Method and system for improving resolution of medium-deep seismic data

Abstract

The invention discloses a medium-deep seismic data resolution improving method and system, and the method comprises the steps: carrying out the preprocessing of three-dimensional seismic data of the same work area, and constructing seismic data through combining with an automatic gain control function; introducing a data augmentation method and a data splicing method; a 3D Cycle-GAN network architecture is adopted to extract features of the shallow high-resolution seismic data; introducing a pair of loss functions combined by generator loss and discriminator loss; and inputting corresponding shallow low-resolution data obtained after low-pass filtering of the shallow seismic data, the deep seismic data and the shallow high-resolution data into the network for training, inputting the deep low-resolution data into the network, and performing AGC inverse transformation on the output of the deep low-resolution data to finally realize improvement of the resolution of the deep seismic data. According to the method, the deep layer and the shallow layer in the same region have very similar geologic structure characteristics, weak supervised learning is carried out by using unpaired deep layer and shallow layer data, and the resolution of deep layer seismic data can be effectively improved.

Description

Background Art

[0002] Oil and gas resources are the foundation of modern industry and play an irreplaceable role in the national economy and people's livelihood. Seismic exploration is a technology used to reveal underground structural characteristics, infer stratum properties and their morphology. Seismic data can be obtained through seismic detection methods. These data can infer the depth and ups and downs of the stratum, and the changes in wave impedance in the stratum can be inferred from them to reveal the geological structure and stratum composition, which has important reference value for the exploration of oil and gas resources. In particular, the high-frequency information usually contains key stratum details, such as stratification and fault information. Therefore, obtaining high-resolution seismic data is extremely critical for accurately interpreting underground structures. However, since the acquisition process of seismic data is affected by various factors such as stratum attenuation, the bandwidth of the obtained raw data is reduced and the data resolution is low. Although modern seismic data acquisition technology has made significant progress, the quality and resolution of the acquired data have been greatly improved, but many acquired seismic data still face the problem of low resolution. And with the development of exploration technology over the years, the current exploration target has shifted to some deep strata with more complex geological structures, the difficulty of data acquisition has increased, and the requirements for data accuracy have also increased. High-resolution seismic data are essential for accurate interpretation of geological structures, so it is necessary to study high-resolution methods for seismic data. Research on improving the resolution of seismic data has attracted widespread attention from researchers.

[0003] In recent years, with the rapid development of deep learning, deep learning-based technologies have been widely used in many fields and have achieved remarkable results. Deep learning can automatically extract relevant features from learning targets through training based on selected input data and learning targets, and is very suitable for processing complex seismic data. At present, deep learning-based technologies have been applied in many directions of seismic data processing, such as noise suppression, interpolation, fault detection, resolution enhancement, etc.

[0004] Seismic data processing methods based on deep learning have been in a stage of rapid development in recent years. Scholars have done a lot of work in the fields of noise suppression, interpolation, and high resolution. Research has proved the feasibility of deep learning in seismic data processing. However, the current research on its application in the field of actual seismic data processing is still limited by sample quality and is in its early stages. In order to avoid the limitation of training samples, self-supervised learning has received widespread attention from scholars in the fields of image restoration, image completion, and image generation. Research on how to construct unpaired datasets suitable for network weakly supervised learning can apply deep learning technology to a wider range of actual seismic data high-resolution processing, which has valuable research value.

[0005] Existing deep learning methods usually require completely corresponding paired data and labels for training. However, since strong supervised learning has very strict requirements on paired samples, each low-resolution data needs to be completely paired with the high-resolution label it is paired with. Therefore, the quality of the high-resolution label directly determines the effect of the high-resolution processing results. Therefore, if you want to apply strong supervised learning methods to the field of high-resolution processing of seismic data, obtaining good quality labels is a major difficulty. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for improving the resolution of medium-deep seismic data in view of the deficiencies in the above-mentioned prior art, so as to solve the technical problem that the prior method requires paired data for strong supervised learning.

[0007] The present invention adopts the following technical solutions:

[0008] A method for improving the resolution of medium-deep seismic data comprises the following steps:

[0009] S1. Preprocess the 3D seismic data of the same work area; construct seismic data by combining the automatic gain control function, and introduce data augmentation method and data splicing method to supplement the seismic data;

[0010] S2, build a 3D Cycle-GAN network architecture to extract shallow high-resolution seismic data features;

[0011] S3, introducing a pair of loss functions composed of a generator loss and a discriminator loss into the 3D Cycle-GAN network obtained in step S2;

[0012] S4, input the shallow and deep seismic data obtained in step S1 and the shallow high-resolution data obtained after low-pass filtering to the corresponding shallow low-resolution data obtained in step S3 into the 3D Cycle-GAN network for training, input the deep low-resolution data into the trained 3D Cycle-GAN network, and perform AGC inverse transformation on the output to improve the resolution of deep seismic data.

[0013] Preferably, in step S1, for the input seismic data D in Each of the i , the result d′ after the signal is preprocessed using the automatic gain control algorithm i as follows:

[0014]

[0015] Among them, W triangle (L) is a triangular window with a length of L.

[0016] Preferably, the data augmentation method is specifically:

[0017] Flip the Inline and Crossline spatial dimensions, swap the Inline and Crossline spatial dimensions, reverse the data as a whole, and get the augmented data Each curly bracket indicates that one of the items is randomly selected with equal probability.

[0018] Preferably, the data splicing method is specifically as follows:

[0019] When dividing the input data into blocks, set a 50% overlap on the data;

[0020] After high-resolution processing is completed, the overlapping parts between high-resolution data blocks are multiplied by the set weights through the windowing method, and then spliced ​​back to the original shape to ensure that the sum of the weights of the spliced ​​data at each position is 1;

[0021] Expand the triangular window to three-dimensional space, and the expression of the window function is as follows:

[0022]

[0023] Among them, the length of the window function is 2N, i, j, k are indexes of three dimensions respectively, 0≤i,j,k≤2N.

