Earthquake denoising method based on tensor decomposition and LP pseudo norm mixed domain constraint

By introducing mixed domain constraints of LP pseudo-norm and short-time Fourier transform in tensor decomposition, combined with the total anisotropy variation, the problems of large edge error and noise residual caused by CP tensor decomposition are solved, and the seismic recording is improved signal-to-noise ratio and continuity.

CN118938322AActive Publication Date: 2024-11-12CHENGDU UNIVERSITY OF TECHNOLOGY

Patent Information

Application Number
CN202411303482.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-18
Publication Date
2024-11-12
Estimated Expiration
2044-09-18

AI Technical Summary

Technical Problem

After decomposition of CP tensors, seismic records have problems such as large edge error, noise residual and effective information loss in the denoising result.

Method used

Using a method based on tensor decomposition and LP pseudo-normal hybrid domain constraint, the sparse information recorded in frequency domain in LP pseudo-normal and short-time Fourier transform mining is introduced, and boundary information is constructed by introducing the total variation of anisotropy. Mixed domain constraint terms, iterative calculations are performed using alternating direction multipliers to reduce information loss.

Benefits of technology

The signal-to-noise ratio and boundary continuity of earthquake records are improved, noise is effectively reduced, and noise residues and information loss in the denoising results are reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938322A_ABST
    Figure CN118938322A_ABST
Patent Text Reader

Abstract

The invention provides an earthquake denoising method based on tensor decomposition and LP pseudo norm mixed domain constraint, and relates to the field of oil-gas exploration and earthquake data processing, and the method comprises the steps: 1, carrying out the preprocessing of earthquake data, and obtaining an earthquake record S of a time-space domain; 2, extracting non-local related information and boundary information of the time-space domain seismic record by utilizing CP tensor decomposition and anisotropic total variation, and constructing a time-space domain constraint term; 3, mining sparse information of the time-frequency domain seismic record by using an LP pseudo norm and short-time Fourier transform, and constructing a time-frequency domain constraint term; and 4, constructing an objective function by combining the seismic record and the constraint term, and carrying out iterative calculation on the objective function by adopting an alternating direction multiplier method to obtain a seismic record X after mixed domain denoising. According to the invention, random noise in the seismic record can be eliminated, the loss of effective signals is reduced, and the signal-to-noise ratio of the seismic record is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of oil and gas exploration and seismic data processing, and proposes a method based on tensor decomposition and L P Seismic denoising method constrained by pseudo-norm mixed domain. Background Art

[0002] Seismic exploration technology plays a vital role in the exploration and development of oil, natural gas and mineral resources. However, the original seismic data obtained during seismic exploration are often affected by various noises. Therefore, how to effectively reduce the noise in seismic data and improve the signal-to-noise ratio of seismic records has become one of the important research directions in the field of seismic processing.

[0003] There is a problem of large edge errors in seismic records after CP tensor decomposition, and total variation constraints can effectively deal with the problem of large edge errors. Chen Hui et al. used tensor decomposition combined with isotropic total variation to perform time domain denoising on seismic data, and achieved certain results, proving the effectiveness of this method. However, total variation only solves the problem of large errors in seismic records after CP decomposition, resulting in residual noise in the denoising results and loss of effective information. Therefore, a denoising method is needed to overcome the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a method based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint solves the problem that the edge error of seismic records is large due to the use of CP tensor decomposition, and the noise in the denoising result is still residual and the effective information is lost. P Pseudo-norm and short-time Fourier transform are used to mine the sparse information of seismic records in the time-frequency domain to reduce the loss of effective information. Boundary information is introduced through anisotropic total variation to improve the boundary continuity of seismic records.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical means:

[0006] The present invention provides a method based on tensor decomposition and L P The seismic denoising method in pseudo-norm mixed domain includes the following steps:

[0007] Step 1: Preprocess the seismic data and obtain the seismic records S in the time and space domain;

[0008] Step 2: Use CP tensor decomposition and anisotropic total variation to extract non-local related information and boundary information of seismic records in the time and space domains and construct constraints in the time and space domains;

[0009] Step 3: Utilize L PPseudo-norm and short-time Fourier transform are used to mine the sparse information of seismic records in the time-frequency domain and construct constraints in the time-frequency domain;

[0010] Step 4: Construct the objective function by combining the seismic records and the constraints, and use the alternating direction multiplier method to iteratively calculate the objective function to obtain the seismic record X after denoising in the mixed domain.

