A seismic data denoising method based on curvelet transform-joint bilateral filtering
The seismic data is processed through curved wave transformation and combined bilateral filtering algorithms, and the problem of sparse coefficient discontinuity in the prior art is solved, the continuity and signal-to-noise ratio of weak in-phase axis are improved, and more efficient noise suppression and signal protection are achieved.
Patent Information
- Application Number
- CN202210477803.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-05
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-05-05
AI Technical Summary
The existing seismic data denoising method results in discontinuity of sparse coefficients after processing, affecting the fidelity of seismic data and weak in-phase axes blurring, reducing the signal-to-noise ratio.
By using curved wave transform combined with a combined bilateral filtering algorithm, the sparse coefficients of different scales are processed to ensure that the sparse coefficients are continuous at the threshold. The combined bilateral filtering algorithm is introduced to complete the missing in-phase axis information and improve the continuity of weak in-phase axis. .
While effectively removing noise, it protects weak in-phase signal, improving the signal-to-noise ratio and fidelity of seismic data, which is significantly better than traditional methods.
Smart Images

Figure CN114779343B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of signal processing, and in particular to a seismic data denoising method combining a traditional curvelet transform algorithm with a joint bilateral filtering algorithm. Background Art
[0002] With the continuous development of seismic exploration in my country, the areas involved in seismic exploration projects are often complex areas. In seismic data acquisition, due to the less than ideal surface layer excitation conditions in complex areas and the interference of field environmental noise, the resolution and signal-to-noise ratio of the collected seismic data are relatively low, and seismic data interpreters are prone to make wrong judgments, which affects the exploration accuracy of the oil layer location, increases the cost of oil development, and causes huge economic losses. Therefore, seismic data denoising is still one of the difficulties and focuses of seismic data processing and analysis.
[0003] There are many existing methods for seismic denoising, which mainly make full use of the sparse characteristics in the time-frequency transform domain, perform threshold processing on the coefficients in the transform domain, and only retain the sparse coefficients with larger amplitudes, and set the sparse coefficients with smaller amplitudes to 0, thereby achieving the denoising effect. However, after this processing method, the sparse coefficients are discontinuous at the threshold, resulting in the fuzzification of some weak event axes in the seismic data imaging after denoising, which reduces the continuity and affects the fidelity of the seismic data. Summary of the invention
[0004] In order to solve the problems of existing seismic denoising methods, the present invention provides a seismic data denoising method based on curvelet transform-joint bilateral filtering. By improving the curvelet transform threshold denoising method, the effect of protecting weak events while denoising is achieved. In the curvelet domain, a joint bilateral filtering algorithm is introduced to ensure that the sparse coefficients are continuous at the threshold and improve the continuity of weak events.
[0005] The specific technical scheme adopted by the present invention is: first, the noisy seismic data is subjected to a curvelet transform to obtain the original curvelet coefficients. Secondly, the original curvelet coefficients are subjected to a scale analysis, and the first scale curvelet coefficients are used as a coarse-scale signal, which mainly contains the main curvelet coefficients and characterizes the characteristics of the seismic data structure; the last two scale curvelet coefficients are used as fine-scale signals, which usually contain noise signals; the remaining scale curvelet coefficients are used as intermediate-scale signals, which usually contain a large number of weak event axis signals. Then, since the coarse-scale signal contains low-frequency information and the event axis is basically continuous. Therefore, the coarse-scale signal is processed by bilateral filtering to obtain new coarse-scale curve coefficients; the intermediate-scale signal contains a large amount of weak event information, and different thresholds are set for the curve coefficients of the intermediate-scale signal of different scales. The curve coefficients with a maximum of 25% of large amplitudes are screened out and retained through different thresholds of each scale, and the threshold coefficients are obtained and used as the guiding curve coefficients; the intermediate-scale signal curve coefficients are used as the processed curve coefficients, and the joint bilateral filtering algorithm is used; after the joint bilateral filtering, the threshold coefficients and the curve coefficients after the joint bilateral filtering are recombined to obtain new intermediate-scale curve coefficients; the fine-scale signal mainly contains noise signals, and the fine-scale signal is threshold processed to obtain new fine-scale curve coefficients. Finally, the new coarse-scale curve coefficients, the new intermediate-scale curve coefficients and the new fine-scale curve coefficients are recombined, and the de-noised seismic data are obtained by using the inverse curve transform.
