Seismic data noise suppression method and system based on deep learning

Abstract

The invention relates to the field of seismic data analysis, in particular to a seismic data noise suppression method and system based on deep learning. The method comprises the following steps: acquiring original seafloor seismic data of a target area and marine environment parameters which are in space-time correlation with the original seafloor seismic data; the marine environment parameters are mapped to generate a space-time continuous physical guidance noise field, the generation process of the physical guidance noise field is constrained by physical consistency, and the physical guidance noise field has low-frequency, coherent and smooth physical attributes of flow-induced noise; and taking the physical guidance noise field as physical guidance information, adaptively separating an effective seismic signal from the original seabed seismic data, and reconstructing and outputting a seabed seismic data set after noise suppression. Through combination of physical reasoning and deep learning, the model can distinguish signals like noise and noise of the image signals, and the high fidelity and generalization ability of seismic data noise reduction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of seismic data analysis, and in particular to a seismic data noise suppression method and system based on deep learning. BACKGROUND

[0002] Traditional seismic data analysis techniques are often disturbed by strong coherent noise related to sea water movement when dealing with submarine seismic data, especially data collected by submarine nodes, such as persistent low-frequency noise generated by the interaction of ocean currents and submarine topography. Traditional filtering methods rely on the simple separation assumption of signal and noise in the transform domain, and are difficult to handle complex flow-induced noise that is non-stationary, spatially varying and overlaps with weak geological signal frequency bands, and are extremely vulnerable to effective signals.

[0003] Existing deep learning-based seismic denoising methods mostly use end-to-end supervised learning, whose effectiveness is heavily dependent on the "pure signal" label. Such methods usually treat noise as a statistical phenomenon for "memory" removal, lacking understanding of the physical source of noise. When facing new types of interference generated by specific ocean dynamics processes such as laminar flow under specific temperature-salinity structure that have not appeared in the training data, the generalization ability decreases dramatically. SUMMARY

[0004] The present application provides a deep learning-based seismic data noise suppression method and system to solve the above problems.

[0005] In a first aspect, the present application provides a deep learning-based seismic data noise suppression method, which comprises: obtaining original submarine seismic data of a target area, and marine environmental parameters that are spatio-temporally correlated with the original submarine seismic data; mapping the marine environmental parameters to generate a spatio-temporally continuous physical guided noise field, the generation process of the physical guided noise field being subject to a physical consistency constraint, the physical guided noise field having the physical properties of low frequency, coherence and smoothness of flow-induced noise; and adaptively separating effective seismic signals from the original submarine seismic data using the physical guided noise field as physical guiding information, and reconstructing and outputting a submarine seismic data set after noise suppression.

[0006] By the technical solution, high-fidelity noise suppression is realized through a physical guidance mechanism, which is significantly better than traditional methods and general deep learning models. First, the physically consistent constraint forces the generated physically guided noise field to have the physical properties of flow-induced noise, enabling the signal extraction network to accurately distinguish between signals that look like noise and noise that looks like signals, significantly improving the protection of weak geological features such as faults and pinch-outs while suppressing strong coherent noise. Second, the method has strong generalization and adaptability. Even in ocean current patterns that have not appeared in the training data, the noise field generator can generate reasonable noise estimates based on the input marine environment parameters, enabling physical reasoning rather than pattern memorization, thereby avoiding the problem of a sharp decline in the generalization ability of traditional deep learning models. In addition, the generated physically guided noise field serves as an intermediate product, providing explainability and being useful for noise source analysis. Finally, by using weakly supervised labels, the method eliminates the dependence on absolute pure signals, which are difficult to obtain, greatly enhancing its practicality and improving the efficiency and reliability of data processing.

[0007] Optionally, the mapping of the marine environment parameters to generate a spatiotemporally continuous physically guided noise field comprises: inputting the marine environment parameters into an environment encoder to encode and generate a high-dimensional environment context feature vector representing the physical state of the current acquisition environment; constructing a spatiotemporal coordinate tensor corresponding to the dimension of the original seabed seismic data, each element in the spatiotemporal coordinate tensor being associated with the time and spatial trace index information of the corresponding sampling point in the marine environment parameters; inputting the environment context feature vector and the spatiotemporal coordinate tensor into a noise field generator, which outputs a preliminary noise field that is spatiotemporally continuous through its internal differentiable neural network mapping; during the training process of the noise field generator, the preliminary noise field is constrained by a physical consistency loss function to force its energy distribution and spatiotemporal variation pattern to conform to the low-frequency, coherence, and smoothness physical priors of flow-induced noise, thereby generating the physically guided noise field.

[0008] By the technical solution, the marine environment parameters and spatiotemporal coordinates are used as inputs, and a physical consistency constraint is combined to enhance the explainability and generalization ability of the noise field generator. First, the model no longer relies solely on statistical learning, but is constrained by physical priors to ensure that the generated noise field is physically reasonable, which enables the model to reason based on physical principles when faced with changes in the acquisition environment, thereby having strong generalization ability. Second, the generated noise field is spatiotemporally continuous and can be accurately matched to each sampling point of the original seismic data, providing high-precision noise estimates. Finally, the environment encoder abstracts complex marine environment parameters into high-dimensional feature vectors, effectively incorporating physical information into the deep learning framework and providing accurate physical guidance for subsequent signal separation.

[0009] Optionally, the physical consistency loss function comprises at least one constraint of a low-frequency spectrum constraint, a space-time smoothness constraint, and a range of apparent velocity constraint; the low-frequency spectrum constraint is to punish the energy of a high-frequency part of the preliminary noise field after frequency domain transformation; the space-time smoothness constraint is to calculate the gradient of the preliminary noise field in the time dimension and the space dimension and punish its large amplitude change; and the range of apparent velocity constraint is to transform the preliminary noise field to a tau-p domain and encourage its energy to be concentrated in a low-speed interval corresponding to the ocean current interference.

[0010] By the above technical solution, three physical consistency constraints are introduced, so that the generated physical guide noise field has extremely high physical rationality, thereby improving the de-noising performance of the subsequent signal extraction network. The low-frequency spectrum constraint avoids the effective signal being mistaken for noise. The space-time smoothness constraint ensures the coherence of the noise field, which is beneficial to the model accurately capturing the wave field characteristics of the flow-induced noise. The range of apparent velocity constraint accurately limits the noise field from the propagation characteristics, ensuring its corresponding relationship with the actual ocean current interference. This multi-dimensional physical constraint improves the generalization ability of the model, so that it can still generate accurate noise prior when facing different marine environments.

[0011] Optionally, the effective seismic signal is adaptively separated from the original seabed seismic data based on the physical guide noise field as the physical guide information, and the seabed seismic data set after noise suppression is reconstructed and output, comprising: splicing the physical guide noise field and the original seabed seismic data in the channel dimension to form a combined data; inputting the combined data into a signal extraction network, the signal extraction network explicitly distinguishes and processes information from the noise field channel and the data channel through its initial attention layer, and uses the physical guide noise field as a spatial attention guide condition to make the signal extraction network focus on the data area with a noise field form similarity greater than a preset similarity, and complete the noise prior; the signal extraction network learns to decouple and separate the component related to the physical guide noise field from the original seabed seismic data under the guidance of the noise prior through a hierarchical Transformer block, while retaining and reconstructing the effective signal component related to geological reflection, to obtain a preliminary de-noised seismic signal; and based on the original seabed seismic data, the preliminary de-noised seismic signal is analyzed to construct a seabed seismic data set.

