A multi-feature extraction and fusion method for magnetic anomaly detection

Abstract

The application discloses a multi-feature extraction and fusion method for magnetic anomaly detection, which is a method for pre-processing data of a magnetic anomaly signal, selecting time-frequency features, statistical features and target magnetic moment features of the magnetic anomaly signal as multi-features of the magnetic anomaly signal, performing multi-feature fusion in a feature layer by using a kernel principal component analysis method, performing data processing on the magnetic anomaly signal, and providing effective input information for a neural network framework. The method can detect the magnetic anomaly signal under a low signal-to-noise ratio, improves detection efficiency, reduces a false rate, and improves performance of the neural network.

Description

Technical Field

[0001] This invention belongs to the field of magnetic anomaly detection technology, specifically relating to a method for multi-feature extraction and fusion in magnetic anomaly detection. Background Technology

[0002] Magnetic anomaly detection is a technique for discovering ferromagnetic objects hidden within a strong and complex magnetic background. In many practical situations, the target is very far from the detection sensor, making detection difficult under low signal-to-noise ratio conditions. Most current methods determine the presence of the target through signal analysis-based approaches, such as orthogonal basis function detection and minimum entropy detection. However, while these methods are resource-efficient, their detection performance is not satisfactory. In recent years, due to advancements in computing power, more sophisticated methods have become applicable to magnetic anomaly detection. Neural networks have found wider application in magnetic anomaly detection. For neural network frameworks, data preprocessing of the magnetic anomaly signal plays a crucial role; multi-feature extraction and feature fusion are key factors in improving the detection performance of neural networks. Summary of the Invention

[0003] The purpose of this invention is to provide a method for multi-feature extraction and fusion in magnetic anomaly detection.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for multi-feature extraction and fusion in magnetic anomaly detection, comprising the following steps:

[0006] Step 1: Determine the time-frequency characteristics of the magnetic anomaly signal;

[0007] Step 2: Determine the statistical characteristics of the magnetic anomaly signal;

[0008] Step 3: Determine the target magnetic moment signal of the magnetic anomaly signal;

[0009] Step 4: At the feature layer, use kernel component analysis to fuse eigenvectors and reduce nonlinear redundancy;

[0010] Step 5: The processed feature information is used as input to the neural network architecture.

[0011] Furthermore, step 1 utilizes wavelet transform to obtain the time-frequency characteristics of the magnetic anomaly signal; the wavelet transform is:

[0012]

[0013] in, This is a magnetic anomaly signal. It is a wavelet function.

[0014] Furthermore, the statistical characteristics of step 2, such as the mean, variance, extreme values, and peak values, are calculated using the following formula:

[0015]

[0016]

[0017] β3=max{S}

[0018] β4=min{S}

[0019]

[0020] Where S represents the measured signal and N represents the number of samples of the signal.

[0021] Furthermore, in step 3, based on prior knowledge of the target magnetic moment, the magnetic anomaly signal T is... r Decomposed into a set of orthogonal basis functions, the three coefficients are as follows:

[0022]

[0023] In the formula f i (τ)(i=1,2,3) are the orthogonal basis functions of the decomposition.

[0024]

[0025]

[0026]

[0027] The coefficients α of the basis functions i (i = 1, 2, 3) can be obtained using the following formula:

[0028]

[0029] Where μ0 = 4π × 10 -7 H / m.

