Artificial source electromagnetic exploration denoising method, system, terminal and medium based on STL-FMD decomposition algorithm

Abstract

The present invention discloses a method, system, terminal, and medium for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm. The method comprises: obtaining observation signals from an original reference station and a noisy observation station and performing downsampling to obtain the reference station and the first observation signal of the observation station; performing STL decomposition on the first observation signal of the observation station and eliminating the trend component to obtain the second observation signal of the observation station; extracting noise using the signal variance ratio of the synchronously observed first observation signal of the reference station and the second observation signal of the observation station to obtain the noisy observation signal of the observation station; performing low-pass FMD decomposition on the noisy observation signal to obtain a noise signal; removing the noise signal to obtain a valid signal; and replacing the noisy portion of the noisy second observation signal of the observation station with the valid signal to obtain a denoised time domain signal. The combined STL and FMD decomposition effectively eliminates strong electromagnetic noise and baseline drift, improving the denoising effect and signal-to-noise ratio of the observation signal.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial source electromagnetic exploration data processing, and in particular to an artificial source electromagnetic exploration denoising method, system, terminal and medium based on an STL-FMD decomposition algorithm. Background Art

[0002] The CSEM (Constructed Source Electromagnetic Method) is a geophysical exploration method characterized by a controllable source, known signal, and strong anti-interference capabilities. It has been widely used in energy and mineral resource exploration, geological surveys, and disaster prevention and control. In field observations, strong electromagnetic interference is a major factor affecting electromagnetic exploration. Furthermore, due to the complexity and uncontrollability of noise, single denoising methods often fail to achieve satisfactory results. Therefore, effectively suppressing strong electromagnetic interference remains a research hotspot and a challenge in this field.

[0003] In the field of electromagnetic exploration data processing, empirical mode decomposition (EMD) can separate the different components of the original signal into a series of intrinsic mode functions (IMFs) based on the signal's local timescale characteristics. These components are then analyzed in the time-frequency domain to identify and separate the signal from noise. This method is considered one of the most effective tools for multi-component signal analysis and is widely used to suppress noise such as strong interference and baseline drift. However, EMD and its improved methods (such as EEMD and LMD) suffer from problems such as mode aliasing, endpoint effects, over- and under-envelopes, and susceptibility to noise. Variational mode decomposition (VMD) transforms mode decomposition into a variational problem, effectively addressing the shortcomings of methods like EMD. However, the algorithm parameters, such as the number of modal components K and the penalty factor α, determine decomposition performance. Therefore, using VMD without any prior knowledge still results in significant decomposition errors, thus reducing the denoising effect. Summary of the Invention

[0004] To address the shortcomings of existing artificial source electromagnetic exploration denoising technologies, the present invention provides an artificial source electromagnetic exploration denoising method, system, terminal and medium based on the STL-FMD decomposition algorithm. The method utilizes the STL (Seasonal-Trend decomposition procedure using Loess, a time series decomposition algorithm based on local weighted regression) decomposition to effectively eliminate the long-period nonlinear baseline drift (trend component) and large-scale step noise of the observation data. Furthermore, the low-pass FMD (Feature Mode Decomposition) decomposition is used to obtain the low-frequency noise of the processed signal, ultimately achieving effective suppression of strong electromagnetic noise and improving the signal-to-noise ratio of the observation data.

[0005] In a first aspect, a method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm is provided, comprising the following steps:

[0006] S1: Obtain the observation signals of the original reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station;

[0007] S2: Perform STL decomposition on the first observation signal of the noisy station and eliminate the trend component to obtain the second observation signal of the noisy station;

[0008] S3: noise identification and positioning is achieved by using the signal variance ratio of the first observation signal of the reference station with high signal-to-noise ratio synchronous observation and the second observation signal of the noisy measurement station, and the noisy observation signal of the noisy measurement station is extracted;

[0009] S4: Perform low-pass FMD decomposition on the noisy observation signal of the noisy observation station to obtain the noise signal;

[0010] S5: Subtract the noise signal from the noisy observation signal of the noisy observation station to obtain the effective signal after denoising;

[0011] S6: Replace the noisy part of the second observation signal of the noisy measuring station with the valid signal to obtain a complete time domain signal after denoising.

[0012] Furthermore, the downsampling process in step S1 includes:

[0013] The Nyquist sampling theorem is used to downsample the original observation signal, that is, to ensure that f s ≥2f max , and the error between the downsampled main frequency amplitude curve and the original main frequency amplitude curve is less than 1%; where f s is the sampling frequency, f max is the highest frequency of the original observed signal.

