Magnetic anomaly sensing method based on improved RLMD and support vector machine and related device

Through the improved RLMD and support vector machine methods, the arrangement entropy screening and signal reconstruction are used to solve the problem that artificial threshold setting in the prior art is difficult to suppress noise, and the recognition accuracy and detection ability of magnetic abnormal signals are improved.

CN118897326AActive Publication Date: 2024-11-05BEIJING AUTOMATION CONTROL EQUIP INST

Patent Information

Application Number
CN202410913670.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-05
Estimated Expiration
2044-07-09

AI Technical Summary

Technical Problem

In the prior art, when identifying magnetic abnormal signals, it is difficult to fully suppress noise by artificially setting the screening threshold, resulting in low target recognition accuracy, especially when the detection distance increases, the signal intensity decreases and is easily flooded by background noise.

Method used

The improved robust local mean decomposition (RLMD) algorithm is used to decompose the nonlinear signal into product function (PF), and adaptive screening is performed by calculating the arrangement entropy of PF, and signal reconstruction and feature extraction are combined with support vector machines to construct the morphology. Learn the group of feature vectors to judge the existence of magnetic abnormal signals.

Benefits of technology

It effectively improves the data signal-to-noise ratio, reduces manual intervention, fully suppresses background noise, improves the accuracy of magnetic anomaly detection and perceived ability in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118897326A_ABST
    Figure CN118897326A_ABST
Patent Text Reader

Abstract

The invention provides a magnetic anomaly sensing method based on improved RLMD and a support vector machine and a related device, and the method comprises the steps: obtaining original data s (t), employing an RLMD algorithm to carry out the decomposition of the original data s (t), and obtaining a plurality of PF components; the permutation entropy of the multiple PF components is calculated, all permutation entropy values are arranged from large to small, the last n PF components are screened out, n is determined in a three-quantile self-adaption mode, linear reconstruction is conducted on the PF components to obtain linear reconstruction signals, and trend terms of the linear reconstruction signals are removed to obtain processed magnetostatic signals m (t); constructing a corresponding feature vector group [D, K, P]; establishing a training sample data set; and inputting the feature vector group corresponding to the signal into a trained support vector machine, and judging whether target magnetic anomaly exists or not. According to the technical scheme, the technical problem that in the prior art, when signal recognition is carried out, the screening threshold value is generally set manually, noise is difficult to fully suppress, and consequently the target recognition accuracy is low is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of aeromagnetic anomaly detection signal processing, and in particular to a magnetic anomaly sensing method based on improved RLMD and support vector machine and a related device. Background Art

[0002] Targets with certain ferromagnetism, such as pipelines and mineral deposits, will cause local magnetic field distortion under the magnetization of the geomagnetic field, namely magnetic anomalies. As the exposure source of invisible targets, accurate measurement of target magnetic anomalies is of great significance for the detection and identification of targets. With the continuous improvement of magnetic sensor technology in recent years, magnetic anomaly sensing technology has become a bottleneck restricting detection capabilities. Magnetic anomaly sensing technology faces two major practical problems: first, the intensity of the target magnetic anomaly signal decreases rapidly with the increase of detection distance, resulting in a very limited detection distance; second, the target magnetic anomaly signal is easily affected by various factors such as aviation platforms, geological signals and geomagnetic daily changes, and is usually submerged in background noise. These factors limit the further development and application of magnetic anomaly detection. Therefore, how to perceive the weak magnetic anomaly signal of the target from the strong noise background is the key to magnetic anomaly detection technology.

[0003] As an adaptive signal processing method, Robust Local Mean Decomposition (RLMD) has been widely used in fault diagnosis and other aspects. Existing technologies in the prior art include CN112557966 A, a transformer winding loosening identification method based on local mean decomposition and support vector machine, CN115792523 A, a low-voltage DC arc fault detection method, and CN 110636053A, a network attack detection method based on local mean decomposition and support vector machine. The above methods can realize the detection and identification of signals. However, in the prior art, when performing signal identification, the screening threshold is usually set manually. This method is difficult to fully suppress noise, resulting in low target identification accuracy. Summary of the invention

[0004] The present invention provides a magnetic anomaly sensing method and related devices based on improved RLMD and support vector machine, which can solve the technical problem in the prior art that when performing signal recognition, the screening threshold is usually set manually, and this method is difficult to fully suppress noise, resulting in low target recognition accuracy.

[0005] According to one aspect of the present invention, a magnetic anomaly sensing method based on improved RLMD and support vector machine is provided. The magnetic anomaly sensing method based on improved RLMD and support vector machine comprises: step 1, using an aeromagnetic survey platform to fly along a preset route, obtaining raw data s(t) that may contain a target static magnetic anomaly signal, and using the RLMD algorithm to decompose the raw data s(t) to obtain a plurality of PF components; step 2, respectively calculating the permutation entropy of the obtained multiple PF components, arranging the permutation entropy values ​​of all PF components from large to small, screening out the last n PF components, wherein n is adaptively determined by the tertile of all permutation entropy values, linearly reconstructing the PF components according to the last n PF components to obtain a linear reconstructed signal, removing the trend term from the linear reconstructed signal to obtain a processed signal; The static magnetic signal m(t) is processed; Step three, the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed static magnetic signal m(t) are calculated, and the corresponding feature vector group [D, K, P] is constructed; Step four, a training sample data set is established, including various situations such as the presence or absence of ferromagnetic targets, ferromagnetic targets at different distances and different speeds of the detection platform, and a label S is set for the signal. The label S of the signal with the target magnetic anomaly is set to 1, and the label S of the signal without the target magnetic anomaly is set to 0. The support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy; Step five, the feature vector group corresponding to the processed static magnetic signal m(t) is input into the trained support vector machine, and the output result is used to determine whether there is a target magnetic anomaly in the original signal to achieve magnetic anomaly perception.

