Wavelet domain multi-scale seismic data missing region self-adaptive intelligent repairing method and device

Abstract

The embodiment of the invention discloses a wavelet domain multi-scale seismic data missing region adaptive intelligent restoration method and device, and the method comprises the steps: obtaining to-be-restored seismic data; inputting the seismic data into a pre-trained first generative network to obtain restored seismic data; the first generative network comprises an encoder, a residual module and a decoder, and wavelet convolution modules are arranged in the encoder and the decoder. The generator network is designed into a network structure embedded with wavelet convolution, so that the smoothness and the accuracy of an interpolation result are enhanced, and details are reserved while the reconstruction precision of low-frequency information is improved.

Description

Technical Field

[0001] The present invention relates to the field of seismic data processing, specifically to the field of multi-scale seismic data restoration in the wavelet domain, and particularly to a method and device for adaptively and intelligently restoring missing areas of multi-scale seismic data in the wavelet domain. Background Art

[0002] In actual seismic data acquisition, ideally, the acquired data should be of high quality, regular, and dense to support subsequent imaging, interpretation, and modeling tasks. However, affected by economic costs, environmental conditions, and equipment limitations, the acquired seismic data usually appears sparse, irregular, or missing. This not only affects the usability of seismic data but also poses challenges to data processing and geological interpretation. Summary of the Invention

[0003] The present invention provides a method and device for adaptively and intelligently restoring missing areas of multi-scale seismic data in the wavelet domain, which can optimize the acquired original seismic data and restore the seismic data after optimization to obtain high-quality seismic data.

[0004] On the one hand, the present invention provides a method for adaptively and intelligently restoring missing areas of multi-scale seismic data in the wavelet domain, which is applied to a system for adaptively and intelligently restoring missing areas of multi-scale seismic data in the wavelet domain. The method includes:

[0005] Obtain the seismic data to be restored;

[0006] Input the seismic data into a pre-trained first generation network to obtain the restored seismic data;

[0007] The first generation network includes an encoder, a residual module, and a decoder:

[0008] Among them, wavelet convolution is used in the encoder and decoder to restore and extract data of a multi-level complex architecture;

[0009] Among them, the training method of the first generation network is as follows:

[0010] Obtain training data, where the training data includes missing seismic data and its corresponding complete seismic data;

[0011] Input the missing seismic data in the training data into a generator to generate first generation data; calculate the loss of a discriminator using the first generation data and the complete seismic data and update the parameters of the discriminator;

[0012] Calculate the loss of the generator and update the parameters of the generator until the model reaches a preset performance index;

[0013] Use the trained generator as the first generation network.

[0014] Through wavelet transform, local analysis of signals can be performed at different scales, which is beneficial to extracting seismic signals at different frequency levels. After introducing wavelet convolution into the model, detailed features can be extracted at different scales, which helps to improve the fine-grained recovery of seismic data interpolation and its performance in multi-level complex structures. Therefore, for data such as seismic signals with multi-scale complex patterns, designing the generator network into a U-Net structure embedded with wavelet convolution helps to enhance the smoothness and accuracy of the interpolation results, improves the reconstruction accuracy of low-frequency information while retaining details.

[0015] In a possible implementation, the first generation network specifically includes four downsampling modules in the encoder and four upsampling modules in the decoder, using wavelet convolutional layers, specifically adopting the db2 wavelet basis, and setting the decomposition level to 2.

[0016] In a possible implementation, the implementation of the four downsampling modules in the encoder is as follows: at each layer of downsampling, 2-level decomposition is performed using the db2 wavelet basis to obtain low-frequency and high-frequency feature maps, and feature maps with 1024 channels are calculated layer by layer; the residual module uses 3x3 convolution; the implementation of the four upsampling modules in the decoder is as follows: at each layer of upsampling, upsampling is performed using the db2 wavelet basis, and the data image is restored to the same spatial resolution as the original image layer by layer.

