Time-frequency domain adaptive processing combined with adjacent channel constraint seismic signal denoising method

Abstract

The application discloses a time-frequency domain self-adaptive processing combined with adjacent track constraint seismic signal denoising method, relates to the technical field of seismic signal processing, and comprises the following steps: S1, obtaining noisy seismic data; S2, performing short-time Fourier transform on the noisy seismic data to obtain a complex time-frequency spectrum; S3, extracting a time-frequency amplitude spectrum and phase information; S4, constructing an initial time-frequency mask; S5, constructing a dynamic noise bottom threshold surface to generate a local self-adaptive time-frequency mask; S6, correcting the local self-adaptive time-frequency mask; S7, determining an effective signal protection time-frequency point set; S8, compressing the time-frequency amplitude spectrum; S9, recombining the compressed time-frequency amplitude spectrum and the phase information to obtain a denoised complex time-frequency spectrum; S10, performing inverse short-time Fourier transform on the denoised complex time-frequency spectrum and reconstructing in the time domain to obtain denoised seismic data; and the method can more effectively suppress random noise, local strong interference and non-stationary noise, while retaining more effective reflection signals, phase axis continuity and waveform structure information, and has remarkable advantages.

Description

Technical Field

[0001] This invention relates to the field of seismic signal processing technology, and in particular to a seismic signal denoising method that combines time-frequency domain adaptive processing with adjacent trace constraints. Background Technology

[0002] In seismic data processing, noise suppression is a crucial step in improving the signal-to-noise ratio (SNR) of seismic data, enhancing the continuity of effective reflection phase axes, and improving the quality of subsequent interpretation and imaging. Actual seismic data is typically affected by multiple factors simultaneously, including random noise, local anomalies, low-frequency coherent noise, and environmental vibrations. These factors mask effective reflection signals, distort waveforms, and further reduce the reliability of fault identification, horizon tracking, and structural imaging. Especially in complex surface conditions or low SNR areas, noise often exhibits significant non-stationarity, local abrupt changes, and spatial inhomogeneity, making it difficult for traditional denoising methods to simultaneously achieve effective noise suppression and effective signal protection. Currently, seismic data denoising methods mainly include median filtering, wavelet transform, transform domain predictive filtering, mode decomposition, and time-frequency analysis. Median filtering methods have good suppression capabilities for impulse noise and local outliers, but they tend to over-smooth effective reflection events, blurring the boundaries of in-phase axes. Wavelet transform methods can suppress noise through multi-scale analysis and are widely used in seismic data processing, but their denoising effect largely depends on the selection of wavelet basis functions, decomposition levels, and threshold functions. When the noise has significant non-stationarity and local time-varying characteristics, fixed threshold strategies often fail to achieve stable results. While mode decomposition-based methods can separate signal and noise components to some extent, parameter selection and mode discrimination remain quite sensitive.

[0003] Chinese invention patent application CN109085649A discloses "A Seismic Data Denoising Method Based on Wavelet Transform Optimization." This method addresses the problems of discontinuity at threshold points and bias in reconstructed signals caused by traditional wavelet threshold functions, proposing an improved threshold function to enhance the denoising effect of seismic profiles. While this type of method can improve reconstruction accuracy to some extent, its core remains wavelet domain threshold compression, which is insufficient for complex non-stationary noise and spatial continuity information.

[0004] Chinese invention patent application CN108919347A discloses a "random noise suppression method for seismic signals based on Variational Mode Decomposition (VMD)". This method utilizes variational mode decomposition to decompose seismic signals, thereby improving the ability to suppress random noise and preserve amplitude. This type of method demonstrates the application value of VMD in seismic denoising; however, its performance largely depends on the setting of mode decomposition parameters, and it primarily processes noise from the perspective of mode separation. Its ability to adaptively suppress local time-varying noise and noise at different time-frequency locations remains limited.

