Method and device for improving seismic profile resolution based on FSRCNN algorithm

Abstract

The present disclosure relates to a method and apparatus for improving the resolution of seismic profiles based on the FSRCNN algorithm, including: acquiring the original seismic profile of the work area and performing preprocessing to obtain the processed seismic profile; using the processed seismic profile as the FSRCNN input data, and performing super-resolution processing on the processed seismic profile through the FSRCNN to obtain the super-resolved seismic profile; performing resampling processing on the super-resolved seismic profile; using the original seismic profile to perform quality control on the resampled seismic profile. If the quality control passes, the resampled seismic profile is output. If the quality control fails, the super-resolution processing of the processed seismic profile is performed again, and sensitive variables are adjusted during the super-resolution processing. This is to solve the problem that when processing 3D seismic data in old areas in the past, due to the small number of vertical sampling points and the large bin size in the horizontal direction of the seismic profile, it is impossible to further improve the resolution of the seismic profile, and the cost of re-acquiring seismic data is too high.

Description

Technical Field

[0001] The present disclosure relates to the technical field of seismic data processing, and particularly to a method and device for improving the resolution of seismic profiles based on the FSRCNN algorithm. Background Art

[0002] In the middle and late stages of oilfield development, the accurate seismic identification of thin reservoirs and small geological targets is of great significance. Higher seismic resolution on the basis of amplitude preservation means that the potential of remaining oil can be more accurately explored. An important index for evaluating the results of development seismic processing is resolution. Seismic resolution includes horizontal resolution and vertical resolution. The vertical apparent resolution is usually one-fourth of the seismic wavelength, and through the inversion technology controlled by well data (pre-stack waveform inversion PWI technology), thin layers with a wavelength of one-eighth or even one-tenth can already be identified. The horizontal resolution determines the distribution range of geological bodies that can be identified in space. Under the same bin situation, the ability to identify horizontal geological bodies is determined by the amplitude preservation of the processing results. However, if the bin size can be reduced at the same horizontal spatial distance, it provides the possibility for the spatial identification of smaller geological bodies and lays a data foundation. Therefore, higher-resolution seismic processing results on the basis of amplitude preservation are the eternal pursuit. However, the cost of re-collecting existing 3D data in old oilfields is high. If the old profiles with a larger bin size and slightly lower resolution can have better resolution capabilities both vertically and horizontally, it is possible to provide strong support for accurate seismic prediction and remaining oil potential tapping in the middle and late stages of old oilfield development. Summary of the Invention

[0003] The present disclosure proposes a method and device for improving the resolution of seismic profiles based on the FSRCNN algorithm to solve the problem that when processing 3D seismic data in old areas in the past, due to the small number of vertical sample points and large bin size in the horizontal direction of the seismic profile, it is impossible to further improve the resolution of the seismic profile, and re-collecting seismic data has too high a cost.

[0004] According to one aspect of the present disclosure, a method for improving the resolution of a seismic profile based on the FSRCNN algorithm is provided, including:

[0005] Obtain the original seismic profile of the work area, and preprocess the original seismic profile to obtain a processed seismic profile;

[0006] Take the processed seismic profile as the FSRCNN input data, and perform super-resolution processing on the processed seismic profile through FSRCNN to obtain a super-resolved seismic profile;

[0007] Perform resampling processing on the super-resolved seismic profile to restore the length and interval of the super-resolved seismic profile to be the same as those of the original seismic profile;

[0008] Using the original seismic profile, perform quality control on the resampled seismic profile. If the quality control passes, output the resampled seismic profile. If the quality control fails, perform super-resolution processing on the processed seismic profile again and adjust the sensitive variables during the super-resolution processing.

[0009] Preferably, the preprocessing includes:

[0010] According to the original seismic profile, use the f-k domain noise suppression method to suppress the coherent noise in the seismic profile, and use the random noise suppression method to remove the low-frequency noise in the seismic profile to obtain a denoised seismic profile.