[0012] By the technical solution, the physical guided noise field is explicitly introduced into the signal extraction network as a noise prior, realizing adaptive and high-fidelity separation of the noise. The design of the initial attention layer enables the model to accurately focus on the noise area, avoiding excessive processing of the effective signal area, thereby reducing the risk of damage to the effective signal. The powerful long-distance dependency modeling capability of the hierarchical Transformer block ensures that the continuity and integrity of the geological reflection events are preserved while removing coherent noise. This physical guided attention mechanism improves the denoising performance and generalization ability of the model in complex marine environments.

[0013] Optionally, the signal extraction network is a deep learning model with a visual Transformer structure as the backbone; the signal extraction network processes the input of the combined data through a self-attention mechanism and a cross-attention mechanism, wherein the physical guided noise field is used as the key and value of the cross-attention mechanism to guide the model to focus on the context related to the noise prior in the data.

[0014] By the technical solution, the cross-attention mechanism is used to enhance the noise suppression capability of the signal extraction network. The physical guided noise field is used as the key and value of the cross-attention mechanism, providing an efficient and interpretable way to inject physical prior information into the deep learning model. This enables the model to adaptively decouple features according to the physical form of the noise, improving the accuracy of denoising and the protection capability for weak signals. This structure avoids the confusion between noise and signal features in traditional methods, ensuring high-fidelity reconstruction of effective signals.

[0015] Optionally, based on the original seabed seismic data, the preliminary denoised seismic signal is analyzed to construct a seabed seismic data set, including: calculating the initial residual between the original seabed seismic data and the preliminary denoised seismic signal; inputting the initial residual, the preliminary denoised seismic signal and the physical guided noise field into a residual optimization model; the residual optimization model is a lightweight convolutional neural network, which is used to analyze the residual noise contained in the initial residual, the effective signal components that are excessively removed, and the noise parts that cannot be completely modeled by the physical guided noise field, and output a fine residual correction amount; adding the residual correction amount and the preliminary denoised seismic signal to obtain the seabed seismic data set after noise suppression.

[0016] By the technical solution, the residual optimization model is introduced to realize secondary fine-tuning of the denoising result, and the signal-to-noise ratio and fidelity of the final output are significantly improved. The residual learning mechanism can focus on processing the subtle errors of the signal extraction network, effectively recover the effective signal components that are excessively removed, and further suppress the residual noise, especially the complex noise that cannot be completely modeled by the physically guided noise field. The design of the lightweight CNN ensures the efficiency of the correction process and avoids introducing excessive computational burden.

[0017] Optionally, the training of the residual optimization model is performed jointly with the training of the signal extraction network, and in the total loss function of the joint training, a signal reconstruction loss acting on the final output and a sparsity constraint loss acting on the final residual are included; the signal reconstruction loss is used to measure the difference between the final output and the weakly supervised label; and the sparsity constraint loss is used to encourage the noise to be completely separated and the effective signal to be completely retained by imposing an L1 norm penalty on the final residual between the original seabed seismic data and the final output, so that the final residual does not contain structural information.

[0018] By the technical solution, the joint training mechanism ensures that the optimization objectives of the two stages of signal extraction and residual optimization are consistent, and end-to-end optimization is realized. The introduction of the sparsity constraint loss is the key to guaranteeing high-fidelity denoising. It prevents effective signal components from being incorrectly classified as noise and removed by forcing the final residual to contain only noise. This constraint mechanism improves the model's ability to protect weak geological signals, especially the reconstruction of details such as faults and pinch-outs.

[0019] Optionally, the joint training process of the signal extraction network and the residual optimization model includes: preparing a training data set, the training data set containing multiple groups of samples, each group of samples including: an original seabed seismic data segment, a spatiotemporally correlated marine environment parameter vector, and a low-disturbance seismic signal data as the weakly supervised label; using the training data set to perform end-to-end joint training of the signal extraction network and the residual optimization model with the total loss function as the target, so that the signal extraction network and the residual optimization model learn the difference between the original seismic data and the pure seismic signal data under different marine environment influences, and further enable the signal extraction network and the residual optimization model to accurately separate and reconstruct the effective seismic signal under the guidance of the noise field.

[0020] By the technical solution, the end-to-end joint training strategy ensures that each component of the model can be optimized cooperatively to achieve global optimization. By using the marine environment parameter vector with spatio-temporal correlation, the model learns the mapping relationship between noise and the physical environment, so that it can still generate a reasonable physical guided noise field when facing environmental conditions that do not appear in the training set, and exhibits strong generalization ability. The weakly supervised learning strategy significantly reduces the dependence on expensive and difficult-to-obtain "pure signal" labels, greatly improving the practicality of the method.

[0021] Optionally, the low-disturbance seismic signal data is seismic data obtained by applying far-offset stack processing to original seabed seismic data; and the signal-to-noise ratio of the low-disturbance seismic signal data is higher than that of the original seabed seismic data.

[0022] By the technical solution, the weakly supervised label generated by far-offset stack processing has significant practicality and technical advantages. The conventional process of seismic data processing is used, which is low in cost and easy to implement, and avoids the dependence on expensive "pure" labels. Far-offset stack uses the difference in velocity between the signal and flow-induced noise to ensure that the generated label has high fidelity in terms of effective signal, which is crucial for training the signal extraction network to recover geological reflections. Finally, the quality of such weakly supervised label is sufficient to drive the joint training process, combined with physical guidance and sparsity constraints, to ultimately achieve high-fidelity noise suppression.

[0023] In a second aspect, the present application provides a deep learning-based seismic data noise suppression system, which comprises: a data acquisition module configured to acquire original seabed seismic data of a target area and marine environment parameters that are spatio-temporally correlated with the original seabed seismic data; a data mapping module configured to map the marine environment parameters to generate a spatio-temporally continuous physical guided noise field, wherein the generation of the physical guided noise field is subject to a physical consistency constraint, and the physical guided noise field has the physical properties of low frequency, coherence and smoothness of flow-induced noise; a data reconstruction module configured to use the physical guided noise field as physical guidance information to adaptively separate effective seismic signals from the original seabed seismic data, and reconstruct and output a seabed seismic data set after noise suppression. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0025] Figure 1 An application scenario diagram provided for an embodiment of the present application.

[0026] Figure 2 A flowchart of a seismic data noise suppression method based on deep learning provided for an embodiment of the present application.

[0027] Figure 3 A structural diagram of a seismic data noise suppression system based on deep learning provided for an embodiment of the present application. DETAILED DESCRIPTION

[0028] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following will be combined with the accompanying drawings for the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0029] In addition, the term “and / or” in the present document is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the character “ / ” in the present document generally represents an “or” relationship between the associated objects unless otherwise specified.

[0030] The embodiments of the present application will be further described in detail below in combination with the accompanying drawings of the specification.

[0031] The existing deep learning-based seismic denoising methods mostly use end-to-end supervised learning, and their effectiveness is heavily dependent on the “pure signal” label. Such methods usually consider noise as a statistical phenomenon for “memory” removal, lacking understanding of the physical source of noise. When facing new types of interference generated by specific ocean dynamics processes (such as laminar flow under specific temperature-salinity structure) that have not appeared in the training data, the generalization ability will sharply decrease.

