Underwater target shaft frequency magnetic field signal identification method based on unsupervised machine learning

Abstract

The invention provides an underwater target shaft frequency magnetic field signal identification method based on unsupervised machine learning, and the method comprises the steps: obtaining a plurality of groups of synchronously observed shaft frequency magnetic field data time sequences through an array type magnetic float or an unmanned plane cluster, and constructing a multi-channel observation data set; performing principal component analysis on the multi-channel observation data set, and extracting and separating large-scale low-frequency interference signals in the original data; segmenting the observation data of each channel after the low-frequency interference signals are removed according to a time sequence, and performing clustering analysis on a signal matrix formed by time slices; selecting the shaft frequency magnetic field observation data corresponding to the center point of each cluster and performing correlation analysis on the shaft frequency magnetic field observation data and the shaft frequency signals generated by simulation; and selecting the time slice with the highest correlation coefficient to carry out power spectrum analysis, if the time slice accords with the line spectrum characteristics of the shaft frequency signal, determining that a target signal exists, otherwise, determining that the target signal does not exist. By applying the technical scheme of the invention, the technical problem of low signal detection efficiency in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of geophysical electromagnetism and underwater target electromagnetic detection technology, and in particular to a method for identifying underwater target axial frequency magnetic field signals based on unsupervised machine learning. Background Art

[0002] When a ship is sailing in the marine environment, corrosion current will be generated in the seawater due to the electrochemical reaction of different metal structural parts. At the same time, in order to prevent the corrosion of the hull, various artificial external protection systems will also output anti-corrosion current into the seawater. Due to the good conductivity of seawater, the main part of the corrosion and anti-corrosion current will form a closed loop through the seawater, the hull, the propeller and the propeller shaft. The rotation of the propeller causes the contact resistance of the shaft system to change periodically, which makes the corrosion and anti-corrosion current in the seawater also pulsate periodically, thereby exciting an extremely low-frequency alternating electromagnetic field in the seawater, which is called the shaft-frequency electromagnetic field. The shaft-frequency electromagnetic field has obvious line spectrum and harmonic wave characteristics, which is obviously different from the environmental magnetic interference such as geology and daily changes. It is an important physical feature for underwater target detection and identification. At present, there are many studies on the shaft-frequency electric field at home and abroad, while the research on the shaft-frequency magnetic field is relatively weak.

[0003] In the early days, shaft frequency magnetic field detection was mainly carried out through inductive magnetic sensors fixed on the seabed. In recent years, with the advancement of small-volume, high-sensitivity atomic magnetometer payloads and observation platform technologies such as aviation and buoys, the application styles of shaft frequency magnetic fields have become more diverse, and the scale of data collection has been further expanded. By deploying multiple groups of magnetic buoys underwater, a normalized observation array can be formed to collect multi-channel shaft frequency magnetic field signals at the same time. Through a multi-machine cluster of unmanned aerial vehicles, an observation array can also be formed in the air, which can not only collect multi-channel shaft frequency magnetic field signals at the same time, but also dynamically track the target with the advantages of strong maneuverability and fast movement speed of the aviation platform. For the above application modes, how to quickly and accurately identify shaft frequency magnetic field signals from observation data is an urgent problem to be solved. Traditional shaft frequency signal recognition methods include adaptive filtering, high-order spectrum analysis, wavelet transform, etc., which are all for single-channel observation data and have not yet been applied to array observation signals, resulting in low signal detection efficiency. Summary of the invention

[0004] The present invention provides a method for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning, which can solve the technical problem of low signal detection efficiency in the prior art.

[0005] According to one aspect of the present invention, a method for identifying shaft-frequency magnetic field signals of underwater targets based on unsupervised machine learning is provided. The method for identifying shaft-frequency magnetic field signals of underwater targets based on unsupervised machine learning includes: step one, obtaining multiple groups of synchronously observed shaft-frequency magnetic field data time series through array-type magnetic buoys or drone clusters, and constructing a multi-channel observation data set; step two, performing principal component analysis on the multi-channel observation data set, extracting and separating large-scale low-frequency interference signals in the original data; step three, segmenting the observation data of each channel after removing the low-frequency interference signal in chronological order, and performing cluster analysis on the signal matrix composed of time slices; step four, selecting the shaft-frequency magnetic field observation data corresponding to the center point of each cluster and the shaft-frequency signal generated by simulation for correlation analysis; step five, selecting the time slice with the highest correlation coefficient for power spectrum analysis, if it meets the line spectrum characteristics of the shaft-frequency signal, it is considered that the target signal exists, otherwise it is considered that it does not exist.