[0024] Preferably, in step S2, the 3D Cycle-GAN network architecture includes a forward loop and a reverse loop, the forward loop and the reverse loop use two sets of generators and two sets of discriminators, the forward and reverse networks of the two sets of generators and discriminators maintain the same architecture, share the same architecture and are jointly trained, and the parameters are updated independently;

[0025] The generator includes a forward generator G LR2HR and the reverse generator G HR2LR , the forward generator G HR2LR Map LR data to HR data, the reverse generator G LR2HR Map HR data to LR data;

[0026] The discriminator includes LR discriminator D LR and HR discriminator D HR , LR discriminator D LR and HR discriminator D HR They are used to determine whether the input data is real LR or HR data.

[0027] Preferably, the forward generator G LR2HR and the reverse generator G HR2LR It is a three-dimensional network that processes two-dimensional data and three-dimensional data at the same time and uses the same network architecture;

[0028] The data is first input by the first convolutional layer and then passes through 16 residual blocks connected end to end in the main part. Each residual block contains two convolutional layers and residual connections. The outer layer of the residual block group has a cross-layer connection.

[0029] Finally, the data goes through multiple convolutional layers and is output through the tanh activation function.

[0030] Preferably, no pooling layers are included in the generator network; residual blocks are used and no batch normalization layers are used.

[0031] Preferably, the LR discriminator D LR and HR discriminator D HR It is a two-dimensional network, using the same network architecture;

[0032] The data first enters the input convolution layer and passes through 5 convolution blocks;

[0033] Each convolutional block contains a convolutional layer, a batch normalization layer, and a Leaky ReLU activation function;

[0034] The data then passes through the Sigmoid activation function and outputs a probability between 0 and 1;

[0035] The discriminator adopts the PatchGAN architecture and replaces the last fully connected layer with a convolutional layer.

[0036] Preferably, in step S3, the loss function is decomposed into generator loss and discriminator loss;

[0037] The generator loss is:

[0038] L Gen =w 1 L GAN +w 2 L cyc +w 3 L idt +w 4 L TV

[0039] Among them, w 1 、w 2 、w 3 and w 4 represents the weight assigned to each loss function, L GAN To combat the loss, L cyc is the cycle consistency loss, L idt is the constant loss, L TV is the total variation loss;

[0040] The discriminator loss is:

[0041]

[0042] Among them, N is the number of training samples in one round of network training, D HR is the HR discriminator, D LR is the LR discriminator, G LR2HR (x i ) is the low-resolution data χ, which is the high-resolution data obtained by the input forward generator, G HR2LR (y i ) is the low-resolution data obtained by inputting the high-resolution data y into the reverse generator.

[0043] In a second aspect, an embodiment of the present invention provides a system for improving the resolution of mid-deep seismic data, comprising:

[0044] The data module pre-processes the 3D seismic data of the same work area; constructs seismic data by combining the automatic gain control function, and introduces data augmentation and data splicing methods to supplement the seismic data;

[0045] Network module, building a 3D Cycle-GAN network architecture to extract shallow high-resolution seismic data features;

[0046] Function module, which introduces a pair of loss functions composed of generator loss and discriminator loss into the 3D Cycle-GAN network;

[0047] The improvement module inputs the shallow and deep seismic data and the shallow high-resolution data obtained after low-pass filtering into the 3D Cycle-GAN network obtained in step S3 for training, inputs the deep low-resolution data into the trained 3D Cycle-GAN network, and performs AGC inverse transformation on the output to improve the resolution of deep seismic data.

[0048] In a third aspect, a computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for improving the resolution of mid-deep seismic data when executing the computer program.

[0049] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for improving the resolution of mid-deep seismic data.

[0050] Compared with the prior art, the present invention has at least the following beneficial effects:

[0051] A method for improving the resolution of medium-deep seismic data is proposed based on the Cycle-GAN network structure. A two-dimensional to three-dimensional Cycle-GAN architecture is proposed. Since the deep gully area in the loess plateau has been eroded by water for a long time, resulting in the loss of surface loess, the impact on seismic waves is greatly reduced. Therefore, the resolution of the curved line seismic data collected along the deep gully area is higher than that of the three-dimensional loess plateau data. Since the loess plateau data and the curved line seismic data are collected at the same geographical location, they actually reflect the same stratigraphic characteristics, so the data of the two are highly correlated. Based on this feature, a deep learning method is used, and a weakly supervised learning method is used to establish a feature distribution mapping between the three-dimensional loess plateau data and the two-dimensional curved line seismic data; at the same time, data augmentation, automatic gain control and unique data splicing processing are added to the three-dimensional seismic data to improve the resolution of the entire loess plateau data.

[0052] Furthermore, due to the unpaired nature of the 3D work area data and the 2D bendline seismic data, and the use of weakly supervised learning methods to learn the distribution and mapping relationship between the two, but due to the different dimensions of the two, we designed a convolutional neural network that can process both 2D and 3D data simultaneously, thereby better preserving the spatial information of the seismic data.

[0053] Furthermore, since the loess plateau data and the curved line seismic data have the same geographical location but different dimensions, there is a complex correspondence between the two. In order to solve the problem of neural network learning difficulties, bidirectional cycle consistency is introduced, and the forward and reverse networks maintain the same architecture, share the same architecture and are jointly trained. Based on the above conditions, a unified learning process is ensured. Since the bidirectional cycle introduces additional constraints, the stability in training is significantly improved, the artifacts of the output data are reduced, and the overall quality of the generated HR seismic data is improved.

[0054] Furthermore, in actual seismic data, the reflection coefficients of different strata may vary greatly, and the seismic signals received by the detector also vary within a large range. If the selected partial reflection signal is too weak, it is difficult for the network to learn the characteristics of the weak reflection signal. In order to adjust the amplitude of the signal to a suitable range, the automatic gain control algorithm (AGC) is introduced to pre-process the signal. The amplitude of the signal is first adjusted to a range suitable for network training. After the network processing is completed, the AGC inverse transform is used to restore the amplitude of the signal. The AGC method is helpful for the network to capture the characteristics of weak reflection signals.

[0055] Furthermore, in order to solve the problem of improving seismic resolution, a pair of loss functions are designed according to the characteristics of seismic data. The loss function consists of two components: generator loss and discriminator loss. Gen It consists of four parts: adversarial loss LGAN , cycle consistency loss L cyc , constant loss L idt and total variation loss L TV The discriminator loss is also defined by the negative log-likelihood to reduce the prediction error and prevent overfitting.

