Seismic data random noise processing method and device, electronic equipment and storage medium

Abstract

The invention provides a seismic data random noise processing method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a seismic data sample containing random noise; inputting the seismic data sample into a preset global masker to obtain a global mask body of the seismic data sample; determining blind spot mapping data of the seismic data sample according to the global mask body; inputting the seismic data sample into a denoising network to obtain a sample denoising result; according to the blind spot mapping data and the sample denoising result, optimizing parameters of the denoising network to obtain a target denoising network; and based on the target denoising network, performing denoising processing on the target seismic data to obtain a target denoising result of the target seismic data. According to the method provided by the scheme, the parameters of the de-noising network are automatically optimized according to the blind spot mapping data and the sample de-noising result, so that the finally obtained target de-noising network does not depend on hypothesis / prior, and the de-noising precision of the target de-noising network on the seismic data is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for processing random noise in seismic data. Background Art

[0002] Noise from various sources is inevitable during seismic data collection. Noise will reduce the quality of the collected data and will have an adverse effect on subsequent processing and interpretation. Seismic data collected in the field often contain two types of noise: coherent noise and random noise. Coherent noise has regular differences from effective signals in terms of speed, direction, frequency, etc., and can be removed by traditional methods such as Fourier transform and curvelet transform. Random noise does not have a fixed waveform, and its speed, direction, frequency, etc. are all random, without fixed distribution characteristics, and is not easy to separate from effective signals.

[0003] In the prior art, the denoising of seismic data is usually achieved by studying the difference in properties between effective signals and random noise in seismic data and designing a corresponding noise attenuation algorithm based on the difference in properties between the two.

[0004] However, existing techniques rely on certain assumptions / priors and involve a large number of parameters that need to be manually adjusted, all of which affect the accuracy of traditional denoising methods. Summary of the invention

[0005] The present application provides a method, device, electronic device and storage medium for processing random noise in seismic data to solve the defects of the prior art such as low accuracy.

[0006] The first aspect of the present application provides a method for processing random noise in seismic data, which is characterized by comprising:

[0007] Obtaining seismic data samples containing random noise;

[0008] Inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample;

[0009] Determining blind spot mapping data of the seismic data sample according to the global mask volume;

[0010] Inputting the seismic data sample into a denoising network to obtain a sample denoising result;

[0011] Optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising results to obtain a target denoising network;

[0012] Based on the target denoising network, the target seismic data is denoised to obtain a target denoising result of the target seismic data.

[0013] Optionally, the step of inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample includes:

[0014] Based on the preset global masker, the odd and even channels in the seismic data sample are masked, and the pixel values ​​at the blind spot are interpolated according to the adjacent pixel values ​​to obtain a global mask body of the seismic data sample composed of two mask units.

[0015] Optionally, determining blind spot mapping data of the seismic data sample according to the global mask volume includes:

[0016] Inputting the global mask volume into the denoising network to obtain a corresponding denoised volume;

[0017] The denoised volume is input into a preset global mask mapper to collect blind spots in the denoised volume based on the preset global mask mapper, and the collected blind spots are projected into the same channel to obtain blind spot mapping data of the seismic data sample.

[0018] Optionally, optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising result includes:

[0019] Performing a weighted combination of the blind spot mapping data and the sample denoising results to obtain a combined denoising result;

[0020] Based on a preset loss function, determining a current loss value of the denoising network according to the combined denoising result and the sample denoising result;

[0021] Optimizing parameters of the denoising network according to a current loss value of the denoising network.

[0022] Optionally, the determining, based on a preset loss function and according to the combined denoising result and the sample denoising result, a current loss value of the denoising network includes:

[0023] Based on the following preset loss function, the current loss value of the denoising network is determined according to the combined denoising result and the sample denoising result:

[0024]

[0025] Among them, Loss represents the current loss value of the denoising network, L rec represents the reconstruction loss, L rev Indicates that you can see the loss again, L ort represents the orthogonal convolution regularization term, h(f θ (Ω(y))) represents the sample denoising result, (h(f θ (Ω(y)))+λ·f θ(y)) / (λ+1) represents the combined denoising result, λ is a hyperparameter used to measure the reproducibility of the denoising result, η represents the weight of the orthogonal convolution regularization term, and y represents the seismic data sample containing random noise.

[0026] Optionally, the denoising process is performed on the target seismic data based on the target denoising network to obtain a target denoising result of the target seismic data, including:

[0027] Normalizing the target seismic data to obtain normalized target seismic data;

[0028] Inputting the normalized target seismic data into the target denoising network, extracting seismic data features from the normalized target seismic data based on the target denoising network, and integrating seismic data features of each channel to obtain a target denoised volume;

[0029] The target denoising volume is subjected to inverse normalization processing to obtain a target denoising result of the target seismic data.