[0011] In the above scheme, step 1 comprises the following steps:

[0012] Step 1.1: Input the noisy seismic records S in the time and space domain;

[0013] Step 1.2: Construct the fidelity term in the x=argmin objective function based on the seismic records S in the time-space domain, as shown in Formula 1:

[0014]

[0015] Among them, S represents the earthquake record with noise in the time and space domain, X represents the earthquake record after denoising, represents the Frobenius norm, represents the minimization operator;

[0016] In the above scheme, step 2 comprises the following steps:

[0017] Step 2.1: Use a 6×6 matrix to segment the seismic record S to obtain segmented seismic record blocks, [S1, S2, S3…S n ];

[0018] Step 2.2: For each small block, use cosine similarity to calculate the similarity between blocks, and select the first 10,000 blocks in descending order of similarity, and stack them into a three-dimensional tensor X;

[0019] Step 2.3: Use CP decomposition to decompose the tensor X to obtain the sum of multiple rank 1 tensors. If the third-order tensor Its decomposition form is as follows:

[0020]

[0021] in represents the outer product, R represents its CP rank and R>0, in Represents the factor matrix a on the first dimension r space, Represents the factor matrix b on the second dimension r space, Represents the factor matrix c on the third dimension r space;

[0022] Step 2.4: According to the input rank R, select the first R groups to stack and obtain a new tensor as the seismic record S after CP tensor decomposition CP , and construct the CP tensor decomposition constraint term

[0023]

[0024] Where λ1 represents the weight coefficient of the low-rank constraint of CP decomposition;

[0025] Step 2.5: Based on Formula 3, the boundary information of the seismic records in the time and space domain is extracted using the anisotropic total variation method, and the time and space domain constraint terms are constructed, as shown in Formula 4:

[0026]

[0027] Where α represents the weight coefficient of the ATV constraint, and ATV represents anisotropic total variation, and its expression is shown in Formula 5:

[0028] ATV(X)=||D x X||1+||XD y ||1 Formula 5

[0029] Where ||*||1 represents the L1 norm, D x represents the longitudinal difference operator, D y represents the horizontal difference operator, and the definition of the vertical difference operator is shown in Formula 6:

[0030]

[0031] The definition of the lateral difference operator is shown in Formula 7:

[0032]

[0033] In the above scheme, step 3 comprises the following steps:

[0034] Step 3.1: Use the short-time Fourier transform forward solution matrix F to obtain the time-frequency domain seismic record from the seismic record, as shown in Formula 8:

[0035]

[0036] Among them, the short-time Fourier transform positive solution matrix F is shown in formula 9:

[0037]

[0038] in, represents a column vector, τ d Control window length, ω b Control frequency range, represents the Fourier transform factor;

[0039] The simplified short-time Fourier transform positive solution matrix F is shown in formula 10:

[0040]

[0041] Step 3.2: Utilize L P The pseudo-norm mines the sparse information of seismic records in the time-frequency domain, as shown in Formula 11:

[0042]

[0043] Where λ2 is L P The weight coefficient of the pseudo-norm sparsity constraint, Indicates L P Pseudo-norm, its expression is shown in formula 12:

[0044]

[0045] Where a, b, m and n represent the ath row, bth column, total number of rows and total number of columns of the time-frequency seismic record, respectively. Representation Elements The Pth power;

