Data processing method, electronic device, storage medium and computer program product

Abstract

The invention discloses a data processing method, electronic equipment, a storage medium and a computer program product, and the method comprises the steps: converting first seismic data into corresponding first wave field data; curvelet transformation is carried out on the first wave field data to obtain a first curvelet coefficient set; extracting one or more first curvelet coefficients corresponding to a first diffracted wave from the first curvelet coefficient set through a first model; and processing the one or more first curvelet coefficients to obtain an image of the first diffracted wave. Through the method and the device, the diffracted waves in the seismic data can be separated more accurately, so that the imaging precision of the diffracted waves is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, electronic device, storage medium, and computer program product. Background Technology

[0002] Diffraction waves are the seismic responses of subsurface discontinuities such as scatterers, fractures, faults, and river channels, and are closely related to hydrocarbon migration and accumulation. However, the energy of diffracted waves is usually much weaker than that of reflected waves generated by large-scale reflecting structures. Therefore, subsurface discontinuities are inevitably masked by large-scale reflectors in seismic migration imaging. To characterize subsurface discontinuities, it is necessary to separate diffracted waves from seismic data and perform diffracted wave imaging. However, the accuracy of diffracted wave separation is low, resulting in low precision of the final diffracted wave imaging. Summary of the Invention

[0003] To address the related technical problems, embodiments of this application provide a data processing method, an electronic device, a storage medium, and a computer program product.

[0004] The technical solution of this application embodiment is implemented as follows: This application provides a data processing method, the method comprising: The first earthquake data is converted into the corresponding first wavefield data; Perform a curvelet transform on the first wave field data to obtain the first curvelet coefficient set; One or more first curvature coefficients corresponding to the first diffraction wave are extracted from the first curvature coefficient set using the first model; The first curvature coefficients are processed to obtain an image of the first diffracted wave.

[0005] In the above scheme, processing the one or more first curvature coefficients to obtain an image of the first diffracted wave includes: Perform inverse curvature transformation on the one or more first curvature coefficients to obtain second wavefield data; Least-squares reverse time-shift imaging is performed on the second wavefield data to obtain an image of the first diffracted wave.

[0006] In the above scheme, the first model is obtained by training one or more first training data and a first label corresponding to each first training data. The first training data is the data in the first dataset that has undergone curvelet transformation to obtain a set of curvelet coefficients. The first label is used to identify the curvelet coefficients corresponding to the diffracted waves in the first training data.

[0007] In the above scheme, the first model includes one or more of the following modules: A channel attention module is used to enhance one or more first channels, which are used to identify diffraction wave related features. A spatial attention module is used to retain the spatial location information related to the first diffracted wave.

[0008] In the above scheme, the step of performing a curvelet transform on the first wavefield data to obtain the first curvelet coefficient set includes: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; The step of performing an inverse curvature transform on the one or more first curvature coefficients to obtain second wavefield data includes: The second wavefield data is obtained by performing an inverse generalized curvilinear transform on one or more of the first curvilinear coefficients.

[0009] The method in the above scheme further includes: The first parameter is adjusted according to the first indicator, wherein the first indicator includes one or more of the following: the ability of the generalized curvelet transform to distinguish different angular directions; the ability of the generalized curvelet transform to distinguish different scale components; the remaining computing resources; the computing resources occupied due to running the first model; the remaining memory resources; the memory resources occupied due to running the first model; and the first parameter includes one or more of the following: radial scaling parameter; angular scaling parameter. The step of performing curvelet transform on the first wavefield data includes: Curved wave transformation is performed on the first wavefield data based on a generalized mother curved wave, wherein the frequency domain representation of the generalized mother curved wave is determined based on the adjusted first parameter.

[0010] In the above scheme, after converting the first seismic data into corresponding first wavefield data, the method further includes: The first seismic data is subjected to a first processing to obtain a corresponding first processing result. The first processing includes one or more of the following: τ-p transform; singular spectrum analysis; The transform domain data corresponding to the second diffracted wave is extracted from the first processing result using the first model. Wherein, if the first processing includes τ-p transform, the transform domain data corresponding to the second diffracted wave includes τ-p domain data corresponding to the second diffracted wave; if the first processing includes singular spectrum analysis, the transform domain data corresponding to the second diffracted wave includes singular spectrum space data corresponding to the second diffracted wave. The transform domain data corresponding to the second diffracted wave is processed to obtain an image of the second diffracted wave.

[0011] In the above scheme, the first dataset includes one or more of the following: The data obtained by performing forward modeling on the second model, which only includes the reflection layer and diffraction points; Data obtained from forward modeling of salt body structures; Wavefield data obtained by converting the second earthquake data.

[0012] This application also provides a data processing apparatus, including: The first conversion unit is used to convert the first seismic data into the corresponding first wavefield data; The second conversion unit is used to perform a curvilinear transformation on the first wave field data to obtain a first set of curvilinear coefficients. The first extraction unit is used to extract one or more first curvature coefficients corresponding to the first diffraction wave from the first curvature coefficient set using a first model. The first processing unit is used to process the one or more first curve coefficients to obtain an image of the first diffracted wave.

[0013] This application also provides an electronic device, including: a processor and a memory for storing a computer program capable of running on the processor. The processor is used to execute the steps of any of the above-mentioned technical solutions when running the computer program.

[0014] This application also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above methods.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above methods.

[0016] As can be seen from the above, in this embodiment, the first seismic data is converted into corresponding first wavefield data, and a curvelet transform is performed on the first wavefield data to obtain a first curvelet coefficient set. One or more first curvelet coefficients corresponding to the first diffracted wave are extracted from the first curvelet coefficient set using a first model, and these one or more first curvelet coefficients are processed to obtain an image of the first diffracted wave. It is evident that this embodiment combines a first model and curvelet transform to separate the curvelet coefficients corresponding to the diffracted wave in the seismic data, and processes these curvelet coefficients to obtain an image of the diffracted wave. Due to the characteristics of curvelet transform, it can more accurately capture the relevant features of the diffracted wave, while the first model can reduce human intervention in the diffracted wave process. The combination of the two can more accurately separate the diffracted wave in the seismic data, thereby improving the accuracy of diffracted wave imaging. Attached Figure Description

[0017] Figure 1 A schematic flowchart illustrating the data processing method provided in the application embodiments of this application; Figure 2 A schematic diagram of the architecture of the first model provided for the application embodiments of this application; Figure 3 A flowchart illustrating Embodiment 1 provided for application of this application; Figure 4 A schematic diagram illustrating an example of the generalized curvilinear transform provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] It should be understood that the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, the term "one or more" in this document is an exemplary expression and can be replaced with any possible expression, such as one or more, at least one, or at least one item, etc.

[0020] It should also be understood that the term "instruction" mentioned in the embodiments of this application can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0021] It should also be understood that the term "correspondence" mentioned in the embodiments of this application may indicate a direct or indirect correspondence between the two, or an association between the two, or a relationship of instruction and being instructed, configuration and being configured, etc.