[0006] Furthermore, in step one, for underwater detection, multiple groups of magnetic buoys equipped with underwater shaft-frequency magnetic field detection sensors are deployed to form an observation array, and data is collected synchronously and continuously; for aerial detection, multiple drones equipped with aerial shaft-frequency magnetic field detection sensors form a cluster, search side by side along a predetermined route, and collect data synchronously and dynamically. Both of the above methods can obtain multi-channel shaft-frequency magnetic field observation data and summarize them to the ship data processing terminal, and then through timing correction and interpolation, the observation data of each channel are unified to the same time base to form a standard data set.

[0007] Furthermore, step 2 specifically includes: principal component analysis is implemented by singular value decomposition algorithm, and the matrix Z is composed of multiple data sets, with a dimension of L*n, where n represents the number of channels and L represents the number of sampling points of each data. First, the matrix Z is converted into a square matrix C by matrix transposition multiplication, that is, C=Z*Z T ; Solve for the eigenvalue λ of the square matrix C 1 ,λ 2 ,...λ m and the corresponding eigenvector υ 1 ,υ 2 ,...υ m , the eigenvector υ 1 ,υ 2 ,...υ m After normalization, the matrix V is formed, which is an n×n orthogonal matrix. The singular values ​​are obtained by taking the square root of the eigenvalues. And arrange them in order from small to large into a diagonal matrix Σ, which is a L×n matrix. The elements on the diagonal are singular values, and the rest of the elements are 0; calculate The column vectors of U are obtained to form an L×L matrix U, and U is also an orthogonal matrix. After obtaining U, Σ, V, they can be reconstructed according to the inverse process of decomposition to obtain the original data matrix; the singular values ​​in the diagonal matrix Σ represent the weights of different signal components in the original data matrix. The first largest singular value represents the large-scale low-frequency interference in each column, which is retained, and the remaining singular values ​​are set to 0 to obtain the processed diagonal matrix Σ k , reconstructed, the reconstructed matrix is ​​as follows: Z k =U*Σ k *V T ; Matrix Z k Represents large-scale low-frequency interference, from the original data matrix Z k Subtract Z from k , and get the data after separating the interference: Z 0 =ZZ k .

[0008] Furthermore, step three specifically includes: processing the preprocessed multi-channel axial frequency magnetic field observation data Z 0 The data are divided in chronological order, and each channel data consists of multiple groups of time slices; all data slices are combined into a data matrix S, with a total of n*m segments, and the total number of sampling points of each channel data is L, which is divided into m segments, and each segment is l=L / m sampling points; cluster analysis is performed on the data sample S, and the number of clusters is K. K data samples are randomly selected from the axial frequency magnetic field data samples as initial cluster centers, and the Euclidean distance of each data sample to the K cluster centers is calculated; each data sample is assigned to the category to which the cluster center closest to it belongs according to the Euclidean distance, and for each category, its cluster center is recalculated; the above steps of calculating the Euclidean distance, assigning data samples, and updating the cluster centers are repeated until the stopping condition is met. The stopping condition is usually that the cluster center no longer changes significantly, or the preset maximum number of iterations is reached.

[0009] Furthermore, in step 3, the Euclidean distance calculation formula is Among them, S i =(s 1 ,s 2 ,s 3 ,...,s l ) is a data sample, C j =(c 1 ,c 2 ,c 3 ,…,c l ) is the cluster center.

[0010] Furthermore, in step 3, for each category, its cluster center is recalculated. If there are h data samples in category j, the new cluster center is calculated as follows: Among them, S k is a vector containing all the features.

[0011] Furthermore, step 4 specifically includes: simulating and generating a set of shaft frequency magnetic field signals with different fundamental frequencies. Selecting the data sample at the center of each cluster as the representative sample of the cluster, and performing correlation analysis with the shaft frequency magnetic field signal generated by simulation in turn, the calculation formula is as follows: Where A is the standard shaft frequency signal generated by simulation, B is a data sample of a cluster center, ρ(A,B) is the correlation coefficient, μ() is the mean, σ(A) is the standard deviation, and for each cluster representative signal, record the maximum correlation coefficient ρ 0 And the fundamental frequency f of the simulated shaft frequency signal corresponding to the maximum value 0 , and then compare the correlation coefficients of all cluster representative signals, and take the cluster representative sample with the largest correlation coefficient as the final representative signal among all the above samples.

