De-noising method and de-noising device for two-dimensional seismic data, equipment and medium

Abstract

The invention relates to a denoising method and device for two-dimensional seismic data, equipment and a medium, and relates to the technical field of seismic data processing, and the denoising method comprises the steps: enabling a denoising network to obtain an initial denoising network through generative adversarial learning; performing supervised iterative learning on the obtained initial denoising network to obtain a target denoising network; denoising the two-dimensional seismic data by using the obtained target denoising network; in the generative adversarial learning, the weight of the denoising network F is frozen in the learning stage of the discriminator network D, and the weight of the discriminator network D is frozen in the learning stage of the denoising network F. According to the denoising method provided by the invention, a generative adversarial learning framework is used to design a noise adversarial extraction learning process, then a supervised iterative learning mode is adopted to carry out further iterative updating on the denoising network, and finally the denoising network with stronger generalization on actual seismic data to be processed is obtained. Therefore, denoising processing is carried out to realize noise suppression of the seismic data.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data processing, and specifically to a denoising method and a denoising device, equipment, and medium for two-dimensional seismic data, and in particular to a denoising method and a denoising device, equipment, and medium for two-dimensional seismic data based on anti-noise extraction. Background Art

[0002] Traditional seismic processing algorithms are mainly based on the structural, continuous, sparsity or similarity assumptions of effective seismic signals, and use time-domain structural guidance filtering, fx-domain predictive filtering, dictionary learning, low-rank fitting, and local similarity filtering algorithms such as NLM and BM3D to suppress coherent noise. Although these algorithms can remove coherent noise to a certain extent, the processing effect on specific data is very dependent on the setting of algorithm parameters. In response to this, a large number of seismic noise suppression methods based on deep learning have been proposed one after another.

[0003] However, most existing deep learning seismic denoising methods use supervised learning to learn the denoiser network model, which requires the provision of a large amount of labeled sample data that is consistent with the seismic data signal and noise pattern to be processed.

[0004] At present, the research mainly synthesizes samples based on the denoising results of traditional methods and the forward response synthesis of artificially generated formation models. The quality of the processing results of traditional methods will directly determine the realism and rationality of the samples synthesized by this method, and often have a great impact on the performance of the final denoising network.

[0005] Therefore, a series of seismic denoising methods based on self-supervised learning and CycleGAN-based unpaired learning have emerged.

[0006] For example, CN116203628A discloses a self-supervised seismic data denoising method, which includes: preprocessing seismic data, dividing each preprocessed seismic data into a plurality of patches in turn, and then randomly extracting seismic data from adjacent positions of each patch, and stacking them in the order of patches to generate noise training image pairs S1(yi) and S2(yi); constructing a self-supervised denoising algorithm and quantitative evaluation indicators; according to the generated noise training image pairs (S1(yi) and S2(yi)), using S1(yi) as input data of the self-supervised denoising network model and S2(yi) as label data, performing self-supervised training of the model based on the self-supervised denoising algorithm, gradually separating noise and signal during multiple rounds of iterations, stopping training when the loss function meets the preset conditions, and obtaining a trained U-Net model; using the trained model to denoise the noisy seismic data in the work area to obtain seismic data with a high signal-to-noise ratio, and evaluating the denoising effect according to the quantitative evaluation indicators.

[0007] However, self-supervised learning seismic denoising often assumes that it can only process random noise and cannot effectively suppress coherent noise in seismic data. At the same time, denoising algorithms based on unpaired learning often have problems such as unstable learning process and the need for manual determination of the training termination time.

[0008] In summary, the current denoising methods for seismic data cannot effectively suppress various noise interferences with complex structures in the original seismic data, and still have the defect of poor denoising effect. Summary of the invention

[0009] In view of the problems existing in the prior art, the purpose of the present invention is to provide a denoising method and denoising device, equipment, and medium for two-dimensional seismic data, so as to solve the problem that the current denoising methods for seismic data cannot effectively suppress various types of noise interference with complex structures in the original seismic data, and still have the defect of poor denoising effect.

[0010] To achieve this object, the present invention adopts the following technical solutions:

[0011] In a first aspect, the present invention provides a denoising method for two-dimensional seismic data, the denoising method comprising:

[0012] The denoising network is used to generate adversarial learning to obtain the initial denoising network;

[0013] The obtained initial denoising network is subjected to supervised iterative learning to obtain the target denoising network;

[0014] De-noising the two-dimensional seismic data using the obtained target de-noising network;

[0015] The learning phase of the discriminator network D in the generative adversarial learning freezes the weights of the denoising network F, and the discriminant loss function used is as follows:

[0016]

[0017] In the formula, X syn refers to the synthesis of noise-free seismic data images; Y field Refers to the actual seismic data image; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network to output the corresponding discrimination result; D(Y field ) refers to inputting the actual seismic data image into the discriminator network and outputting the corresponding discrimination result; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result

[0018] The learning phase of the denoising network F in the generative adversarial learning freezes the weights of the discriminator network D, and the discriminant loss function used is as follows:

[0019] L gen (F|D)=E[(F(Y syn )-X syn ) 2 -(1-D(F(Y field ))) 2 +(1+D(Y syn )) 2 ]

[0020] In the formula, X syn refers to the synthesis of noise-free seismic data images; Y syn Refers to artificially synthesized noisy seismic data images; Y field Refers to the actual seismic data image; F(Y syn ) refers to inputting the artificially synthesized noisy seismic data image into the denoising network and outputting the denoised image; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result.

