A method and system for suppressing seismic signal noise based on self-supervised learning

Abstract

The application discloses a seismic signal noise suppression method and system based on self-supervised learning, which processes a single noisy seismic signal; after reading the original seismic signal, data normalization to the [0, 1] interval is first performed; then, multiple Bernoulli sampling experiments are performed on the normalized seismic signal to construct data pairs of auxiliary tasks; the overall structure of a denoising network is constructed, and the optimization target of the network is determined; the constructed Bernoulli sampling data pairs are trained by using the denoising network to learn the mapping relationship between the noisy seismic signal and the clean seismic signal; after the network iteration converges, the parameters of the network model are saved, and the input single noisy seismic signal is recovered. The application effectively solves the problem that the clean seismic signal label is difficult to obtain, and on the basis of effectively suppressing noise, the damage to the useful signal is reduced as much as possible, and the application has good fidelity and practicability.

Description

Technical Field

[0001] This invention belongs to the field of seismic signal processing technology, specifically relating to a method and system for suppressing seismic signal noise based on self-supervised learning. Background Technology

[0002] In field seismic exploration, due to the influence of exploration techniques, equipment, and environmental factors, the acquired seismic signals inevitably contain a large amount of random noise. Noise can severely affect the resolution of seismic signals, the accuracy of reservoir prediction, and subsequent interpretation of seismic data. Therefore, research on seismic signal noise suppression methods has always been a hot topic in the field of seismic signal processing.

[0003] Traditional seismic signal noise suppression methods have achieved very good results. These methods often require explicit physical information when establishing mathematical models, and then separate the useful signal and random noise through mathematical optimization. Traditional seismic signal noise suppression methods can be broadly classified into three categories: The first category is filtering-based methods. These methods use two-dimensional Fourier transforms to transform the useful signal and noise into the FK domain, and construct suitable pass / stop regions in the transform domain to separate the two signals. The second category is matrix factorization-based denoising methods. Based on the physical characteristics of the useful signal and noise, local SVD is used to decompose the noisy signal into different components, and the effective components are selected to reconstruct the useful signal. The third category is based on sparse signal representation. These methods typically search for a suitable dictionary in the transform domain to sparsely represent the useful signal, but cannot sparsely represent random noise.

[0004] In recent years, with the successful application of deep learning technologies, represented by convolutional neural networks, in image processing and computer vision, deep learning has also provided new inspiration for noise suppression in seismic signals. Supervised deep learning methods, such as DnCNN and 3D-DnCNN, construct sample-label datasets and perform data augmentation, then train end-to-end convolutional neural networks to enable them to separate noise, successfully recovering clean and useful signals.

[0005] Supervised denoising networks' ability to separate random noise is highly dependent on the training dataset. However, in actual seismic exploration, clean seismic signal labels are often difficult to obtain, requiring significant human, material, and financial resources. Typically, traditional denoising methods are used to process noisy seismic signals to construct the dataset, thus limiting the performance of the denoising network to the preprocessing methods. Preprocessing massive amounts of seismic data also results in very slow processing speeds. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for suppressing seismic signal noise based on self-supervised learning, which addresses the shortcomings of the prior art and solves the technical problem of difficulty in obtaining clean seismic signals in actual seismic exploration.

[0007] The present invention adopts the following technical solution:

[0008] A seismic signal noise suppression method based on self-supervised learning includes the following steps:

[0009] The individual seismic signals were normalized, and Bernoulli sampling was used to process the normalized seismic signals. Bernoulli sampling data pairs for the auxiliary task were constructed through repeated experiments.

[0010] Construct a denoising network based on an encoding / decoding structure;

[0011] Determine the optimization objective function of the denoising network;

[0012] The denoising network is trained until convergence based on the optimization objective function;

[0013] The constructed Bernoulli sampling data is used to learn the input denoising network to recover individual noisy seismic signals and reconstruct clean, useful signals.

[0014] Specifically, the data pairs for auxiliary tasks as follows:

[0015]

[0016]

[0017] Where, m i Let y be the mask used for the i-th Bernoulli sampling, ⊙ be the Hadamard product, N be the number of Bernoulli sampling data pairs, and y be the mask used for the i-th Bernoulli sampling. norm This is the normalized, noisy seismic signal.