[0005] Chinese invention patent application CN106338769A discloses a "method and system for denoising seismic data." This method transforms time-domain seismic data to the frequency domain and uses prediction operators to denoise the frequency-domain data, thereby achieving time-domain data reconstruction. This type of method has certain effectiveness in frequency domain prediction and multichannel data processing, but it still has limitations in handling strong non-stationary noise, local anomaly interference, and preserving local details under complex noise backgrounds.

[0006] Chinese patent CN104502968B proposes a "controllable source seismic data detection method based on threshold multi-level median filtering," which uses threshold and multi-level median filtering to process impulse noise and background noise, demonstrating that median filtering has good anti-outlier capabilities in seismic data processing. However, this type of method is more suitable for specific noise scenarios and has limited adaptability to complex time-frequency variation noise. Chinese invention patent application CN115808717A discloses a "seismic data denoising method and system based on adaptive equalization," which achieves noise attenuation through smoothing and adaptive equalization. This also indicates that existing technologies have begun to pay attention to local differences and adaptive processing issues in seismic data, but the combined utilization of time-frequency domain local noise floor variations and spatial continuity protection is still insufficient.

[0007] As can be seen from the above existing technologies, although existing seismic data denoising methods can suppress noise to a certain extent, they still have the following shortcomings: First, for complex scenarios where random noise, strong local interference, and non-stationary noise coexist, their adaptive ability is insufficient, and they are prone to problems such as large local residual noise or false attenuation of effective signals; Second, they do not make full use of the continuity of effective reflection events in the spatial direction, which may lead to breakage or local distortion of the in-phase axis after denoising; Third, hard thresholding or fixed thresholding methods are prone to causing abrupt changes in time-frequency boundaries, which in turn introduce waveform distortion or artifacts during signal reconstruction. Summary of the Invention

[0008] The purpose of this invention is to design a seismic signal denoising method that combines time-frequency domain adaptive processing with adjacent trace constraints in order to solve the above-mentioned problems.

[0009] The present invention achieves the above objectives through the following technical solutions:

[0010] A seismic signal denoising method combining time-frequency domain adaptive processing and adjacent trace constraints includes:

[0011] S1. Acquire noisy seismic data , represented as: ,in, A noise-free, effective seismic signal. This is a noise signal. For time parameters, Earthquake track number;

[0012] S2. Perform a short-time Fourier transform on the noisy seismic data to obtain the complex-time spectrum. , For indexing time windows, For frequency parameters;

[0013] S3. Extract the time-frequency amplitude spectrum of the complex time spectrum. and phase information ;

[0014] S4. Based on the time-frequency amplitude spectrum, calculate the median amplitude along the channel direction at the same time window and frequency position to obtain the constrained amplitude of the median amplitude of adjacent channels. Construct the initial time-frequency mask ;

[0015] S5. Based on the initial time-frequency mask, the dynamic noise floor is estimated using the local window statistical method to construct the dynamic noise floor threshold surface. And generate a local adaptive time-frequency mask;

[0016] S6. Amplitude constraint through adjacent channel median The local adaptive time-frequency mask is modified to obtain the time-frequency mask to be suppressed.

[0017] S7. Determine the set of effective signal protection time and frequency points that need to be protected;

[0018] S8. Compress the time-frequency amplitude spectrum according to the time-frequency mask to be suppressed and the effective signal region to obtain the compressed time-frequency amplitude spectrum. ;

[0019] S9. Compress the time-frequency amplitude spectrum With phase information Recombining them yields the denoised complex-time spectrum. ;

[0020] S10. The denoised complex-time spectrum is subjected to inverse short-time Fourier transform, and time-domain reconstruction is performed using an overlapping and adding method to obtain the denoised seismic data. .