[0021] The denoising method provided by the present invention uses a generative adversarial learning framework to design a noise adversarial extraction learning process. A denoising network is initially learned, and it is required that the noise extracted by the network can be consistent with the actual noisy seismic data pattern when added to artificially synthesized data. Then, a supervised iterative learning method is adopted. Based on the denoising result of the denoising network and the extracted noise, realistic paired samples are generated, and the denoising network is further iterated and updated. Finally, a denoising network with stronger generalization on the actual seismic data to be processed is obtained. De-noising processing is performed based on the network, thereby achieving noise suppression of the seismic data.

[0022] As a preferred technical solution of the present invention, the implementation of the discriminator network D includes: taking the seismic image Y as input, and after processing in a series of convolutional layers and pooling layers, using tanh activation to convert it into a responsive discriminant result as output.

[0023] As a preferred technical solution of the present invention, the de-noising network F is used in the generative adversarial learning to denoise the actual seismic data Y field Processing to obtain denoising results and the extracted noise estimate

[0024] As a preferred technical solution of the present invention, the generative adversarial learning uses a wavelet estimation method based on autocorrelation function to estimate the actual seismic data Y field The wavelet is estimated and the synthetic noise-free seismic data image is obtained based on it. Then, the noise-free seismic data image X synand the extracted noise estimate The artificially synthesized noisy seismic data image Y is obtained by mixing syn .

[0025] As a preferred technical solution of the present invention, the denoising result is and artificially synthesized noisy seismic data image Y syn The synthetic noise-free seismic data image X syn and the actual earthquake data Y field They correspond to the real noise-free image and the real noisy image respectively.

[0026] Preferably, in the generative adversarial learning, the learning of the discriminator network D and the learning of the denoising network F are performed alternately and iteratively.

[0027] As a preferred technical solution of the present invention, the supervised iterative learning is performed for T rounds of iterations. In the tth round of iterative learning, t≤T, each pair of samples randomly intercepts two local small blocks from different positions in the seismic data to be processed. Then, the network F obtained by the t-1th round of iterative learning is applied (t-1) Perform signal-to-noise estimation:

[0028]

[0029] , and then by Mix and get supervised samples as follows:

[0030]

[0031] For the supervised sample, As iterative learning The synthetic noise-free seismic data image labels are The denoising network F is formed in the form of (t) of learning.

[0032] As a preferred technical solution of the present invention, the supervised iterative learning is combined with the discriminator network D to construct a denoising network F for generative adversarial learning. (t) of learning.

[0033] Preferably, the denoising network F in the supervised iterative learning (t) The loss function for learning is:

[0034]

[0035] In the formula, Refers to the t-1th iteration of the actual seismic data image; Refers to the supervised sample of the t-1th iteration; The supervised loss term of ; as the adversarial supervision loss term.

[0036] Preferably, the learning loss function of the discriminator network D in the supervised iterative learning is:

[0037]

[0038] In the formula, X syn Refers to synthesizing noise-free seismic data images; Refers to the supervised sample of the t-1th iteration; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result.

[0039] In a second aspect, the present invention provides a denoising device for two-dimensional seismic data, the denoising device comprising:

[0040] An initial denoising network acquisition module is used to obtain an initial denoising network by using generative adversarial learning;

[0041] A target denoising network acquisition module is used to perform supervised iterative learning on the obtained initial denoising network to obtain a target denoising network;

[0042] The denoising module is used to denoise the two-dimensional seismic data using the obtained target denoising network.

[0043] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0044] at least one processor; and a memory communicatively coupled to the at least one processor;

[0045] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the denoising method for two-dimensional seismic data described in the first aspect.

[0046] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and when the computer executable instructions are executed by a processor, the denoising method for two-dimensional seismic data described in the first aspect is implemented.

[0047] Compared with the prior art solutions, the present invention has the following beneficial effects:

[0048] (1) The denoising method provided by the present invention is combined with the generative adversarial learning framework to introduce seismic noise adversarial extraction learning, so that the denoising network can be directly used to extract noise based on the actual data to be processed. Then, on this basis, realistic paired samples are generated through the denoising results of the current denoising network, the extracted noise and artificially synthesized data, and the final denoising network is established in a supervised iterative learning manner, and noise suppression is performed on the seismic data to be processed.

