A magnetic anomaly signal extraction and detection method based on improved EEMD

Abstract

This invention discloses a method for magnetic anomaly signal extraction and detection based on an improved EEMD. By introducing the Teager transient energy operator, a mutation energy gating mechanism, and a dynamic continuous soft mask based on Z-Score and Sigmoid functions, combined with orthogonal basis functions and constant false alarm rate detection, this invention achieves high-fidelity extraction and adaptive detection of magnetic anomaly signals under extremely low signal-to-noise ratios. The proposed mutation energy gating mechanism can directly intercept and eliminate high-frequency pure noise components or low-frequency trend interference in the early decomposition stage, fundamentally avoiding misjudging noise spikes as targets, while significantly reducing the subsequent computational burden on the system. The robust noise standard deviation calculated based on the median absolute deviation effectively suppresses background noise while perfectly preserving the low-frequency time-domain characteristics and time-frequency continuity of the target signal.

Description

Technical Field

[0001] This invention belongs to the field of weak signal processing and magnetic anomaly detection technology. In detection scenarios with complex geomagnetic background noise and strong environmental interference, it introduces transient change energy gating screening and adaptive continuous soft mask mechanism, combined with orthogonal basis matched filtering to optimize signal reconstruction and target detection strategies, to achieve efficient extraction of weak magnetic anomaly features. This maximizes the target detection probability and reduces the false alarm rate of the system in environments with extremely low signal-to-noise ratios. Specifically, it involves a magnetic anomaly signal extraction and detection method based on improved EEMD (Ensemble Empirical Mode Decomposition). Background Technology

[0002] With the continuous development of magnetic sensor technology, magnetic anomaly detection (MAD), as a core technology for intelligent sensing and weak signal processing, has shown great application potential in underwater target detection, unexploded ordnance disposal, geological exploration, and anti-submarine warfare. MAD can achieve highly concealed passive detection by detecting minute disturbances in the Earth's magnetic field caused by ferromagnetic targets. However, traditional weak signal processing methods (such as low-pass filtering and band-pass filtering) suffer from severe waveform distortion when processing non-stationary and nonlinear signals, making it difficult to simultaneously achieve noise removal and target feature preservation, and thus failing to meet the requirements for high-fidelity extraction in complex environments. To overcome these limitations, Empirical Mode Decomposition (EMD) and its derivative algorithms have been introduced into non-stationary signal processing, adaptively decomposing complex signals into multiple modal components to accurately extract hidden target features, especially in nonlinear signal analysis scenarios, further reducing the reliance on traditional frequency domain filters.

[0003] However, despite the advantages of EMD and Ensemble Empirical Mode Decomposition (EEMD) in processing non-stationary signals, the problems in magnetic anomaly signal extraction are becoming increasingly prominent as the actual engineering detection environment becomes more complex. Targets in magnetic anomaly detection are typically far away, resulting in extremely weak signals that are easily overwhelmed by complex geomagnetic background noise, sensor platform motion noise, and environmental electromagnetic interference. Especially in dynamically fluctuating geomagnetic environments, effectively suppressing strong background noise while fully preserving weak target features has become a significant challenge in magnetic anomaly signal processing. The inherent mode aliasing problem in standard EMD algorithms, and the severe interference of complex environmental noise on the target waveform, further increase the complexity of high-fidelity extraction of magnetic anomaly signals.

[0004] Traditional denoising methods based on EEMD rarely consider the continuity of the transient energy change sensing and reconstruction process of signals. They often employ mechanical hard or semi-soft thresholding, which can easily truncate target waveforms with continuous physical characteristics, causing reconstruction distortion. Furthermore, they are prone to misjudging high-frequency spike noise as target signals, leading to an abnormally high false alarm rate. In addition, the environmental adaptability of threshold settings in target detection must be considered. Existing fixed threshold decisions are insufficient to cope with dynamically changing, highly cluttered environments. It is necessary to dynamically adjust the continuous soft mask weighting strategy and adaptive detection threshold to optimize signal reconstruction effectiveness, thereby improving the ultimate detection range and overall detection efficiency of the magnetic anomaly detection system. Summary of the Invention

[0005] To address the technical problems of modal aliasing, high false alarm rate, reconstruction distortion, and poor environmental adaptability in existing technologies, this invention proposes a magnetic anomaly signal extraction and detection method based on an improved EEMD. This invention introduces the Teager transient energy operator, a mutation energy gating mechanism, and a dynamic continuous soft mask based on Z-score and Sigmoid functions, combined with orthogonal basis functions and constant false alarm rate detection, achieving high-fidelity extraction and adaptive detection of magnetic anomaly signals under extremely low signal-to-noise ratios. To achieve the above objectives, this invention adopts the following technical solution:

[0006] In a first aspect, embodiments of this application provide a method for extracting and detecting magnetic anomaly signals based on an improved EEMD, comprising the following steps:

[0007] S1. Input Data Acquisition:

[0008] Obtain discrete observation data from actual detection, including the total number of sampling points observed. sampling frequency , and the three-component sequence actually measured by the fluxgate sensor.

[0009] S2. Calculate the input signal based on the three-component sequence.

[0010] S3. Decompose the input signal using the Ensemble Empirical Mode Decomposition (EEMD) algorithm to obtain multiple intrinsic mode functions (EMFs). ) Quantity.

