Method and device for improving seismic data resolution based on wavelet domain residual network

Abstract

The invention relates to the technical field of oil field seismic data processing, in particular to a method and device for improving seismic data resolution based on a wavelet domain residual network. The method comprises the steps that acquired data containing existing seismic data samples are input into a U-Net network for training, and the U-Net network comprises a residual module, a global context module and wavelet transformation; determining a total loss function of the trained U-Net network model; and according to the total loss function, adjusting and optimizing the U-Net network model to obtain an optimal model, and performing resolution improvement processing on the to-be-processed seismic data by using the optimal model to obtain a final processing result. The problems that in an existing seismic data resolution improving method, wavelet extraction difficulty is large, stratum quality factors are uncertain, well information depends on, parameters are complex and the like, component information and global context information in sample data and label data are not comprehensively utilized, and consequently the seismic data resolution cannot be further improved are solved.

Description

Technical Field

[0001] The invention relates to the technical field of oilfield seismic data processing, and in particular to a method and a device for improving seismic data resolution based on a wavelet domain residual network. Background Art

[0002] Improving the resolution of seismic data can improve the accuracy of oil and gas exploration and reduce exploration costs. Since the shorter the wavelength of the seismic wavelet, the higher the resolution, the traditional method is to improve the resolution by compressing the wavelet. In practical applications, wavelet extraction is difficult, so scholars have proposed methods based on time-frequency analysis. In order to shorten the calculation time, generalized S transform, Shearlet transform, Chrip-Z transform and other methods have been proposed to process seismic data. Inverse Q filtering and deconvolution have always been important means to improve the resolution of seismic data. They can effectively correct the attenuation and dispersion of seismic waves and obtain high-resolution seismic data. However, the deconvolution and Q compensation methods are limited by the assumptions that the wavelet is the minimum phase and the reflection coefficient is white noise, and complex parameters need to be obtained, which is not convenient in practical applications.

[0003] In addition, traditional methods for improving seismic data resolution include translational Gaussian window decomposition, well-seismic joint stacking, adaptive time sampling, matching pursuit, etc. These traditional methods also have problems such as difficulty in wavelet extraction, uncertainty in formation quality factors, dependence on well information, and complex parameters.

[0004] With the widespread application of neural networks, methods for improving the resolution of seismic data have also been developed. In recent years, convolutional neural networks have developed rapidly. They can better extract image features and learn the nonlinear relationship between input and target, making it possible to use deep neural networks to improve seismic resolution. Deep learning methods use data-driven methods to adaptively characterize the relationship between input and target and have good autonomous learning capabilities. However, the current methods for improving seismic data resolution based on deep learning do not fully utilize the component information and global context information in sample data and label data. Summary of the invention

[0005] The present invention proposes a method and device for improving the resolution of seismic data based on a wavelet domain residual network, so as to solve the problems in existing methods for improving the resolution of seismic data, such as the difficulty in wavelet extraction, uncertainty in formation quality factors, dependence on well information, and complex parameters, and the incomplete use of component information and global context information in sample data and label data, resulting in the inability to further improve the resolution of seismic data.

[0006] According to one aspect of the present invention, a method for improving the resolution of seismic data based on a wavelet domain residual network is provided, comprising:

[0007] Obtaining sample data including existing earthquake data;

[0008] Inputting the sample data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform;

[0009] Determine the total loss function of the trained U-Net network model;

[0010] According to the total loss function, the U-Net network model is tuned to obtain an optimal model, and the optimal model is used to improve the resolution of the seismic data to be processed to obtain a final processing result.

[0011] Preferably, the residual module is used to obtain a first feature map after performing a 3*3 convolution and an activation operation on the sample data, and to obtain a second feature map and a fourth feature map after performing a 1*1 convolution on the sample data;

[0012] Performing an addition operation on the first feature map and the second feature map to form a first residual block, and performing convolution and activation operations on the first residual block to obtain a third feature map;

[0013] The third feature map and the fourth feature map are added to obtain the output result of the residual module.