[0011] Perform spectral analysis on the denoised seismic profile, and select the seismic profile with frequencies above a predetermined decibel in the analysis result as the preprocessed seismic profile.

[0012] Preferably, the method for performing super-resolution processing on the processed seismic profile by FSRCNN includes:

[0013] Set the number of convolution kernels to d, and use d convolution kernels to perform feature extraction processing on the input processed seismic profile to obtain a d-dimensional feature map.

[0014] Set the number of filter kernels to s, and use s filter kernels to perform dimensionality reduction and compression processing on the d-dimensional feature map to obtain an s-dimensional image.

[0015] Set the number of convolutional layers to m, and use m convolutional layers to perform non-linear mapping processing on the s-dimensional image to obtain a super-resolved s-dimensional image.

[0016] Use the d convolution kernels to perform expansion and reconstruction on the super-resolved s-dimensional image to obtain a d-dimensional image.

[0017] Perform deconvolution processing on the d-dimensional image to obtain a super-resolved seismic profile.

[0018] Preferably, the method for performing resampling processing on the super-resolved seismic profile includes:

[0019] According to the original record length and original sampling step of the original seismic profile, use the step length calculation formula to determine the resampling step length.

[0020] According to the original profile width and original trace interval of the original seismic profile, use the trace interval calculation formula to determine the compressed trace interval.

[0021] Use the resampling step length and the compressed trace interval to perform resampling processing on the super-resolved seismic profile to obtain a super-resolved seismic profile with the same record length and trace interval as the original seismic profile.

[0022] Preferably, the step size calculation formula includes:

[0023] Resampling step size = (original record length / length after super-resolution) * original sampling step size;

[0024] The trace interval calculation formula includes:

[0025] Compressed trace interval = (original section width / super-resolution section width) * original trace interval.

[0026] Preferably, the method for quality control of the resampled seismic profile using the original seismic profile includes:

[0027] Compare the high-frequency end frequency on the resampled seismic profile with the high-frequency end frequency at the same position on the original seismic profile;

[0028] Determine whether the high-frequency end frequency on the resampled seismic profile is greater than the high-frequency end frequency at the same position on the original seismic profile. If so, the quality control passes; otherwise, the quality control fails.

[0029] Preferably, the sensitive variables include: the number of convolutional kernels d, the number of filter kernels s, and the number of convolutional layers m.

[0030] Preferably, before resuper-resolving the processed seismic profile, it further includes:

[0031] Preprocess the original seismic data again.

[0032] The present disclosure provides a device for improving the resolution of seismic profiles based on the FSRCNN algorithm, including:

[0033] An acquisition unit for acquiring the original seismic profile of the work area, preprocessing the original seismic profile to obtain a processed seismic profile;

[0034] A super-resolution processing unit for using the processed seismic profile as the FSRCNN input data and performing super-resolution processing on the processed seismic profile through FSRCNN to obtain a super-resolved seismic profile;

[0035] A resampling processing unit for resampling the super-resolved seismic profile to restore the length and trace interval of the super-resolved seismic profile to be consistent with the original seismic profile;

[0036] A quality control unit for using the original seismic profile to perform quality control on the resampled seismic profile. If the quality control passes, the resampled seismic profile is output; if the quality control fails, the processed seismic profile is resuper-resolved, and the sensitive variables are readjusted during super-resolution processing.

[0037] The present invention has at least the following beneficial effects:

[0038] The present disclosure provides a method and device for improving the resolution of seismic profiles based on the FSRCNN algorithm. By taking a seismic profile as input into the FSRCNN network, a super-resolved seismic profile with "larger size and more sample points" is generated. Then, through upsampling vertically and reducing the trace interval horizontally, the size of the super-resolved seismic data volume is restored to the same size as that before super-resolution, achieving the effect of efficiently improving the resolution of seismic profiles using the FSRCNN algorithm. By performing quality control on the processed seismic profile, if it fails the quality control, it is re-super-resolved, and finally, seismic data with truly improved resolution is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings herein are incorporated into the specification and form a part of this specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to explain the technical solutions of the present disclosure.