[0018] Furthermore, Bernoulli's experiment was repeated 100 times.

[0019] Specifically, the denoising network based on the encoder-decoder structure includes a data processing module, an encoder, a decoder, and a residual noise separation module; the data processing module includes normalization operations and Bernoulli sampling; the two-dimensional seismic signal encoded by the encoder is transformed into a value equal to the size of the actual input seismic signal. The two-dimensional seismic signal is restored to its original size after being decoded; the residual noise separation module is used to calculate the difference between the input noisy seismic signal and the clean useful signal predicted by the network, to obtain the noise separated by the network, and the prior noise with a mean of 0 is used as a regularization constraint.

[0020] Furthermore, the encoder has 5 encoding modules, each including partial convolution, dilated convolution, residual learning units, and max pooling; the decoder has 5 decoding modules, each including upsampling with a scaling factor of 2, skip connections, and standard convolutional layers with Dropout.

[0021] Specifically, the optimization objective function L of the denoising network total Specifically:

[0022] L total =L target +αL zm +βL tv

[0023] Among them, L target Let L be the target loss function, α be the weighting coefficient of the noise zero-mean loss function, and L be the weighting coefficient of the noise zero-mean loss function. zm Let L be the zero-mean loss function for noise, β be the weighting coefficient of the total variation loss function, and L be the weighting coefficient of the total variation loss function. tv This is the total variation loss function.

[0024] Specifically, training the denoising network involves the following steps:

[0025] The Adam gradient descent algorithm is used to optimize the entire objective function L of the denoising network. total To optimize, the initial learning rate was set to 0.0001, the Epoch was set to 15000, Dropout was enabled during training, and the parameters of the denoising network were saved after the objective function converged.

[0026] Specifically, a prediction is made for a single noisy seismic signal. Dropout is enabled during the prediction process, and the experiment is repeated N times. The average value of each experiment is then selected as the final result.

[0027] Furthermore, the average value x′ of each experimental result is calculated as follows:

[0028]

[0029] in, This represents the result of the denoising network recovering the noisy seismic signal in the i-th Bernoulli experiment.

[0030] Secondly, embodiments of the present invention provide a seismic signal noise suppression system based on self-supervised learning, comprising:

[0031] The data module normalizes individual seismic signals, processes the normalized seismic signals using Bernoulli sampling, and constructs Bernoulli sampling data pairs for auxiliary tasks through repeated experiments.

[0032] Build modules to construct denoising networks based on encoder-decoder structures;

[0033] The function module determines the optimization objective function of the denoising network obtained from the construction module;

[0034] The training module trains the denoising network obtained from the construction module until convergence, based on the optimized objective function obtained from the function module.

[0035] The suppression module uses the Bernou sampling data obtained from the data module to learn the denoising network obtained from the training module, and then recovers the single noisy seismic signal to reconstruct a clean and useful signal.

[0036] Compared with the prior art, the present invention has at least the following beneficial effects:

[0037] A self-supervised learning-based method for suppressing seismic signal noise is proposed. Due to the large data range of actual seismic signals, to maintain data dimension consistency, normalization is performed to map all acquired seismic signals to the [0, 1] interval. Simultaneously, to address the difficulty in obtaining clean seismic signals, Bernoulli sampling is used to construct auxiliary task data pairs, transforming the original problem, which cannot be directly solved, into a dual problem. Since the original and dual problems have the same solution, an optimization objective function can be constructed and a denoising network trained even when the clean label is unknown, achieving blind denoising of individual seismic signals. In constructing the denoising network structure, the encoder-decoder network can extract high-level semantic features from the data through the encoder. The decoder uses upsampling and skip connections to convert the extracted features into target task data, and a denoising network based on the encoder-decoder structure is constructed to optimize the objective function. After a certain number of iterations, the denoising network finally separates the useful signal and random noise at the output.

[0038] Furthermore, constructing Bernoulli sampling data pairs With Bernoulli sampling data pairs, the primal problem can be transformed into a dual problem, yielding the optimization objective of the self-supervised network. Furthermore, Bernoulli sampling addresses the issue of insufficient training samples from a single seismic dataset.