[0022] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0023] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0024] Diffraction waves are the seismic responses of subsurface discontinuities such as scatterers, fractures, faults, and river channels, and are closely related to hydrocarbon migration and accumulation. However, the energy of diffracted waves is usually much weaker than that of reflected waves generated by large-scale reflecting structures; therefore, subsurface discontinuities are inevitably masked by large-scale reflectors in seismic migration imaging. To characterize subsurface discontinuities, diffraction wave separation and imaging are important problems that need to be solved.

[0025] Among related technologies, a data-domain diffraction wave separation scheme can be employed. This scheme converts pre-stack seismic data into common-offset gathers or plane wave gathers to highlight the geometric differences between diffracted and reflected waves. Subsequently, techniques such as dip filtering, plane wave deconstruction filtering, Generalized Curvelet Transform (GCT), or Radon transform are used to separate the diffracted wave field from the seismic data. This scheme can also be based on a diffraction wave separation method using superimposed common reflection elements to separate diffracted waves from seismic data with low signal-to-noise ratios. Deep learning techniques can also be used for diffraction wave separation. Labels for the training set can be generated through physical modeling, synthetic data summation, or by directly applying a plane wave deconstruction filter to the entire wavefield. Network architectures include U-net, generative adversarial networks, and other convolutional neural networks (CNNs).

[0026] Besides using data-domain diffraction separation schemes, imaging-domain schemes can also be employed to separate diffraction waves. In these schemes, diffraction waves are separated using different gathers, such as dip-domain common imaging point gathers, local imaging matrices, and reflection angle gathers. These methods identify diffraction and reflection waves within the migration gathers and superimpose the diffraction waves to generate an image of subsurface discontinuities. Since diffraction waves are correlated with positive and negative dip structure images, diffraction imaging can be obtained through dip-correlated wavefield multiplication and used to calculate other seismic properties to characterize discontinuities. However, migration imaging is typically generated by conventional ray-based or wave equation-based migration operators, which treat the adjoint of the forward operator as its inverse. Due to irregular acquisition geometries and the narrow frequency bands of seismic data, these methods can produce low-resolution migration images with artifacts. Compared to these migration methods, Least-Squares Reverse Time Migration (LSRTM) can suppress artifacts and improve image resolution by iteratively updating the imaging results to fit the observation data.

[0027] However, deep learning-based diffraction wave separation schemes are highly dependent on the quality and quantity of training data. For complex seismic data, the separation effect is unstable. Furthermore, the "black box" nature of neural network models limits the understanding and debugging of their internal working principles, affecting their reliability and interpretability in practical applications. GCT-based diffraction wave separation methods require empirical selection of the coefficient matrices of diffracted and reflected waves for signal reconstruction. Since the distribution of curve wave coefficients changes with seismic data, achieving good separation results requires time-windowing for large-scale data and significant time investment in selecting curve wave coefficients, which cannot meet the demands of efficient processing. Because reflected and diffracted waves with similar angles and scales are difficult to distinguish, the separation results may contain significant residual reflected waves or lose some diffraction wave information. When migrating and imaging the separated diffracted waves, conventional reverse-time migration is essentially the convolution of the subsurface reflection coefficient and the Hessian matrix. Because the calculation uses the conjugate transpose of the matrix for approximation, the results suffer from migration noise and low resolution, leading to low accuracy in the final diffraction wave imaging.

[0028] To address at least one of the aforementioned technical problems, this application proposes a data processing method. In this embodiment, first seismic data is converted into corresponding first wavefield data; a curvelet transform is performed on the first wavefield data to obtain a first set of curvelet coefficients; one or more first curvelet coefficients corresponding to a first diffracted wave are extracted from the first set of curvelet coefficients using a first model; and the one or more first curvelet coefficients are processed to obtain an image of the first diffracted wave. It is evident that this application combines a first model and curvelet transform to separate the curvelet coefficients corresponding to diffracted waves in seismic data, and processes these coefficients to obtain an image of the diffracted wave. Due to the characteristics of curvelet transform, it can more accurately capture the relevant features of diffracted waves, while the first model can reduce human intervention in the diffracted wave process. The combination of these two methods can more accurately separate diffracted waves from seismic data, thereby improving the accuracy of diffracted wave imaging.

[0029] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0030] This application uses an electronic device as an example of an execution subject, and a data processing device as one of the manifestations of an electronic device. This application does not limit the manifestation of electronic devices.

[0031] Please see Figure 1 The data processing method provided in this application includes: Step 101: Convert the first seismic data into the corresponding first wavefield data; For example, the seismic data (e.g., first seismic data) in the embodiments of this application are also referred to as seismic data. Seismic data refers to the original data collected in seismic exploration or seismic detection, such as pre-stack seismic data, or the original seismic data recorded by a detector.

[0032] For example, the wavefield data (e.g., first wavefield data, second wavefield data, etc.) in the embodiments of this application include one or more of the following: plane wave gathers, common offset gathers, tilt gathers, and common reflection point gathers.

[0033] Step 102: Perform a curvelet transform on the first wave field data to obtain the first curvelet coefficient set.

[0034] For example, the set of curve wave coefficients in the embodiments of this application is an exemplary statement and can be replaced with any possible statement, such as a curve wave coefficient matrix, such as multiple curve wave coefficients, etc.

[0035] For example, the step of performing a curvelet transform on the first wavefield data to obtain a first set of curvelet coefficients includes one or more of the following: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; Perform a uniform discrete curvilinear transformation on the first wavefield data to obtain the first curvilinear coefficient set; Subband decomposition and ridge wave analysis are performed on the first wavefield data to determine the curved singular features in the first wavefield data, thereby obtaining the first set of curved wave coefficients.

[0036] Understandably, the generalized curvelet transform can preserve the geometric structure information in seismic data, and it can more effectively capture the response of curved or edge-shaped geological structures, such as small-scale heterogeneous structures like diffraction waves, thereby improving the accuracy of diffraction wave separation. The uniform discrete curvelet transform can control redundancy and improve efficiency. The method of subband decomposition and ridge wave analysis has strong nonlinear approximation capabilities, which improves the accuracy of diffraction wave separation.

[0037] Understandably, in practical applications, the values ​​of u and v will affect the scale and angular resolution of GCT. When both approach the optimal value... At this time, the resolution is high, but it is accompanied by a large amount of computation and storage. Therefore, u and v can be adjusted according to specific problems (such as the first index below) to change the anisotropy relationship.