[0032] Based on this, this application provides a deep learning-based method and system for noise suppression of seismic data. Through a physical guidance mechanism, it achieves high-fidelity noise suppression, significantly outperforming traditional methods and general deep learning models. The physically guided noise field, generated by physical consistency constraints, possesses the physical properties of flow-induced noise, enabling the signal extraction network to accurately distinguish between "noise-like signals" and "signal-like noise." This significantly improves the protection of weak geological features such as faults and pinch-outs while suppressing strong coherent noise. Secondly, this method exhibits strong generalization and adaptability. Even under ocean current patterns not present in the training data, the noise field generator can generate reasonable noise estimates based on the input ocean environmental parameters, achieving "physical inference" rather than "pattern memory," thus avoiding the problem of a sharp decline in generalization ability in traditional deep learning models. Furthermore, the generated physically guided noise field, as an intermediate product, provides interpretability and can be used for noise source analysis. Finally, by employing weakly supervised labels, this method eliminates the dependence on elusive "absolutely pure" signals, greatly enhancing its practicality and improving the efficiency and reliability of data processing.

[0033] Figure 1 This is a schematic diagram of an application scenario provided by this application. When analyzing submarine seismic data, the method provided in this application is applied to achieve high-fidelity noise suppression.

[0034] Specifically, the method provided in this application can be applied to any server. The server interacts with sensor arrays and environmental sensors to obtain raw seafloor seismic data provided by the sensor arrays and marine environmental parameters provided by the environmental sensors. Based on real-time signal processing and physics-guided deep learning theory, a noise field generation module with physical constraints is constructed by introducing marine environmental parameters as prior information. This solves the limitations of traditional methods and general deep learning methods in dealing with complex flow-induced noise. Through the combination of "physical reasoning + deep learning", the model can distinguish between "signals that look like noise" and "noise that looks like signals", significantly improving the high fidelity and generalization ability of noise reduction. This method enables monitoring personnel to obtain seafloor seismic datasets with high reference value.

[0035] For specific implementation details, please refer to the following examples.

[0036] Figure 2 This application provides a flowchart of a deep learning-based seismic data noise suppression method according to an embodiment of the present application. The method of this embodiment can be applied to servers in the above scenarios, such as... Figure 2 As shown, the method includes: S201. Acquire the raw submarine seismic data of the target area, as well as marine environmental parameters that are spatially and temporally correlated with the raw submarine seismic data; S202, mapping the marine environment parameters to generate a spatiotemporally continuous physically guided noise field, the generation process of the physically guided noise field being subject to a physical consistency constraint, the physically guided noise field having the low-frequency, coherent and smooth physical properties of flow-induced noise; S203, using the physically guided noise field as physical guidance information, adaptively separating effective seismic signals from the original seabed seismic data, reconstructing and outputting the seabed seismic data set after noise suppression.

[0037] Based on real-time signal processing and physically guided deep learning theory, the present application introduces marine environment parameters as prior information, constructs a noise field generation module with physical constraints, and solves the limitations of traditional methods and general deep learning methods in processing complex flow-induced noise. The core principle of the method is that the physical properties of flow-induced noise such as low frequency, coherence and smoothness are strongly correlated with marine environment parameters such as flow velocity, flow direction and temperature and salinity. The marine environment parameters obtained in step S201 are used to represent the physical origin of noise. In step S202, an environment parameter is mapped to a spatiotemporally continuous physically guided noise field by a noise field generator subject to a physical consistency constraint. The noise field has the physical prior properties of flow-induced noise, and can generate reasonable noise estimates according to physical reasoning even in the case of weak or no labels. In step S203, the signal extraction network uses the physically guided noise field as explicit noise prior information to guide the model to adaptively focus on areas similar to the noise field in the original data, thereby realizing high-fidelity separation and reconstruction of effective signals. The combination of physical reasoning and deep learning enables the model to distinguish between signals like noise and noise like signals, significantly improving the high-fidelity and generalization ability of noise reduction.

[0038] The method includes three main steps S201, S202 and S203, corresponding to Figure 1The overall flowchart is shown. In step S201, the original seabed seismic data of the target area is obtained by the sensor array arranged in the target area, which is usually common receiver gather or common shot gather, and the dimension is time-space trace index. At the same time, the marine environmental parameters closely related to the data acquisition time and space position are obtained by the environmental sensor, such as surface flow velocity, flow direction, water depth, thermocline depth, wave height, etc. These parameters constitute the characteristic vector representing the physical origin of flow noise. In step S202, the marine environmental parameters are input into the deep learning network to generate a physically guided noise field. The core of the generation process is the physical consistency constraint, which ensures that the generated noise field has the low frequency, coherence and smoothness of flow noise. The physically guided noise field as an interpretable intermediate product directly shows the noise form that the model "thinks", and enhances the credibility of the model. In step S203, the signal extraction network receives the original seabed seismic data and the physically guided noise field, uses the latter as powerful noise prior information, and decouples the effective seismic signal from the original data through an adaptive separation mechanism, and reconstructs and outputs the seabed seismic data set after noise suppression. The role of the signal extraction network is to use the physically guided information to perform accurate signal-noise separation. In the above scheme, the original seabed seismic data is the noisy data to be processed, the marine environmental parameters are the physical causes of the flow noise, and the physically guided noise field is the bridge connecting the physical world and the deep learning model. The signal extraction network is responsible for completing the core noise suppression and signal reconstruction task under the guidance of physics.

[0039] In the above embodiments, the selection of marine environmental parameters can be adjusted according to the actual work area conditions. For example, in addition to the surface flow velocity, flow direction, water depth, thermocline depth, seabed topographic slope, seabed sediment type, and tidal period can also be included as parameters to more comprehensively characterize the physical causes of flow-induced noise. These parameters can be further processed by the encoder to generate a higher-dimensional environmental context feature vector. In step S202, the generation model of the physically guided noise field can be replaced by other generation models, such as a structure based on a generative adversarial network (GAN) or a variational autoencoder (VAE), but must be constrained by the physical consistency constraint in its training process. For example, a loss term based on a physical equation (such as a simplified form of the Navier-Stokes equation) can be used to more strictly constrain the spatio-temporal evolution of the noise field. In step S203, the signal extraction network can be replaced by other advanced deep learning structures, such as a U-Net structure based on a convolutional neural network (CNN) or a topological structure based on a graph neural network (GNN), but the core requirement is that the network must be able to explicitly receive and utilize the physically guided noise field as prior information to guide the signal separation process. For example, the noise prior can be introduced by introducing an attention mechanism in the feature fusion layer of the CNN to weight and fuse the feature maps of the noise field with the feature maps of the original data.

[0040] In some embodiments, marine environmental parameters are input into an environmental encoder, which encodes to generate a high-dimensional environmental context feature vector representing the physical state of the current acquisition environment; a spatio-temporal coordinate tensor corresponding to the dimension of the original seabed seismic data is constructed, each element in the spatio-temporal coordinate tensor is associated with the time and spatial channel index information of the corresponding sampling point in the marine environmental parameters; the environmental context feature vector and the spatio-temporal coordinate tensor are jointly input into a noise field generator, which outputs a preliminary noise field through its internal differentiable neural network mapping; during the training process of the noise field generator, the preliminary noise field is constrained by a physical consistency loss function, forcing its energy distribution and spatio-temporal variation pattern to conform to the low-frequency, coherence, and smoothness physical priors of flow-induced noise, thereby generating a physically guided noise field.

