Pre-stack common image point gather enhancement method and device

Abstract

This specification relates to the technical field of seismic data processing, and specifically discloses a method and apparatus for enhancing pre-stack common imaging point gathers. The method includes: obtaining a pre-stack common imaging point gather, and extracting a common offset seismic data set from the pre-stack common imaging point gather; generating a standard image based on the common offset seismic data in the pre-stack common imaging point gather with offsets within a preset range; generating a to-be-registered image based on the common offset seismic data in the pre-stack common imaging point gather with offsets outside the preset range; using a deep learning data-driven model to learn the global feature relationship between the standard image and the to-be-registered image to obtain the deformation parameters of the to-be-registered image; registering the to-be-registered image using the deformation parameters to obtain a registered image; and generating an enhanced common imaging point gather based on the standard image and the registered image. The above solution can improve the accuracy and efficiency of pre-stack common imaging point gather enhancement.

Description

Technical Field

[0001] This specification relates to the technical field of seismic data processing, and particularly relates to a method and device for enhancing pre-stack common image gather. Background Art

[0002] In the process of seismic data processing, achieving accurate and clear imaging of geological structures is an important task, which plays a very important role in subsequent studies of underground geological structures. With the development of oil exploration and development technologies, small-structure complex hydrocarbon reservoirs and lithologic hydrocarbon reservoirs that were previously restricted and undeveloped have also become exploration targets. These unconventional hydrocarbon reservoirs have stronger concealment and more complex hydrocarbon accumulation laws, which require a greater improvement in the accuracy of seismic exploration. In the past, researchers often used post-stack seismic data with high signal-to-noise ratio. Post-stack seismic data can reflect the strength and continuity of seismic events, but it cannot obtain information on how the reflection coefficient changes with the offset, that is, information on how the amplitude of the reflected wave changes with the offset. Pre-stack gathers, on the other hand, have not been stacked, have a large amount of data, and contain rich information, especially offset information, which is closely related to reservoir lithology and fluid properties.

[0003] Ideally, the common image gather (CIG gather) generated by pre-stack migration corresponds to the same geological formation position and geometric structure. It is expected that the traces in the common image gather maintain a high degree of waveform and phase consistency. However, due to the influence of migration velocity errors, anisotropy, etc., the geological structure is not correctly imaged. The actual imaging points deviate from the true imaging points in the horizontal and vertical directions, resulting in ineffective flattening of seismic events in the common image gather. The common offset gather (COG gather) extracted from the common image gather can be regarded as multiple observations of the same underground imaging. The horizontal and vertical misalignments of imaging points in different COG gathers will reduce the resolution and signal-to-noise ratio of the stacked section, and lead to a decrease in the accuracy of subsequent seismic attribute analysis, seismic interpretation, and seismic inversion. Therefore, it is necessary to enhance the common image gather, that is, to process seismic data to reduce the horizontal and vertical misalignments of imaging points in the common offset gather.

[0004] Currently, traditional methods for enhancing common image gathers mainly rely on methods such as single-trace cross-correlation and time-frequency analysis to adaptively estimate and eliminate the correction amount in the vertical direction, and have a certain effect of eliminating residual moveout. However, these methods focus more on the misalignment between each imaging point and the true imaging point in the vertical direction, ignoring the impact of horizontal misalignment on the stacking quality. The subsequent image calibration technology performs local cross-correlation calculations from a two-dimensional perspective, which can only eliminate misalignment to a certain extent and cannot take into account the overall structural structure, resulting in insufficient accuracy and efficiency in enhancing the common image gather, and having certain limitations.

[0005] For the above problems, no effective solutions have been proposed yet. Summary of the Invention

[0006] An embodiment of this specification provides a method and device for enhancing pre-stack common imaging point gathers to solve the problem of low accuracy in the pre-stack common imaging point gather enhancement method in the prior art.

[0007] An embodiment of this specification provides a method for enhancing pre-stack common imaging point gathers, including:

[0008] Obtain a pre-stack common imaging point gather, and extract a common offset seismic data set from the pre-stack common imaging point gather;

[0009] Based on the common offset seismic data with offsets within a preset range in the common offset seismic data set, generate a standard image; based on the common offset seismic data with offsets outside the preset range in the common offset seismic data set, generate an image to be registered;

[0010] Use a deep learning data-driven model to learn the global feature relationship between the standard image and the image to be registered, and obtain the deformation parameters of the image to be registered;

[0011] Register the image to be registered using the deformation parameters to obtain a registered image; based on the standard image and the registered image, generate an enhanced common imaging point gather.

[0012] In one embodiment, obtaining a pre-stack common imaging point gather and extracting a common offset seismic data set from the pre-stack common imaging point gather includes:

[0013] Obtain a pre-stack seismic data set; perform migration on the pre-stack seismic data set to obtain a pre-stack common imaging point gather;

[0014] Extract a common offset seismic data set from the pre-stack common imaging point gather;

[0015] Perform preprocessing on the common offset seismic data set to obtain a preprocessed common offset seismic data set.

[0016] In one embodiment, during the iterative training process of the deep learning data-driven model, input the standard image and the image to be registered into the deep learning data-driven model to obtain a preliminary registration result, calculate the loss value of the preliminary registration result through a loss function, determine whether the loss value meets a preset condition, if it does not meet the preset condition, update the model parameters of the deep learning data-driven model, and enter the next iteration cycle until the obtained loss value meets the preset condition.

[0017] In one embodiment, the deep learning data-driven model includes a siamese neural network; the siamese neural network includes an encoder and a decoder;

[0018] The encoder is used to extract seismic features of the standard image and the image to be registered; the seismic features include similarity features and difference features between the standard image and the image to be registered;

[0019] The decoder is used to reconstruct the extracted seismic features to obtain the deformation parameters of the image to be registered.

[0020] In one embodiment, the loss function is:

[0021]

[0022] where L(f, m, φ) is the loss function, is the similarity evaluation loss function, L smooth (φ) is the smooth regularization term loss function, f represents the standard image, m represents the image to be registered, φ represents the deformation parameter, and λ is a hyperparameter.

[0023] In one embodiment, registering the image to be registered using the deformation parameter to obtain a registered image includes:

[0024] Registering the image to be registered using the deformation parameter through a spatial transformation network to obtain a registered image; wherein, the spatial transformation network includes a grid generator and a resampler; the grid generator is used to generate a deformation field using the deformation parameter; the resampler is used to apply the deformation field to the image to be registered to obtain a registered image.