[0017] The network of the present invention adopts the db2 wavelet basis and sets the decomposition level to 2. The db2 wavelet basis (Daubechies 2) can better extract the details and overall trends of seismic signals through its balanced low-frequency and high-frequency characteristics. Compared with higher-order wavelet bases (such as db4 or db6), db2 has a lower computational complexity and is suitable for processing large-scale seismic data. The two-level decomposition helps to extract multi-scale features, enhance the noise suppression ability of the model, and provides a balance between computational efficiency and feature extraction ability, enabling the generator to more effectively capture key features when processing complex seismic data.

[0018] In a possible implementation, the discriminator includes a convolutional layer and seven spectrally-normalized convolutional blocks connected in sequence; each convolutional block includes a convolutional layer, an SN layer, and a Leaky ReLU activation function; the kernel size of each convolution is 3×3, the number of feature maps increases from 64 to 512, and convolution with a stride of 2 is used.

[0019] The seismic data of the present invention sequentially passes through a convolutional layer and seven spectral normalization convolutional blocks (Conv SD-Block). Each convolutional block includes a convolutional layer, an SN layer, and a LeakyReLU activation function. The kernel size of each convolution is 3×3, the number of feature maps increases from 64 to 512, and convolution with a stride of 2 is used to enable the network to focus on a larger range of data and extract more global features; it should be noted that the SN layer is used to replace the BN layer. By restricting the spectral norm of the weights of each layer, the discriminator can be made more sensitive to data details, overfitting can be reduced, and the learning effect of local features can be improved.

[0020] In a possible implementation manner, the loss function in the training method of the first generation network includes:

[0021] l G = l rec + αl per + βl adv + λ ltv

[0022] where l G is the loss function of the generator, and α, β, and λ are the weights corresponding to each loss term, which are used to balance the influence of the loss on training; l rec represents the reconstruction loss calculated using the mean square error MSE, which is used to measure the pixel difference between the generated data and the target data. What is calculated is the MSE of the generated data and its corresponding complete data d R and is expressed as:

[0023]

[0024] W and H represent the dimensions of the input missing seismic data d M and t represents the scale of the resolution improvement from d M to d R ; is the corresponding complete data;

[0025] The perceptual loss l per is a loss function that measures the perceptual quality difference between the generated data and the target data and is expressed as:

[0026]

[0027] where φ i，j represents the operation of extracting the feature map of the jth convolution before the ith max-pooling layer, and W i，j and H i，j represent the dimensions of each feature map within the VGG network;

[0028] l adv is the reconstruction loss; ltv is the total variation loss;

[0029] The reconstruction loss is expressed as:

[0030]

[0031] where N represents the number of samples of the missing data d M ; denotes the probability that the discriminator regards the reconstructed data as real complete seismic data; To obtain better gradient behavior, minimize instead of

[0032] the total variation loss l tv is expressed as:

[0033]

[0034] where M is the total number of pixels of the data image for normalization; X is the input data or feature map; (i, j) represents the pixel coordinates; The first term (X i+1，j - X i，j ) 2 calculates the pixel change amount in the vertical direction; The second term (X i，j+1 - X i，j ) 2 calculates the pixel change amount in the horizontal direction.

[0035] It should be noted that the loss function of the generator is crucial for the quality of the generated data. Using only pixel-level losses (such as L1 loss) in seismic data interpolation often results in blurred and over-smoothed results, which will lead to inaccuracies in subsequent imaging. Therefore, the VGG loss is integrated as a perceptual loss into the loss function of the generator. The feature maps extracted by the pre-trained 16-layer VGG network are used to calculate the loss. It maps the seismic data to a higher-dimensional feature space and compares the distance between the generated data and the target data in this space. This process effectively utilizes the high-level features of the VGG network in the feature representation space to more precisely evaluate the structural similarity between the reconstructed data and the real data. Therefore, the VGG loss can encourage the model to generate reconstruction results with richer details and higher perceptual quality during the optimization process. The total variation loss l tv , is a loss function used to measure the smoothness of data or feature maps, mainly used to reduce noise and artifacts in the data while maintaining the smoothness and structural continuity of the data. When the total variation loss is used in a generative adversarial network, it can reduce the clutter and irregular textures of the generated data and improve the data quality.

