A mixing wave interference suppression method and device, storage medium and electronic equipment

Abstract

The application discloses a mixed wave interference suppression method. The method comprises the following steps: obtaining target mixed wave interference seismic data; performing data domain conversion on the target mixed wave interference seismic data to obtain converted seismic data; training a neural network model according to the converted seismic data to obtain a trained neural network model; and performing mixed wave suppression processing on the converted seismic data by using the trained neural network model to obtain seismic data after mixed wave interference of the target mixed wave interference seismic data is removed. Without synthesizing a large amount of mixed wave sample seismic data, the method directly starts from the mixed wave seismic data itself, trains a network model according to the mixed wave seismic data to be processed in an unsupervised manner, directly uses the trained network model for mixed wave interference processing of the mixed wave seismic data, is more widely applicable, can more effectively protect the seismic data, avoids loss of effective seismic data, does not cause structural false images, and improves the signal-to-noise ratio.

Description

Technical Field

[0001] This application relates to the field of earthquake data processing technology, and in particular to a method, apparatus, computer-readable storage medium, and electronic device for suppressing wave interference. Background Technology

[0002] High-efficiency seismic acquisition technology is being used more and more widely in onshore or offshore seismic exploration. Because this technology allows adjacent shot points to be fired simultaneously or with a very short firing interval, it can greatly improve acquisition efficiency and significantly increase the shot-detector density of seismic acquisition.

[0003] High-efficiency seismic acquisition technology significantly reduces the cost of high-density exploration, making it the preferred acquisition technology for large-scale seismic exploration projects. However, high-efficiency seismic acquisition technologies such as dynamic sliding scan, independent synchronous scan, and ultra-high-efficiency acquisition all introduce noise interference to varying degrees, which can significantly affect the quality of seismic data from adjacent shots. Therefore, it is necessary to specifically remove noise interference through noise suppression.

[0004] Conventional methods for suppressing seismic interference rely on inversion and reconstruction techniques based on sparse transforms to separate seismic information from the muddy waves and effective reflections. However, these methods are prone to structural artifacts and have relatively low signal-to-noise ratio improvements.

[0005] In the past two years, with the popularization of deep learning artificial intelligence algorithms, some researchers have begun to use deep learning methods to attenuate and suppress noise in seismic data mixing, and have achieved certain processing results.

[0006] However, the main characteristic of current deep neural network-based seismic mixing suppression technology is that it requires supervised training of the network model. This means that a large amount of clean seismic data and seismic data containing seismic interference need to be synthesized as labels to train the model. The seismic mixing suppression effect may be better in synthetic seismic data, but in actual seismic data processing, due to the large differences in seismic data samples, the seismic mixing suppression effect is poor, and the adaptability of this type of method is limited to a certain extent. Summary of the Invention

[0007] To address the aforementioned issues, this application proposes a method, apparatus, storage medium, and electronic device for suppressing noise interference. By utilizing unsupervised deep learning technology, it achieves efficient autonomous suppression of noise interference during seismic acquisition. By driving the network model to update hyperparameters based on the seismic data itself, it realizes intelligent noise suppression.

[0008] A first aspect of this application provides a method for suppressing network interference, the method comprising:

[0009] Acquire target ripple interference seismic data;

[0010] The target turbulent seismic data is subjected to data domain transformation to obtain transformed seismic data;

[0011] The neural network model is trained based on the converted seismic data to obtain a trained neural network model;

[0012] The trained neural network model is used to perform noise suppression processing on the converted seismic data to obtain the target noise-interference seismic data after noise removal.

[0013] In some embodiments, training the neural network model based on the converted seismic data to obtain a trained neural network model includes:

[0014] The transformed seismic data is divided into patches to obtain a patch dataset;

[0015] The fragment dataset is augmented and expanded to obtain the expanded fragment dataset.

[0016] Training data is obtained based on the expanded patch dataset;

[0017] The neural network model is trained based on the training data to obtain the trained neural network model.

[0018] In some embodiments, the training data includes model input data, and obtaining the training data based on the expanded patch dataset includes:

[0019] Based on a preset masking strategy, the fragment data in the expanded fragment dataset are masked to obtain the masked fragment dataset.

[0020] The masked patch dataset is used as the input data for the model.

[0021] In some embodiments, the training data further includes model label data, and the step of obtaining training data based on the expanded patch dataset further includes:

[0022] The expanded patch dataset is used as the model label data.

[0023] In some embodiments, the preset masking strategy includes at least one of a dotted masking strategy, a block masking strategy, and a striped masking strategy. The masking process performed on the patch data in the expanded patch dataset based on the preset masking strategy includes:

[0024] Obtain the signal-to-noise ratio of the data segment;

[0025] When the signal-to-noise ratio of the patch data is greater than a first preset threshold, the patches in the expanded patch dataset are masked based on a dotted masking strategy.

[0026] If the signal-to-noise ratio of the patch data is less than a second preset threshold, the patches in the expanded patch dataset are masked based on a strip masking strategy.

[0027] When the signal-to-noise ratio of the patch data is not greater than a first preset threshold and not less than a second preset threshold, the patches in the expanded patch dataset are masked based on a block masking strategy.

[0028] In some embodiments, the data domain conversion includes:

[0029] Any one of the following: common receiver point domain conversion, common shot-receiver distance domain conversion, and cross-shaped domain conversion.

[0030] In some embodiments, training the neural network model based on the training data to obtain the trained neural network model includes:

[0031] Set the initial network parameters of the neural network model;

[0032] The neural network model is trained based on the model input data and model label data of the training data to obtain optimized network parameters;

[0033] The optimized network parameters are used as the network parameters of the neural network model to obtain the trained neural network model.

[0034] A second aspect of this application provides a noise suppression device, the device comprising:

[0035] The acquisition module is used to acquire target ripple interference seismic data;

[0036] The conversion module is used to perform data domain conversion on the target ripple interference seismic data to obtain the converted seismic data;

[0037] The training module is used to train the neural network model based on the converted seismic data to obtain a trained neural network model.

[0038] The interference suppression module is used to perform noise suppression processing on the converted seismic data through the trained neural network model to obtain the target noise interference seismic data after removing the noise.

[0039] A third aspect of this application provides a computer-readable storage medium storing a computer program that can be executed by one or more processors to implement the mixing interference suppression method described above.

[0040] A fourth aspect of this application provides an electronic device including a memory and one or more processors, wherein a computer program is stored on the memory, and the memory and the one or more processors are communicatively connected to each other, wherein when the computer program is executed by the one or more processors, it implements the mixing interference suppression method as described above.

[0041] Compared with the prior art, the technical solution of this application has the following advantages or beneficial effects:

[0042] (1) There is no need to synthesize a large number of seismic data containing turbulent samples. The network model is directly trained in an unsupervised manner based on the turbulent seismic data to be processed. The trained network model is then directly used for turbulent interference processing of the turbulent seismic data. Compared with the traditional sparse transform turbulent suppression method, it has the characteristics of wider applicability.