[0046] In the above scheme, step 4 includes the following steps:

[0047] Step 4.1: Construct the objective function by combining the seismic records and constraints as shown in Formula 13:

[0048]

[0049] Step 4.2: Based on Formula 13, introduce the Lagrange multiplier terms R and R x and R y , the dual terms C, C x and C y , we get the seismic denoising objective function based on hybrid domain constraints, as shown in Formula 14:

[0050]

[0051] Where μ and β represent the weight coefficients of the dual terms;

[0052] Step 4.3: Set the weight coefficients λ1, λ2, α, μ and β of the low-rank constraint, sparse constraint, ATV constraint and dual term respectively. P The value of P in the pseudonorm;

[0053] Step 4.4: Set the initial value R of the Lagrange multiplier 1 =0, and its dual term initial value C 1 =0,

[0054] Step 4.5: Update the seismic record X to be denoised in the time domain according to the alternating direction multiplier algorithm and the Sylvester equation algorithm i+1 , the sub-objective function is shown in Formula 15:

[0055]

[0056] The calculation formula is shown in Formula 16:

[0057]

[0058] in Denotes the Kronecker product, matrices A, B, and C are given by Equation 17:

[0059]

[0060] Where E represents the identity matrix, * T represents transpose, E m represents the m-order identity matrix, E n represents the n-order identity matrix, vec(*) represents vectorization;

[0061] Step 4.6: Update the Lagrange multiplier term R according to the alternating direction multiplier algorithm and the soft threshold shrinkage algorithm i+1 , and As shown in formulas 18, 19 and 20:

[0062]

[0063] Where P is a given parameter, the value is less than 1, C x i Represents the dual term C x The data of the i-th iteration, X i represents the data of the i-th iteration of the earthquake record X, C y i Represents the dual term C y The data of the i-th iteration;

[0064] Step 4.7: Update the dual term C according to the alternating direction multiplier algorithm and Fermat's lemma i+1 , and As shown in formulas 21, 22, and 23:

[0065] C i+1 =C i +FX i+1 -R i+1Formula 21

[0066]

[0067] Step 4.8: Determine ||X i+1 -X i ||2 / ||X i ||2>tol, if yes, return to step 4.5 to loop, if no, end the loop.

[0068] Because the present invention adopts the above technical means, it has the following beneficial effects:

[0069] The present invention provides a seismic denoising method based on tensor decomposition and Lp pseudo-norm mixed domain constraint, which mines the sparse information of seismic records in the time-frequency domain through Lp pseudo-norm and short-time Fourier transform, and mines the boundary information of seismic records in the time-space domain through ATV, thereby solving the problem that the use of CP tensor decomposition leads to large edge errors of seismic records, and the denoising result still has residual noise and loss of effective information. The effect of improving the signal-to-noise ratio and continuity of seismic records is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flowchart of the method of the present invention;

[0071] Figure 2 A noise-free seismic profile of the present invention;

[0072] Figure 3 A noisy seismic profile according to the present invention;

[0073] Figure 4 This is a schematic diagram of the denoising result of combining CP tensor decomposition with ATV;

[0074] Figure 5 It is a schematic diagram of the denoising result of the present invention. DETAILED DESCRIPTION

[0075] The following is a detailed description of the embodiments of the present invention. Although the present invention will be described and illustrated in conjunction with some specific embodiments, it should be noted that the present invention is not limited to these embodiments. On the contrary, modifications or equivalent substitutions made to the present invention should all be included in the scope of the claims of the present invention.

[0076] In addition, in order to better illustrate the present invention, numerous specific details are given in the following specific embodiments. It will be understood by those skilled in the art that the present invention can also be implemented without these specific details.

[0077] Combine the following Figure 1-5 The present invention is described in detail.

[0078] The technical problem solved by the present invention is as follows: the problem that the use of CP tensor decomposition leads to large edge errors in seismic records and that noise in denoising results is still residual and effective information is lost is solved, thereby improving the denoising effect.