[0011] S4. Extract each using the Teager Energy Operator (TEO). The transient energy envelope of the component is obtained and smoothed to suppress high-frequency noise.

[0012] S5. Based on the multiple relationship between the global maximum and median of the smoothed transient energy envelope, for the corresponding... The components are subjected to mutation energy gating screening to remove pure noise or trend interference components that do not contain the target features, and retain the effective components that contain mutation features.

[0013] S6. For the components retained through gating, calculate the adaptive noise threshold, and combine Z-Score normalization and the Sigmoid function to construct a continuous soft mask for weighted clipping and reconstruction to obtain the reconstructed enhanced signal.

[0014] S7. Using the orthogonal basis function (OBF) constructed based on the analytical features of magnetic anomalies of the mobile detection platform relative to the stationary magnetic dipole target, the extracted enhanced signal is subjected to multi-scale convolution filtering. The energy is fused by summing the squares of the outputs of each orthogonal branch, and the maximum energy value under all scanning scales is extracted at each discrete sampling point to obtain a one-dimensional matching detection energy sequence.

[0015] S8. Based on the one-dimensional matching detection energy sequence, the adaptive decision threshold is dynamically calculated using the constant false alarm rate (GO-CFAR) algorithm to complete the final magnetic anomaly target detection.

[0016] In one possible implementation, in step S2, the modulus of the three-component sequence actually measured by the fluxgate sensor is calculated to obtain the scalar total magnetic field, and the discrete sequence composed of the scalar total magnetic field values ​​of all sampling points is then obtained. As the input signal sequence:

[0017] (1)

[0018] in, For discrete sampling point numbers, The fluxgate sensor is at the 1st The triaxial magnetic fields measured at each sampling point are mutually orthogonal. axis, shaft and The sequence of magnetic field components in the three directions of the axis. In the first The scalar total magnetic field is obtained by taking the modulus of the three-component sequence of each sampling point.

[0019] In one possible implementation, in step S3, the discrete input signal sequence is... Gaussian white noise sequences with amplitudes conforming to a preset standard deviation ratio were added multiple times. Multiple empirical mode decompositions (EMDs) are performed, and finally, the modal components of corresponding orders obtained from the multiple decompositions are ensemble averaged to obtain the desired result. eigenmode function components of each order :

[0020] (2)

[0021] (3)

[0022] in, The preset average number of sets, The number of iterations in the current loop. , For the first The Gaussian white noise sequence added in the next iteration; For the first Noisy auxiliary signal of secondary construction To The first one obtained by EMD decomposition First-order modal components, The order of the component is denoted as .

[0023] In one possible implementation, step S4 involves calculating the components of each order of intrinsic mode function. In the sampling points Discrete output of the Teager energy operator ,right The absolute value is determined by setting the window length to 1. The Gaussian smoothing function is used for smoothing, followed by square root dimensionality reduction to obtain the smoothed transient energy envelope. :

[0024] (4)

[0025] (5)

[0026] Here, Gaussian represents the Gaussian smoothing function.

[0027] In one possible implementation, in step S5, each order is calculated. Given the global maximum and median of the transient energy envelope of the components, establish corresponding abrupt energy gating conditions:

[0028] (6)

[0029] in, For the first Step The discrete sequence formed by the transient energy envelopes of the components at all sampling points. and They represent the first Step transient energy envelope of the component The global maximum and median, This is a preset mutation energy threshold multiple. To prevent computational overflow, a minimal constant is used. .

[0030] The intrinsic mode function components of each order are screened using a mutation energy gating condition: if the mutation energy gating condition is met, it indicates that the current component does not contain significant local energy mutations and belongs to pure noise or extremely low-frequency trend interference components, which are directly set to zero and removed, skipping the subsequent calculation of that component; if the mutation energy gating condition is not met, it is determined that the component contains effective mutation features and is retained to proceed to the next step. This mechanism can significantly reduce the false alarm rate and computational load of the system.

[0031] In one possible implementation, in step S6, for the valid components retained through gating screening... Calculate the transient energy envelope. The median absolute deviation is used to estimate the magnitude of the pure background noise fluctuation unaffected by abnormal target spikes, i.e., the system robust noise standard deviation. .

[0032] (7)

[0033] The robust noise standard deviation Multiply by dynamic weighting factor This is then superimposed on the energy median to calculate an adaptive noise threshold suitable for the current environment. .

[0034] (8)

[0035] in, This is a weighting factor that is dynamically adjusted with the order.

[0036] The transient energy envelope value of the current sampling point Subtract the adaptive noise threshold The energy deviation is obtained, and then the energy deviation is divided by the system robust noise standard deviation and the minimum constant. The sum of these values ​​is used to calculate the Z-score normalized distance of each point in the energy envelope relative to the threshold. And substitute it into the Sigmoid activation function to generate continuous soft mask weights. For the portion of the calculation result less than zero, a semi-soft threshold pruning (i.e., forced setting to 0) is performed, while for the portion greater than or equal to zero, a non-linear smooth transition mapping is performed.

[0037] (9)

[0038] (10)

[0039] in, This is the normalized spatial kurtosis parameter.