[0039] Furthermore, N=100 Bernoulli samplings are repeated to construct data pairs. On the one hand, it solves the problem of insufficient training samples for a single earthquake data set; on the other hand, by averaging the processed results of each Bernoulli sample data set by the denoising network, the accuracy of the network prediction results can be improved.

[0040] Furthermore, the entire denoising network adopts an encoding-decoding-based structure, including a data processing module, an encoder, a decoder, and a residual noise separation module. The data processing module constructs Bernoulli sampled data pairs through normalization and Bernoulli sampling operations. The encoder effectively extracts high-level semantic features from the input seismic data through step-by-step convolution and pooling operations. The decoder, through concatenation and upsampling operations, fully integrates the features extracted by deep and shallow networks, transforming high-level semantic features into data suitable for the target task. The residual noise separation module calculates the residual between the actual input seismic signal and the network's predicted useful signal to obtain separated noise, and uses prior information with independent noise and zero mean as constraints. The introduction of the residual noise module also helps to avoid network overfitting to some extent.

[0041] Furthermore, the encoder aims to extract high-level semantic features from actual seismic data. Each encoding module includes partial convolution, dilated convolution, residual learning units, and max pooling operations. Partial convolution selectively acquires contextual information through a mask, enabling the restoration of imaging results. Dilated convolution introduces holes into ordinary convolution, increasing the network's receptive field. Residual learning units are introduced to prevent network degradation. Pooling operations compress the image and extract semantic features. The decoder includes upsampling with a scaling factor of 2, and skip connections fuse information from shallow and deep networks, progressively restoring the details and size of the target seismic image.

[0042] Furthermore, in the network optimization objective L target Introducing a regularization term can prevent overfitting and improve the fidelity of the recovered clean seismic signal, minimizing L. target On the one hand, it makes the predicted value of the denoising network Get as close as possible to the true value x i On the other hand, it minimizes the noise n i Energy. L zm This ensures that the noise separated by the denoising network has a mean of zero, preventing useful signals from being mixed into the noise separated by the network, and improving the fidelity of recovering the useful signal. tv By constraining gradient changes in the horizontal and vertical directions, noise can be suppressed to some extent. These three loss functions together form the optimization objective of the entire network, improving its overall denoising performance.

[0043] Furthermore, the Adam gradient descent method is used to optimize the loss function of the entire network, with an initial learning rate of 0.0001. The purpose of gradient descent is to find the minimum value of the network's loss function and determine the optimal parameters for the entire denoising network. In addition, Adam can automatically adjust the learning rate for each input variable of the loss function and update the variables by using a moving average that exponentially decreases the gradient.

[0044] Furthermore, the prediction of a single noisy seismic signal is performed by repeating Bernoulli sampling N times, constructing data pairs and feeding them into a denoising network for learning. The average value of the prediction results is taken as the final result. Multiple experiments not only expand the data but also improve the accuracy of the network.

[0045] Furthermore, multiple Bernoulli sampling experiments are equivalent to training multiple denoising networks, and the average of the prediction results of each denoising network for a single noisy seismic signal is used as the final result. This can improve the accuracy of the network in predicting clean and useful signals.

[0046] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0047] In summary, this invention transforms a problem that cannot be directly solved by utilizing the mathematical characteristics of noise and Bernoulli sampling. It then obtains an approximate optimal solution to the original problem by solving the dual problem. This invention saves a significant amount of time required to construct artificially labeled clean seismic signals and also solves the problem of obtaining clean seismic signals in actual seismic exploration.

[0048] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0049] Figure 1 Flowchart for invention;

[0050] Figure 2 This is a diagram of the entire denoising neural network structure of the present invention;

[0051] Figure 3 The diagram shows the processing results of actual earthquake data, where (a) is the actual acquired earthquake signal, (b) is the useful earthquake signal recovered by the network of the present invention, and (c) is the noise separated by the network of the present invention.