[0012] Furthermore, step five specifically includes: selecting the final representative signal with the strongest correlation with the simulation signal for power spectrum analysis, and checking the power spectrum at f 0 ,2f 0 ,3f 0 ,4f 0 ,5f 0 The amplitude at the equal frequency point value is consistent with the axis frequency line spectrum characteristics. If it is consistent, it is considered that there is an axis frequency magnetic field signal in the data sample. Otherwise, it is considered that there is no axis frequency signal in the entire data set. 0 is the fundamental frequency of the simulated shaft frequency signal corresponding to the maximum value of the correlation coefficient.

[0013] Furthermore, the final representative signal with the strongest correlation with the simulation signal is selected for power spectrum analysis, which specifically includes: normalizing the Hanning window; multiplying the final representative signal with the strongest correlation with the simulation signal with the normalized Hanning window, and then performing Fourier transform, calculating the power spectrum density, and converting the square root into the amplitude spectrum to obtain the energy spectrum of the final representative signal.

[0014] According to another aspect of the present invention, a system for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning is provided. The system for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning uses the underwater target shaft frequency magnetic field signal identification method based on unsupervised machine learning as described above to perform underwater target shaft frequency magnetic field signal identification.

[0015] The technical solution of the present invention is applied to provide a method for identifying shaft-frequency magnetic field signals of underwater targets based on unsupervised machine learning. The method processes and analyzes multi-channel shaft-frequency magnetic field data observed in an array manner through two unsupervised machine learning algorithms, principal component analysis and cluster analysis, to extract target signals. First, through principal component analysis, large-scale low-frequency interference that may exist in each channel is extracted and separated. Then, through cluster analysis, the array data set is divided into several clusters. Then, samples at the center of the clusters are selected for correlation analysis and power spectrum analysis to identify shaft-frequency magnetic field signals. This can greatly improve the signal recognition efficiency and accuracy in the shaft-frequency magnetic field array detection method of underwater targets. 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 flowchart of a method for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning according to a specific embodiment of the present invention is shown;

[0018] Figure 2 A schematic diagram of an array-type magnetic buoy and a drone cluster observation method provided in a specific embodiment of the present invention is shown;

[0019] Figure 3 A schematic diagram of multi-channel observation data slicing segmentation provided according to a specific embodiment of the present invention is shown. DETAILED DESCRIPTION

[0020] 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.

[0021] 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.

[0022] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​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, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method 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.

[0023] like Figure 1 As shown, according to a specific embodiment of the present invention, a method for identifying shaft-frequency magnetic field signals of underwater targets based on unsupervised machine learning is provided, and the method for identifying shaft-frequency magnetic field signals of underwater targets based on unsupervised machine learning includes: step one, obtaining multiple groups of synchronously observed shaft-frequency magnetic field data time series through an array magnetic buoy or a drone cluster, and constructing a multi-channel observation data set; step two, performing principal component analysis on the multi-channel observation data set, extracting and separating large-scale low-frequency interference signals in the original data; step three, segmenting the observation data of each channel after removing the low-frequency interference signal in chronological order, and performing cluster analysis on the signal matrix composed of time slices; step four, selecting the shaft-frequency magnetic field observation data corresponding to the center point of each cluster and the shaft-frequency signal generated by simulation for correlation analysis; step five, selecting the time slice with the highest correlation coefficient for power spectrum analysis, if it meets the line spectrum characteristics of the shaft-frequency signal, it is considered that the target signal exists, otherwise it is considered that it does not exist.

[0024] By applying this configuration, a method for underwater target shaft frequency magnetic field signal recognition based on unsupervised machine learning is provided. This method processes and analyzes the multi-channel shaft frequency magnetic field data of array observation through two unsupervised machine learning algorithms, principal component analysis and cluster analysis, to extract the target signal. First, through principal component analysis, the large-scale low-frequency interference that may exist in each channel is extracted and separated. Then, through cluster analysis, the array data set is divided into several clusters. Then, the samples at the center of the clusters are selected for correlation analysis and power spectrum analysis to identify the shaft frequency magnetic field signal, which can greatly improve the signal recognition efficiency and accuracy in the underwater target shaft frequency magnetic field array detection method.