[0021] The beneficial effects of this invention are as follows: It employs short-time Fourier transform to convert seismic data to the time-frequency domain, utilizing the differences in local energy distribution of seismic signals and noise at different times and frequency locations to construct an adaptive time-frequency mask based on dynamic noise floor estimation. Furthermore, it introduces median constraints between adjacent traces and, combined with coherence protection and first-wave protection, corrects the mask results to better conform to the spatial continuity characteristics of the effective reflection phase axis of the seismic signal. Subsequently, the time-frequency amplitude is smoothly compressed, and inverse short-time Fourier reconstruction is completed while maintaining the original phase essentially unchanged. This method, with similar computational efficiency, can more effectively suppress random noise, strong local interference, and non-stationary noise, while retaining more effective reflection signals, phase axis continuity, and waveform structure information. It reduces the over-smoothing problem easily caused by traditional median filtering, the insufficient adaptability of traditional wavelet transform to complex local noise, and the potential for residual noise or local signal attenuation in complex noise backgrounds using the VMD-WPT method. It exhibits good stability and robustness, and is particularly suitable for noise suppression of seismic data under complex noise conditions, demonstrating significant advantages. Attached Figure Description

[0022] Figure 1 Flowchart of a seismic signal denoising method with time-frequency domain adaptive processing and adjacent trace constraints;

[0023] Figure 2 This is a single-channel raw seismic signal;

[0024] Figure 3 This is a waveform diagram of a noisy seismic signal;

[0025] Figure 4 This is a schematic diagram of the time-frequency distribution of a single-channel raw seismic signal;

[0026] Figure 5 This is a schematic diagram of the time-frequency distribution of a noisy seismic signal;

[0027] Figure 6 This is a global mask image;

[0028] Figure 7 This is a comparison image of local adaptive masks;

[0029] Figure 8 An amplitude diagram of the original seismic record simulating earthquake data;

[0030] Figure 9 An amplitude map of a noisy seismic record used to simulate seismic data;

[0031] Figure 10 This is a graph showing the denoising result of simulated earthquake data using median filtering;

[0032] Figure 11This is a diagram showing the denoising result of simulated earthquake data using wavelet transform;

[0033] Figure 12 This is a denoising result of simulated seismic data using the VMD-WPT method;

[0034] Figure 13 This is a denoising result diagram of the seismic signal denoising method of the present invention, which uses time-frequency domain adaptive processing combined with adjacent trace constraints on simulated seismic data;

[0035] Figure 14 An amplitude diagram of the original seismic record from actual earthquake data;

[0036] Figure 15 This is a denoising result of actual earthquake data using median filtering;

[0037] Figure 16 This is a denoising result of actual earthquake data using wavelet transform;

[0038] Figure 17 This is a denoising result of actual seismic data using the VMD-WPT method;

[0039] Figure 18 This is a denoising result diagram of actual seismic data using the seismic signal denoising method of the present invention, which combines time-frequency domain adaptive processing with adjacent trace constraints. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0041] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0042] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0043] In the description of this invention, it should be understood that the terms "upper," "lower," "inner," "outer," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used to facilitate the description of this invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0044] Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0045] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, terms such as "set" and "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0046] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0047] A seismic signal denoising method combining time-frequency domain adaptive processing and adjacent trace constraints includes:

[0048] S1. Acquire noisy seismic data , represented as: ,in, A noise-free, effective seismic signal. This is a noise signal. For time parameters, This refers to the seismic trace number. Noise includes one or more of the following: random noise, localized strong interference, anomalously high-energy noise, and coherent interference. In practical applications, the processing range can be selected based on the number of sampling points and traces in the seismic record. For example, in the code implementation, the time sampling range is 1:1000, and the trace range is 1:240. Mean reduction processing is performed on each trace to reduce the impact of the DC component and overall offset on subsequent time-frequency analysis results.

[0049] S2. Perform a short-time Fourier transform on the noisy seismic data to obtain the complex-time spectrum. , represented as: , For indexing time windows, For frequency parameters; The window function is preferred, ideally the Hanning window. The unit is the imaginary unit. In actual implementation, the window length can be set to 32, and the number of Fourier transform points can be set to 32 or an integer power of 2 corresponding to the window length, to ensure a balance between time-frequency resolution and computational efficiency. This method uses short-time Fourier transform to convert the seismic signal to the time-frequency domain, making the local energy distribution of signal and noise at different times and frequency locations clearer. Figure 3 The time-frequency distribution diagrams of the single-channel raw signal and the noisy signal are given. It can be seen that the noisy signal has obvious local high-energy noise regions and background energy rise regions in the time-frequency plane.