[0025] In one embodiment, after generating an enhanced common image point gather based on the standard image and the registered image, it further includes:

[0026] Generating a pre-registration horizontal stack profile and / or full stack profile according to the standard image and the image to be registered; generating a post-registration horizontal stack profile and / or full stack profile according to the standard image and the registered image;

[0027] Determining the enhancement effect of the common image point gather according to the pre- and post-registration common image point gathers and the pre- and post-registration horizontal stack profiles and / or full stack profiles.

[0028] This embodiment of the specification also provides a pre-stack common image point gather enhancement device, including:

[0029] An extraction module, configured to obtain a pre-stack common image point gather and extract a common offset seismic data set from the pre-stack common image point gather;

[0030] A generation module, configured to generate a standard image based on the common-offset seismic data in the common-offset seismic dataset with the offset within a preset range; and further configured to generate a to-be-registered image based on the common-offset seismic data in the common-offset seismic dataset with the offset outside the preset range.

[0031] A learning module, configured to learn the global feature relationship between the standard image and the to-be-registered image by using a deep learning data-driven model, and obtain the deformation parameters of the to-be-registered image.

[0032] A registration module, configured to register the to-be-registered image by using the deformation parameters to obtain a registered image; and further configured to generate an enhanced common image point gather based on the standard image and the registered image.

[0033] An embodiment of this specification further provides a computer device, including a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the steps of the pre-stack common image point gather enhancement method described in any of the above embodiments are implemented.

[0034] An embodiment of this specification further provides a computer-readable storage medium, on which computer instructions are stored. When the instructions are executed, the steps of the pre-stack common image point gather enhancement method described in any of the above embodiments are implemented.

[0035] In the embodiments of this specification, the global feature relationship between the standard image and the image to be registered is learned by a deep learning data-driven model, which can give play to the advantages of seismic big data, and can take into account geological laws such as overall structural consistency and local waveform similarity, and can fully utilize the deep learning data-driven model's ability to extract seismic big data features, so as to achieve the purpose of promoting the high-precision and high-efficiency enhancement of pre-stack common imaging point gathers through artificial intelligence. By using the deep learning data-driven model to extract features of profile images of different offsets, deformation parameters are generated, and the calibration of the image to be registered is completed, thereby enhancing the consistency of the common phase axis in the pre-stack common imaging point gathers, and improving the resolution of the stacked profile, which has a strong industrial application value. Through the above scheme, the problem that the traditional gather enhancement method only performs calculations based on low-dimensional local similarity, resulting in residual time differences in the vertical direction in the common imaging point gathers, can be alleviated. The deep learning data-driven model based on mining high-dimensional information based on seismic data features is combined with geological knowledge, and with the help of image registration technology, a set of high-precision and high-efficiency pre-stack common imaging point gather enhancement methods that consider geological structure consistency and local waveform similarity is constructed. Under the guidance of geological knowledge such as geological structure consistency and local waveform similarity, this method gives full play to the powerful information mining ability and feature extraction advantages of the deep learning data-driven model, and can specifically analyze the characteristics between seismic profiles at different offset distances, laying the foundation for the subsequent generation of deformation fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of this specification, constitute a part of this specification, and do not constitute a limitation of this specification. In the drawings:

[0037] Figure 1 A flow chart of a method for enhancing pre-stack common imaging point gathers in an embodiment of this specification is shown;

[0038] Figure 2 A flow chart showing a method for enhancing pre-stack common imaging point gathers in an embodiment of this specification is shown;

[0039] Figure 3 shows an unsupervised image registration structure diagram in an embodiment of this specification;

[0040] Figure 4 A schematic diagram of a twin neural network in an embodiment of this specification is shown;

[0041] Figure 5 A schematic diagram of a space transformation network in an embodiment of the present specification is shown;

[0042] Figure 6 The Sigsbee2B seismic data in one embodiment of this specification is shown;

[0043] Figure 7 shows the seismic profiles at different offsets in an embodiment of this specification;

[0044] Figure 8 shows a comparison diagram of the image registration effect in an embodiment of this specification;

[0045] Figure 9 shows a vector visualization schematic diagram of the deformation field in an embodiment of this specification;

[0046] Figure 10 shows the horizontal stacked profiles before and after registration in an embodiment of this specification;

[0047] Figure 11 shows the full stacked profiles before and after registration in an embodiment of this specification;

[0048] Figure 12 shows Figure 11 a schematic diagram shown by some samples in;

[0049] Figure 13 shows the common image point gather before (left) and after (right) registration of the 20th trace in the test set in an embodiment of this specification;

[0050] Figure 14 shows a schematic diagram of a pre-stack common image point gather enhancement device in an embodiment of this specification;

[0051] Figure 15 shows a schematic diagram of a computer device in an embodiment of this specification. Detailed implementation manners

[0052] The principles and spirit of this specification will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only to enable those skilled in the art to better understand and then implement this specification, and do not limit the scope of this specification in any way. On the contrary, these embodiments are provided to make the disclosure of this specification more thorough and complete, and to convey the scope of this disclosure fully to those skilled in the art.

[0053] Those skilled in the art know that the embodiments of this specification can be implemented as a system, device, equipment, method, or computer program product. Therefore, the disclosure of this specification can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0054] In order to enhance the pre-stack common imaging point gather and further improve the subsequent stacking imaging quality, the inventors found through research that the advantages of efficiently extracting seismic data features by artificial intelligence can be combined with geological knowledge. With the help of unsupervised image registration technology, a pre-stack common imaging point gather enhancement method and device considering geological structure consistency and local waveform similarity are proposed. The method and device rely on a deep learning data-driven model to analyze different common offset gather data, deeply explore the similarities and differences in morphology, structure, phase, etc. between the to-be-registered images with medium and far offsets and the standard images with near offsets, and then generate a deformation field. At the same time, relying on a spatial transformation network, efficient and accurate calibration of the to-be-registered images is achieved; by combining the advantages of deep learning in artificial intelligence in mining large-scale seismic data with the geological laws summarized by experts and scholars, the pre-stack common imaging point gather is enhanced with the help of image registration technology. Through this solution, important support can be provided for subsequent work such as seismic imaging, structural interpretation, and oil and gas reservoir prediction.