[0025] Furthermore, in the present invention, in step one, for underwater detection, multiple groups of magnetic buoys equipped with underwater shaft-frequency magnetic field detection sensors are deployed to form an observation array, and data is collected synchronously and continuously; for aerial detection, multiple unmanned aerial vehicles equipped with aerial shaft-frequency magnetic field detection sensors form a cluster, search side by side according to a predetermined route, and collect data synchronously and dynamically. Both of the above methods can obtain multi-channel shaft-frequency magnetic field observation data and summarize them to the ship data processing terminal, and then through timing correction and interpolation, the observation data of each channel are unified to the same time base to form a standard data set.

[0026] In the present invention, step 2 specifically includes: principal component analysis is implemented by singular value decomposition algorithm, and a matrix Z is formed by multiple data sets, and the dimension is L*n, where n represents the number of channels and L represents the number of sampling points of each data. First, the matrix Z is converted into a square matrix C by matrix transposition multiplication, that is, C=Z*Z T ; Solve for the eigenvalue λ of the square matrix C 1 ,λ 2 ,…λ m and the corresponding eigenvector υ 1 ,υ 2 ,...υ m , the eigenvector υ 1 ,υ 2 ,...υ m After normalization, the matrix V is formed, which is an n×n orthogonal matrix. The singular values ​​are obtained by taking the square root of the eigenvalues. And arrange them in order from small to large into a diagonal matrix Σ, which is a L×n matrix. The elements on the diagonal are singular values, and the rest of the elements are 0; calculate The column vectors of U are obtained to form an L×L matrix U, and U is also an orthogonal matrix. After obtaining U, Σ, V, they can be reconstructed according to the inverse process of decomposition to obtain the original data matrix; the singular values ​​in the diagonal matrix Σ represent the weights of different signal components in the original data matrix. The first largest singular value represents the large-scale low-frequency interference in each column, which is retained, and the remaining singular values ​​are set to 0 to obtain the processed diagonal matrix Σ k, reconstructed, the reconstructed matrix is ​​as follows: Z k =U*Σ k *V T ; Matrix Z k Represents large-scale low-frequency interference, from the original data matrix Z k Subtract Z from k , and get the data after separating the interference: Z 0 =ZZ k .

[0027] Furthermore, step three specifically includes: processing the preprocessed multi-channel axial frequency magnetic field observation data Z 0 The data are divided in chronological order, and each channel data consists of multiple groups of time slices; all data slices are combined into a data matrix S, with a total of n*m segments, and the total number of sampling points of each channel data is L, which is divided into m segments, and each segment is l=L / m sampling points; cluster analysis is performed on the data sample S, and the number of clusters is K. K data samples are randomly selected from the axial frequency magnetic field data samples as initial cluster centers, and the Euclidean distance of each data sample to the K cluster centers is calculated; each data sample is assigned to the category to which the cluster center closest to it belongs according to the Euclidean distance, and for each category, its cluster center is recalculated; the above steps of calculating the Euclidean distance, assigning data samples, and updating the cluster centers are repeated until the stopping condition is met. The stopping condition is usually that the cluster center no longer changes significantly, or the preset maximum number of iterations is reached.

[0028] Among them, in step 3, the Euclidean distance calculation formula is Among them, S i =(s 1 ,s 2 ,s 3 ,…,s l ) is a data sample, C j =(c 1 ,c 2 ,c 3 ,…,c l ) is the cluster center. In step 3, for each category, its cluster center is recalculated. If there are h data samples in category j, the new cluster center is calculated as follows: Among them, S k is a vector containing all the features.

[0029] Furthermore, step 4 specifically includes: simulating and generating a set of shaft frequency magnetic field signals with different fundamental frequencies. Selecting the data sample at the center of each cluster as the representative sample of the cluster, and performing correlation analysis with the shaft frequency magnetic field signal generated by simulation in turn, the calculation formula is as follows: Where A is the standard shaft frequency signal generated by simulation, B is a data sample of a cluster center, ρ(A,B) is the correlation coefficient, μ() is the mean, σ(A) is the standard deviation, and for each cluster representative signal, record the maximum correlation coefficient ρ 0 And the fundamental frequency f of the simulated shaft frequency signal corresponding to the maximum value 0 , and then compare the correlation coefficients of all cluster representative signals, and take the cluster representative sample with the largest correlation coefficient as the final representative signal among all the above samples.