[0040] Figure 1 FIG. shows a flowchart of a method for improving the resolution of seismic profiles based on the FSRCNN algorithm according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0041] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. Like reference numerals in the drawings denote functionally identical or similar elements. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0042] The term "exemplary" used herein means "serving as an example, embodiment, or illustration". Any embodiment described as "exemplary" herein is not necessarily to be construed as superior or better than other embodiments.

[0043] The term "and / or" herein merely describes an association relationship of associated objects and indicates that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set composed of A, B, and C.

[0044] In addition, for a better illustration of the present disclosure, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present disclosure can also be implemented without some of these specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail so as to highlight the gist of the present disclosure.

[0045] Figure 1 The flowchart shows a method for improving the resolution of seismic profiles based on the FSRCNN algorithm according to an embodiment of the present disclosure. As Figure 1 shown, a method for improving the resolution of seismic profiles based on the FSRCNN algorithm includes: Step S01: Obtain the original seismic profile of the work area, preprocess the original seismic profile to obtain a processed seismic profile; Step S02: Use the processed seismic profile as the FSRCNN input data, and perform super-resolution processing on the processed seismic profile through the FSRCNN to obtain a super-resolved seismic profile; Step S03: Perform resampling processing on the super-resolved seismic profile to restore the trace length and trace interval of the super-resolved seismic profile to be the same as those of the original seismic profile; Step S04: Use the original seismic profile to perform quality control on the resampled seismic profile. If the quality control passes, output the resampled seismic profile. If the quality control fails, re-perform super-resolution processing on the processed seismic profile and adjust the sensitive variables during the super-resolution processing.

[0046] The method for improving the resolution of seismic profiles based on the FSRCNN algorithm provided by the embodiments of the present invention specifically includes the following steps:

[0047] Step S01: Obtain the original seismic profile of the work area, preprocess the original seismic profile to obtain a processed seismic profile.

[0048] In the present disclosure, the preprocessing includes: According to the original seismic profile, use the f-k domain noise suppression method to suppress the coherent noise in the seismic profile, and use the random noise suppression method to remove the low-frequency noise in the seismic profile to obtain a denoised seismic profile; perform spectral analysis on the denoised seismic profile, and select the seismic profile with a frequency above a predetermined decibel in the analysis result as the preprocessed seismic profile.

[0049] In the embodiments of the present disclosure, the method for suppressing the coherent noise in the seismic profile by using the f-k domain noise suppression method includes: Sort the obtained original seismic profile data into the geophone line domain to obtain sorted seismic data; use the sorted seismic data to perform discrete two-dimensional frequency-wavenumber filtering in the main survey line and the connecting line directions respectively by using a discretized two-dimensional time-space filtering function.

[0050]

[0051] In the formula, n and m are the starting and ending values of the seismic data wavenumber in the f-k domain, f is the frequency, Δt is the time sampling interval, Δx is the spatial sampling interval, Δτ is the sampling interval of the filtering factor in time, Δξ represents the sampling interval of the filtering factor in space, and p and q represent the starting and ending values of the frequency of the seismic data in the f-k domain.

[0052] Input the sampling interval of seismic data, the frequency and wavenumber of noise into formula (1), and perform coherent noise suppression through a discretized two-dimensional time-space filtering function to obtain the seismic data after coherent noise suppression.

[0053] In the embodiment of the present disclosure, the method for removing low-frequency noise in a seismic profile by using a random noise suppression method to obtain a denoised seismic profile includes: sorting the obtained seismic data after coherent noise suppression into the shot domain, OVT domain (common offset vector domain), Tau-p domain, and random seismic trace arrangement domain, and using Fourier transform to convert the seismic data sorted into the shot domain, OVT domain, Tau-p domain, and random seismic trace arrangement domain to the frequency domain.

[0054] Among them, the Fourier transform formula is:

[0055]

[0056] In the formula, D(x, y, ω) is the forward Fourier transform of the seismic data d(x, y, t) in the time domain, and ω is the frequency.