[0006] Compared with the existing methods, the present invention introduces a joint bilateral filtering algorithm in the curvelet transform domain. The joint bilateral filtering has a guide graph, combines the intermediate scale curvelet coefficients and their similarity, and completes the missing event information, which can improve the continuity of the weak event. The curvelet transform threshold method can suppress noise well. The present invention protects the weak event signal while effectively removing noise. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Figure 1 is a flow chart of the method of the present invention;
[0008] Figure 2 Spectral plots for each scale analysis of noisy seismic data;
[0009] Figure 3 It is a flowchart of the specific implementation steps of the seismic data denoising method based on curvelet transform-joint bilateral filtering;
[0010] Figure 4 The global profile and its local magnified image are compared with the noise suppression results of actual seismic data;
[0011] Figure 5 The cross-section diagram and its local magnification diagram are the comparison results of different methods for suppressing the noise of actual seismic data;
[0012] Figure 6 Single-channel diagram comparing the noise suppression results of different methods on actual seismic data;
[0013] Figure 7 The following is a spectrum diagram comparing the noise suppression results of actual seismic data using different methods. DETAILED DESCRIPTION
[0014] The present invention discloses a seismic data denoising method based on curvelet transform-joint bilateral filtering, and belongs to the field of seismic data processing. The present invention improves the curvelet transform threshold method, and introduces joint bilateral filtering in the curvelet domain to solve the problem of discontinuity at the threshold of the curvelet transform hard threshold method, thereby achieving seismic data denoising while improving the continuity of weak event axes. The following introduces the specific implementation steps of the present invention, and further describes it in detail in combination with examples and drawings. According to the following description, the content, features, advantages and effects of the present invention will become clearer.
[0015] Figure 1 The method flow chart of the present invention is as follows:
[0016] Step 101: Perform a curvelet transform on the noisy seismic data. The actual collected seismic data contains effective signals and noise, and the composition of the seismic data can be described by the following expression:
[0017] f(t)=o(t)+n(t)
[0018] Where f(t) represents the noisy seismic data, o(t) represents the original seismic data without noise, and n(t) represents the noise. Where x is the spatial domain factor, j is the scale factor, j = 1, 2, ..., r. The total scale number r is determined by the number of seismic traces M and the number of sampling points N of the original seismic data f(t).
[0019]
[0020] in, This is a round-up operation. Fourier transform of With the Fourier window function U j (ω)Satisfaction ω is the frequency domain factor, and all curve functions can be obtained through Translation and rotation are obtained. Introduce the displacement factor sequence k=(k1,k2)∈Z 2 , rotation angle and location Where 0≤θ l <2π, l is the angle factor, and l=0,1,…, is a floor operation, then the curve function It can be expressed as:
[0021]
[0022] In the formula, the rotation matrix
[0023]
[0024] For the original seismic signal f(t), the original curvelet coefficient c is obtained by curvelet forward transformation in the real domain: j,l,k ,
[0025]
[0026] In the formula, is the Fourier transform of f(t).
[0027] Step 102: Perform scale analysis on the original curvelet coefficients. Perform inverse curvelet transform on each scale curvelet coefficient of the seismic data to obtain the spectrum of each scale information. Figure 2 As shown, it can be seen that the first scale only contains low-frequency information, characterizing the characteristics of the seismic data structure, and is classified as a coarse-scale signal; the second scale to the third-to-last scale contain multiple angles, which contain a large number of effective signals and are classified as intermediate-scale signals; the last two scales are mainly composed of high-frequency noise signals, so they are classified as fine-scale signals.