[0079] Technical means: A method based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint includes the following steps:

[0080] Step 1: Preprocess the seismic data and obtain the seismic records S in the time and space domain;

[0081] Step 2: Use CP tensor decomposition and anisotropic total variation to extract non-local related information and boundary information of seismic records in the time and space domains and construct constraints in the time and space domains;

[0082] Step 3: Utilize L P The pseudo-norm mines the sparse information of seismic records in the time-frequency domain and constructs constraints in the time-frequency domain;

[0083] Step 4: Use the alternating direction multiplier method to iteratively calculate the objective function to obtain the seismic record X after hybrid domain denoising.

[0084] Step 1 includes the following steps:

[0085] Step 1.1: Input the noisy seismic records S in the time and space domain;

[0086] Step 1.2: Construct the fidelity term in the objective function based on the seismic records S in the time and space domain, as shown in Formula 1:

[0087]

[0088] Among them, S represents the earthquake record with noise in the time and space domain, X represents the earthquake record after denoising, represents the Frobenius norm, Indicates the value of X when the objective function is minimized;

[0089] Step 2 includes the following steps:

[0090] Step 2.1: Use a 6×6 matrix to segment the seismic record S to obtain segmented seismic record blocks, [S1, S2, S3…S n ];

[0091] Step 2.2: For each small block, use cosine similarity to calculate the similarity between blocks, and select the first 10,000 blocks in descending order of similarity, and stack them into a three-dimensional tensor X;

[0092] Step 2.3: Use CP decomposition to decompose the tensor X to obtain the sum of multiple rank 1 tensors. If the third-order tensor Its decomposition form is as follows:

[0093]

[0094] in represents the outer product, R represents its CP rank and R>0, ,in Represents the factor matrix a on the first dimension r space, Represents the factor matrix b on the second dimension r space, Represents the factor matrix c on the third dimension r space;

[0095] Step 2.4: According to the input rank R, select the first R groups to stack and obtain a new tensor as the seismic record S after CP tensor decomposition CP , and construct the CP tensor decomposition constraint term

[0096]

[0097] Where λ1 represents the weight coefficient of the low-rank constraint of CP decomposition;

[0098] Step 2.5: Based on Formula 3, the boundary information of the seismic records in the time and space domain is extracted using the anisotropic total variation method, and the time and space domain constraint terms are constructed, as shown in Formula 4:

[0099]

[0100] Where α represents the weight coefficient of the ATV constraint, and ATV represents anisotropic total variation, and its expression is shown in Formula 5:

[0101] ATV(X)=||D x X||1+||XD y ||1 Formula 5

[0102] Where ||*||1 represents the L1 norm, D x represents the longitudinal difference operator, D y represents the horizontal difference operator, and the definition of the vertical difference operator is shown in Formula 6:

[0103]

[0104] The definition of the lateral difference operator is shown in Formula 7:

[0105]

[0106] Step 3 includes the following steps:

[0107] Step 3.1: Use the short-time Fourier transform forward solution matrix F to obtain the time-frequency domain seismic record from the seismic record, as shown in Formula 8:

[0108]

[0109] Among them, the short-time Fourier transform positive solution matrix F is shown in formula 9:

[0110]

[0111] in, represents a column vector, τ d Control window length, ω b Control frequency range, represents the Fourier transform factor;

[0112] The simplified short-time Fourier transform positive solution matrix F is shown in formula 10:

[0113]

[0114] Step 3.2: Using L P The pseudo-norm mines the sparse information of seismic records in the time-frequency domain, as shown in Formula 11:

[0115]

[0116] Where λ2 is L P The weight coefficient of the pseudo-norm sparsity constraint, Indicates L P Pseudo-norm, its expression is shown in formula 12:

[0117]