[0043] (2) Compared with the traditional sparse transform mixing suppression method, this application can more effectively protect seismic data, avoid loss of effective seismic data information, and will not cause tectonic artifacts, while also improving the signal-to-noise ratio. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art are briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0045] Figure 1 A flowchart of a method for suppressing interference provided in an embodiment of this application;

[0046] Figure 2 A Unet network structure model diagram provided for an embodiment of this application;

[0047] Figure 3 A schematic diagram illustrating the effect of suppressing interference is provided for an embodiment of this application;

[0048] Figure 4 This is a schematic diagram of the structure of a noise suppression device provided in an embodiment of this application;

[0049] Figure 5 This is a connection block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0050] The following detailed description of the embodiments of this application, in conjunction with the accompanying drawings, will provide a thorough understanding of how this application uses technical means to solve technical problems and achieve corresponding technical effects, enabling its implementation. The embodiments of this application and the various features within them can be combined with each other without conflict, and all resulting technical solutions are within the protection scope of this application.

[0051] As can be seen from the background technology, high-efficiency seismic acquisition technology is being used more and more widely in onshore or offshore seismic exploration. Because this technology allows adjacent shot points to be fired simultaneously or with a very short firing interval, it can greatly improve acquisition efficiency and also significantly increase the shot-detector density of seismic acquisition.

[0052] High-efficiency seismic acquisition technology significantly reduces the cost of high-density exploration, making it the preferred acquisition technology for large-scale seismic exploration projects. However, high-efficiency seismic acquisition technologies such as dynamic sliding scan, independent synchronous scan, and ultra-high-efficiency acquisition all introduce noise interference to varying degrees, which can significantly affect the quality of seismic data from adjacent shots. Therefore, it is necessary to specifically remove noise interference through noise suppression.

[0053] Conventional methods for suppressing seismic interference rely on inversion and reconstruction techniques based on sparse transforms to separate seismic information from the muddy waves and effective reflections. However, these methods are prone to structural artifacts and have relatively low signal-to-noise ratio improvements.

[0054] In the past two years, with the popularization of deep learning artificial intelligence algorithms, some researchers have begun to use deep learning methods to attenuate and suppress noise in seismic data mixing, and have achieved certain processing results.

[0055] However, the main characteristic of current deep neural network-based seismic mixing suppression technology is that it requires supervised training of the network model. This means that a large amount of clean seismic data and seismic data containing seismic interference need to be synthesized as labels to train the model. The seismic mixing suppression effect may be better in synthetic seismic data, but in actual seismic data processing, due to the large differences in seismic data samples, the seismic mixing suppression effect is poor, and the adaptability of this type of method is limited to a certain extent.

[0056] In view of this, this application discloses a method, device, storage medium and electronic device for suppressing turbulence interference. It removes turbulence noise interference by using an unsupervised autonomous network model, without the need to synthesize a large amount of turbulent sample seismic data. It starts directly from the turbulent seismic data itself, and through the unsupervised autonomous seismic data feature structure, the trained model can be directly used for turbulent seismic data processing, making it more widely applicable.

[0057] Compared with traditional sparse transform mixing suppression methods, this application can more effectively protect seismic data, avoid loss of effective seismic information, avoid causing tectonic artifacts, and also improve the signal-to-noise ratio.

[0058] Example 1

[0059] This embodiment provides a method for suppressing video mixing interference. Figure 1 A flowchart of a method for suppressing interference provided in an embodiment of this application is shown below. Figure 1 As shown, the method in this embodiment includes:

[0060] S110, Acquire target ripple interference seismic data.

[0061] Optionally, target ripple interference seismic data can be obtained from seismic data of the target mining area.

[0062] S120. Perform data domain transformation on the target turbulent interference seismic data to obtain the transformed seismic data.

[0063] In some embodiments, the data domain conversion includes:

[0064] Any one of the following: common receiver point domain conversion, common shot-receiver distance domain conversion, and cross-shaped domain conversion.

[0065] Optionally, the target mixed-wave interference seismic data can be transformed from the common shot point domain to other data domains such as the common receiver point domain, the common shot-receiver distance domain, or the cross domain to obtain transformed seismic data. This transforms the mixed-wave coherent noise interference that affects the effective reflection seismic information into strong-energy random noise interference in other data domains such as the common receiver point domain, resulting in characteristic differences from the effective reflection phase axis, which can be used for subsequent processing.

[0066] S130. Train the neural network model based on the converted seismic data to obtain a trained neural network model.

[0067] A U-shaped network structure, such as Unet or Unet++, can be used, which is built based on an encoder layer and a decoder layer. Figure 2 A Unet network structure model diagram provided for embodiments of this application, such as... Figure 2 The Unet network structure model shown uses a multi-layer deep network encoder to extract patch feature structures. As the number of layers increases and the network becomes deeper, the feature structures of compressed low-dimensional seismic data are obtained. Then, the high-dimensional seismic data is continuously reconstructed and restored through the decoder layer to obtain the reconstructed and restored seismic data.

[0068] To improve the suppression of noise while preserving effective seismic signals, skip connections are used to establish connections between the coding and decoding layers. Effective seismic information extracted from the coding layer at the same height is directly spliced ​​into the decoding layer at the same height. This ensures that the deep network does not lose higher-order features during downsampling and upsampling, and can recover the effective information loss caused during downsampling.

[0069] To improve the generalization ability of the network architecture and increase the diversity of the network model, random dropout is used to randomly deactivate neurons in the multi-layer network, so that the network model can effectively retain the characteristics of seismic data while discarding noise features.

[0070] In some embodiments, training the neural network model based on the converted seismic data to obtain a trained neural network model includes:

[0071] The transformed seismic data is divided into patches to obtain a patch dataset;

[0072] The fragment dataset is augmented and expanded to obtain the expanded fragment dataset.

[0073] Training data is obtained based on the expanded patch dataset;

[0074] The neural network model is trained based on the training data to obtain the trained neural network model.

[0075] Optionally, the piece-cutting and segmentation process includes the following steps:

[0076] Set the patch size to M*N, then slide the patches across the seismic data from top to bottom and from left to right, setting the sliding step size, including a spatial step size x_step and a time step size t_step. The patch size and sliding step size in different directions are adjusted based on the actual size of the seismic data profile. If the seismic data to be processed is large, the patch size and sliding step size can be appropriately increased. When the seismic data size is insufficient for patch division, zero-padding is required to ensure the seismic data can be completely divided. This step transforms a turbulent seismic data profile into thousands of patches of the same size, forming a patch dataset B.

[0077] Optionally, augmentation processing includes the following steps:

[0078] Unsupervised mixing and interference suppression techniques were employed, eliminating the need to synthesize large amounts of seismic and label data. To expand the seismic dataset, the patch seismic dataset obtained in the previous steps required augmentation. Seismic data augmentation methods included polarity reversal, rotation at different angles (e.g., 90°, 180°, and 270° counterclockwise), and vertical or horizontal flipping. Furthermore, to enhance data diversity, mirroring can be used to enlarge the seismic data size; for example, a 32x32 seismic dataset can be mirrored to 56x56, and then further cropped to obtain different types of 32x32 seismic datasets. Through these augmentation methods, the seismic data can be expanded by several times, or even tens of times, to obtain a new patch dataset B'. Due to the increased diversity of sample data, the deep neural network is less likely to fall into local extremum traps, thus avoiding overfitting issues in the network model.

[0079] In some embodiments, the training data includes model input data, and obtaining the training data based on the expanded patch dataset includes:

[0080] Based on a preset masking strategy, the fragment data in the expanded fragment dataset are masked to obtain the masked fragment dataset.

[0081] The masked patch dataset is used as the input data for the model.