[0028] Based on this analysis, the original curvelet coefficients are divided into coarse-scale signal curvelet coefficients Q j,l,k , the curvelet coefficient S of the intermediate scale signal component j,l,k and the curvelet coefficient T of the fine-scale signal component j,l,k .
[0029]
[0030] Step 103: Perform bilateral filtering on the coarse-scale signal. The coarse-scale signal contains low-frequency information, and the event axis is basically continuous, so there is no need to complete the missing event axis. In this paper, bilateral filtering is used to process the curvature coefficients of the coarse-scale signal. The similarity between the curvature coefficients of the weak event axis is large, and the similarity between the curvature coefficients of the noise signal is small. Bilateral filtering is performed based on the similarity information between the curvature coefficients, retaining the curvature coefficients of the weak event axis and reducing the curvature coefficients of the noise signal. The curvature coefficient Q of the coarse-scale signal is j,l,k Perform bilateral filtering to obtain new curvelet coefficients
[0031]
[0032] in, is the normalized coefficient of the bilateral filtering formula,
[0033]
[0034] In the formula, k q1 , k q2 is the coarse-scale signal curve coefficient Q j,l,k The displacement factor in the frequency domain space, k q =(k q3 ,k q4 ), and k q3 , k q4 for The displacement factor of the curvelet coefficient of the coarse-scale signal in the frequency domain space within the range; Indicated by (k q1 ,k q2 ) is the center, the bilateral filtering range is n1×n1, n1 is the side length of the bilateral filtering and n1=2a1+1, a1=0,1,2…;
[0035] W d1 (k q3 ,k q4 ) is the spatial domain weight of bilateral filtering,
[0036]
[0037] W r1 (k q3 ,k q4 ) is the range weight of bilateral filtering,
[0038]
[0039] In the formula, is the variance of the geometric distance of the bilateral filtering algorithm, is the variance of the bilateral filtering grayscale similarity.
[0040] Step 104: Perform joint bilateral filtering on the intermediate-scale signal. j,l,k Contains a lot of information about the event axis. In order to improve the continuity of the curve coefficient at the threshold and suppress the noise coefficient, different thresholds λ are set for the curve coefficients of the intermediate-scale signals of different scales. j ', filter out up to 25% of the larger amplitude curve coefficients through different thresholds at each scale and retain them to obtain the threshold coefficient and as the guiding curve coefficient,
[0041]
[0042] The intermediate scale signal curve coefficient S j,l,kAs the processed curvelet coefficient, the joint bilateral filtering algorithm is used to complete the missing event information in combination with the similarity of the curvelet coefficient. The curvelet coefficient after joint bilateral filtering is obtained.
[0043]
[0044] in, is the normalized coefficient of the joint bilateral filtering formula,
[0045]
[0046] In the formula, k s1 , k s2 is the intermediate scale signal curve coefficient S j,l,k The displacement factor in the frequency domain space, k s =(k s3 ,k s4 ), k s3 , k s4 for The displacement factor of the curvelet coefficient of the intermediate scale signal within the range in the frequency domain space; Indicated by (k s1 ,k s2 ) is the center, the joint bilateral filtering range of size n2×n2, n2 is the side length of the joint bilateral filtering and n2=2a2+1, a2=0,1,2…; is the spatial domain weight of the joint bilateral filtering,
[0047]
[0048] is the range weight of the joint bilateral filtering,
[0049]
[0050] In the formula, is the variance of the geometric distance of the joint bilateral filtering algorithm, is the variance of the joint bilateral filtering grayscale similarity. and They are (k s1 ,k s2 ) points and (k s3 ,k s4 ) point. After the combined bilateral filtering process, the large curve coefficient (threshold coefficient) and the curvelet coefficients after joint bilateral filtering Reorganize to obtain the denoised intermediate scale signal curve coefficients
[0051]
[0052] Step 105: Perform hard threshold processing on the fine-scale signal. In the fine-scale, most of the signals are noise signals, so only hard threshold processing is required for the fine-scale signal. The hard threshold formula is as follows:
[0053]
[0054] Step 106: Recombining the new coarse-scale signal, the new intermediate-scale signal and the new fine-scale signal to obtain new curvelet coefficients. After denoising the curvelet coefficients of different scales using bilateral filtering, joint bilateral filtering and hard threshold processing, the denoised curvelet coefficients are constructed.