[0030] Furthermore, step five specifically includes: selecting the final representative signal with the strongest correlation with the simulation signal for power spectrum analysis, and checking the power spectrum at f 0 ,2f 0 ,3f 0 ,4f 0 ,5f 0 The amplitude at the equal frequency point value is consistent with the axis frequency line spectrum characteristics. If it is consistent, it is considered that there is an axis frequency magnetic field signal in the data sample. Otherwise, it is considered that there is no axis frequency signal in the entire data set. 0 is the fundamental frequency of the simulated shaft frequency signal.

[0031] Among them, selecting the final representative signal with the strongest correlation with the simulation signal for power spectrum analysis specifically includes: normalizing the Hanning window; multiplying the final representative signal with the strongest correlation with the simulation signal with the normalized Hanning window, and then performing Fourier transform, calculating the power spectrum density, and converting the square root into the amplitude spectrum to obtain the energy spectrum of the final representative signal.

[0032] According to another aspect of the present invention, a system for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning is provided. The system for identifying underwater target shaft frequency magnetic field signals based on unsupervised machine learning uses the underwater target shaft frequency magnetic field signal identification method based on unsupervised machine learning as described above to perform underwater target shaft frequency magnetic field signal identification.

[0033] By applying this configuration, a shaft frequency magnetic field signal recognition system for underwater targets based on unsupervised machine learning is provided. The system processes and analyzes the multi-channel shaft frequency magnetic field data of array observation through two unsupervised machine learning algorithms, principal component analysis and cluster analysis, to extract target signals. First, through principal component analysis, large-scale low-frequency interference that may exist in each channel is extracted and separated. Then, through cluster analysis, the array data set is divided into several clusters. Then, the samples at the center of the clusters are selected for correlation analysis and power spectrum analysis to identify shaft frequency magnetic field signals. This can greatly improve the signal recognition efficiency and accuracy in the shaft frequency magnetic field array detection method for underwater targets.

[0034] In order to further understand the present invention, the following Figures 1 to 3The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning provided by the present invention is described in detail.

[0035] like Figures 1 to 3 As shown, according to a specific embodiment of the present invention, a method and system for underwater target shaft frequency magnetic field signal recognition based on unsupervised machine learning is provided, which can improve the signal recognition efficiency and accuracy in the underwater target shaft frequency magnetic field array detection method. The method specifically includes the following steps.

[0036] Step 1: Obtain multiple sets of synchronously observed axial frequency magnetic field data time series through array magnetic buoys or drone clusters to construct a multi-channel observation data set.

[0037] The present invention is suitable for aviation and underwater shaft frequency magnetic field detection applications. Figure 2 As shown in the figure: For underwater detection, multiple groups of magnetic buoys equipped with underwater shaft frequency magnetic field detection sensors are deployed to form an observation array to synchronously and continuously collect data; for aerial detection, multiple drones equipped with aviation shaft frequency magnetic field detection sensors form a cluster, search side by side according to the predetermined route, and synchronously and dynamically collect data. Both of the above methods can obtain multi-channel shaft frequency magnetic field observation data and summarize them to the ship data processing terminal, and then through timing correction and interpolation, the observation data of each channel are unified to the same time reference to form a standard data set. The measured data at sea include not only the shaft frequency magnetic field signals generated by underwater targets, but also various electromagnetic noise interferences, such as environmental magnetic interference caused by geology, daily changes, waves, lightning, etc., and artificial electromagnetic interference caused by offshore drilling platforms, submarine transportation pipelines, electromagnetic equipment of surface ships, carrier platforms, etc.

[0038] Step 2: Perform principal component analysis on the multi-channel observation data set to extract and separate large-scale low-frequency interference signals from the original data.

[0039] For multi-channel observation data sets, the components with the strongest energy in each signal are low-frequency, large-scale interference caused by geology, diurnal changes, etc., which need to be separated first. Since environmental interference such as geology and diurnal changes are distributed over a large range in space, the above interference observed in each channel has a strong correlation and can be extracted and separated by principal component analysis.