[0041] The present embodiment focuses on the generation process of physically guided noise field, involving environment encoder, spatio-temporal coordinate tensor, noise field generator and physical consistency constraint. Among them, the environment encoder: adopts a small feedforward neural network (such as three-layer MLP), receives the input marine environment parameters such as flow velocity, flow direction, water depth, thermocline depth, and encodes them into an environmental context feature vector with dimension D (for example, D = 64 or ). The feature vector carries the physical cause information of flow noise. Spatio-temporal coordinate tensor: it is a tensor with the same time and spatial channel dimensions as the original ocean bottom seismic data. Each element in the tensor is a multi-dimensional vector containing the time index, spatial channel index such as receiver position coordinates and acquisition timestamp information associated with environmental parameters of the corresponding sampling point. Noise field generator: adopts a deep learning model based on MLP or Siren. It receives the spliced environmental context feature vector and spatio-temporal coordinate tensor, performs differentiable mapping to output the preliminary noise field consistent with the original data dimension. Physical consistency constraint: when training the noise field generator, the preliminary noise field is constrained by adding a physical consistency loss term such as spectrum loss, gradient loss, etc. in the total loss function. For example, the low-frequency spectrum constraint punishes high-frequency energy, forcing the noise field to maintain low-frequency properties; the spatio-temporal smoothness constraint punishes sharp spatio-temporal changes, forcing the noise field to maintain coherence and smoothness. These constraints ensure that the finally generated physically guided noise field conforms to the physical prior of flow noise in both time and frequency domains.

[0042] In the above embodiments, the environmental encoder can employ a Transformer-based structure to better capture the non-linear interactions between different ocean environmental parameters such as current velocity, thermocline depth. For example, a self-attention mechanism can be employed to weight the importance of different environmental parameters, thereby generating a more representative environmental context feature vector. The noise field generator can be replaced by an Implicit Neural Representation (INR)-based model, such as a Siren network using a periodic activation function such as a sine function, which helps to generate a spatio-temporally continuous noise field with higher resolution and stronger smoothness, especially suitable for describing the coherence of flow-induced noise. The encoding method of the spatio-temporal coordinate tensor can be optimized. In addition to directly using the time index and spatial channel index, the absolute position coordinates such as latitude, longitude, water depth and relative time such as relative to the tidal period can be Fourier feature encoded (Positional Encoding) to improve the modeling ability of the noise field generator for high-frequency details while maintaining the capture of low-frequency coherence. The implementation of the physical consistency constraint can be diversified. In addition to the loss function imposing constraints, physical priors can be used as regularization terms, or physical model layers can be embedded in the architecture of the noise field generator, for example, by a special convolutional layer to simulate the diffusion or advection process of the flow field in time and space, to more closely integrate physical laws.

[0043] In some embodiments, the physical consistency loss function comprises at least one constraint of low-frequency spectrum constraint, spatio-temporal smoothness constraint, and apparent velocity range constraint; the low-frequency spectrum constraint is to penalize the energy of the high-frequency part of the preliminary noise field after frequency domain transformation; the spatio-temporal smoothness constraint is to calculate the gradient of the preliminary noise field in the time dimension and the spatial dimension and penalize its large amplitude change; the apparent velocity range constraint is to transform the preliminary noise field to the τ-p domain and encourage its energy to concentrate in the low-speed interval corresponding to the ocean current disturbance.

[0044] The present embodiment enforces the preliminary noise field generated to comply with the physical priors of flow-induced noise from three dimensions of frequency domain, spatio-temporal domain, and apparent velocity domain by introducing three physical consistency constraints. The low-frequency spectrum constraint ensures that the noise field has low-frequency properties by penalizing high-frequency energy, avoiding the model misjudging high-frequency geological signals as noise. The spatio-temporal smoothness constraint ensures that the noise field is continuous and coherent in time and space, in line with the characteristics of fluid disturbance waves, preventing the generator from outputting high-frequency, irregular noise. The apparent velocity range constraint is unique to flow-induced noise, which has extremely low apparent velocity. By transforming the data to the τ-p domain, the noise field generator is encouraged to concentrate its energy in the low-speed interval corresponding to the ocean current disturbance, thereby generating a noise field that is more consistent with the characteristics of flow-induced noise. The domain (slowness-time domain) can clearly separate energy of different speeds. The constraint encourages the energy of the noise field to concentrate in the area with large slowness, i.e. the low-speed interval, thereby accurately simulating the characteristics of the ocean current disturbance. The three constraints together constitute a powerful physical prior, significantly improving the physical reasonableness of the noise field and the guidance accuracy of the signal extraction network.

[0045] In the above implementation process, the low-frequency spectrum constraint can use a wavelet transform-based constraint, which penalizes the coefficients in the wavelet coefficients corresponding to high-frequency components to achieve more fine-grained frequency domain control. The spatio-temporal smoothness constraint can be replaced by a total variation (TV) based constraint, i.e. penalizing the L1 norm of the gradient, which is beneficial to preserving the sharp boundaries that may exist in the noise field (e.g. the fluid disturbance boundary caused by the sudden change of the seabed topography), while maintaining the smoothness of most areas. The range of apparent velocity constraint can be enhanced by introducing a sparse representation-based constraint. For example, in the domain, in addition to penalizing high-speed energy, the sparsity of low-speed energy can also be encouraged, i.e. imposing an L1 norm penalty in the low-speed interval, so that the flow-induced noise appears as a few concentrated energy groups in the domain. In addition, more complex wave field decomposition methods (such as high-order Radon transform) can be used to replace the standard transform to improve the capture accuracy of non-linear coherent noise.

[0046] In some embodiments, the physical guided noise field is spliced with the original seabed seismic data in the channel dimension to form combined data; the combined data is input into the signal extraction network, which explicitly distinguishes and processes information from the noise field channel and the data channel through its initial attention layer, and uses the physical guided noise field as a spatial attention guidance condition to make the signal extraction network focus on the data region with a similarity greater than a preset similarity to the noise field form, and complete the noise prior; the signal extraction network learns to decouple and separate the component related to the physical guided noise field from the original seabed seismic data through the hierarchical Transformer block under the guidance of the noise prior, while preserving and reconstructing the effective signal component related to geological reflection, to obtain a preliminary denoised seismic signal; based on the original seabed seismic data, the preliminary denoised seismic signal is analyzed to construct a seabed seismic data set.

[0047] The working principle of the signal extraction network is as follows: first, by splicing the original data and the physical guided noise field in the channel dimension into combined data, the model obtains the physical prior information of the noise in the input stage. Second, the initial attention layer explicitly distinguishes the information of the two channels, and uses the noise field as a spatial attention guide condition. This means that the model will calculate the similarity between each spatio-temporal position in the original data and the noise field in form, and concentrate attention resources on the data area with high similarity (i.e. high noise possibility). This completes the introduction of the noise prior, enabling the model to accurately know "where the noise is". Subsequently, the hierarchical Transformer block uses its powerful long-distance dependence modeling capability to learn the decoupling function of the signal and the noise under the guidance of the noise prior. The Transformer structure can capture the coherence of the seismic data in time and space, thereby accurately preserving and reconstructing the effective signal components such as geological reflections while separating the noise components, and finally outputting the preliminary denoised seismic signal. Finally, through residual analysis, the preliminary denoised signal is optimized to construct the final seabed seismic data set.

[0048] The core of the present embodiment is the structure and function of the signal extraction network, which adopts a Transformer-based architecture to achieve accurate guidance of the noise prior. Combined data: formed by splicing the original seabed seismic data ( ) and the physical guided noise field (dimension ) in the channel dimension, forming an input tensor with a dimension of . Initial attention layer: this is a customized attention module whose function is to explicitly introduce the noise prior. This layer first maps the combined data to a high-dimensional feature space. Then, it uses the feature map of the noise field as "query" or "key / value" to calculate the similarity matrix between the original data channel feature map and the noise field channel feature map. Through this similarity matrix, a spatial attention weight map is generated, which indicates which regions in the original data are highly similar in form to the physical guided noise field. Noise prior: i.e. the spatial attention weight map generated by the initial attention layer. This weight map assigns high weights to regions with high noise possibility, thereby guiding the subsequent hierarchical Transformer block to focus on processing these regions. Hierarchical Transformer block: composed of multiple Transformer encoders, adopting a hierarchical design (such as a pyramid structure) to capture spatio-temporal features at different scales. Within each Transformer block, the self-attention mechanism learns how to decompose the feature vectors in the original data into signal components and noise components under the guidance of the noise prior, achieving decoupling and separation. Preliminary denoised seismic signal: the direct output of the signal extraction network, is the seismic data after preliminary noise suppression.