[0055]
[0056] Step 107: Perform an inverse curvelet transform on the new curvelet coefficients to obtain the denoised curvelet coefficients. The inverse curvelet transform formula is as follows:
[0057]
[0058] Finally, the present invention evaluates the denoising result through a spectrum diagram.
[0059] The implementation process of the present invention is demonstrated below in conjunction with specific embodiments. Figure 3 The present invention provides specific implementation steps.
[0060] Step 1: This example uses the actual earthquake data CMP_INL2000_stk_org_gain.segy. Figure 4 The actual seismic data and the seismic data profile after denoising of the present invention are shown in Figure 1. (1-1) and (1-2) are the entire actual seismic data and the actual seismic data in the red box, respectively.
[0061] Step 2: Submit the data into the curvelet transform-joint bilateral filtering algorithm programming.
[0062] Step 3: Divide the seismic data into 3 scale parts.
[0063] Step 4: Use bilateral filtering to process the coarse-scale signal.
[0064] Step 5: First set the threshold coefficient and use it as a guide map, then combine it with the intermediate-scale signal and use joint bilateral filtering for processing.
[0065] Step 6: Use hard thresholding to denoise the fine-scale signal.
[0066] Step 7: Recombine the results of bilateral filtering, joint bilateral filtering and hard threshold processing, and then use inverse curvelet transform to obtain denoised seismic data.
[0067] Step 8: Compare the denoising results of the present invention with those of wavelet transform and curvelet transform. Figure 5 The following are the profiles of seismic data after denoising by various methods. Among them, (1-1) and (1-2) are respectively the whole seismic data after denoising by wavelet transform and the local enlarged images shown in red boxes; (2-1) and (2-2) are respectively the whole seismic data after denoising by curvelet transform and the local enlarged images shown in red boxes; (3-1) and (3-2) are respectively the whole seismic data after denoising by the present invention and the local enlarged images shown in red boxes. Figure 5 It can be seen that after wavelet transform and curvelet transform, weak signals are eliminated and the resolution is reduced. While removing noise, the weak signals are retained and enhanced, especially the deep weak signals are rich in information, providing high-quality seismic data for subsequent modeling and interpretation of deep data. The local magnification diagram and single-channel comparison both demonstrate the effectiveness of the method in this paper. Among them, the single-channel comparison diagram is shown in Figure 6 As shown, (1-1), (1-2), (1-3) and (1-4) are respectively a single-channel graph of actual noisy seismic data, a single-channel graph of seismic data after wavelet transform denoising, a single-channel graph of seismic data after curvelet transform denoising and a single-channel graph of seismic data after denoising of the present invention.
[0068] Step 9: Calculate the frequency spectrum of the entire noisy actual seismic data and the denoised seismic data. Figure 7 The denoised spectra of each method are shown in Figure 2. Figure 7 It can be seen that for the high-frequency part, the three methods can suppress the noise well. For the low-frequency part, the result of wavelet transform contains less energy information, and obviously, most of the event axis signals are suppressed; the result of curvelet transform is better than that of wavelet transform, but compared with the actual data, some weak event axis signals are still suppressed; the result of the method proposed in this paper is obviously better than that of wavelet and curvelet.
[0069] Through the above process, it can be seen that the present invention provides a seismic denoising method based on curvelet transform-joint bilateral filtering, which solves the problem of discontinuity at the threshold of the hard threshold method. While denoising, it improves the continuity of weak event axes and significantly improves the signal-to-noise ratio. Its effect is better than that of traditional methods.