[0036] On the other hand, an embodiment of the present invention provides a wavelet domain multi-scale seismic data missing area adaptive intelligent repair device.

[0037] The seismic data processing device executes a wavelet domain multi-scale seismic data missing area adaptive intelligent repair method, and the steps include: acquiring seismic data to be repaired; inputting the seismic data into a pre-trained first generation network to obtain repaired seismic data; the first generation network includes an encoder, a residual module, and a decoder: among them, wavelet convolution modules are provided in the encoder and the decoder; among them, the training method of the first generation network is: acquiring training data, the training data includes missing seismic data and its corresponding complete seismic data; inputting the missing seismic data in the training data into a generator to generate first generated data; calculating the loss of a discriminator using the first generated data and the complete seismic data and updating the parameters of the discriminator; calculating the loss of the generator through the discriminator and updating the parameters of the generator until the model reaches a preset performance index; using the trained generator as the first generation network.

[0038] On the other hand, the present invention provides an electronic device, including: a processor, adapted to implement one or more instructions; and a computer storage medium, the computer storage medium stores one or more instructions, and the one or more instructions are adapted to be loaded and executed by the processor to perform the above-mentioned wavelet domain multi-scale seismic data missing area adaptive intelligent repair method.

[0039] On the other hand, the present invention provides a computer storage medium, characterized in that the computer storage medium stores one or more instructions, and the one or more instructions are adapted to be loaded and executed by a processor to perform the above-mentioned wavelet domain multi-scale seismic data missing area adaptive intelligent repair method.

[0040] It can be seen that the present invention uses wavelet convolution in the network to extract detailed features at different scales, which helps in the performance of seismic data interpolation that requires fine-grained restoration and multi-level complex structures. Specifically, for the characteristics of seismic data, the db2 wavelet basis is adopted, and the decomposition level is set to 2, which can better extract the details and overall trends of seismic signals. Compared with higher-order wavelet bases (such as db4 or db6), db2 has a lower computational complexity and is suitable for processing large-scale seismic data. The two-level decomposition helps to extract multi-scale features, enhance the noise suppression ability of the model, and provides a balance between computational efficiency and feature extraction ability, enabling the generator to more effectively capture key features when processing complex seismic data;

[0041] Furthermore, in the present invention, a spectral normalization convolutional block is used. By restricting the spectral norm of each layer of weights, the discriminator can be made more sensitive to data details, reduce overfitting, and improve the learning effect of local features.

[0042] In addition, in the setting of the loss function, in view of the problem that using only pixel-level losses (such as L1 loss) in seismic data interpolation often results in blurred and over-smoothed results, the VGG loss is integrated as a perceptual loss into the loss function of the generator, effectively utilizing the high-level features of the VGG network in the feature representation space to more precisely evaluate the structural similarity between the reconstructed data and the real data; when the total variation loss is used in the generative adversarial network, it can reduce the clutter and irregular textures of the generated data and improve the data quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 It is a schematic structural diagram of the generator of the WTCGAN network designed by the present invention provided by the embodiments of the present invention;

[0045] Figure 2 It is a schematic flowchart of an adaptive intelligent repair method for missing regions of multi-scale seismic data in the wavelet domain provided by the embodiments of the present invention;

[0046] Figure 3 It is a schematic structural diagram of a discriminator network provided by the embodiments of the present invention;

[0047] Figure 4 It is a schematic diagram of seismic data provided by the embodiments of the present invention; Figure 4 (a) is ideal seismic data; Figure 4 (b) is seismic data with 60% randomly missing; Figure 4 (c) is seismic data with 70% randomly missing;

[0048] Figure 5 It is a schematic F-K spectrum diagram of seismic data provided by the embodiments of the present invention; Figure 5 (a)-(c) are respectively Figure 4 the F-K spectra in (a)-(c);

[0049] Figure 6 It is the reconstruction result of 60% random missing provided by the embodiments of the present invention; Figure 6 (a) is the reconstruction result of the traditional U-Net, SNR = 16.34 dB; Figure 6 (b) is the reconstruction result through U-NetGAN, SNR = 23.07 dB; Figure 6(c) is the reconstruction result by WTCGAN, SNR = 29.11dB;

[0050] Figure 7 is the reconstruction result of 70% random missing provided by the embodiment of the present invention; Figure 7 (a) is the reconstruction result of traditional U-Net, SNR = 12.60dB; Figure 7 (b) is the reconstruction result by U-NetGAN, SNR = 21.04dB; Figure 7 (c) is the reconstruction result by WTCGAN, SNR = 28.18dB.