[0040] Using this continuous soft mask weight sequence For the retained effective components Values ​​at each sampling point Perform point-by-point weighted product to obtain the denoised and cropped components. :

[0041] (11)

[0042] Finally, all weighted effective components Along order Summing in each direction yields the result of the first... Enhanced signal at each sampling point Enhanced signal at all sampling points This results in a reconstructed enhanced signal with wide bandwidth preservation and an extremely high signal-to-noise ratio. :

[0043] (12)

[0044] In one possible implementation, in step S7, based on the analytical characteristics of the magnetic anomaly signal (i.e., the Anderson magnetic dipole model waveform characteristics) collected when the mobile detection platform moves at a constant linear speed past a stationary typical ferromagnetic target, considering that the unknown platform speed and minimum intersection distance in actual detection will cause changes in the width of the received target signal on the time axis, a time scale factor is introduced. and define the normalized time variable. Construct a set of mutually orthogonal basis function waveforms The reconstructed enhanced signal With at a given scale The basis function waveforms Perform convolution filtering, then square the convolution outputs corresponding to the waveforms of each basis function, add them together, and calculate the current scale. The overall output energy below At each discrete sampling point At this point, it iterates through all preset scan scales. Compare the combined output energies across all scanning scales, and select the maximum value as the matching detection energy at that sampling point. The matching detection energy of all sampling points constitutes a one-dimensional matching detection energy sequence. :

[0045] (13)

[0046] (14)

[0047] (15)

[0048] (16)

[0049] (17)

[0050] in, This represents the discrete convolution operation. Indicated in scale The basis function waveforms are as follows. This indicates that at each discrete sampling point At different scales The maximum energy value under the specified conditions.

[0051] In one possible implementation, in step S8, the one-dimensional matching detection energy sequence extracted in step S7 is... With the current discrete sampling points to be detected A sliding window is set up around the center. Within this window, the average power estimates of the local background noise in the training units located to the left and right of the discrete sampling point to be detected are calculated respectively. and The training unit is a data sequence interval within a sliding window specifically used to estimate the local environmental background noise, with a single-sided length of [missing information]. To prevent energy leakage at the target edge from causing a masking effect, the larger of the power estimates on both sides is selected as the local noise reference power at the current moment. According to the preset system allowable false alarm probability and training unit length Calculate the corresponding threshold multiplier factor. The threshold multiplier factor is multiplied by the local noise reference power to generate the adaptive decision threshold for the current sampling point. Finally, the adaptive decision threshold is... Matching detection energy with corresponding sampling points Compare point by point:

[0052] (18)

[0053] (19)

[0054] (20)

[0055] If at a certain discrete sampling point Satisfying If the magnetic anomaly is detected at the actual observation time corresponding to the sampling point, it is determined that the magnetic anomaly target was successfully detected; otherwise, it is determined that the current sampling point is only experiencing fluctuations in environmental background noise.

[0056] Secondly, embodiments of this application provide a magnetic anomaly signal extraction and detection system based on an improved EEMD, comprising the following modules:

[0057] Data acquisition module: Used to acquire discrete observation data from actual detection, including the total number of sampling points observed. sampling frequency , and the three-component sequence actually measured by the fluxgate sensor.

[0058] Input signal calculation module: Calculates the input signal based on a three-component sequence.

[0059] Decomposition Module: This module uses the Ensemble Empirical Mode Decomposition (EEMD) algorithm to decompose the input signal, obtaining multiple intrinsic mode functions (EMFs). ) Quantity.

[0060] Gated filtering module: Utilizes the Teager Energy Operator (TEO) to extract each The transient energy envelope of the component is obtained and smoothed to suppress high-frequency noise. Based on the smoothed transient energy envelope, the corresponding... The components are screened using mutation energy gating.

[0061] Signal reconstruction module: For the components retained through gating, calculate the adaptive noise threshold, and combine Z-Score normalization and the Sigmoid function to construct a continuous soft mask for weighted clipping and reconstruction to obtain the reconstructed enhanced signal.

[0062] Fusion module: Using orthogonal basis functions (OBF), the extracted enhanced signal is subjected to multi-scale convolution filtering to obtain a one-dimensional matching detection energy sequence.

[0063] Detection module: Based on the one-dimensional matching detection energy sequence, combined with the constant false alarm rate (GO-CFAR) algorithm, the adaptive decision threshold is dynamically calculated to complete the final detection of magnetic anomaly targets.

[0064] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory;

[0065] The memory is used to store computer programs.

[0066] When the processor executes the program stored in the memory, it implements any of the magnetic anomaly signal extraction and detection methods described in this application.

[0067] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the magnetic anomaly signal extraction and detection methods described in this application.

[0068] Fifthly, embodiments of this application provide a computer program product containing instructions that, when run on a computer, cause the computer to execute any of the magnetic anomaly signal extraction and detection methods described in this application.

[0069] Compared with the prior art, the present invention has the following technical effects:

[0070] (1) A mutation energy gating mechanism is proposed. By utilizing the relationship between the global maximum value and the median of the energy envelope, high-frequency pure noise components or low-frequency trend interference can be directly intercepted and eliminated in the early decomposition stage. This fundamentally avoids misjudging noise spikes as targets and significantly reduces the subsequent computational burden of the system.

[0071] (2) This method abandons the traditional hard thresholding method and innovatively adopts robust noise standard deviation calculated based on the absolute deviation of the median. It also designs a continuous soft mask combining Z-Score normalization and the Sigmoid activation function. This method achieves the processing of various orders of noise levels. The smooth weighted cropping of components effectively suppresses background noise while perfectly preserving the low-frequency time-domain characteristics and time-frequency continuity of the target signal.