[0049] In the above embodiments, the concatenation manner of combining data can be replaced by feature fusion manner. For example, the original data and the physical guided noise field can be respectively extracted through independent convolution layers, and then weighted summation or element-level multiplication is performed in the feature space as the input of the signal extraction network. The initial attention layer can adopt a more complex cross-attention mechanism. For example, a double-flow network can be designed, one of which processes the original data and the other of which processes the noise field, and then through the cross-attention mechanism, the features of the noise field are taken as the key and value, and the features of the original data are taken as the query, to realize more fine noise guidance. The hierarchical Transformer block can be replaced by other types of Transformer structure, such as SwinTransformer or ConvNeXt, to improve the calculation efficiency and feature extraction ability. Inside the Transformer block, a gating mechanism can be introduced to dynamically control the influence degree of the noise prior on the feature decoupling process, and ensure that the influence of the noise prior is minimized in the signal region. The implementation manner of the noise prior can adopt a probability density function. That is, the output of the initial attention layer is not a hard attention weight map, but a probability distribution map representing the probability of each space-time point containing noise, and the subsequent Transformer block performs feature weighting based on this probability distribution map.

[0050] In some embodiments, the signal extraction network is a deep learning model with a visual Transformer structure as the backbone; the signal extraction network processes the input of the combined data through a self-attention mechanism and a cross-attention mechanism, wherein the physical guided noise field is used as the key and value of the cross-attention mechanism to guide the model to pay attention to the context related to the noise prior in the data.

[0051] In seismic data processing, coherent noise such as flow noise has long-range spatio-temporal correlation across the entire gather, which is difficult for traditional CNN to capture effectively. The embodiment adopts ViT as the backbone of the signal extraction network to fully utilize its global modeling capability. The working principle of the network is as follows: first, the combined data (containing original data and noise field) is divided into Token sequence and position encoding. Second, through the self-attention mechanism (Self-Attention), the model captures the long-range spatio-temporal correlation of signals and noise in the entire data set. The core innovation lies in the introduction of cross-attention mechanism (Cross-Attention). In the cross-attention mechanism, the features from the original data (as the query Q) interact with the features from the physically guided noise field (as the key K and the value V). Since the noise field has been ensured to be physically reasonable through physical constraints, using it as K and V can enable the original data Q to query the context information similar to the physical noise pattern. This mechanism explicitly injects noise prior information into the feature extraction process, guiding the model to accurately decouple the noise component in the feature space, thereby realizing high-precision signal separation.

[0052] Signal extraction network: the backbone adopts ViT structure, for example, hierarchical ViT (such as SwinTransformer) can be used to balance local feature extraction and global context modeling. Self-attention mechanism: applied to the inside of the combined data, used to capture the mutual relationship and long-range dependence of seismic data and noise field in space and time. This mechanism ensures that the model can understand the coherence of flow noise across the entire gather. Cross-attention mechanism: is the core component to realize physical guidance. In the cross-attention layer: query (Q): feature representation from the original seabed seismic data. Key (K) and value (V): feature representation from the physically guided noise field. Cross-attention calculation , by calculating the similarity between the original data features Q and the noise field features K, an attention weight is generated. This weight indicates which features in the original data should be "affected" or "guided" by the features V of the noise field. Since K and V come from the noise field with physical constraints, this guidance forces the model to focus on the context information related to the physical noise, thereby achieving accurate import of noise prior. The output of the network is the feature representation of the preliminary denoised seismic signal, which is then reconstructed into a time-domain signal through a decoder (such as MLP or deconvolution layer).

[0053] In the above implementation process, the backbone of the signal extraction network can be replaced by a structure based on the hybrid of convolutional neural network and Transformer (such as ConvNeXt or CoaT) to improve the efficiency of local feature extraction while maintaining the ability of global context modeling. The implementation of cross-attention mechanism can be adjusted. For example, multi-head cross-attention can be used, in which different heads focus on different physical properties of the noise field (such as low-frequency components, coherent paths, etc.), thereby achieving more fine-grained guidance. The role of the physically guided noise field in the cross-attention mechanism can be extended. In addition to being the key K and value V, it can also control its influence on the query Q through a gating mechanism. For example, by a learnable gating parameter , dynamically adjust the guidance strength under different signal-to-noise ratio conditions. In the ViT structure, the Tokenization process can use Overlapping Patch Embedding to preserve the local continuity of seismic data in space, which is crucial for processing the wave field characteristics of coherent noise.

[0054] In some embodiments, an initial residual error between the original seafloor seismic data and the preliminary denoised seismic signal is calculated; the initial residual error, the preliminary denoised seismic signal, and the physically guided noise field are jointly input into a residual optimization model; the residual optimization model is a lightweight convolutional neural network for analyzing the residual noise contained in the initial residual error, the effective signal components that are over-removed, and the noise parts that the physically guided noise field fails to completely model, and outputting a fine residual correction; the residual correction is added to the preliminary denoised seismic signal to obtain the seafloor seismic data set after noise suppression.

[0055] Although signal extraction networks utilize physical guidance, two types of errors may still exist in complex real-world data: residual noise (noise that has not been completely removed) and excessive removal of the effective signal (signal impairment). This implementation introduces a residual optimization model to refine the initial denoising results. Its working principle is based on the idea of ​​residual learning: the difference between the initial denoising result and the original data, i.e., the initial residual, is used as the correction target. The residual optimization model receives three inputs: the initial residual containing error information, the initial denoised signal (providing signal context), and the physically guided noise field (providing noise prior). The role of the residual optimization model (lightweight CNN) is to analyze the composition of the initial residual. For example, if there are high-frequency, incoherent components in the residual, it may correspond to residual random noise; if there are structures in the residual similar to the in-phase axis of the effective signal, it may correspond to excessively removed effective signal. By combining the contextual information of the initial denoised signal and the noise field, the model can distinguish these components and output a refined residual correction. This correction includes the recovery of the effective signal and further removal of residual noise. Finally, the correction is added back to the initial denoised signal to obtain the final high-fidelity submarine earthquake dataset.

[0056] Initial residual: The calculation formula is as follows ,in It is raw undersea earthquake data. This is the seismic signal after preliminary denoising. The residual represents the sum of noise and error perceived by the signal extraction network. The residual optimization model employs a lightweight convolutional neural network (such as a ResNet block or a simplified version of U-Net), designed to efficiently process the input tensor and output a correction. The model receives a multi-channel input tensor, with channels including... , and (Physically guided noise field). Output of the correction: Output of the residual optimization model. This is the precise residual correction amount. If If it is positive, it means that it needs to be obtained from Subtract residual noise from the middle; if A negative value indicates that the effective signal from the excessive removal needs to be added back. In the middle. Final output: Noise-suppressed submarine seismic dataset. The calculation formula is .