[0057] For any spectral component ω in the frequency domain (f-x domain) i , the data D(x, y, ω i ) given by formula (2) is only a function of spatial position, denoted as f(x, y). Design a rectangular box (i.e., a two-dimensional prediction operator). Suppose there are n groups of in-phase axes R i (t) with different velocities in this window, and the time differences of each in-phase axis in the x and y directions are Δt x (i) and Δt y (i), where i = 1, 2,..., n. Then the two-dimensional seismic data f(x, y) of a specific frequency component can be expressed in the form of a two-dimensional Z transform as:

[0058]

[0059] In the formula, F(Z1, Z2) is the Z transform result of the signal f(x, y);

[0060] Use the following formula (4) to obtain F(Z1, Z2) when the Z transform result E(Z1, Z2) of the prediction error is the smallest;

[0061] E(Z1, Z2) = F(Z1, Z2)P(Z1, Z2); (4)

[0062] In the formula, E(Z1, Z2) is the Z transform result of the prediction error, P(Z1, Z2) is the prediction error filter, and F(Z1, Z2) is the input signal.

[0063] In formula (4), the prediction error filter P(Z1, Z2) is actually unknown. Assume that after the three-dimensional seismic data of P(Z1, Z2) is predicted by a rectangular prediction operator, the output prediction error is minimized. A target function is constructed and its partial derivative with respect to the prediction operator is set to 0 to obtain matrix equation (5). According to matrix equation (5), each component column matrix P of the prediction operator is determined;

[0064] R H ·P = R; (5)

[0065] In the formula, R H is the multi-channel autocorrelation Hermite matrix of the f - x domain seismic data, R is the multi-channel autocorrelation column matrix of the f - x domain seismic data, and P is each component column matrix of the prediction operator;

[0066] Substitute each component column matrix P of the prediction operator obtained from formula (5) into formula (4). By adjusting F(Z1, Z2), the Z - transform result E(Z1, Z2) of the prediction error obtained from formula (4) is minimized, thereby obtaining the adjusted F(Z1, Z2).

[0067] Perform an inverse Fourier transform on the obtained adjusted F(Z1, Z2) to obtain the seismic data after random noise suppression in the time domain (f - x domain), that is, the seismic profile after denoising processing.

[0068] In the embodiment of the present disclosure, the predetermined decibel is - 20 db. Perform a spectral analysis on the seismic profile after the denoising processing, and select the seismic profile with a frequency above - 20 db in the analysis result as the processed seismic profile.

[0069] Step S02: Use the processed seismic profile as the FSRCNN input data, and perform super - resolution processing on the processed seismic profile through FSRCNN to obtain the super - resolved seismic profile.

[0070] In the present disclosure, the method for performing super - resolution processing on the processed seismic profile through FSRCNN includes: setting the number of convolution kernels to d, using d convolution kernels to perform feature extraction processing on the input processed seismic profile to obtain a d - dimensional feature map; setting the number of filter kernels to s, using s filter kernels to perform dimensionality reduction and compression processing on the d - dimensional feature map to obtain an s - dimensional image; setting the number of convolutional layers to m, using m convolutional layers to perform non - linear mapping processing on the s - dimensional image to obtain a super - resolved s - dimensional image; using the d convolution kernels to perform expansion and reconstruction on the super - resolved s - dimensional image to obtain a d - dimensional image; performing a deconvolution process on the d - dimensional image to obtain the super - resolved seismic profile.

[0071] In an embodiment of the present disclosure, d convolutional kernels are set in the FSRCNN (image super-resolution network). The processed seismic profile is used as input data, and the d convolutional kernels are used to perform a convolutional operation on the processed seismic profile to extract sub-block information (feature maps) from the seismic profile through feature extraction. This process is the feature extraction process. Among them, the number d of convolutional kernels determines the dimension d of the feature map output by this convolution.

