prior datasets for multiple modes. Secondly, based on this dataset, a distance waveform time-series mapping model is trained to establish signal feature input segments. and zero-crossing position Mapping relationship between , To preset the input signal segment length, its main function is to input a segment of rising data information before the target intersection. The distance waveform time-series mapping model can perform sequence prediction and obtain zero-crossing position information. Then, the actual collected rising segment information before the target intersection is decomposed to extract low-frequency morphological feature components. The input is fed into the distance waveform time-series mapping model to obtain the number of remaining sampling points before the zero-crossing position after sequence prediction. The zero-crossing time gap is calculated based on the sampling rate information; finally, the appropriate distance from the target position is inverted based on the motion parameters of the detector itself to obtain the final distance information.

[0107] In one embodiment, the present invention also provides a moving target detection method based on multiphysics fields, such as... Figure 9 As shown, it specifically includes:

[0108] S901, Modeling Simulation and Dataset Generation.

[0109] S902, training distance waveform time-series mapping model.

[0110] S903, real-time signal acquisition and preprocessing.

[0111] S904, zero-crossing position prediction.

[0112] S905, real-time distance inversion calculation.

[0113] S906, Results Optimization and Response.

[0114] The continuous ranging results are weighted and averaged to improve accuracy. When the distance meets the requirements, the final response maneuver is triggered.

[0115] In one embodiment, the method designed in this invention can rely on the following multiphysics-based moving target detection system, including a sensor driver interface, signal conditioning, embedded data acquisition and processing, power supply module, etc. Each system adopts an embedded, customizable design, allowing the algorithm to be directly integrated into the main control unit's data analysis module during development. The specific structure of the multiphysics-based moving target detection system is as follows: Figure 10 As shown, the system includes a magnetic detection system, an electrostatic field detection system, and an embedded data acquisition and processing system. The magnetic detection system includes a tunneling magnetoresistance (TMR) sensor, signal filtering, signal conditioning, and analog-to-digital (A / D) conversion. The electrostatic field detection system includes a sensing electrode, charge-to-voltage (Q / V) conversion, signal conditioning, voltage tracking, and A / D conversion. The embedded data acquisition and processing system includes a main control unit data acquisition subsystem, a main control unit data analysis module, and a detection actuator. The main control unit data analysis module is the main carrier of the signal processing algorithm mentioned in the method of this invention.

[0116] In operation, the hardware system of this structure firstly utilizes a TMR sensor as its core component to acquire magnetic field information in real time. The TMR sensor output signal, after hardware processing, is input to an embedded data acquisition and processing system via an A / D conversion module for further processing and analysis of the magnetic field data acquired by the TMR sensor. Secondly, the electrostatic field detection module operates based on electrostatic field theory, monitoring the charge on the sensing electrode in real time. It determines the detector's position relative to the target based on changes in the Q / V conversion voltage and controls the output of the detonation signal. A base voltage power supply adjustment module outputs a small current to detect the base signal and continuously adjusts the base signal current based on the state of the sensing electrode to stabilize its voltage output. This is primarily used to achieve stable operation under different environmental conditions and improve the system's adaptability. This process is repeated until the acquired magnetic signal meets the target's magnetic characteristics and conforms to the magnetic target identification criteria.

[0117] This invention innovatively integrates a data-driven distance waveform timing mapping model with magnetic field and electrostatic field composite detection technology. By constructing a dynamic calculation framework of "zero-crossing prediction-distance inversion," it overcomes the static limitations of traditional ranging methods. Specifically, its core includes: based on the regular timing characteristics of magnetic field disturbances and electrostatic field vector changes during the intersection of the target and the detector (such as the evolution trend of the signal waveform rising first and then falling, with the intersection point being the zero-crossing point), a neural network model is used to learn the signal evolution law in the non-intersection stage to achieve dynamic prediction of the zero-crossing timing; combined with a dual-physics field collaborative mechanism, the complementary characteristics of the magnetic field disturbance signal and electrostatic field vector change data are used to suppress environmental noise, improve the recognition of signal characteristics in complex environments, and reduce noise interference from a single physics field; finally, the distance is calculated in reverse based on the zero-crossing timing and target motion parameters, forming a closed-loop ranging system of "prediction-verification-correction." This method overcomes the static limitations of the fixed threshold method, significantly improves ranging accuracy and response speed through dynamic prediction, and provides an efficient solution for real-time detection and interception of moving targets.

[0118] In addition, this invention takes the information from the detection front end as input, realizes zero-crossing prediction through the distance waveform time-series mapping model, and then combines the motion parameters of the current motion detector itself to finally realize the inversion method of the distance to be detected. The target detection and ranging process improved by this method can effectively improve the ranging accuracy.

[0119] The core advantage of this technical solution lies in its improved ranging accuracy. Traditional ranging methods use fixed signal thresholds, whose accuracy is easily affected by target speed, approach angle, and environmental noise. This solution applies a data-driven neural network model to dynamically predict the zero-crossing position of the signals by learning the temporal characteristics of the magnetic and electrostatic field signals before the targets intersect. This method transforms the ranging criterion from static amplitude judgment to dynamic temporal analysis, aiming to improve the accuracy and stability of the ranging results.