[0014] Furthermore, step S2 specifically includes:

[0015] S2.1, STL decomposes the inner loop, uses the local weighted regression algorithm to fit the periodic component, and then solves the trend component;

[0016] S2.2, STL decomposes the outer loop, obtains the residual, and calculates the robust weight in each loop, updates the weight of the local weighted regression algorithm, and uses it for the next inner loop calculation;

[0017] S2.3. When the iteration result meets the convergence condition, the decomposition is completed and the final trend component T is obtained. t ; The convergence condition is defined as:

[0018]

[0019] in, and Represents the trend component or cycle component of continuous iteration;

[0020] S2.4, the first observation signal X of the noisy station t Eliminate the trend component T t , the second observation signal of the noisy station is recorded as XN t , namely XN t =X t -T t .

[0021] Furthermore, STL decomposes the inner loop: Assume that the trend component and cycle component in the k-th iteration result are represented by T t k 、 The initial value of the trend component is set to 0, that is, T t 0 =0; the decomposition process is as follows:

[0022] S2.1.1. Detrending: Use the first observation signal X of the noisy station at the current moment t Subtract the trend component after the last inner cycle to get the detrended sequence, expressed as X t -T t k ;

[0023] S2.1.2. Periodic subsequence smoothing: Divide the sequence obtained in S2.1.1 into multiple subsequences, perform local weighted regression smoothing on each subsequence, and combine and restore the smoothed subsequences. The resulting periodic subsequences are recorded as

[0024] S2.1.3. Periodic Subsequences Low-pass filtering: For the periodic subsequence obtained in S2.1.2 Perform sliding average processing and then perform local weighted regression algorithm smoothing to obtain sequence;

[0025] S2.1.4. Detrending periodic subsequences: using As a periodic component

[0026] S2.1.5. De-periodicization of the original sequence: as the sequence following the deperiodic component;

[0027] S2.1.6, Trend smoothing: The sequence obtained in S2.1.5 is smoothed by local weighted regression algorithm to obtain the trend component T t k+1 ;

[0028] STL decomposition outer loop: trend component T obtained according to S2.1.6 t k+1 , calculate the residual Furthermore, the robust weight ρ is calculated using the residual t , the calculation formula is as follows:

[0029] in

[0030] Where R t Represents the iterative residual of this cycle, that is B(·) represents the bisquare weight function; median represents the median, and || represents the absolute value of the data;

[0031] Get the robust weight ρ t After that, the weight of the local weighted regression algorithm is updated in the inner loop, that is, the weight of the local weighted regression algorithm is multiplied by ρ t , completing the next iteration of the inner loop.

[0032] Furthermore, step S3 specifically includes:

[0033] S3.1. The signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy station for high signal-to-noise ratio synchronous observation is calculated by the following formula:

[0034]

[0035] Where, ROV i is the signal variance ratio corresponding to the i-th window, var represents the signal variance, e r represents the first observation signal of the reference station, el represents the second observation signal of the noisy station, (t i ,t i+1 ) represents the data segment corresponding to the i-th window, and Win_Nums is the number of windows;

[0036] S3.2, use the density-based clustering algorithm (DBSCAN) to perform cluster analysis on the signal variance ratio obtained in S3.1, complete the identification and positioning of the noise, and extract the noisy observation signal x t .

[0037] Furthermore, step S4 specifically includes:

[0038] S4.1, parameter initialization: set the initialization FMD parameters, decomposition modulus Mode_Nums, filter length L, maximum number of iterations Max_IterNums, low-pass filter cutoff frequency f c ;

[0039] S4.2, FIR filter design: Design M Hanning window low-pass filter banks, M = 5-10, and initialize the number of iterations j = 1;

[0040] S4.3. Noisy observation signal x t Filter decomposition: The low-pass filter designed by S4.2 is used to decompose the noisy observation signal x t Perform filtering to obtain the M modal components of the jth iteration Its expression is as follows:

[0041]

[0042] Where, represents the mth modal component in the jth iteration, represents the filter coefficient of the mth low-pass filter in the jth iteration (m=1,2,…M), and “*” represents the convolution operation;

[0043] S4.4. Filter coefficient update: using the noisy observation signal x t , modal components and estimated signal period Implement filter coefficient update and set j=j+1;

[0044] S4.5, Iteration determination: Determine whether the current iteration has reached the maximum number of iterations; if so, proceed to the next step S4.6; otherwise, return to S4.3;

[0045] S4.6, Modal component selection: Calculate the correlation coefficients of all modal components and construct the correlation matrix CC K×K , select the two modal components with the largest correlation coefficient and calculate the failure cycle by Calculate their related kurtosis; then select the modal component with larger related kurtosis as the decomposed modal component, and set M = M-1;

[0046] S4.7, FMD decomposition result judgment: judge whether the current mode number reaches the set mode number Mode_Nums. If not, return to step S4.3; otherwise, stop iteration and obtain the final FMD decomposition result; the final low-frequency noise signal can be obtained, recorded as u t .