[0014] Preferably, the global context module is used to perform 1*1 convolution and normalization processing on the output result of the residual module to obtain a processing result;

[0015] Performing a pixel-based multiplication operation on the processing result and the residual module output to obtain global context information;

[0016] The acquired global context information is sequentially subjected to 1*1 convolution, layer normalization, activation, and 1*1 convolution to obtain a new feature map.

[0017] Preferably, the wavelet transform method comprises:

[0018] The data output after the residual module and the global context module are subjected to a two-dimensional discrete wavelet transform using formula (1):

[0019]

[0020] Where m and n are the parameters of the mother function, W is the discrete transformation function, a is the scale parameter, and b is 1 is the first dimension translation parameter, b 2 is the second dimension translation parameter, t 1 is the first dimension time, t2 is the second dimension time, and b is the generating function parameter.

[0021] Preferably, the method for determining the total loss function of the trained U-Net network model comprises:

[0022] The loss between the output data of the U-Net network and the sample data before wavelet transformation is defined as the first part of the loss;

[0023] The loss between the output data of the U-Net network after wavelet transformation and the sample data after wavelet transformation is defined as the second part of the loss;

[0024] Determine the loss function used for the first part of the loss and the second part of the loss;

[0025] The total loss function is determined according to the loss functions used by the first part of the loss and the second part of the loss.

[0026] Preferably, the loss function used for the first part of the loss includes:

[0027] L MIX =λ·L MS-SSIM +(1-λ)·L 1 (2);

[0028] Where, L 1 is the minimum absolute error function, L MS-SSIM is the multi-scale structural similarity loss function, λ is 0.6;

[0029] The loss functions used in the second part of the loss include:

[0030]

[0031] Where W SR , W HR are the four wavelet coefficient components after wavelet decomposition of the output data of the U-Net network and the high-resolution label data, δ i are the weights of the wavelet coefficients.

[0032] Preferably, the method for determining the total loss function according to the loss function used by the first part of the loss and the second part of the loss includes:

[0033] Determine the total loss function using formula (4);

[0034]

[0035] Where, L WAVE is the loss function of the second part, L MIXis the loss function of the first part, is the weight coefficient.

[0036] According to one aspect of the present invention, a device for improving the resolution of seismic data based on a wavelet domain residual network is provided, comprising:

[0037] An acquisition unit, used for acquiring sample data including existing seismic data;

[0038] A model training unit, used for inputting the sample data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform;

[0039] A loss function determination unit, used to determine the total loss function of the trained U-Net network model;

[0040] The optimal model generating unit is used to tune the U-Net network model according to the total loss function to obtain the optimal model, and use the optimal model to improve the resolution of the seismic data to be processed to obtain the final processing result.

[0041] The present invention has at least the following beneficial effects:

[0042] The present invention proposes a method and device for improving the resolution of seismic data based on a wavelet domain residual network. By adding a residual structure to the neural network, the gradient vanishing and network degradation are reduced; by adding a global context module and wavelet transform, the context and texture information of seismic data at different levels are learned. The training network is adjusted using a total loss function, thereby effectively improving the main frequency and bandwidth of the seismic data, making the event axis of the data clearer, and effectively improving the resolution of the seismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and, together with the specification, are used to explain the technical solutions of the present invention.

[0044] Figure 1 A flow chart showing a method for improving seismic data resolution based on a wavelet domain residual network according to an embodiment of the present invention;

[0045] Figure 2 A schematic diagram showing a method for improving seismic data resolution based on a wavelet domain residual network according to an embodiment of the present invention;

[0046] Figure 3 A schematic diagram showing a residual module according to an embodiment of the present invention is shown;

[0047] Figure 4 A schematic diagram showing a global context module according to an embodiment of the present invention is shown;

[0048] Figure 5 A schematic diagram of a two-dimensional discrete wavelet decomposition process according to an embodiment of the present invention is shown;

[0049] Figure 6 Showing the spectrum diagram of seismic data before and after processing according to an embodiment of the present invention;

[0050] Figure 7 A cross-sectional view showing seismic data before processing according to an embodiment of the present invention is shown;

[0051] Figure 8 A cross-sectional diagram of seismic data after processing according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0052] Various exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0053] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0054] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0055] In addition, in order to better illustrate the present invention, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present invention.