[0038] In practical applications, the method further includes: The first parameter is adjusted according to the first indicator, wherein the first indicator includes one or more of the following: the ability of the generalized curvelet transform to distinguish different angular directions; the ability of the generalized curvelet transform to distinguish different scale components; the remaining computing resources; the computing resources occupied due to running the first model; the remaining memory resources; the memory resources occupied due to running the first model; and the first parameter includes one or more of the following: radial scaling parameter; angular scaling parameter. The step of performing curvelet transform on the first wavefield data includes: Curved wave transformation is performed on the first wavefield data based on a generalized mother curved wave, wherein the frequency domain representation of the generalized mother curved wave is determined based on the adjusted first parameter.

[0039] In this embodiment, by performing a curvelet transform on the first wavefield data based on a generalized mother curvelet, and making the frequency domain representation of the generalized mother curvelet dependent on the adjusted first parameter, the curvelet transform process can improve its flexibility and adaptability, thereby better matching the resource conditions and separation requirements under different application scenarios, and thus improving the robustness and accuracy of the diffraction wave separation process.

[0040] For example, the generalized mother curve wave can be represented in the frequency domain as: (3) in, Represents the scaling matrix of the parabola; Indicated by radians Rotate; Indicates rotation angle The unit vector; u and v represent the radial scaling parameter and the angular scaling parameter, respectively.

[0041] For example, adjusting the first parameter includes adjusting u and v in the above formula.

[0042] For example, in high-resolution scenarios, the ability to distinguish between components at different scales can be enhanced by increasing the number of scale sampling layers and combining them with a narrower radial window function; while in resource-constrained scenarios, the computational and storage overhead can be reduced while ensuring the ability to extract the main features, by decreasing the number of scale layers or increasing the bandwidth of the radial window function.

[0043] For example, in order to achieve better wavefield separation, the first wavefield data can be divided into time windows.

[0044] Step 103: Extract one or more first curvilinear coefficients corresponding to the first diffraction wave from the first curvilinear coefficient set using the first model.

[0045] For example, diffracted waves refer to the seismic response caused by underground discontinuities (such as cracks, faults, river channels, etc.).

[0046] For example, the first model includes one or more of the following: U-Net network; generative adversarial model; ResNet, DenseNet; Transformer model.

[0047] For example, a typical application of this application embodiment is that the first model is a U-Net network. Other deep learning models, such as ResNet, DenseNet or Transformer, can be introduced on the basis of the U-Net network to form a hybrid network structure. Through the residual connection or attention mechanism of the network, the diffraction wave related features can be extracted more efficiently.

[0048] For example, one or more first curvature coefficients corresponding to the first diffracted wave are extracted from the first curvature coefficient set using a first model, including one or more of the following: The first set of curvelet coefficients is convolved by the convolutional layer of the encoder of the first model; The result of the convolution operation is downsampled using a downsampling layer. The features associated with one or more first curve coefficients corresponding to the first diffracted wave are enhanced by using a Convolutional Block Attention Module (CBAM). The original size of the feature map obtained by the encoder is restored by upsampling, and the features from different layers are fused by skip connections.

[0049] For example, taking the U-Net network, which is an Encoder-Decoder architecture for semantic segmentation, as an example, the U-Net network includes an encoder, a convolutional attention module, a decoder, etc.

[0050] For example, the encoder of the first model includes convolutional layers and downsampling layers.

[0051] For example, the above convolutional layer includes a modified linear unit. For instance, each convolutional layer performs two convolution operations, with a kernel size of 3×3 and a stride of 1. Zero padding is used during the operation to ensure that the feature map sizes are consistent, and the number of feature maps is doubled after the first convolution operation.

[0052] For example, the downsampling layer can perform downsampling operations, such as using a convolution operation with a kernel size of 4×4 and a stride of 2×2 instead of traditional max pooling.

[0053] For example, the convolutional attention module of the first model includes a channel attention module and / or a spatial attention module, as can be seen in the following practical applications: In practical applications, the first model includes one or more of the following modules: A channel attention module is used to enhance one or more first channels, which are used to identify diffraction wave related features. A spatial attention module is used to retain the spatial location information related to the first diffracted wave.

[0054] In this embodiment, by introducing a channel attention module and a spatial attention module into the first model, automatic enhancement of diffraction wave features and accurate preservation of spatial location information are achieved. The channel attention module and the spatial attention module complement each other, together forming a complete attention mechanism system. This attention mechanism system further improves the overall performance of the model, thereby enhancing the automation and reliability of diffraction wave separation, and thus providing strong support for high-resolution imaging of underground geological structures.

[0055] For example, in the embodiments of this application, the channel attention module can enhance the first channel that is sensitive to diffraction waves, the first channel identifying the relevant features of the curvature sparsity of the diffraction waves.

[0056] For example, a channel attention module is an attention mechanism component in a neural network. In this embodiment, the channel attention module calculates the importance weights of different feature channels, weighting and enhancing the feature channels used to identify diffraction waves, thereby improving the model's ability to perceive target features (such as diffraction waves). Specifically, the channel attention module typically combines global average pooling and a multilayer perceptron (MLP) to generate a channel attention map. This channel attention map is then multiplied by the original feature map to amplify key features related to diffraction waves and suppress irrelevant information. In seismic data processing, the channel attention module can enhance features such as frequency and scale related to diffraction waves, making these features easier to separate and identify in subsequent processing.

[0057] For example, the spatial attention module is also an attention mechanism component in a neural network. Its main function is to model the spatial distribution of input features, identify regions associated with diffracted waves, and retain the spatial location information related to the first diffracted wave. Specifically, the spatial attention module typically generates a spatial attention map based on the local neighborhood relationships of the feature map using convolution operations. This spatial attention map is then multiplied with the original feature map to highlight important regions and remove background noise. For instance, when processing seismic data, the spatial attention module can help retain the geometric structure information of the first diffracted wave at a specific location, avoiding positional shifts or loss caused by reflected wave interference.

[0058] For example, the decoder of the first model includes an upsampling layer to recover the original size of the feature map and to fuse features from different layers using skip connections.

[0059] For example, please refer to Figure 2 , Figure 2 Taking the U-net network as an example, the first model includes four crossover lines. The input to the convolutional layer is the concatenation result of the feature map from the previous layer and the feature map from the encoder. After the first convolution operation, the number of feature maps is halved, and after the second operation, the number remains unchanged. After the upsampling layer performs a transpose convolution operation on the feature map, the number of feature maps is halved, and the size is doubled. The last layer obtains the curve coefficients corresponding to the diffracted wave by convolving with a 1×1 convolution kernel.

[0060] The training process of the first model in the embodiments of this application is described below: In practical applications, the first model is obtained by training one or more first training data and a first label corresponding to each first training data. The first training data is obtained by performing curvelet transformation on the data in the first dataset to obtain a curvelet coefficient set. The first label is used to identify the curvelet coefficient corresponding to the diffraction wave in the first training data.

[0061] In this embodiment, the training data needs to be pre-converted to be curved, so that the data used for model training is consistent with the "curved wave transformation + first model" architecture in this embodiment, thereby improving the training effect of the model and further improving the accuracy of the model in separating diffracted waves.