[0056] Furthermore, in order to prevent overfitting due to insufficient training data, a large amount of training data is required. However, due to the limitations of exploration technology, the old data used for high-resolution processing is often small in size, which exacerbates the problem of insufficient training data. Therefore, we introduced a series of data augmentation methods. Through this data augmentation method, the number of training samples is effectively expanded. Different data augmentation methods can be used for input data and labels, allowing the network to access the entire data set from multiple perspectives, which provides greater flexibility than the data augmentation method of paired training data.

[0057] Furthermore, due to GPU memory limitations, the network cannot input the entire 3D seismic data volume at once, so the data needs to be subdivided into fixed-size blocks for high-resolution processing, and the traditional direct splicing strategy may introduce discontinuities at the boundaries of the processed blocks. To address this challenge and enhance the overall integrity of seismic data, we proposed an improved boundary processing strategy. This improved boundary processing strategy effectively reduces the artifacts that may be generated by the direct splicing method. By adopting our designed overlapping sampling and data windowing technology, our method ensures that high-resolution seismic features are more accurately retained, ensuring the effect of 3D seismic data processing to ultimately output high-resolution data.

[0058] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0059] In summary, based on the Cycle-GAN network structure, the present invention proposes a two-dimensional-three-dimensional Cycle-GAN architecture; utilizing the fact that the loess tableland work area data and the curved line seismic data are geographically collected at the same location, and the two actually reflect the same stratigraphic characteristics and have strong correlation characteristics, a deep learning method is used to establish a characteristic distribution mapping between the three-dimensional loess tableland data of the work area and the two-dimensional curved line seismic data, and at the same time, data augmentation, automatic gain control and unique data splicing processing are added to the seismic data, thereby improving the resolution of the entire loess tableland data.

[0060] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings to be used in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0062] Figure 1 A schematic diagram of the three-dimensional seismic data network training process proposed by the present invention;

[0063] Figure 2 Schematic diagram of the generator in the 3D Cycle-GAN architecture;

[0064] Figure 3 Schematic diagram of the discriminator in the 3D Cycle-GAN architecture;

[0065] Figure 4 is a comparison of the results before and after AGC processing, where (a) is the original seismic data and (b) is the data after AGC processing;

[0066] Figure 5 This is a schematic diagram of data block processing;

[0067] Fig. 6 is a schematic diagram of the three-dimensional window function used for data splicing, where (a) is a three-dimensional structure, (b) is a two-dimensional section with z = 0, and (c) is z = 0 with different values ​​of y;

[0068] Figure 7 Normalize amplitude spectra of simulated seismic data for deep and shallow layers;

[0069] FIG8 is a comparison of the results of the simulated seismic data Inline section (Inline=200), wherein (a) is the original deep data, (b) is the corresponding high-resolution data, (c) is the result of the spectrum whitening method, and (d) is the result of the method of the present invention;

[0070] Fig. 9 Comparison of the results of the simulated seismic data Crossline profile (Crossline = 200), where (a) is the original deep data, (b) is the corresponding high-resolution data, (c) is the result of the spectrum whitening method, and (d) is the result of the method of the present invention;

[0071] Fig.10 Comparison of the results of simulated seismic data isochronous slices (t = 450ms), where (a) is the original deep data, (b) is the corresponding high-resolution data, (c) is the result of the spectrum whitening method, and (d) is the result of the method of the present invention;

[0072] Fig.11 To simulate the comparison of multi-channel normalized amplitude spectra of seismic data;

[0073] Fig.12 (a) is a crossline profile of real seismic data. The red box highlights the shallow part, which has a relatively high resolution and is used as the network input. In contrast, the blue box marks the deep part, which has a lower resolution due to seismic wave attenuation and is selected as the training target for weakly supervised learning. (b) is a comparison of the normalized amplitude spectra between shallow high-resolution (HR) data (red) and deep low-resolution (LR) data (blue).

[0074] FIG13 is a normalized amplitude spectrum, wherein (a) is the normalized amplitude spectrum of the Inline section in FIG14 , and (b) is the normalized amplitude spectrum of the Crossline section in FIG15 ;

[0075] FIG14 is an Inline profile of actual seismic data, where (a) is deep LR seismic data, (b) is the HR result obtained by the traditional time-varying spectrum whitening method, and (c) is the result obtained by the method of the present invention. The green box shows that the method of the present invention significantly improves the resolution, demonstrating its robustness in revealing geological structures.

[0076] FIG15 is a crossline section of actual seismic data, where (a) is deep LR seismic data, (b) is the HR result obtained by the traditional time-varying spectrum whitening method, and (c) is the result obtained by the method of the present invention. The yellow box indicates that the method of the present invention significantly improves the resolution under noise conditions, indicating that it has a strong anti-interference ability to noise;

[0077] FIG16 is a time section of actual seismic data at t=2284 ms, wherein (a) is the original LR seismic data, (b) is the HR result obtained by the traditional time-varying spectrum whitening method, and (c) is the result obtained by the method of the present invention, as shown in the yellow rectangle. The method of the present invention depicts a clearer geological structure than the spectrum whitening method;

[0078] FIG17 is a time section of actual seismic data at t=2504 ms, wherein (a) is the original LR seismic data, (b) is the HR result obtained by the traditional time-varying spectrum whitening method, and (c) is the result obtained by the method of the present invention, as shown in the yellow rectangle, the method of the present invention reveals more subtle geological structures;

[0079] Fig.18 It is a flowchart of the present invention;

[0080] Fig.19 A schematic diagram of a computer device provided by an embodiment of the present invention;

[0081] Fig. 20 The present invention is a block diagram of an electronic device provided according to an embodiment. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0084] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0085] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. 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 character " / " in the present invention generally indicates that the associated objects are in an "or" relationship.