[0056] Figure 1 A flow chart showing a method for improving seismic data resolution based on a wavelet domain residual network according to an embodiment of the present invention; Figure 2 A schematic diagram showing a method for improving seismic data resolution based on a wavelet domain residual network according to an embodiment of the present invention; Figure 3 A schematic diagram showing a residual module according to an embodiment of the present invention is shown; Figure 4 A schematic diagram showing a global context module according to an embodiment of the present invention is shown; Figure 5 A schematic diagram of a two-dimensional discrete wavelet decomposition process according to an embodiment of the present invention is shown; Figure 6 Showing the spectrum diagram of seismic data before and after processing according to an embodiment of the present invention; Figure 7 A cross-sectional view showing seismic data before processing according to an embodiment of the present invention is shown; Figure 8 FIG. 2 shows a cross-sectional view of seismic data after processing according to an embodiment of the present invention. Figure 1-8 As shown, a method for improving the resolution of seismic data based on a wavelet domain residual network includes: step S01: obtaining sample data containing existing seismic data; step S02: inputting the sample data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform; step S03: determining a total loss function of the trained U-Net network model; step S04: according to the total loss function, tuning the U-Net network model to obtain an optimal model, and using the optimal model to improve the resolution of the seismic data to be processed to obtain a final processing result.

[0057] The method for improving the resolution of seismic data based on wavelet domain residual network provided by the embodiment of the present invention specifically includes the following steps:

[0058] Step S01: Acquire sample data including existing earthquake data.

[0059] In the embodiment of the present invention, the sample data of seismic data includes the low-resolution seismic profiles and their corresponding high-resolution seismic profiles that have been processed in the past. The resolution is the degree to which texture information can be distinguished and identified on the seismic profile. The higher the resolution of the seismic profile, the more detailed texture information can be identified. Therefore, improving the resolution of the seismic profile can make the seismic profile clearer and make the identification of information such as the common axis more accurate and complete.

[0060] Step S02: input the sample data into a U-Net network, train the U-Net network, and obtain a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform.

[0061] In the embodiment of the present invention, a residual module and a global context module are added to the U-Net network (semantic segmentation network) to form a global context residual U-Net network (Global Context Residuals U-Net, GCRU-Net). In order to learn the context and texture information of seismic data at different levels, a wavelet transform is added to the GCRU-Net to form a wavelet domain global context residual U-Net network (Wavelet domain Global Context Residuals U-Net, WGCRU-Net). The overall network model is as follows: Figure 2 shown.

[0062] The model mainly consists of U-Net network, sub-pixel convolution of residual module, global context module and wavelet transform. The U-Net network can realize feature extraction and feature recovery, and its unique long connection can reduce information loss and enable better feature fusion.

[0063] In the present invention, the residual module is used to obtain a first feature map after 3*3 convolution and activation operation on the sample data, and to obtain a second feature map and a fourth feature map after 1*1 convolution on the sample data; add the first feature map and the second feature map to form a first residual block, and convolve and activate the first residual block to obtain a third feature map; add the third feature map and the fourth feature map to obtain the output result of the residual module.

[0064] In the embodiments of the present invention and Figure 3 In the paper, residual convolution modules are added to the downsampling and upsampling of the U-Net network, which can better extract features and alleviate the problem of network gradient disappearance, thereby improving the performance of the network.

[0065] The seismic profile in the sample data, i.e., the source feature map R, is input into the residual module for 3*3 convolution Conv and activation (ReLU function) operations to obtain the first feature map R 1 ; The source feature map R is then convolved by 1*1 Conv1x1 to obtain the second feature map R 2 And the fourth characteristic map R 4 ; R 1 and R 2 The addition operation is performed to form the first residual block; the first residual block is input into the second convolution again for 3*3 convolution Conv, and after activating ReLU, the third feature map R is formed. 3 , R 3 and R 4 The addition operation is performed to form the final output.