[0051] Figure 8 is a schematic structural diagram of an electronic device provided by the embodiment of the present invention. Detailed implementation manners

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0053] It should be noted that the terms "first", "second", etc. in the description and claims of the embodiments of the present invention and the above accompanying drawings are used to distinguish different objects, rather than to describe a specific order.

[0054] Based on the above description, an adaptive intelligent repair method for missing regions of wavelet-domain multi-scale seismic data is proposed in the embodiment of the present invention, which is applied to an adaptive intelligent repair system for missing regions of wavelet-domain multi-scale seismic data.

[0055] The method includes:

[0056] S201, obtaining the seismic data to be repaired;

[0057] S202, inputting the seismic data into a pre-trained first generation network to obtain the repaired seismic data;

[0058] The first generation network includes an encoder, a residual module, and a decoder:

[0059] Among them, wavelet convolution is used in the encoder and decoder to recover and extract data of a multi-level complex architecture; the schematic diagram of the first generation network is as Figure 1 shown.

[0060] Among them, the training method of the first generation network is:

[0061] Obtaining training data, where the training data includes missing seismic data and its corresponding complete seismic data;

[0062] Input the missing seismic data in the training data into a generator to generate first generated data; calculate the loss of a discriminator using the first generated data and the complete seismic data and update the parameters of the discriminator;

[0063] Calculate the loss of the generator and update the parameters of the generator until the model reaches a preset performance metric;

[0064] Use the trained generator as a first generation network.

[0065] Through wavelet transform, local analysis of a signal can be performed at different scales, which is particularly suitable for seismic signals containing different frequency levels; after introducing wavelet convolution into the model, detailed features can be extracted at different scales, which helps in the performance of seismic data interpolation that requires fine-grained restoration and multi-level complex structures. Therefore, for data such as seismic signals with multi-scale complex patterns, designing the generator network into a UNet structure embedded with wavelet convolution helps to enhance the smoothness and accuracy of the interpolation results, improve the reconstruction accuracy of low-frequency information while retaining details. The introduction of a residual module further enhances the non-linear modeling ability, making the interpolation results more consistent in high-frequency and low-frequency features, and avoiding false oscillations or waveform distortions.

[0066] In a possible implementation, the first generation network specifically includes four downsampling modules of an encoder and four upsampling modules of a decoder, uses wavelet convolutional layers, and specifically adopts a db2 wavelet basis with the decomposition level set to 2.

[0067] In a possible implementation, the implementation of the four downsampling modules of the encoder is as follows: at each layer of downsampling, perform 2-level decomposition using the db2 wavelet basis to obtain low-frequency and high-frequency feature maps, and layer by layer calculate to obtain a feature map with 1024 channels; the residual module uses 3x3 convolution; the implementation of the four upsampling modules of the decoder is as follows: at each layer of upsampling, perform upsampling using the db2 wavelet basis, and layer by layer calculate to restore the data image to the same spatial resolution as the original image.

[0068] The network of the present invention adopts a db2 wavelet basis with the decomposition level set to 2. The db2 wavelet basis (Daubechies 2) can better extract the details and overall trends of seismic signals through its balanced low-frequency and high-frequency characteristics. Compared with higher-order wavelet bases (such as db4 or db6), db2 has a lower computational complexity and is suitable for processing large-scale seismic data. The two-level decomposition helps to extract multi-scale features, enhance the noise suppression ability of the model, and provides a balance between computational efficiency and feature extraction ability, enabling the generator to more effectively capture key features when processing complex seismic data.

[0069] In a possible implementation, the discriminator includes a convolutional layer and seven spectral normalization convolutional blocks connected in sequence; each convolutional block includes a convolutional layer, an SN layer, and a Leaky ReLU activation function; the kernel size of each convolution is 3×3, the number of feature maps increases from 64 to 512, and convolution with a stride of 2 is used to enable the network to focus on a larger range of data.