[0086] It should be understood that, although the terms first, second, third, etc. may be used to describe preset ranges, etc. in the embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are only used to distinguish preset ranges from each other. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0087] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0088] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0089] The present invention provides a method for improving the resolution of medium-deep seismic data. Based on the Cycle-GAN network structure, a 3D Cycle-GAN architecture is proposed. Taking advantage of the fact that the deep and shallow layers in the same work area have very similar geological structural characteristics, and due to the unpaired characteristics of deep and shallow data, a deep learning method is adopted, and a weakly supervised learning method is used to learn the distribution and mapping relationship of deep and shallow data; at the same time, data augmentation, automatic gain control and unique data splicing processing are added to the three-dimensional seismic data, so that the resolution of deep seismic data can be effectively improved.

[0090] See also Fig.18 The present invention provides a method for improving the resolution of medium-deep seismic data. Based on deep learning, a 3D Cycle-GAN architecture is proposed on the basis of the Cycle-GAN network structure. The weakly supervised learning method is used to learn the distribution and mapping relationship between deep and shallow data, which can effectively improve the resolution of deep seismic data. The specific implementation steps are as follows:

[0091] S1. Preprocess the 3D seismic data of the same work area, including dividing the deep and shallow layers based on the time window; combining the automatic gain control function to build a more uniform energy distribution, which is convenient for the network to learn the input seismic data; introducing the data augmentation method to build a larger data set; introducing the data splicing method to enhance the overall integrity of the 3D seismic data;

[0092] Please refer to Figure 4. In actual seismic data, since the reflection coefficients of different strata may vary greatly, the seismic signals received by the detector also vary within a large range. During the training process, only a small amount of data is intercepted each time to form a sample set and sent to the network for training. If the selected partial reflection signal is too weak, use L 2The error obtained when using loss functions such as norm is very small, so it is difficult for the network to learn the characteristics of weak reflection signals. In order to adjust the amplitude of the signal to a suitable range, attenuate large signals, and amplify small signals, the automatic gain control algorithm (AGC) can be used to pre-process the signal. First, adjust the amplitude of the signal to a range suitable for network training. After the network processing is completed, use the AGC inverse transform to restore the amplitude of the signal. The AGC method is helpful for the network to capture the characteristics of weak reflection signals. For the input seismic data D in Each of the i , the AGC result is calculated by the following formula:

[0093]

[0094] Where d′ i is d i The AGC result, W triangle (L) is a triangular window with a length of L. The results of AGC processing of actual seismic data are shown in Figure 4. The energy of the original data is concentrated in the time window range of t = 1.2 to 1.6 s, and the energy in the time window range of t = 0.5 to 1.2 s is much lower than the former; the energy distribution of the data after AGC processing is more uniform, which is convenient for the network to learn the characteristics of the input seismic data.

[0095] Since a 3D network has more parameters than a 2D network for the same depth of convolutional neural network, a large amount of training data is required to prevent overfitting due to insufficient training data. However, due to the limitations of exploration technology, the old data used for high-resolution processing is often small in scale, which further exacerbates the problem of insufficient training data. For training data D with a 3D structure, the present invention uses the following three-point data augmentation method:

[0096] 1) Flip the Inline and Crossline spatial dimensions

[0097]

[0098] in N i and N x The length of the data in Inline and Crossline.

[0099] 2) Swap the two spatial dimensions of Inline and Crossline

[0100]

[0101] 3) Reverse the overall data

[0102]

[0103] in, is the augmented data, and each curly bracket in the formula represents randomly selecting one of them with equal probability. Through this data augmentation method, the number of training samples is effectively expanded. Different data augmentation methods can be used for input data and labels, allowing the network to access the entire data set from multiple angles, which provides greater flexibility than the data augmentation method of paired training data.

[0104] See also Figure 5 Due to the limitation of GPU video memory, the network cannot input the entire 3D seismic data volume at once, so the data needs to be subdivided into fixed-size blocks for high-resolution processing. The network processes each block of data in blocks and then splices it to its original size, so the selected boundary processing method is crucial to the quality of the output data. The traditional direct splicing strategy may introduce discontinuities at the boundaries of the processed blocks. In order to meet this challenge and enhance the overall integrity of seismic data, the present invention proposes an improved boundary processing strategy.

[0105] When the input data is divided into blocks, a 50% overlap is set for the data. After the high-resolution processing is completed, the overlapping parts between the high-resolution data blocks are multiplied by appropriate weights by the windowing method, and they are spliced ​​back to the original shape, and the sum of the weights of the spliced ​​data at each position is guaranteed to be 1. The present invention extends the triangular window to three-dimensional space, making it suitable for splicing three-dimensional data. Please refer to Figure 6 for a schematic diagram of the three-dimensional window function. The expression of the window function is as follows:

[0106]

[0107] Among them, the length of the window function is 2N, i, j, k are the indexes of the three dimensions respectively, 0≤i,j,k≤2N,

[0108]

[0109] By adopting the design of overlapping sampling and data windowing technology, the final output of high-resolution data by three-dimensional seismic data processing is guaranteed.

[0110] S2. A 3D Cycle-GAN network architecture is proposed to extract features of shallow high-resolution seismic data;

[0111] See also Figure 1, a schematic diagram of the three-dimensional seismic data network training process proposed in the present invention. Due to the unpaired characteristics of deep and shallow data, it is difficult to extract paired labels for training using strong supervision methods. The present invention uses weak supervision learning methods to learn the distribution and mapping relationship between deep and shallow data. In order to improve the resolution of deep data based on the resolution of shallow data, this chapter uses Cycle-GAN as the backbone network and uses unpaired deep and shallow data for training. Since there is a complex corresponding relationship between the shallow and deep layers in the data collected at the same time, in order to solve the problem of neural network learning difficulties, the present invention introduces bidirectional cycle consistency. Because the network performance is unstable when only the forward cycle is used, especially when facing certain complex data sets, HR outputs with misleading features are often generated. After the introduction of the bidirectional loop, the schematic diagram of the network training process is as follows Figure 1 As shown in FIG, the specific training process includes a forward loop (LR→HR→LR) and a reverse loop (HR→LR→HR).

[0112] Such improvements enable the network in this chapter to more comprehensively understand the subtle correlations in seismic data. The bidirectional cycle uses two sets of generators (forward generator G LR2HR and the reverse generator G HR2LR ) and two sets of discriminators (LR discriminator D LR and HR discriminator D HR ), the forward and reverse networks of both keep the same architecture, share the same architecture and are trained jointly, but their parameters are updated independently.