[0118] Where a, b, m and n represent the ath row, bth column, total number of rows and total number of columns of the time-frequency seismic record, respectively. Representation elements The Pth power;

[0119] Step 4 includes the following steps:

[0120] Step 4.1: Construct the objective function by combining the seismic records and constraints as shown in Formula 13:

[0121]

[0122] Step 4.2: Based on Formula 13, introduce the Lagrange multiplier terms R and R x and R y , the dual terms C, C x and C y, we get the seismic denoising objective function based on hybrid domain constraints, as shown in Formula 14:

[0123]

[0124] Where μ and β represent the weight coefficients of the dual terms;

[0125] Step 4.3: Set the weight coefficients λ1, λ2, α, μ and β of the low-rank constraint, sparse constraint, ATV constraint and dual term respectively. P The value of P in the pseudonorm;

[0126] Step 4.4: Set the initial value R of the Lagrange multiplier 1 =0, and its dual term initial value C 1 =0,

[0127] Step 4.5: Update the seismic record X to be denoised in the time domain according to the alternating direction multiplier algorithm and the Sylvester equation algorithm i+1 , the sub-objective function is shown in Formula 15:

[0128]

[0129] The calculation formula is shown in Formula 16:

[0130]

[0131] Where matrices A, B and C are given by formula 17:

[0132]

[0133] Where E represents the identity matrix, * T represents transpose, E m represents the m-order identity matrix, E n represents the n-order identity matrix, vec(*) represents vectorization;

[0134] Step 4.6: Update the Lagrange multiplier term R according to the alternating direction multiplier algorithm and the soft threshold shrinkage algorithm i+1 , and As shown in formulas 18, 19 and 20:

[0135]

[0136] Where P is a given parameter, the value is less than 1, C x i Represents the dual term C x The data of the i-th iteration, X irepresents the data of the i-th iteration of the earthquake record X, C y i Represents the dual term C y The data of the i-th iteration;

[0137] Step 4.7: Update the dual term C according to the alternating direction multiplier algorithm and Fermat's lemma i+1 , and As shown in formulas 21, 22, and 23:

[0138] C i+1 =C i +FX i+1 -R i+1 Formula 21

[0139]

[0140] Step 4.8: Determine ||X i+1 -X i ||2 / ||X i ||2>tol, if yes, return to step 4.5 to loop, if no, end the loop.

[0141] Technical effect: The present invention adopts CP tensor decomposition with anisotropic total variation constraints and introduces frequency domain information through L pseudo-norm, which solves the problems of large edge errors of seismic records caused by using CP tensor decomposition and residual noise in denoising results and loss of effective information, thereby improving the denoising effect.

[0142] Example 1

[0143] like Figure 1-5 As shown, the mixed domain constrained denoising is as follows:

[0144] Step 1.1: Input the noisy seismic records S in the time and space domain;

[0145] Step 1.2: Construct the fidelity term in the objective function based on the seismic records S in the time and space domain, as shown in Formula 1:

[0146]

[0147] Among them, S represents the earthquake record with noise in the time and space domain, X represents the earthquake record after denoising, represents the Frobenius norm, It represents the value of X when the objective function is minimized.

[0148] Step 2.1: Use a 6×6 matrix to segment the seismic record S to obtain segmented seismic record blocks, [S1, S2, S3…S n ];

[0149] Step 2.2: For each small block, use cosine similarity to calculate the similarity between blocks, and select the first 10,000 blocks in descending order of similarity, and stack them into a three-dimensional tensor X;

[0150] Step 2.3: Use CP decomposition to decompose the tensor X to obtain the sum of multiple rank 1 tensors. If the third-order tensor Its decomposition form is as follows:

[0151]

[0152] in represents the outer product, R represents its CP rank and R>0, ,in Represents the factor matrix a on the first dimension r space, Represents the factor matrix b on the second dimension r space, Represents the factor matrix c on the third dimension r space;