[0047] Furthermore, the correlation coefficient between any two modal components in step S4.6 is calculated as follows:

[0048]

[0049] Where u p 、u q represents the pth and qth decomposition moduli, represents the mean of the p-th and q-th decomposition moduli, n is the sampling point, and N is the signal length.

[0050] Secondly, a denoising system for artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm is provided, comprising the following steps:

[0051] A downsampling module is used to obtain the observation signals of the original reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station;

[0052] An STL decomposition module is used to perform STL decomposition on the first observation signal of the noisy measurement station and eliminate the trend component therein to obtain the second observation signal of the noisy measurement station;

[0053] A noise identification and positioning module is used to realize noise identification and positioning by using the signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy measurement station observed synchronously with a high signal-to-noise ratio, and extract the noisy observation signal of the noisy measurement station;

[0054] The FMD decomposition module is used to perform low-pass FMD decomposition on the noisy observation signal of the noisy observation station to obtain the noise signal;

[0055] The denoising module is used to subtract the noise signal from the noisy observation signal of the noisy observation station to obtain the effective signal after denoising;

[0056] The time domain signal splicing module is used to replace the noisy part of the second observation signal of the noisy measuring station with the valid signal to obtain the complete time domain signal after denoising.

[0057] In a third aspect, an electronic terminal is provided, comprising:

[0058] a memory having a computer program stored thereon;

[0059] The processor is used to load and execute the computer program to implement the artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm as described above.

[0060] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm as described above is implemented.

[0061] The present invention proposes a denoising method, system, terminal and medium for artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm. Aiming at large-scale electromagnetic interference and long-period nonlinear baseline drift in artificial source electromagnetic method (CSEM) observation data, the original data is downsampled under the premise of following the Nyquist sampling theorem and not affecting the main frequency amplitude curve to eliminate part of the random interference and improve the speed of subsequent processing; STL decomposition is used to eliminate the baseline drift and large-scale step noise of the observation signal; the variance ratio of the synchronously observed high signal-to-noise ratio reference signal and the observation station signal is used to realize rapid and accurate identification and positioning of noise, and the noisy signal is extracted and post-processed by DBSCAN as a technical means; for the small-scale noise retained after STL processing, low-pass FMD decomposition is used to obtain a low-frequency noise signal, thereby realizing the extraction of the effective signal; finally, the noise-free signal and the processed signal are spliced ​​in the time domain for subsequent processing. The advantage of the present invention is that it uses the time-domain signal variance ratio to achieve rapid and accurate noise identification and positioning for complex and uncontrollable CSEM strong electromagnetic interference. For the first time in the field of electromagnetic exploration data processing, the combination of STL and low-pass FMD decomposition is used to effectively eliminate strong electromagnetic noise and baseline drift, thereby improving the denoising effect and signal-to-noise ratio of the observation signal, and opening up a new research path for CSEM signal-noise separation. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a flow chart of a method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm provided by an embodiment of the present invention;

[0064] Figure 2 The time domain waveforms and main frequency curves of the measured original reference station and the strong interference station provided by the embodiment of the present invention; (a) is the time domain waveform of the measured reference station, (b) is the time domain waveform of the strong interference station, and (c) is the main frequency curve of the measured reference station and the strong interference station;

[0065] Figure 3 : This is a comparison of the time domain waveforms and main frequency curves before and after downsampling processing of a strongly interfering station provided by an embodiment of the present invention; (a) is the time domain waveform before downsampling processing of a strongly interfering station, (b) is the time domain waveform after downsampling processing of a strongly interfering station, and (c) is the main frequency curve before and after downsampling processing of a strongly interfering station;

[0066] Figure 4: This is a comparison of the time domain waveforms and main frequency curves before and after STL decomposition of a strong interference station provided by an embodiment of the present invention; wherein (a) is the time domain waveform of the strong interference station before STL decomposition, (b) is the time domain waveform of the trend component after STL decomposition of the strong interference station, (c) is the time domain waveform of the strong interference station after STL decomposition with the trend component removed, and (d) is the main frequency curve before and after STL decomposition of the strong interference station;