[0006] Furthermore, in step 1, the original data s(t) is decomposed by using the RLMD algorithm to obtain multiple PF components, which specifically includes: step 1.1, determining all local extreme points in the original data s(t); step 1.2, calculating and obtaining the first local mean function m according to the multiple local extreme points 11 (t) and the first local envelope estimate function a 11 (t); Step 1.3, the first local mean function m 11 (t) is subtracted from the original data s(t) to obtain the first intermediate function h 11 (t); Step 1.4, use the first intermediate function h 11 (t) divided by the first local envelope estimate function a 11 (t) Get the first demodulated function s after demodulation 11 (t); Step 1.5, the first intermediate function h 11 (t) is used as the original data, and the steps from the extreme point to the subtraction of the local mean are performed, and steps 1.1 to 1.4 are repeated to obtain the second local mean function m 12 (t), the second local envelope estimation function a 12 (t), the second intermediate function h12 (t) and the second demodulated function s 12 (t); Step 1.6, the second intermediate function h 12 (t) is used as the original data, and the steps from the extreme point to the subtraction of the local mean are performed. Steps 1.1 to 1.4 are repeated to obtain the third local mean function m 13 (t), the third local envelope estimation function a 13 (t), the third intermediate function h 13 (t) and the third demodulated function s 13 (t); Step 1.7, and so on, repeat the steps from the extreme point to subtracting the local mean until a pure FM signal s appears 1n (t), based on multiple local envelope estimation functions and pure FM signal s 1n (t) Calculate and obtain the first PF component; Step 1.8, u 1 (t) = s(t) - PF 1 (t) is used as the original signal for the next decomposition, and steps 1.1 to 1.7 are repeated to obtain the second PF component; step 1.9, steps 1.1 to 1.8 are repeated to obtain the third PF component, ..., the Nth PF component in turn, and u is calculated to obtain 2 (t),…,u k (t), until u k (t) Stop until there is only one extreme point left or it becomes a monotonic function.

[0007] Furthermore, the local mean m in any local mean function 1i (t) can be based on Calculate and obtain the local envelope estimate a of any local envelope estimate function 11 (t) can be based on Calculate and obtain, where n k is the kth extreme point, n k+1 is the k+1th extreme point.

[0008] Furthermore, the permutation entropy PE of any PF component can be calculated according to Calculate and obtain, where p j is the probability of occurrence of arrangement j in the PF component, k=1, 2, ..., N, and arrangement j is the jth arrangement of the PF component.

[0009] Furthermore, the processed static magnetoelectric signal m(t) can be calculated and obtained according to m(t)=lm(t)-L(t), where: lm(t) is the linear reconstruction signal and L(t) is the linear trend term.

[0010] Furthermore, 2(t), ..., u k (t) can be based on Calculate and obtain.

[0011] Furthermore, the morphological fractal dimension D in the feature vector group [D, K, P] = log 2 N m (δ m ) / (-log 2 δ m ), where N m (δ m ) represents the number of non-empty boxes corresponding to , and the side length of the box is δ m (m=1,2,...,M); Kurtosis value in the eigenvector group [D,K,P] Where x is the amplitude of the processed static magnetostatic signal m(t), μ is the average value of the processed static magnetostatic signal m(t), and σ is the standard deviation of the processed static magnetostatic signal m(t); the peak-to-peak value P in the eigenvector group [D, K, P] is max(x)-min(x), where max(x) is the maximum value of the processed static magnetostatic signal m(t), and min(x) is the minimum value of the processed static magnetostatic signal m(t).

[0012] According to another aspect of the present invention, a magnetic anomaly sensing system based on VMD-1DCNN is provided. The magnetic anomaly sensing system based on VMD-1DCNN uses the magnetic anomaly sensing method based on improved RLMD and support vector machine as described above to sense magnetic anomalies. The magnetic anomaly sensing system includes: a PF component calculation module, the PF component calculation module is used to use the aeromagnetic survey platform to fly along a preset route, obtain raw data s(t) that may contain target static magnetic anomaly signals, and use the RLMD algorithm to decompose the raw data s(t) to obtain multiple PF components; a static magnetic signal acquisition module, the static magnetic signal acquisition module is used to calculate the permutation entropy of the obtained multiple PF components, arrange the permutation entropy values ​​of all PF components from large to small, screen out the last n PF components, wherein n is adaptively determined by the tertile of all permutation entropy values, linearly reconstruct the PF components according to the last n PF components to obtain a linear reconstructed signal, and remove the trend from the linear reconstructed signal The potential term obtains the processed static magnetic signal m(t); a feature vector group construction module, the feature vector group construction module is used to calculate the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed static magnetic signal m(t), and construct the corresponding feature vector group [D, K, P]; a training sample data set construction module, the training sample data set construction module is used to establish a training sample data set, including various situations such as whether there is a ferromagnetic target, ferromagnetic targets at different distances and different speeds of the detection platform, set a label S for the signal, the signal label S with the target magnetic anomaly is set to 1, and the label S without the target magnetic anomaly signal is set to 0, and the support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy; a magnetic anomaly perception judgment module, the magnetic anomaly perception judgment module is used to input the feature vector group corresponding to the processed static magnetic signal m(t) into the trained support vector machine, and judge whether there is a target magnetic anomaly in the original signal according to the output result to realize magnetic anomaly perception.