[0030] Optionally, the denoising network includes an encoder, a decoder and a skip connection block;

[0031] The encoder and decoder both include a random dropout layer.

[0032] The second aspect of the present application provides a seismic data random noise processing device, comprising:

[0033] An acquisition module, used for acquiring seismic data samples containing random noise;

[0034] A mask module, used for inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample;

[0035] A blind spot module, used to determine blind spot mapping data of the seismic data sample according to the global mask volume;

[0036] A first denoising module, used for inputting the seismic data sample into a denoising network to obtain a sample denoising result;

[0037] An optimization module, used to optimize the parameters of the denoising network according to the blind spot mapping data and the sample denoising result, so as to obtain a target denoising network;

[0038] The second denoising module is used to perform denoising processing on the target seismic data based on the target denoising network to obtain a target denoising result of the target seismic data.

[0039] Optionally, the mask module is specifically used to:

[0040] Based on the preset global masker, the odd and even channels in the seismic data sample are masked, and the pixel values ​​at the blind spot are interpolated according to the adjacent pixel values ​​to obtain a global mask body of the seismic data sample composed of two mask units.

[0041] Optionally, the blind spot module is specifically used for:

[0042] Inputting the global mask volume into the denoising network to obtain a corresponding denoised volume;

[0043] The denoised volume is input into a preset global mask mapper to collect blind spots in the denoised volume based on the preset global mask mapper, and the collected blind spots are projected into the same channel to obtain blind spot mapping data of the seismic data sample.

[0044] Optionally, the optimization module is specifically used to:

[0045] Performing a weighted combination of the blind spot mapping data and the sample denoising results to obtain a combined denoising result;

[0046] Based on a preset loss function, determining a current loss value of the denoising network according to the combined denoising result and the sample denoising result;

[0047] Optimizing parameters of the denoising network according to a current loss value of the denoising network.

[0048] Optionally, the optimization module is specifically used to:

[0049] Based on the following preset loss function, the current loss value of the denoising network is determined according to the combined denoising result and the sample denoising result:

[0050]

[0051] Among them, Loss represents the current loss value of the denoising network, L rec represents the reconstruction loss, L rev Indicates that you can see the loss again, L ort represents the orthogonal convolution regularization term, h(f θ (Ω(y))) represents the sample denoising result, (h(f θ (Ω(y)))+λ·f θ (y)) / (λ+1) represents the combined denoising result, λ is a hyperparameter used to measure the reproducibility of the denoising result, η represents the weight of the orthogonal convolution regularization term, and y represents the seismic data sample containing random noise.

[0052] Optionally, the second denoising module is specifically used to:

[0053] Normalizing the target seismic data to obtain normalized target seismic data;

[0054] Inputting the normalized target seismic data into the target denoising network, extracting seismic data features from the normalized target seismic data based on the target denoising network, and integrating seismic data features of each channel to obtain a target denoised volume;

[0055] The target denoising volume is subjected to inverse normalization processing to obtain a target denoising result of the target seismic data.

[0056] Optionally, the denoising network includes an encoder, a decoder and a skip connection block;

[0057] The encoder and decoder both include a random dropout layer.

[0058] A third aspect of the present application provides an electronic device, comprising: at least one processor and a memory;

[0059] The memory stores computer-executable instructions;

[0060] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the method described in the first aspect and various possible designs of the first aspect.

[0061] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in the first aspect and various possible designs of the first aspect are implemented.

[0062] The technical solution of this application has the following advantages:

[0063] The present application provides a method, device, electronic device and storage medium for processing random noise in seismic data, the method comprising: obtaining seismic data samples containing random noise; inputting the seismic data samples into a preset global masker to obtain a global mask body of the seismic data samples; determining blind spot mapping data of the seismic data samples according to the global mask body; inputting the seismic data samples into a denoising network to obtain sample denoising results; optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising results to obtain a target denoising network; based on the target denoising network, denoising the target seismic data to obtain a target denoising result of the target seismic data. The method provided by the above scheme automatically optimizes the parameters of the denoising network according to the blind spot mapping data and the sample denoising results, so that the target denoising network finally obtained no longer relies on assumptions / priors, thereby improving the denoising accuracy of the target denoising network for seismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application, and a person skilled in the art can also obtain other drawings based on these drawings.