[0113] Among them, the generator G LR2HR Map the LR data to HR data, and the generator G HR2LR Map HR data to LR data; Discriminator D LR and HR are used to determine whether the input data is real LR or HR data. The consistent architecture of the forward and reverse components ensures a unified learning process. The bidirectional loop significantly improves the stability of training, reduces the artifacts of the output data, and improves the overall quality of the generated HR seismic data due to the introduction of additional constraints.

[0114] Specifically, for the forward loop, the forward generator G LR2HR Learn a mapping from deep LR seismic data to shallow HR data. Input For deep LR seismic data, label is the HR seismic data of the corresponding shallow layer. It should be noted that the LR data and HR data used here are both three-dimensional data volumes. LR2HR The output is the high-resolution result of the deep data χ HR . Reverse generator G HR2LR Analog G LR2HRThe inverse process is essentially mapping the HR seismic data back to the original LR data. LR2HR The output of HR Input reverse generator G HR2LR , get the original LR data Approximate results of By using and Consistency constraints are imposed to ensure that G LR2HR High resolution reversibility. HR discriminator D HR To evaluate and improve G LR2HR The generation performance of G LR2HR Generated HR data χ HR and real labeled HR data The probabilities of indicating whether the generated HR data and the target HR data are real label data are estimated respectively, guiding the training of the generator. This adversarial training process forces G LR2HR Synthesize more realistic HR data, and finally enable it to generate HR data closely related to the target HR data, achieving a meaningful resolution improvement of LR seismic data. The reverse cycle is exactly the opposite of the forward cycle. All shallow and deep data in the forward cycle are swapped. The forward generator G LR2HR and the reverse generator G HR2LR Interchange, LR discriminator D LR and HR discriminator D HR Interchange is the reverse cycle process.

[0115] See also Figure 2 , which is the generator part in the 3D Cycle-GAN network architecture proposed in the present invention. For the generator, the data is first input by the first convolution layer and passes through the 16 residual blocks connected end to end in the main part. Each residual block contains two convolution layers and residual connections respectively, and the outer layer of the residual block group has a larger cross-layer connection. Finally, the data passes through several convolution layers and is output through the tanh activation function. In order to prevent the loss of useful information, the generator network does not include the pooling layer; residual blocks are used to enhance the performance of deep networks; batch normalization layers are not used. This is because when the network is deepened, especially when training the GAN network, the BN layer may introduce artifacts and reduce the generalization ability of the network. Removing the BN layer helps to improve the stability and consistency of training.

[0116] See also Figure 3, which is the discriminator part in the 3D Cycle-GAN network architecture proposed in this invention. For the discriminator, the data first enters from the input 3D convolution layer and passes through 5 convolution blocks, of which the first 2 are 3D convolutions and the last 3 are 2D convolutions. Each convolution block contains a 3D convolution layer, a batch normalization layer and a Leaky ReLU activation function (α=0.2);

[0117] The data is then passed through a Sigmoid activation function, which outputs a probability between 0 and 1. Unlike traditional discriminators, the discriminator of the present invention uses a PatchGAN architecture, replacing the last fully connected layer with a convolutional layer. This allows the discriminator to focus on local areas of the data, which is critical for high-resolution processing of seismic data. In addition, this architecture reduces the number of model parameters and enables the discriminator to converge quickly, resulting in data with higher resolution and higher accuracy.

[0118] S3. Introduce a pair of loss functions composed of generator loss and discriminator loss to reduce prediction error and prevent overfitting;

[0119] Loss functions are crucial in deep learning because they determine how to penalize network outputs during weight optimization. Choosing a suitable loss function is very important to reduce prediction errors and prevent overfitting. In order to solve the problem of improving seismic resolution, the present invention designs a pair of loss functions based on the characteristics of seismic data. The loss function can be decomposed into two components: generator loss and discriminator loss.

[0120] 1) Generator Loss

[0121] The forward and reverse generators have the same network architecture and use the same loss function. The generator loss is:

[0122] L Gen =w 1 L GAN +w 2 L cyc +w 3 L idt +w 4 L TV

[0123] Among them, w 1 、w 2 、w 3 and w 4 represents the weight assigned to each loss function, L GAN To combat the loss, L cyc is the cycle consistency loss, L idt is the constant loss, L TVis the total variation loss. Ideally, it is hoped that the individual contributions of the four loss functions to the final loss function are synergistic and work together to improve the resolution of seismic data.

[0124] Adversarial loss is used to ensure that G LR2HR With G HR2LR The output can deceive D as much as possible HR and D LR When χ HR D HR The more likely it is to be classified as real HR data, the smaller the value of the adversarial loss. Using negative log-likelihood as adversarial loss L GAN , which is defined as follows:

[0125]

[0126] Where N is the number of training samples that the network is trained on in one round.

[0127] The cycle consistency loss aims to ensure that the process of the high-resolution forward generator and the reverse generator is reversible, thereby ensuring the consistency of the data before and after passing through the generator; the cycle consistency loss is defined as follows:

[0128]

[0129] Using L 1 loss and the multi-scale structural similarity (MS-SSIM) loss L MS-SSIM A comprehensive method to define the identity loss L idt , as shown below:

[0130]

[0131] in, The HR data y is low-pass filtered to obtain the corresponding degraded data, L 1 The loss is a measure of the difference between two seismic images, given two 3D seismic data and The calculation formula of L1 loss is as follows:

[0132]

[0133] For the 2D seismic sections U and V intercepted from the 3D data, different multi-scale structural self-similarity is combined to calculate MS-SSIM as follows:

[0134]

[0135] Where M represents the maximum scale, l(U,V), c(U,V) and s(U,V) represent the multi-scale self-similarity of U and V in terms of brightness, contrast and structure; α, β and γ are the weights corresponding to each item. The definitions of l(U,V), c(U,V) and s(U,V) are as follows:

[0136]

[0137] Where μ and σ represent the mean and standard deviation respectively; C 1 , C 2 and C 3 Used to stabilize division and prevent the denominator from being zero. When the Inline length of the input data is less than the Crossline length, the MS-SSIM loss is defined as follows:

[0138]

[0139] in, and Represents three-dimensional data and Randomly selected Inline 2D section, r i is the randomly selected Inline index. When the Inline length of the input data is greater than the Crossline length, similarly, a Crossline 2D section is randomly selected. and Calculate according to the same formula.