[0082] In some embodiments, the training data further includes model label data, and the step of obtaining training data based on the expanded patch dataset further includes:

[0083] The expanded patch dataset is used as the model label data.

[0084] In some embodiments, the preset masking strategy includes at least one of a dotted masking strategy, a block masking strategy, and a striped masking strategy. The masking process performed on the patch data in the expanded patch dataset based on the preset masking strategy includes:

[0085] Obtain the signal-to-noise ratio of the data segment;

[0086] When the signal-to-noise ratio of the patch data is greater than a first preset threshold, the patches in the expanded patch dataset are masked based on a dotted masking strategy.

[0087] If the signal-to-noise ratio of the patch data is less than a second preset threshold, the patches in the expanded patch dataset are masked based on a strip masking strategy.

[0088] When the signal-to-noise ratio of the patch data is not greater than a first preset threshold and not less than a second preset threshold, the patches in the expanded patch dataset are masked based on a block masking strategy.

[0089] Optionally, after augmenting the seismic sample dataset, masking can be performed. Masking includes the following steps:

[0090] Randomly select one of the fragments b from the expanded fragment dataset. i Where i = 1, 2, ..., N, and N is the total number of fragment samples, the fragment is processed by a masking strategy to obtain the masked fragment b′. i As input to a deep network, the original piece b i The desired output of the deep network is used to train the network model parameters.

[0091] Optionally, the signal-to-noise ratio of the patch data can be obtained through mean-variance analysis. Specifically, for patch data b containing seismic data noise from turbulence, i To conduct a preliminary assessment of noise intensity, the first step is to acquire patch data b. i The mean and standard deviation are then used to obtain the data b of the patch using the following formula. i Signal-to-noise ratio:

[0092] mse=10*(b i _mean / b i _sd)

[0093] Among them, b i _mean represents the piece of data b i The mean, b i _sd represents the fragment data b i The standard deviation, mse, is the standard deviation of the fragmented data b. i The signal-to-noise ratio result value.

[0094] Optionally, the masking strategy includes the following three methods, which can be flexibly selected according to actual needs or factors such as the strength and distribution characteristics of the noise:

[0095] (1) Dot masking strategy

[0096] For a certain piece b i Perform dotted masking processing, randomly select a certain proportion of data points p_point in the patch, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around p_point within the set radius range to replace the value of that point.

[0097] The randomly selected data points p_point can also be zeroed out or filled with random noise.

[0098] It should be noted that the random point sampling ratio p_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0099] (2) Block masking strategy

[0100] For a certain piece b i Perform block masking processing, randomly select a certain percentage of data blocks b_block from the block, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around b_block within the set radius range to replace the values ​​in the data block.

[0101] Seismic data within randomly selected data blocks (b_block) can also be zeroed out or filled with random noise.

[0102] It should be noted that the random block extraction ratio b_percent can be selected from a few tenths of a percent to several tens of percent, depending on the characteristics of the actual seismic data.

[0103] (3) Striped masking strategy

[0104] For a certain piece b i Masking is performed by randomly selecting a certain percentage of strip seismic traces (s_trace) from the patch in the spatial direction. Then, a mask radius (m_r) is set within a certain range, and other strip seismic traces are randomly selected around s_trace within the set radius to replace the original seismic trace.

[0105] The randomly selected strip-shaped seismic traces s_trace can also be zeroed out or filled with random noise.

[0106] It should be noted that the random seismic trace extraction ratio s_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0107] It should be further noted that when masking a certain patch, the point, block, and strip seismic data are randomly extracted and exhibit a random distribution pattern.

[0108] Optionally, when the noise interference energy of the target mixed seismic data is relatively weak and close to random noise (e.g., when the signal-to-noise ratio is greater than the first preset threshold), a dotted masking strategy is adopted; when the noise interference energy of the target mixed seismic data is very strong (e.g., when the signal-to-noise ratio is less than the second preset threshold), a striped masking strategy is adopted; when the noise interference energy of the target mixed seismic data is between relatively weak and very strong (e.g., when the signal-to-noise ratio is not greater than the first preset threshold and not less than the second preset threshold), a block masking strategy is adopted.

[0109] For example, when the interference energy of the mixing noise is relatively weak and close to that of random noise, the signal-to-noise ratio of the seismic data is high, such as when MSE > 20 dB, a point masking strategy is adopted; when the interference energy of the mixing noise is very strong, the signal-to-noise ratio of the seismic data is low, such as when MSE < 10 dB, a strip masking strategy is adopted; when the interference energy of the mixing noise is at a medium intensity level, the signal-to-noise ratio of the seismic data is moderate, such as when 10 dB ≤ MSE ≤ 20 dB, a block masking strategy is adopted.

[0110] It should be noted that, provided the first preset threshold is greater than the second preset threshold, both the first and second preset thresholds can be set according to the user's actual needs, and no special restrictions are made here.

[0111] Existing conventional methods for demixing using deep learning require synthesizing a large amount of labeled sample data for training. This application employs a masking strategy that starts with the seismic data itself. It randomly selects a certain proportion of point-like, block-like, and strip-like seismic data from different segmented seismic data blocks for masking processing. The masked data serves as the network input, along with the original block-like data. i The data is used as the network output to train the network model parameters. This approach abandons conventional supervised learning methods and utilizes unsupervised learning, avoiding the need to synthesize large amounts of ripple sample data. This makes ripple suppression processing more flexible and allows for a wider range of seismic data to be processed.

[0112] In some embodiments, training the neural network model based on the training data to obtain the trained neural network model includes:

[0113] Set the initial network parameters of the neural network model;

[0114] The neural network model is trained based on the model input data and model label data of the training data to obtain optimized network parameters;

[0115] The optimized network parameters are used as the network parameters of the neural network model to obtain the trained neural network model.

[0116] Optionally, after building the deep U-shaped network model, you can set network model training parameters such as batch size, epochs, learning rate l_r, loss function, and optimizer. These parameters can be flexibly adjusted according to the characteristics of the seismic data.

[0117] Optionally, during training, a specific piece of data b from the augmented patch dataset can be used. i The seismic data b′ after masking i and b i The network is divided into training samples, test samples, and validation samples according to a certain ratio. The training samples are used to train the network model and adjust the parameters. The test samples are used to evaluate the generalization performance of the trained network model to determine whether to retrain the model, and are not used for adjusting the model parameters. The validation samples are used to test the performance of the network model, and the model hyperparameters are adjusted based on the validation results.

[0118] For example, since the seismic data before and after masking are in one-to-one correspondence, such as b′ after 1000 masking processes... i b before processing 1000 masks i The data, namely 1000 pairs of sample data, is divided into 800 pairs of training samples, 200 pairs of test samples, and 100 pairs of validation samples in a ratio of 8:1:1.

[0119] The masked seismic data is used as the model input, and the unmasked seismic data is used as the model label data. The neural network model is continuously trained. As the number of training iterations increases, the loss functions of both the training and validation sets decrease and stabilize. When the training results meet the preset requirements, stable network parameters are obtained, which can then be used for subsequent processing of seismic data with ripple interference.

[0120] S140. The converted seismic data is subjected to noise suppression processing using the trained neural network model to obtain the target noise interference seismic data after noise removal.