[0052] Figure 4 for Figure 3 The FK spectrum diagrams are shown, where (a) is the FK spectrum corresponding to the actual seismic signal, (b) is the FK spectrum of the useful seismic signal, and (c) is the spectrum of noise.

[0053] Figure 5 The diagram shows the processing results of actual earthquake data, where (a) is the actual acquired earthquake signal, (b) is the useful earthquake signal recovered by the network of the present invention, and (c) is the noise separated by the network of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0056] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0057] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" relationship.

[0058] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0059] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0060] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0061] This invention provides a method for suppressing seismic signal noise based on self-supervised learning. Before processing with an end-to-end convolutional neural network, it eliminates the need to construct clean labels. Instead, it uses an auxiliary task to extract mathematical features from the data itself as supervisory information (not the original task labels, but constructed auxiliary task labels), thereby enabling the convolutional neural network to separate useful signals from noise. Due to the uneven distribution of seismic signal data, a normalization operation is first used to normalize the input seismic signal to the [0, 1] interval. Then, Bernoulli sampling is repeated multiple times on the normalized seismic signal to construct labels for the auxiliary task. An end-to-end deep learning network based on an encoding / decoding structure is built and trained on the data pairs constructed using Bernoulli sampling. After the network converges after a certain number of iterations, the network model parameters are saved, and a mapping from a single noisy seismic signal to a clean, useful signal is achieved.

[0062] Please see Figure 1 This invention discloses a seismic signal noise suppression method based on self-supervised learning, comprising the following steps:

[0063] S1. Normalize the input single seismic signal to the [0, 1] interval;

[0064] First, the actual earthquake data y is read and normalized as shown in equation (1):

[0065]

[0066] Among them, y min and y max These represent the minimum and maximum values ​​of the two-dimensional seismic signal, respectively. The normalization operation transforms the input noisy seismic signal to the [0, 1] interval.

[0067] S2. Use Bernoulli sampling to process seismic signals and repeatedly conduct experiments to construct data pairs for auxiliary tasks;

[0068] Multiple Bernoulli samplings were performed on the normalized seismic signal to construct data pairs. As shown below:

[0069]

[0070]

[0071] Where, m i Let y be the mask for the i-th Bernoulli sampling, ⊙ be the Hadamard product, and y be the mask for the i-th Bernoulli sampling. norm This is the normalized, noisy seismic signal.

[0072] Bernoulli sampling is equivalent to applying a binary mask consisting only of 0s and 1s to a two-dimensional seismic signal. Each position on the mask has a probability of being preserved of 0.7. Bernoulli sampling data pairs are constructed through this alternating pass-through and block-through method. Furthermore, to address the insufficient data volume of a single noisy seismic signal, 100 Bernoulli experiments are repeated.

[0073] S3. Construct a denoising network based on an encoding / decoding structure;

[0074] Please see Figure 2 The denoising network based on the encoder-decoder structure is similar to the traditional encoder-decoder network UNet, mainly consisting of a data processing module, an encoder, a decoder, and a residual noise separation module. The data processing module includes normalization operations and Bernoulli sampling. The encoder has five encoding modules, each including partial convolution, dilated convolution, residual learning units, and max pooling. After passing through the encoder, the encoded 2D seismic signal is finally transformed into a signal of the input size... The decoder also has five decoding modules, each including upsampling with a scaling factor of 2, skip connections, and standard convolutional layers with Dropout. After passing through the decoder, the 2D seismic signal is finally restored to its original size. Finally, there is a residual noise separation module, which calculates the difference between the input noisy seismic signal and the clean signal predicted by the network to obtain the noise separated by the network. The noise prior with a mean of 0 is used as a regularization constraint to prevent the network from overfitting. In this way, the denoising network can be used to optimize the solution of the objective function.