[0140] For the output data D of the forward generator or the reverse generator, L TV The definition is as follows:

[0141]

[0142] in, and are the gradients of D in three dimensions respectively.

[0143] Following the core idea of ​​GAN, this paper uses the discriminator D HR and D LR To the generator G LR2HR and G HR2LR Conduct adversarial training.

[0144] Specifically, the discriminator’s loss function is used to ensure that it can correctly determine the difference between real and fake data; it outputs 1 when it determines that the input data is real data, and outputs 0 when it determines that it is the output of the generator.

[0145] The discriminator loss is also defined by the negative log-likelihood as follows:

[0146]

[0147] S4. The shallow and deep seismic data and the corresponding shallow low-resolution data obtained after low-pass filtering of the shallow high-resolution data are input into the network for training. After the training is completed, the deep low-resolution data are input into the network, and the output is subjected to AGC inverse transformation, ultimately achieving an improvement in the resolution of deep seismic data.

[0148] After completing data preprocessing and model training, the deep low-resolution three-dimensional seismic data is sent to the network, and the network output is subjected to AGC inverse transformation to improve the resolution of deep seismic data.

[0149] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as systems, methods or program products. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits", "modules" or "platforms".

[0150] In another embodiment of the present invention, a system for improving the resolution of medium-deep seismic data is provided. The system can be used to implement the above-mentioned method for improving the resolution of medium-deep seismic data. Specifically, the system for improving the resolution of medium-deep seismic data includes a data module, a network module, a function module and an improvement module.

[0151] Among them, the data module pre-processes the three-dimensional seismic data of the same work area; combines the automatic gain control function to construct seismic data, and introduces data augmentation methods and data splicing methods to supplement the seismic data;

[0152] Network module, building a 3D Cycle-GAN network architecture to extract shallow high-resolution seismic data features;

[0153] Function module, which introduces a pair of loss functions composed of generator loss and discriminator loss into the 3D Cycle-GAN network;

[0154] In the improvement module, the shallow and deep seismic data as well as the shallow high-resolution data are low-pass filtered to obtain the corresponding shallow low-resolution data, which are input into the 3D Cycle-GAN network for training, and the deep low-resolution data are input into the trained 3D Cycle-GAN network, and the output is subjected to AGC inverse transformation to improve the resolution of deep seismic data.

[0155] In another embodiment of the present invention, a terminal device is provided, the terminal device includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions; the processor described in the embodiment of the present invention can be used for the operation of the method for improving the resolution of medium-deep seismic data, including:

[0156] Preprocess the three-dimensional seismic data of the same work area; construct seismic data in combination with the automatic gain control function, and introduce data augmentation methods and data splicing methods to supplement the seismic data; construct a 3D Cycle-GAN network architecture to extract the characteristics of shallow high-resolution seismic data; introduce a pair of loss functions composed of generator loss and discriminator loss into the 3D Cycle-GAN network; input the shallow and deep seismic data and the corresponding shallow low-resolution data obtained after low-pass filtering of shallow high-resolution data into the 3D Cycle-GAN network for training, input the deep low-resolution data into the trained 3D Cycle-GAN network, and perform AGC inverse transformation on the output to improve the resolution of deep seismic data.

[0157] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device, and can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that more specific examples (non-exhaustive list) of the computer-readable storage medium here include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0158] Computer readable storage media also include data signals propagated in baseband or as part of a carrier wave, which carry readable program codes. Such propagated data signals can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0159] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0160] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for improving the resolution of medium-deep seismic data in the above embodiment; the processor may load and execute the following steps:

[0161] Preprocess the three-dimensional seismic data of the same work area; construct seismic data in combination with the automatic gain control function, and introduce data augmentation methods and data splicing methods to supplement the seismic data; construct a 3D Cycle-GAN network architecture to extract the characteristics of shallow high-resolution seismic data; introduce a pair of loss functions composed of generator loss and discriminator loss into the 3D Cycle-GAN network; input the shallow and deep seismic data and the corresponding shallow low-resolution data obtained after low-pass filtering of shallow high-resolution data into the 3D Cycle-GAN network for training, input the deep low-resolution data into the trained 3D Cycle-GAN network, and perform AGC inverse transformation on the output to improve the resolution of deep seismic data.

[0162] See also Fig.19 , the terminal device is a computer device, and the computer device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the method for improving the resolution of medium-deep seismic data in the embodiment is implemented. To avoid repetition, it is not described one by one here. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the system for improving the resolution of deep seismic data in the embodiment are implemented. To avoid repetition, it is not described one by one here.

[0163] The computer device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Fig.19 This is only an example of the computer device 60 and does not constitute a limitation of the computer device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0164] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, central processing units, graphics processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, data processing logic based on quantum computing, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0165] The memory 62 may be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the computer device 60.

[0166] Furthermore, the memory 62 may include both an internal storage unit of the computer device 60 and an external storage device. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0167] Any reference to memory, database or other media used in the embodiments provided in the present application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0168] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0169] See also Fig. 20 The terminal device 600 is an electronic device, and the electronic device is in the form of a general computing device. The components of the electronic device may include but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0170] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above method section of this specification. For example, the processing unit 610 can perform the following steps: Fig.18 Follow the steps shown in .

[0171] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0172] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0173] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0174] The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 650. Furthermore, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 via a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0175] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings here can usually be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0176] See also Figure 7 In order to evaluate the effectiveness of the proposed method in this experiment for simulating seismic data, the present invention evaluates the results from two perspectives: profile results and multi-channel average amplitude spectrum. To verify the effectiveness and superiority of the network method.

[0177] Please refer to Figure 8, which shows the results of simulating seismic data in the Inline section (Inline = 200), where Figure 8 (a) and Figure 8 (b) are the original data of deep low resolution and corresponding regional high resolution, respectively, Figure 8 (c) is the result of deep low resolution data after spectral whitening, and Figure 8 (d) is the result of deep low resolution data after network high resolution processing. From the comparison of the experimental results of these two groups, it can be seen that the resolution of the deep part of the data is effectively improved after network processing, and the fault information in the original data is effectively retained. Compared with the spectral whitening method, it is closer to the corresponding high-resolution data, which fully demonstrates the effectiveness of the method of the present invention in correctly improving the resolution of seismic data.