[0067] Figure 5 is the noise identification result of the strong interference station provided by the embodiment of the present invention; wherein (a) is the DBSCAN clustering result, and (b) is the noisy observation data of the strong interference station;

[0068] Figure 6 : This is a comparison of the time domain waveforms and main frequency curves before and after the low-pass FMD decomposition processing of the strong interference measurement station provided by an embodiment of the present invention; wherein (a) is the time domain waveform of the strong interference measurement station before the low-pass FMD decomposition processing, (b) is the time domain waveform of the low-frequency noise signal after the low-pass FMD decomposition processing of the strong interference measurement station, (c) is the time domain waveform after the time domain signals of the strong interference measurement station are spliced, and (d) is the main frequency curve before and after the low-pass FMD decomposition processing of the strong interference measurement station. DETAILED DESCRIPTION

[0069] To make the objectives, technical solutions, and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other implementations obtained by those of ordinary skill in the art without inventive effort are within the scope of protection of the present invention.

[0070] As a novel vibration signal processing and detection method, Feature Mode Decomposition (FMD) leverages the superiority of signal correlation kurtosis and decomposes different modes using a designed adaptive finite impulse response (FIR) filter. This decomposition also considers the impulsiveness and periodicity of fault signals, resulting in more robust decomposition results. The Seasonal-Trend decomposition procedure using Loess (STL) based on local weighted regression is a general and robust time series decomposition method that decomposes time series into a quantitative trend component (baseline drift), a periodic component (seasonal component), and a residual. This effectively eliminates long-term nonlinear baseline drift in time series.

[0071] To address the poor denoising effects of existing artificial source electromagnetic exploration denoising technologies, this present invention applies the STL and FMD decomposition algorithms to artificial source electromagnetic exploration signal processing for the first time. This effectively separates signal from noise in strong interference environments, improving the denoising effect and signal-to-noise ratio of the observed data. The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but is not intended to limit the invention.

[0072] The embodiment of the present invention provides a method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm, using the wide field electromagnetic method (WFEM) as the exploration method, such as Figure 1 As shown, the following steps are included:

[0073] S1. Downsampling of original observation signals: Obtain the original observation signals of the reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station.

[0074] Specifically, the Nyquist sampling theorem is used to downsample the original observation signal, that is, to ensure that f s ≥2f max , and the error between the downsampled main frequency amplitude curve and the original main frequency amplitude curve is less than 1%; where f s is the sampling frequency, f max is the highest frequency of the original observed signal.

[0075] Figure 2 The time domain waveform and main frequency curve of the measured original reference station and the strong interference station (ie, the noisy station) provided by the embodiment of the present invention, the time domain signal of the reference station (see Figure 2 In (a), except for some pulse interference, there is no large-scale step noise and it is relatively stable. Its main frequency amplitude curve (see Figure 2 (c)) is continuous, smooth, and has no distortion points, which meets the basic requirements of the reference signal; while the station signal (see Figure 2 (b) is obviously affected by large-scale step noise such as square wave and triangle wave. The main frequency amplitude curve (see Figure 2 (c) The low frequency band (<32Hz) has severe distortion. Figure 3 The time domain waveform and main frequency curve comparison of the strong interference station before and after downsampling processing provided in the embodiment of the present application show that some random pulse interference is eliminated after downsampling processing (see Figure 3 (b)), and has no significant effect on the main frequency amplitude curve (see Figure 3 (c)).

[0076] S2, STL decomposition: Perform STL decomposition on the first observation signal of the noisy measurement station, and convert the first observation signal X t Indicated as X t =T t +St +R t , where T t represents the trend component (baseline drift), S t represents the periodic component (seasonal component), R t Represents the residual, t represents the time; the trend component in the first observation signal of the noisy measurement station is eliminated to obtain the second observation signal of the noisy measurement station.

[0077] The STL decomposition consists of two parts: inner loop and outer loop, which are implemented using MATLAB functions. The inner loop uses a local weighted regression algorithm to fit the periodic component S t , and then solve the trend component T t ; Obtain the residual R through the outer loop t , and calculate the robust weight in each loop for the next inner loop calculation. The specific steps include:

[0078] S2.1, STL decomposition inner loop: Assume that the trend component and cycle component in the k-th iteration result are represented by T t k 、 The initial value of the trend component is set to 0, that is, T t 0 =0; the decomposition process is as follows:

[0079] S2.1.1. Detrending: Use the first observation signal X of the noisy station at the current moment t Subtract the trend component after the last inner cycle to get the detrended sequence, expressed as X t -T t k ;