[0066] In the present invention, the global context module is used to perform 1*1 convolution and normalization processing on the output result of the residual module to obtain a processing result; perform a pixel-based multiplication operation on the processing result and the output of the residual module to obtain global context information; and perform 1*1 convolution, layer normalization, activation, and 1*1 convolution on the obtained global context information in sequence to obtain a new feature map.

[0067] In the embodiments of the present invention and Figure 4 In the figure, the output result of the residual module is subjected to 1*1 convolution Conv1x1 and normalized Softmax operation, and then pixel-based multiplication operation is performed with the output result of the residual module to obtain global context information; the global context information is then subjected to 1*1 convolution Conv1x1, layer normalization LN, activation ReLU, and 1*1 convolution Conv1x1 in sequence to achieve feature conversion; among them, layer normalization can simplify the optimization process and act as a regularizer, and finally the output result of the residual module is added to obtain a new feature map.

[0068] In the present invention, the wavelet transform method comprises: performing a two-dimensional discrete wavelet transform on the data output after passing through the residual module and the global context module using formula (1):

[0069]

[0070] Where m and n are the parameter pair of the mother function ψ, W is the discrete transformation function, a is the scale parameter, and b is 1 is the first dimension translation parameter, b 2 is the second dimension translation parameter, W(a,b 1 ,b 2 ) is the coefficient obtained by two-dimensional discrete wavelet transform, indicating that at scale a and position (b 1 ,b 2 ) on the wavelet coefficients, t 1 is the first dimension time variable, t 2 is the second-dimensional time variable, ψ a,b is the generating function, and b is the generating function parameter.

[0071] In the embodiment of the present invention, wavelet transform is added to the output of the last layer of the U-Net network. Wavelet transform overcomes the limitation of Fourier transform for non-stationary signals by automatically adjusting the time and frequency windows, and can convert data from the spatial domain to the wavelet domain. By combining wavelet transform with deep learning, subgraphs at different levels can be trained separately to obtain more detailed information. The local idea in wavelet transform can refine the local analysis of data. After wavelet transform, more details can be observed, solving the endpoint and continuity problems ignored by other analysis methods, and bringing convenience in analyzing subtle time domain signals and frequency domain signals.

[0072] Signal f(t)∈L 2 The continuous wavelet transform of (R) is defined as:

[0073]

[0074] In the formula, a is the scale factor, b is the position factor, |a| -1 / 2 is the normalization factor, tb is the translation factor, t is a continuous time variable, and the signal f(t) can be any function, representing the change in time. 2 (R) represents the signal space, that is, the set of square integrable functions, and R is the entire real number axis.

[0075] The above formula (5) can be simplified as:

[0076]

[0077] in, It is called wavelet basis function.

[0078] In wavelet transform, due to the correlation of wavelet basis functions, the coefficient information is redundant. In order to reduce the redundancy and parameter quantity of wavelet transform while retaining the integrity of the original signal, a discrete wavelet function is proposed. The definition is as follows:

[0079]

[0080] In the formula, m and n are the parameter pair of ψ, a 0 is the scale parameter, b 0 is the translation parameter.

[0081] The discrete wavelet transform is defined as follows:

[0082]

[0083] In image processing, the data to be processed is usually two-dimensional or multi-dimensional. Applying the principle of wavelet analysis, the discrete wavelet transform is extended to two dimensions, which is defined as formula (1).

[0084] The inverse transform is defined as:

[0085]

[0086] Where, t 1 is the first dimension time, t 2 is the second dimension time, C is the inverse function of the generating function, a is the scale parameter, b 1 is the first dimension translation parameter, b 2 is the second-dimensional translation parameter, ψ a, b is the generating function.