[0072] (3) By deeply integrating orthogonal basis function (OBF) multi-scale matched filtering with constant false alarm rate (CFAR) technology, the system can dynamically calculate the local environmental benchmark and adaptively adjust the detection threshold based on the real-time noise power distribution on both sides of the target edge. This mechanism effectively overcomes the failure problem of traditional fixed threshold in non-stationary clutter environments and still has excellent detection performance under extremely low signal-to-noise ratio and extreme intersection distance. Attached Figure Description

[0073] Figure 1 This is a flowchart of a magnetic anomaly signal extraction and detection method based on an improved EEMD;

[0074] Figure 2 This is a spatial geometric relationship model diagram for magnetic anomaly target detection in an embodiment of the present invention;

[0075] Figure 3 The graph shows the change in detection probability as a function of power signal-to-noise ratio for each method. Detailed Implementation

[0076] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0077] Example:

[0078] like Figure 1 As shown, the magnetic anomaly signal extraction and detection method based on the improved EEMD in this embodiment is based on orthogonal basis matched filtering and constant false alarm rate detection technology. This model maximizes the signal-to-noise ratio of the extracted signal, maximizes the detection probability, and minimizes the false alarm rate, thus obtaining the optimal detection scheme.

[0079] Including the following steps:

[0080] S1. Input data acquisition.

[0081] S2. Calculate the input magnetic anomaly signal.

[0082] Furthermore, step S2 includes the following steps:

[0083] S2-1 Calculates the measured background scalar total field;

[0084] S2-2 Calculates and simulates the target magnetic anomaly characteristic signal;

[0085] S2-3 mixes to generate the input magnetic anomaly signal.

[0086] S3. Extract intrinsic mode functions using ensemble empirical mode decomposition (EEMD).

[0087] Furthermore, step S3 includes the following steps:

[0088] S3-1 constructs an analysis signal containing auxiliary white noise;

[0089] S3-2 performs EMD decomposition on the noisy auxiliary signal;

[0090] S3-3 Set-averaged reconstructed stable eigenmode functions Quantity.

[0091] S4. Extract the envelope using the Teager Energy Operator (TEO) and perform smoothing.

[0092] S5, Mutation Energy Gated Screening.

[0093] S6, Adaptive noise threshold and continuous soft mask weighted reconstruction.

[0094] Furthermore, step S6 includes the following steps:

[0095] S6-1 Calculates the robust noise standard deviation and adaptive threshold of the system;

[0096] S6-2 constructs a continuous soft mask for Z-Score;

[0097] S6-3 performs weighted cropping and reconstruction on the retained intrinsic mode function components.

[0098] S7. Use orthogonal basis functions (OBF) for matched filtering.

[0099] S8. Combine constant false alarm rate (CFAR) to complete target detection.

[0100] This embodiment is applicable to the verification and evaluation of magnetic anomaly detection algorithms based on complex geomagnetic backgrounds. In this embodiment, the spatial geometric relationship model for magnetic anomaly target detection, on which the simulated magnetic anomaly characteristic signals are based, is as follows: Figure 2 As shown. In this model, a fluxgate sensor device is first used to collect a complex geomagnetic background noise sequence in actual space. Then, based on... Figure 2 The relative motion relationship and physical model between the sensor and the target are shown, and a simulated magnetic anomaly characteristic signal is calculated and generated. The measured geomagnetic background noise and the simulated magnetic anomaly characteristic signal are linearly superimposed according to a preset signal-to-noise ratio to construct a noisy hybrid magnetic anomaly data sequence, which is used as the final algorithm input for subsequent mode decomposition, feature extraction, and target determination. This embodiment can accurately evaluate the detection performance of the algorithm in scenarios with high dynamics and extremely low signal-to-noise ratio requirements. This embodiment makes the following assumptions:

[0101] (1) Based on Figure 2 The physical model assumes that the magnetic dipole target in real space remains stationary, while the detection platform equipped with a magnetic sensor moves in uniform linear motion relative to the target.

[0102] (2) Ignore the extremely low frequency baseline fluctuations caused by the nonlinear drift of the actual detection platform (such as UAV, AUV) itself, or assume that they have been filtered out by a high-pass filter in the early stage of acquisition.

[0103] (3) Assume that the actual ambient background noise collected mainly consists of white noise that conforms to a Gaussian distribution and noise with a high degree of uniformity. It is composed of fractal noise superposition with spectral characteristics.

[0104] Specifically, the magnetic anomaly signal extraction and detection method based on the improved EEMD in this embodiment includes the following steps:

[0105] S1. Input data acquisition.

[0106] The acquired input data and related physical parameters specifically include: the total number of sampling points observed by the system. sampling frequency Corresponding time series The physical parameters required to generate the simulation scene: the equivalent magnetic dipole moment of the stationary target. The linear motion speed of the detection platform equipped with a magnetic sensor The time it takes for the magnetic sensor to reach the nearest observation point to the target The closest intersection distance between the magnetic sensor and the stationary target Vacuum permeability Preset target power signal-to-noise ratio The fluxgate sensor measured the three-component sequence of the geomagnetic background. ;

[0107] S2. Calculate the input magnetic anomaly signal.