[0055] Based on this, an embodiment of this specification provides a pre-stack common imaging point gather enhancement method. Figure 1 The flowchart of the pre-stack common imaging point gather enhancement method in an embodiment of this specification is shown. Although this specification provides method operation steps or device structures as shown in the following embodiments or drawings, more or fewer operation steps or module units may be included in the method or device based on routine or non-creative labor. In steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure described in the embodiments of this specification and shown in the drawings. When the method or module structure is applied to an actual device or terminal product, it can be executed sequentially or in parallel according to the method or module structure connection shown in the embodiments or drawings (for example, in an environment of parallel processors or multi-threaded processing, or even a distributed processing environment).

[0056] Specifically, as Figure 1 shown, the pre-stack common imaging point gather enhancement method provided in an embodiment of this specification may include the following steps.

[0057] Step S101, obtain a pre-stack common imaging point gather, and extract a common offset seismic data set from the pre-stack common imaging point gather.

[0058] The method in the embodiment of this specification can be applied to a computer device. A pre-stack common imaging point gather can be obtained, and a common offset seismic data set can be extracted from the pre-stack common imaging point gather.

[0059] In some embodiments of the present specification, obtaining a pre-stack common image gather and extracting a common offset seismic data set from the pre-stack common image gather may include: obtaining a pre-stack seismic data set; performing migration on the pre-stack seismic data set to obtain a pre-stack common image gather; extracting a common offset seismic data set from the pre-stack common image gather; and performing preprocessing on the common offset seismic data set to obtain a preprocessed common offset seismic data set.

[0060] Specifically, a pre-stack seismic data set can be obtained. The pre-stack seismic data set has not been stacked, has a large amount of data, and contains rich information, especially offset information, which is closely related to reservoir lithology and fluid characteristics. Perform migration on the pre-stack seismic data set to obtain a pre-stack common image gather. A common offset seismic data set can be extracted from the pre-stack common image gather. Data augmentation and normalization and other preprocessing can be performed on the common offset seismic data set to obtain a preprocessed common offset seismic data set.

[0061] Step S102: Generate a standard image based on the common offset seismic data in the pre-set range of offsets in the common offset seismic data set; generate a to-be-registered image based on the common offset seismic data outside the pre-set range of offsets in the common offset seismic data set.

[0062] After obtaining the common offset seismic data set, a standard image and a to-be-registered image can be established. The images corresponding to the common offset seismic data within the pre-set range can be stacked to obtain a standard image. In one embodiment, the multiple common offset seismic data within the near offset range can be stacked as the standard image. The images corresponding to each offset seismic data among the multiple common offset seismic data within the far offset range can be used as the to-be-registered images. Exemplarily, in one embodiment, the offset range in the common offset seismic data set is 200 - 4000 meters, with a total of 39 offsets; that is, starting from an offset of 200 meters, there is an offset interval of every 100 meters. In this exemplary embodiment, the first three offset profiles are selected and stacked as the standard image, that is, 200 - 500 meters, that is, the near offset range can be, for example, 200 - 500 meters. Correspondingly, the images corresponding to each offset within the offset range of 600 - 4000 meters can be used as the to-be-registered images.

[0063] Step S103: Use a deep learning data-driven model to learn the global feature relationship between the standard image and the to-be-registered image to obtain the deformation parameters of the to-be-registered image.

[0064] After obtaining the standard image and the image to be registered, a deep learning data-driven model can be used to learn the global feature relationship between the standard image and the image to be registered, and the deformation parameters between the standard image and the image to be registered can be obtained. Among them, the deformation parameters are used to characterize the horizontal correction amount and vertical correction amount of each imaging point corresponding to the image to be registered, and are used to eliminate the horizontal displacement and vertical displacement of the image to be registered. The global feature relationship can include the similarities and differences in features such as morphology, structure, and phase between the standard image and the image to be registered.

[0065] Overall, different offset profiles (i.e., the image to be registered and the standard image) should be consistent for the underground geological structure; locally, there should also be a certain similarity in the polarity of the seismic waveforms at the same position. Therefore, under the guidance of the above geological knowledge, combined with the powerful feature extraction ability of the deep learning data-driven model, similarity features and difference features are extracted from the pre-stack seismic profiles with different offsets, and the deformation parameters are obtained therefrom.

[0066] Step S104, register the image to be registered by using the deformation parameters to obtain a registered image; generate an enhanced common image point gather based on the standard image and the registered image.

[0067] After obtaining the deformation parameters of the image to be registered, the deformation parameters can be used to register the image to be registered to obtain a registered image. That is, applying the horizontal correction amount and vertical correction amount of each imaging point in the deformation parameters to the corresponding imaging point of the image to be registered can obtain the corresponding registered image. Based on the standard image and the registered image, an enhanced common image point gather can be generated.

[0068] In the above embodiment, the global feature relationship between the standard image and the image to be registered is learned by the deep learning data-driven model, which can give play to the advantages of seismic big data, and can take into account geological laws such as overall structural consistency and local waveform similarity, and can fully utilize the deep learning data-driven model's ability to extract seismic big data features, so as to achieve the purpose of promoting the high-precision and high-efficiency enhancement of pre-stack common imaging point gathers through artificial intelligence. By using the deep learning data-driven model to extract features of profile images of different offset distances, the similarity between them is obtained and deformation parameters are generated, and the calibration of the image to be registered is completed, thereby enhancing the consistency of the common phase axis in the pre-stack common imaging point gathers, and improving the resolution of the stacked profile, which has a strong industrial application value. Through the above scheme, the problem that the traditional gather enhancement method only performs calculations based on low-dimensional local similarity, resulting in residual time differences in the vertical direction in the common imaging point gathers, can be alleviated. The deep learning data-driven model based on mining high-dimensional information based on seismic data features is combined with geological knowledge, and with the help of image registration technology, a set of high-precision and high-efficiency pre-stack common imaging point gather enhancement methods that consider geological structure consistency and local waveform similarity is constructed. Under the guidance of geological knowledge such as geological structure consistency and local waveform similarity, this method gives full play to the powerful information mining ability and feature extraction advantages of the deep learning data-driven model, and can specifically analyze the characteristics between seismic profiles at different offset distances, laying the foundation for the subsequent generation of deformation fields.