[0049] (2) The denoising method provided by the present invention effectively suppresses various types of noise interference with complex structures in the original seismic data. In particular, for the bottom area of ​​the image, the clarity of the corresponding stratum and structural reflection response is significantly improved. This shows that the proposed deep learning denoising process can establish an effective denoising network, suppress the signal-to-noise ratio in actual noisy seismic data, and provide input data with higher signal-to-noise ratio for subsequent seismic interpretation and inversion work. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a denoising method for two-dimensional seismic data provided by an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of a denoising device for two-dimensional seismic data provided by an embodiment of the present invention;

[0052] Figure 3 is a schematic diagram of an electronic device provided by an embodiment of the present invention;

[0053] Figure 4 is a general technical flow chart of counteracting noise generation in Embodiment 1 of the present invention;

[0054] Figure 5 is a schematic diagram of the network structure of the noisy seismic image discriminator in Example 1 of the present invention;

[0055] Figure 6 1 is a schematic diagram of the result of extracting noise from actual data through seismic noise adversarial extraction learning in Example 1 of the present invention;

[0056] Figure 7 This is the result diagram of the original noisy seismic data in Example 1 of the present invention;

[0057] Figure 8 This is a diagram of seismic data results after denoising in Example 1 of the present invention.

[0058] In the figure: 100-initial denoising network acquisition module, 200-target denoising network acquisition module, 300-denoising module;

[0059] 10 - electronic device, 11 - processor, 12 - ROM, 13 - RAM, 14 - bus, 15 - I / O interface, 16 - input unit, 17 - output unit, 18 - storage unit, 19 - communication unit.

[0060] The present invention is further described in detail below. However, the following examples are only simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims. DETAILED DESCRIPTION

[0061] To better illustrate the present invention and facilitate understanding of the technical solution of the present invention, typical but non-limiting embodiments of the present invention are as follows:

[0062] This embodiment provides a denoising method for two-dimensional seismic data. The process is as follows: Figure 1 As shown, the denoising method includes:

[0063] The denoising network is used to generate adversarial learning to obtain the initial denoising network;

[0064] The obtained initial denoising network is subjected to supervised iterative learning to obtain the target denoising network;

[0065] The two-dimensional seismic data is denoised using the obtained target denoising network.

[0066] Among them, the weight of the denoising network F is frozen in the learning stage of the discriminator network D in the generative adversarial learning, and the discriminant loss function used is as follows:

[0067]

[0068] In the formula, X syn refers to the synthesis of noise-free seismic data images; Y field Refers to the actual seismic data image; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network to output the corresponding discrimination result; D(Y field ) refers to inputting the actual seismic data image into the discriminator network and outputting the corresponding discrimination result; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result Among them, (1+D(X syn )) 2 The sum of terms (1-D(Y field )) 2 The term requires that the discrimination results of the real noise-free image and the real noisy image be as close to -1 or 1 as possible; and D(F(Y field )) 2 The item requires the actual seismic data image Y field The judgment result is as close to 0 as possible; at the same time, D(Xsyn +(Y field -F(Y field ))) 2 The item requires artificial synthesis of noisy seismic data image Y syn The discrimination result is also as close to 0 as possible. The discriminator established with this loss function will be able to effectively distinguish between real noise-free images, real noisy images, false noise-free images and false noisy images, and will be used to assist the learning of the denoising network F.

[0069] Among them, the weight of the discriminator network D is frozen in the learning stage of the denoising network F in the generative adversarial learning, and the discriminant loss function used is as follows:

[0070] L gen (F|D)=E[(F(Y syn )-X syn ) 2 -(1-D(F(Y field ))) 2 +(1+D(Y syn )) 2 ]

[0071] In the formula, X syn refers to the synthesis of noise-free seismic data images; Y syn Refers to artificially synthesized noisy seismic data images; Y field Refers to the actual seismic data image; F(Y syn ) refers to inputting the artificially synthesized noisy seismic data image into the denoising network and outputting the denoised image; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network to output the corresponding discrimination result; where (F(Y syn )-X syn ) 2 Item Y syn →X syn The supervised learning method requires the denoising network F to artificially synthesize noisy seismic data image Y syn The denoising result should be as close as possible to the synthetic noise-free seismic data image X syn Consistent; (1-D(F(Y field ))) 2 In the form of adversarial discriminant loss, the denoising network F is required to be able to detect the actual seismic data Y field The denoising result of (1+D(Y syn )) 2 The loss term requires the use of a denoising network F based on the actual seismic data Y field The resulting extracted noise estimate Can synthesize noise-free seismic data images X syn Mixed to obtain artificially synthesized noisy seismic data image Y syn =X syn +(Y field -F(Y field )) is as consistent as possible in style with real noisy seismic images.