[0075] S4. Determine the optimization objective function of the denoising network obtained in step S3;

[0076] Considering clean seismic signals x i Since it is unknowable, it is impossible to directly solve equation (4) using network optimization:

[0077]

[0078] Considering the data pairs constructed by Bernoulli sampling in step 02 and the characteristics of noise independence and zero mean, equation (4) is transformed into equations (5a) and (5b):

[0079]

[0080]

[0081] To avoid any useful signal being mixed into the separated noise, the prior information, which is independent of the noise and has a mean of 0, is incorporated as part of the optimization objective in the form of a regularization penalty. Simultaneously, the total variation loss function is introduced as a regularization term to further suppress the noise by constraining gradient changes in the horizontal and vertical directions. The final optimization objective function of the entire denoising network is determined as shown in equations (6a)-(6d):

[0082] L total =L target +aL zm +βL tv (6a)

[0083]

[0084]

[0085]

[0086] Among them, L target Let L be the target loss function. zm Let L be the zero-mean loss function for noise, and α be L. zm The weighting coefficient, L tv Let β be the total variation loss function, and L be the total variation loss function. tv The weighting coefficients are: H is the height of the 2D seismic data (number of sampling points), W is the width of the 2D seismic data (number of traces), and F is the weighting coefficient. θ (·) represents the noise reduction network. and, Data pairs constructed for Bernoulli sampling, m i Let n′ be the mask used for the i-th Bernoulli sampling, θ be the parameters of the denoising network, and n′ be the denoising network. h，w x′ is the value of the noise n′ separated by the denoising network at the position (h, w). h，w The value of the clean seismic signal x′ recovered by the denoising network at the location (h, w).

[0087] S5. Train the denoising network obtained in step S3 based on the optimized objective function obtained in step S4 until convergence, and save the parameters of the denoising network.

[0088] The objective function (6a) was optimized using the Adam gradient descent algorithm. The initial learning rate was set to 0.0001 and the epoch size was 15000. Dropout was enabled during training, and the parameters of the network model were saved after the objective function converged.

[0089] S6. Input the noisy seismic signal into the denoising network obtained in step S5 for learning, recover the individual noisy seismic signal, and reconstruct a clean and useful signal.

[0090] For a single noisy input seismic signal, prediction is performed. Dropout is enabled during the prediction process, and the experiment is repeated N times. The average value of each experiment result is then selected as the final result, as shown below:

[0091]

[0092] in, This represents the result of the denoising network recovering the noisy seismic signal in the i-th Bernoulli experiment.

[0093] The denoising network of this invention can perform end-to-end processing on input noisy seismic signals, mine the characteristics of random noise, and ultimately map the noisy seismic signals into clean, useful signals.

[0094] In another embodiment of the present invention, a seismic signal noise suppression system based on self-supervised learning is provided. This system can be used to implement the above-mentioned seismic signal noise suppression method based on self-supervised learning. Specifically, the seismic signal noise suppression system based on self-supervised learning includes a data module, a construction module, a function module, a training module, and a suppression module.

[0095] The data module normalizes individual seismic signals, processes the normalized seismic signals using Bernoulli sampling, and constructs Bernoulli sampling data pairs for auxiliary tasks through repeated experiments.

[0096] Build modules to construct denoising networks based on encoder-decoder structures;

[0097] The function module determines the optimization objective function of the denoising network obtained from the construction module;

[0098] The training module trains the denoising network obtained from the construction module until convergence, based on the optimized objective function obtained from the function module.

[0099] The suppression module uses the Bernou sampling data obtained from the data module to learn the denoising network obtained from the training module, and then recovers the single noisy seismic signal to reconstruct a clean and useful signal.

[0100] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a seismic signal noise suppression method based on self-supervised learning, including:

[0101] A single seismic signal is normalized, and Bernoulli sampling is used to process the normalized seismic signal. Bernoulli sampling data pairs for auxiliary tasks are constructed through repeated experiments. A denoising network based on an encoding / decoding structure is constructed. The optimization objective function of the denoising network is determined. The denoising network is trained until convergence based on the optimization objective function. The constructed Bernoulli sampling data pairs are input into the denoising network for learning, and a single noisy seismic signal is recovered to reconstruct a clean and useful signal.

[0102] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0103] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the seismic signal noise suppression method based on self-supervised learning in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0104] A single seismic signal is normalized, and Bernoulli sampling is used to process the normalized seismic signal. Bernoulli sampling data pairs for auxiliary tasks are constructed through repeated experiments. A denoising network based on an encoding / decoding structure is constructed. The optimization objective function of the denoising network is determined. The denoising network is trained until convergence based on the optimization objective function. The constructed Bernoulli sampling data pairs are input into the denoising network for learning, and a single noisy seismic signal is recovered to reconstruct a clean and useful signal.