[0069] In some embodiments of the present specification, during the iterative training process of the deep learning data-driven model, the standard image and the image to be registered are input into the deep learning data-driven model to obtain a preliminary registration result, and the loss value of the preliminary registration result is calculated by a loss function to determine whether the loss value meets a preset condition. If the preset condition is not met, the model parameters of the deep learning data-driven model are updated and the next iteration cycle is entered until the obtained loss value meets the preset condition.

[0070] The deep learning data-driven model in this embodiment can be iteratively trained through unsupervised learning algorithms. Specifically, during the iterative training of the model, the to-be-registered image and the standard image are fed into the deep learning data-driven model to obtain a preliminary registration result. Subsequently, the loss value is calculated through the loss function, and the parameters of the deep learning data-driven model are updated and enter the next iteration cycle. By iterating the above process multiple times, a refined registration result can be finally obtained. By using an unsupervised method, the workload of seismic processing personnel can be reduced, and the purpose of promoting high-precision and high-efficiency enhancement of pre-stack common image point gathers by artificial intelligence can be achieved. By using the deep learning data-driven model to extract the features of two different offset section images, the similarity between them is obtained and the deformation parameters are generated. At the same time, combined with the unsupervised image registration method, the calibration of the to-be-registered image can be completed without manually making labels, thereby enhancing the coherence of the in-phase axis in the pre-stack common image point gather and improving the resolution of the stacked section, which has strong industrial application value.

[0071] In some embodiments of this specification, the deep learning data-driven model includes a siamese neural network; the siamese neural network includes an encoder and a decoder; the encoder is used to extract the seismic features of the standard image and the to-be-registered image; the seismic features include the similarity features and the difference features between the standard image and the to-be-registered image; the decoder is used to reconstruct the data of the extracted seismic features to obtain the deformation parameters of the to-be-registered image.

[0072] In this embodiment, the deep learning data-driven model used can adopt a siamese neural network. The siamese neural network architecture mainly consists of an encoder and a decoder. Among them, the encoder part can select the convolutional part of the classic convolutional neural network VGG-16, which includes a total of 5 convolutional stages (conv1 to conv5), and each convolutional stage consists of a convolutional layer and a pooling layer. Among them, the network parameter sharing is not set in the conv1 stage of the network, so that the two coupled networks can learn their respective features separately; in the conv2 and conv3 stages of the network, the parameter sharing of the network allows the network to jointly learn common features; then the feature maps obtained by the two coupled networks are concatenated and jointly trained for conv4 and conv5, and finally the upsampling of the network is realized through a completely symmetric decoder to finally obtain the deformation parameters required by the subsequent network. The decoder part can be completely symmetric with the encoder part and consists of a convolutional layer and an upsampling layer. The encoder function is mainly responsible for extracting seismic features with different levels and different semantics of the input data, and also includes the similarity and difference features between the two input data. The decoder function mainly reconstructs the extracted seismic features to obtain the deformation parameters of the image to be registered. Through the siamese neural network, a lightweight deep learning data-driven model can be constructed, which can improve the efficiency of pre-stack common image gather enhancement.

[0073] Those skilled in the art can understand that the deep learning data-driven model can also adopt other forms of neural networks, such as the U-Net neural network. Correspondingly, different neural networks have different advantages and disadvantages, and the effects that can be achieved are also different.

[0074] Considering that the deformation parameter itself has physical meaning, therefore, in addition to the original similarity evaluation loss function, a smooth regularization term loss function is also required to constrain the deformation parameter, and the two are controlled by a hyperparameter λ. Therefore, in some embodiments of this specification, the loss function can be:

[0075]

[0076] where L(f, m, φ) is the loss function, is the similarity evaluation loss function, L smooth (φ) is the smooth regularization term loss function, f represents the standard image, m represents the image to be registered, φ represents the deformation parameter, λ is the hyperparameter, represents registering the image to be registered m using the deformation parameter φ to obtain the registered image.

[0077] In some embodiments of this specification, registering the to-be-registered image by using the deformation parameter to obtain the registered image may include: registering the to-be-registered image by using the deformation parameter through a spatial transformation network to obtain the registered image; wherein, the spatial transformation network includes a grid generator and a resampler; the grid generator is configured to generate a deformation field by using the deformation parameter; and the resampler is configured to apply the deformation field to the to-be-registered image to obtain the registered image.

[0078] The deformation parameter can be applied to the to-be-registered image through a spatial transformation network to obtain the registered image. Specifically, the spatial transformation network responsible for completing the registration work mainly consists of two parts: a Grid generator and a Sample. The Grid generator can use the deformation parameter predicted by the siamese neural network to generate a deformation field, that is, the displacement direction of each imaging point in the to-be-registered image, which can also be understood as the horizontal and vertical correction amounts of the imaging point. The Sample can apply the deformation field generated by the Grid generator to the to-be-registered image to obtain the registered image. Through the spatial transformation network, efficient and accurate calibration of the to-be-registered image can be achieved.

[0079] In some embodiments of this specification, after generating the enhanced common image point gather based on the standard image and the registered image, it may further include: generating a pre-registration horizontal stack profile and / or full stack profile according to the standard image and the to-be-registered image; generating a post-registration horizontal stack profile and / or full stack profile according to the standard image and the registered image; and determining the enhancement effect of the common image point gather based on the pre- and post-registration common image point gathers and the pre- and post-registration horizontal stack profiles and / or full stack profiles.

[0080] After obtaining the enhanced common image point gather, it is necessary to evaluate the enhancement effect. In this embodiment, the enhancement effect of the deformation parameter generated by the deep learning data-driven model on the to-be-registered image can be evaluated by comparing the pre- and post-registration horizontal stack profiles and / or full stack profiles and observing the changes in the pre- and post-registration common image point gathers. Specifically, a pre-registration horizontal stack profile and / or full stack profile can be generated according to the standard image and the to-be-registered image. A post-registration horizontal stack profile and / or full stack profile can also be generated according to the standard image and the registered image. The pre-registration common image point gather is the acquired pre-stack common image point gather, and the post-registration common image point gather is the enhanced common image point gather. Then, the enhancement effect of the common image point gather can be determined based on the pre- and post-registration common image point gathers and the pre- and post-registration horizontal stack profiles and / or full stack profiles.

[0081] In one embodiment, the horizontal stacked profiles before and after registration can be compared. If the result shows that the in-phase axis part of the result after registration has a good horizontal stacking effect and the signal-to-noise ratio also increases significantly in the non-in-phase axis part, it can be shown that the common image point gather enhancement effect is better.