[0108] Furthermore, step S2 includes the following steps:

[0109] S2-1 Calculates the measured background scalar total field;

[0110] S2-2 Calculates and simulates the target magnetic anomaly characteristic signal;

[0111] S2-3 mixes to generate the input magnetic anomaly signal.

[0112] And in step S2-1:

[0113] The three-component sequence of the geomagnetic background actually measured by the fluxgate sensor is in the first... The value of each sampling point Perform the modulo operation to obtain the first... scalar total geomagnetic field at each sampling point :

[0114] (twenty one)

[0115] A discrete sequence is constructed based on the scalar total geomagnetic field values ​​of all sampling points. .

[0116] In step S2-2:

[0117] Assuming the simulated target (magnetic dipole) remains stationary, based on the uniform linear motion model of the detection platform, the magnetic sensor operates at discrete sampling points. corresponding time Actual displacement relative to the target Instantaneous distance from magnetic sensor to stationary target Calculate using the following formula:

[0118] (twenty two)

[0119] (twenty three)

[0120] By combining the physical characteristic equations of a magnetic dipole, the magnetic sensor's measurement of the first [value] along the motion trajectory is obtained. Simulated ideal magnetic anomaly target signal at each sampling point :

[0121] (twenty four)

[0122] A discrete sequence is constructed based on the simulated ideal magnetic anomaly target signal values ​​from all sampling points. .

[0123] In step S2-3:

[0124] Based on the preset target power signal-to-noise ratio For the simulated ideal magnetic anomaly target signal sequence After performing the necessary amplitude scaling, it is then compared with the measured background scalar total field sequence. Linear superposition generates a noisy mixed input magnetic anomaly signal for subsequent algorithm verification. :

[0125] (25)

[0126] S3. Extract intrinsic mode functions using ensemble empirical mode decomposition (EEMD).

[0127] Furthermore, step S3 includes the following steps:

[0128] S3-1 constructs an analysis signal containing auxiliary white noise;

[0129] S3-2 performs EMD decomposition on the noisy auxiliary signal;

[0130] S3-3 Set-averaged reconstructed stable eigenmode functions Quantity.

[0131] In step S3-1:

[0132] Set the ensemble number of empirical mode decompositions to be In the Second-rate In the decomposition loop, the discrete sequence magnetic anomaly signal generated in step S2 is... Add a Gaussian white noise sequence with an amplitude that matches a preset standard deviation ratio. A noisy auxiliary signal was constructed for this decomposition. :

[0133] (26)

[0134] In step S3-2:

[0135] Noisy auxiliary signal for construction Perform standard Empirical Mode Decomposition (EMD), extracting local features sequentially from high frequency to low frequency to obtain the modal components corresponding to each order of the current cyclic decomposition. The maximum order of the decomposition and extraction is set to be... Right now .

[0136] In step S3-3:

[0137] Repeat steps S3-1 and S3-2 until all settings are completed. This is a decomposition cycle. Utilizing the zero-mean statistical property of white noise, all... The corresponding order obtained from the second decomposition The modal components are ensemble-averaged to cancel out the effects of the added auxiliary white noise, ultimately obtaining stable, aliased eigenmode function components of each order with clear physical frequency band meaning. :

[0138] (27)

[0139] S4. Extract each using the Teager Energy Operator (TEO). The transient energy envelope of the component.

[0140] Calculate the components of each order of intrinsic mode function In the sampling points Discrete output of the Teager energy operator To suppress the amplification effect of high-frequency white noise, The absolute value is determined by setting the window length to 1. The transient energy envelope is obtained by smoothing the data using a Gaussian smoothing function and then performing square root dimensionality reduction. :

[0141] (28)

[0142] (29)

[0143] In equation (29), gaussian is the Gaussian smoothing function.

[0144] S5, Mutation Energy Gated Screening.

[0145] Calculate each order The global maximum and median of the transient energy envelope of the components are determined, and a sudden energy gating condition is established:

[0146] (30)

[0147] in, For the first Step The discrete sequence formed by the transient energy envelopes of the components at all sampling points. and They represent the first Step transient energy envelope of the component The global maximum and median, As a preset mutation energy threshold multiple, this embodiment preferably uses Set it to 2.5. To prevent computational overflow, a minimal constant is used. If this condition is true, it means that... If a component does not contain significant local energy abrupt changes and is considered pure noise or trend interference, it is directly set to zero and removed, and subsequent calculations of that component are skipped. If the gating condition is not met, the component is determined to contain effective abrupt change features, is retained, and enters the adaptive mask weighting process, thereby significantly reducing the false alarm rate and computational load of the system.

[0148] S6, Adaptive noise threshold and continuous soft mask weighted reconstruction.

[0149] Furthermore, step S6 includes the following steps:

[0150] S6-1 Calculates the robust noise standard deviation and adaptive threshold of the system;

[0151] S6-2 constructs a continuous soft mask for Z-Score;

[0152] S6-3 performs weighted cropping and reconstruction on the retained intrinsic mode function components.

[0153] In step S6-1:

[0154] For the valid components retained through gating screening Based on its transient energy envelope The median absolute deviation calculation system robust noise standard deviation And use this to calculate the adaptive noise threshold. :

[0155] (31)

[0156] (32)

[0157] in, This is a weighting factor that is dynamically adjusted with the order.