[0105] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0106] Please see Figure 3 , Figure 3 (a) shows the actual acquired two-dimensional seismic signal, totaling 500 channels, each with 800 sampling points, and a sampling time interval of 2ms. It can be seen that the horizontal phase axis is strongly interfered with by random noise, severely affecting the continuity and smoothness of the useful signal phase axis, and also reducing the signal-to-noise ratio. The input signal is processed using a seismic signal noise suppression method based on self-supervised learning according to this invention.

[0107] Figure 3 (b) and Figure 3 (c) shows the useful signal reconstructed by the denoising network of this invention and the separated random noise, respectively. It can be seen that the denoising network of this invention significantly suppresses random noise, restoring the continuity and smoothness of the phase axis of the seismic signal. Furthermore, by observing the noise separated by the network, we cannot see any structure of the phase axis. That is, our network has good fidelity and basically does not damage the useful signal. In addition, observation... Figure 3(c) The present invention also suppresses coherent noise in the tilt direction to a certain extent. To further analyze the effectiveness of the denoising network of the present invention, Figure 3 (a) Figure 3 (b) and Figure 3 (c) Transform to the FK domain respectively, such as Figure 4 (a) Figure 4 (b) and Figure 4 As shown in (c), it can be seen that the energy of the useful signal is mainly concentrated in the middle part, while the random noise is mainly distributed around it. The useful signal and random noise separated by the denoising network are also well separated in the transform domain, which indirectly demonstrates the effectiveness of the network of this invention.

[0108] Figure 5 (a) Figure 5 (b) and Figure 5 (c) represents another set of actual seismic data, consisting of noisy seismic signals, the useful signals recovered by this invention, and separated noise. The actual seismic signals comprise a total of 1000 channels, each with 3001 sampling points, and a sampling time interval of 2 ms. Figure 5 In (b), it can be seen that the denoising network of this invention can effectively suppress noise in both weakly noisy and strongly noisy data regions, ensuring the continuity of the seismic signal phase axis. Furthermore, observation... Figure 5 (c) No horizontal phase axis structure can be seen, indicating that the recovered useful signal has almost no energy residue and has good fidelity.

[0109] In summary, this invention presents a seismic signal noise suppression method and system based on self-supervised learning. By utilizing the mathematical characteristics of noise and Bernoulli sampling, the problem, which was originally unsolvable, is transformed into a solution by solving the dual problem, thus obtaining an approximate optimal solution to the original problem. This addresses the difficulty of obtaining clean seismic signals in practical seismic exploration. Furthermore, to avoid overfitting of individual input signals during training, residual learning units are introduced when constructing the denoising network, improving network performance. Dilated convolutions are used to increase the receptive field of the entire network and better extract features from seismic data. Finally, a residual noise separation module is added. By calculating the difference between the noisy seismic signal and the clean, useful signal predicted by the network, random noise is separated. The independent and zero-mean characteristics of the noise are used as a regularization constraint to prevent the separated noise from being mixed with relevant information of the useful signal, ensuring the fidelity of the entire network. Experimental results show that this invention not only effectively removes random noise but also has minimal damage to the recovered useful signal, exhibiting good fidelity and practicality.

[0110] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0111] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0112] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0113] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0117] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A seismic signal noise suppression method based on self-supervised learning, characterized in that, The method comprises the following steps: The single seismic signal is normalized, and the normalized seismic signal is processed by using Bernoulli sampling, and a Bernoulli sampling data pair for an auxiliary task is constructed through repeated experiments; The denoising network based on the coding and decoding structure comprises a data processing module, an encoder, a decoder and a residual noise separation module. The data processing module comprises a normalization operation and Bernoulli sampling. The two-dimensional seismic signal coded by the encoder becomes an input actual seismic signal size ; the two-dimensional seismic signal processed by the decoder is restored to the original size; The residual noise separation module is used to calculate the difference between the input noisy seismic signal and the clean useful signal predicted by the network, obtain the noise separated by the network, and take the noise prior and mean value of 0 as a regularization constraint; Determining an optimization objective function for a denoising network , in particular: wherein, is a target loss function, is a weight coefficient of the noise zero-mean loss function, is a noise zero-mean loss function, is a weight coefficient of the total variation loss function, is a total variation loss function; The denoising network is trained based on the optimization objective function until convergence; The constructed Bernoulli sampling data pair is input into the denoising network for learning, and a single noisy seismic signal is recovered to reconstruct a clean useful signal.