[0087] The specific steps of using the Haar wavelet basis function to perform two-dimensional discrete wavelet decomposition on the seismic data, that is, the output data of the global context module, are as follows: first, perform one-dimensional discrete wavelet decomposition on each row of the seismic data to obtain the low-frequency component L and high-frequency component H in the horizontal direction, and then perform one-dimensional discrete wavelet decomposition on each column of L and H to obtain the low-frequency component LL, horizontal high-frequency component LH, vertical high-frequency component HL, and diagonal high-frequency component HH of the original seismic data (output data of the global context module). The transformation process is as follows: Figure 5 The seismic profiles before and after the two-dimensional discrete wavelet decomposition are shown in Figure 5 shown.

[0088] Figure 2 This is a schematic diagram of the internal operation principle of the U-Net network of the present invention. Figure 2 The numbers in the figure are the number of neurons in each layer.

[0089] Step S03: Determine the total loss function of the trained U-Net network model.

[0090] In the present invention, the method for determining the total loss function of the trained U-Net network model includes: defining the loss between the output data of the U-Net network before wavelet transformation and the sample data as the first partial loss; defining the loss between the output data of the U-Net network after wavelet transformation and the sample data after wavelet transformation as the second partial loss; determining the loss functions used for the first partial loss and the second partial loss; and determining the total loss function based on the loss functions used for the first partial loss and the second partial loss.

[0091] In an embodiment of the present invention, after adding the two-dimensional discrete wavelet transform, the loss function of the U-Net network includes two parts. The first part is the loss between the output of the U-Net network before the wavelet transform and the label data in the sample data; the second part is the wavelet loss between the output of the U-Net network after the wavelet transform and the label data in the data after the wavelet transform.

[0092] Different network models require different loss functions. Choosing a suitable loss function can affect the performance of the model on the task.

[0093] In the present invention, the loss function used in the first part of the loss includes:

[0094] L MIX =λ·L MS-SSIM +(1-λ)·L 1 (2);

[0095] Where, L 1 is the minimum absolute error function, LMS-SSIM is the multi-scale structural similarity loss function, and λ is 0.6.

[0096] In the embodiment of the present invention, the loss function used in the first part of the loss is L 1 The linear combination of the loss function and the MS-SSIM loss function. 1 The loss function is also called the minimum absolute error function. Its purpose is to minimize the absolute error between the estimated value and the target value. It is defined as:

[0097]

[0098] Where m is the total number of pixels, x and y are the super-resolution image and the high-resolution image, respectively, and θ is the parameter set.

[0099] MS-SSIM is a multi-scale structural similarity loss function. The scale factor is added to SSIM to better approach the human visual system and achieve subjective consistency. It is defined as:

[0100]

[0101]

[0102] In the formula, μ x , μ y is the mean of x and y, σ xy is the covariance, is the variance of x and y, c 1 、c 2 、c 3 There are three constants, n represents the scale. Generally, the scale number is set to 5. α, β, and γ are the weights of brightness, contrast, and structure, and the weight values ​​are {0.0448, 0.2856, 0.3001, 0.2363, 0.1333}.

[0103] The loss function L used in the first part of the loss is MIX It is defined as formula (2).

[0104] In the present invention, the loss function used by the second part of the loss includes:

[0105]

[0106] Where W SR , W HR are the four wavelet coefficient components after wavelet decomposition of the output data of the U-Net network and the high-resolution label data, δ i are the weights of the wavelet coefficients.

[0107] In the embodiment of the present invention, the data output after being processed by the U-Net network and the high-resolution label data are each subjected to wavelet decomposition to obtain wavelet coefficients, which are expressed as:

[0108] W SR =(LL 1 ,LH 1 ,HL 1 ,HH 1 ) (10);

[0109] W HR =(LL 2 ,LH 2 ,HL 2 ,HH 2 ) (11);

[0110] In the formula, LL i , LH i , HL i , HH i They are respectively the low-frequency component, horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component of the data (sample data) after wavelet transform (two-dimensional discrete wavelet transform).

[0111] The second part of the loss (wavelet coefficient loss) is calculated using the MSE (root mean square) function and can be expressed as shown in formula (3).