[0057] In the above implementation process, the residual optimization model can be replaced with a sequence model based on Gated Recurrent Units (GRUs) or Long Short-Term Memory (LSTM) networks, especially when the residual noise exhibits time-series correlation, as such models can better capture its dynamic characteristics. The input to the residual optimization model can be optimized. In addition to the initial residual, the initial denoised signal, and the noise field, an environmental context feature vector (from the environmental encoder) can be used as a conditional input to help the model better understand the correlation between residual noise and environmental changes. The method of applying corrections can be replaced with a weighted average. For example, the final output... It can be ,in It is a weighted graph output by the model, dynamically determining the balance between the initial denoising result and the result of subtracting the noise field from the original data. The initial residual can be calculated using frequency domain residuals. That is, calculated in the frequency domain. This is then fed into the residual optimization model to focus more on correcting frequency domain errors, such as high-frequency random noise or low-frequency residual coherent noise.

[0058] In some embodiments, the training of the residual optimization model is performed jointly with the training of the signal extraction network, and the total loss function of the joint training includes a signal reconstruction loss applied to the final output and a sparsity constraint loss applied to the final residual; the signal reconstruction loss is used to measure the difference between the final output and the weakly supervised label; the sparsity constraint loss is used to encourage complete noise separation and complete preservation of effective signals by applying an L1 norm penalty to the final residual between the original seafloor seismic data and the final output, so that the final residual does not contain structural information.

[0059] To ensure the signal extraction network and residual optimization model work together effectively and achieve high-fidelity denoising, this implementation employs an end-to-end joint training mechanism. This joint training allows both modules to jointly optimize their parameters under the guidance of a global loss function, achieving optimal signal separation and reconstruction results. Total Loss Function It comprises two key components: signal reconstruction loss and sparsity constraint loss Signal reconstruction loss Its purpose is to ensure that the final output is as close as possible to the weakly supervised label (such as low-perturbation seismic signal data). This provides the main supervisory signal, guiding the model to learn how to recover the effective signal. (Sparseness constraint loss) This is one of the key innovations of this invention. Final residual This represents all the noise components as perceived by the model. Its physical meaning is that if the noise is completely separated and the effective signal is fully preserved, then... It should contain only noise and not any structural information (such as phase axes, reflecting interfaces, etc.). Through the analysis of... Applying an L1 norm penalty (sparseness constraint) forces the model to retain all structural information in the final output, while concentrating unstructured noise components. In the middle, the L1 norm (compared to the L2 norm) naturally encourages sparsity, which helps to completely separate noise.

[0060] Joint training: All trainable parameters of the signal extraction network and the residual optimization model are updated in the same iteration, ensuring collaborative optimization of the two modules. Total loss function. Defined as ,in and These are weighting coefficients. This is a loss of physical consistency. Signal reconstruction loss. The L1 or L2 norm is typically used to measure the final output. With weak supervision label The differences between them.

[0061] Sparse constraint loss : Acts on the final residual By imposing L1 norm penalties, we encourage... The sparsity, i.e. The energy in the signal should be concentrated as much as possible in the noise component and should not contain structural signal information.

[0062] Through joint training and sparsity constraints, the model is forced to learn a decomposition function that decomposes the original data into... (High-fidelity signal) and (Sparse noise), thus avoiding excessive damage to the effective signal.

[0063] During the above implementation process, signal reconstruction loss Perceptual loss or structural similarity loss (SSIMLoss) can be used to better measure the visual and structural similarity of the final output to the weakly supervised label, especially in maintaining the continuity of the phase axis. Sparse constraint loss. Non-L1 norm sparsity measures, such as entropy-based constraints, can be used to encourage sparsity in the final residual. The probability distribution of the signal has high entropy (i.e., strong randomness and lack of structural information). During joint training, a multi-scale loss mechanism can be employed. That is, the signal reconstruction loss is calculated separately at different levels of the network. The loss function is then weighted and summed to ensure that the model can accurately separate signal and noise at different scales. To enhance the robustness of the model, an adversarial loss can be added to the total loss function, in which a discriminator is used to distinguish the final output from the weakly supervised labels, thereby further improving the realism and signal-to-noise ratio of the final output.

[0064] In some embodiments, a training dataset is prepared, which contains multiple sets of samples. Each set of samples includes: raw submarine seismic data segments, spatiotemporally correlated marine environmental parameter vectors, and low-disturbance seismic signal data as the weak supervision label. Using the total loss function as the objective, the signal extraction network and the residual optimization model are jointly trained end-to-end using the training dataset. This enables the signal extraction network and the residual optimization model to learn the differences between raw seismic data and clean seismic signal data under different marine environmental influences, thereby enabling the signal extraction network and the residual optimization model to accurately separate and reconstruct effective seismic signals under the guidance of a noise field.

[0065] The performance of deep learning models heavily depends on the quality of training data and the training strategy. A major challenge in processing submarine seismic data is the difficulty in obtaining "absolutely pure" signal labels. This implementation employs a weakly supervised learning strategy, utilizing "low-disturbance seismic signal data" as weakly supervised labels, and combines end-to-end joint training to maximize the model's practicality and generalization ability. The joint training process works as follows: each set of samples in the training dataset contains raw data (noisy), marine environmental parameters (the physical causes of the noise), and low-disturbance signal data (weakly supervised labels). By inputting the environmental parameters into a noise field generator, a physically guided noise field is generated. Then, a noise field-guided signal extraction network and a residual optimization model process the raw data. The total loss function is used... With this goal in mind, the model learns during training how to generate reasonable noise estimates based on the marine environmental parameter E. and utilize Accurately extract effective signals from raw data , making The training strategy aims to obtain labels that are as close as possible to those with weak supervision. This end-to-end training approach enables the model to explicitly learn and adapt to the effects of noise on different marine environments, thus giving it strong generalization capabilities in practical applications.

[0066] Training dataset: contains Groups of samples, each group of samples By triplet Composition, among which, This is a raw submarine seismic data segment. It is a vector of marine environmental parameters with spatiotemporal correlation. This corresponds to low-disturbance seismic signal data (weakly supervised label). The size of the data segment is typically a common-detector point gather or a subset thereof. Marine environmental parameter vector. It must include key parameters such as flow velocity, flow direction, water depth, and thermocline depth to drive the noise field generator. (Weak supervision tag) The signal-to-noise ratio (SNR) of the obtained signal data is higher than that of the original data after applying relatively reliable denoising processes such as long-offset superposition and high-precision FK filtering. However, it may still contain a small amount of residual noise. End-to-end joint training: Minimize the total loss function using optimizers such as Adam or SGD. In each training iteration, the data and The inputs to the entire network architecture include an environmental encoder, a noise field generator, a signal extraction network, and a residual optimization model. Loss function Based on the final output and The differences and sparsity of the final residuals are calculated, and all network parameters are updated by backpropagation.

[0067] In summary, the preparation of the training dataset can employ data augmentation techniques, such as random time-shifting, amplitude scaling, or adding random background noise to the original data, to improve the model's robustness. The joint training process can utilize a staged training strategy. For example, the noise field generator can be trained independently first (using physical consistency loss and a small amount of supervision information), then frozen, and finally trained end-to-end for the signal extraction network and the residual optimization model. The optimizer can be an adaptive learning rate optimizer (such as AdamW) combined with a learning rate scheduling strategy (such as Cosine Annealing) to ensure the stability and convergence speed of the training process. During training, model uncertainty estimation can be introduced. For example, Monte Carlo Dropout or Bayesian neural networks can be used to output an uncertainty map along with the final result to evaluate the reliability of the denoising results.