[0072] s filter kernels are set in the FSRCNN. Using the s filter kernels, the d-dimensional feature map obtained through the convolutional operation is subjected to dimensionality reduction and compression processing. The dimension refers to the number of channels of the feature map. The dimensionality reduction and compression processing is mainly to reduce the number of operation parameters in subsequent super-resolution and speed up the operation speed. s 1×1 filter kernels can be used to perform dimensionality reduction processing on the d-dimensional feature map through convolution, that is, from d dimensions to s dimensions, to obtain an s-dimensional image. This process is the dimensionality reduction and compression process. Since s is much smaller than d, a huge number of operation parameters can be reduced after dimensionality reduction.

[0073] m convolutional layers are set in the FSRCNN. Using the m convolutional layers, super-resolution processing is performed on the s-dimensional image obtained after dimensionality reduction and compression to obtain a super-resolved s-dimensional image. This process is the non-linear mapping process. Among them, the m convolutional layers determine the accuracy and complexity of the super-resolution network.

[0074] For the super-resolved s-dimensional image, d convolutional kernels are used for extended reconstruction to restore the s-dimensional image after dimensionality reduction to the original dimension, obtaining an extended reconstructed d-dimensional image.

[0075] Deconvolution processing is performed on the extended reconstructed d-dimensional image, which is the inverse process of the above process. Through deconvolution processing, the input small-sized original seismic profile is restored to the super-resolved large-sized seismic profile, that is, the extended reconstructed d-dimensional (d channels) image is restored to a 1-dimensional (1 channel) image. Deconvolution is different from traditional interpolation methods. The reconstruction results of traditional interpolation methods all use a set of common reconstruction formulas, while the kernel for deconvolution in the present disclosure needs to be learned through FSRCNN, and more accurate results can be obtained for different images.

[0076] Step S03: Resample the super-resolved seismic profile so that the trace length and trace interval of the super-resolved seismic profile are restored to be the same as those of the original seismic profile.

[0077] In the present disclosure, the method for resampling the super-resolved seismic profile includes: determining the resampling step size according to the original recording length of the original seismic profile and the original sampling step size by using a step size calculation formula; determining the compressed trace interval according to the original profile width and the original trace interval of the original seismic profile by using a trace interval calculation formula; and performing resampling processing on the super-resolved seismic profile by using the resampling step size and the compressed trace interval to obtain a super-resolved seismic profile with the same trace length and trace interval as the original seismic profile.

[0078] In the present disclosure, the step size calculation formula includes: resampling step size = (original recording length / super-resolved trace length) × original sampling step size; the trace interval calculation formula includes: compressed trace interval = (original profile width / super-resolved profile width) × original trace interval.

[0079] In an embodiment of the present disclosure, the resampling step size is for resampling the obtained large-size super-resolved seismic profile. After resampling, the seismic propagation time of a single trace on the seismic profile remains unchanged, but the data content increases. For example, the original single-trace data of 4 s has only 50 data points, and the single-trace data of 4 s on the super-resolved and resampled seismic profile has 200 data points, increasing the detailed information.

[0080] After resampling processing, the super-resolved seismic profile is restored to the size with the same vertical trace length and horizontal width as the original seismic profile. For example, if the original seismic profile has a recording length of 5 s and a sampling length of 1 ms, and the super-resolved trace length increases to 10 s, then the vertical resampling length of this step is 0.5 ms, and the trace length after resampling is exactly the same as that of the original seismic profile.

[0081] Step S04: Using the original seismic profile, perform quality control on the resampled seismic profile. If the quality control passes, output the resampled seismic profile; if the quality control fails, re-perform super-resolution processing on the processed seismic profile and adjust the sensitive variables during the super-resolution processing.

[0082] In the present disclosure, the method for performing quality control on the resampled seismic profile by using the original seismic profile includes: comparing the high-frequency end frequency on the resampled seismic profile with the high-frequency end frequency at the same position on the original seismic profile; judging whether the high-frequency end frequency on the resampled seismic profile is greater than the high-frequency end frequency at the same position on the original seismic profile. If so, the quality control passes; otherwise, the quality control fails.

[0083] In the present disclosure, the sensitive variables include: the number of convolution kernels d, the number of filter kernels s, and the number of convolutional layers m.