[0062] For example, the first label can serve as a supervision label for the first model. The first label can be generated using methods such as plane wave deconstruction filters.

[0063] The following example illustrates how the first model in this application is trained: For example, before training the first model, it is necessary to construct a first dataset (i.e., a training dataset). The process of constructing the first dataset can refer to the following practical application: In practical applications, the first dataset includes one or more of the following: The data obtained by performing forward modeling on the second model, which only includes the reflection layer and diffraction points; Data obtained from forward modeling of salt body structures; Wavefield data obtained by converting the second earthquake data.

[0064] The different data sources in the first dataset have different adaptation conditions and application scenarios: In this embodiment of the application, by defining and constructing a diverse first dataset, this embodiment of the application can provide richer training samples, improve the model's adaptability and generalization ability to different geological conditions, and achieve more accurate diffraction wave separation effect.

[0065] It should be noted that the embodiments of this application do not limit the conversion of the second seismic data into wavefield data. For example, if the second seismic data is already wavefield data, it is not necessary to convert the second seismic data into wavefield data.

[0066] For example, the second model is a simplified geological model, typically containing only reflection layers and diffraction points, without complex reflection structures. By performing forward modeling on the second model, a relatively pure diffraction wave signal can be obtained, improving the network's ability to identify diffraction wave features. Using the data obtained from forward modeling the second model to train the first model makes it suitable for detailed exploration tasks with high separation requirements.

[0067] For example, salt body construction models include one or more of the following: Sigsbee2A model, 2D SEG / EAGE salt dome model.

[0068] For example, the data obtained by forward modeling a salt body structure model refers to seismic response data generated from the salt body structure model through numerical simulation methods. A salt body structure model is a more complex geological model that includes heterogeneous structures such as salt domes and salt walls. By performing forward modeling on the salt body structure model, seismic responses containing rich geological information can be simulated. This seismic response data is suitable for training models to identify diffraction wave characteristics in complex geological backgrounds. Training a first model with data obtained from forward modeling a salt body structure model makes the first model suitable for diffraction wave extraction tasks in high-noise environments in actual exploration scenarios.

[0069] For example, the second seismic data is actual acquired seismic data, such as a set of marine carbonate reservoir data from the Gulf of Mexico. By preprocessing and transforming this raw seismic data, clearer wavefield data can be obtained. The wavefield data obtained through transformation of the second seismic data can be applied to the model training process. Using this type of wavefield data when training a neural network model can improve the model's adaptability to real-world data.

[0070] For example, the first label of the curve coefficient corresponding to the diffracted wave can be obtained by using a plane wave deconstruction filter.

[0071] For example, after constructing the first dataset, it is necessary to perform curvelet transform on the data in the first dataset to obtain the first training data. The specific curvelet transform process can be referred to in step 102. The first data includes reflected wave data and diffracted wave data.

[0072] For example, with the first training data as input and the curve coefficients corresponding to the diffracted wave as the desired output, the mean square error can be used as the loss function during the training process of the first model. The initial learning rate is set to 0.001, and the learning rate is decayed using a cosine annealing algorithm with hot restart. The batch size is set to 48, the epoch is 120, and the Adam algorithm is used to optimize the learning objective.

[0073] In practical applications, after converting the first seismic data into corresponding first wavefield data, the method further includes: The first seismic data is subjected to a first processing to obtain a corresponding first processing result. The first processing includes one or more of the following: τ-p transform; singular spectrum analysis; The transform domain data corresponding to the second diffracted wave is extracted from the first processing result using the first model. Wherein, if the first processing includes τ-p transform, the transform domain data corresponding to the second diffracted wave includes τ-p domain data corresponding to the second diffracted wave; if the first processing includes singular spectrum analysis, the transform domain data corresponding to the second diffracted wave includes singular spectrum space data corresponding to the second diffracted wave. The transform domain data corresponding to the second diffracted wave is processed to obtain an image of the second diffracted wave.

[0074] In this embodiment, high-quality first processing results are extracted by performing τ-p transform and / or singular spectrum analysis on the first wavefield data. Then, the generalized curvelet transform coefficients corresponding to the second diffracted wave are automatically identified and extracted using a first model. Finally, a high-resolution image of the second diffracted wave is obtained through inverse transform and LSRTM imaging. This method effectively separates the weak-energy second diffracted wave from the first wavefield data, improving the accuracy and reliability of imaging and thus better characterizing the distribution characteristics of small-scale discontinuities in the subsurface. Furthermore, this method can provide more precise technical support for fields such as geophysical exploration and geological hazard monitoring.

[0075] It is understood that the embodiments of this application are not limited to converting the first seismic data into a first set of curvature coefficients through curvature transformation. In the embodiments of this application, step 102 can also be replaced by "performing a first process on the first seismic data". Correspondingly, step 103 can be replaced by "extracting the transform domain data corresponding to the second diffraction wave from the first processing result through a first model". Step 104 can be replaced by "processing the transform domain data corresponding to the second diffraction wave to obtain the image of the second diffraction wave".

[0076] For example, in the embodiments of this application, the curve wave transform and the first processing can be combined. For instance, the first model can extract the transform domain data corresponding to the second diffraction wave from the first processing result, and the first model can also extract one or more first curve wave coefficients corresponding to the first diffraction wave from the first curve wave coefficient set. Then, the transform domain data corresponding to the second diffraction wave and one or more first curve wave coefficients corresponding to the first diffraction wave are combined to obtain the image corresponding to the diffraction wave in the first seismic data.

[0077] For example, the slope-intercept (τ-p) transform is a method for converting common-detector point gathers into plane wave gathers. The core idea of ​​the τ-p transform is to time-shift the first wavefield data and map it to the slope-intercept domain (τ-p domain), thereby separating wavefield components with different propagation directions. The τ-p transform can effectively highlight the geometric differences between the second diffracted wave and the reflected wave, and it can improve the interpretability of the data.

[0078] For example, Singular Spectrum Analysis (SSA) is a non-parametric signal decomposition technique that can extract the main trends, periodicities, and noise components from complex time series. SSA reconstructs and reduces noise by performing Singular Value Decomposition (SVD) on the data matrix. In this embodiment, SSA can be used to extract local structural features from the first wavefield data and to enhance the identifiability of the second diffracted wave.

[0079] Step 104: Process the one or more first curve coefficients to obtain an image of the first diffracted wave.

[0080] In practical applications, processing the one or more first curve coefficients to obtain an image of the first diffracted wave includes: Perform inverse curvature transformation on the one or more first curvature coefficients to obtain second wavefield data; Least-squares reverse time-shift imaging is performed on the second wavefield data to obtain an image of the first diffracted wave.