[0080] S2.1.2. Periodic subsequence smoothing: Divide the sequence obtained in S2.1.1 into multiple subsequences, perform local weighted regression smoothing on each subsequence, and combine and restore the smoothed subsequences. The resulting periodic subsequences are recorded as

[0081] S2.1.3. Periodic Subsequences Low-pass filtering: For the periodic subsequence obtained in S2.1.2 Perform sliding average processing and then perform local weighted regression algorithm smoothing to obtain sequence;

[0082] S2.1.4. Detrending periodic subsequences: using As a periodic component

[0083] S2.1.5. De-periodicization of the original sequence: as the sequence following the deperiodic component;

[0084] S2.1.6, Trend smoothing: The sequence obtained in S2.1.5 is smoothed by local weighted regression algorithm to obtain the trend component T t k+1 ;

[0085] S2.2, STL decomposition outer loop: According to S2.1.6, the trend component T t k+1 , calculate the residual Furthermore, the robust weight ρ is calculated using the residual t , the calculation formula is as follows:

[0086]

[0087] Where R t Represents the iterative residual of this cycle, that is B(·) represents the bisquare weight function; median represents the median, and || represents the absolute value of the data;

[0088] Get the robust weight ρ t After that, the weight of the local weighted regression algorithm is updated in the inner loop, that is, the weight of the local weighted regression algorithm is multiplied by ρ t , complete the next iteration of the inner loop;

[0089] S2.3. When the iteration result meets the convergence condition, the decomposition is completed and the final trend component T is obtained. t ; The convergence condition is defined as:

[0090]

[0091] in, and Represents the trend component or cycle component of continuous iteration;

[0092] S2.4, the first observation signal X of the noisy station t Eliminate the trend component T t , the second observation signal is recorded as XN t , namely XN t =X t -T t .

[0093] Figure 4 The comparison of the time domain waveform and main frequency curve before and after the STL decomposition of the strong interference station provided in the embodiment of the present application. Figure 4 (a)) is decomposed by STL to obtain its trend components (see Figure 4(b)), further, by subtracting the trend component from the downsampled data, we can obtain a processed signal that eliminates the trend and large-scale step noise (see Figure 4 In (c), the main frequency amplitude curves before and after processing are compared, and it can be seen that the low-frequency data is not significantly improved after eliminating baseline drift and large-scale step noise (see Figure 4 In (d), further processing is required to remove low-frequency interference.

[0094] S3. Noise identification and positioning: The noise is quickly and accurately identified and positioned using the signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy station observed synchronously with a high signal-to-noise ratio, and the noisy observation signal of the noisy station is extracted. This specifically includes the following steps:

[0095] S3.1. The signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy station for high signal-to-noise ratio synchronous observation is calculated by the following formula:

[0096]

[0097] Where, ROV i is the signal variance ratio corresponding to the i-th window, var represents the signal variance, e r represents the first observation signal of the reference station, el represents the second observation signal of the noisy station, (t i ,t i+1 ) represents the data segment corresponding to the i-th window, and Win_Nums is the number of windows;

[0098] S3.2. Use the density-based clustering algorithm (DBSCAN) to perform cluster analysis on the signal variance ratio obtained in S3.1. The discrete points are noise points. The noise is identified and located, and the noisy observation signal x is extracted. t .

[0099] Figure 5 This is the strong interference station noise identification result provided by the embodiment of this application. DBSCAN clustering is performed using the variance ratio of the reference signal and the station signal (see Figure 5 (a)), accurate noise localization can be obtained. In subsequent processing, only the noisy data (see Figure 5 (b)) to preserve the original information to the greatest extent possible.

[0100] S4: Low-pass FDM decomposition of the noisy observation signal at the noisy station: Perform low-pass FDM decomposition on the noisy observation signal to obtain the noise signal. This is implemented using a MATLAB function. The source code is modified and the filter is set to low-pass filtering. The FMD decomposition process is as follows:

[0101] S4.1, parameter initialization: set the initialization FMD parameters, decomposition modulus Mode_Nums, filter length L, maximum number of iterations Max_IterNums, low-pass filter cutoff frequency f c .

[0102] In this embodiment, Mode_Nums=7, filter length L=25, maximum number of iterations Max_IterNums=30, low-pass filter cutoff frequency f c =100Hz.

[0103] S4.2. FIR filter design: Design M Hanning window low-pass filter banks, M = 5 to 10, and initialize the number of iterations j = 1. In this embodiment, M is set to 10.

[0104] S4.3. Noisy observation signal x t Filter decomposition: Low-pass filter designed by S4.2 (cut-off frequency is f c ) for the noisy observation signal x t Perform filtering to obtain the M modal components of the jth iteration Its expression is as follows:

[0105]

[0106] Where, represents the mth modal component in the jth iteration, represents the filter coefficient of the mth low-pass filter in the jth iteration (m=1, 2, ...M), and "*" represents the convolution operation.