[0120] Regarding response speed, this technology is optimized for detecting high-speed moving targets. Traditional technologies are mostly passive responses, with inherent response delays, which is a limiting factor in high-speed target detection. This scheme utilizes the look-ahead inference capability of sequence prediction networks (such as LSTM) to predict the zero-crossing time gap and inverts the distance by combining it with the detector's motion parameters. This predictive working mode shortens the system response time and is suitable for the real-time ranging requirements of high-speed moving targets.

[0121] This scheme enhances the system's environmental adaptability through a dual-field fusion strategy. In complex electromagnetic environments, single-physical-field detection signals are susceptible to noise interference or changes in target characteristics, leading to decreased reliability. This scheme employs a dual-field fusion strategy, utilizing the complementarity of the two physical mechanisms in their detection principles to mutually verify and correct target information. This composite detection method improves the system's anti-interference capability and target identification accuracy in complex environments.

[0122] This invention represents a breakthrough in target ranging technology within the context of composite detection. It creatively transforms the zero-crossing phenomenon during the intersection of a moving target and the detector into a predictable mathematical model, and integrates neural networks and multi-physics field collaborative mechanisms to solve the dynamic response lag problem caused by the reliance on static thresholds in traditional methods. Utilizing common phenomena in dual-mode composite detection of magnetic and electric fields, the invention uncovers the temporal pattern of the intersection signal's "rise followed by fall," establishing a distance-waveform mapping model. Furthermore, it proposes using sequence prediction networks (such as Long Short-Term Memory networks) to learn the evolution characteristics of magnetic field disturbances and electrostatic field vectors during the non-intersection phase, achieving dynamic prediction of the zero-crossing position. This mechanism overcomes the static limitations of traditional threshold responses, effectively improving the ranging accuracy between the moving target and the detector. Simultaneously, by suppressing environmental noise through composite field collaboration, it provides a new paradigm for composite detection in complex scenarios.

[0123] The novelty of this invention lies in two aspects: a revolutionary prediction paradigm and a multi-physics field fusion mechanism design. Firstly, in terms of prediction paradigm, existing technologies such as magnetic ranging or electrostatic ranging rely on instantaneous signal inversion and fail to utilize temporal characteristics for forward-looking inference. Secondly, in the design of data fusion mechanisms for combined magnetic and electrostatic field detection, traditional combined detection methods often employ simple superposition or threshold fusion, failing to address the issue of feature complementarity between physical fields. This invention and its design method introduce the cross-application concept of "temporal prediction + physical constraint inversion," identifying the evolutionary patterns of the signal sequence before convergence through model training, enabling advance calculation of distance parameters, and possessing robust mutual verification between two independent physical fields. This significantly improves the accuracy of target ranging from both the principle and anti-interference aspects.

[0124] At the engineering application level, this technology has a low deployment threshold. Its core innovation lies primarily in the algorithm, which is compatible with mainstream sensors (such as TMR magnetic sensors and sensing electrodes) without requiring hardware modifications. The method provided by this invention can be directly embedded into existing hardware platforms such as field-programmable gate arrays, microcontrollers, or digital signal processors, resulting in a wide range of applications and low implementation costs. Specifically, the algorithm's functional modules are designed and embedded within the system under development, achieving functional deployment only through software control module upgrades, making implementation easy. This characteristic reduces the implementation cost and deployment difficulty of technology upgrades, enabling its widespread application in existing systems. This invention can be applied in resource exploration and industrial inspection fields.

[0125] When applying the multi-physics-based moving target detection method provided by this invention, it is not necessary to consider... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this invention does not impose any restrictions on it.

[0126] The above describes a multi-physics-based moving target detection method provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding multi-physics-based moving target detection device, which includes:

[0127] The acquisition module is used to acquire the magnetic field signal and electrostatic field signal of the detector detecting the moving target under multi-physics field conditions before the detector intersects with the moving target; the multi-physics field includes magnetic field and electrostatic field.

[0128] The prediction module is used to fuse the magnetic field signal and the electrostatic field signal to obtain a fused signal, and based on the fused signal, predict the number of remaining sampling points before the detector reaches the zero-crossing position from the current position; the zero-crossing position is the position of the detector when the distance between the detector and the moving target is the shortest in the future.

[0129] The calculation module is used to determine the time gap before the detector and the moving target intersect at the current moment based on the detector's sampling rate and the number of remaining sampling points; and to determine the distance between the detector and the zero-crossing position at the current moment based on the detector's motion parameters and the time gap.

[0130] Specific limitations regarding the multiphysics-based moving target detection device can be found in the limitations of the multiphysics-based moving target detection method described above, and will not be repeated here. Each module in the aforementioned multiphysics-based moving target detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0131] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A moving target detection method based on multiphysics is provided.

[0132] The present invention also provides Figure 11 The schematic diagram of the computer device shown is as follows: Figure 11 As shown, at the hardware level, this computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 A moving target detection method based on multiphysics is provided.