[0153] Step 2.4: According to the input rank R, select the first R groups to stack and obtain a new tensor as the seismic record S after CP tensor decomposition CP , and construct the CP tensor decomposition constraint term

[0154]

[0155] Where λ1 represents the weight coefficient of the low-rank constraint of CP decomposition;

[0156] Step 2.5: Based on Formula 3, the boundary information of the seismic records in the time and space domain is extracted using the anisotropic total variation method, and the time and space domain constraint terms are constructed, as shown in Formula 4:

[0157]

[0158] Where α represents the weight coefficient of the ATV constraint, and ATV represents anisotropic total variation, and its expression is shown in Formula 5:

[0159] ATV(X)=||D x X||1+||XD y ||1 Formula 5

[0160] Where ||*||1 represents the L1 norm, D x represents the longitudinal difference operator, D y represents the horizontal difference operator, and the definition of the vertical difference operator is shown in Formula 6:

[0161]

[0162] The definition of the lateral difference operator is shown in Formula 7:

[0163]

[0164] Step 3.1: Use the short-time Fourier transform forward solution matrix F to obtain the time-frequency domain seismic record from the seismic record, as shown in Formula 8:

[0165]

[0166] Among them, the short-time Fourier transform positive solution matrix F is shown in formula 9:

[0167]

[0168] in, represents a column vector, τ d Control window length, ω b Control frequency range, represents the Fourier transform factor;

[0169] The simplified short-time Fourier transform positive solution matrix F is shown in formula 10:

[0170]

[0171] Step 3.2: Utilize L P The pseudo-norm mines the sparse information of seismic records in the time-frequency domain, as shown in Formula 11:

[0172]

[0173] Where λ2 is L P The weight coefficient of the pseudo-norm sparsity constraint, Indicates L P Pseudo-norm, its expression is shown in formula 12:

[0174]

[0175] Where a, b, m and n represent the ath row, bth column, total number of rows and total number of columns of the time-frequency seismic record, respectively. Representation Elements The Pth power;

[0176] Step 4.1: Construct the objective function by combining the seismic records and constraints as shown in Formula 13:

[0177]

[0178] Step 4.2: Based on Formula 13, introduce the Lagrange multiplier terms R and R x and Ry , the dual terms C, C x and C y , we get the seismic denoising objective function based on hybrid domain constraints, as shown in Formula 14:

[0179]

[0180] Where μ and β represent the weight coefficients of the dual terms;

[0181] Step 4.3: Set the weight coefficients λ1, λ2, α, μ and β of the low-rank constraint, sparse constraint, ATV constraint and dual term respectively, L P The value of P in the pseudonorm;

[0182] Step 4.4: Set the initial value R of the Lagrange multiplier 1 =0, and its dual term initial value C 1 =0,

[0183] Step 4.5: Update the seismic record X to be denoised in the time domain according to the alternating direction multiplier algorithm and the Sylvester equation algorithm i+1 , the sub-objective function is shown in Formula 15:

[0184]

[0185] The calculation formula is shown in Formula 16:

[0186]

[0187] Where matrices A, B and C are given by formula 17:

[0188]

[0189] Where E represents the identity matrix, * T represents transpose, E m represents the m-order identity matrix, E n represents the n-order identity matrix, vec(*) represents vectorization;

[0190] Step 4.6: Update the Lagrange multiplier term R according to the alternating direction multiplier algorithm and the soft threshold shrinkage algorithm i+1 , and As shown in formulas 18, 19 and 20:

[0191]

[0192] Where P is a given parameter, the value is less than 1, C x i represents the dual term C xThe data of the i-th iteration, X i represents the data of the i-th iteration of the earthquake record X, C y i Represents the dual term C y The data of the i-th iteration;

[0193] Step 4.7: Update the dual term C according to the alternating direction multiplier algorithm and Fermat's lemma i+1 , and As shown in formulas 21, 22, and 23:

[0194] C i+1 =C i +FX i+1 -R i+1 Formula 21

[0195]

[0196] Step 4.8: Determine ||X i+1 -X i ||2 / ||X i ||2>tol, if yes, return to step 4.5 to loop, if no, end the loop.