[0072] The implementation of the discriminator network D includes: taking the seismic image Y as input, and after applying a series of convolutional layers and pooling layers, using tanh activation to convert the corresponding discriminant results as output:

[0073] Q=D(Y)∈[-1,1],

[0074] Where D(Y) means inputting the seismic image Y into the discriminator network D and outputting the discrimination result.

[0075] Among them, the discriminator network D includes 5 convolutional layers, the first two of which are convolutional layers + PReLU activation functions. The convolutional layer is used to extract the features of the image. The PReLU (Parametric Rectified Linear Unit) activation function is used to increase the nonlinear ability of the network and help the model learn more complex features. The third is the LayerNormalization layer feature standardization layer. This layer is used to normalize the features, which helps to speed up the training process and improve the stability of the model. The fourth and fifth are convolutional layers + PReLU activation functions. The maximum pooling layer is used to reduce the dimension of the feature map, reduce the amount of calculation, and retain the most important feature information. The global average pooling layer compresses the feature map into a single value to capture the global features of the entire image. The output of the global average pooling layer is sent to the fully connected layer, and the tanh activation function is used to limit the output between -1 and 1.

[0076] In the present invention, the discriminator network D has three functions:

[0077] 1) Determine whether the input seismic image Y is a true noise-free image or a true noisy image. When the seismic image Y is a true noise-free image, the output judgment result of D(Y) is required to be -1, and when the seismic image Y is a true noisy image, the output judgment result of D(Y) is required to be 1;

[0078] 2) Determine whether the input seismic image Y is a real noise-free image or a network denoising result. If the seismic image Y is a network denoising result, the output judgment result of D(Y) is required to be negative and as close to 0 as possible, such as -0.01. <D(Y)<0;

[0079] 3) Determine whether the input seismic image Y is a real noisy seismic image or an artificially synthesized noisy seismic data image. If the seismic image Y is an artificially synthesized noisy seismic data image, the output judgment result of D(Y) is required to be positive and as close to 0 as possible, for example, 0 <D(Y)<0.01。

[0080] Among them, the generator implementation method of the denoising network F can be implemented using any existing end-to-end denoising network structure.

[0081] Exemplarily, the denoising network structure of the end-to-end structure may be selected as a common network structure such as DnCNN.

[0082] In the present invention, a denoising network F is used to denoise the actual seismic data Y field Processing to obtain denoising results and the extracted noise estimate

[0083]

[0084] Where Y field Refers to the actual seismic data image; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result Refers to the noise estimate obtained by subtracting the two.

[0085] Based on this, the artificially synthesized reflection coefficient sequence R syn To perform artificial synthesis of noisy seismic images, we use the wavelet estimation method based on the autocorrelation function to estimate the actual seismic data Y field The wavelet is estimated and synthesized to obtain the synthetic noise-free seismic data image Then, the noise-free seismic data image X syn and the extracted noise estimate The artificially synthesized noisy seismic data image Y is obtained by mixing syn ,Right now:

[0086]

[0087] Where Y field Refers to the actual seismic data image; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result Refers to the noise estimate obtained by subtracting the two; R syn Refers to the artificially synthesized reflection coefficient sequence; W field Refers to the wavelet estimation obtained from actual seismic data; X syn Refers to synthesizing noise-free seismic data images.

[0088] In this process, the present invention will remove the noise result and artificially synthesized noisy seismic data image Y syn They correspond to the false noise-free image and the false noisy image generated by the denoising network F respectively; the synthesized noise-free seismic data image X syn and the actual earthquake data Y field They correspond to the real noise-free image and the real noisy image respectively.

[0089] In the generative adversarial learning, the learning of the discriminator network D and the learning of the denoising network F are performed alternately and iteratively.

[0090] For example, the discriminator network D will perform targeted discriminant learning according to the processing results of the current denoising network F, and the denoising network F will be further adjusted according to the updated discriminator network D; when the generative adversarial learning reaches the loss function convergence constraint (during the training process, if the loss on the verification set does not decrease significantly within a certain number of epochs (such as the loss of the current result compared with the previous loss, that is, |current result loss-last result loss| / last result loss×100%≤10% change, the current result loss and the last result loss can correspond to the loss function output value, or can be the result of normalizing the loss function output value), the training can be terminated in advance), the denoising network F will be able to achieve the following functions:

[0091] 1) Through Y syn →X syn With supervised learning, the denoising network F will have preliminary denoising capabilities, which can be used to extract the actual seismic data Y field More effective extraction of clean signals and noise estimation;

[0092] 2) Through (1+D(Y syn )) 2 The constraint of the loss term, based on the noise extracted by the denoising network F, can effectively synthesize sample data consistent with the pattern of the seismic data to be processed, which will help further iterative updates of the subsequent denoising network.