[0070] The seismic data of the present invention sequentially passes through a convolutional layer and seven spectral normalization convolutional blocks (Conv SD-Block). Each convolutional block includes a convolutional layer, an SN layer, and a Leaky ReLU activation function. The kernel size of each convolution is 3×3, the number of feature maps increases from 64 to 512, and convolution with a stride of 2 is used to enable the network to focus on a larger range of data and extract more global features; it should be noted that the SN layer is used to replace the BN layer, and by restricting the spectral norm of the weights of each layer, the discriminator can be made more sensitive to data details, reduce overfitting, and improve the learning effect of local features. The network structure of the discriminator is as Figure 3 shown.

[0071] In a possible implementation, the loss function in the training method of the first generation network includes: lG = lrec + αlper + βladv + λltv

[0072] l G = l rec + αl per + βl adv + λl tv

[0073] where l G is the loss function of the generator, and α, β, and λ are the weights corresponding to each loss term, used to balance the influence of the loss on training; l rec represents the reconstruction loss calculated using the mean squared error MSE, used to measure the pixel difference between the generated data and the target data, and calculates the MSE of the generated data and its corresponding complete data d R , expressed as:

[0074]

[0075] W and H represent the dimensions of the input missing seismic data d M , and t represents the scale of the resolution improvement from d M to d R ; is the corresponding complete data;

[0076] The perceptual loss l per is a loss function that measures the perceptual quality difference between the generated data and the target data, expressed as:

[0077]

[0078] where φ i,j represents the operation of extracting the feature map of the j-th convolution before the i-th max-pooling layer, and W i,j and H i，j represent the dimensions of the respective feature maps within the VGG network;

[0079] l adv is the reconstruction loss; l tv is the total variation loss;

[0080] The reconstruction loss is expressed as:

[0081]

[0082] where N represents the number of samples of the missing data d M , represents the probability that the discriminator regards the reconstructed data as real complete seismic data; to obtain better gradient behavior, minimize instead of

[0083] The total variation loss l tv is expressed as:

[0084]

[0085] where M is the total number of pixels of the data image for normalization; X is the input data or feature map; (i, j) represents the pixel coordinates; the first term (X i+1，j -X i，j ) 2 calculates the pixel change amount in the vertical direction; the second term (X i，j+1 -X i，j ) 2 calculates the pixel change amount in the horizontal direction.

[0086] It should be noted that the loss function of the generator is crucial for the quality of the generated data. In seismic data interpolation, using only pixel-level losses (such as L1 loss) often results in blurred and over-smoothed results, which will lead to inaccuracies in subsequent imaging. Therefore, the VGG loss is integrated as a perceptual loss into the loss function of the generator. We use the feature maps extracted by a pre-trained 16-layer VGG network to calculate the loss. It maps the seismic data into a higher-dimensional feature space and compares the distances between the generated data and the target data in this space. This process effectively utilizes the high-level features of the VGG network in the feature representation space to more precisely evaluate the structural similarity between the reconstructed data and the real data. Therefore, the VGG loss can encourage the model to generate reconstructed results with richer details and higher perceptual quality during the optimization process.

[0087] l adv is defined based on the probability of recognizing the reconstructed data as real data. To deceive the discriminator, the generator learns the distribution of real, complete, and non-missing seismic data as much as possible.

[0088] Finally, the total variation loss l tv , is a loss function used to measure the smoothness of data or feature maps, mainly used to reduce noise and artifacts in the data while maintaining the smoothness and structural continuity of the data. When used in a generative adversarial network, the total variation loss can reduce clutter and irregular textures in the generated data and improve data quality.

[0089] In a possible implementation, signal-to-noise ratio (SNR) and structural similarity (SSIM) metrics are used to evaluate the quality of the interpolated seismic data. The calculation formulas for SNR and SSIM are as follows:

[0090] SNR(dB) = 10lgy 2 / (y - G(x)) 2

[0091]

[0092] where y and G(x) represent the ideal and restored seismic data respectively; μ y and μ G(x) represent the average values of the data image pixels of y and G(x) respectively; σ y and σ G(x) represent the variances of the data image pixels of y and G(x) respectively; σ yG(x) represents the covariance of the data image pixels of y and G(x); c1 and c2 represent the stable constants for calculation.