[0013] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the magnetic anomaly perception method based on improved RLMD and support vector machine as described above.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the magnetic anomaly perception method based on improved RLMD and support vector machine are implemented as described above.

[0015] The technical solution of the present invention is applied to provide a magnetic anomaly perception method based on improved RLMD and support vector machine. In the method, RLMD can decompose nonlinear and non-stationary signals into a series of product functions (PF), each PF containing physical information of a specific frequency. Reasonable reconstruction of the PF after RLMD processing can effectively improve the data signal-to-noise ratio, which is conducive to subsequent feature extraction. Magnetic anomaly perception can be regarded as a binary classification problem, and support vector machine is a binary classification method with good performance and small sample demand. The idea of ​​support vector machine is to find an optimal classification hyperplane that can accurately divide data based on the principle of structural minimization. The signal after PF reconstruction is feature extracted and input into the support vector machine, and the binary judgment of whether the original signal has a target is output, so that the target magnetic anomaly signal can be perceived. In addition, the present invention uses the permutation entropy indicator when screening the PF components. The core idea is that the target magnetic anomaly signal has different physical properties from the background noise, and the permutation entropy reflects this feature. By calculating the permutation entropy of all PF components and sorting them from large to small, a group of PF components with the smallest permutation entropy is determined according to the tertile for reconstruction. The method has obvious adaptability, fully considers the difference between the target magnetic anomaly signal and the background noise at the physical level, and does not need to manually set the screening threshold, so as to achieve the purpose of fully suppressing the background noise. Therefore, compared with the prior art, the magnetic anomaly perception method based on improved RLMD and support vector machine provided by the present invention reduces the manual intervention in feature extraction, can improve the magnetic anomaly perception ability in complex environments, fully considers the difference between the target magnetic anomaly signal and the background noise at the physical level, can effectively improve the accuracy of magnetic anomaly detection in the prior art, and effectively solves the technical problem that the prior art usually sets the screening threshold manually when performing signal recognition, which is difficult to fully suppress noise and leads to low target recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flow chart of a magnetic anomaly perception method based on improved RLMD and support vector machine provided according to a specific embodiment of the present invention is shown. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0020] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps described in these embodiments do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0021] like Figure 1As shown, according to a specific embodiment of the present invention, a magnetic anomaly sensing method based on improved RLMD and support vector machine is provided, and the magnetic anomaly sensing method based on improved RLMD and support vector machine includes: step 1, using an aeromagnetic survey platform to fly along a preset route, obtaining raw data s(t) that may contain a target static magnetic anomaly signal, and using the RLMD algorithm to decompose the raw data s(t) to obtain multiple PF components; step 2, respectively calculating the permutation entropy of the obtained multiple PF components, arranging the permutation entropy values ​​of all PF components from large to small, screening out the last n PF components, wherein n is adaptively determined by the tertile of all permutation entropy values, linearly reconstructing the PF components according to the last n PF components to obtain a linear reconstructed signal, and removing the trend term from the linear reconstructed signal to obtain a processed Step 3: calculate the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed static magnetic signal m(t), and construct the corresponding feature vector group [D, K, P]; Step 4: establish a training sample data set, including various situations such as the presence or absence of ferromagnetic targets, ferromagnetic targets at different distances and different speeds of the detection platform, set a label S for the signal, the signal label S with the target magnetic anomaly is set to 1, and the label S without the target magnetic anomaly signal is set to 0, and the support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy; Step 5: input the feature vector group corresponding to the processed static magnetic signal m(t) into the trained support vector machine, and judge whether there is a target magnetic anomaly in the original signal according to the output result to realize magnetic anomaly perception.

[0022] By applying this configuration, a magnetic anomaly perception method based on improved RLMD and support vector machine is provided. In this method, RLMD can decompose nonlinear and non-stationary signals into a series of product functions (PF), each PF containing physical information of a specific frequency. Reasonable reconstruction of the PF after RLMD processing can effectively improve the data signal-to-noise ratio, which is conducive to subsequent feature extraction. Magnetic anomaly perception can be regarded as a binary classification problem, and support vector machine is a binary classification method with good performance and small sample demand. The idea of ​​support vector machine is to find an optimal classification hyperplane that can accurately divide data based on the principle of structural minimization. The signal reconstructed by PF is feature extracted and input into the support vector machine, and the binary judgment of whether the original signal has a target is output, so that the target magnetic anomaly signal can be perceived. In addition, the present invention uses the permutation entropy indicator when screening the PF components. The core idea is that the target magnetic anomaly signal has different physical properties from the background noise, and the permutation entropy reflects this feature. By calculating the permutation entropy of all PF components and sorting them from large to small, a group of PF components with the smallest permutation entropy is determined according to the tertile for reconstruction. The method has obvious adaptability, fully considers the difference between the target magnetic anomaly signal and the background noise at the physical level, and does not need to manually set the screening threshold, so as to achieve the purpose of fully suppressing the background noise. Therefore, compared with the prior art, the magnetic anomaly perception method based on improved RLMD and support vector machine provided by the present invention reduces the manual intervention in feature extraction, can improve the magnetic anomaly perception ability in complex environments, fully considers the difference between the target magnetic anomaly signal and the background noise at the physical level, can effectively improve the accuracy of magnetic anomaly detection in the prior art, and effectively solves the technical problem that the prior art usually sets the screening threshold manually when performing signal recognition, which is difficult to fully suppress noise and leads to low target recognition accuracy.