[0093] The generative adversarial learning of the initial denoising network is mainly based on artificially synthesized Y syn →X syn However, due to the variability of underground structure and seismic response, the noise-free seismic data image X syn It may not cover all situations, which may lead to insufficient generalization of the seismic denoising network obtained in the previous step for the processed seismic data. In this regard, the present invention uses the denoising network obtained by seismic noise adversarial extraction learning as the initial denoising network F (0), and further improve its effect on the seismic data to be processed through supervised iterative update learning. During the training process, if the loss on the validation set does not decrease significantly within a certain number of epochs, such as the loss of the current result compared with the previous loss, that is, |current result loss-last result loss| / last result loss×100%≤10%, the current result loss and the last result loss can correspond to the output value of the loss function, or can be the result of normalizing the output value of the loss function, the training learning can be terminated early.

[0094] The denoising network iterative update learning will be carried out for T rounds of iterations. In the specific t-th round of iterative learning, each pair of samples will randomly intercept two local small blocks from different positions in the seismic data to be processed. Then, the network F obtained by the t-1th round of iterative learning is applied (t-1) To estimate the signal-to-noise ratio:

[0095]

[0096] In the formula, Refers to the t-1th iteration of the actual seismic data image; Refers to the local screenshot image at position i of the actual seismic data image; Refers to the local screenshot image at position j of the actual seismic data image; Refers to the noise estimate at the t-1th iteration.

[0097] pass A supervised sample can be obtained by mixing:

[0098]

[0099] In the formula, Refers to the t-1th iteration of the actual seismic data image; refers to the noise estimate of the t-1th iteration; Refers to the local screenshot image at position i of the actual seismic data image; Refers to the local screenshot image at position j of the actual seismic data image; Refers to the supervised samples generated at the t-1th iteration.

[0100] For this sample, we can As iterative learning The synthetic noise-free seismic data image labels are The denoising network F is formed in the form of (t) of learning.

[0101] However, due to Only F (t-1)The processing result is still different from the real synthetic noise-free seismic data image. Therefore, we can further combine the previously obtained discriminator network D to construct a generative adversarial denoising network F. (t) Learning; Specific denoising network F (t) The loss function for learning is:

[0102]

[0103] In the formula, Refers to the t-1th iteration of the actual seismic data image; Refers to the supervised samples of the t-1th iteration.

[0104] in, There is a supervised loss item; is the adversarial supervision loss term; it further requires The style is more like a synthetic noise-free seismic data image. Correspondingly, the learning loss function of the discriminator network D is:

[0105]

[0106] In the formula, X syn Refers to synthesizing noise-free seismic data images; Refers to the supervised sample of the t-1th iteration; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result.

[0107] Since at this stage the discriminator network D only needs to be able to distinguish the synthetic noise-free seismic data image X syn And iterative denoising results Therefore, its loss function is slightly simplified in form. Each round of iterative update will also alternately iterate the learning of the discriminator network D and the denoising network F. (t) of learning.

[0108] It can be seen that, unlike the seismic noise adversarial extraction learning stage, the denoising network F in this stage (t) It is completely based on the actual seismic data to be processed, and only the learning process of the discriminator D involves synthetic data; therefore, as the number of iterations proceeds, the denoising network will be able to estimate the noise based on more and more accurate signal Perform more effective sample synthesis and network learning to obtain a more effective denoising network F (t) Used for processing target area data.

[0109] Furthermore, this embodiment provides a denoising device for two-dimensional seismic data, such as Figure 2 As shown, the denoising device comprises:

[0110] An initial denoising network acquisition module 100 is used to obtain an initial denoising network by using generative adversarial learning;

[0111] A target denoising network acquisition module 200 is used to perform supervised iterative learning on the obtained initial denoising network to obtain a target denoising network;

[0112] The denoising module 300 is used to denoise the two-dimensional seismic data using the obtained target denoising network.

[0113] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0114] Further, the present invention provides an electronic device, which is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0115] like Figure 3 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The I / O interface 15 is also connected to the bus 14.

[0116] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0117] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as a denoising method for two-dimensional seismic data.

[0118] In some embodiments, the denoising method for two-dimensional seismic data can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the denoising method for two-dimensional seismic data described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the denoising method for two-dimensional seismic data in any other appropriate manner (e.g., by means of firmware).

[0119] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0120] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0121] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0122] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0123] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0124] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0125] The server provided in this embodiment includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a denoising method for two-dimensional seismic data is implemented.

[0126] Unless otherwise specifically stated, terms such as processing, computing, calculating, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate and convert data represented as physical (e.g., electronic) quantities within registers or memories of a processing system into other data similarly represented as physical quantities within memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals may be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0127] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments of the present invention may all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person may implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present invention.