[0030] Furthermore, after multi-feature extraction in step 4, redundant information in the magnetic anomaly signal is reduced, computational power is decreased, and effective information is retained. However, redundancy issues can arise in the nonlinear relationships between features in the feature layer, affecting the results of magnetic anomaly detection. Therefore, in the feature layer, kernel component analysis is used to fuse feature vectors to reduce nonlinear redundancy, as shown below:

[0031] The magnetic feature sample set is X m×n The nonlinear kernel function is φ(x), and its covariance matrix can be written as:

[0032]

[0033] The eigenset λ corresponding to the covariance matrix C, and the eigenvector set α, are:

[0034] Cα=λα

[0035] For the above formula, use φ(x) k Taking the inner product of (k = 1, 2, ..., n), we get:

[0036]

[0037] The computation of the inner product is quite complex. Here, we can introduce a kernel function to obtain the kernel matrix K(x). i ,x j )=<φ(x i ),φ(x j By simplifying the inner product using the method of )>, the above expression can be simplified to

[0038] nλα=Kα

[0039] The feature vector set can be represented as α=(α1,α2,···,α n ) T ,make For the non-zero feature set of K, the above formula can be simplified to:

[0040]

[0041] Will sorted in descending order The corresponding eigenvectors are α1>α2>α3>···>α n Given a contribution rate, the required number of feature values ​​is k. The first k feature values ​​are selected to form the fusion B. m×k The feature sample set Y after kernel principal component analysis k×n for:

[0042]

[0043] The beneficial effects of this invention are as follows:

[0044] The time-frequency characteristics, statistical characteristics, and target magnetic moment characteristics of the magnetic anomaly signal are selected as multiple features of the signal. Kernel principal component analysis is used to perform multi-feature fusion at the feature layer to process the magnetic anomaly signal data, providing effective input information for the neural network framework. Furthermore, it can detect magnetic anomaly signals under low signal-to-noise ratio conditions, improving detection efficiency, reducing false positives, and enhancing the performance of the neural network. Attached Figure Description

[0045] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0046] The present invention will now be further described with reference to the accompanying drawings.

[0047] To enable those skilled in the art to better understand the solutions proposed in this invention, the following detailed description of the implementation examples of this discovery is provided in conjunction with the accompanying drawings and embodiments.

[0048] This invention provides a multi-feature extraction and fusion method for magnetic anomaly detection. The basic idea of ​​this method is to select the time-frequency features, statistical features, and target magnetic moment features of the magnetic anomaly signal as multiple features, and then use kernel principal component analysis to perform multi-feature fusion at the feature layer to process the magnetic anomaly signal data, providing effective input information for a neural network framework. Figure 1 The flowchart is shown below:

[0049] 1. Determine the time-frequency characteristics of the magnetic anomaly signal: Currently, wavelet transform has become a powerful tool for processing non-stationary signals. The magnetic anomaly signal generated by the target is a non-stationary signal. Therefore, this invention uses wavelet transform to obtain the time-frequency characteristics of the magnetic anomaly signal.

[0050] The wavelet transform is:

[0051] in, This is a magnetic anomaly signal. It is a wavelet function.

[0052] 2. Determine the statistical characteristics of the magnetic anomaly signal: When affected by the target magnetic moment, the original magnetic field will change. Statistical characteristics such as mean, variance, extrema, and peak value can all be used as statistical features of the neural network. The mean describes the overall characteristics of the magnetic anomaly signal. Extrema can represent peak values. Variance and peak value reflect the differences in peak values. These statistical characteristics can be calculated using the following formula.

[0053]

[0054]

[0055] β3=max{S}

[0056] β4=min{S}

[0057]

[0058] Where S represents the measured signal and N represents the number of samples of the signal.

[0059] 3. Determine the target magnetic moment signal of the magnetic anomaly signal. Based on prior knowledge of the target magnetic moment, its characteristics can be represented by a set of coefficients decomposed into orthogonal basis functions. According to OBF theory, the magnetic anomaly signal T can be... r Decomposed into a set of orthogonal basis functions, the three coefficients are as follows:

[0060]

[0061] In the formula f i (τ)(i=1,2,3) are the orthogonal basis functions of the decomposition.