[0040] Principal component analysis is implemented through the singular value decomposition algorithm. The multi-channel data set constitutes a matrix Z with a dimension of L*n, where n represents the number of channels and L represents the number of sampling points for each channel of data. First, the matrix Z is converted into a square matrix C by matrix transposition multiplication, that is:

[0041] C=Z*Z T (1)

[0042] Then solve the eigenvalue λ of the square matrix C 1 ,λ2 ,…λ m and the corresponding eigenvector υ 1 ,υ 2 ,…υ m . The eigenvector υ 1 ,υ 2 ,…υ m After normalization, the matrix V is formed, which is an n×n orthogonal matrix. The singular value is obtained by taking the square root of the eigenvalue. And arrange them in order from small to large into a diagonal matrix Σ, which is a L×n matrix. The elements on the diagonal are singular values, and the rest of the elements are 0. Then calculate The column vectors of U are obtained to form an L×L matrix U, which is also an orthogonal matrix. After obtaining U,Σ,V, we can reconstruct it according to the inverse process of decomposition to obtain the original matrix.

[0043] In the above process, the singular values ​​in the diagonal matrix Σ represent the weights of different signal components in the original data matrix. The first largest singular value represents the large-scale low-frequency interference existing in each column, which is retained, and the remaining singular values ​​are set to 0 to obtain the processed diagonal matrix Σ k , reconstructed, the reconstructed matrix is ​​as follows:

[0044] Z k =U*Σ k *V T (2)

[0045] Matrix Z k Represents large-scale low-frequency interference, from the original data matrix Z k Subtract Z from k , and get the data after the interference is separated:

[0046] Z 0 =ZZ k (3)

[0047] Step three, segment the observation data of each channel in time sequence, and perform cluster analysis on the signal matrix composed of time slices.

[0048] The preprocessed multi-channel axial frequency magnetic field observation data Z0 are segmented in time sequence, and each data is composed of multiple groups of time slices. Figure 3 As shown, S 11 ,S 12 ,S 13 ,…is the data slice obtained by segmenting the first channel data, S 21 ,S 22 ,S 23 ,…is the data slice obtained by segmenting the second channel data, S 31 ,S 32 ,S33 ,…are the data slices obtained by segmenting the third channel data.

[0049] Then all the above data slices are formed into a data matrix S, which has n*m segments in total. The total number of sampling points of each channel data is L, which is divided into m segments, and each segment has l=L / m sampling points.

[0050] S=[S 11 ,S 12 ,…,S 1m ,S 21 ,S 22 ,…,S 2m ,S n1 ,S n2 ,…,S nm ] (4)

[0051] Then perform cluster analysis on the data sample S, the number of clusters is K, and K can usually be set to 5. Randomly select K data samples from the axial frequency magnetic field data samples as the initial cluster centers. Calculate the Euclidean distance from each data sample to the K cluster centers. For a data sample S i =(s 1 ,s 2 ,s 3 ,…,s l ) and cluster center C j =(c 1 ,c 2 ,c 3 ,…,c l ), the distance calculation formula is as follows:

[0052]

[0053] Assign each data sample to the category to which the cluster center closest to it belongs according to the distance. For each category, recalculate its cluster center. If there are h data samples in category j, the new cluster center is calculated as follows:

[0054]

[0055] Among them, S k is a vector containing all features. Repeat the above steps of calculating Euclidean distance, assigning data samples, updating cluster centers, etc. until the stopping condition is met. The stopping condition is usually that the cluster center no longer changes significantly or reaches the preset maximum number of iterations.

[0056] Step 4: Select the shaft frequency magnetic field observation data corresponding to the center point of each cluster and perform correlation analysis with the shaft frequency signal generated by simulation.

[0057] At this point, the original n*m data slices have been divided into K clusters, and the data samples in each cluster have similar characteristics. By looking at the data sample characteristics in each cluster, we can understand the differences between different categories and infer the different states or sources of the signal.

[0058] The simulation generates a set of shaft frequency magnetic field signals with different fundamental frequencies. The data sample at the center of each cluster is selected as the representative sample of the cluster, and the correlation analysis is performed with the shaft frequency magnetic field signal generated by the simulation. The calculation formula is as follows:

[0059]

[0060] Where A is the standard shaft frequency signal generated by simulation, B is a data sample of a cluster center, ρ(A,B) is the correlation coefficient, μ() is the mean, and σ(A) is the standard deviation. For each cluster representative signal, record its maximum correlation coefficient ρ 0 And the fundamental frequency f of the simulated shaft frequency signal corresponding to the maximum value 0 Then compare the correlation coefficients of the representative signals of all clusters, and take the representative sample of the cluster with the largest correlation coefficient as the final representative signal among all the above samples.