[0084] In the present disclosure, before the super-resolution processing of the processed seismic profile, it further includes: preprocessing the original seismic data again.

[0085] In an embodiment of the present disclosure, the high-frequency end frequency is the maximum frequency f among the effective frequencies of the seismic profile. max . If the maximum frequency f on the super-resolved seismic profile max is greater than f at the same position of the original seismic profile max , it indicates that the frequency of the super-resolved and reconstructed seismic profile has been improved, useful information has been increased, and the quality control passes.

[0086] If the quality control fails, the super-resolution process of step S02 needs to be repeated, and during the super-resolution process, the values of the number d of convolution kernels of the sensitive variable, the number s of filtering kernels, and the number m of convolutional layers are adjusted. Among them, when adjusting the sensitive variable, it is first necessary to determine whether the super-resolved seismic profile output in step S02 is in an overfitting state or an underfitting state; the overfitting state means that only a small amount of data in the output seismic profile has good effects, and the underfitting state means that the effects of the output seismic profile are all not good; if it is in the overfitting state, when adjusting the sensitive variable, the values of the number d of convolution kernels, the number s of filtering kernels, and the number m of convolutional layers are all reduced. The number d of convolution kernels and the number s of filtering kernels can be reduced by 20 - 30, and the value of the number m of convolutional layers can be reduced by 1 - 2; if it is in the underfitting state, the number d of convolution kernels and the number s of filtering kernels are increased by 20 - 30, and the value of the number m of convolutional layers is increased by 1 - 2. After adjusting the sensitive variable, the super-resolved seismic profile output is resampled and then quality-controlled again. If the quality control fails, step S02 is re-executed to adjust the sensitive variable and iteratively perform super-resolution processing until the quality control passes.

[0087] In an embodiment of the present disclosure, if the output seismic profile still cannot pass the quality control after the sensitive variable is adjusted beyond the limit threshold, step S01 needs to be re-executed to preprocess the original seismic profile again to further improve the signal-to-noise ratio of the seismic profile and reduce noise; the re-preprocessed seismic profile continues to execute the super-resolution processing of step S02, the resampling processing of step S03, and the quality control of step S04.

[0088] It can be understood that the above-mentioned various method embodiments mentioned in the present disclosure can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present disclosure will not elaborate further.

[0089] The execution entity of the method for improving the resolution of seismic profiles based on the FSRCNN algorithm can be a device for improving the resolution of seismic profiles based on the FSRCNN algorithm. For example, the method for improving the resolution of seismic profiles based on the FSRCNN algorithm can be executed by a terminal device, a server, or other processing devices. Among them, the terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the method for improving the resolution of seismic profiles based on the FSRCNN algorithm can be implemented by a processor invoking computer-readable instructions stored in a memory.

[0090] Those skilled in the art can understand that in the above method of the specific implementation manner, the writing order of each step does not mean a strict execution order and constitutes any limitation to the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.

[0091] The present disclosure provides a device for improving the resolution of seismic profiles based on the FSRCNN algorithm, including: an acquisition unit, configured to acquire the original seismic profile of the work area, and perform preprocessing on the original seismic profile to obtain a processed seismic profile; a super-resolution processing unit, configured to use the processed seismic profile as the FSRCNN input data, and perform super-resolution processing on the processed seismic profile through the FSRCNN to obtain a super-resolved seismic profile; a resampling processing unit, configured to perform resampling processing on the super-resolved seismic profile to restore the trace length and trace interval of the super-resolved seismic profile to be consistent with the original seismic profile; a quality control unit, configured to use the original seismic profile to perform quality control on the resampled seismic profile. If the quality control passes, the resampled seismic profile is output. If the quality control fails, the processed seismic profile is re-super-resolved, and the sensitive variables are readjusted during the super-resolution processing.