[0112] In the present invention, the method for determining the total loss function based on the loss function used by the first part of the loss and the second part of the loss includes: determining the total loss function using formula (4);

[0113]

[0114] Among them, L WAVE is the loss function of the second part, L MIX is the loss function of the first part, is the weight coefficient.

[0115] In the embodiment of the present invention, the difference between the trained network model and the sample data can be determined based on the total loss function. The smaller the total loss function is, the higher the accuracy of the trained network model is and the more accurate the prediction result is.

[0116] Step S04: According to the total loss function, the U-Net network model is tuned to obtain an optimal model, and the optimal model is used to improve the resolution of the seismic data to be processed to obtain a final processing result.

[0117] In the embodiment of the present invention, the training process of the network model is an optimization problem, and the goal is to minimize the loss function. By continuously adjusting the parameters of the model, the total loss function is gradually reduced, and the prediction ability of the model is gradually improved.

[0118] According to the total loss function, the U-Net network model parameters are adjusted and trained until the total loss function of the trained model is minimized, thus obtaining the final optimal model.

[0119] In an embodiment of the present invention, the seismic profile whose resolution is to be improved is input into the optimal model obtained after training to perform resolution improvement processing, and after running, a seismic profile with higher clarity will be obtained, that is, the final processing result.

[0120] Seismic data contains complex noise that can reflect the real underground structure. Compared with synthetic seismic data, it is much more difficult to improve the resolution and signal-to-noise ratio. In order to test the generalization ability of the optimal network model obtained by the method of the present invention, the actual seismic data is processed to improve the resolution. The spectrum before and after processing is shown in the figure. Figure 6 As shown in the spectrum diagram, it can be seen that after the actual seismic data is processed by the optimal model of the present invention, the bandwidth is widened, the main frequency is also improved, and the low-frequency information is well preserved. The spectrum change trend of the seismic data before and after processing is the same, indicating that this method is real and effective in improving the resolution of seismic data.

[0121] The seismic sections before and after processing are as follows: Figure 7 and 8 As shown, Figure 7 To process the previous seismic profile, the profile contains a lot of noise and some artifacts. From the local magnified image, it can be seen that the event axis is not clear and the texture detail information is not rich enough. Figure 8 For the processed seismic profile, it can be clearly seen that the texture detail information in the profile is richer, some noise is suppressed, and artifacts are weakened. From the local magnified image, it can be seen that the phase axis becomes clearer and the resolution is greatly improved, indicating that the network model of the present invention has good generalization ability when processing actual seismic data.

[0122] It can be understood that the above-mentioned various method embodiments mentioned in the present invention can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present invention will not go into details.

[0123] The execution subject of the method for improving the resolution of seismic data based on wavelet domain residual network can be a device for improving the resolution of seismic data based on wavelet domain residual network. For example, the method for improving the resolution of seismic data based on wavelet domain residual network can be executed by a terminal device or a server or other processing device, wherein 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 implementations, the method for improving the resolution of seismic data based on wavelet domain residual network can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0124] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0125] The present invention provides a device for improving the resolution of seismic data based on a wavelet domain residual network, comprising: an acquisition unit, used for acquiring sample data of seismic data; a model training unit, used for inputting the sample data and label data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network comprises a residual module, a global context module, and a wavelet transform; a loss function determination unit, used for determining a total loss function of the trained U-Net network model; and an optimal model generation unit, used for tuning the U-Net network model according to the total loss function, obtaining an optimal model, and using the optimal model to perform resolution improvement processing on seismic data to be processed, and obtaining a final processing result.

[0126] In some embodiments, the functions or modules included in the device provided by the embodiments of the present invention can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0127] The present invention relates to a method for improving the resolution of seismic data based on a wavelet domain residual network. The residual structure is added to the network to reduce gradient vanishing and network degradation. Then, wavelet transform and global context modules are added to learn the context and texture information of seismic data at different levels. Then, the network is trained using wavelet loss and data loss to improve the robustness of the model. The processing results of simulated data and actual data show that the method of the present invention improves the main frequency and bandwidth of seismic data, makes the event axis of the data clearer, effectively improves the resolution, and can further improve the accuracy of oil and gas exploration and reduce exploration costs.