[0081] In this embodiment, inverse warp transform can more accurately reconstruct signals with complex geometries, such as diffracted waves generated by underground discontinuities. Using inverse warp transform helps improve the resolution and accuracy of subsequent imaging. Least squares reverse time migration imaging can effectively suppress imaging noise and improve image resolution. Furthermore, it can gradually correct imaging results through multiple iterations, thereby further improving imaging accuracy.

[0082] For example, the inverse curvature transform is an operation that restores data in the curvature domain (e.g., one or more first curvature coefficients) to spatiotemporal domain data (e.g., second wavefield data).

[0083] For example, least squares reverse time migration imaging of the second wavefield data includes: continuously adjusting the parameters of the imaging model according to the objective function of least squares reverse time migration and using a numerical optimization algorithm, and obtaining the image of the first diffracted wave based on the second wavefield data through the imaging model.

[0084] For example, the objective function of plane wave least squares reverse time migration is defined as the difference between diffracted waves predicted by diffracted waves separated from the first seismic data (e.g., diffracted waves corresponding to the second wavefield data obtained through inverse generalized curve wave transform) using diffracted wave imaging results. Plane wave least squares reverse time migration can be solved by the conjugate gradient method.

[0085] For example, as described above, the second wave field data can be obtained through steps 101-103 and then through the inverse generalized curvelet transform. The first model and the curvelet transform (e.g., the generalized curvelet transform) constitute the diffraction wave separation operator. Taking the second wave field data as a plane wave channel set as an example, the process of processing through the inverse curvelet transform in steps 101-103 is equivalent to abstracting the diffraction wave separation process into a diffraction wave separation operator acting on the plane wave channel set, thereby obtaining the separated diffraction wave.

[0086] For example, the prediction of diffraction wave imaging results can be obtained by calculating the Born forward modeling operator for predicting diffraction waves.

[0087] In practical applications, the step of performing a curvelet transform on the first wavefield data to obtain the first curvelet coefficient set includes: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; The step of performing an inverse curvature transform on the one or more first curvature coefficients to obtain second wavefield data includes: The second wavefield data is obtained by performing an inverse generalized curvilinear transform on one or more of the first curvilinear coefficients.

[0088] In this embodiment, by introducing generalized curvelet transform and inverse generalized curvelet transform, automatic and efficient separation and reconstruction of diffracted wave components in the seismic wavefield are achieved. By introducing generalized curvelet transform and inverse generalized curvelet transform and combining them with the U-Net network to separate diffracted wave components, weak-energy diffracted wave signals can be effectively extracted, thereby improving the detection accuracy of underground discontinuities and providing more reliable technical support for geophysical exploration, engineering geology and other fields.

[0089] It should be noted that the inverse generalized curvelet transform is the reverse process of the generalized curvelet transform. The inverse generalized curvelet transform is used to remap the curvelet coefficients corresponding to the separated diffracted waves back to the spatiotemporal domain, generating clear diffracted wave data.

[0090] For example, the inverse generalized curvilinear transform can be performed based on a generalized mother curvilinear wave, which could be, for instance, a generalized mother curvilinear wave. .

[0091] As can be seen from the above, the embodiments of this application propose a diffraction wave separation method that integrates curvelet transform (e.g., GCT) with a neural network model (i.e., the first model, such as a U-Net network). The curvelet transform is used to extract features at different scales, and the first model learns the corresponding patterns and features through training data, thereby achieving automatic and efficient separation of the curvelet coefficients of diffracted and reflected waves. This reduces manual intervention in the selection of curvelet coefficients, thereby improving the separation accuracy. The separated data is then processed by LSRTM to obtain a high-resolution image with a small-scale heterogeneous structure.

[0092] The present application will be further described in detail below with reference to application examples.

[0093] Example 1 Example 1 uses the wavefield data as a plane wave gather or a common offset gather as an example. Example 1 uses the curve transform as a generalized curve transform. Example 1 uses the inverse curve transform as an inverse generalized curve transform as an example. Example 1 uses the first model as a U-Net network as an example.

[0094] Please see Figure 3 , Figure 3 Here is an exemplary flowchart of Example 1. Figure 3 Includes the following steps: S301: Acquire first earthquake data; For example, converting the first seismic data into corresponding first wavefield data includes: calculating the plane wavelet set according to the following formula:

[0095] in, It is by The detector at that location records a plane wave channel gather, which is a wave gather with an exit angle of . and frequency The seismic wave field reflected back to the Earth's surface from the plane wave source; It is located in The common detector point gather received by the detector at the location; This indicates a time shift of the common detector gather; These are the horizontal coordinates of the earthquake source; is the ray parameter of the plane wave channel set; v is the near-surface velocity.

[0096] S302: Convert the first seismic data into a plane wave gather or a common offset gather; S303: Perform a generalized curvilinear transform on the plane wave gather or common offset gather to obtain the first curvilinear coefficient set; For example, performing a generalized curvelet transform on a plane wave channel gather or a common offset channel gather includes: performing a generalized curvelet transform on each plane wave channel gather according to the following equation (2): (2) in, The first set of curve coefficients is calculated, where j, l, and k represent the scale, angle, and position, respectively; The plane wave channel set calculated in equation (1) is... The generalized "mother" curve wave; its frequency domain representation is shown in equation (3): (3) in, Represents the scaling matrix of the parabola; Indicated by radians Rotate; Indicates rotation angle The unit vector; u and v represent the radial scaling parameter and the angular scaling parameter, respectively.

[0097] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating an example of the generalized curvilinear transform provided in an embodiment of this application. Figure 4 In the figure, (a) represents the angle of departure. The planar channel set, (a) describes the relationship between distance and time. Figure 4 In the diagram, (b) represents the curve coefficient corresponding to the plane wave channel gather (a), and (c) represents the exit angle. The plane wave channel set, (d) is the curve coefficient corresponding to the plane wave channel set (c). Figure 4 It shows the launch angle as and The planar wave channel set and its curvature coefficients are shown, where diffracted and reflected waves differ significantly in angle and scale. The curvature coefficients of reflected waves are concentrated within the angular range corresponding to the exit angle, while the curvature coefficients of diffracted waves are divided into two halves due to the decomposition of the angle.

[0098] Understandably, in practical applications, the values ​​of u and v will affect the scale and angular resolution of GCT. When both approach a certain value... At this time, the resolution is high, but it is accompanied by a large amount of computation and storage. Therefore, u and v can be adjusted according to specific problems (such as the first indicator) to change the anisotropy relationship.

[0099] For example, in order to achieve better wavefield separation, the first wavefield data can be divided into time windows.

[0100] S304: Framework for building U-Net networks; For example, the U-Net network can be an Encoder-Decoder based U-Net network, which can be used to identify and separate curvelet coefficients. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a schematic diagram of the U-Net network provided in an embodiment of this application. Figure 2 The input to the model is a graph showing the relationship between the curvature angle or curvature degree and the scale.