[0121] After training the neural network model, the trained neural network model is used to suppress the turbulent seismic data to be processed, thereby obtaining the turbulent-free seismic data corresponding to the turbulent seismic data to be processed.

[0122] This embodiment proposes a method for suppressing seismic data noise mixing based on a masking strategy and an unsupervised learning-based U-shaped network. First, the seismic data to be suppressed undergoes data domain transformation. Then, the seismic data is divided into patch seismic data through sliding segmentation. Seismic data augmentation techniques are then used to significantly increase the sample size and enrich sample diversity. Next, different masking techniques are employed to randomly replace the masked patches. This method abandons conventional supervised learning methods, avoiding the need to synthesize a large amount of mixed sample data, thus broadening the scope of seismic data to be processed. The data before and after masking are used as the input and expected output of a deep network. A network model based on Unet or Unet++ is used for both input and output of the seismic data, continuously training the network parameters to extract effective information features from the seismic data. By constructing an unsupervised seismic attenuation network model driven by the characteristics of the seismic data itself, it is directly applied to seismic data with mixed noise interference. The output of the network model is the seismic data after removing the mixed noise.

[0123] The noise suppression method disclosed in this embodiment does not require synthesizing a large number of noise-containing seismic data samples when suppressing noise interference in noise-affected seismic data. Instead, it directly trains an unsupervised network model based on the noise-affected seismic data to be processed, and then directly uses the trained network model for noise interference processing of the noise-affected seismic data. Compared with the traditional sparse transform noise suppression method, it has a wider range of applicability.

[0124] Compared with traditional sparse transform mixing suppression methods, this application can more effectively protect seismic data, avoid loss of effective seismic data information, and avoid causing tectonic artifacts, while also improving the signal-to-noise ratio.

[0125] Example 2

[0126] This embodiment provides a specific example, verifying the effectiveness of the interference suppression method. The method of this embodiment may include the following steps:

[0127] The first step is to acquire target ripple interference seismic data.

[0128] Optionally, target ripple interference seismic data can be obtained from seismic data of the target mining area.

[0129] The second step is to perform data domain transformation on the target muddy interference seismic data to obtain the transformed seismic data.

[0130] In some embodiments, the data domain conversion includes:

[0131] Any one of the following: common receiver point domain conversion, common shot-receiver distance domain conversion, and cross-shaped domain conversion.

[0132] Optionally, the target mixed-wave interference seismic data can be transformed from the common shot point domain to other data domains such as the common receiver point domain, the common shot-receiver distance domain, or the cross domain to obtain transformed seismic data. This transforms the mixed-wave coherent noise interference that affects the effective reflection seismic information into strong-energy random noise interference in other data domains such as the common receiver point domain, resulting in characteristic differences from the effective reflection phase axis, which can be used for subsequent processing.

[0133] The third step is to train the neural network model based on the converted seismic data to obtain a trained neural network model.

[0134] Optionally, a U-shaped network structure such as Unet or Unet++, built based on an encoder layer and a decoder layer, can be used. Figure 2 A Unet network structure model diagram provided for embodiments of this application, such as... Figure 2 The Unet network structure model shown uses a multi-layer deep network encoder to extract patch feature structures. As the number of layers increases and the network becomes deeper, the feature structures of compressed low-dimensional seismic data are obtained. Then, the high-dimensional seismic data is continuously reconstructed and restored through the decoder layer to obtain the reconstructed and restored seismic data.

[0135] To improve the suppression of noise while preserving effective seismic signals, skip connections are used to establish connections between the coding and decoding layers. Effective seismic information extracted from the coding layer at the same height is directly spliced ​​into the decoding layer at the same height. This ensures that the deep network does not lose higher-order features during downsampling and upsampling, and can recover the effective information loss caused during downsampling.

[0136] To improve the generalization ability of the network architecture and increase the diversity of the network model, random dropout is used to randomly deactivate neurons in the multi-layer network, so that the network model can effectively retain the characteristics of seismic data while discarding noise features.

[0137] In some embodiments, training the neural network model based on the converted seismic data to obtain a trained neural network model includes:

[0138] The transformed seismic data is divided into patches to obtain a patch dataset;

[0139] The fragment dataset is augmented and expanded to obtain the expanded fragment dataset.

[0140] Training data is obtained based on the expanded patch dataset;

[0141] The neural network model is trained based on the training data to obtain the trained neural network model.

[0142] Optionally, the piece-cutting and segmentation process includes the following steps:

[0143] Set the patch size to M*N, then slide the patches across the seismic data from top to bottom and from left to right. Set the spatial step size to x_step and the time step size to t_step. The patch size and sliding step size in different directions are adjusted according to the actual size of the seismic data profile. If the seismic data to be processed is large, the patch size and sliding step size can be appropriately increased. When the seismic data size is insufficient for patch division, zero-padding is required to ensure that the seismic data can be completely divided. This step transforms a turbulent seismic data profile into a dataset B of thousands of patches of the same size.

[0144] Optionally, augmentation processing includes the following steps:

[0145] Unsupervised mixing and interference suppression techniques were employed, eliminating the need to synthesize large amounts of seismic and label data. To expand the seismic dataset, the patch seismic dataset obtained in the previous steps required augmentation. Seismic data augmentation methods included polarity reversal, rotation at different angles (e.g., 90°, 180°, and 270° counterclockwise), and vertical or horizontal flipping. Furthermore, to enhance data diversity, mirroring can be used to enlarge the seismic data size; for example, a 32x32 seismic dataset can be mirrored to 56x56, and then further cropped to obtain different types of 32x32 seismic datasets. Through these augmentation methods, the seismic data can be expanded by several times, or even tens of times, to obtain a new patch dataset B'. Due to the increased diversity of sample data, the deep neural network is less likely to fall into local extremum traps, thus avoiding overfitting issues in the network model.

[0146] In some embodiments, the training data includes model input data, and obtaining the training data based on the expanded patch dataset includes:

[0147] Based on a preset masking strategy, the fragment data in the expanded fragment dataset are masked to obtain the masked fragment dataset.

[0148] The masked patch dataset is used as the input data for the model.

[0149] In some embodiments, the training data further includes model label data, and the step of obtaining training data based on the expanded patch dataset further includes:

[0150] The expanded patch dataset is used as the model label data.

[0151] In some embodiments, the preset masking strategy includes at least one of a dotted masking strategy, a block masking strategy, and a striped masking strategy. The masking process performed on the patch data in the expanded patch dataset based on the preset masking strategy includes:

[0152] Obtain the signal-to-noise ratio of the data segment;

[0153] When the signal-to-noise ratio of the patch data is greater than a first preset threshold, the patches in the expanded patch dataset are masked based on a dotted masking strategy.

[0154] If the signal-to-noise ratio of the patch data is less than a second preset threshold, the patches in the expanded patch dataset are masked based on a strip masking strategy.

[0155] When the signal-to-noise ratio of the patch data is not greater than a first preset threshold and not less than a second preset threshold, the patches in the expanded patch dataset are masked based on a block masking strategy.

[0156] Optionally, after augmenting the seismic sample dataset, masking can be performed. Masking includes the following steps:

[0157] Randomly select one of the fragments b from the expanded fragment dataset. i Where i = 1, 2, ..., N, and N is the total number of fragment samples, the fragment is processed by a masking strategy to obtain the masked fragment b′. i As input to a deep network, the original piece b i The desired output of the deep network is used to train the network model parameters.