[0128] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0129] For software implementation, the techniques described in the present invention can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in the present application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0130] Further, in order to illustrate the good denoising effect that can be achieved by the denoising method for two-dimensional seismic data provided by the present invention, a practical example is used for illustration, as follows:

[0131] Example 1

[0132] This embodiment provides a denoising method for two-dimensional seismic data, which is as follows:

[0133] 1. Seismic noise adversarial extraction learning stage

[0134] The goal of network learning at this stage is to use the generative adversarial learning framework. Figure 4 As shown, through noise adversarial extraction learning, at the scale of unequal treatment of processed seismic data and synthetic seismic data, on the one hand, the initial denoising network is preliminarily learned, and on the other hand, based on the obtained target denoising network, noise is extracted from the actual data to synthesize more realistic denoising samples.

[0135] The specific implementation mainly involves the adversarial learning process of the denoising network F and the discriminator network D.

[0136] The discriminator network D adopts Figure 5 It is implemented by the network structure shown in , which takes the seismic image Y as input and, after applying a series of convolutional layers and pooling layers, uses tanh activation to transform it into a discriminant result of the response as output:

[0137] Q=D(Y)∈[-1,1],

[0138] The discrimination results of the discriminator network D are as follows:

[0139] 1) Determine whether the input seismic image Y is a real noise-free image or a real noisy image. When the seismic image Y is a real noise-free image, the output judgment result of D(Y) is required to be -1, and when the seismic image Y is a real noisy image, the output judgment result of D(Y) is required to be 1.

[0140] 2) Determine whether the input seismic image Y is a real noise-free image or a network denoising result. If Y is a network denoising result, the output judgment result of D(Y) is required to be negative and as close to 0 as possible, -0.01 <D(Y)<0。

[0141] 3) Determine whether the input seismic image Y is a real noisy seismic image or an artificially synthesized noisy seismic data image. If Y is an artificially synthesized noisy seismic data image, the output judgment result of D(Y) is required to be positive and as close to 0 as possible. <D(Y)<0.01。

[0142] The specific noise generation adversarial extraction process will use the denoising network F to implement the generator. Here, the denoising network F can be implemented using any existing end-to-end denoising network structure. This embodiment focuses on the subsequent noise adversarial extraction process.

[0143] In this embodiment, the denoising network F is first used to denoise the actual seismic data Y field Processing to obtain denoising results and the extracted noise estimate

[0144]

[0145] Based on the artificially synthesized reflection coefficient sequence R syn To perform artificial synthesis of noisy seismic images, we first use the wavelet estimation method based on the autocorrelation function to estimate the actual seismic data Y field The wavelet is estimated and synthesized to obtain the synthetic noise-free seismic data image Then, the noise-free seismic data image X syn and the extracted noise estimate The artificially synthesized noisy seismic data image Y is obtained by mixing syn ,Right now:

[0146]

[0147] In this process, this embodiment will remove the noise result and artificially synthesized noisy seismic data image Y synAs the false noise-free image and false noisy image generated by the denoising network F, the synthetic noise-free seismic data image X syn and actual seismic data as the true noise-free image and the true noisy image.

[0148] In this setting, the learning of the discriminator network D will freeze the weights of the denoising network F, and then use the following discriminative loss function:

[0149]

[0150] Among them, (1+D(X syn )) 2 The sum of terms (1-D(Y field )) 2 The term requires that the discrimination results of the real noise-free image and the real noisy image are as close to -1 or 1 as possible, and D(F(Y field )) 2 Item requires actual earthquake data Y field The judgment result is as close to 0 as possible. At the same time, D(X syn +(Y field -F(Y field ))) 2 The item requires artificial synthesis of noisy seismic data image Y syn The discrimination result is also as close to 0 as possible. The discriminator established with this loss function will be able to effectively distinguish between real noise-free images, real noisy images, false noise-free images and false noisy images, and is used to assist the learning of the denoising network F.

[0151] Correspondingly, in the learning stage of the denoising network F, this embodiment freezes the weights of the discriminator network D and incorporates them into the loss function of the denoising network F learning, that is:

[0152] L gen (F|D)=E[(F(Y syn )-X syn ) 2 -(1-D(F(Y field ))) 2 +(1+D(Y syn )) 2 ]

[0153] In the above loss function, (F(Y syn )-X syn ) 2 Item Y syn →X syn The supervised learning method requires the denoising network F to artificially synthesize noisy seismic data image Y syn The denoising result should be as close as possible to the synthetic noise-free seismic data image Y synConsistent, and (1-D(F(Y field ))) 2 In the form of adversarial discriminant loss, the denoising network F is required to be able to detect the actual seismic data Y field The denoising result of (1+D(Y syn )) 2 The loss term requires the use of a denoising network F based on the actual seismic data Y field Extracted noise Can synthesize noise-free seismic data images X syn Mix and synthesize noisy image Y syn =X syn +(Y field -F(Y field )) is as consistent as possible in style with real noisy seismic images.