[0178] See also Fig. 9 , which shows the result of simulating seismic data in the Crossline section (Crossline = 200), where Fig. 9 (a) and Fig. 9 (b) The original data of deep layer low resolution and corresponding area high resolution, Fig. 9 (c) is the result of spectral whitening of deep low-resolution data. Fig. 9 (d) is the result of the network high-resolution processing of the deep low-resolution data. From the comparison of the experimental results of the two groups, it can be seen that the spectral whitening method recovers the fuzzy phase axis of the fault part here, and the phase axis structure here is almost invisible; and the network output result can more effectively recover the phase axis of the original data compared with the spectral whitening method, which shows that the method of the present invention can more effectively improve the resolution of seismic signals compared with the traditional method.

[0179] See also Fig.10 , which shows the result of simulating seismic data in isochronous slices (t = 450ms), where Fig.10 (a) and Fig.10 (b) The original data of deep layer low resolution and corresponding area high resolution, Fig.10 (c) is the result of spectral whitening of deep low-resolution data. Fig.10(d) is the result of deep low-resolution data after network high-resolution processing. From the comparison of the experimental results of these two groups, it can be seen that the network output result is more similar to the corresponding high-resolution data part compared with the spectral whitening method while fully retaining the fault information. This shows that the method of the present invention can more effectively improve the resolution of seismic signals compared with traditional methods.

[0180] See also Fig.11 , which shows the comparison of multi-channel amplitude spectra of simulated seismic data processed by spectral whitening method and network method, where the solid line is the original low-resolution data, the dashed line is the corresponding high-resolution data, the dotted line is the result of spectral whitening method, and the dot-dash line is the high-resolution result of network processing. It can be seen from the spectrum that both the network method and the spectral whitening method can effectively improve the high-frequency components of the data, and the bandwidth expansion of the two is basically the same. However, compared with the spectral whitening method, the spectrum obtained by processing the data using the network method is closer to the original data in low frequency, especially in the low-frequency part below 20Hz. The network method better retains the low-frequency information of the deep data, while the spectral whitening method has the problem of too high low frequency.

[0181] Compared with synthetic data, real data presents greater complexity and is significantly affected by noise. In order to further evaluate the effectiveness of the high-resolution method proposed in this invention, the spectral whitening method is selected as the benchmark method, and a three-dimensional post-stack seismic data set acquired in a certain work area is used for comparison. The Inline and Crossline sizes of the data are 931 and 2471 respectively, and the time sampling rate of the data is 2ms. The 600×1800 part close to the center with a higher signal-to-noise ratio is selected as the training data, and the entire data is tested after the training is completed.

[0182] See also Fig.12 , the shallow part is selected from the time window of 0.6–1.6 s, while the deep part is selected from the relatively high signal-to-noise ratio region of 1.9–2.9 s. After training, the network is applied to the deep data to enhance the resolution. Fig.12 (a) shows the crossline profile of an actual earthquake. Fig.12 (b) shows the normalized multi-channel amplitude spectra of deep and shallow layers.

[0183] Please refer to Figure 14, which shows the results of the actual seismic data in the Inline section (Inline = 8km). Figure 14 (a) shows the original deep LR data, which is distorted by some noise, resulting in discontinuity. Figure 14 (b) shows the enhanced result after spectral whitening, and Figure 14 (c) shows the result after network processing. The results show that the resolution of the original seismic data has been significantly improved, the reflections are finer, and the geological structure is finer. In addition, it is observed that the two methods have great structural similarities, which confirms the reliability of our method.

[0184] Please refer to Figure 15, which shows the results of the actual seismic data in the Crossline section (Crossline = 6km). Figure 15(a) is the original Crossline section, which shows a relatively wide feature due to the lack of high-frequency energy. Compared with the original section, the spectral whitening technique in Figure 15(b) and our method in Figure 15(c) have obvious resolution improvement and effectively recover valuable high-frequency information. However, spectral whitening has difficulty in maintaining spatial continuity and energy consistency, resulting in increased noise. In contrast, our method significantly enhances lateral continuity and improves spatial energy consistency. Even under noisy conditions, the network can still maintain excellent enhancement performance, demonstrating its strong anti-interference ability to noise, as shown. This enhancement is crucial for reliable stratigraphic tracking and detailed stratigraphic interpretation.

[0185] Please refer to Figure 13, which is a normalized amplitude spectrum, wherein Figure 13(a) is the normalized amplitude spectrum of the Inline section in Figure 14, and Figure 13(b) is the normalized amplitude spectrum of the Crossline section in Figure 15. They represent the original LR data, the HR result after spectral whitening, and the method of the present invention, respectively. Both our method and spectral whitening effectively enhance the high-frequency components and achieve considerable bandwidth expansion. They are closely combined with the original LR data, retaining the inherent characteristics while improving the resolution. However, compared with the spectral whitening method, the data processed by this method corresponds better to the original data. Spectral whitening shows a significant deviation from the original spectrum when it exceeds 20Hz, and cannot follow the original frequency conversion at around 35Hz, and loses signal integrity in the mid-frequency range, highlighting the advantages of the method of the present invention in smoothly extending the bandwidth, accurately enhancing the high-frequency components, and maintaining the overall data integrity, making it more effective in improving the resolution of seismic data.

[0186] Please refer to Figures 16 and 17, which show the results of the actual seismic data at isochronous slices (t=2284ms) and (t=2504ms), respectively. The ability of the method of the present invention to improve resolution is further emphasized. After network processing, the improvement of deep time slices is more obvious than spectral whitening. As shown in the box, the method of the present invention not only achieves a higher clarity in depicting geological structures, but also reveals clearer details that cannot be captured by spectral whitening. The spatial relationship between different strata can be more accurately depicted, making the geological interface clearer. This detailed representation not only helps to interpret the underground geology more accurately, but also greatly enhances our understanding of complex stratigraphic relationships, thereby providing an important tool for advancing seismic exploration and analysis.