[0158] In step S6-2:

[0159] The energy deviation is obtained by subtracting the adaptive noise threshold from the transient energy envelope value of the current sampling point. This energy deviation is then divided by the sum of the system robust noise standard deviation and the minimum constant to calculate the Z-score normalized distance of each point in the energy envelope relative to the threshold. And substitute it into the Sigmoid activation function to generate continuous soft mask weights. For the portion of the calculation result less than zero, a semi-soft threshold pruning (i.e., forcibly setting it to 0) is performed; for the portion greater than or equal to zero, a non-linear smooth transition mapping is applied.

[0160] (33)

[0161] (34)

[0162] in, In this embodiment, the normalized spatial steepness parameter is set to 4.

[0163] In step S6-3:

[0164] Using this soft mask weight For the retained effective components Values ​​at each sampling point Perform point-by-point weighted product to obtain the denoised and cropped components. Finally, all weighted effective components are... Summing along the order direction yields the th order. Enhanced signal at each sampling point Enhanced signal at all sampling points This results in an enhanced signal that retains a wide bandwidth and has an extremely high signal-to-noise ratio. :

[0165] (35)

[0166] (36)

[0167] S7. Use orthogonal basis functions (OBF) for matched filtering.

[0168] Based on the analytical characteristics of magnetic anomaly signals generated by typical ferromagnetic targets undergoing uniform linear motion (i.e., the waveform characteristics of the Anderson magnetic dipole model), a time scale factor is introduced to adapt to the variation in the target signal's time width caused by unknown target velocity and intersection distance in actual detection. and define the normalized time variable. Construct a set of mutually orthogonal basis function waveforms The enhanced magnetic anomaly signal reconstructed in step S6 With at a given scale The basis function waveforms Perform convolution filtering, then square the convolution outputs corresponding to the waveforms of each basis function, add them together, and calculate the current scale. The overall output energy below At each discrete sampling point At this point, it iterates through all preset scan scales. Compare the combined output energy across all scanning scales. The maximum value among them is selected as the matching detection energy sequence at that sampling point. The matching detection energy of all sampling points constitutes a one-dimensional matching detection energy sequence. :

[0169] (37)

[0170] (38)

[0171] (39)

[0172] (40)

[0173] (41)

[0174] in, This represents the discrete convolution operation. Indicated in scale The basis function waveforms are as follows. This indicates that at each discrete sampling point At different scales The maximum energy value under the specified conditions.

[0175] S8. Combine constant false alarm rate (CFAR) to complete target detection.

[0176] Energy sequences for one-dimensional matching detection Based on the current sampling point to be detected A sliding window is set up around the center. Within this window, the average power estimates of the local background noise in the training units to the left and right of the test point are calculated respectively. and The training unit is a data sequence interval within a sliding window specifically used to estimate the local environmental background noise, with a one-sided length of . To prevent energy leakage at the target edge from causing a masking effect, the larger of the power estimates on both sides is selected as the local noise reference power at the current moment. According to the preset system allowable false alarm probability and training unit length Calculate the corresponding threshold multiplier factor. Threshold multiplier factor Multiply by the local noise reference power to generate the adaptive decision threshold for the current sampling point. Finally, the adaptive decision threshold is... Matching detection energy with corresponding sampling points Compare point by point:

[0177] (42)

[0178] (43)

[0179] (44)

[0180] If at a certain discrete sampling point Satisfying If the system detects a magnetic anomaly at the actual observation time corresponding to the sampling point, it is determined that the system has successfully detected the magnetic anomaly target; otherwise, it is determined that the current sampling point is only experiencing fluctuations in environmental background noise.

[0181] like Figure 3 The figure shows Monte Carlo simulation curves comparing the target detection probability (Pd) of the proposed TEO-based continuous soft thresholding method with traditional EMD, EEMD, and AWF adaptive window hard thresholding methods under different power signal-to-noise ratio (Power SNR) environments. The experimental test range is -20 dB to -6 dB. The following significant experimental results can be obtained from the figure:

[0182] In the extremely low signal-to-noise ratio range of -20 dB to -14 dB, the detection performance of traditional methods drops precipitously, while the method of this invention always maintains its leading position, demonstrating an absolute advantage in harsh environments with extremely low signal-to-noise ratios.

[0183] Taking -18 dB as an example, the detection probabilities of traditional EMD and EEMD are only about 27% and 42% respectively, and the AWF adaptive window hard threshold method is about 48%. However, the detection probability of the TEO continuous soft threshold method proposed in this invention is close to 70%, which shows extremely strong anti-noise interference capability.

[0184] When the SNR is increased to -16 dB, the detection probability of the method of this invention has exceeded 85%, while the AWF adaptive window hard threshold method, which performs second best, is less than 80%, and the traditional EMD method is only about 45%.

[0185] With the improvement of signal-to-noise ratio, the detection probability curve of this invention rises most rapidly. At -12 dB, the method of this invention has already achieved a near 100% full probability detection state; while the traditional EMD method still hovers around 70% and cannot converge at -6 dB, and the EEMD and AWF adaptive window hard threshold methods need to reach around -7 dB to barely achieve a similar detection level. This shows that this invention is the first to achieve convergence of high probability detection.