[0154] The learning of the above discriminator network D and the learning of the denoising network F will be carried out alternately and iteratively. In this process, the discriminator network D will perform targeted discriminant learning based on the processing results of the current denoising network F, and the denoising network F will be further adjusted according to the updated discriminator network D. When this generative adversarial learning reaches a balance (the loss of the current result is compared with the loss of the previous result, that is, |current result loss-last result loss| / last result loss×100%≤10%, and the current result loss and the last result loss correspond to the output value of the loss function), the denoising network F will be able to achieve the following functions:

[0155] 1) Through Y syn →X syn With supervised learning, the denoising network F will have preliminary denoising capabilities, which can be used to extract the actual seismic data Y field More effective extraction of clean signals and noise estimation;

[0156] 2) Through (1+D(Y syn )) 2 The constraint of the loss term, based on the noise extracted by the denoising network F, can effectively synthesize sample data consistent with the pattern of the seismic data to be processed, which will help further iterative updates of the subsequent denoising network.

[0157] 2. Denoising network iterative update learning phase

[0158] The denoising network obtained based on the aforementioned seismic noise adversarial extraction learning is mainly based on the artificially synthesized Y syn →X syn However, due to the variability of underground structure and seismic response, the noise-free seismic data image X synIt may not cover all situations, which may lead to insufficient generalization of the seismic denoising network obtained in the previous step for the processed seismic data. In this regard, the present invention uses the denoising network obtained by seismic noise adversarial extraction learning as the initial denoising network F (0) , and further improve its effect on the seismic data to be processed through iterative update learning.

[0159] The denoising network iterative update learning will be carried out for T rounds of iterations. In the specific t-th round of iterative learning, each pair of samples will randomly intercept two local small blocks from different positions in the seismic data to be processed. Then, the network F obtained by the t-1th round of iterative learning is applied (t-1) To estimate the signal-to-noise ratio:

[0160]

[0161] pass A mixture of , we can get a supervised sample:

[0162]

[0163] For this sample, As iterative learning The synthetic noise-free seismic data image labels are The denoising network F is formed in the form of (t) of learning.

[0164] However, due to Only F (t-1) The processing result is still different from the real synthetic noise-free seismic data image. Therefore, we can further combine the previously obtained discriminator network D to construct a generative adversarial denoising network F. (t) Learning, specifically denoising network F (t) The loss function for learning is:

[0165]

[0166] in, There is a supervised loss term, and To counter the supervision loss term, it further requires The style is more like a synthetic noise-free seismic data image; correspondingly, the learning loss function of the discriminator network is:

[0167]

[0168] Since at this stage the discriminator network D only needs to be able to distinguish the real noise-free image X syn And iterative denoising results Therefore, its loss function is slightly simplified in form. Each round of iterative update will also alternately iterate the learning of the discriminator network D and the denoising network F. (t) It can be seen that, unlike the seismic noise adversarial extraction learning stage, the denoising network F in this stage (t) The process is based entirely on the actual seismic data to be processed. Only the learning process of the discriminator D involves synthetic data. Therefore, as the number of iterations proceeds, the denoising network will be able to estimate the noise based on more and more accurate signal-to-noise More effective sample synthesis and network learning are performed. When the loss of the current result is compared with the loss of the previous result, that is, |current result loss - previous result loss| / last result loss × 100% ≤ 10%, the current result loss and the previous result loss correspond to the output value of the loss function, thereby obtaining a more effective denoising network F. (t) Used for processing target area data.

[0169] 3. Experimental simulation results

[0170] The denoising network is implemented using the MWCNN network structure, such as Figure 6 As shown, the seismic noise adversarial extraction learning process of the present invention is then used to obtain a denoising network, and noise extraction and denoising processing results are performed on multiple noisy actual seismic data. Figure 6 It can be seen from the figure that the target denoising network can effectively extract different modes of data noise in seismic data, thereby performing effective sample synthesis and assisting the subsequent iterative learning of the denoising network; on this basis, according to the process involved in the present invention, the denoising network can be continuously updated for specific actual seismic data. Figure 7 The actual noisy seismic profile of a certain work area shown in is processed using the final target denoising network. The results are as follows Figure 8 shown.

[0171] contrast Figure 7 and Figure 8 It can be seen from the seismic profile in the figure that the denoising method in the present invention effectively suppresses various noise interferences with complex structures in the original seismic data, especially for the bottom area of ​​the image, the clarity of the corresponding stratum and structural reflection responses is significantly improved, which shows that the proposed deep learning denoising process can establish an effective denoising network, suppress the signal-to-noise ratio in actual noisy seismic data, and provide input data with higher signal-to-noise ratio for subsequent seismic interpretation and inversion work.