[0050] S3. Extract the time-frequency amplitude spectrum of the complex time spectrum. and phase information , represented as: , The amplitude spectrum is used to construct the time-frequency mask and estimate the noise floor, while the phase spectrum remains unchanged in the subsequent reconstruction in order to preserve the waveform structure characteristics of the original seismic signal as much as possible.

[0051] S4. Construct the initial time-frequency mask based on the time-frequency amplitude spectrum. Specifically, it includes:

[0052] S401. Perform median statistics on time-frequency amplitudes along the channel direction to obtain the median constrained amplitude of adjacent channels. This enhances the robustness of time-frequency domain background energy estimation, as expressed as: , The median of the amplitude across all channels at the same time window and frequency location is used. Since effective reflection events usually have strong continuity between adjacent channels, while local random noise and isolated outliers often lack this consistency, using the cross-channel median can reduce the impact of outliers on background estimation; the adjacent channel median constrains the amplitude. The purpose is to ensure that when seismic reflection events have good continuity between traces, the median result is close to the original amplitude, so it will not weaken the effective events too much; however, for random noise and isolated high-energy anomalies, they lack continuity in the trace direction, so they will be significantly suppressed after median constraint.

[0053] S402, constrain amplitude based on adjacent channel median values Analyze global median threshold The global median threshold is analyzed using either the first or second analysis method. ;

[0054] The first analysis method is to directly constrain the amplitude using the median values ​​of adjacent channels. Analyze global median threshold , represented as: ;

[0055] The second analysis method involves constraining the amplitude of the median values ​​of adjacent channels along the time direction. Median smoothing is performed to obtain the smoothed median constraint amplitude of adjacent channels. And constrain the amplitude based on the smoothed median of adjacent channels. Analyze global median threshold , is represented as: is represented as: , ; The adjustment coefficient is used to smooth the median constraint amplitude of adjacent channels along the time direction, resulting in the smoothed median constraint amplitude of adjacent channels. To reduce the impact of local spikes on the dynamic noise floor estimation, the code uses a time sliding median window of length 5 to smooth the cross-channel median results.

[0056] S403, Based on the global median threshold Constructing the initial time-frequency mask , represented as: .

[0057] S5. Based on the initial time-frequency mask, the dynamic noise floor is estimated using the local window statistical method to construct the dynamic noise floor threshold surface. And generate a local adaptive time-frequency mask; the local adaptive time-frequency mask is used to mark candidate time-frequency points to be suppressed, specifically including:

[0058] S501, using the time frequency of the nth earthquake. Select a local neighborhood around the center Analyze the local background median and local median absolute deviation , represented as: , , Used to characterize local background energy estimation Used to characterize the estimation of local discreteness, p∈m, q∈f;

[0059] S502, Based on the local background median and local median absolute deviation Constructing a dynamic noise bottom threshold surface , represented as: , The adjustment coefficient is obtained through the dynamic noise bottom threshold surface. This allows the mask discrimination criteria on different times, frequencies, and seismic traces to change with local background noise, thereby improving the adaptability to non-stationary noise.

[0060] S503, Based on the dynamic noise bottom threshold surface Generate a local adaptive time-frequency mask , represented as: When the amplitude at a certain time-frequency point is greater than the dynamic noise floor multiplied by a threshold multiple, it indicates that the location is more likely to belong to an abnormally high-energy region and requires further suppression; otherwise, the original state is maintained. It should be noted that the mask here does not directly and simply distinguish between "signal" and "noise," but is used to mark time-frequency regions that need further suppression or correction.