[0186] This application also provides a magnetic anomaly signal extraction and detection system based on an improved EEMD, comprising the following modules:

[0187] Data acquisition module: Used to acquire discrete observation data from actual detection, including the total number of sampling points observed. sampling frequency , and the three-component sequence actually measured by the fluxgate sensor.

[0188] Input signal calculation module: Calculates the input signal based on a three-component sequence.

[0189] Decomposition Module: This module uses the Ensemble Empirical Mode Decomposition (EEMD) algorithm to decompose the input signal, obtaining multiple intrinsic mode functions (EMFs). ) Quantity.

[0190] Gated filtering module: Utilizes the Teager Energy Operator (TEO) to extract each The transient energy envelope of the component is calculated and smoothed to suppress high-frequency noise. Based on the multiple relationship between the global maximum and median of the smoothed transient energy envelope, the corresponding... The components are subjected to mutation energy gating screening to remove pure noise or trend interference components that do not contain the target features, and retain the effective components that contain mutation features.

[0191] Signal reconstruction module: For the components retained through gating, calculate the adaptive noise threshold, and combine Z-Score normalization and the Sigmoid function to construct a continuous soft mask for weighted clipping and reconstruction to obtain the reconstructed enhanced signal.

[0192] Fusion Module: The orthogonal basis function (OBF) constructed based on the analytical features of magnetic anomalies of the mobile detection platform relative to the stationary magnetic dipole target is used to perform multi-scale convolution filtering on the extracted enhanced signal. The energy is fused by summing the squares of the outputs of each orthogonal branch, and the maximum energy value under all scanning scales is extracted at each discrete sampling point to obtain a one-dimensional matching detection energy sequence.

[0193] Detection module: Based on the one-dimensional matching detection energy sequence, combined with the constant false alarm rate (GO-CFAR) algorithm, the adaptive decision threshold is dynamically calculated to complete the final detection of magnetic anomaly targets.

[0194] This application also provides an electronic device, including a processor and a memory.

[0195] The memory is used to store computer programs.

[0196] When the processor executes a program stored in the memory, it implements any of the methods described in this application.

[0197] In one possible implementation, the electronic device of this application embodiment further includes a communication interface and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0198] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0199] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0200] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0201] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0202] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements any of the methods described in this application.

[0203] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the methods described in this application.

[0204] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0205] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0206] The various embodiments in this specification are described in a related manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other.

[0207] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A method for extracting and detecting magnetic anomaly signals based on an improved EEMD, characterized in that, Includes the following steps: S1. Input Data Acquisition: Obtain discrete observation data from actual detection, including the total number of sampling points observed. sampling frequency , and the three-component sequence actually measured by the fluxgate sensor; S2. Calculate the input signal based on the three-component sequence; S3. The input signal is decomposed using the ensemble empirical mode decomposition (EEMD) algorithm to obtain multiple intrinsic mode functions. Quantity; S4. Extract each energy using the Teager energy operator. The transient energy envelope of the component is obtained and smoothed to suppress high-frequency noise; S5. Based on the smoothed transient energy envelope, for the corresponding Mutation energy-gated screening of components; S6. For the components retained through gating, calculate the adaptive noise threshold, and combine Z-Score normalization and Sigmoid function to construct a continuous soft mask for weighted clipping and reconstruction to obtain the reconstructed enhanced signal. S7. Using orthogonal basis functions, multi-scale convolution filtering is performed on the extracted enhanced signal to obtain a one-dimensional matching detection energy sequence. S8. Based on the one-dimensional matching detection energy sequence, combined with the maximum constant false alarm rate algorithm, the adaptive decision threshold is dynamically calculated to complete the final magnetic anomaly target detection.

2. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 1, characterized in that, In step S2, the modulus of the three-component sequence actually measured by the fluxgate sensor is calculated to obtain the scalar total magnetic field, and the discrete sequence composed of the scalar total magnetic field values ​​of all sampling points is then obtained. As the input signal sequence: （1） in, For discrete sampling point numbers, The fluxgate sensor is at the 1st The samples measured at each sampling point are mutually orthogonal in a spatial rectangular coordinate system. axis, shaft and The sequence of magnetic field components in the three directions of the axis. In the first The scalar total magnetic field is obtained by taking the modulus of the three-component sequence of each sampling point.

3. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 2, characterized in that, In step S3, the discrete input signal sequence is... Gaussian white noise sequences with amplitudes conforming to a preset standard deviation ratio were added multiple times. Multiple empirical mode decompositions are performed, and finally, the modal components of corresponding orders obtained from the multiple decompositions are ensemble averaged to obtain... eigenmode function components of each order : （2） （3） in, The preset average number of sets, This represents the current iteration number of the loop. For the first The Gaussian white noise sequence added in the next iteration; For the first Noisy auxiliary signal of secondary construction To The first one obtained by EMD decomposition First-order modal components, The order of the component is denoted as .

4. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 3, characterized in that, In step S4, the components of each eigenmode function are calculated. In the sampling points Discrete output of the Teager energy operator ,right The absolute value is determined by setting the window length to 1. The Gaussian smoothing function is used for smoothing, followed by square root dimensionality reduction to obtain the smoothed transient energy envelope. : （4） （5） Here, Gaussian represents the Gaussian smoothing function.

5. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 4, characterized in that, In step S5, each order is calculated. The global maximum and median of the transient energy envelope of the components are used to establish corresponding abrupt energy gating conditions: （6） in, For the first Step The discrete sequence formed by the transient energy envelopes of the components at all sampling points. and They represent the first Step transient energy envelope of the component The global maximum and median, This is a preset mutation energy threshold multiple. A very small constant to prevent computational overflow; The intrinsic mode function components of each order are screened by the mutation energy gating condition: if the mutation energy gating condition is met, the current component is set to zero and removed, and the subsequent calculation of the component is skipped; if the mutation energy gating condition is not met, the component is determined to contain a valid mutation feature and is retained to proceed to the next step.

6. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 5, characterized in that, In step S6, the effective components retained after gating are... Calculate the transient energy envelope. The median absolute deviation is used to estimate the magnitude of the pure background noise fluctuation unaffected by abnormal target spikes, i.e., the system robust noise standard deviation. The robust noise standard deviation is multiplied by a dynamic weighting factor and then added to the energy median to calculate the adaptive noise threshold applicable to the current environment. The energy deviation is obtained by subtracting the adaptive noise threshold from the transient energy envelope value of the current sampling point. The energy deviation is then divided by the sum of the system robust noise standard deviation and the minimum constant to calculate the Z-Score normalized distance of each point of the energy envelope relative to the threshold. This distance is then substituted into the Sigmoid activation function to generate continuous soft mask weights. For the part of the calculation result that is less than zero, a semi-soft threshold is pruned, i.e., forced to 0. For the part that is greater than or equal to zero, a nonlinear smooth transition mapping is performed. Using this continuous soft mask weight sequence, the values ​​of the retained effective components at each sampling point are weighted and multiplied point by point to obtain the noise-reduced and cropped components; Finally, summing all the weighted effective components along the order direction yields the [number of weighted components]. The enhanced signals at each sampling point constitute the reconstructed enhanced signal.

7. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 6, characterized in that, In step S7, a time scale factor is introduced based on the waveform characteristics of the Anderson magnetic dipole model. and define the normalized time variable. Construct a set of mutually orthogonal basis function waveforms The reconstructed enhanced signal With at a given scale The basis function waveforms Perform convolution filtering, then square the convolution outputs corresponding to the waveforms of each basis function, add them together, and calculate the current scale. The overall output energy below At each discrete sampling point At this point, it iterates through all preset scan scales. Compare the combined output energies across all scanning scales, and select the maximum value as the matching detection energy at that sampling point. The matching detection energy of all sampling points constitutes a one-dimensional matching detection energy sequence. : （13） （14） （15） （16） （17） in, This represents the discrete convolution operation. Indicated in scale The basis function waveforms are as follows. This indicates that at each discrete sampling point At different scales The maximum energy value under the given conditions.

8. The method for extracting and detecting magnetic anomaly signals based on an improved EEMD according to claim 7, characterized in that, In step S8, the energy sequence for one-dimensional matching detection is... With the current discrete sampling points to be detected A sliding window is set up around the center. Within this window, the average power estimates of the local background noise in the training units located to the left and right of the discrete sampling point to be detected are calculated respectively. and The training unit is a data sequence interval within a sliding window specifically used to estimate the local environmental background noise, with a single-sided length of [missing information]. ; The larger of the two power estimates is selected as the local noise reference power at the current moment. According to the preset system allowable false alarm probability and training unit length Calculate the corresponding threshold multiplier factor. The threshold multiplier factor is multiplied by the local noise reference power to generate the adaptive decision threshold for the current sampling point. Finally, the adaptive decision threshold is... Matching detection energy with corresponding sampling points Compare point by point: （18） （19） （20） If at a certain discrete sampling point Satisfying If the magnetic anomaly is detected at the actual observation time corresponding to the sampling point, it is determined that the magnetic anomaly target was successfully detected; otherwise, it is determined that the current sampling point is only experiencing fluctuations in environmental background noise.

9. A magnetic anomaly signal extraction and detection system based on an improved EEMD, characterized in that, Includes the following modules: Data acquisition module: used to acquire discrete observation data from actual detection, including the total number of sampling points, sampling frequency, and the three-component sequence actually measured by the fluxgate sensor; Input signal calculation module: Calculates the input signal based on a three-component sequence; Decomposition Module: Utilizes the ensemble empirical mode decomposition algorithm to decompose the input signal, obtaining multiple intrinsic mode functions (EMFs). Quantity; Gated filtering module: Utilizes the Teager energy operator to extract each The transient energy envelope of the component is obtained and smoothed to suppress high-frequency noise; based on the smoothed transient energy envelope, the corresponding... Mutation energy-gated screening of components; Signal reconstruction module: For the components retained by gating, calculate the adaptive noise threshold, and combine Z-Score normalization and Sigmoid function to construct a continuous soft mask for weighted clipping and reconstruction to obtain the reconstructed enhanced signal; Fusion module: Using orthogonal basis functions, the extracted enhanced signal is subjected to multi-scale convolution filtering to obtain a one-dimensional matching detection energy sequence; Detection module: Based on the one-dimensional matching detection energy sequence, combined with the maximum constant false alarm rate algorithm, the adaptive decision threshold is dynamically calculated to complete the final magnetic anomaly target detection.

10. An electronic device, characterized in that, Including processor and memory; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements any of the magnetic anomaly signal extraction and detection methods described in this application.

Images