Claims
1. A curvelet transform-bilateral filtering seismic data denoising method, characterized in that: This method is an improved curvelet transform threshold method for denoising. Curvelet transform can handle multi-directional linear variation characteristics. Curvelet coefficient scale analysis shows that the coarse-scale signal contains the main curvelet coefficient, which characterizes the characteristics of seismic data structure. The medium-scale signal contains a large number of weak event signals; the fine-scale signal contains noise signals; Bilateral filtering is a method developed on the basis of Gaussian filtering theory. The filter coefficient design of this method takes into account the spatial domain and range factors, and can effectively protect the weak signal information in the image. The joint bilateral filtering algorithm is an improved bilateral filtering algorithm. Based on the bilateral filtering theory, the algorithm introduces the guidance map of the joint bilateral filtering, combines the similarity of the effective information to complete the missing effective weak signal information, and improves the continuity of the weak event axis. The steps are as follows: First, the noisy seismic data is subjected to curvelet transform to obtain the original curvelet coefficients; the original curvelet coefficients are subjected to scale analysis to obtain coarse-scale signals, intermediate-scale signals and fine-scale signals; the coarse-scale signals are subjected to bilateral filtering to obtain new coarse-scale signals; the intermediate-scale signals are subjected to joint bilateral filtering to obtain new intermediate-scale signals; the fine-scale signals are subjected to hard threshold processing to obtain new fine-scale signals; the new coarse-scale signals, the new intermediate-scale signals and the new fine-scale signals are recombined to obtain new curvelet coefficients; the new curvelet coefficients are subjected to inverse curvelet transform to obtain denoised seismic data; The process of bilateral filtering the coarse-scale signal is as follows: The curvelet coefficient Q of the coarse-scale signal j,l,k Perform bilateral filtering to obtain new curvelet coefficients in, is the normalized coefficient of the bilateral filtering formula, In the formula, kq1, k q2 is the coarse-scale signal curve coefficient Q j,l,k The displacement factor in the frequency domain space is kq = (kq3, kq4), and k q3 , kq4 is The displacement factor of the curvelet coefficient of the coarse-scale signal in the frequency domain space within the range; It represents the bilateral filtering range of n1×n1 centered at (kq1, kq2), where n1 is the side length of the bilateral filtering and n1=2a1+1, a1=0,1,2…; W d1 (k q3 ,k q4 ) is the spatial domain weight of bilateral filtering, W r1 (k q3 ,k q4 ) is the range weight of bilateral filtering, In the formula, is the variance of the geometric distance of the bilateral filtering algorithm, is the variance of the bilateral filtering grayscale similarity; The process of joint bilateral filtering of the intermediate scale signal is: The intermediate scale signal curve coefficient S j,l,k As the processed curvelet coefficient, the joint bilateral filtering algorithm is used to complete the missing event axis information in combination with the similarity of the curvelet coefficient; the curvelet coefficient after joint bilateral filtering is obtained in, is the normalized coefficient of the joint bilateral filtering formula, In the formula, k s1 , ks2 is the intermediate scale signal curve coefficient S j,l,k The displacement factor in the frequency domain space, k s =(k s3 ,k s4 ), k s3 , ks4 is The displacement factor of the curvelet coefficient of the intermediate scale signal within the range in the frequency domain space; Indicated by (k s1 ,k s2 ) is the center, the joint bilateral filtering range of size n2×n2, n2 is the side length of the joint bilateral filtering and n2=2a2+1, a2=0,1,2…; is the spatial domain weight of the joint bilateral filtering, is the range weight of the joint bilateral filtering, In the formula, is the variance of the geometric distance of the joint bilateral filtering algorithm, is the variance of the grayscale similarity of the joint bilateral filtering; and They are (k s1 ,k s2 ) points and (k s3 ,k s4 ) point corresponding to the curvelet coefficient.
2. The method for denoising seismic data by using curvelet transform and joint bilateral filtering according to claim 1, characterized in that: It is necessary to set a threshold curvelet coefficient as a guide map for joint bilateral filtering. In the threshold curvelet coefficient, the threshold is set to no more than 25% of the total number of curvelet coefficients at the current scale.
Citation Information
Patent Citations
Weak seismic signal reconstruction method based on curvelet transform
CN104007469A
3D earthquake data de-noising method based on 2D curvelet transform
CN105676292A