[0023] In order to realize magnetic anomaly perception based on improved RLMD and support vector machine, it is first necessary to use the aeromagnetic platform to fly along the preset route to obtain the original data s(t) that may contain the target static magnetic anomaly signal, and then use the RLMD algorithm to decompose the original data s(t) to obtain multiple PF components.

[0024] As a specific embodiment of the present invention, the original data s(t) in step 1 is a segment of level flight data containing about 3000 sampling points. The original data s(t) is decomposed by the RLMD algorithm to obtain multiple PF components, including:

[0025] Step 1.1, determine all local extreme points n in the original data s(t) k (k=1,2,...), including the maximum and minimum values;

[0026] Step 1.2, calculate and obtain the first local mean function m based on multiple local extreme points 11 (t) and the first local envelope estimate function a 11 (t). The local mean m in any local mean function 1i (t) can be based on After obtaining multiple local means, multiple local means are fitted together to form a first local mean function; the local envelope estimation value a in any envelope estimation function 1i (t) can be based on Calculate and obtain, where n k is the kth extreme point, n k+1 is the k+1th extreme point.

[0027] Step 1.3: The first local mean function m 11 (t) is subtracted from the original data s(t) to obtain the first intermediate function h 11 (t);

[0028] Step 1.4, use the first intermediate function h 11 (t) divided by the first local envelope estimate function a 11 (t) Get the first demodulated function s after demodulation 11 (t);

[0029] Step 1.5: transform the first intermediate function h 11 (t) is used as the original data, and the steps from the extreme point to the subtraction of the local mean are performed, and steps 1.1 to 1.4 are repeated to obtain the second local mean function m 12 (t), the second local envelope estimation function a 12 (t), the second intermediate function h 12 (t) and the second demodulated function s 12 (t). Specifically, the first intermediate function h 11 (t) as the original data, determine all the local extreme value points in the original data, calculate and obtain the second local mean function and the second local envelope estimation function according to the multiple local extreme value points, and convert the second local mean function from the original data (i.e., the first intermediate function h 11 (t)) to obtain the second intermediate function h 12 (t), using the second intermediate function h 12 (t) divided by the second local envelope estimate function a 12 (t) Get the second demodulated function s after demodulation 12 (t);

[0030] Step 1.6: Set the second intermediate function h 12(t) is used as the original data, and the steps from the extreme point to the subtraction of the local mean are performed. Steps 1.1 to 1.4 are repeated to obtain the third local mean function m 13 (t), the third local envelope estimation function a 13 (t), the third intermediate function h 13 (t) and the third demodulated function s 13 (t). Specifically, the second intermediate function h 12 (t) as the original data, determine all the local extreme value points in the original data, calculate and obtain the third local mean function and the third local envelope estimation function according to the multiple local extreme value points, and convert the third local mean function from the original data (i.e., the second intermediate function h 12 (t)) to obtain the third intermediate function h 13 (t), using the third intermediate function h 13 (t) divided by the third local envelope estimate function a 13 (t) Get the third demodulated function s after demodulation 13 (t);

[0031] Step 1.7, and so on, repeat the steps from the extreme point to subtracting the local mean until a pure FM signal s appears. 1n (t), based on multiple local envelope estimation functions and pure FM signal s 1n (t) Calculate and obtain the first PF component. In the present invention, s 1n (t) = h 1n (t) / a 1n (t), n=1,2,3,…, envelope function of the first PF component Multiply the pure FM signal with the envelope function to get the first PF component PF 1 (t) = s 1n (t)a 1 (t);

[0032] Step 1.8, change u 1 (t) = s(t) - PF 1 (t) As the original signal for the next decomposition, repeat steps 1.1 to 1.7 to obtain the second PF component;

[0033] Step 1.9, and so on, repeat steps 1.1 to 1.8, and obtain the third PF component, ..., Nth PF component in turn, and calculate u 2 (t),…,u k (t), until u k (t) until there is only one extreme point left or it is a monotonic function. 2 (t),…,u k (t) can be based on Calculate and obtain.

[0034] Specifically, u 2 (t) = u 1 (t)-PF 2 (t) is used as the original signal for the next decomposition, and steps 1.1 to 1.7 are repeated to obtain the third PF component; u 3 (t) = u 2 (t)-PF 3 (t) is used as the original signal for the next decomposition, and steps 1.1 to 1.7 are repeated to obtain the fourth PF component; u 4 (t) = u 3 (t)-PF 4 (t) is used as the original signal for the next decomposition, and steps 1.1 to 1.7 are repeated to obtain the fourth PF component; and so on until u k (t) until there is only one extreme point left or it is a monotonic function.

[0035] Furthermore, after obtaining multiple PF components, the permutation entropy can be calculated for each of the multiple PF components, the permutation entropy values ​​of all PF components are arranged from large to small, and the last n PF components are screened out, where n is adaptively determined by the ternary digit of all permutation entropy values. The PF components are linearly reconstructed according to the last n PF components to obtain a linear reconstructed signal, and the trend term of the linear reconstructed signal is removed to obtain the processed static magnetic signal m(t).