[0065] Figure 1 This is a schematic diagram of the structure of the seismic data random noise processing system based on the embodiment of the present application;

[0066] Figure 2 A schematic diagram of a flow chart of a method for processing random noise in seismic data provided in an embodiment of the present application;

[0067] Figure 3 A schematic diagram of a global masker and a global mask mapper provided in an embodiment of the present application;

[0068] Figure 4 A schematic diagram of an exemplary global masker provided in an embodiment of the present application;

[0069] Figure 5 The overall principle diagram of the random noise processing method for seismic data provided by the embodiment of the present application;

[0070] Figure 6 The ELU activation function image provided by the embodiment of the present application;

[0071] Figure 7 A schematic diagram of a denoising network structure provided in an embodiment of the present application;

[0072] Figure 8 A topological diagram of a standard neural network and a neural network after applying a random dropout layer provided in an embodiment of the present application;

[0073] Fig. 9 (a)~ Fig. 9 (o) is a comparison chart of the results of different methods on sigmoid synthetic seismic data;

[0074] Fig.10 The loss and signal-to-noise ratio change curves of the method provided in the embodiment of the present application after 5000 rounds of training on synthetic data;

[0075] Fig.11 (a)~ Fig.11 (i) is a comparison chart of the results of different methods on actual seismic data;

[0076] Fig.12 (a)~ Fig.12 (i) is a comparison chart of the results of different methods on actual seismic data;

[0077] Fig.13 A schematic diagram of the structure of a random noise processing device for seismic data provided in an embodiment of the present application;

[0078] Fig.14 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0079] The above drawings have shown clear embodiments of the present application, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0080] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0081] In addition, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. In the description of the following embodiments, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0082] At present, oil and gas exploration has shifted from structural oil and gas reservoirs to stratigraphic-lithological oil and gas reservoirs, which are characterized by wide distribution, large number, complex causes, various types, and hidden forms. Therefore, there are higher and more comprehensive requirements for seismic technology and seismic data quality. However, in the process of seismic data acquisition, noise from various sources is inevitable. Noise will reduce the quality of the collected data and will have an adverse effect on subsequent processing and interpretation. Seismic data collected in the field often contain two types of noise: coherent noise and random noise. Coherent noise has regular differences from effective signals in terms of speed, direction, frequency, etc., which can be removed by traditional methods such as Fourier transform and curvelet transform. Random noise has no fixed waveform, and its speed, direction, frequency, etc. are all random, without fixed distribution characteristics, and is not easy to separate from effective signals. Therefore, the study of efficient random noise attenuation methods for seismic data has always attracted much attention from the industry and academia.

[0083] Random noise attenuation in seismic data aims to improve the signal-to-noise ratio of seismic data and enhance the efficiency and accuracy of subsequent seismic data processing and interpretation. The basic principles of most denoising methods include two steps: (1) studying the differences in the properties of effective signals and random noise in seismic data; (2) designing corresponding noise attenuation algorithms to achieve seismic data denoising. Model-driven methods include but are not limited to prediction filter-based, sparse transformation-based, and decomposition-based methods, which can be applied to denoising problems under various noise levels and improve denoising effects to varying degrees. However, these methods usually rely on certain assumptions / priors and involve a large number of parameters that need to be manually adjusted, all of which affect the quality and efficiency of traditional denoising methods.

[0084] In recent years, the rapid development of the artificial intelligence industry has led to the widespread application of deep learning algorithms in seismic data processing. Compared with model-driven methods, data-driven methods have shown great potential in denoising effects. The random noise attenuation method for seismic data based on deep learning does not need to rely on accurate prior assumptions. It uses a large amount of sample data and the powerful feature extraction ability of convolutional neural networks, and adjusts network parameters through the nonlinear approximation ability of deep learning, thereby establishing a complex network model to complete the mapping from noisy data to clean data. However, for seismic data denoising, it is difficult to obtain the noise-free label corresponding to the actual noisy seismic data. The unsupervised method understands the data by mining the inherent characteristics of the data, and only uses the noisy data to train the denoising network, which can achieve a denoising effect similar to that of the supervised denoising method. Since the unsupervised denoising method usually uses the noisy data itself as a label to train the network, one problem that needs to be solved is to prevent the network from eventually learning the identity mapping, that is, overfitting occurs, and the network output is the original noisy image. The unsupervised denoising methods that only use a single noisy data can be divided into two categories: methods based on deep image priors and methods based on blind spot networks. Deep image priors believe that convolutional neural networks are naturally low-impedance to seismic signals and high-impedance to noise, that is, in the fitting process, the model first learns seismic signals and then learns noise. This method usually causes the signal-to-noise ratio to rise rapidly, and then drops sharply due to overfitting. The key to this type of algorithm is to design an appropriate stopping criterion to prevent overfitting and output a relatively good denoising result. The blind spot network randomly extracts some pixels from the noisy image as "blind spots", and the original pixel values ​​corresponding to these blind spots are disrupted or replaced with random values. By establishing a corresponding denoising model, the values ​​of the pixels at the blind spot can be predicted based on other pixels around the blind spot. The learning process of the blind spot network actually assumes that the effective signal in the image has spatial autocorrelation, so its value can be predicted based on the adjacent pixels, while random noise without spatial autocorrelation cannot be predicted. However, due to the considerable information loss of the blind spot network during input or network design, its performance is severely limited.