[0062]

[0063]

[0064]

[0065] The coefficients α of the basis functions i (i = 1, 2, 3) can be obtained using the following formula:

[0066]

[0067] Where μ0 = 4π × 10 -7 H / m

[0068] 4. After multi-feature extraction, redundant information in the magnetic anomaly signal can be reduced, computational power can be decreased, and effective information can be retained. However, redundancy issues can arise in the nonlinear relationships between features in the feature layer, affecting the results of magnetic anomaly detection. Therefore, kernel component analysis can be used to fuse feature vectors in the feature layer to reduce nonlinear redundancy. The process is as follows:

[0069] The magnetic feature sample set is X m×n The nonlinear kernel function is φ(x), and its covariance matrix can be written as:

[0070]

[0071] The eigenset λ corresponding to the covariance matrix C, and the eigenvector set α, are:

[0072] Cα=λα

[0073] For the above formula, use φ(x) k Taking the inner product of (k = 1, 2, ..., n), we get:

[0074]

[0075] The computation of the inner product is quite complex. Here, we can introduce a kernel function to obtain the kernel matrix K(x). i ,x j )=<φ(x i ),φ(x j By simplifying the inner product using the method of )>, the above expression can be simplified to

[0076] nλα=Kα

[0077] The feature vector set can be represented as α=(α1,α2,···,α n ) T ,make For the non-zero feature set of K, the above formula can be simplified to:

[0078]

[0079] Will sorted in descending order The corresponding eigenvectors are α1>α2>α3>···>α n Given a contribution rate, the required number of feature values ​​is k. The first k feature values ​​are selected to form the fusion B. m×k The feature sample set Y after kernel principal component analysis k×n for:

[0080]

[0081] 5. The processed feature information is then used as input to the neural network architecture.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for multi-feature extraction and fusion in magnetic anomaly detection, characterized in that: The steps are as follows: Step 1: Determine the time-frequency characteristics of the magnetic anomaly signal; Step 2: Determine the statistical characteristics of the magnetic anomaly signal; Step 3: Determine the target magnetic moment signal of the magnetic anomaly signal; Step 4: In the feature layer, use kernel component analysis to fuse the feature vectors and reduce nonlinear redundancy; After multi-feature extraction, the redundant information of the magnetic anomaly signal is reduced, the computing power is reduced, and the effective information is preserved. However, in the feature layer, the nonlinear relationship of each feature may cause redundancy problems, which may affect the results of magnetic anomaly detection. Therefore, in the feature layer, kernel component analysis is used to fuse the feature vectors to reduce nonlinear redundancy, and the process is as follows: The magnetic signature sample set is , the nonlinear kernel function is , and its covariance matrix is written as ； The set of eigenvectors corresponding to the covariance matrix C The set of eigenvectors is There are: ； Taking the inner product of the above equation with we obtain: ； The calculation of inner product is very complex, by introducing the kernel function, the kernel matrix The method simplifies the inner product, and the above formula is simplified as: ； The set of eigenvectors is denoted as , let , be the non-zero eigenset of K, then the above equation simplifies to: ； Will descending order , the corresponding feature vector is The number of required eigenvalues is k under the given contribution rate, and the first k eigenvalues are selected to form the fusion , the feature sample set after kernel principal component analysis : ； Step 5: The processed feature information is used as the input of the neural network and enters the neural network architecture. 2.The method of claim 1, wherein: The step 1 uses wavelet transform to obtain the time-frequency characteristics of the magnetic anomaly signal; The wavelet transform is as follows: ； wherein is the magnetic anomaly signal.

3. The method of claim 1, wherein the method comprises: The statistical characteristics of the step 2, such as mean, variance, extreme value and peak value, are calculated by the following formula: ； ； ； ； ； wherein denotes the measurement signal, denotes the number of samples of the signal.

4. The method of claim 1, wherein the method comprises: The step 3 decomposes the magnetic anomaly signal into a set of orthogonal base functions according to the prior knowledge about the target magnetic moment The three coefficients are expressed as follows: ； wherein , are the decomposed orthogonal basis functions; ； ； ； Coefficients of the basis functions By the formula ； wherein .

Images