[0036] The permutation entropy PE of any PF component can be calculated according to Calculate and obtain, where p j is the probability of occurrence of arrangement j in the PF component, k=1, 2, ..., N, arrangement j is the jth arrangement of the PF component, which is a known parameter.

[0037] Since the larger the PE value, the higher the proportion of noise in the modal component and the more random it is, in order to fully suppress the background noise, the PE values ​​of all modal components are arranged from large to small, and the last n components are screened out, where n is adaptively determined by the tertile of all PE values ​​to obtain the linear reconstructed signal The linear trend term L(t) of the linear reconstructed signal lm(t) is obtained by fitting.

[0038] The purpose of detrending is to remove the linear trend term of the reconstructed data. The core is to fit the least squares line segment of the original data and then subtract it from the original data to make the mean of the processed data close to 0. The calculation process can be described as: estimate the linear function parameters k, b, so that they satisfy Then k, b are calculated as: Among them, xi ∈(t 1 ,t 2 ,…,t n ), y i ∈(lm(t 1 ),lm(t 2 ),…lm(t n )), pinv is a subfunction for finding the generalized inverse of a matrix, which is implemented by matrix singular value decomposition. For example, if the inverse of matrix X is required, first perform singular value decomposition on it: [U, S, V] = SVD(X), where X = U*S*V T , and U*U T =I, V*U T =I, then the generalized inverse of matrix X is X -1 For: X -1 =pinv(X)=V*S -1 *U T Therefore, L(t)=k*lm(t)+b.

[0039] The processed static magnetic signal m(t) is obtained by subtracting the linear trend term L(t) from the linear reconstructed signal lm(t), that is, m(t)=lm(t)-L(t), where lm(t) is the linear reconstructed signal and L(t) is the linear trend term.

[0040] Furthermore, after obtaining the processed magnetostatic signal m(t), a feature vector group is constructed for the processed magnetostatic signal m(t). Specifically, the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed magnetostatic signal m(t) are calculated, and the corresponding feature vector group [D, K, P] is constructed.

[0041] In particular, the method for constructing a set of feature vectors for the processed magnetostatic signal m(t) is:

[0042] Calculate the morphological fractal dimension D of m(t), the morphological fractal dimension D in the feature vector group [D, K, P] = log 2 N m (δ m ) / (-log 2 δ m ), where N m (δ m ) represents the number of non-empty boxes corresponding to , and the side length of the box is δ m (m=1,2,…,M);

[0043] Calculate the kurtosis value K of m(t), the kurtosis value in the eigenvector group [D, K, P] Where x is the amplitude of the processed static magnetostatic signal m(t), μ is the average value of the processed static magnetostatic signal m(t), and σ is the standard deviation of the processed static magnetostatic signal m(t);

[0044] Calculate the peak-to-peak value P of m(t), the peak-to-peak value P in the eigenvector group [D, K, P] = max(x)-min(x), where max(x) is the maximum value of the processed static magnetostatic signal m(t), and min(x) is the minimum value of the processed static magnetostatic signal m(t).

[0045] Thus, the set of eigenvectors corresponding to m(t) is [D, K, P].

[0046] Furthermore, after constructing the corresponding feature vector group [D, K, P], a training sample data set can be established, including various situations with or without ferromagnetic targets, ferromagnetic targets at different distances, and different speeds of the detection platform. A label S is set for the signal, and the label S of the signal with target magnetic anomaly is set to 1, and the label S of the signal without target magnetic anomaly is set to 0. The support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy. In particular, the sample data set is generated using the Monte Carlo method, which basically covers various situations that may occur in the target to be detected.

[0047] After the training sample data set is established, the feature vector group corresponding to the processed static magnetic signal m(t) can be input into the trained support vector machine, and the output result can be used to determine whether the target magnetic anomaly exists in the original signal to achieve magnetic anomaly perception.

[0048] According to another aspect of the present invention, a magnetic anomaly sensing system based on VMD-1DCNN is provided. The magnetic anomaly sensing system based on VMD-1DCNN uses the magnetic anomaly sensing method based on improved RLMD and support vector machine as described above to sense magnetic anomalies. The magnetic anomaly sensing system includes a PF component calculation module, a static magnetic signal acquisition module, a feature vector group construction module, a training sample data set construction module and a magnetic anomaly sensing judgment module. The PF component calculation module is used to use an aeromagnetic survey platform to fly along a preset route to obtain raw data s(t) that may contain a target static magnetic anomaly signal, and use the RLMD algorithm to decompose the raw data s(t) to obtain multiple PF components. The static magnetic signal acquisition module is used to calculate the permutation entropy of the obtained multiple PF components, arrange the permutation entropy values ​​of all PF components from large to small, and screen out the last n PF components, where n is adaptively determined by the tertile of all permutation entropy values, and the last n PF components are selected according to the permutation entropy values. The PF component is linearly reconstructed to obtain a linear reconstructed signal, and the trend term is removed from the linear reconstructed signal to obtain the processed static magnetic signal m(t). The feature vector group construction module is used to calculate the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed static magnetic signal m(t), and construct the corresponding feature vector group [D, K, P]. The training sample data set construction module is used to establish a training sample data set, including various situations such as the presence or absence of ferromagnetic targets, ferromagnetic targets at different distances, and different speeds of the detection platform. A label S is set for the signal, and the signal label S with the target magnetic anomaly is set to 1, and the label S without the target magnetic anomaly is set to 0. The support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy. The magnetic anomaly perception judgment module is used to input the feature vector group corresponding to the processed static magnetic signal m(t) into the trained support vector machine, and judge whether there is a target magnetic anomaly in the original signal according to the output result to realize magnetic anomaly perception.