[0093] The experiment used relevant datasets and selected seismic data of 80 shots. The dataset was divided into a training set of 64 shots, a validation set of 8 shots, and a test set of 8 shots according to the ratios of 80%, 10%, and 10%. The random seed was fixed to randomly miss 60% of the data to test the interpolation ability of the model. To test the interpolation ability and robustness of the model under more demanding conditions with more severe data missing, 70% random missing was also performed on the data. Then, the training set and the validation set were split into seismic data of size 256×256. The actual dataset used in the experiment was an ocean acquisition dataset. 120 shots of data were randomly selected from the dataset. Similarly, the dataset was divided into a training set of 96 shots, a validation set of 12 shots, and a test set of 12 shots according to the ratios of 80%, 10%, and 10%. The random seed was fixed to perform 60% and 70% missing on the data. Then, the dataset was split into data blocks of 512×120. The batch size of the experiment was set to 8, the number of training epochs was set to 100, and the learning rates of the generator and the discriminator were set to 0.0004 and 0.0008 respectively to ensure that the generator could be gradually optimized during training while reducing the risk of training instability and mode collapse. In the loss function, the adversarial loss weight coefficient of the generator was 0.05, the perceptual loss weight coefficient was 0.04, and the total variational loss weight coefficient was 0.001. In addition, the Adam optimizer was used for parameter update during the training process. The input and output sizes of the network were 256×256 slice data. The experiment was implemented based on the PyTorch deep learning framework and trained using an RTX 4060Ti GPU. To evaluate the performance of the proposed method, we designed two groups of experiments for two cases of randomly missing 60% and 70% of the seismic data. On this basis, we used the UNET and UNETGAN networks to perform the same interpolation task for comparative analysis.

[0094] This application conducted comparative experiments on the UNET-based method, UNETGAN, and the WTCGAN of this application on complex synthetic seismic records. To simulate data missing, Figure 4 60% of the traces in (a) were randomly deleted, as shown in Figure 4 (b), while the more severe case of 70% random missing is shown in Figure 4 (c). Further analysis of the impact of random missing on the frequency-domain characteristics Figure 5 shows the Figure 4 (a), Figure 4 (b), and Figure 4 (c) corresponding F-K spectra. Figure 5 (a) is the F-K spectrum of the complete seismic record, reflecting the frequency characteristics and energy distribution of the original data. Figure 5 (b) is the F-K spectrum of the 60% randomly missing data. It can be observed that significant frequency information loss and spectral structure damage are caused by the missing traces. Figure 5(c) The F-K spectrum with 70% randomly missing data shows more serious information loss in the frequency domain compared to 60% random missing, especially in the high-frequency part, exhibiting more obvious energy attenuation and aliasing phenomena. Figure 6 The reconstruction results with 60% random missing are shown: Figure 6 (a) is the result of traditional U-Net reconstruction, SNR = 16.34 dB; Figure 6 (b) is the result of reconstruction by U-NetGAN, SNR = 23.07 dB; Figure 6 (c) is the result of reconstruction by the proposed method of this application, SNR = 29.11 dB. Figure 7 The reconstruction results with 70% random missing are shown: Figure 7 (a) is the result of traditional U-Net reconstruction, SNR = 12.60 dB; Figure 7 (b) is the result of reconstruction by U-NetGAN, SNR = 21.04 dB; Figure 7 (c) is the result of reconstruction by WTCGAN, SNR = 28.18 dB. It can be seen that the method of this application exhibits superior reconstruction characteristics.

[0095] On the other hand, the embodiment of the present invention provides a wavelet-domain multi-scale seismic data missing area adaptive intelligent repair device.