[0092] In some embodiments, the functions or modules included in the device provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0093] Due to the need for timeliness in the efficient exploration and development of current oilfields, when selecting a super-resolution neural network, methods with too many preprocessing operation layers cannot be chosen, and at the same time, the duration of sample learning and training must be significantly shortened. As an algorithm for improving the resolution of seismic processing profiles, FSRCNN has two obvious advantages compared with conventional CNN and SRCNN networks: First, FSRCNN directly uses seismic profiles with small sizes, few sample points, and low resolution (i.e., original seismic profiles) as inputs, without the need to perform multiple interpolations on low-resolution images like the SRCNN algorithm to obtain seismic profiles (low-resolution profiles) with the same number of sample points and size as the expected output as inputs; Second, FSRCNN uses the deconvolution layer of the fifth layer at the end of the network to perform upsampling, significantly shortening the training time while obtaining the super-resolution result. It is applicable to the high rhythm of oilfield benefit exploration and precise development and the high requirements for high resolution of seismic profiles.

[0094] Based on the concept of the neural network super-resolution algorithm, this disclosure innovatively uses the preprocessed seismic profile with a higher signal-to-noise ratio as a picture containing certain feature vectors as the input of the FSRCNN network. After entering the FSRCNN network, a super-resolved seismic profile with "larger size and more sample points" is generated through machine learning, and then through the method of upsampling vertically and reducing the trace interval horizontally, the size of the super-resolved seismic data volume is restored to the same size as that before super-resolution. The number of seismic traces inside the data volume increases after super-resolution and resampling, and the number of vertical sample points increases, enabling the reconstruction of more details and achieving the effect of efficiently improving the resolution of seismic profiles based on the FSRCNN algorithm. This disclosure also innovatively adds a quality control step, and the seismic profiles processed by FSRCNN are quality controlled by using methods such as spectral analysis comparison and evaluation of the signal-to-noise ratio of the peak frequency. If the quality control is not passed, iterative loop processing is required until high-resolution seismic data that truly improves the resolution and serves oilfield benefit exploration and precise development is obtained.

[0095] The various embodiments of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary skill in the art in the technical field to understand the disclosed embodiments.

Claims

1. A method for improving the resolution of seismic profiles based on the FSRCNN algorithm, characterized in that, Including: Obtain the original seismic profile of the work area, preprocess the original seismic profile to obtain the processed seismic profile; Use the processed seismic profile as the input data of FSRCNN, and perform super-resolution processing on the processed seismic profile through FSRCNN to obtain the super-resolved seismic profile; Perform resampling processing on the super-resolved seismic profile to restore the trace length and trace interval of the super-resolved seismic profile to be the same as those of the original seismic profile; The method of the resampling processing includes: according to the original record trace length and the original sampling step length of the original seismic profile, use the step length calculation formula to determine the resampling step length; according to the original profile width and the original trace interval of the original seismic profile, use the trace interval calculation formula to determine the compressed trace interval; use the resampling step length and the compressed trace interval to perform resampling processing on the super-resolved seismic profile to obtain the super-resolved seismic profile with the same trace length and trace interval as the original seismic profile; the step length calculation formula includes: resampling step length = (original record trace length / super-resolved trace length) * original sampling step length; the trace interval calculation formula includes: compressed trace interval = (original profile width / super-resolved profile width) * original trace interval; Use the original seismic profile to perform quality control on the seismic profile after resampling processing. If the quality control passes, output the seismic profile after resampling processing. If the quality control fails, re-perform super-resolution processing on the processed seismic profile and adjust the sensitive variables during super-resolution processing; The method of the quality control includes: compare the high-frequency end frequency on the seismic profile after resampling processing with the high-frequency end frequency at the same position on the original seismic profile; judge whether the high-frequency end frequency on the seismic profile after resampling processing is greater than the high-frequency end frequency at the same position on the original seismic profile. If it is, the quality control passes; otherwise, the quality control fails; The sensitive variables include: the number of convolution kernels d, the number of filter kernels s, and the number of convolutional layers m; when adjusting the sensitive variables, judge whether the output super-resolved seismic profile is in an overfitting state or an underfitting state. If it is overfitting, reduce the values of the number of convolution kernels d, the number of filter kernels s, and the number of convolutional layers m when adjusting the sensitive variables. If it is underfitting, increase the number of convolution kernels d and the number of filter kernels s; 2. The method for improving the resolution of seismic profiles based on the FSRCNN algorithm according to claim 1, characterized in that, The preprocessing includes: According to the original seismic profile, use the f-k domain noise suppression method to suppress the coherent noise in the seismic profile, and use the random noise suppression method to remove the low-frequency noise in the seismic profile to obtain the denoised seismic profile; Perform spectral analysis on the denoised seismic profile, and select the seismic profile with a frequency above a predetermined decibel in the analysis result as the preprocessed seismic profile; 3. The method for improving the resolution of seismic profiles based on the FSRCNN algorithm according to claim 1, wherein The method of performing super-resolution processing on the processed seismic profile through FSRCNN includes: Set the number of convolution kernels to d, and use d convolution kernels to perform feature extraction processing on the input processed seismic profile to obtain a d-dimensional feature map; Set the number of filter kernels to s, and use s filter kernels to perform dimensionality reduction and compression processing on the d-dimensional feature map to obtain an s-dimensional image; Set the number of convolutional layers to m, and use m convolutional layers to perform non-linear mapping processing on the s-dimensional image to obtain the super-resolved s-dimensional image; Use the d convolutional kernels to perform extended reconstruction on the super-resolved s-dimensional image to obtain a d-dimensional image; Perform deconvolution processing on the d-dimensional image to obtain the super-resolved seismic profile.