[0082] In one embodiment, the full stacked profiles before and after registration can be compared. If the result shows that the in-phase axis can be depicted more clearly in the result after registration and the resolution and signal-to-noise ratio of the seismic profile both increase to a certain extent, it can be shown that the common image point gather enhancement effect is better.

[0083] In one embodiment, the changes in the common image point gathers before and after registration can be observed. If the result after registration can flatten the in-phase axis of the far offset to a certain extent, the purpose of pre-stack common image point gather enhancement is achieved.

[0084] In some embodiments of this specification, after determining the common image point gather enhancement effect based on the common image point gathers before and after registration and the horizontal stacked profiles and / or full stacked profiles before and after registration, it may further include: determining whether to adjust the hyperparameters in the loss function according to the common image point gather enhancement effect.

[0085] Specifically, the comparison of the full stacked profiles before and after registration and the observation of the changes in the common image point gathers can be used to evaluate whether to adjust the hyperparameters of the network model. In the case of a good enhancement effect, the hyperparameters may not be adjusted. In the case of a poor enhancement effect, the hyperparameters can be adjusted until the enhancement effect is good.

[0086] The following hyperparameters can be adjusted: learning rate, the weight of the similarity evaluation function in the loss function, the weight of the smoothing regularization term, batch size, etc. The evaluation is carried out based on the changes in the full stacked profiles and the common image point gathers before and after registration. For example, observe whether the in-phase axis of the full stacked profile is depicted more clearly and whether the profile resolution increases; whether the far offset part in the common image point gather is flattened.

[0087] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. Specifically, reference can be made to the description of the relevant processing-related embodiments above, and details will not be repeated here.

[0088] The above description is of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The above method will be described below in conjunction with a specific embodiment. However, it should be noted that this specific embodiment is only for better illustrating this specification and does not constitute an improper limitation of this specification.

[0090] In this specific embodiment, a method for enhancing pre-stack common imaging point gathers is provided. In this embodiment, by constructing a deep learning data-driven model, considering the overall geological structure consistency and local waveform similarity, the horizontal error and vertical error of different offset profiles are reduced by using the unsupervised image registration technology of deep learning, effectively enhancing the coherence of the event axes in the pre-stack common imaging point gathers, and thus improving the resolution of the stacked seismic data.

[0091] Please refer to Figure 2 , which shows the flow chart of the method for enhancing pre-stack common imaging point gathers in this specific embodiment. As Figure 2 shown, when performing pre-stack common imaging point gather enhancement, first, high-quality image data is selected and generated. A common offset dataset is extracted from the pre-stack common imaging point gathers, and preprocessing such as data augmentation and normalization is performed on the common offset seismic dataset to obtain the preprocessed common offset seismic dataset, that is, the preprocessed common offset seismic dataset. A standard image and an image to be registered are established. Generally, the superposition of several common offset seismic data within a small offset range (for example, 200 - 500 meters) is used as the standard image, and the remaining offset profiles are used as the images to be registered.

[0092] Secondly, registration relationship learning is performed. By constructing a deep learning data-driven model, features such as the geological structure consistency and local waveform difference between the standard image and the image to be registered are learned, thereby obtaining the deformation field of the image to be registered. The deformation field is applied to the image to be registered through a spatial transformation network, and finally, the quality of the gather enhancement effect is evaluated by observing the stacked section and the common imaging point gathers. Please refer to Figure 3, which shows the structure diagram of unsupervised image registration in this embodiment. It includes two parts: a deep learning network model and a spatial transformation network. The deep learning network model in this embodiment uses a siamese convolutional neural network, which is mainly responsible for extracting features such as geological structure consistency and local waveform similarity between the images to be registered and the standard image to generate a deformation field. The spatial transformation network uses a Grid generator and a Sampler to resample the images to be registered according to the spatial deformation field, thereby obtaining the registered images.

[0093] In this specific embodiment, geological knowledge is introduced into the deep learning unsupervised image registration technology, aiming to automatically estimate the horizontal and vertical correction amounts of imaging points in different offset sections by using the powerful information mining function of the deep neural network for seismic big data. At the same time, the deep neural network can also achieve end-to-end, image-to-image, and pixel-to-pixel feature learning. The deep learning data-driven model used in this patent is a siamese neural network. The siamese neural network architecture mainly consists of an encoder and a decoder. The encoder part selects the convolutional part of the classic convolutional neural network VGG-16, which includes a total of 5 convolutional stages: conv1-conv5. Each convolutional stage consists of a convolutional layer and a pooling layer; the decoder part is completely symmetric to the encoder part and consists of a convolutional layer and an upsampling layer. The encoder function is mainly responsible for extracting seismic features with different levels and different semantics of the input data, and also includes the similarity and difference features between the two input data; the decoder function mainly reconstructs the extracted seismic features to obtain the deformation parameters of the images to be registered. Please refer to Figure 4 , which shows the schematic diagram of the siamese neural network in this embodiment. As Figure 4 shown, this network uses VGG-16 as the backbone network of the siamese neural network. At the same time, the network parameters in the conv1 stage of the network are not shared, so that the two coupled networks can learn their respective features separately; in the conv2 and conv3 stages of the network, sharing the network parameters allows the network to jointly learn common features; then the feature maps obtained by the two coupled networks are concatenated and jointly trained for conv4 and conv5, and finally the upsampling of the network is realized through a completely symmetric decoder, and finally the deformation parameters required for the subsequent network are obtained.

[0094] The spatial transformation network responsible for completing the registration work mainly consists of two parts: Grid generator and Sample. The Grid generator is a grid generator that uses the deformation parameters predicted by the Siamese neural network to generate a deformation field, that is, the displacement direction of each imaging point in the image to be registered, which can also be understood as the horizontal and vertical correction amounts of the imaging points. Sample is a resampler that applies the deformation field generated by the Grid generator to the image to be registered to obtain the registered image. Please refer to Figure 5 , which shows the spatial transformation network in this embodiment. Figure 5 In which, U represents the image to be registered, V represents the registered image, θ represents the deformation field, and T θ (G) represents the grid generator. As Figure 5 shown, this network uses the grid generator and the resampler to calculate the transformation matrix according to the deformation field generated by the previous deep learning network model, and performs spatial transformation on the input image to be registered according to the transformation matrix to obtain the registered image.