[0101] For example, the encoder of the U-Net network includes convolutional layers and downsampling layers: For example, the convolutional layer can obtain the feature map corresponding to the first set of curvelet coefficients using the following formula: (4) in, This represents the feature map of the k-th layer; BN() represents modified linear unit; BN() represents batch normalization. This represents the weight matrix of the k-th layer; This represents the bias matrix of the k-th layer; * represents the convolution operation. Each layer performs two convolution operations with a kernel size of 3×3 and a stride of 1. Zero padding is used during the operation to ensure that the feature map sizes are consistent. The number of feature maps is doubled after the first convolution operation.

[0102] For example, a downsampling layer can perform a downsampling operation using the following formula: (5) Here, Down() represents the downsampling operation. To retain more information, a convolution operation with a kernel size of 4×4 and a stride of 2 is used instead of traditional max pooling.

[0103] For example, to improve the feature extraction capability of the model, a convolutional attention module is integrated into the U-Net network. The convolutional attention module of the U-Net network includes a channel attention module and / or a spatial attention module.

[0104] For example, channel attention can be used to process relevant data using the following formula:

[0105] in, Indicates channel attention; represents the sigmoid function; AvgPool() represents average pooling; MaxPool() represents max pooling; MLP() represents a multilayer perceptron.

[0106] For example, the spatial attention module can process relevant data using the following formula: in, This indicates channel attention; Cat() indicates the concatenation operation. This indicates a convolution operation with a 7×7 kernel.

[0107] For example, an attention mechanism can be applied to feature map a: (8) in, The calculation can be referenced from formula (6). The calculation can be referenced from formula (7).

[0108] For example, the decoder includes an upsampling layer that restores the original size of the feature map and utilizes skip connections to fuse features from different layers.

[0109] For example, the U-Net network includes four crossover lines. The input to the convolutional layer is the concatenation result of the feature map from the previous layer and the feature map from the encoder, which can be represented as: (9) in, This represents the feature map of the l-th layer in the decoder; This represents the feature map from the encoder. In the convolution operation, the kernel size is 3×3, and the stride is 1. After the first convolution operation, the number of feature maps is halved; after the second operation, the number remains unchanged. The upsampling layer performs a transpose convolution on the feature maps, halving the number of feature maps and doubling their size. The final layer convolves with a 1×1 kernel to obtain the curve coefficients corresponding to the diffracted wave.

[0110] S305: Construct the first dataset and train the U-Net network; For example, when constructing the first dataset, models containing only reflection layers and only diffraction points are first forward modeled to obtain reflected wave data and diffracted wave data. Labels used to identify diffracted wave data (e.g., the curve coefficients corresponding to diffracted waves) are used as labels for the first training data. The first dataset includes superimposed data of diffracted and reflected waves. Secondly, in order to improve the network's data processing capabilities, data obtained by forward modeling more complex Sigsbee2A models and 2D SEG / EAGE salt dome models are used as part of the first dataset for training the U-Net network. The first labels for the curve coefficients corresponding to diffracted waves are obtained by using a plane wave deconstruction filter. Finally, a set of marine carbonate reservoir data from the Gulf of Mexico is used to increase the diversity of the first dataset and improve the generalization ability of the U-Net network.

[0111] The following examples illustrate how the U-Net network in this application is trained: The first set of curvature coefficients obtained from Formula 2 It also contains reflected wave data. and diffraction wave data : (10) by As input, the diffraction wave curvature coefficient To obtain the desired output, we can derive the following formula: (11) in, W and b are both neural network parameters; Unet() represents the processing procedure of the U-Net network.

[0112] During the training of the U-Net network, mean squared error is used as the loss function: (12) in, Represents the loss function; Denotes the Frobenius norm; This represents N training data-label pairs. The initial learning rate is set to 0.001, and the learning rate is decayed using a cosine annealing algorithm with hot restart. The batch size is set to 48, the epoch is 120, and the Adam algorithm is used to optimize the learning objective.

[0113] S306: Extract one or more first curvature coefficients corresponding to the first diffraction wave from one or more first curvature coefficients using a U-Net network; For example, the trained U-net network can extract the curve coefficients corresponding to the diffracted wave, as shown in the following formula: (13) in, The curve coefficients corresponding to the separated diffracted wave; The full-wave field to be separated corresponds to the curved wave coefficients (i.e., the first set of curved wave coefficients). This is the trained U-Net network.

[0114] S307: Perform inverse generalized curvilinear transform on one or more first curvilinear coefficients corresponding to the first diffracted wave; The inverse curvature transform can be performed on one or more first curvature coefficients using the following formula:

[0115] in, This represents the separated spatiotemporal diffraction wave data. For generalized mother curve wave; This represents the first curvature coefficient.

[0116] S308: Perform least-squares reverse time-shift imaging on the results obtained from the inverse generalized curve transform.

[0117] It is understandable that the result obtained from the inverse generalized curve transform is the second wave field data.

[0118] For example, the objective function of plane wave least-squares reverse time migration is defined as the difference between the diffracted wave separated from observed seismic data and the diffracted wave predicted using diffracted wave imaging results: (15) in, Let L be the objective function for least-squares reverse-time migration of diffracted waves in the plane wave domain; L is the Born forward modeling operator used to predict diffracted waves; m is the diffracted wave imaging result; F is the diffracted wave separation operator composed of GCT and U-Net network; D is the plane wave channel set synthesized from observation data; this equation can be solved using the conjugate gradient method.

[0119] As can be seen from the above, this application proposes a method for diffraction wave separation using a U-net network in the curvelet domain. First, the spatiotemporal domain data (i.e., wavefield data) is transformed from GCT to the curvelet domain, decomposed into curvelet coefficient matrices of different scales and angles, and the curvelet coefficients corresponding to the diffracted and reflected waves are separated using the U-Net model. The separated diffracted wave data is then transformed back to the spatiotemporal domain, and finally, LSRTM is used for migration imaging.

[0120] More specifically, this application proposes a diffraction wave separation scheme combining generalized curvelet transform and U-Net network, and presents an algorithm flow and strategy for U-Net network training and inverse transform processing based on the curvelet domain. Compared with related technologies, the U-Net network in this application can automatically learn the diffraction wave features extracted from the curvelet domain through training, realizing the identification and separation of curvelet coefficients, enhancing separation accuracy and reducing dependence on manual coefficient selection. It can also handle large-scale data, improving computational efficiency. Generalized curvelet transform can effectively capture multi-scale and multi-directional features of data and perform sparse representation of the data, highlighting features crucial for diffraction wave separation, reducing the learning burden of the U-Net network, and improving training speed. Compared with purely data-driven methods, the method combined with a physical model has stronger interpretability and credibility.