[0158] Optionally, the signal-to-noise ratio of the patch data can be obtained through mean-variance analysis. Specifically, for patch data b containing seismic data noise from turbulence, i To conduct a preliminary assessment of noise intensity, the first step is to acquire patch data b. i The mean and standard deviation are then used to obtain the data b of the patch using the following formula. i Signal-to-noise ratio:

[0159] mse=10*(b i _mean / b i _sd)

[0160] Among them, b i _mean represents the piece of data b i The mean, b i _sd represents the fragment data b iThe standard deviation, mse, is the standard deviation of the fragmented data b. i The signal-to-noise ratio result value.

[0161] Optionally, the masking strategy includes the following three methods, which can be flexibly selected according to actual needs or factors such as the strength and distribution characteristics of the noise:

[0162] (1) Dot masking strategy

[0163] For a certain piece b i Perform dotted masking processing, randomly select a certain proportion of data points p_point in the patch, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around p_point within the set radius range to replace the value of that point.

[0164] The randomly selected data points p_point can also be zeroed out or filled with random noise.

[0165] It should be noted that the random point sampling ratio p_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0166] (2) Block masking strategy

[0167] For a certain piece b i Perform block masking processing, randomly select a certain percentage of data blocks b_block from the block, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around b_block within the set radius range to replace the values ​​in the data block.

[0168] Seismic data within randomly selected data blocks (b_block) can also be zeroed out or filled with random noise.

[0169] It should be noted that the random block extraction ratio b_percent can be selected from a few tenths of a percent to several tens of percent, depending on the characteristics of the actual seismic data.

[0170] (3) Striped masking strategy

[0171] For a certain piece b i Masking is performed by randomly selecting a certain percentage of strip seismic traces (s_trace) from the patch in the spatial direction. Then, a mask radius (m_r) is set within a certain range, and other strip seismic traces are randomly selected around s_trace within the set radius to replace the original seismic trace.

[0172] The randomly selected strip-shaped seismic traces s_trace can also be zeroed out or filled with random noise.

[0173] It should be noted that the random seismic trace extraction ratio s_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0174] It should be further noted that when masking a certain patch, the point, block, and strip seismic data are randomly extracted and exhibit a random distribution pattern.

[0175] Optionally, when the noise interference energy of the target mixed seismic data is relatively weak and close to random noise (e.g., when the signal-to-noise ratio is greater than the first preset threshold), a dotted masking strategy is adopted; when the noise interference energy of the target mixed seismic data is very strong (e.g., when the signal-to-noise ratio is less than the second preset threshold), a striped masking strategy is adopted; when the noise interference energy of the target mixed seismic data is between relatively weak and very strong (e.g., when the signal-to-noise ratio is not greater than the first preset threshold and not less than the second preset threshold), a block masking strategy is adopted.

[0176] For example, when the interference energy of the mixing noise is relatively weak and close to that of random noise, the signal-to-noise ratio of the seismic data is high, such as when MSE > 20 dB, a point masking strategy is adopted; when the interference energy of the mixing noise is very strong, the signal-to-noise ratio of the seismic data is low, such as when MSE < 10 dB, a strip masking strategy is adopted; when the interference energy of the mixing noise is at a medium intensity level, the signal-to-noise ratio of the seismic data is moderate, such as when 10 dB ≤ MSE ≤ 20 dB, a block masking strategy is adopted.

[0177] It should be noted that, provided the first preset threshold is greater than the second preset threshold, both the first and second preset thresholds can be set according to the user's actual needs, and no special restrictions are made here.

[0178] Existing conventional methods for demixing using deep learning require synthesizing a large amount of labeled sample data for training. This application employs a masking strategy that starts with the seismic data itself. It randomly selects a certain proportion of point-like, block-like, and strip-like seismic data from different segmented seismic data blocks for masking processing. The masked data serves as the network input, along with the original block-like data. i The data is used as the network output to train the network model parameters. This approach abandons conventional supervised learning methods and utilizes unsupervised learning, avoiding the need to synthesize large amounts of ripple sample data. This makes ripple suppression processing more flexible and allows for a wider range of seismic data to be processed.

[0179] In some embodiments, training the neural network model based on the training data to obtain the trained neural network model includes:

[0180] Set the initial network parameters of the neural network model;

[0181] The neural network model is trained based on the model input data and model label data of the training data to obtain optimized network parameters;

[0182] The optimized network parameters are used as the network parameters of the neural network model to obtain the trained neural network model.

[0183] Optionally, after building the deep U-shaped network model, you can set network model training parameters such as batch size, epochs, learning rate l_r, loss function, and optimizer. These parameters can be flexibly adjusted according to the characteristics of the seismic data.

[0184] Optionally, during training, a specific piece of data b from the augmented patch dataset can be used. i The seismic data b′ after masking i and b i The network is divided into training samples, test samples, and validation samples according to a certain ratio. The training samples are used to train the network model and adjust the parameters. The test samples are used to evaluate the generalization performance of the trained network model to determine whether to retrain the model, and are not used for adjusting the model parameters. The validation samples are used to test the performance of the network model, and the model hyperparameters are adjusted based on the validation results.

[0185] For example, since the seismic data before and after masking are in one-to-one correspondence, such as b′ after 1000 masking processes... i b before processing 1000 masks i The data, namely 1000 pairs of sample data, is divided into 800 pairs of training samples, 200 pairs of test samples, and 100 pairs of validation samples in a ratio of 8:1:1.

[0186] The masked seismic data is used as the model input, and the unmasked seismic data is used as the model label data. The neural network model is continuously trained. As the number of training iterations increases, the loss functions of both the training and validation sets decrease and stabilize. When the training results meet the preset requirements, stable network parameters are obtained, which can then be used for subsequent processing of seismic data with ripple interference.

[0187] The fourth step involves using the trained neural network model to perform noise suppression processing on the converted seismic data to obtain the target noise-interference seismic data after noise removal.

[0188] After training the neural network model, the trained neural network model is used to suppress the turbulent seismic data to be processed, thereby obtaining the turbulent-free seismic data corresponding to the turbulent seismic data to be processed.

[0189] like Figure 3 As shown, Figure 3 This application provides a schematic diagram illustrating the effect of suppressing interference. Figure 3 The left part of the image contains the original seismic data with severe noise interference, making it impossible to effectively identify in-phase axes; Figure 3 The middle part of the image represents the seismic data after removing the noise using this invention; Figure 3 The right side of the image shows the removed muddy interference, and it can be seen that the suppressed muddy interference does not damage the effective seismic reflection signal.

[0190] This embodiment demonstrates the effectiveness of the noise suppression method provided in this application. When suppressing noise in noise-affected seismic data, the noise suppression method disclosed in this embodiment does not require the synthesis of a large number of noise-containing seismic data samples. Instead, it directly trains an unsupervised network model based on the noise-affected seismic data to be processed, and then directly applies the trained network model to the noise-affected seismic data. Compared with the traditional sparse transform noise suppression method, it has a wider range of applicability.

[0191] Compared with traditional sparse transform mixing suppression methods, this application can more effectively protect seismic data, avoid loss of effective seismic data information, and avoid causing tectonic artifacts, while also improving the signal-to-noise ratio.