[0061] Step 5: Select the time slice with the highest correlation coefficient for power spectrum analysis. If it meets the line spectrum characteristics of the axis frequency signal, it is considered that the target signal exists, otherwise it is considered that it does not exist.

[0062] The final representative signal with the strongest correlation with the simulation signal obtained in the previous step is selected for power spectrum analysis. The calculation process includes: adding Hanning window, Fourier transform, and calculating power spectrum. The Hanning window formula is as follows:

[0063] w 0 (k)=0.54-0.46*cos(2kπ / l), 0≤k≤l (8)

[0064] Then the Hanning window is normalized:

[0065]

[0066] Then, the final representative signal with the strongest correlation with the simulation signal obtained in step 4 is multiplied by the normalized Hanning window, and then Fourier transform is performed to calculate the power spectrum density and square root conversion to amplitude spectrum. The formula is as follows:

[0067]

[0068] In the above formula, k is the sampling point number, fft() is the fast Fourier transform function, and conj() is the function for finding the conjugate complex number. The P obtained at this time is 0 is the amplitude spectrum of the final representative signal with the strongest correlation with the simulation signal, fs is the sampling rate of the final representative signal that has the strongest correlation with the simulated signal.

[0069] Then check the amplitude spectrum at f 0 ,2f 0 ,3f 0 ,4f 0 ,5f 0 , the amplitude at the equal frequency point value is consistent with the axis frequency line spectrum characteristics. If it is consistent, it is considered that there is an axis frequency magnetic field signal in the data sample. Otherwise, it is considered that there is no axis frequency signal in the entire data set. 0 is the fundamental frequency of the simulated shaft frequency signal corresponding to the maximum value of the correlation coefficient.

[0070] The present invention processes and analyzes the multi-channel axial frequency magnetic field data of array observation through two unsupervised machine learning algorithms, principal component analysis and cluster analysis, to extract the target signal. First, through principal component analysis, the large-scale low-frequency interference that may exist in each channel is extracted and separated, and then through cluster analysis, the array data set is divided into several clusters, and then the cluster center samples are selected for correlation analysis and power spectrum analysis to identify the axial frequency magnetic field signal, which can greatly improve the efficiency of axial frequency magnetic field signal detection.

[0071] 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.

[0072] 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.

[0073] 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 method for identifying shaft frequency magnetic field signals of underwater targets based on unsupervised machine learning, characterized in that: The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning includes: Step 1: Acquire multiple sets of synchronously observed axial frequency magnetic field data time series through array magnetic buoys or drone clusters to construct a multi-channel observation data set; Step 2: Perform principal component analysis on the multi-channel observation data set to extract and separate large-scale low-frequency interference signals from the original data; Step 3: Segment the observation data of each channel after removing the low-frequency interference signal in time sequence, and perform cluster analysis on the signal matrix composed of time slices; Step 4: Select the shaft frequency magnetic field observation data corresponding to the center point of each cluster and the shaft frequency signal generated by simulation for correlation analysis; Step 5: Select the time slice with the highest correlation coefficient for power spectrum analysis. If it meets the line spectrum characteristics of the axis frequency signal, it is considered that the target signal exists, otherwise it is considered that it does not exist.

2. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 1 is characterized in that: In the step one, for underwater detection, multiple groups of magnetic buoys equipped with underwater shaft-frequency magnetic field detection sensors are deployed to form an observation array, and data is collected synchronously and continuously; for aerial detection, multiple drones equipped with aerial shaft-frequency magnetic field detection sensors are formed into a cluster, and they search side by side along a predetermined route and collect data synchronously and dynamically. Both of the above methods can obtain multi-channel shaft-frequency magnetic field observation data and summarize them to the ship data processing terminal, and then through timing correction and interpolation, the observation data of each channel are unified to the same time base to form a standard data set.

3. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 1 is characterized in that: The step 2 specifically includes: Principal component analysis is implemented through the singular value decomposition algorithm. The matrix Z is composed of multiple data sets with a dimension of L*n, where n represents the number of channels and L represents the number of sampling points of each data. First, the matrix Z is converted into a square matrix C by matrix transposition multiplication, that is, C=Z*Z T ; Solve for the eigenvalues ​​λ1,λ2,…λ of the square matrix C m and the corresponding eigenvectors υ1,υ2,…υ m , the eigenvectors υ1,υ2,...υ m After normalization, the matrix V is formed, which is an n×n orthogonal matrix. The singular values ​​are obtained by taking the square root of the eigenvalues. And arrange them in ascending order into a diagonal matrix Σ, which is an L×n matrix. The elements on the diagonal are singular values, and the rest of the elements are 0; calculate Get the column vector of U to form an L×L matrix U, and U is also an orthogonal matrix. After getting U,Σ,V, we can reconstruct it according to the inverse process of decomposition to get the original data matrix; The singular values ​​in the diagonal matrix Σ represent the weights of different signal components in the original data matrix. The first largest singular value represents the large-scale low-frequency interference in each column. It is retained and the remaining singular values ​​are set to 0 to obtain the processed diagonal matrix Σ. k , reconstructed, the reconstructed matrix is ​​as follows: Z k =U*Σ k *V T ; Matrix Z k Represents large-scale low-frequency interference, from the original data matrix Z k Subtract Z from k , get the data after separating the interference: Z0=ZZ k .

4. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 3 is characterized in that: The step three specifically includes: The preprocessed multi-channel axial frequency magnetic field observation data Z0 are segmented in time sequence, and each channel data consists of multiple groups of time slices; All data slices are formed into a data matrix S, with a total of n*m segments. The total number of sampling points of each channel data is L, which is divided into m segments, and each segment has l=L / m sampling points; Performing cluster analysis on the data sample S, the number of clusters is K, randomly selecting K data samples from the axial frequency magnetic field data samples as initial cluster centers, and calculating the Euclidean distance from each data sample to the K cluster centers; Assign each data sample to the category to which the cluster center closest to it belongs according to the Euclidean distance, and recalculate the cluster center for each category; Repeat the above steps of calculating the Euclidean distance, allocating data samples, and updating the cluster center until the stopping condition is met. The stopping condition is usually that the cluster center no longer changes significantly or the preset maximum number of iterations is reached.

5. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 4 is characterized in that: In step 3, the Euclidean distance calculation formula is: Among them, S i =(s1,s2,s3,...,s l ) is a data sample, C j =(c1,c2,c3,...,c l ) is the cluster center.

6. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 5 is characterized in that: In step 3, for each category, its cluster center is recalculated. If there are h data samples in category j, the new cluster center is calculated as follows: Among them, S k is a vector containing all the features.

7. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 6 is characterized in that: The step 4 specifically includes: simulating and generating a set of shaft frequency magnetic field signals with different fundamental frequencies. Selecting the data sample at the center of each cluster as the representative sample of the cluster, and sequentially performing correlation analysis with the shaft frequency magnetic field signals generated by simulation, the calculation formula is as follows: Where A is the standard shaft frequency signal generated by simulation, B is a data sample at the center of a cluster, ρ(A, B) is the correlation coefficient, μ() is the mean, σ(A) is the standard deviation, and for each cluster representative signal, record the maximum correlation coefficient ρ0 and the simulated shaft frequency signal fundamental frequency f0 corresponding to the maximum value, and then compare the correlation coefficients of all cluster representative signals, and take the cluster representative sample with the largest correlation coefficient as the final representative signal among all the above samples.

8. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 7 is characterized in that: The step five specifically includes: selecting the final representative signal with the strongest correlation with the simulation signal for power spectrum analysis, and checking whether the amplitude of the power spectrum at frequency points such as f0, 2f0, 3f0, 4f0, 5f0 is consistent with the shaft frequency line spectrum characteristics. If it is consistent, it is considered that there is a shaft frequency magnetic field signal in the data sample; otherwise, it is considered that there is no shaft frequency signal in the entire data set, and f0 is the fundamental frequency of the simulation shaft frequency signal corresponding to the maximum correlation coefficient.

9. The underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning according to claim 8 is characterized in that: The final representative signal with the strongest correlation with the simulation signal is selected for power spectrum analysis, specifically including: Normalize the Hanning window; The final representative signal with the strongest correlation with the simulation signal is multiplied by the normalized Hanning window, and then Fourier transformed, the power spectrum density is calculated, and the square root is converted into the amplitude spectrum to obtain the energy spectrum of the final representative signal.

10. An underwater target shaft frequency magnetic field signal recognition system based on unsupervised machine learning, characterized in that: The underwater target shaft frequency magnetic field signal recognition system based on unsupervised machine learning uses the underwater target shaft frequency magnetic field signal recognition method based on unsupervised machine learning as described in any one of claims 1 to 9 to perform underwater target shaft frequency magnetic field signal recognition.

Images