[0068] In some embodiments, the low-disturbance seismic signal data is seismic data obtained by applying long-offset stacking processing to the original seafloor seismic data; the signal-to-noise ratio of the low-disturbance seismic signal data is higher than that of the original seafloor seismic data.

[0069] Weakly supervised learning requires labeled data It possesses a relatively high signal-to-noise ratio (SNR), but does not need to be "absolutely pure." Flow-induced noise typically exists in a low-velocity, coherent form in near-offset data. However, in far-offset stacking, effective signals such as formation reflections have high velocities, and their in-phase axes are effectively enhanced during stacking. Meanwhile, low-velocity flow-induced noise is effectively suppressed due to its low-velocity characteristics and spatial coherence. Therefore, seismic data obtained by applying far-offset stacking to raw submarine seismic data has a significantly higher SNR than the original data, making it an economical and practical weakly supervised label. Although such labels may still contain a small amount of residual noise, their effective signal components are greatly preserved and enhanced, thus meeting the requirements for signal reconstruction loss. Requirements for label quality.

[0070] Low-disturbance seismic signal data Generation Process: Data Screening: First, the raw seafloor seismic data is screened by offset, selecting only seismic traces with offsets greater than a preset threshold (e.g., greater than 1000 meters). This is because flow-induced noise is usually strongest at near offsets. Velocity Analysis: A detailed velocity analysis is performed on the screened far-offset data to determine the superposition velocity of effective signals such as first-order reflected waves. Superposition Processing: Conventional dynamic correction and superposition processing (such as CMP superposition or common-detector superposition) are applied. Due to the extremely low apparent velocity of flow-induced noise, its energy is effectively suppressed during superposition due to insufficient coherence. Output: The superimposed data is the low-disturbance seismic signal data. Signal-to-noise ratio improvement: Through long-offset superposition processing, the amplitude of the effective signal is enhanced, while the amplitude of flow-induced noise is relatively weakened, thereby improving the signal-to-noise ratio. The signal-to-noise ratio is significantly higher than that of the original submarine seismic data. For example, the signal-to-noise ratio can be improved by 5dB to 10dB.

[0071] Low-disturbance seismic signal data Other advanced denoising methods can be used to generate weakly supervised labels. For example, methods based on High-Order Singular Value Decomposition (HOSVD) or sparse representation can be used to preprocess the original data to generate higher-quality weakly supervised labels. The offset threshold can be dynamically adjusted according to the water depth and ocean current intensity of the actual work area. In deep-water work areas, the influence of flow-induced noise is wider, and the offset threshold may need to be set higher. The overlay processing can be replaced by high-precision denoising methods, such as methods based on 3D block matching and 3D filtering (BM3D), to further suppress random noise in the far-offset data to improve the purity of the weakly supervised labels. To improve the reliability of the labels, an ensemble strategy of multiple weakly supervised labels can be adopted. For example, far-offset overlay labels and HOSVD-based denoised labels can be generated simultaneously, and the signal reconstruction loss can be minimized. The training employs a weighted average method for supervision to improve its robustness.

[0072] Figure 3 A schematic diagram of a deep learning-based seismic data noise suppression system provided in one embodiment of this application is shown below. Figure 3 As shown, a seismic data noise suppression system 300 based on deep learning in this embodiment includes: a data acquisition module 301, a data mapping module 302, and a data reconstruction module 303.

[0073] The data acquisition module 301 is used to acquire the original submarine seismic data of the target area, as well as marine environmental parameters that are spatiotemporally related to the original submarine seismic data. The data mapping module 302 is used to map the marine environmental parameters to generate a spatiotemporally continuous physical guidance noise field. The generation process of the physical guidance noise field is subject to physical consistency constraints, and the physical guidance noise field has the physical properties of low frequency, coherence and smoothness of flow-induced noise. The data reconstruction module 303 is used to adaptively separate effective seismic signals from the original submarine seismic data using the physical guidance noise field as physical guidance information, and to reconstruct and output the noise-suppressed submarine seismic dataset.

[0074] Optionally, the data mapping module 302 is specifically used for: inputting the marine environmental parameters into an environmental encoder to encode and generate a high-dimensional environmental context feature vector characterizing the physical state of the current acquisition environment; constructing a spatiotemporal coordinate tensor corresponding to the dimension of the original seafloor seismic data, wherein each element in the spatiotemporal coordinate tensor is associated with the temporal and spatial trace index information of the corresponding sampling point in the marine environmental parameters; inputting the environmental context feature vector and the spatiotemporal coordinate tensor together into a noise field generator, wherein the noise field generator outputs a spatiotemporally continuous preliminary noise field through its internal differentiable neural network mapping; during the training process of the noise field generator, constraining the preliminary noise field through a physical consistency loss function, forcing its energy distribution and spatiotemporal variation pattern to conform to the low-frequency, coherent, and smooth physical priors of flow-induced noise, thereby generating the physical guidance noise field.

[0075] Optionally, in the data mapping module 302, the physical consistency loss function includes at least one of the following constraints: low-spectrum constraint, spatiotemporal smoothness constraint, and apparent velocity range constraint; the low-spectrum constraint is to penalize the high-frequency energy of the initial noise field after frequency domain transformation; the spatiotemporal smoothness constraint is to calculate the gradient of the initial noise field in the time and space dimensions and penalize its large changes; the apparent velocity range constraint is to transform the initial noise field to the τ-p domain and encourage its energy to concentrate in the low-speed range corresponding to ocean current interference.

[0076] Optionally, the data reconstruction module 303 is specifically used for: splicing the physical guidance noise field with the original submarine seismic data in the channel dimension to form combined data; inputting the combined data into a signal extraction network, which explicitly distinguishes and processes information from the noise field channel and the data channel through its initial attention layer, and uses the physical guidance noise field as a spatial attention guidance condition to make the signal extraction network focus on the data region with a morphological similarity to the noise field greater than a preset similarity, thus completing the noise prior; the signal extraction network, guided by the noise prior, learns to decouple and separate the components related to the physical guidance noise field from the original submarine seismic data through a hierarchical Transformer block, while retaining and reconstructing effective signal components related to geological reflection, thus obtaining a preliminary denoised seismic signal; and analyzing the preliminary denoised seismic signal based on the original submarine seismic data to construct the submarine seismic dataset.

[0077] Optionally, in the data reconstruction module 303, the signal extraction network is a deep learning model with a visual Transformer structure as its backbone; the signal extraction network processes the input of the combined data through a self-attention mechanism and a cross-attention mechanism, wherein the physical guidance noise field is used as the key and value of the cross-attention mechanism to guide the model to focus on the context related to the noise prior in the data.

[0078] Optionally, when constructing the submarine earthquake dataset based on the original submarine earthquake data and analyzing the pre-denoised seismic signal, the data reconstruction module 303 is specifically used to: calculate the initial residual between the original submarine earthquake data and the pre-denoised seismic signal; input the initial residual, the pre-denoised seismic signal, and the physical guidance noise field into the residual optimization model; the residual optimization model is a lightweight convolutional neural network used to analyze the residual noise, over-removed effective signal components, and noise components that the physical guidance noise field fails to fully model in the initial residual, and output a fine residual correction amount; add the residual correction amount to the pre-denoised seismic signal to obtain the noise-suppressed submarine earthquake dataset.

[0079] Optionally, in the data reconstruction module 303, the training of the residual optimization model and the training of the signal extraction network are performed jointly, and the total loss function of the joint training includes a signal reconstruction loss applied to the final output and a sparsity constraint loss applied to the final residual; the signal reconstruction loss is used to measure the difference between the final output and the weakly supervised label; the sparsity constraint loss is used to encourage complete separation of noise and complete preservation of effective signal by applying L1 norm penalty to the final residual between the original seafloor seismic data and the final output, so that the final residual does not contain structural information.