[0192] Example 3

[0193] This embodiment provides a noise suppression device. This device embodiment can be used to execute the method embodiment of this application. For details not disclosed in this device embodiment, please refer to the method embodiment of this application. Figure 4 This is a schematic diagram of a noise suppression device provided in an embodiment of this application, as shown below. Figure 4 As shown, the device 400 provided in this embodiment includes:

[0194] Module 401 is used to acquire target ripple interference seismic data;

[0195] The conversion module 402 is used to perform data domain conversion on the target turbulent interference seismic data to obtain the converted seismic data;

[0196] The training module 403 is used to train the neural network model based on the converted seismic data to obtain a trained neural network model.

[0197] The interference suppression module 404 is used to perform noise suppression processing on the converted seismic data through the trained neural network model to obtain the target noise interference seismic data after removing the noise.

[0198] In some embodiments, the data domain conversion includes:

[0199] Any one of the following: common receiver point domain conversion, common shot-receiver distance domain conversion, and cross-shaped domain conversion.

[0200] Optionally, target ripple interference seismic data can be obtained from seismic data of the target mining area.

[0201] Optionally, the target mixed-wave interference seismic data can be transformed from the common shot point domain to other data domains such as the common receiver point domain, the common shot-receiver distance domain, or the cross domain to obtain transformed seismic data. This transforms the mixed-wave coherent noise interference that affects the effective reflection seismic information into strong-energy random noise interference in other data domains such as the common receiver point domain, resulting in characteristic differences from the effective reflection phase axis, which can be used for subsequent processing.

[0202] Optionally, a U-shaped network structure such as Unet or Unet++, built based on an encoder layer and a decoder layer, can be used. Figure 2 A Unet network structure model diagram provided for embodiments of this application, such as... Figure 2 The Unet network structure model shown extracts patch feature structures through multiple deep network encoder layers. As the number of layers increases and the network deepens, it obtains the feature structures of compressed low-dimensional seismic data. Then, the high-dimensional seismic data is continuously reconstructed and restored through decoder layers to obtain the reconstructed seismic data.

[0203] To improve the suppression of noise while preserving effective seismic signals, a skip connection is used to connect the coding and decoding layers. Effective seismic information extracted from the coding layer at the same height is directly concatenated into the decoding layer at the same height. This ensures that the deep network does not lose high-order features during downsampling and upsampling, and can recover the effective information loss caused during downsampling.

[0204] To improve the generalization ability of the network architecture and increase the diversity of the network model, random dropout is used to randomly deactivate neurons in the multi-layer network, so that the network model can effectively retain the characteristics of seismic data while discarding noise features.

[0205] In some embodiments, the training module 403 includes: a segmentation unit, an expansion unit, an acquisition unit, and a training unit; wherein...

[0206] A segmentation unit is used to segment the converted seismic data into patches to obtain a patch dataset.

[0207] An augmentation unit is used to augment and expand the fragment dataset to obtain an augmented fragment dataset.

[0208] The acquisition unit is used to acquire training data based on the expanded patch dataset.

[0209] The training unit is used to train the neural network model based on the training data to obtain the trained neural network model.

[0210] Optionally, the piece-cutting and segmentation process includes the following steps:

[0211] Set the patch size to M*N, then slide the patches across the seismic data from top to bottom and from left to right. Set the step size along the spatial direction as x_step and the step size along the time direction as t_step. The patch size and sliding step size in different directions are adjusted according to the actual size of the seismic data profile. If the seismic data to be processed is large, the patch size and sliding step size can be appropriately increased. When the seismic data size is insufficient for patch division, zero-padding is required to ensure that the seismic data can be completely divided. This step transforms a turbulent seismic data profile into a dataset B of thousands of patches of the same size.

[0212] Optionally, augmentation processing includes the following steps:

[0213] Unsupervised mixing and interference suppression techniques were employed, eliminating the need to synthesize large amounts of seismic and label data. To expand the seismic dataset, the patch seismic dataset obtained in the previous steps required augmentation. Seismic data augmentation methods included polarity reversal, rotation at different angles (e.g., 90°, 180°, and 270° counterclockwise), and vertical or horizontal flipping. Furthermore, to enhance data diversity, mirroring can be used to enlarge the seismic data size; for example, a 32x32 seismic dataset can be mirrored to 56x56, and then further cropped to obtain different types of 32x32 seismic datasets. Through these augmentation methods, the seismic data can be expanded by several times, or even tens of times, to obtain a new patch dataset B'. Due to the increased diversity of sample data, the deep neural network is less likely to fall into local extremum traps, thus avoiding overfitting issues in the network model.

[0214] In some embodiments, the training data includes model input data, and the acquisition unit includes: a mask subunit and a first confirmation subunit; wherein...

[0215] The masking subunit is used to perform masking processing on the piece data in the expanded piece dataset based on a preset masking strategy to obtain the masked piece dataset.

[0216] The first confirmation subunit is used to take the masked patch dataset as the model input data.

[0217] In some embodiments, the training data further includes model label data, and the acquisition unit further includes a second confirmation subunit, used to use the expanded patch dataset as the model label data.

[0218] In some embodiments, the preset masking strategy includes at least one of a dotted masking strategy, a block masking strategy, and a striped masking strategy, and the masking subunit includes an acquisition part, a first processing part, a second processing part, and a third processing part; wherein...

[0219] The acquisition section is used to acquire the signal-to-noise ratio of the fragment data;

[0220] The first processing part is used to perform masking processing on the blocks in the expanded block dataset based on a dotted masking strategy when the signal-to-noise ratio of the block data is greater than a first preset threshold.

[0221] The second processing part is used to perform masking processing on the blocks in the expanded block dataset based on a strip masking strategy when the signal-to-noise ratio of the block data is less than a second preset threshold.

[0222] The third processing part is used to perform masking processing on the blocks in the expanded block dataset based on a block masking strategy when the signal-to-noise ratio of the block data is not greater than a first preset threshold and not less than a second preset threshold.

[0223] Optionally, after augmenting the seismic sample dataset, masking can be performed. Masking includes the following steps:

[0224] Randomly select one of the fragments b from the expanded fragment dataset. i Where i = 1, 2, ..., N, and N is the total number of fragment samples, the fragment is processed by a masking strategy to obtain the masked fragment b′. i As input to a deep network, the original piece b i The desired output of the deep network is used to train the network model parameters.

[0225] Optionally, the signal-to-noise ratio of the patch data can be obtained through mean-variance analysis. Specifically, for patch data b containing seismic data noise from turbulence, i To conduct a preliminary assessment of noise intensity, the first step is to acquire patch data b.i The mean and standard deviation are then used to obtain the data b of the patch using the following formula. i Signal-to-noise ratio:

[0226] mse=10*(b i _mean / b i _sd)

[0227] Among them, b i _mean represents the piece of data b i The mean, b i _sd represents the fragment data b i The standard deviation, mse, is the standard deviation of the fragmented data b. i The signal-to-noise ratio result value.

[0228] Optionally, the masking strategy includes the following three methods, which can be flexibly selected according to actual needs or factors such as the strength and distribution characteristics of the noise:

[0229] (1) Dot masking strategy

[0230] For a certain piece b i Perform dotted masking processing, randomly select a certain proportion of data points p_point in the patch, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around p_point within the set radius range to replace the value of that point.

[0231] The randomly selected data points p_point can also be zeroed out or filled with random noise.