[0187] In summary, the present invention provides a method and system for improving the resolution of medium-deep seismic data, designs a three-dimensional cyclic generative adversarial network structure, and a corresponding loss function, data augmentation and data splicing method, to achieve the purpose of improving the resolution of deep seismic data. The feasibility and effectiveness of the method of the present invention are also demonstrated in experiments with simulated seismic data and actual seismic data. The results obtained are compared with the traditional spectral whitening method, and the continuity and high-frequency details of the event axis are improved.

[0188] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0189] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0190] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0191] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0192] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0193] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0194] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0195] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0196] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0198] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for improving the resolution of medium-deep seismic data, characterized in that: The following steps are involved: S1. Preprocess the 3D seismic data of the same work area; construct seismic data by combining the automatic gain control function, and introduce data augmentation method and data splicing method to supplement the seismic data; S2, build a 3D Cycle-GAN network architecture to extract shallow high-resolution seismic data features; S3, introducing a pair of loss functions composed of a generator loss and a discriminator loss into the 3D Cycle-GAN network obtained in step S2; S4, input the shallow and deep seismic data obtained in step S1 and the shallow high-resolution data obtained after low-pass filtering to the corresponding shallow low-resolution data obtained in step S3 into the 3D Cycle-GAN network for training, input the deep low-resolution data into the trained 3D Cycle-GAN network, and perform AGC inverse transformation on the output to improve the resolution of deep seismic data.

2. The method for improving the resolution of mid-deep seismic data according to claim 1, characterized in that: In step S1, for the input earthquake data D in Each of the i , the result d′ after the signal is preprocessed using the automatic gain control algorithm i as follows: Among them, W triangle (L) is a triangular window with a length of L.

3. The method for improving the resolution of mid-deep seismic data according to claim 2, characterized in that: The data augmentation method is as follows: Flip the Inline and Crossline spatial dimensions, swap the Inline and Crossline spatial dimensions, reverse the data as a whole, and get the augmented data Each curly bracket indicates that one of the items is randomly selected with equal probability.

4. The method for improving the resolution of mid-deep seismic data according to claim 2, characterized in that: The specific data splicing method is: When dividing the input data into blocks, set a 50% overlap on the data; After high-resolution processing is completed, the overlapping parts between high-resolution data blocks are multiplied by the set weights through the windowing method, and then spliced ​​back to the original shape to ensure that the sum of the weights of the spliced ​​data at each position is 1; Expand the triangular window to three-dimensional space, and the expression of the window function is as follows: Among them, the length of the window function is 2N, i, j, k are indexes of three dimensions respectively, 0≤i,j,k≤2N.

5. The method for improving the resolution of mid-deep seismic data according to claim 1, characterized in that: In step S2, the 3D Cycle-GAN network architecture includes a forward loop and a reverse loop. The forward loop and the reverse loop use two sets of generators and two sets of discriminators. The forward and reverse networks of the two sets of generators and discriminators maintain the same architecture, share the same architecture and are jointly trained, and the parameters are updated independently. The generator includes a forward generator G LR2HR and the reverse generator G HR2LR , the forward generator G HR2LR Map LR data to HR data, the reverse generator G LR2HR Map HR data to LR data; The discriminator includes LR discriminator D LR and HR discriminator D HR , LR discriminator D LR and HR discriminator D HR They are used to determine whether the input data is real LR or HR data.

6. The method for improving the resolution of mid-deep seismic data according to claim 5, characterized in that: Forward generator G LR2HR and the reverse generator G HR2LR It is a three-dimensional network that processes two-dimensional data and three-dimensional data at the same time and uses the same network architecture; The data is first input by the first convolutional layer and then passes through 16 residual blocks connected end to end in the main part. Each residual block contains two convolutional layers and residual connections. The outer layer of the residual block group has a cross-layer connection. Finally, the data goes through multiple convolutional layers and is output through the tanh activation function.

7. The method for improving the resolution of mid-deep seismic data according to claim 6, characterized in that: No pooling layers are included in the generator network; residual blocks are used and no batch normalization layers are used.

8. The method for improving the resolution of mid-deep seismic data according to claim 5, characterized in that: LR Discriminator D LR and HR discriminator D HR It is a two-dimensional network, using the same network architecture; The data first enters the input convolution layer and passes through 5 convolution blocks; Each convolutional block contains a convolutional layer, a batch normalization layer, and a Leaky ReLU activation function; The data then passes through the Sigmoid activation function and outputs a probability between 0 and 1; The discriminator adopts the PatchGAN architecture and replaces the last fully connected layer with a convolutional layer.

9. The method for improving the resolution of mid-deep seismic data according to claim 1, characterized in that: In step S3, the loss function is decomposed into generator loss and discriminator loss; The generator loss is: L Gen =w1L GAN +w2L cyc +w3L idt +w4L TV Among them, w1, w2, w3 and w4 represent the weights assigned to each loss function, L GAN To combat the loss, L cyc is the cycle consistency loss, L idt is the constant loss, L TV is the total variation loss; The discriminator loss is: Among them, N is the number of training samples in one round of network training, D HR is the HR discriminator, D LR is the LR discriminator, G LR2HR (x i ) is the low-resolution data χ, which is the high-resolution data obtained by the input forward generator, G HR2LR (y i ) is the low-resolution data obtained by inputting the high-resolution data y into the reverse generator.

10. A system for improving the resolution of medium-deep seismic data, characterized in that: include: Data module, pre-processing the 3D seismic data of the same work area; Combined with the automatic gain control function, seismic data is constructed, and data augmentation and data splicing methods are introduced to supplement the seismic data; Network module, building a 3D Cycle-GAN network architecture to extract features of shallow high-resolution seismic data; Function module, which introduces a pair of loss functions composed of generator loss and discriminator loss into the 3D Cycle-GAN network; In the improvement module, the shallow and deep seismic data as well as the shallow high-resolution data are low-pass filtered to obtain the corresponding shallow low-resolution data, which are input into the 3D Cycle-GAN network for training, and the deep low-resolution data are input into the trained 3D Cycle-GAN network, and the output is subjected to AGC inverse transformation to improve the resolution of deep seismic data.

Images