[0061] S6. Correct the local adaptive time-frequency mask by means of adjacent channel median constraints, and constrain the amplitude by means of adjacent channel median constraints. Constraints are applied to modify the local adaptive time-frequency mask to obtain the time-frequency mask to be suppressed, thereby reducing the influence of abnormal high-energy points and local isolated noise points on the masking results. The amplitude is constrained by the median of adjacent channels. This can enhance spatial continuity during the denoising process and reduce the impact of isolated outliers on the reconstruction results. In the program implementation, the half-width of the channel window can be set to 15, meaning the total window length is... Specifically, the amplitude is constrained based on the median values ​​of adjacent channels. The candidate time-frequency points to be suppressed in the local adaptive time-frequency model are subjected to secondary discrimination, and effective time-frequency points with continuity between adjacent channels are eliminated to obtain the amplitude constrained by the median of adjacent channels. Constraint-corrected time-frequency mask to be suppressed;

[0062] S7. Determine the effective signal area that needs protection; specifically including:

[0063] S701, Using adjacent channel median values ​​to constrain amplitude Construct a locally coherent reference , The coherence gating factor is preferably 1.6;

[0064] S702, Local coherent reference By comparing with the original amplitude, a coherent gated mask is obtained. , represented as: The coherent gating mechanism further prevents effective reflection in-phase axes from being misjudged as areas requiring strong suppression. When coherent gating is enabled, only time-frequency points that simultaneously meet the conditions of having an amplitude exceeding the dynamic noise floor threshold and being significantly higher than the local median reference are marked.

[0065] S703, Analyze the protection cutoff time for the nth earthquake. The first-wave protection mechanism does not participate in masking suppression for time-frequency points earlier than the protection deadline, thereby preventing the effective energy of the first wave and its vicinity from being mistakenly weakened during the denoising process; specifically:

[0066] (1) Extract the time series of the noisy seismic data of the nth channel, and subtract its own median from the time series to obtain the preprocessed signal. Calculate the envelope sequence of the preprocessed signal.

[0067] (2) Within the preset early-stage noise window, calculate the median absolute deviation of the envelope sequence, and estimate the early-stage noise level of the nth earthquake based on the median absolute deviation of the envelope sequence.

[0068] (3) Determine the first wave detection threshold based on the noise level in the early period, compare the envelope sequence with the first wave detection threshold, and when the envelope value continuously exceeds the first wave detection threshold, take the moment when it first exceeds the first wave detection threshold as the arrival time of the first wave. ;

[0069] (4) Based on the arrival time of the first wave Determine the protection deadline , represented as: ,in, To protect the extension duration;

[0070] S704, coherent gate mask The marked time and frequency points and protection cutoff times The previous time and frequency points are used as the effective signal protection time and frequency point set.

[0071] S8. Based on the corrected time-frequency mask and the set of effective signal-protected time-frequency points, compress the time-frequency amplitude spectrum to obtain the compressed time-frequency amplitude spectrum. Specifically:

[0072] The time-frequency amplitude spectrum is compressed using a soft mask to obtain the compressed time-frequency amplitude spectrum. , represented as: ,in, Here are the soft mask weighting coefficients, and For the time-frequency points marked by the suppression time-frequency mask and the time-frequency points of the effective signal protection time-frequency point set, the original amplitude remains unchanged. When performing compression processing, the hard threshold method of directly setting to zero is not used. Instead, the soft mask method is used to smoothly compress the amplitude to reduce the risk of edge abrupt changes and waveform distortion. Moreover, the reconstruction result using the soft mask method is smoother.

[0073] Alternatively, the amplitudes of the time-frequency points marked by the time-frequency mask to be suppressed and the time-frequency points outside the effective signal protection time-frequency point set can be replaced with the median constraint amplitudes of adjacent channels. The original amplitude remains unchanged for the time-frequency points marked by the suppression time-frequency mask and the time-frequency points of the effective signal protection time-frequency point set.

[0074] S9. Compress the time-frequency amplitude spectrum With phase information Recombining them yields the denoised complex-time spectrum. , represented as: Since the phase information remains essentially unchanged, this method can effectively preserve the waveform structure characteristics of seismic signals while suppressing noise.