[0107] S4.4. Filter coefficient update: using the noisy observation signal x t , modal components and estimated signal period Implement filter coefficient update and set j = j + 1. Specifically include:

[0108] S4.4.1. Signal period estimation: For each modal component Perform Hilbert transform to obtain the analytical signal Calculate analytical signals The time when the autocorrelation spectrum reaches the local maximum after passing through the zero point is the signal iteration period

[0109] S4.4.2. Filter coefficient update: for modal components The relevant kurtosis is maximized and constrained decomposition is performed, where the relevant kurtosis is calculated as follows:

[0110]

[0111] Among them, f m =[f m (1)f m (2)…f m (L)],R XWX =x t T W H x t , R XX =x t T x t , f m represents the filter coefficient vector of the mth filter, f m (L) represents the filter coefficient of the Lth point of the mth filter.

[0112]

[0113] Among them, R XWX represents the weighted correlation matrix; R XX represents the correlation matrix; W H represents the weighting matrix; H represents the shift order of the correlation kurtosis, which is set to 1; N is the signal length; L is the filter length; and T represents the conjugate transpose sign.

[0114] Furthermore, solving the generalized eigenvalue problem R XWX f m =λR XX f m ,λ is its generalized eigenvalue, select the maximum value λ among the generalized eigenvalues max The corresponding eigenvector updates the filter coefficients.

[0115] S4.5, Iteration judgment: judge whether the current iteration has reached the maximum number of iterations; if so, execute the next step S4.6; otherwise, return to S4.3.

[0116] S4.6, Modal component selection: Calculate the correlation coefficients of all modal components and construct the correlation matrix CC K×K , select the two modal components with the largest correlation coefficient and calculate the failure cycle by Calculate their related kurtosis; then select the modal component with larger related kurtosis as the decomposed modal component, and set M = M-1.

[0117] The correlation coefficient between any two modal components is calculated as follows:

[0118]

[0119] Where u p 、u q represents the pth and qth decomposition moduli, represents the mean of the p-th and q-th decomposition moduli, n is the sampling point, and N is the signal length.

[0120] S4.7, FMD decomposition result judgment: judge whether the current mode number reaches the set mode number Mode_Nums. If not, return to step S4.3; otherwise, stop iteration and obtain the final FMD decomposition result; the final low-frequency noise signal can be obtained, recorded as u t .

[0121] S5, noisy signal removal: the noisy observation signal x t Subtract the noise signal u t , get the effective signal after denoising, expressed as Y t =x t -u t .

[0122] S6. Evaluation of time domain signal splicing and processing effect: The second observation signal X t The noisy part in is replaced by the effective signal Y t , obtain the complete time domain signal after denoising. Perform FFT transformation on the processed result to evaluate the continuity of its main frequency amplitude in the frequency domain.

[0123] Figure 6 The comparison results of the time domain waveform and main frequency curve before and after the low-pass FMD decomposition processing of the strong interference station provided in the embodiment of this application are shown. t (See Figure 6 (a) Perform low-pass FMD decomposition to obtain its low-frequency noise u t (See Figure 6 (b)); after removing the low-frequency noise, it is further spliced ​​with the noise-free data to obtain the complete time domain signal after processing (see Figure 6 (c)); and then the main frequency amplitude curves before and after processing (see Figure 6 The signal processed by the present invention eliminates the large-scale step noise of the original observation signal, and the main frequency amplitude curve is continuous and smooth, with significantly improved quality. This shows that the artificial source electromagnetic exploration denoising method provided by the present invention is feasible and has good denoising effect.

[0124] The above-mentioned embodiment proposes a denoising method and system for artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm. For large-scale electromagnetic interference and long-period nonlinear baseline drift in artificial source electromagnetic method (CSEM) observation data, the original data is downsampled while complying with the Nyquist sampling theorem and without affecting the main frequency amplitude curve to eliminate some random interference and improve the speed of subsequent processing. STL decomposition is used to eliminate baseline drift and large-scale step noise in the observation signal. The variance ratio of the synchronously observed high signal-to-noise ratio reference signal and the observation station signal is used to achieve rapid and accurate noise identification and positioning, and DBSCAN is used as a technical means to extract the noisy signal for post-processing. For the small-scale noise retained after STL processing, low-pass FMD decomposition is used to obtain a low-frequency noise signal, thereby realizing the extraction of the effective signal. Finally, the noise-free signal and the processed signal are spliced ​​in the time domain for subsequent processing.