4. The method for improving the resolution of seismic profiles based on the FSRCNN algorithm according to claim 1, wherein, Before re-performing super-resolution processing on the processed seismic profile, it also includes: Re-preprocess the original seismic data.

5. An apparatus for improving the resolution of seismic profiles based on the FSRCNN algorithm, characterized in that, It includes: An acquisition unit for acquiring the original seismic profile of the work area, preprocessing the original seismic profile to obtain the processed seismic profile; A super-resolution processing unit for using the processed seismic profile as the FSRCNN input data, and performing super-resolution processing on the processed seismic profile through FSRCNN to obtain the super-resolved seismic profile; A resampling processing unit for performing resampling processing on the super-resolved seismic profile to restore the trace length and trace interval of the super-resolved seismic profile to be the same as those of the original seismic profile; The method of the resampling processing includes: according to the original recording trace length and the original sampling step of the original seismic profile, using the step calculation formula to determine the resampling step; according to the original profile width and the original trace interval of the original seismic profile, using the trace interval calculation formula to determine the compressed trace interval; using the resampling step and the compressed trace interval to perform resampling processing on the super-resolved seismic profile to obtain a super-resolved seismic profile with the same trace length and trace interval as the original seismic profile; the step calculation formula includes: resampling step = (original recording trace length / super-resolved trace length) * original sampling step; the trace interval calculation formula includes: compressed trace interval = (original profile width / super-resolved profile width) * original trace interval; A quality control unit for using the original seismic profile to perform quality control on the resampled seismic profile. If the quality control passes, the resampled seismic profile is output. If the quality control fails, the processed seismic profile is re-super-resolved, and the sensitive variables are re-adjusted during the super-resolution processing; The method of the quality control includes: comparing the high-frequency end frequency on the resampled seismic profile with the high-frequency end frequency at the same position on the original seismic profile; judging whether the high-frequency end frequency on the resampled seismic profile is greater than the high-frequency end frequency at the same position on the original seismic profile. If it is, the quality control passes, otherwise, the quality control fails; The sensitive variables include: the number of convolutional kernels d, the number of filter kernels s, and the number of convolutional layers m; when adjusting the sensitive variables, judge whether the output super-resolved seismic profile is in an overfitting state or an underfitting state. If it is overfitting, when adjusting the sensitive variables, reduce the values of the number of convolutional kernels d, the number of filter kernels s, and the number of convolutional layers m. If it is underfitting, increase the number of convolutional kernels d and the number of filter kernels s.

Images