[0049] By applying this configuration, a magnetic anomaly perception system based on improved RLMD and support vector machine is provided. In this system, RLMD can decompose nonlinear and non-stationary signals into a series of product functions (PF), each PF containing physical information of a specific frequency. Reasonable reconstruction of the PF after RLMD processing can effectively improve the data signal-to-noise ratio, which is conducive to subsequent feature extraction. Magnetic anomaly perception can be regarded as a binary classification problem, and support vector machine is a binary classification method with good performance and small sample demand. The idea of ​​support vector machine is to find an optimal classification hyperplane that can accurately divide data based on the principle of structural minimization. The signal reconstructed by PF is feature extracted and input into the support vector machine, and the binary judgment of whether the original signal has a target is output, so that the target magnetic anomaly signal can be perceived. In addition, the present invention uses the permutation entropy indicator when screening the PF components. The core idea is that the target magnetic anomaly signal has different physical properties from the background noise, and the permutation entropy reflects this feature. By calculating the permutation entropy of all PF components and sorting them from large to small, a group of PF components with the smallest permutation entropy is determined according to the tertile for reconstruction. The system has obvious adaptability and fully considers the physical differences between the target magnetic anomaly signal and the background noise. It does not require artificial setting of the screening threshold and achieves the purpose of fully suppressing the background noise.

[0050] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the magnetic anomaly perception method based on improved RLMD and support vector machine as described above.

[0051] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program, characterized in that when the computer program is executed by a processor, the steps of the magnetic anomaly perception method based on improved RLMD and support vector machine are implemented as described above.

[0052] In order to further understand the present invention, the following Figure 1 The magnetic anomaly perception method based on improved RLMD and support vector machine provided by the present invention is described in detail.

[0053] like Figure 1 As shown in the figure, focusing on the civilian and military needs such as mineral resource exploration, shipwreck archaeology and underwater target detection, the present invention proposes a magnetic anomaly perception method based on improved RLMD and support vector machine, which can effectively improve the accuracy of magnetic anomaly detection in the prior art to achieve effective detection of weak magnetic anomaly signals. The complete process is as follows Figure 1 As shown, the following steps are included:

[0054] Step S01, performing a flight test, collecting raw data that may contain target magnetic anomalies, and intercepting the level flight segment for subsequent processing;

[0055] Step S02, using RLMD to decompose the original data s(t) to obtain a number of PF components. In particular, for any original data s(t), the method of decomposing it by RLMD is:

[0056] Use spline interpolation to process all the maximum and minimum values ​​of s(t) respectively, construct the upper and lower envelope functions and calculate the local mean function m 11 (t) and the local envelope function a 11 (t);

[0057] M 11 (t) is separated from s(t) to obtain h 11 (t), using h 11 (t) divided by the envelope function a 11 (t) Get the demodulated s 11 (t), repeat the above steps until a pure FM signal s appears 1n (t), where s 1n (t) = h 1n (t) / a 1n (t),n=1,2,3,…。

[0058] Envelope signal of the first PF component

[0059] Multiply the pure FM signal with the envelope function to get the first PF component PF 1 (t) = s 1n (t)a 1 (t);

[0060] will u 1 (t) = s(t) - PF 1 (t) is used as the initial signal for the next cycle until u k (t) Stop when there is only one extreme point left or it is a monotonic function.

[0061]

[0062] Through the above cycle, the original data s(t) is decomposed into k PF components and a residual function, which can be expressed as: Performing RLMD decomposition on the original data can not only suppress the modal aliasing of other adaptive decomposition algorithms, but also has higher computational efficiency and obtains physical information with more obvious local characteristics.

[0063] Step S03, calculate the permutation entropy of the PF component, and perform adaptive screening of the PF component according to the tertile of the permutation entropy, reconstruct the screened component and remove the linear trend to obtain the reconstructed static magnetostatic data signal m(t). In particular, the method for screening the PF component is:

[0064] Calculate the permutation entropy PE of each PF component:

[0065]

[0066] where p j Represents the probability of occurrence of permutation j in the PF component.

[0067] Arrange the PE values ​​of all PF components from large to small, and select the last n PF components, where n is adaptively determined by the tertile of the PE value, to obtain the linear reconstructed signal

[0068] The linear trend term of lm(t) is fitted and subtracted to obtain the processed static magnetostatic data signal m(t).

[0069] When screening the PF components, the permutation entropy indicator is used. Its core idea is that the target magnetic anomaly signal and the background noise have different physical properties, and the permutation entropy reflects this characteristic. By calculating the permutation entropy of all PF components and sorting them from large to small, a group of PF components with the smallest permutation entropy is determined according to the tertile for signal reconstruction. This method has obvious adaptability and fully considers the physical differences between the target magnetic anomaly signal and the background noise. There is no need to manually set the screening threshold, and the purpose of suppressing background noise is achieved.