[0133] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0134] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

Claims

1. A moving target detection method based on multiphysics, characterized in that, The method includes: Before the detector intersects with the moving target, the magnetic field signal and electrostatic field signal of the detector on the moving target under multi-physics field are acquired; the multi-physics field includes magnetic field and electrostatic field. The magnetic field signal and the electrostatic field signal are fused to obtain a fused signal. Empirical mode decomposition (EMD) is then performed on the fused signal to extract low-frequency topographic feature components. These low-frequency feature components are input into a pre-trained range waveform time-series mapping model to obtain the number of remaining sampling points before the detector reaches its zero-crossing position from its current position. The range waveform time-series mapping model is calculated using either matrix multiplication or polynomial methods. During the intersection process between the detector and the moving target, a signal waveform is generated that first rises and then falls. The point where the detector and the moving target intersect is the intersection point, i.e., the zero-crossing position. The zero-crossing position represents the detector's position at the time when the distance between the detector and the moving target is minimized. Based on the detector's sampling rate and the number of remaining sampling points, determine the time gap before the detector and the moving target intersect at the current moment; Based on the detector's motion parameters and time interval, determine the distance between the detector and the zero-crossing position at the current moment.

2. The method according to claim 1, characterized in that, The training process of the distance waveform time-series mapping model includes: Acquire prior datasets under various modes; the prior datasets are the rising data information of the sample detector before it intersects with the sample moving target; The number of remaining sampling points before reaching the zero point is used as the output label. The pruned long short-term memory network is trained using the prior dataset, and the trained network is tested until the validation error of the trained network is within the preset range. Extract the parameter matrix of the network whose verification error is within a preset range, and transform the network calculation process into matrix multiplication or polynomial calculation mode based on the parameter matrix to obtain the distance waveform time-series mapping model.

3. The method according to claim 2, characterized in that, Obtain prior datasets for various modes, including: Establish a dual-mode detection model for magnetic and electrostatic fields in the digital space; Construct a scenario of intersection between the sample detector and the moving sample target, including the spatial coordinate system, initial position and trajectory, as well as parameter combinations of different speed ranges, different detection axes and different intersection angles; According to the orthogonal experimental design method, a multi-parameter combined simulation was carried out on the dual-mode detection model based on magnetic field and electrostatic field. The magnetic field signal and electrostatic field signal of the sample were recorded during the intersection of the sample detector and the sample moving target. The corresponding distance information, time information and the label information of the number of remaining sampling points before reaching the zero-crossing position were marked for each sampling point. For any combination of parameters, the sample magnetic field signal and the sample electrostatic field signal are preprocessed separately, and the preprocessed sample magnetic field signal and sample electrostatic field signal are fused to obtain the sample fused signal; Modal decomposition is performed on the fused sample signal to obtain the low-frequency morphological feature components of the sample; The low-frequency morphological feature components of the sample and the number of remaining sampling points before reaching the zero-crossing position are determined as the prior dataset.

4. The method according to claim 3, characterized in that, The magnetic field detection model is based on the magnetic dipole model, which considers the magnitude and direction of the magnetic moment of the target object and establishes a mathematical model of the change of magnetic field strength with distance; the electrostatic field detection model is based on the charge distribution model, which considers the charge quantity and distribution characteristics of the target object and establishes a mathematical model of the change of electric field strength with distance.

5. The method according to claim 1, characterized in that, Based on the detector's sampling rate and the number of remaining sampling points, determine the time gap before the detector and the moving target intersect at the current moment, including: The product of the detector's sampling rate and the number of remaining sampling points is used to determine the time gap before the detector and the moving target intersect at the current moment.

6. The method according to claim 1, characterized in that, Motion parameters include velocity; based on the detector's motion parameters and time interval, the distance between the detector and the zero-crossing position at the current moment is determined, including: The product of the detector's current velocity and the time gap is used to determine the distance between the detector and the zero-crossing position at the current moment.

7. A moving target detection device based on multiphysics, characterized in that, include: The acquisition module is used to acquire the magnetic field signal and electrostatic field signal of the detector detecting the moving target under multi-physics field conditions before the detector intersects with the moving target; the multi-physics field includes magnetic field and electrostatic field. The prediction module is used to fuse the magnetic field signal and the electrostatic field signal to obtain the fused signal, and to perform empirical mode decomposition on the fused signal to extract low-frequency morphological feature components. The low-frequency topographic feature components are input into a pre-trained range waveform time-series mapping model to obtain the number of remaining sampling points before the detector reaches the zero-crossing position from its current position. The calculation mode of the range waveform time-series mapping model is either matrix multiplication mode or polynomial calculation mode. During the intersection motion between the detector and the moving target, a signal waveform with a trend of first rising and then falling is generated. The position where the detector and the moving target intersect is the intersection point, i.e., the zero-crossing position. The zero-crossing position is the position of the detector when the distance between the detector and the moving target is the shortest in the future. The calculation module is used to determine the time gap before the detector and the moving target intersect at the current moment based on the detector's sampling rate and the number of remaining sampling points; and to determine the distance between the detector and the zero-crossing position at the current moment based on the detector's motion parameters and the time gap.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images