[0095] During the iterative training of the model, the image to be registered and the standard image are sent into the deep learning data-driven model to obtain a preliminary registration result. Subsequently, the loss value is calculated through the loss function, and the parameters of the deep learning data-driven model are updated accordingly and enter the next iteration cycle. By iterating the above process multiple times, a fine registration result can be finally obtained. The unsupervised registration method is used in this embodiment, so the loss function used in the entire model training iteration process is as follows:

[0096]

[0097] Among them, L(f, m, φ) represents the total loss value, f represents the standard image, m represents the image to be registered, and φ represents the deformation parameter. represents the similarity evaluation function between the standard image and the predicted image, and L smooth (φ) represents the smoothing regularization term for the deformation parameter, and λ controls the composition ratio between the similarity evaluation function and the smoothing regularization term. The similarity evaluation function selects the MSE loss function, and the MSE loss function is more suitable for the case where the images have a high similarity distribution; the smoothing regularization term is calculated using the gradient of the deformation parameter, which can ensure that the generated deformation field has a certain physical meaning. The specific formulas of the two parts of the loss function are as follows:

[0098]

[0099]

[0100] Among them, p represents the pixel point, and f(p) represents a pixel value in the standard image. Represents a pixel value in the predicted image, and Ω represents the total number of pixels. After the model training is completed, the model is applied to the validation set to obtain the registration result. By comparing the full stack profiles before and after the registration and observing the changes in the common imaging point gathers, it is possible to evaluate whether to adjust the network model hyperparameters.

[0101] In this embodiment, a pre-stack common imaging point gather enhancement method that considers geological structure consistency and local waveform similarity is provided. The feature extraction of two profile images with different offset distances is performed using a deep learning data-driven model to obtain the similarity between them and generate deformation parameters. At the same time, in combination with the unsupervised image registration method, the calibration of the image to be registered can be completed using a spatial transformation network without the need for manual labeling, thereby enhancing the consistency of the in-phase axes in the pre-stack common imaging point gathers and improving the resolution of the stacked profile. It has a strong industrial application value.

[0102] In the process of seismic data processing, although the effects of different offset profiles before stack are different, they correspond to the same seismic profile, but are mapped from different angles. Therefore, overall, different offset profiles should be consistent with the underground geological structure; and locally, the polarity of seismic waveforms at the same position should also have certain similarities. Therefore, under the guidance of the above geological knowledge, combined with the powerful feature extraction ability of deep learning data-driven models, similarity features and difference features are extracted from seismic profiles with different offsets before stack, and deformation fields are obtained; the deformation field is applied to the medium and long offset profiles through a spatial transformation network, which can ultimately realize the enhancement of prestack common imaging point gathers, alleviating the problem that the traditional gather enhancement method only calculates based on low-dimensional local similarity, so that the common imaging point gathers still have residual time difference in the vertical direction. Combining the deep learning data-driven model based on seismic data feature mining of high-dimensional information with geological knowledge, and with the help of image registration technology, a set of high-precision and high-efficiency prestack common imaging point gather enhancement methods that consider geological structure consistency and local waveform similarity are constructed. Guided by geological knowledge such as geological structure consistency and local waveform similarity, this method leverages the powerful information mining and feature extraction capabilities of deep learning data-driven models, and can conduct targeted analysis of the characteristics between seismic profiles at different offsets, laying the foundation for the subsequent generation of deformation fields. At the same time, it leverages the advantage of unsupervised methods that they can efficiently complete target tasks without the use of labels, providing technical support for the realization of refined and efficient seismic data processing.

[0103] The method in the above embodiment is applied to a specific scenario. Figure 6, showing Sigsbee2B seismic data. It is a deep-water model in the Gulf of Mexico released by the two major exploration geophysical societies, SEG in the United States and EAGE in Europe, and is a standard data used for seismic data processing internationally. The offset of this data is 200 - 4000m, with a total of 39 offsets. Figure 6 (a) of Figure 6 is the velocity model of Sigsbee2B, Figure 6 (b) of Figure 6 is the seismic profile at zero offset obtained by migration imaging of the Sigsbee2B velocity model. Figure 6 In the horizontal axis in it is CDP (common depth point), and the vertical axis is Time (time). In the left part, the time is 1000 - 2300ms and the CDP is 0 - 300, mainly including planar reflectors, faults, and diffraction points. The formation is relatively flat, and its imaging result is accurate, which can reflect the real situation of the velocity model. The right part of the model is the salt dome part with complex geometry, the seismic waveform is chaotic, and the diffracted wave at the salt dome boundary does not converge. The bottom of the model is the high-velocity layer, resulting in poor migration imaging effect within the seabed range, weak signal, and chaotic waveform.

[0104] Please refer to Figure 7 , showing the seismic profiles at different offsets. Figure 7 (a)(b)(c)(d) of Figure 7 are the seismic profiles at offsets of 200 - 400m, 1200 - 1400m, 2200 - 2400m, and 3200 - 3400m respectively. It can be observed that within the time range of 1000 - 2300ms and the CDP range of 0 - 50, as the offset increases, the imaging result becomes weaker and weaker, and the waveform almost disappears at 3200 - 3400m; while in the salt dome part within the time range of 600 - 1500ms and the CDP range of 300 - 900, as the offset increases, the imaging of the salt dome edge becomes chaotic due to the stretching distortion caused by NMO, and the whole salt dome gradually moves upward with the increase of the offset, showing dislocation.

[0105] The prestack common image point gather enhancement method in this embodiment can be applied to the above seismic data. Please refer to Figure 8 , showing the comparison diagram of image registration effects. Figure 8 (a)(b)(c)(d) of Figure 8 are the image to be registered, the registered image, the standard image, and the deformation field respectively. From Figure 8 it can be seen that the method in this embodiment can improve the resolution of the seismic profile in-phase axis to a certain extent, making the in-phase axis clearer and more continuous.

[0106] Please refer to Figure 9 , showing the vector visualization schematic diagram of the deformation field. One vector can be displayed every 2 pixel points. As Figure 9As shown, the vector visualization of the deformation field can clearly show the vector displacement direction of the pixel points. At the same time, the displacement directions of the same event axis shown in the vector field are consistent, indicating that the registration effect in this embodiment has physical significance and conforms to certain geological significance.