[0096] The wavelet-domain multi-scale seismic data missing area adaptive intelligent repair system executes the wavelet-domain multi-scale seismic data missing area adaptive intelligent repair method, and the steps include:

[0097] Obtain the seismic data to be repaired;

[0098] Input the seismic data into the pre-trained first generation network to obtain the repaired seismic data;

[0099] The first generation network includes an encoder, a residual module, and a decoder:

[0100] Among them, wavelet convolution modules are set in the encoder and the decoder;

[0101] Among them, the training method of the first generation network is:

[0102] Obtain training data, and the training data includes missing seismic data and its corresponding complete seismic data;

[0103] Input the missing seismic data in the training data into the generator to generate the first generated data; calculate the loss of the discriminator using the first generated data and the complete seismic data and update the parameters of the discriminator;

[0104] Calculate the loss of the generator through the discriminator and update the parameters of the generator until the model reaches a preset performance index;

[0105] Use the trained generator as the first generation network.

[0106] An embodiment of the present invention also provides a computer storage medium (Memory). The computer storage medium is a memory device in an electronic device for storing programs and data. It can be understood that the computer storage medium here can include both the built-in storage medium in the electronic device and, of course, the extended storage medium supported by the electronic device. The computer storage medium provides a storage space that stores the operating system of the electronic device. And, one or more instructions suitable for being loaded and executed by the processor are stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory; optionally, it can also be at least one computer storage medium located far from the aforementioned processor.

[0107] Figure 8 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. The electronic device includes: at least one processor 301, such as a central processing unit (CPU), at least one memory 302, and at least one bus 303.

[0108] Program instructions can be stored in the aforementioned memory 302, and the aforementioned processor 301 can be used to call the program instructions to execute a method for adaptively and intelligently repairing missing areas of multi-scale seismic data in the wavelet domain.

[0109] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and this program can be stored in a computer-readable storage medium. The storage medium includes read only memory (ROM), random access memory (RAM), programmable read only memory (PROM), erasable programmable read only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), solid state disk (SSD), or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium capable of carrying or storing data.

[0110] It should be noted that all steps of the embodiments of this application are executed under legal and compliant conditions, that is, all steps of the embodiments of this application are executed with authorization.

[0111] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for adaptive intelligent repair of missing regions of multi-scale seismic data in wavelet domain, characterized in that: Applied to a wavelet domain multi-scale seismic data missing area adaptive intelligent repair system, the method comprises: Obtaining seismic data to be restored; Inputting the seismic data into a pre-trained first generation network to obtain repaired seismic data; The first generation network includes an encoder, a residual module, and a decoder: Wherein, a wavelet convolution module is provided in the encoder and the decoder; Among them, the training method of the first generation network is: Acquire training data, wherein the training data includes missing seismic data and corresponding complete seismic data; Inputting the missing seismic data in the training data into the generator to generate first generated data; using the first generated data and the complete seismic data to calculate the loss of the discriminator and update the parameters of the discriminator; Calculating the loss of the generator through the discriminator and updating the parameters of the generator until the model reaches a preset performance indicator; The trained generator is used as the first generation network.

2. The adaptive intelligent repair method for missing region of multi-scale seismic data in wavelet domain according to claim 1, characterized in that: The first generation network specifically includes a four-layer downsampling module of the encoder and a four-layer upsampling module of the decoder, uses a wavelet convolution layer, specifically adopts a db2 wavelet basis, and the number of decomposition layers is set to 2.

3. The adaptive intelligent repair method for missing region of multi-scale seismic data in wavelet domain according to claim 2, characterized in that: The four-layer downsampling module of the encoder is implemented as follows: at each downsampling layer, a two-level decomposition is performed using a db2 wavelet basis to obtain low-frequency and high-frequency feature maps, and a feature map of 1024 channels is obtained by calculating layer by layer; the residual module uses 3x3 convolution; the four-layer upsampling module of the decoder is implemented as follows: at each upsampling layer, a db2 wavelet basis is used for upsampling, and a data image is restored to the same spatial resolution as the original image by calculating layer by layer.