[0075] S10. The denoised complex-time spectrum is subjected to inverse short-time Fourier transform, and time-domain reconstruction is performed using an overlapping and adding method to obtain the denoised seismic data. , represented as: The program implementation uses window functions, window lengths, overlap lengths, and zero-padding parameters corresponding to the short-time Fourier transform to ensure that the signal can be stably reconstructed.

[0076] The denoised seismic data output by this method Compared with the comparison method:

[0077] The denoised seismic data obtained by this method The output can also calculate the difference between the original data and the denoised result: ;

[0078] And further calculate the pseudo signal-to-noise ratio index. : Used to evaluate the noise reduction effect of the method of the present invention.

[0079] Coherent protection and first-wave protection mechanisms are introduced to protect the effective signal area. The coherent protection mechanism is used to reduce the masking degree for reflection events with strong inter-channel continuity, while the first-wave protection mechanism is used to preserve the time and frequency points within a preset time range of the arrival time of the first wave, thereby preventing the effective energy of the first wave and the early period from being erroneously weakened during noise suppression.

[0080] like Figure 5 and Figure 6As shown, this method addresses the complex situation of coexisting random noise, strong local interference, and non-stationary noise. Analysis and calculation results show that while traditional median filtering can reduce random noise to some extent, it can easily cause excessive smoothing of effective reflection events. Wavelet transform has a certain effect in suppressing high-frequency noise, but its adaptability to complex non-stationary noise is insufficient. The VMD-WPT method can improve some noise suppression effects, but residual noise or attenuation of effective signals may still exist in complex local areas. This invention, by constructing a multi-layered processing flow including dynamic noise floor estimation, adaptive time-frequency masking, adjacent trace median constraints, coherence gating, and first-wave protection, makes the noise suppression process more consistent with the time-frequency distribution and spatial continuity characteristics of seismic signals. Therefore, it performs better in preserving the continuity of effective reflection phase axes, reducing background noise, and mitigating local anomalous interference, exhibiting better stability and robustness, and is particularly suitable for noise suppression of seismic data under complex noise backgrounds.

[0081] This method uses short-time Fourier transform to convert seismic data to the time-frequency domain. It utilizes the differences in local energy distribution of seismic signals and noise at different times and frequencies to construct an adaptive time-frequency mask based on dynamic noise floor estimation. Furthermore, it introduces adjacent trace median constraints and combines coherence protection and first-wave protection to correct the masking results, making them more consistent with the spatial continuity of the effective reflection phase axis of the seismic data. Subsequently, a soft mask is used to smooth and compress the time-frequency amplitude, and inverse short-time Fourier reconstruction is completed while maintaining the original phase essentially unchanged. This method, with similar computational efficiency, can more effectively suppress random noise, strong local interference, and non-stationary noise, while retaining more effective reflection signals, phase axis continuity, and waveform structure information. It reduces the over-smoothing problem easily caused by traditional median filtering, the insufficient adaptability of traditional wavelet transform to complex local noise, and the potential for residual noise or local signal attenuation in complex noise backgrounds by the VMD-WPT method. It exhibits good stability and robustness, and is particularly suitable for noise suppression of seismic data under complex noise conditions, demonstrating significant advantages.

[0082] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A seismic signal denoising method combining time-frequency domain adaptive processing and adjacent trace constraints, characterized in that, include: S1. Acquire noisy seismic data , represented as: ,in, A noise-free, effective seismic signal. This is a noise signal. For time parameters, Earthquake track number; S2. Perform a short-time Fourier transform on the noisy seismic data to obtain the complex-time spectrum. , For indexing time windows, For frequency parameters; S3. Extract the time-frequency amplitude spectrum of the complex time spectrum. and phase information ; S4. Based on the time-frequency amplitude spectrum, calculate the median amplitude along the channel direction at the same time window and frequency position to obtain the constrained amplitude of the median amplitude of adjacent channels. Construct the initial time-frequency mask ; S5. Based on the initial time-frequency mask, the dynamic noise floor is estimated using the local window statistical method to construct the dynamic noise floor threshold surface. And generate a local adaptive time-frequency mask; S6. Amplitude constraint through adjacent channel median The local adaptive time-frequency mask is modified to obtain the time-frequency mask to be suppressed. S7. Determine the set of effective signal protection time and frequency points that need to be protected; S8. Compress the time-frequency amplitude spectrum based on the time-frequency mask to be suppressed and the set of effective signal protection time-frequency points to obtain the compressed time-frequency amplitude spectrum. ; S9. Compress the time-frequency amplitude spectrum With phase information Recombining them yields the denoised complex-time spectrum. ; S10. The denoised complex-time spectrum is subjected to inverse short-time Fourier transform, and time-domain reconstruction is performed using an overlapping and adding method to obtain the denoised seismic data. .

2. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 1, characterized in that, S4 includes: S401. Perform median statistics on time-frequency amplitudes along the channel direction to obtain the median constrained amplitude of adjacent channels. , represented as: ; S402, constrain amplitude based on adjacent channel median values Analyze global median threshold ; S403, Based on the global median threshold Constructing the initial time-frequency mask , represented as: .

3. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 2, characterized in that, In S402, the amplitude of the median value of adjacent channels is constrained along the time direction. Median smoothing is performed to obtain the smoothed median constraint amplitude of adjacent channels. And constrain the amplitude based on the smoothed median of adjacent channels. Analyze global median threshold , represented as: , , This is the adjustment coefficient.

4. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 1, characterized in that, S5 includes: S501, using the time frequency of the nth earthquake. Select a local neighborhood around the center Analyze the local background median and local median absolute deviation ; S502, Based on the local background median and local median absolute deviation Constructing a dynamic noise bottom threshold surface ; S503, Based on the initial time-frequency mask Utilizing dynamic noise low threshold surface Generate a local adaptive time-frequency mask , represented as: .

5. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 1, characterized in that, S7 includes: S701, Using adjacent channel median values ​​to constrain amplitude Construct a locally coherent reference , This is the coherent gating factor; S702, Local coherent reference By comparing with the original amplitude, a coherent gated mask is obtained. , represented as: ; S703, Analyze the protection cutoff time for the nth earthquake. ; S704, coherent gate mask The marked time and frequency points and protection cutoff times The previous time and frequency points are used as the effective signal protection time and frequency point set.

6. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 5, characterized in that, S703 includes: (1) Extract the time series of the noisy seismic data of the nth channel, and subtract its own median from the time series to obtain the preprocessed signal. Calculate the envelope sequence of the preprocessed signal. (2) Within the preset early-stage noise window, calculate the median absolute deviation of the envelope sequence, and estimate the early-stage noise level of the nth earthquake based on the median absolute deviation of the envelope sequence. (3) Determine the first wave detection threshold based on the noise level in the early period, compare the envelope sequence with the first wave detection threshold, and when the envelope value continuously exceeds the first wave detection threshold, take the moment when it first exceeds the first wave detection threshold as the arrival time of the first wave. ; (4) Based on the arrival time of the first wave Determine the protection deadline , represented as: ,in, To protect the extension duration.

7. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 5, characterized in that, Coherent gating multiple The value is 1.

6.

8. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 1, characterized in that, In S9, soft mask compression is performed on the time-frequency points marked by the time-frequency mask to be suppressed and the time-frequency points outside the effective signal protection time-frequency point set to obtain the compressed time-frequency amplitude spectrum. , represented as: ,in, Here are the soft mask weighting coefficients, and The original amplitude remains unchanged for the time-frequency points marked by the suppression time-frequency mask and the time-frequency points of the effective signal protection time-frequency point set.

9. The seismic signal denoising method based on time-frequency domain adaptive processing and adjacent trace constraints according to claim 1, characterized in that, In S9, the amplitudes of the time-frequency points marked by the time-frequency mask to be suppressed and the time-frequency points outside the effective signal protection time-frequency point set are replaced with the median constraint amplitudes of adjacent channels. The original amplitude remains unchanged for the time-frequency points marked by the suppression time-frequency mask and the time-frequency points of the effective signal protection time-frequency point set.

Images