[0197] Effect analysis: Figure 2-5 As shown in the figure, the denoised seismic record profile can more clearly reflect the underground layer and structure than the noisy seismic record profile, which proves the correctness of this method. Figure 2 and Figure 5 , we can find that the denoised earthquake record is very close to the real noise-free earthquake record. Figure 4 and Figure 5 It can be found that this method is more effective in suppressing noise, and effective information is better preserved, which further proves the correctness of the method. The present invention adopts CP tensor decomposition with anisotropic total variation constraints, and introduces frequency domain information through L pseudo-norm, and proposes a seismic denoising method based on mixed domain constraints in combination with the alternating direction multiplier method. The noisy seismic records are denoised by the denoising method mentioned above, and the optimal denoising results are output through repeated iterations. The problem of large edge errors of seismic records caused by the use of CP tensor decomposition and residual noise in the denoising results and loss of effective information is solved, thereby improving the denoising effect.

Claims

1. A method based on tensor decomposition and L P The seismic denoising method in pseudo-norm mixed domain is characterized by: The steps include: Step 1: Preprocess the seismic data and obtain the seismic records S in the time and space domain; Step 2: Use CP tensor decomposition and anisotropic total variation to extract non-local related information and boundary information of seismic records in the time and space domains and construct constraints in the time and space domains; Step 3: Utilize L P Pseudo-norm and short-time Fourier transform are used to mine the sparse information of seismic records in the time-frequency domain and construct constraints in the time-frequency domain; Step 4: Construct the objective function by combining the seismic records and the constraints, and use the alternating direction multiplier method to iteratively calculate the objective function to obtain the seismic record X after denoising in the mixed domain.

2. A method according to claim 1 based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint is characterized by: The step 1 comprises the following steps: Step 1.1: Input the noisy seismic records S in the time and space domain; Step 1.2: Construct the fidelity term in the x=argmin objective function based on the seismic records S in the time and space domain, as shown in Formula 1: Among them, S represents the earthquake record with noise in the time and space domain, X represents the earthquake record after denoising, represents the Frobenius norm, Represents the minimization operator.

3. A method according to claim 2 based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint is characterized by: The step 2 comprises the following steps: Step 2.1: Use a 6×6 matrix to segment the seismic record S to obtain segmented seismic record blocks, [S1, S2, S3…S n ]; Step 2.2: For each small block, use cosine similarity to calculate the similarity between blocks, and select the first 10,000 blocks in descending order of similarity, and stack them into a three-dimensional tensor X; Step 2.3: Use CP decomposition to decompose the tensor X to obtain the sum of multiple rank 1 tensors. If the third-order tensor Its decomposition form is as follows: in represents the outer product, R represents its CP rank and R>0, in Represents the factor matrix a on the first dimension r space, Represents the factor matrix b on the second dimension r space, Represents the factor matrix c on the third dimension r space; Step 2.4: According to the input rank R, select the first R groups to stack and obtain a new tensor as the seismic record S after CP tensor decomposition CP , and construct the CP tensor decomposition constraint term Where λ1 represents the weight coefficient of the low-rank constraint of CP decomposition; Step 2.5: Based on Formula 3, the boundary information of the seismic records in the time and space domain is extracted using the anisotropic total variation method, and the time and space domain constraint terms are constructed, as shown in Formula 4: Where α represents the weight coefficient of the ATV constraint, and ATV represents anisotropic total variation, and its expression is shown in Formula 5: ATV(X)=||D x X||1+||XD y ||1 Formula 5 Where ||*||1 represents the L1 norm, D x represents the longitudinal difference operator, D y represents the horizontal difference operator, and the definition of the vertical difference operator is shown in Formula 6: The definition of the lateral difference operator is shown in Formula 7:

4. A method according to claim 3 based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint is characterized by: The step 3 comprises the following steps: Step 3.1: Use the short-time Fourier transform forward solution matrix F to obtain the time-frequency domain seismic record from the seismic record, as shown in Formula 8: Among them, the short-time Fourier transform positive solution matrix F is shown in formula 9: in, represents a column vector, τ D Control window length, ω b Control frequency range, represents the Fourier transform factor; The simplified short-time Fourier transform positive solution matrix F is shown in formula 10: Step 3.2: Using L P The pseudo-norm mines the sparse information of seismic records in the time-frequency domain, as shown in Formula 11: Where λ2 is L P The weight coefficient of the pseudo-norm sparsity constraint, Indicates L P Pseudo-norm, its expression is shown in formula 12: Where a, b, m and n represent the ath row, bth column, total number of rows and total number of columns of the time-frequency seismic record, respectively. Representation elements The Pth power.

5. A method according to claim 4 based on tensor decomposition and L P The seismic denoising method based on pseudo-norm mixed domain constraint is characterized by: The step 4 comprises the following steps: Step 4.1: Construct the objective function by combining the seismic records and constraints as shown in Formula 13: Step 4.2: Based on Formula 13, introduce the Lagrange multiplier terms R and R x and R y , the dual terms C, C x and C y , we get the seismic denoising objective function based on hybrid domain constraints, as shown in Formula 14: Where μ and β represent the weight coefficients of the dual terms; Step 4.3: Set the weight coefficients λ1, λ2, α, μ and β of the low-rank constraint, sparse constraint, ATV constraint and dual term respectively, L P The value of P in the pseudonorm; Step 4.4: Set the initial value R of the Lagrange multiplier 1 =0, and its dual term initial value C 1 =0, Step 4.5: Update the seismic record X to be denoised in the time domain according to the alternating direction multiplier algorithm and the Sylvester equation algorithm i+1 , the sub-objective function is shown in Formula 15: The calculation formula is shown in Formula 16: in Denotes the Kronecker product, matrices A, B, and C are given by Equation 17: Where E represents the identity matrix, * T represents transpose, E m represents the m-order identity matrix, E n represents the n-order identity matrix, vec(*) represents vectorization; Step 4.6: Update the Lagrange multiplier term R according to the alternating direction multiplier algorithm and the soft threshold shrinkage algorithm i+1 , and As shown in formulas 18, 19 and 20: Where P is a given parameter, the value is less than 1, C x i represents the dual term C x The data of the i-th iteration, X i represents the data of the i-th iteration of the earthquake record X, C y i Represents the dual term C y The data of the i-th iteration; Step 4.7: Update the dual term C according to the alternating direction multiplier algorithm and Fermat's lemma i+1 , and As shown in formulas 21, 22, and 23: C i+1 =C i +FX i+1 -R i+1 Formula 21 Step 4.8: Determine ||X i+1 -X i ||2 / ||X i ||2>tol, if yes, return to step 4.5 to loop, if no, end the loop.

Citation Information

Patent Citations

  • Seismic signal random noise suppression method based on vmd

    CN108919347A

  • Seismic data denoising method combining tensor decomposition and total variation

    CN112596104A

  • Low-rank tensor seismic data denoising method based on variational mode decomposition

    CN113204051A

  • Seismic denoising method based on Lp pseudo norm and gamma norm sparse low-rank constraint

    CN114966861A

  • Three-dimensional tensor random plus footprint noise denoising method based on total variation low-rank approximation

    CN115482168A

Cited By

  • Seismic data denoising method and system based on tensor ring decomposition and unidirectional total variation

    CN120762108A

  • Seismic data denoising method and system based on tensor ring decomposition and one-way total variation

    CN120762108B