[0232] It should be noted that the random point sampling ratio p_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0233] (2) Block masking strategy

[0234] For a certain piece b i Perform block masking processing, randomly select a certain percentage of data blocks b_block from the block, then set a mask radius m_r within a certain range, and randomly select other data point values ​​around b_block within the set radius range to replace the values ​​in the data block.

[0235] Seismic data within randomly selected data blocks (b_block) can also be zeroed out or filled with random noise.

[0236] It should be noted that the random block extraction ratio b_percent can be selected from a few tenths of a percent to several tens of percent, depending on the characteristics of the actual seismic data.

[0237] (3) Striped masking strategy

[0238] For a certain piece b i Masking is performed by randomly selecting a certain percentage of strip seismic traces (s_trace) from the patch in the spatial direction. Then, a mask radius (m_r) is set within a certain range, and other strip seismic traces are randomly selected around s_trace within the set radius to replace the original seismic trace.

[0239] The randomly selected strip-shaped seismic traces s_trace can also be zeroed out or filled with random noise.

[0240] It should be noted that the random seismic trace extraction ratio s_percent is selected from a few tenths of a percent to several tens of percent based on the characteristics of the seismic data.

[0241] It should be further noted that when masking a certain patch, the point, block, and strip seismic data are randomly extracted and exhibit a random distribution pattern.

[0242] Optionally, when the noise interference energy of the target mixed seismic data is relatively weak and close to random noise (e.g., when the signal-to-noise ratio is greater than the first preset threshold), a dotted masking strategy is adopted; when the noise interference energy of the target mixed seismic data is very strong (e.g., when the signal-to-noise ratio is less than the second preset threshold), a striped masking strategy is adopted; when the noise interference energy of the target mixed seismic data is between relatively weak and very strong (e.g., when the signal-to-noise ratio is not greater than the first preset threshold and not less than the second preset threshold), a block masking strategy is adopted.

[0243] For example, when the interference energy of the mixing noise is relatively weak and close to that of random noise, the signal-to-noise ratio of the seismic data is high, such as when MSE > 20 dB, a point masking strategy is adopted; when the interference energy of the mixing noise is very strong, the signal-to-noise ratio of the seismic data is low, such as when MSE < 10 dB, a strip masking strategy is adopted; when the interference energy of the mixing noise is at a medium intensity level, the signal-to-noise ratio of the seismic data is moderate, such as when 10 dB ≤ MSE ≤ 20 dB, a block masking strategy is adopted.

[0244] It should be noted that, provided the first preset threshold is greater than the second preset threshold, both the first and second preset thresholds can be set according to the user's actual needs, and no special restrictions are made here.

[0245] Existing conventional methods for demixing using deep learning require synthesizing a large amount of labeled sample data for training. This application employs a masking strategy that starts with the seismic data itself. It randomly selects a certain proportion of point-like, block-like, and strip-like seismic data from different segmented seismic data blocks for masking processing. The masked data serves as the network input, along with the original block-like data. i The data is used as the network output to train the network model parameters. This approach abandons conventional supervised learning methods and utilizes unsupervised learning, avoiding the need to synthesize large amounts of ripple sample data. This makes ripple suppression processing more flexible and allows for a wider range of seismic data to be processed.

[0246] In some embodiments, the training unit includes: a setting subunit, a training subunit, and a configuration subunit; wherein...

[0247] A setting subunit is used to set the initial network parameters of the neural network model;

[0248] The training subunit is used to train the neural network model based on the model input data and model label data of the training data to obtain optimized network parameters;

[0249] A configuration subunit is used to use the optimized network parameters as the network parameters of the neural network model to obtain the trained neural network model.

[0250] Optionally, after building the deep U-shaped network model, you can set network model training parameters such as batch size, epochs, learning rate l_r, loss function, and optimizer. These parameters can be flexibly adjusted according to the characteristics of the seismic data.

[0251] Optionally, during training, a specific piece of data b from the augmented patch dataset can be used. i The seismic data b′ after masking i and b i The network is divided into training samples, test samples, and validation samples according to a certain ratio. The training samples are used to train the network model and adjust the parameters. The test samples are used to evaluate the generalization performance of the trained network model to determine whether to retrain the model, and are not used for adjusting the model parameters. The validation samples are used to test the performance of the network model, and the model hyperparameters are adjusted based on the validation results.

[0252] For example, since the seismic data before and after masking are in one-to-one correspondence, such as b′ after 1000 masking processes... i b before processing 1000 masks iThe data, namely 1000 pairs of sample data, is divided into 800 pairs of training samples, 200 pairs of test samples, and 100 pairs of validation samples in a ratio of 8:1:1.

[0253] The masked seismic data is used as the model input, and the unmasked seismic data is used as the model label data. The neural network model is continuously trained. As the number of training iterations increases, the loss functions of both the training and validation sets decrease and stabilize. When the training results meet the preset requirements, stable network parameters are obtained, which can then be used for subsequent processing of seismic data with ripple interference.

[0254] After training the neural network model, the trained neural network model is used to suppress the turbulent seismic data to be processed, thereby obtaining the turbulent-free seismic data corresponding to the turbulent seismic data to be processed.

[0255] Those skilled in the field can understand. Figure 4 The structures shown do not constitute a limitation on the apparatus of the embodiments of this application. They may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0256] It should be noted that the above modules / units can be either functional modules or program modules, and can be implemented through software or hardware. For modules / units implemented in hardware, the above modules / units can reside in the same processor; or the above modules / units can be located in different processors in any combination.

[0257] The apparatus disclosed in this embodiment includes: an acquisition module 401 for acquiring target muddy interference seismic data; a conversion module 402 for performing data domain conversion on the target muddy interference seismic data to obtain converted seismic data; a training module 403 for training a neural network model based on the converted seismic data to obtain a trained neural network model; and an interference suppression module 404 for performing muddy suppression processing on the converted seismic data using the trained neural network model to obtain muddy-free seismic data of the target muddy interference seismic data. When performing muddy suppression using the apparatus disclosed in this embodiment, it is not necessary to synthesize a large amount of muddy sample seismic data. Instead, an unsupervised training network model is performed directly based on the muddy seismic data to be processed, and the trained network model is then directly used for muddy interference processing of the muddy seismic data. Compared with traditional sparse transform muddy suppression methods, this application has a wider range of applicability. Compared with traditional sparse transform muddy suppression methods, this application can more effectively protect seismic data, avoid loss of effective seismic data information, prevent tectonic artifacts, and improve the signal-to-noise ratio.

[0258] Example 4

[0259] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement all or part of the steps of the interference suppression method as described in Embodiment 1 or Embodiment 2.

[0260] Computer-readable storage media may individually include computer programs, data files, data structures, etc., or combinations thereof. The computer-readable storage media or computer program may be specifically designed and understood by those skilled in the art of computer software, or the computer-readable storage media may be known and available to those skilled in the art of computer software. Examples of computer-readable storage media include: magnetic media, such as hard disks, floppy disks, and magnetic tapes; optical media, such as CD-ROMs and DVDs; magneto-optical media, such as optical discs; and hardware devices specifically configured to store and execute computer programs, such as read-only memory (ROM), random access memory (RAM), flash memory; or servers, application stores, etc. Examples of computer programs include machine code (e.g., code generated by a compiler) and files containing high-level code that can be executed by a computer using an interpreter. The described hardware devices may be configured to function as one or more software modules to perform the operations and methods described above, and vice versa. Furthermore, computer-readable storage media may be distributed across networked computer systems, allowing for the decentralized storage and execution of program code or computer programs.