[0128] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for improving the resolution of seismic data based on wavelet domain residual network, It is characterized in that include: Obtaining sample data including existing earthquake data; Inputting the sample data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform; Determine the total loss function of the trained U-Net network model; According to the total loss function, the U-Net network model is tuned to obtain an optimal model, and the optimal model is used to improve the resolution of the seismic data to be processed to obtain a final processing result.

2. The method for improving seismic data resolution based on wavelet domain residual network according to claim 1, Features: The residual module is used to obtain a first feature map after performing a 3*3 convolution and an activation operation on the sample data, and to obtain a second feature map and a fourth feature map after performing a 1*1 convolution on the sample data; Performing an addition operation on the first feature map and the second feature map to form a first residual block, and performing convolution and activation operations on the first residual block to obtain a third feature map; The third feature map and the fourth feature map are added to obtain the output result of the residual module.

3. The method for improving seismic data resolution based on wavelet domain residual network according to claim 1, Features: The global context module is used to perform 1*1 convolution and normalization processing on the output result of the residual module to obtain a processing result; Performing a pixel-based multiplication operation on the processing result and the residual module output to obtain global context information; The acquired global context information is sequentially subjected to 1*1 convolution, layer normalization, activation, and 1*1 convolution to obtain a new feature map.

4. The method for improving seismic data resolution based on wavelet domain residual network according to claim 1, It is characterized in that The wavelet transform method comprises: The data output after the residual module and the global context module are subjected to a two-dimensional discrete wavelet transform using formula (1): Where m and n are the parameters of the mother function, W is the discrete transformation function, a is the scale parameter, and b is 1 is the first dimension translation parameter, b 2 is the second dimension translation parameter, t 1 is the first dimension time, t 2 is the second dimension time, and b is the generating function parameter.

5. The method for improving the resolution of seismic data based on wavelet domain residual network according to any one of claims 1 to 4, It is characterized in that The method for determining the total loss function of the trained U-Net network model includes: The loss between the output data of the U-Net network and the sample data before wavelet transformation is defined as the first part of the loss; The loss between the output data of the U-Net network after wavelet transformation and the sample data after wavelet transformation is defined as the second part of the loss; Determine the loss function used for the first part of the loss and the second part of the loss; The total loss function is determined according to the loss functions used by the first part of the loss and the second part of the loss.

6. The method for improving the resolution of seismic data based on wavelet domain residual network according to claim 5, Features: The loss functions used in the first part of the loss include: Where, L 1 is the minimum absolute error function, L MS-SSIM is the multi-scale structural similarity loss function, λ is 0.6; The loss functions used in the second part of the loss include: Where W SR , W HR are the four wavelet coefficient components after wavelet decomposition of the output data of the U-Net network and the high-resolution label data, δ i are the weights of the wavelet coefficients.

7. The method for improving seismic data resolution based on wavelet domain residual network according to claim 5, It is characterized in that The method of determining the total loss function according to the loss function used by the first part of the loss and the second part of the loss includes: Determine the total loss function using formula (4); Where, L WAVE is the loss function of the second part, L MIX is the loss function of the first part, is the weight coefficient.

8. A device for improving the resolution of seismic data based on wavelet domain residual network, It is characterized in that include: An acquisition unit, used for acquiring sample data including existing seismic data; A model training unit, used for inputting the sample data into a U-Net network, training the U-Net network, and obtaining a trained U-Net network model, wherein the U-Net network includes a residual module, a global context module, and a wavelet transform; A loss function determination unit, used to determine the total loss function of the trained U-Net network model; The optimal model generating unit is used to tune the U-Net network model according to the total loss function to obtain the optimal model, and use the optimal model to improve the resolution of the seismic data to be processed to obtain the final processing result.

Images