2. The self-supervised learning based seismic signal noise suppression method of claim 1, wherein, Data pairs for secondary tasks As follows: wherein, is the number of the first Bernoulli sampling, is the mask used for the first Bernoulli sampling, is the Hadamard product, is the number of Bernoulli sampling data pairs, is the normalized noisy seismic signal.

3. The self-supervised learning based seismic signal noise suppression method of claim 2, wherein, The Bernoulli experiment is repeated 100 times.

4. The self-supervised learning based seismic signal noise suppression method of claim 1, wherein, The encoder has 5 encoding modules, each of which includes partial convolution, hollow convolution, residual learning unit and maximum pooling; the decoder has 5 decoding modules, each of which includes up-sampling with a scale factor of 2, jump connection and standard convolution layer with Dropout.

5. The self-supervised learning based seismic signal noise suppression method of claim 1, wherein, The denoising network is trained as follows: The entire optimization objective function of the denoising network is performed using an Adam gradient descent algorithm optimization, the initial learning rate is 0.0001, the Epoch is set to 15000, the Dropout is turned on during the training process, and the parameters of the denoising network are saved after the optimization objective function converges.

6. The self-supervised learning based seismic signal noise suppression method of claim 1, wherein, The prediction is done on the input single noisy seismic signal, again with Dropout turned on and repeated In the second test, the average of the results of each experiment is chosen as the final result.

7. The self-supervised learning based seismic signal noise suppression method of claim 6, wherein, The average of the results of each experiment was calculated Specifically: wherein is the de-noising network in the Bernoulli experiment.

8. A seismic signal noise suppression system based on self-supervised learning, characterized in that, It comprises: The data module normalizes the single seismic signal, processes the normalized seismic signal by using Bernoulli sampling, and constructs a Bernoulli sampling data pair for an auxiliary task through repeated experiments; The construction module constructs a denoising network based on the coding and decoding structure, and the denoising network comprises a data processing module, an encoder, a decoder and a residual noise separation module; The data processing module comprises a normalization operation and Bernoulli sampling; the two-dimensional seismic signal after encoding by the encoder becomes an input actual seismic signal size ; the two-dimensional seismic signal after decoding by the decoder is restored to the original size; the residual noise separation module is used to calculate the difference between the input noisy seismic signal and the network-predicted clean useful signal, obtain the network-separated noise, and take the noise prior and mean value of 0 as a regularization constraint; The function module determines an optimization objective function of the de-noising network obtained by the construction module , and specifically: wherein, is a target loss function, is a weight coefficient of the noise zero-mean loss function, is a noise zero-mean loss function, is a weight coefficient of the total variation loss function, is a total variation loss function; The training module trains the denoising network obtained by the construction module based on the optimization objective function obtained by the function module until convergence; The suppression module inputs the Bernoulli sampling data pair obtained by the data module into the denoising network obtained by the training module for learning, recovers a single noisy seismic signal, and reconstructs a clean useful signal.

9. The self-supervised learning based seismic signal noise suppression system of claim 8, wherein, Data pairs for secondary tasks As follows: wherein, is the number of Bernoulli samples used, is the mask used for the Bernoulli sampling, is the Hadamard product, is the number of Bernoulli sample data pairs, is the normalized noisy seismic signal.

10. The self-supervised learning based seismic signal noise suppression system of claim 8, wherein, The encoder has 5 encoding modules, each of which includes partial convolution, hollow convolution, residual learning unit and maximum pooling; the decoder has 5 decoding modules, each of which includes up-sampling with a scale factor of 2, jump connection and standard convolution layer with Dropout.

Images