[0080] Optionally, in the data reconstruction module 303, the joint training process of the signal extraction network and the residual optimization model is specifically used for: preparing a training dataset, the training dataset containing multiple sets of samples, each set of samples including: original seafloor seismic data segments, spatiotemporally correlated marine environmental parameter vectors, and low-disturbance seismic signal data as the weak supervision label; using the training dataset as the objective of the total loss function, performing end-to-end joint training of the signal extraction network and the residual optimization model, so that the signal extraction network and the residual optimization model learn the differences between the original seismic data and the clean seismic signal data under different marine environmental influences, thereby enabling the signal extraction network and the residual optimization model to accurately separate and reconstruct effective seismic signals under the guidance of a noise field.

[0081] Optionally, in the data reconstruction module 303, the low-disturbance seismic signal data is seismic data obtained by applying long-offset stacking processing to the original seafloor seismic data; the signal-to-noise ratio of the low-disturbance seismic signal data is higher than that of the original seafloor seismic data.

[0082] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A method for suppressing seismic data noise based on deep learning, characterized in that, include: Acquire raw submarine seismic data for the target area, as well as marine environmental parameters that are spatiotemporally correlated with the raw submarine seismic data; The marine environmental parameters are mapped to generate a spatiotemporally continuous physical guidance noise field. The generation process of the physical guidance noise field is constrained by physical consistency. The physical guidance noise field has the physical properties of low frequency, coherence and smoothness of flow-induced noise. Using the physical guidance noise field as physical guidance information, effective seismic signals are adaptively separated from the original submarine seismic data, and the submarine seismic dataset after noise suppression is reconstructed and output.

2. The method according to claim 1, characterized in that, The process of mapping the marine environmental parameters to generate a spatiotemporally continuous physical guidance noise field includes: The marine environmental parameters are input into the environmental encoder to generate a high-dimensional environmental context feature vector that characterizes the physical state of the currently collected environment. Construct a spatiotemporal coordinate tensor corresponding to the dimensions of the original seafloor seismic data, wherein each element in the spatiotemporal coordinate tensor is associated with the temporal and spatial trace index information of the corresponding sampling point in the marine environmental parameters; The environmental context feature vector and the spatiotemporal coordinate tensor are input into the noise field generator, which outputs a spatiotemporally continuous preliminary noise field through its internal differentiable neural network mapping. During the training process of the noise field generator, the initial noise field is constrained by the physical consistency loss function, which forces its energy distribution and spatiotemporal variation pattern to conform to the low-frequency, coherent and smooth physical priors of flow-induced noise, thereby generating the physical guided noise field.

3. The method according to claim 2, characterized in that, The physical consistency loss function includes at least one of the following constraints: low-spectrum constraint, spatiotemporal smoothness constraint, and apparent velocity range constraint; The low-frequency constraint is to penalize the high-frequency portion of the energy after the initial noise field has undergone frequency domain transformation. The spatiotemporal smoothness constraint is: to calculate the gradient of the initial noise field in the time and space dimensions and penalize its large changes; The apparent velocity range constraint is to transform the initial noise field to the τ-p domain and encourage its energy to concentrate in the low-speed range corresponding to ocean current interference.

4. The method according to claim 2, characterized in that, The process of adaptively separating effective seismic signals from the original submarine seismic data using the physical guidance noise field as physical guidance information, and reconstructing and outputting a noise-suppressed submarine seismic dataset includes: The physical guidance noise field is spliced ​​with the original submarine seismic data in the channel dimension to form combined data; The combined data is input into the signal extraction network. The signal extraction network explicitly distinguishes and processes information from the noise field channel and the data channel through its initial attention layer. It uses the physical guidance noise field as a spatial attention guidance condition to make the signal extraction network focus on the data region with a morphological similarity to the noise field greater than a preset similarity, thus completing the noise prior. The signal extraction network, guided by the noise prior, learns to decouple and separate components related to the physical guidance noise field from the original submarine seismic data through hierarchical Transformer blocks, while retaining and reconstructing effective signal components related to geological reflections, thus obtaining a preliminarily denoised seismic signal. Based on the raw submarine earthquake data, the earthquake signals after preliminary denoising are analyzed to construct the submarine earthquake dataset.

5. The method according to claim 4, characterized in that, The signal extraction network is a deep learning model with a visual Transformer structure as its backbone. The signal extraction network processes the input of the combined data through a self-attention mechanism and a cross-attention mechanism, wherein the physical guidance noise field is used as the key and value of the cross-attention mechanism to guide the model to focus on the context in the data that is related to the noise prior.

6. The method according to claim 4, characterized in that, The process of constructing the submarine earthquake dataset based on raw submarine earthquake data, analyzing the preliminarily denoised earthquake signals, includes: Calculate the initial residual between the original submarine seismic data and the pre-denoised seismic signal; The initial residual, the pre-denoised seismic signal, and the physical guidance noise field are all input into the residual optimization model; The residual optimization model is a lightweight convolutional neural network used to analyze the residual noise, over-removed effective signal components, and noise components that the physical guidance noise field failed to fully model in the initial residual, and output a fine residual correction amount. The residual correction amount is added to the pre-denoised seismic signal to obtain the noise-suppressed submarine seismic dataset.

7. The method according to claim 6, characterized in that, The training of the residual optimization model is carried out in conjunction with the training of the signal extraction network, and the total loss function of the joint training includes a signal reconstruction loss that acts on the final output and a sparsity constraint loss that acts on the final residual. The signal reconstruction loss is used to measure the difference between the final output and the weakly supervised label; The sparsity constraint loss is used to encourage complete noise separation and full preservation of effective signals by applying an L1 norm penalty to the final residual between the original seafloor seismic data and the final output, so that the final residual does not contain structural information.

8. The method according to claim 7, characterized in that, The joint training process of the signal extraction network and the residual optimization model includes: Prepare a training dataset, which contains multiple sets of samples. Each set of samples includes: raw submarine seismic data segments, spatiotemporally correlated marine environmental parameter vectors, and low-disturbance seismic signal data as the weak supervision label. Using the total loss function as the objective, the signal extraction network and the residual optimization model are jointly trained end-to-end using the training dataset. This enables the signal extraction network and the residual optimization model to learn the differences between raw seismic data and clean seismic signal data under different marine environmental influences, thereby allowing the signal extraction network and the residual optimization model to accurately separate and reconstruct effective seismic signals under the guidance of a noise field.

9. The method according to claim 8, characterized in that, The low-disturbance seismic signal data is seismic data obtained by applying long-offset stacking processing to the original seafloor seismic data. The signal-to-noise ratio of the low-disturbance seismic signal data is higher than that of the original submarine seismic data.

10. A seismic data noise suppression system based on deep learning, characterized in that, The method applied to any one of claims 1-9 includes: The data acquisition module is used to acquire raw submarine seismic data of the target area, as well as marine environmental parameters that are spatiotemporally correlated with the raw submarine seismic data; The data mapping module is used to map the marine environmental parameters to generate a spatiotemporally continuous physical guidance noise field. The generation process of the physical guidance noise field is subject to physical consistency constraints, and the physical guidance noise field has the physical properties of low frequency, coherence, and smoothness of flow-induced noise. The data reconstruction module is used to adaptively separate effective seismic signals from the original submarine seismic data using the physical guidance noise field as physical guidance information, and to reconstruct and output the noise-suppressed submarine seismic dataset.

Images