[0121] From an application perspective, the curve domain diffraction wave separation scheme based on an Encoder-Decoder network architecture proposed in this application can effectively extract weak energy diffraction waves in the entire wave field, providing technical support for multiple fields such as geophysical exploration, signal processing, and geological disaster monitoring. In ultrasound imaging, by separating and amplifying weak signals, the scheme provided in this application can improve the detectability of lesions and enhance the accuracy of medical images. In acoustic testing, the scheme provided in this application can be used to detect microcracks, defects, or other small anomalies in metal structures or materials, improving the reliability of industrial product quality control and non-destructive testing. In the field of engineering geology, the scheme provided in this application can be used for geological exploration of buildings and infrastructure, such as the detection of fine structures in areas like underground tunnels and bridge foundations, identifying potential hazards such as cracks and cavities, and assessing the stability of the foundation. In the field of marine exploration, shipborne sonar technology is used to measure seabed topography and water depth, obtaining high-resolution seabed information by analyzing sound wave reflections. The scheme provided in this application can help extract detailed information related to geological structures from sonar data, such as scattering points with drastic changes in wave impedance.

[0122] Based on the embodiments described above, this application also provides a data processing apparatus, see [link to previous document]. Figure 5The data processing device includes: The first conversion unit 501 is used to convert the first seismic data into the corresponding first wavefield data; The second conversion unit 502 is used to perform a curvilinear transformation on the first wave field data to obtain a first curvilinear coefficient set. The first extraction unit 503 is used to extract one or more first curvature coefficients corresponding to the first diffraction wave from the first curvature coefficient set using a first model. The first processing unit 504 is used to process the one or more first curve coefficients to obtain an image of the first diffraction wave.

[0123] In one embodiment, the first processing unit 504 processes the one or more first curve coefficients to obtain an image of the first diffracted wave, including: Perform inverse curvature transformation on the one or more first curvature coefficients to obtain second wavefield data; Least-squares reverse time-shift imaging is performed on the second wavefield data to obtain an image of the first diffracted wave.

[0124] In one embodiment, the first model is obtained by training one or more first training data and a first label corresponding to each of the first training data. The first training data is obtained by performing curvelet transformation on the data in the first dataset to obtain a curvelet coefficient set. The first label is used to identify the curvelet coefficient corresponding to the diffraction wave in the first training data.

[0125] In one embodiment, the first model includes one or more of the following modules: A channel attention module is used to enhance one or more first channels, which are used to identify diffraction wave related features. A spatial attention module is used to retain the spatial location information related to the first diffracted wave.

[0126] In one embodiment, the second conversion unit 502 performs a curvelet transform on the first wavefield data to obtain a first set of curvelet coefficients, including: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; The step of performing an inverse curvature transform on the one or more first curvature coefficients to obtain second wavefield data includes: The second wavefield data is obtained by performing an inverse generalized curvilinear transform on one or more of the first curvilinear coefficients.

[0127] In one embodiment, the device further includes a first adjustment unit, the first adjustment unit being configured to: adjust a first parameter according to a first index, wherein the first index includes one or more of the following: the ability of the generalized curvelet transform to distinguish different angular directions; the ability of the generalized curvelet transform to distinguish different scale components; remaining computing resources; computing resources occupied due to running the first model; remaining memory resources; memory resources occupied due to running the first model; and the first parameter includes one or more of the following: radial scaling parameters; angular scaling parameters; The second conversion unit 502 performs curvelet transform on the first wavefield data, including: Curved wave transformation is performed on the first wavefield data based on a generalized mother curved wave, wherein the frequency domain representation of the generalized mother curved wave is determined based on the adjusted first parameter.

[0128] In one embodiment, the device further includes a second processing unit, a second extraction unit, and a third processing unit. After converting the first seismic data into corresponding first wavefield data, the second processing unit performs a first processing on the first seismic data to obtain a corresponding first processing result. The first processing includes one or more of the following: τ-p transform; singular spectrum analysis. The second extraction unit is used to extract the transform domain data corresponding to the second diffracted wave from the first processing result using the first model. Wherein, when the first processing includes τ-p transform, the transform domain data corresponding to the second diffracted wave includes the τ-p domain data corresponding to the second diffracted wave; when the first processing includes singular spectrum analysis, the transform domain data corresponding to the second diffracted wave includes the singular spectrum space data corresponding to the second diffracted wave. The third processing unit is used to process the transform domain data corresponding to the second diffracted wave to obtain an image of the second diffracted wave.

[0129] In one embodiment, the first dataset includes one or more of the following: The data obtained by performing forward modeling on the second model, which only includes the reflection layer and diffraction points; Data obtained from forward modeling of salt body structures; Wavefield data obtained by converting the second earthquake data.

[0130] In practical applications, the first conversion unit 501, the second conversion unit 502, the first extraction unit 503, the first processing unit 504, the first adjustment unit, the second processing unit, the second extraction unit, and the third processing unit can be implemented by the processor in the communication device.

[0131] It should be noted that the data processing apparatus provided in the above embodiments is only illustrated by the division of the above program modules. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the apparatus can be divided into different program modules to complete all or part of the processing described above. In addition, the data processing apparatus and data processing method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0132] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of this application, this application also provides an electronic device, see [link to relevant documentation]. Figure 6 The electronic device includes: The first communication interface 1 is capable of exchanging information with other devices; The first processor 2 is configured to convert the first seismic data into corresponding first wavefield data, perform curvelet transform on the first wavefield data to obtain a first curvelet coefficient set, extract one or more first curvelet coefficients corresponding to the first diffraction wave from the first curvelet coefficient set using a first model, and process the one or more first curvelet coefficients to obtain an image of the first diffraction wave.

[0133] In one embodiment, the first processor 2 processes the one or more first curve coefficients to obtain an image of the first diffracted wave, including: Perform inverse curvature transformation on the one or more first curvature coefficients to obtain second wavefield data; Least-squares reverse time-shift imaging is performed on the second wavefield data to obtain an image of the first diffracted wave.

[0134] In one embodiment, the first model is obtained by training one or more first training data and a first label corresponding to each of the first training data. The first training data is obtained by performing curvelet transformation on the data in the first dataset to obtain a curvelet coefficient set. The first label is used to identify the curvelet coefficient corresponding to the diffraction wave in the first training data.

[0135] In one embodiment, the first model includes one or more of the following modules: A channel attention module is used to enhance one or more first channels, which are used to identify diffraction wave related features. A spatial attention module is used to retain the spatial location information related to the first diffracted wave.

[0136] In one embodiment, the first processor 2 performs a curvelet transform on the first wavefield data to obtain a first set of curvelet coefficients, including: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; The step of performing an inverse curvature transform on the one or more first curvature coefficients to obtain second wavefield data includes: The second wavefield data is obtained by performing an inverse generalized curvilinear transform on one or more of the first curvilinear coefficients.