[0125] The embodiment of the present invention further provides an artificial source electromagnetic exploration denoising system based on the STL-FMD decomposition algorithm, comprising the following steps:

[0126] A downsampling module is used to obtain the observation signals of the original reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station;

[0127] An STL decomposition module is used to perform STL decomposition on the first observation signal of the noisy measurement station and eliminate the trend component therein to obtain the second observation signal of the noisy measurement station;

[0128] A noise identification and positioning module is used to realize noise identification and positioning by using the signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy measurement station observed synchronously with a high signal-to-noise ratio, and extract the noisy observation signal of the noisy measurement station;

[0129] The FMD decomposition module is used to perform low-pass FMD decomposition on the noisy observation signal of the noisy observation station to obtain the noise signal;

[0130] The denoising module is used to subtract the noise signal from the noisy observation signal of the noisy observation station to obtain the effective signal after denoising;

[0131] The time domain signal splicing module is used to replace the noisy part of the second observation signal of the noisy measuring station with the valid signal to obtain the complete time domain signal after denoising.

[0132] It should be understood that the functional unit modules in the embodiments of the present invention can be concentrated in one processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated into one unit module, and can be implemented in the form of hardware or software.

[0133] An embodiment of the present invention further provides an electronic terminal, including:

[0134] a memory having a computer program stored thereon;

[0135] The processor is configured to load and execute the computer program to implement the artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm described in the above embodiment.

[0136] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm described in the above embodiment is implemented.

[0137] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0138] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0139] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0142] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm, characterized in that: The steps include: S1: Obtain the observation signals of the original reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station; S2: Perform STL decomposition on the first observation signal of the noisy station and eliminate the trend component to obtain the second observation signal of the noisy station; S3: noise identification and positioning is achieved by using the signal variance ratio of the first observation signal of the reference station with high signal-to-noise ratio synchronous observation and the second observation signal of the noisy measurement station, and the noisy observation signal of the noisy measurement station is extracted; S4: Perform low-pass FMD decomposition on the noisy observation signal of the noisy observation station to obtain the noise signal; S5: Subtract the noise signal from the noisy observation signal of the noisy observation station to obtain the effective signal after denoising; S6: Replace the noisy part of the second observation signal of the noisy measuring station with the valid signal to obtain a complete time domain signal after denoising.

2. The artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to claim 1 is characterized in that: The downsampling process in step S1 includes: The Nyquist sampling theorem is used to downsample the original observation signal, that is, to ensure that f s ≥2f max , and the error between the downsampled main frequency amplitude curve and the original main frequency amplitude curve is less than 1%; where f s is the sampling frequency, f max is the highest frequency of the original observed signal.

3. The artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to claim 1 is characterized in that: Step S2 specifically includes: S2.1, STL decomposes the inner loop, uses the local weighted regression algorithm to fit the periodic component, and then solves the trend component; S2.2, STL decomposes the outer loop, obtains the residual, and calculates the robust weight in each loop, updates the weight of the local weighted regression algorithm, and uses it for the next inner loop calculation; S2.

3. When the iteration result meets the convergence condition, the decomposition is completed and the final trend component T is obtained. t ; The convergence condition is defined as: in, and Represents the trend component or cycle component of continuous iteration; S2.4, the first observation signal X of the noisy station t Eliminate the trend component T t , the second observation signal of the noisy station is recorded as XN t , namely XN t =X t -T t .

4. The artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to claim 3 is characterized in that: STL decomposition inner loop: Assume that the trend component and cycle component in the k-th iteration result are represented by T t k 、 The initial value of the trend component is set to 0, that is, T t 0 =0; The decomposition process is as follows: S2.1.

1. Detrending: Use the first observation signal X of the noisy station at the current moment t Subtract the trend component after the last inner cycle to get the detrended sequence, expressed as X t -T t k ; S2.1.

2. Periodic subsequence smoothing: Divide the sequence obtained in S2.1.1 into multiple subsequences, perform local weighted regression smoothing on each subsequence, and combine and restore the smoothed subsequences. The resulting periodic subsequences are recorded as S2.1.

3. Periodic Subsequences Low-pass filtering: For the periodic subsequence obtained in S2.1.2 Perform sliding average processing and then perform local weighted regression algorithm smoothing to obtain sequence; S2.1.