[0070] Step S04, construct a feature vector group [D, K, P] for the processed data m(t). In particular, each element in the feature vector group is obtained in the following way:

[0071] The morphological fractal dimension of m(t) is D=log 2 N m (δ m ) / (-log 2 δ m ), where N m (δ m ) represents the number of non-empty boxes corresponding to , and the side length of the box is δ m (m=1,2,…,M);

[0072] Kurtosis value of m(t) Where x is the amplitude of the processed static magnetostatic signal m(t), μ is the average value of the processed static magnetostatic signal m(t), and σ is the standard deviation of the processed static magnetostatic signal m(t);

[0073] The peak-to-peak value P of m(t)=max(x)-min(x), where max(x) is the maximum value of the processed magnetostatic signal m(t), and min(x) is the minimum value of the processed magnetostatic signal m(t).

[0074] Step S05, establish a training sample data set, including various situations such as whether there is a ferromagnetic target, ferromagnetic targets at different distances, different platform speeds, etc., set a label S for the signal, and set the signal label S of the signal with target magnetic anomaly to 1, and set the signal label S of the signal without target magnetic anomaly to 0. The sample data set is generated by the Monte Carlo method, which basically covers various situations that may occur in the target to be detected. The support vector machine is trained based on the training sample data set to obtain the optimal magnetic anomaly perception strategy.

[0075] Step S06, input the feature vector group [D, K, P] corresponding to m(t) into the support vector machine, output the judgment of whether m(t) contains the target magnetic anomaly signal, and realize the perception and classification of magnetic anomalies. Support vector machines have obvious advantages in solving small sample, nonlinear and high-dimensional pattern recognition problems. The mathematical model is simple and clear, and it is widely used in scenarios such as pattern recognition. The algorithm of support vector machine can be reduced to solving a constrained QP problem. Its main idea is to map the input vector to a high-dimensional feature space through a nonlinear mapping of some implementation selection, and construct the optimal hyperplane in this space. In this example, the input vector is the feature vector group [D, K, P] corresponding to m(t), and the output includes two cases, namely, m(t) contains the target magnetic anomaly signal and m(t) does not contain the target magnetic anomaly signal, so that the target magnetic anomaly perception is completed.

[0076] In summary, the present invention provides a magnetic anomaly perception method based on improved RLMD and support vector machine. The method reduces manual intervention in feature extraction, can improve the magnetic anomaly perception ability in complex environments, fully considers the physical differences between target magnetic anomaly signals and background noise, can effectively improve the accuracy of magnetic anomaly detection in the prior art, and effectively solves the technical problem that in the prior art, when performing signal recognition, the screening threshold is usually set manually, and this method is difficult to fully suppress noise, resulting in low target recognition accuracy.

[0077] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0078] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A magnetic anomaly perception method based on improved RLMD and support vector machine, characterized in that: The magnetic anomaly perception method based on improved RLMD and support vector machine includes: Step 1: Use the aeromagnetic survey platform to fly along the preset route to obtain the original data s(t) that may contain the target static magnetic anomaly signal, and use the RLMD algorithm to decompose the original data s(t) to obtain multiple PF components; Step 2, respectively calculating the permutation entropy of the obtained multiple PF components, arranging the permutation entropy values ​​of all PF components from large to small, screening out the last n PF components, where n is adaptively determined by the tertile of all permutation entropy values, linearly reconstructing the PF components according to the last n PF components to obtain a linear reconstructed signal, and removing the trend term from the linear reconstructed signal to obtain the processed static magnetic signal m(t); Step 3, calculating the morphological fractal dimension D, kurtosis value K and peak-to-peak value P of the processed static magnetic signal m(t), and constructing the corresponding feature vector group [D, K, P]; Step 4: Establish a training sample data set, including various situations such as the presence or absence of ferromagnetic targets, ferromagnetic targets at different distances, and detection platform speeds. Set a label S for the signal. The label S of the signal with target magnetic anomaly is set to 1, and the label S of the signal without target magnetic anomaly is set to 0. Train the support vector machine based on the training sample data set to obtain the optimal magnetic anomaly perception strategy. Step 5: Input the feature vector group corresponding to the processed static magnetic signal m(t) into the trained support vector machine, and judge whether there is a target magnetic anomaly in the original signal based on the output result to realize magnetic anomaly perception.

2. The magnetic anomaly perception method based on improved RLMD and support vector machine according to claim 1 is characterized in that: In the step 1, the original data s(t) is decomposed by using the RLMD algorithm to obtain multiple PF components, including: Step 1.1, determine all local extreme points in the original data s(t); Step 1.2, calculate and obtain the first local mean function m based on multiple local extreme points 11 (t) and the first local envelope estimate function a 11 (t); Step 1.3: transform the first local mean function m 11 (t) is subtracted from the original data s(t) to obtain the first intermediate function h 11 (t); Step 1.4, use the first intermediate function h 11 (t) divided by the first local envelope estimate function a 11 (t) Get the first demodulated function s after demodulation 11 (t); Step 1.5: transform the first intermediate function h 11 (t) as the original data, perform the steps from the extreme point to subtracting the local mean, repeat the steps 1.1 to 1.4, and obtain the second local mean function m 12 (t), the second local envelope estimation function a 12 (t), the second intermediate function h 12 (t) and the second demodulated function s 12 (t); Step 1.6: transform the second intermediate function h 12 (t) as the original data, perform the steps from the extreme point to subtracting the local mean, repeat the steps 1.1 to 1.4, and obtain the third local mean function m 13 (t), the third local envelope estimation function a 13 (t), the third intermediate function h 13 (t) and the third demodulated function s 13 (t); Step 1.7, and so on, repeat the steps from the extreme point to subtracting the local mean until a pure FM signal s appears. 1n (t), based on multiple local envelope estimation functions and pure FM signal s 1n (t) Calculate and obtain the first PF component; Step 1.8, taking u1(t)=s(t)-PF1(t) as the original signal for the next decomposition, repeating steps 1.1 to 1.7 to obtain the second PF component; Step 1.9, repeating steps 1.1 to 1.8, obtaining the third PF component, ..., Nth PF component in turn, and calculating u2(t), ..., u k (t), until u k (t) Stop until there is only one extreme point left or it becomes a monotonic function.