[0261] Example 5

[0262] Figure 5 A connection block diagram of an electronic device provided in an embodiment of this application, such as... Figure 5 As shown, the electronic device 500 may include: one or more processors 501, memory 502, multimedia components 503, input / output (I / O) interface 504, and communication components 505.

[0263] The memory 502 is used to store various types of data, such as instructions for any application or method in the electronic device, as well as application-related data. One or more processors 501 are used to execute all or part of the steps of the mixing interference suppression method as described in Embodiment 1 or Embodiment 2.

[0264] It should be noted that one or more processors 501 may be implemented as application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, for performing the methods described above.

[0265] The memory 502 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0266] Multimedia component 503 may include a screen, which may be a touchscreen, and an audio component for outputting and / or inputting audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory or transmitted via a communication component. The audio component also includes at least one speaker for outputting audio signals.

[0267] I / O interface 504 provides an interface between one or more processors 501 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons can be virtual buttons or physical buttons.

[0268] The communication component 505 is used for wired or wireless communication between the electronic device 500 and other devices. Wired communication includes communication via network ports, serial ports, etc.; wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, 5G, or one or more combinations thereof. Therefore, the corresponding communication component 505 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0269] In summary, this application discloses a method, apparatus, storage medium, and electronic device for suppressing seismic noise. Specifically, it discloses a method for suppressing seismic data noise based on a masking strategy and an unsupervised learning U-shaped network. First, the seismic data to be suppressed undergoes data domain transformation. Then, the seismic data is divided into patch seismic data through sliding segmentation. Seismic data augmentation techniques are then used to significantly increase the sample size and enrich the sample diversity. Different masking techniques are then used to randomly replace the masked patches. This approach abandons conventional supervised learning methods, avoiding the need to synthesize a large amount of mixed sample data by employing unsupervised learning, thus broadening the scope of seismic data to be processed. The data before and after masking are used as the input and expected output of a deep network. A network model built based on Unet or Unet++ is used for both the input and output of the seismic data, continuously training the network parameters to extract effective information features from the seismic data. By constructing an unsupervised seismic attenuation network model driven by the characteristics of the seismic data itself, it is directly applied to seismic data with mixed noise interference. The output of the network model is the seismic data after removing the mixed noise. By removing turbulence noise interference through an unsupervised autonomous network model, there is no need to synthesize a large amount of turbulent seismic data samples. Instead, the model starts directly from the turbulent seismic data itself and is trained on the characteristic structure of the turbulent seismic data through an unsupervised network. The trained model can then be directly used for turbulent seismic data processing, making it more widely applicable.

[0270] Compared with traditional sparse transform mixing suppression methods, this application can more effectively protect seismic data, avoid loss of effective seismic information, avoid causing tectonic artifacts, and also improve the signal-to-noise ratio.

[0271] It should also be understood that the methods or systems disclosed in the embodiments provided in this application can also be implemented in other ways. The method or system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functions, and operations of possible implementations of methods and apparatus according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, computer program segment, or part of a computer program, which includes one or more computer programs for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings, and may actually be executed substantially in parallel. They may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer programs.

[0272] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, apparatus, or apparatus that includes that element. The use of terms such as "first," "second," etc., is for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly indicating the number or sequence of the indicated technical features. In the description of this application, unless otherwise expressly defined, terms such as "shading interference seismic data," "data domain conversion," "segmentation," "augmentation," and "mask" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of these terms in this application based on the specific content of the technical solution. Furthermore, in the description of this application, unless otherwise stated, the terms "multiple" or "more" mean at least two.

[0273] Finally, it should be noted that in the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "a single example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0274] Although embodiments of this application have been shown and described above, it is to be understood that the above embodiments are exemplary and the content is only for the purpose of facilitating understanding of this application, and is not intended to limit this application. Any person skilled in the art to which this application pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope disclosed in this application, but the scope of protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A method of jamming suppression for a mixer, characterized by, The method comprises: acquiring target mixed interference seismic data; performing data domain conversion on the target mixed interference seismic data to obtain converted seismic data; training a neural network model according to the converted seismic data to obtain a trained neural network model; performing mixed wave suppression processing on the converted seismic data by using the trained neural network model to obtain seismic data after mixed wave removal of the target mixed interference seismic data; wherein the training of the neural network model according to the converted seismic data to obtain the trained neural network model comprises: performing patch cutting division on the converted seismic data to obtain a patch data set; performing augmentation expansion processing on the patch data set to obtain an expanded patch data set; acquiring training data according to the expanded patch data set; wherein the training data comprises model input data, and the acquiring of the training data according to the expanded patch data set comprises: performing mask processing on patch data in the expanded patch data set based on a preset mask strategy to obtain a masked patch data set; taking the masked patch data set as the model input data; training the neural network model according to the training data to obtain the trained neural network model; wherein the preset mask strategy comprises at least one of a point mask strategy, a block mask strategy and a strip mask strategy, and the mask processing on the patch data in the expanded patch data set based on the preset mask strategy comprises: acquiring a signal-to-noise ratio of the patch data; in a case where the signal-to-noise ratio of the patch data is greater than a first preset threshold, performing mask processing on the patch in the expanded patch data set based on the point mask strategy; in a case where the signal-to-noise ratio of the patch data is less than a second preset threshold, performing mask processing on the patch in the expanded patch data set based on the strip mask strategy; in a case where the signal-to-noise ratio of the patch data is not greater than the first preset threshold and not less than the second preset threshold, performing mask processing on the patch in the expanded patch data set based on the block mask strategy.

2. The method of claim 1, wherein, The training data further comprises model label data, and the acquiring of the training data according to the expanded patch data set further comprises: taking the expanded patch data set as the model label data.

3. The method of claim 1, wherein, The data domain conversion comprises: any one of common receiver domain conversion, common offset domain conversion and cross domain conversion.

4. The method of claim 1, wherein, The training of the neural network model according to the training data to obtain the trained neural network model comprises: setting initial network parameters of the neural network model; training the neural network model according to the model input data and the model label data of the training data to obtain optimized network parameters; taking the optimized network parameters as the network parameters of the neural network model to obtain the trained neural network model.

5. A jammer using the method of claim 1, wherein comprises: an acquisition module configured to acquire target mixed interference seismic data; A conversion module is configured to perform data domain conversion on the target mixed-wave interference seismic data to obtain converted seismic data. A training module is configured to train a neural network model according to the converted seismic data to obtain a trained neural network model. An interference suppression module is configured to perform mixed-wave suppression processing on the converted seismic data by using the trained neural network model to obtain mixed-wave removed seismic data of the target mixed-wave interference seismic data.

6. A computer-readable storage medium, characterized in that, The computer program stored in the computer readable storage medium can be executed by one or more processors to implement the mixed-wave interference suppression method according to any one of claims 1-4.

7. An electronic device, comprising: The computer program stored in the computer readable storage medium can be executed by one or more processors to implement the mixed-wave interference suppression method according to any one of claims 1-4. The computer program stored in the computer readable storage medium can be executed by one or more processors to implement the mixed-wave interference suppression method according to any one of claims 1-4.

Images