4. The adaptive intelligent repair method for missing region of multi-scale seismic data in wavelet domain according to claim 1, characterized in that: The discriminator includes a convolution layer and seven spectral normalization convolution blocks connected sequentially; each convolution block includes a convolution layer, an SN layer and a Leaky ReLU activation function; the kernel size of each convolution is 3×3, the number of feature maps increases from 64 to 512, and a convolution with a step size of 2 is used.

5. The method for adaptive intelligent repair of missing regions of multi-scale seismic data in wavelet domain according to claim 4, characterized in that: The loss function in the training method of the first generating network includes: l G =l rec +αl per +βl adv +λl tv Among them, l G is the loss function of the generator, α, β, λ are the weights corresponding to each loss term, which are used to balance the impact of loss on training; l rec Represents the reconstruction loss calculated using mean square error MSE, which is used to measure the pixel difference between the generated data and the target data. and its corresponding complete data d R The MSE of is expressed as: W and H represent the missing earthquake data of the input M The dimension of t is from d M to d R The scale of resolution improvement; is the corresponding complete data; Perceptual loss per is a loss function that measures the difference in perceptual quality between generated data and target data, expressed as: where φ i，j represents the operation of extracting the jth convolution feature map before the i-th maximum pooling layer, W i，j and H i，j Represents the dimensions of each feature map in the VGG network; x and y represent count values; l adv To rebuild the losses; tv is the total variational loss.

6. The method for adaptive intelligent repair of missing regions of multi-scale seismic data in wavelet domain according to claim 5, characterized in that: The reconstruction loss is expressed as: Where N represents missing data d M The number of samples, indicates that the discriminator will reconstruct the data The probability of considering the complete seismic data as true; to obtain better gradient behavior, minimize Rather than The total variational loss l tv It is expressed as: Where M is the total number of pixels in the data image, which is used for normalization; X is the input data or feature map; (i, j) represents the pixel coordinates; the first term (X i+1，j -X i，j ) 2 Calculate the pixel change in the vertical direction; the second term (X i，j+1 -X i，j ) 2 Calculates the horizontal pixel change.

7. The method for adaptive intelligent repair of missing regions of multi-scale seismic data in wavelet domain according to claim 1, characterized in that: The signal-to-noise ratio (SNR) and structural similarity (SSIM) indicators are used to evaluate the quality of the interpolated seismic data; The signal-to-noise ratio and structural similarity are calculated as: SNR(dB)=10lgy 2 / (y-G(x)) 2 where y and G(x) represent the ideal and restored seismic data respectively; μ y and u G(x) Represents the average value of the data image pixels of y and G(x); σ y and σ G(x) Represents the variance of the data image pixels of y and G(x) respectively; σ yG(x) represents the covariance of y and G(x) data image pixels; c1 and c2 represent the calculated stability constants.

8. An adaptive intelligent repair device for missing regions of multi-scale seismic data in wavelet domain, characterized in that: The wavelet domain multi-scale seismic data missing area adaptive intelligent repair device performs a wavelet domain multi-scale seismic data missing area adaptive intelligent repair method, the steps include: Obtaining seismic data to be restored; Inputting the seismic data into a pre-trained first generation network to obtain repaired seismic data; The first generation network includes an encoder, a residual module, and a decoder: Wherein, a wavelet convolution module is provided in the encoder and the decoder; Among them, the training method of the first generation network is: Acquire training data, wherein the training data includes missing seismic data and corresponding complete seismic data; Inputting the missing seismic data in the training data into the generator to generate first generated data; using the first generated data and the complete seismic data to calculate the loss of the discriminator and update the parameters of the discriminator; Calculating the loss of the generator through the discriminator and updating the parameters of the generator until the model reaches a preset performance indicator; The trained generator is used as the first generation network.

9. An electronic device, characterized in that: include: a processor adapted to implement one or more instructions; and, A computer storage medium storing one or more instructions, wherein the one or more instructions are suitable for being loaded by the processor and executed by the method for adaptive intelligent repair of missing areas of multi-scale seismic data in the wavelet domain as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The computer storage medium stores one or more instructions, and the one or more instructions are suitable for being loaded by a processor and executed by a method for adaptive intelligent repair of missing areas of multi-scale seismic data in the wavelet domain as described in any one of claims 1-7.

Images