3. The magnetic anomaly perception method based on improved RLMD and support vector machine according to claim 1 is characterized in that: The local mean m in any of the local mean functions 1i (t) can be based on Calculate and obtain the local envelope estimation value a of any of the local envelope estimation functions 11 (t) can be based on Calculate and obtain, where n k is the kth extreme point, n k+1 is the k+1th extreme point.

4. The magnetic anomaly perception method based on improved RLMD and support vector machine according to claim 3 is characterized in that: The permutation entropy PE of any of the PF components can be calculated according to Calculate and obtain, where p j is the probability of occurrence of arrangement j in the PF component, k = 1, 2, ..., N, and arrangement j is the jth arrangement of the PF component.

5. The magnetic anomaly perception method based on improved RLMD and support vector machine according to claim 4 is characterized in that: The processed static magnetoelectric signal m(t) can be calculated according to m(t)=lm(t)-L(t), where: lm(t) is the linear reconstruction signal and L(t) is the linear trend term.

6. The magnetic anomaly sensing method based on improved RLMD and support vector machine according to claim 1, characterized in that: u2(t),…,u k (t) can be based on Calculate acquisition.

7. The magnetic anomaly sensing method based on improved RLMD and support vector machine according to claim 1, characterized in that: The morphological fractal dimension D in the feature vector set [D, K, P] is log2N m (δ m ) / (-log2δ m ), where N m (δ m ) represents the number of non-empty boxes corresponding to , and the side length of the box is δ m (m=1,2,…,M); Kurtosis value in the eigenvector group [D,K,P] Wherein x is the amplitude of the processed static magnetostatic signal m(t), μ is the average value of the processed static magnetostatic signal m(t), and σ is the standard deviation of the processed static magnetostatic signal m(t); the peak-to-peak value P in the eigenvector group [D, K, P] is max(x)-min(x), wherein max(x) is the maximum value of the processed static magnetostatic signal m(t), and min(x) is the minimum value of the processed static magnetostatic signal m(t).

8. A magnetic anomaly perception system based on VMD-1DCNN, characterized in that: The magnetic anomaly sensing system based on VMD-1DCNN uses the magnetic anomaly sensing method based on improved RLMD and support vector machine as described in any one of claims 1 to 7 to sense magnetic anomalies, and the magnetic anomaly sensing system includes: A PF component calculation module, wherein the PF component calculation module is used to use an aeromagnetic survey platform to fly along a preset route, obtain raw data s(t) that may contain a target static magnetic anomaly signal, and use an RLMD algorithm to decompose the raw data s(t) to obtain multiple PF components; A static magnetostatic signal acquisition module, wherein the static magnetostatic signal acquisition module is used to calculate the permutation entropy of the obtained multiple PF components respectively, arrange the permutation entropy values ​​of all PF components from large to small, screen out the last n PF components, wherein n is adaptively determined by the tertile of all permutation entropy values, linearly reconstruct the PF components according to the last n PF components to obtain a linear reconstructed signal, and remove the trend term from the linear reconstructed signal to obtain a processed static magnetostatic signal m(t); A feature vector group construction module, wherein the feature vector group construction module is used to calculate the morphological fractal dimension D, the kurtosis value K and the peak-to-peak value P of the processed static magnetic signal m(t), and construct a corresponding feature vector group [D, K, P]; A training sample data set construction module, wherein the training sample data set construction module is used to establish a training sample data set, including various situations of the presence or absence of a ferromagnetic target, the ferromagnetic target at different distances, and the detection platform at different speeds, and to set a label S for the signal, wherein the label S of the signal with a target magnetic anomaly is set to 1, and the label S of the signal without a target magnetic anomaly is set to 0, and a support vector machine is trained based on the training sample data set to obtain an optimal magnetic anomaly perception strategy; The magnetic anomaly perception and judgment module is used to input the feature vector group corresponding to the processed static magnetic signal m(t) into the trained support vector machine, and judge whether there is a target magnetic anomaly in the original signal according to the output result to realize magnetic anomaly perception.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the computer program to implement the steps of the magnetic anomaly perception method based on improved RLMD and support vector machine according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the magnetic anomaly sensing method based on improved RLMD and support vector machine as described in any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • Self-adaptive defect detection method, device, recognition system and storage medium

    CN113688807A

  • Magnetic anomaly detection method and system, medium, equipment and terminal

    CN115032704A

  • Power load prediction method based on dynamic decomposition-reconstruction integrated processing

    CN116822742A

  • Transient electromagnetic data processing method based on RLMD-ApEn

    CN117724177A

  • Earthquake magnetic disturbance analysis method based on satellite observation

    CN118151252A

Cited By

  • Aircraft magnetic anomaly classification method and system based on SVM

    CN117390525A

  • A method and system for classifying aircraft magnetic anomalies based on SVM

    CN117390525B