4. Detrending periodic subsequences: using As a periodic component S2.1.

5. De-periodicization of the original sequence: as the sequence following the deperiodic component; S2.1.

6. Trend smoothing: The sequence obtained in S2.1.5 is smoothed by a local weighted regression algorithm to obtain the trend component. STL decomposition outer loop: trend component T obtained according to S2.1.6 t k+1 , calculate the residual Use the residual to calculate its robust weight ρ t , the calculation formula is as follows: in Where R t Represents the iterative residual of this cycle, that is B(·) represents the bisquare weight function; median represents the median, and || represents the absolute value of the data; Get the robust weight ρ t After that, the weight of the local weighted regression algorithm is updated in the inner loop, that is, the weight of the local weighted regression algorithm is multiplied by ρ t , completing the next iteration of the inner loop.

5. The method for denoising artificial source electromagnetic exploration based on the STL-FMD decomposition algorithm according to claim 1, characterized in that: Step S3 specifically includes: S3.

1. The signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy station for high signal-to-noise ratio synchronous observation is calculated by the following formula: Where, ROV i is the signal variance ratio corresponding to the i-th window, var represents the signal variance, e r represents the first observation signal of the reference station, el represents the second observation signal of the noisy station, (t i ,t i+1 ) represents the data segment corresponding to the i-th window, and Win_Nums is the number of windows; S3.

2. Use density-based clustering algorithm to perform cluster analysis on the signal variance ratio obtained in S3.1 to complete the identification and positioning of noise and extract the noisy observation signal x t .

6. The artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to claim 1 is characterized in that: Step S4 specifically includes: S4.1, parameter initialization: set the initialization FMD parameters, decomposition modulus Mode_Nums, filter length L, maximum number of iterations Max_IterNums, low-pass filter cutoff frequency f c ; S4.2, FIR filter design: Design M Hanning window low-pass filter banks, M = 5-10, and initialize the number of iterations j = 1; S4.

3. Noisy observation signal x t Filter decomposition: The low-pass filter designed by S4.2 is used to decompose the noisy observation signal x t Perform filtering to obtain the M modal components of the jth iteration Its expression is as follows: Where, represents the mth modal component in the jth iteration, represents the filter coefficient of the mth low-pass filter in the jth iteration, and "*" represents the convolution operation; S4.

4. Filter coefficient update: using the noisy observation signal x t , modal components and estimated signal period Implement filter coefficient update and set j=j+1; S4.5, Iteration determination: Determine whether the current iteration has reached the maximum number of iterations; if so, proceed to the next step S4.6; otherwise, return to S4.3; S4.6, Modal component selection: Calculate the correlation coefficients of all modal components and construct the correlation matrix CC K×K , select the two modal components with the largest correlation coefficient and calculate the failure cycle by Calculate their related kurtosis; then select the modal component with larger related kurtosis as the decomposed modal component, and set M = M-1; S4.7, FMD decomposition result judgment: judge whether the current mode number reaches the set mode number Mode_Nums. If not, return to step S4.3; otherwise, stop iteration and obtain the final FMD decomposition result; the final low-frequency noise signal can be obtained, recorded as u t .

7. The artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to claim 6 is characterized in that: The correlation coefficient between any two modal components in step S4.6 is calculated as follows: Where u p 、u q represents the pth and qth decomposition moduli, represents the mean of the p-th and q-th decomposition moduli, n is the sampling point, and N is the signal length.

8. An artificial source electromagnetic exploration denoising system based on the STL-FMD decomposition algorithm, characterized in that: The steps include: A downsampling module is used to obtain the observation signals of the original reference station and the noisy measurement station and perform downsampling processing to obtain the first observation signals of the reference station and the noisy measurement station; An STL decomposition module is used to perform STL decomposition on the first observation signal of the noisy measurement station and eliminate the trend component therein to obtain the second observation signal of the noisy measurement station; A noise identification and positioning module is used to realize noise identification and positioning by using the signal variance ratio of the first observation signal of the reference station and the second observation signal of the noisy measurement station observed synchronously with a high signal-to-noise ratio, and extract the noisy observation signal of the noisy measurement station; The FMD decomposition module is used to perform low-pass FMD decomposition on the noisy observation signal of the noisy observation station to obtain the noise signal; The denoising module is used to subtract the noise signal from the noisy observation signal of the noisy observation station to obtain the effective signal after denoising; The time domain signal splicing module is used to replace the noisy part of the second observation signal of the noisy measuring station with the valid signal to obtain the complete time domain signal after denoising.

9. An electronic terminal, characterized in that: include: a memory having a computer program stored thereon; A processor is used to load and execute the computer program to implement the artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the artificial source electromagnetic exploration denoising method based on the STL-FMD decomposition algorithm according to any one of claims 1 to 7 is implemented.

Images