[0137] In one embodiment, the first processor 2 is further configured to: adjust a first parameter according to a first index, wherein the first index includes one or more of the following: the ability of the generalized curvelet transform to distinguish different angular directions; the ability of the generalized curvelet transform to distinguish different scale components; remaining computing resources; computing resources occupied due to running the first model; remaining memory resources; memory resources occupied due to running the first model; and the first parameter includes one or more of the following: radial scaling parameter; angular scaling parameter; The first processor 2 performs curvelet transform on the first wavefield data, including: Curved wave transformation is performed on the first wavefield data based on a generalized mother curved wave, wherein the frequency domain representation of the generalized mother curved wave is determined based on the adjusted first parameter.

[0138] In one embodiment, the first processor 2 is further configured to: The first seismic data undergoes a first processing to obtain a corresponding first processing result. The first processing includes one or more of the following: τ-p transform; singular spectrum analysis; The transform domain data corresponding to the second diffracted wave is extracted from the first processing result using the first model. Wherein, if the first processing includes τ-p transform, the transform domain data corresponding to the second diffracted wave includes τ-p domain data corresponding to the second diffracted wave; if the first processing includes singular spectrum analysis, the transform domain data corresponding to the second diffracted wave includes singular spectrum space data corresponding to the second diffracted wave. The transform domain data corresponding to the second diffracted wave is processed to obtain an image of the second diffracted wave.

[0139] In one embodiment, the first dataset includes one or more of the following: The data obtained by performing forward modeling on the second model, which only includes the reflection layer and diffraction points; Data obtained from forward modeling of salt body structures; Wavefield data obtained by converting the second earthquake data.

[0140] It should be noted that the specific processing procedure of the first communication interface 1 can be understood by referring to the above method.

[0141] Of course, in practical applications, the various components in an electronic device are coupled together through bus system 4. It can be understood that bus system 4 is used to achieve communication and connection between these components. In addition to the data bus, bus system 4 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 6 The general will label all buses as Bus System 4.

[0142] The first memory 3 in this embodiment is used to store various types of data to support operation in the electronic device. Examples of such data include any computer program used to operate on the electronic device.

[0143] The methods disclosed in the embodiments of this application can be applied to the first processor 2, or implemented by the first processor 2. The first processor 2 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware or by instructions in the form of software in the first processor 2. The first processor 2 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The first processor 2 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the first memory 3. The first processor 2 reads the information in the first memory 3 and completes the steps of the aforementioned method in combination with its hardware.

[0144] In an exemplary embodiment, the electronic device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned methods.

[0145] It is understood that the first memory 3 in this application embodiment can be a volatile memory pool or a non-volatile memory pool, or both. The non-volatile memory pool can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory pool can be a disk storage pool or a magnetic tape storage pool. The volatile memory pool can be a random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The storage pools described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of storage pools.

[0146] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a first memory 3 storing a computer program, which can be executed by a first processor 2 to complete the steps described in the aforementioned method.

[0147] In an exemplary embodiment, this application also provides a computer program product, including a computer program that can be executed by a first processor 2 to perform the steps described in the foregoing method.

[0148] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A data processing method, characterized in that, The data processing method includes: The first earthquake data is converted into the corresponding first wavefield data; Perform a curvelet transform on the first wave field data to obtain the first curvelet coefficient set; One or more first curvature coefficients corresponding to the first diffraction wave are extracted from the first curvature coefficient set using the first model; The first curvature coefficients are processed to obtain an image of the first diffracted wave.

2. The method according to claim 1, characterized in that, The step of processing the one or more first curvature coefficients to obtain an image of the first diffracted wave includes: Perform inverse curvature transformation on the one or more first curvature coefficients to obtain second wavefield data; Least-squares reverse time-shift imaging is performed on the second wavefield data to obtain an image of the first diffracted wave.

3. The method according to claim 1, characterized in that, The first model is obtained by training one or more first training data and a first label corresponding to each first training data. The first training data is obtained by performing curvelet transformation on the data in the first dataset to obtain a curvelet coefficient set. The first label is used to identify the curvelet coefficient corresponding to the diffraction wave in the first training data.

4. The method according to claim 3, characterized in that, The first model includes one or more of the following modules: A channel attention module is used to enhance one or more first channels, which are used to identify diffraction wave related features. A spatial attention module is used to retain the spatial location information related to the first diffracted wave.

5. The method according to claim 2, characterized in that, The step of performing a curvelet transform on the first wavefield data to obtain a first set of curvelet coefficients includes: Perform a generalized curvilinear transform on the first wavefield data to obtain the first curvilinear coefficient set; The step of performing an inverse curvature transform on the one or more first curvature coefficients to obtain second wavefield data includes: The second wavefield data is obtained by performing an inverse generalized curvilinear transform on one or more of the first curvilinear coefficients.

6. The method according to claim 1, characterized in that, The method further includes: The first parameter is adjusted according to the first indicator, wherein the first indicator includes one or more of the following: the ability of the generalized curvelet transform to distinguish different angular directions; the ability of the generalized curvelet transform to distinguish different scale components; the remaining computing resources; the computing resources occupied due to running the first model; the remaining memory resources; the memory resources occupied due to running the first model; and the first parameter includes one or more of the following: radial scaling parameter; angular scaling parameter. The step of performing curvelet transform on the first wavefield data includes: Curved wave transformation is performed on the first wavefield data based on a generalized mother curved wave, wherein the frequency domain representation of the generalized mother curved wave is determined based on the adjusted first parameter.

7. The method according to claim 1, characterized in that, After converting the first seismic data into corresponding first wavefield data, the method further includes: The first seismic data is subjected to a first processing to obtain a corresponding first processing result. The first processing includes one or more of the following: τ-p transform; singular spectrum analysis; The transform domain data corresponding to the second diffracted wave is extracted from the first processing result using the first model. Wherein, if the first processing includes τ-p transform, the transform domain data corresponding to the second diffracted wave includes τ-p domain data corresponding to the second diffracted wave; if the first processing includes singular spectrum analysis, the transform domain data corresponding to the second diffracted wave includes singular spectrum space data corresponding to the second diffracted wave. The transform domain data corresponding to the second diffracted wave is processed to obtain an image of the second diffracted wave.

8. The method according to claim 3, characterized in that, The first dataset includes one or more of the following: The data obtained by performing forward modeling on the second model, which only includes the reflection layer and diffraction points; Data obtained from forward modeling of salt body structures; Wavefield data obtained by converting the second earthquake data.

9. A data processing apparatus, characterized in that, include: The first conversion unit is used to convert the first seismic data into the corresponding first wavefield data; The second conversion unit is used to perform a curvilinear transformation on the first wave field data to obtain a first set of curvilinear coefficients. The first extraction unit is used to extract one or more first curvature coefficients corresponding to the first diffraction wave from the first curvature coefficient set using a first model. The first processing unit is used to process the one or more first curve coefficients to obtain an image of the first diffracted wave.

10. An electronic device, characterized in that, include: The processor and the memory used to store computer programs that can run on the processor. When the processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 8.

11. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images