[0107] Please refer to Figure 10 , which shows the horizontal stacked sections before and after registration. Figure 10 In (a) of Figure 10 , it is the horizontal stack of the standard image and the image to be registered, and in (b) of

[0108] Please refer to Figure 11 , which shows the full stacked sections before and after registration. Figure 12 For Figure 11 is a partial sample display in Figure 11 and Figure 12 . The horizontal axis in Figure 11 and Figure 12 is CDP (Common Depth Point), and the vertical axis is Time. Figure 11 and Figure 12 The left side of Figure 11 and Figure 12 is the stacked result before registration, and the right side of

[0109] Please refer to Figure 13 , which shows the common image point gather of the 20th trace in the test set before (left) and after (right) registration. As can be seen from Figure 13 , the result after registration by the method in this embodiment can flatten the event axis of the far offset to a certain extent, achieving the purpose of pre-stack gather enhancement. Among them, the abscissa offset is the offset.

[0110] Based on the same inventive concept, embodiments of this specification also provide a pre-stack common image point gather enhancement device, as described in the following embodiments. Since the principle of the pre-stack common image point gather enhancement device for solving problems is similar to that of the pre-stack common image point gather enhancement method, the implementation of the pre-stack common image point gather enhancement device can refer to the implementation of the pre-stack common image point gather enhancement method, and the repeated parts will not be elaborated. As used hereinafter, the term "unit" or "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated. Figure 14 is a structural block diagram of the pre-stack common image point gather enhancement device according to an embodiment of this specification, as Figure 14 shown, including: an extraction module 141, a generation module 142, a learning module 143, and a registration module 144. The following describes this structure.

[0111] The extraction module 141 is used to obtain a pre-stack common image point gather and extract a common offset seismic data set from the pre-stack common image point gather.

[0112] The generation module 142 is used to generate a standard image based on the common offset seismic data in the common offset seismic data set within a preset range of offsets; and is also used to generate an image to be registered based on the common offset seismic data in the common offset seismic data set outside the preset range of offsets.

[0113] The learning module 143 is used to learn the global feature relationship between the standard image and the image to be registered by using a deep learning data-driven model, and obtain the deformation parameters of the image to be registered.

[0114] The registration module 144 is used to register the image to be registered by using the deformation parameters to obtain a registered image; and is also used to generate an enhanced common image point gather based on the standard image and the registered image.

[0115] In some embodiments of this specification, the extraction module may specifically be used to:

[0116] Obtain a pre-stack seismic data set; perform migration on the pre-stack seismic data set to obtain a pre-stack common image point gather;

[0117] Extract a common offset seismic data set from the pre-stack common image point gather;

[0118] Perform preprocessing on the common offset seismic data set to obtain a preprocessed common offset seismic data set.

[0119] In some embodiments of this specification, during the iterative training process of the deep learning data-driven model, the standard image and the image to be registered are input into the deep learning data-driven model to obtain a preliminary registration result. The loss value of the preliminary registration result is calculated through a loss function, and it is determined whether the loss value meets a preset condition. If the preset condition is not met, the model parameters of the deep learning data-driven model are updated, and the next iteration cycle is entered until the obtained loss value meets the preset condition.

[0120] In some embodiments of this specification, the deep learning data-driven model includes a siamese neural network; the siamese neural network includes an encoder and a decoder;

[0121] The encoder is used to extract seismic features of the standard image and the image to be registered; the seismic features include similarity features and difference features between the standard image and the image to be registered;

[0122] The decoder is used to reconstruct the data of the extracted seismic features to obtain the deformation parameters of the image to be registered.

[0123] In some embodiments of this specification, the loss function is:

[0124]

[0125] where L(f, m, φ) is the loss function, is the similarity evaluation loss function, L smooth (φ) is the smooth regularization term loss function, f represents the standard image, m represents the image to be registered, φ represents the deformation parameter, and λ is a hyperparameter.

[0126] In some embodiments of this specification, the registration module can specifically be used for:

[0127] Register the image to be registered through a spatial transformation network using the deformation parameter to obtain a registered image; wherein, the spatial transformation network includes a grid generator and a resampler; the grid generator is used to generate a deformation field using the deformation parameter; the resampler is used to apply the deformation field to the image to be registered to obtain a registered image.

[0128] In some embodiments of this specification, the device may further include a determination module, and the determination module can specifically be used for:

[0129] Generate a pre-registration horizontal stack section and / or full stack section according to the standard image and the image to be registered; generate a post-registration horizontal stack section and / or full stack section according to the standard image and the registered image;

[0130] Determine the enhancement effect of the common image point gather based on the common imaging point gathers before and after registration, as well as the horizontal stacked profile and / or full stacked profile before and after registration.

[0131] From the above description, it can be seen that the embodiments of this specification achieve the following technical effects: By using a deep learning data-driven model to learn the global feature relationship between the standard image and the image to be registered, the advantages of seismic big data can be exploited, and geological laws such as overall structural consistency and local waveform similarity can be considered. The powerful feature extraction ability of the deep learning data-driven model for seismic big data features can be fully utilized to achieve the purpose of promoting high-precision and high-efficiency prestack common image point gather enhancement through artificial intelligence. By using the deep learning data-driven model to extract the features of different offset profile images, the similarity between them is obtained and deformation parameters are generated to complete the calibration of the image to be registered, thereby enhancing the coherence of the in-phase axis in the prestack common image point gather and improving the resolution of the stacked profile, which has strong industrial application value. Through the above solution, the problem that the common image point gather still has residual moveout in the vertical direction due to the calculation only based on low-dimensional local similarity in the traditional gather enhancement method can be alleviated. By combining the deep learning data-driven model that mines high-dimensional information based on seismic data features with geological knowledge and using image registration technology, a set of high-precision and high-efficiency prestack common image point gather enhancement methods that consider geological structure consistency and local waveform similarity are constructed. Under the guidance of geological knowledge such as geological structure consistency and local waveform similarity, the powerful information mining ability and feature extraction advantages of the deep learning data-driven model are exerted, and the characteristics between different offset seismic profiles can be analyzed targeted, laying a foundation for the subsequent generation of the deformation field.

[0132] The embodiments of this specification also provide a computer device, which can specifically refer to Figure 15 the schematic structural diagram of the computer device based on the prestack common image point gather enhancement method provided by the embodiments of this specification as shown. The computer device can specifically include an input device 151, a processor 152, and a memory 153. Among them, the memory 153 is used to store instructions executable by the processor. When the processor 152 executes the instructions, the steps of the prestack common image point gather enhancement method described in any of the above embodiments are implemented.