[0172] It is stated that the present invention illustrates the detailed structural features of the present invention through the above embodiments, but the present invention is not limited to the above detailed structural features, that is, it does not mean that the present invention must rely on the above detailed structural features to be implemented. Those skilled in the art should understand that any improvement of the present invention, equivalent replacement of the components selected by the present invention, addition of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

[0173] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0174] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0175] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A denoising method for two-dimensional seismic data, characterized in that: The denoising method comprises: The denoising network is used to generate adversarial learning to obtain the initial denoising network; The obtained initial denoising network is subjected to supervised iterative learning to obtain the target denoising network; De-noising the two-dimensional seismic data using the obtained target de-noising network; The learning phase of the discriminator network D in the generative adversarial learning freezes the weights of the denoising network F, and the discriminant loss function used is as follows: Where, X syn refers to the synthesis of noise-free seismic data images; Y field Refers to the actual seismic data image; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result; D(Y field ) refers to inputting the actual seismic data image into the discriminator network and outputting the corresponding discrimination result; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result The learning phase of the denoising network F in the generative adversarial learning freezes the weights of the discriminator network D, and the discriminant loss function used is as follows: L gen (F|D)=E[(F(Y syn )-X syn ) 2 -(1-D(F(Y field ))) 2 +(1+D(Y syn )) 2 ] Where, X syn refers to the synthesis of noise-free seismic data images; Y syn Refers to artificially synthesized noisy seismic data images; Y field Refers to the actual seismic data image; F(Y syn ) refers to inputting the artificially synthesized noisy seismic data image into the denoising network and outputting the denoised image; F(Y field ) refers to inputting the actual seismic data image into the denoising network and outputting the denoising result D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result.

2. The denoising method according to claim 1, characterized in that: The implementation of the discriminator network D includes: taking the seismic image Y as input, and after processing in a series of convolutional layers and pooling layers, using tanh activation to convert the corresponding discrimination result as output.

3. The denoising method according to claim 1 or 2, characterized in that: The generative adversarial learning uses the denoising network F to the actual seismic data Y field Processing to obtain denoising results and the extracted noise estimate 4. The denoising method according to any one of claims 1 to 3, characterized in that: The generative adversarial learning uses the wavelet estimation method based on the autocorrelation function to estimate the actual seismic data Y field The wavelet of the noise-free seismic data is estimated and synthesized based on it. Then, the noise-free seismic data image X syn and the extracted noise estimate The artificially synthesized noisy seismic data image Y is obtained by mixing syn .

5. The denoising method according to any one of claims 1 to 4, characterized in that: The denoising result is and artificially synthesized noisy seismic data image Y syn The synthetic noise-free seismic data image X syn and the actual earthquake data Y field They correspond to the real noise-free image and the real noisy image respectively; Preferably, in the generative adversarial learning, the learning of the discriminator network D and the learning of the denoising network F are performed alternately and iteratively.

6. The denoising method according to any one of claims 1 to 5, characterized in that: The supervised iterative learning is performed for T rounds of iterations. In the tth round of iterative learning, t≤T, each pair of samples randomly selects two local small blocks from different positions in the seismic data to be processed. and Then, the network F obtained by the t-1th round of iterative learning is applied (t-1) Perform signal-to-noise estimation: Afterwards, by and Mix and get supervised samples as follows: For the supervised sample, As iterative learning The synthetic noise-free seismic data image labels are The denoising network F is formed in the form of (t) of learning.

7. The denoising method according to any one of claims 1 to 6, characterized in that: The supervised iterative learning is combined with the discriminator network D to construct a denoising network F for generative adversarial learning. (t) learning; Preferably, the denoising network F in the supervised iterative learning (t) The loss function for learning is: In the formula, Refers to the t-1th iteration of the actual seismic data image; Refers to the supervised sample of the t-1th iteration; for The supervised loss term of ; As a counter-supervision loss term; Preferably, the learning loss function of the discriminator network D in the supervised iterative learning is: Where, X syn Refers to synthesizing noise-free seismic data images; Refers to the supervised sample of the t-1th iteration; D(X syn ) refers to inputting the synthetic noise-free seismic data image into the discriminator network and outputting the corresponding discrimination result.

8. A denoising device for two-dimensional seismic data, characterized in that: The denoising device comprises: An initial denoising network acquisition module is used to obtain an initial denoising network by using generative adversarial learning; A target denoising network acquisition module is used to perform supervised iterative learning on the obtained initial denoising network to obtain a target denoising network; The denoising module is used to denoise the two-dimensional seismic data using the obtained target denoising network.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the denoising method for two-dimensional seismic data described in any one of claims 1-7.

10. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, and when the computer executable instructions are executed by a processor, the denoising method for two-dimensional seismic data described in any one of claims 1 to 7 is implemented.

Images