[0133] In this embodiment, the input device may specifically be one of the main devices for information exchange between a user and a computer system. The input device may include a keyboard, a mouse, a camera, a scanner, a light pen, a handwriting input board, a voice input device, etc.; the input device is used to input original data and programs for processing these data into the computer. The input device may also acquire data transmitted from other modules, units, and devices. The processor may be implemented in any suitable manner. For example, the processor may take the form of, for example, a microprocessor or a processor, a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and a form embedded microcontroller, etc. The memory may specifically be a memory device used to store information in modern information technology. The memory may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function without a physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory module, a TF card, etc.

[0134] In this embodiment, the functions and effects specifically implemented by this computer device can be explained by comparison with other embodiments, and will not be elaborated here.

[0135] This specification embodiment also provides a computer storage medium based on a pre-stack common imaging point gather enhancement method. The computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the pre-stack common imaging point gather enhancement method described in any of the above embodiments are implemented.

[0136] In this embodiment, the above storage medium includes, but is not limited to, a Random Access Memory (RAM), a Read-Only Memory (ROM), a Cache, a Hard Disk Drive (HDD), or a Memory Card. The memory may be used to store computer program instructions. The network communication unit may be set according to the standards specified by the communication protocol and is an interface for network connection communication.

[0137] In this embodiment, the functions and effects specifically implemented by the program instructions stored in this computer storage medium can be explained by comparison with other embodiments, and will not be elaborated here.

[0138] Obviously, those skilled in the art should understand that the various modules or steps of the embodiments of the present specification described above can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a sequence different from that here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the embodiments of the present specification are not limited to any specific combination of hardware and software.

[0139] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and many applications other than the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this specification should not be determined with reference to the above description, but should be determined with reference to the full scope of the foregoing claims and the equivalents thereof that these claims possess.

[0140] The above is only the preferred embodiment of this specification and is not used to limit this specification. For those skilled in the art, various changes and modifications can be made to the embodiments of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the protection scope of this specification.

Claims

1. A pre-stack common image point gather enhancement method, characterized in that, Comprising: Obtain a pre-stack common image gather, and extract a common offset seismic data set from the pre-stack common image gather; Generate a standard image based on the common offset seismic data in the common offset seismic data set whose offset is within a preset range; Generate a to-be-registered image based on the common offset seismic data in the common offset seismic data set whose offset is outside the preset range; Use a deep learning data-driven model to learn the global feature relationship between the standard image and the to-be-registered image, and obtain the deformation parameters of the to-be-registered image; Register the to-be-registered image using the deformation parameters to obtain a registered image; Generate an enhanced common image gather based on the standard image and the registered image.

2. The pre-stack common imaging point gather enhancement method according to claim 1, wherein Obtain a pre-stack common image gather, and extract a common offset seismic data set from the pre-stack common image gather, including: Obtain a pre-stack seismic data set; perform migration on the pre-stack seismic data set to obtain a pre-stack common image gather; Extract a common offset seismic data set from the pre-stack common image gather; Perform preprocessing on the common offset seismic data set to obtain a preprocessed common offset seismic data set.

3. The pre-stack common imaging point gather enhancement method according to claim 1, wherein During the iterative training process of the deep learning data-driven model, input the standard image and the to-be-registered image into the deep learning data-driven model to obtain a preliminary registration result, calculate the loss value of the preliminary registration result through a loss function, and determine whether the loss value meets a preset condition. If the preset condition is not met, update the model parameters of the deep learning data-driven model and enter the next iteration cycle until the obtained loss value meets the preset condition.

4. The pre-stack common image point gather enhancement method according to claim 3, wherein The deep learning data-driven model includes a Siamese neural network; the Siamese neural network includes an encoder and a decoder; The encoder is used to extract seismic features of the standard image and the to-be-registered image; the seismic features include similarity features and difference features between the standard image and the to-be-registered image; The decoder is used to perform data reconstruction on the extracted seismic features to obtain the deformation parameters of the to-be-registered image.

5. The pre-stack common image point gather enhancement method according to claim 4, characterized in that, The loss function is: Among them, \(L(f,m,\varphi)\) is the loss function, which is the similarity evaluation loss function, \(L\) smooth (\varphi)\) is the smoothing regularization term loss function, \(f\) represents the standard image, \(m\) represents the image to be registered, \(\varphi\) represents the deformation parameter, and \(\lambda\) is the hyperparameter.

6. The pre-stack common image point gather enhancement method according to claim 4, wherein Register the to-be-registered image using the deformation parameters to obtain a registered image, including: Register the to-be-registered image using the deformation parameters through a spatial transformation network to obtain a registered image; wherein, the spatial transformation network includes a grid generator and a resampler; the grid generator is used to generate a deformation field using the deformation parameters; the resampler is used to apply the deformation field to the to-be-registered image to obtain a registered image.

7. The pre-stack common image point gather enhancement method according to claim 1, characterized in that After generating an enhanced common image gather based on the standard image and the registered image, further including: Generate a pre-registration horizontal stack section and / or full stack section according to the standard image and the to-be-registered image; generate a post-registration horizontal stack section and / or full stack section according to the standard image and the registered image; Determine the enhancement effect of the common image gather according to the pre- and post-registration common image gathers and the pre- and post-registration horizontal stack sections and / or full stack sections.

8. A pre-stack common image point gather enhancement device, characterized in that, Comprising: An extraction module, configured to obtain a pre-stack common image point gather and extract a common offset seismic data set from the pre-stack common image point gather; A generation module, configured to generate a standard image based on the common offset seismic data in the common offset seismic data set whose offset is within a preset range; and further configured to generate a to-be-registered image based on the common offset seismic data in the common offset seismic data set whose offset is outside the preset range; A learning module, configured to learn the global feature relationship between the standard image and the to-be-registered image by using a deep learning data-driven model to obtain the deformation parameters of the to-be-registered image; A registration module, configured to register the to-be-registered image by using the deformation parameters to obtain a registered image; It is further configured to generate an enhanced common image point gather based on the standard image and the registered image.

9. A computer device, characterized in that, It includes a processor and a memory for storing processor-executable instructions, and when the processor executes the instructions, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that, When the instructions are executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images