[0085] In response to the above problems, the embodiments of the present application provide a method, device, electronic device and storage medium for processing random noise in seismic data, the method comprising: obtaining a seismic data sample containing random noise; inputting the seismic data sample into a preset global masker to obtain a global mask body of the seismic data sample; determining the blind spot mapping data of the seismic data sample based on the global mask body; inputting the seismic data sample into a denoising network to obtain a sample denoising result; optimizing the parameters of the denoising network based on the blind spot mapping data and the sample denoising result to obtain a target denoising network; based on the target denoising network, denoising the target seismic data to obtain a target denoising result of the target seismic data. The method provided by the above scheme automatically optimizes the parameters of the denoising network based on the blind spot mapping data and the sample denoising result, so that the target denoising network finally obtained no longer relies on assumptions / priors, thereby improving the denoising accuracy of the target denoising network for seismic data.

[0086] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0087] Embodiment 1:

[0088] First, the structure of the seismic data random noise processing system based on this application is described:

[0089] The random noise processing method, device, electronic device and storage medium for seismic data provided in the embodiments of the present application are suitable for adjusting the parameters of a denoising network for seismic data to improve the denoising accuracy of the denoising network for seismic data. Figure 1 As shown, it is a structural diagram of the seismic data random noise processing system based on the embodiment of the present application, which mainly includes a data acquisition device, a seismic data random noise processing device and a denoising network. Specifically, the data acquisition device is used to collect seismic data samples containing random noise, and send the obtained seismic data samples to the seismic data random noise processing device, which determines the target parameters of the denoising network by performing a series of processing on the seismic data, and then optimizes the parameters of the denoising network to obtain a target denoising network. When there is target seismic data to be processed for noise in the future, the seismic data random noise processing device can use the target denoising network to perform denoising on the target seismic data to obtain a target denoising result of the target seismic data.

[0090] Embodiment 2:

[0091] The embodiment of the present application provides a method for processing random noise of seismic data, which is used to adjust the parameters of a denoising network of seismic data to improve the denoising accuracy of the denoising network for seismic data. The execution subject of the embodiment of the present application is an electronic device, such as a server, a desktop computer, a laptop computer, a tablet computer, and other electronic devices that can be used for random noise processing of seismic data.

[0092] like Figure 2 FIG. 1 is a flow chart of a method for processing random noise in seismic data provided by an embodiment of the present application, the method comprising:

[0093] Step 201, obtaining seismic data samples containing random noise.

[0094] Step 202: Input the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample.

[0095] Specifically, the seismic data samples may be first normalized to their maximum and minimum values, and then a global masker may be preset to achieve global masking of the seismic data samples by masking different pixel channels of the image (seismic data samples) to obtain a global mask volume of the seismic data samples.

[0096] Step 203: Determine blind spot mapping data of the seismic data sample according to the global mask volume.

[0097] It should be noted that blind spots mainly refer to pixels with empty pixel values ​​in seismic data samples.

[0098] Specifically, since the effective signals of adjacent seismic traces (pixel channels) are correlated, the blind spots can be mapped to the corresponding seismic traces according to the correlation relationship and the global mask volume to obtain the blind spot mapping data of the seismic data sample.

[0099] Step 204: input the seismic data samples into a denoising network to obtain sample denoising results.

[0100] It should be noted that the denoising network is a denoising network to be trained, and the seismic data samples are input into the denoising network, but it does not participate in back propagation to avoid the network learning identity mapping, and at the same time, it can use the complete noisy data without information loss.

[0101] Step 205, optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising results to obtain a target denoising network.

[0102] Specifically, the parameter optimization strategy of the denoising network can be determined according to the fusion result of the blind spot mapping data and the sample denoising result, and then the parameters of the denoising network can be optimized according to the target parameters represented by the parameter optimization strategy to obtain the target denoising network.

[0103] Step 206, based on the target denoising network, denoising the target seismic data to obtain a target denoising result of the target seismic data.

[0104] Among them, the target denoising network is the seismic data to be denoised currently. Specifically, the target seismic data can be denoised based on the target denoising network to eliminate the noise data in the target seismic data and obtain the target denoising result of the target seismic data. The target denoising result is the noise-free seismic data after denoising.

[0105] Based on the above embodiments, Figure 3 As shown, it is a schematic diagram of a global masker and a global mask mapper provided in an embodiment of the present application. As an implementable method, based on the above embodiment, in one embodiment, a seismic data sample is input into a preset global masker to obtain a global mask body of the seismic data sample, including:

[0106] Step 2021, based on a preset global masker, mask the odd and even channels in the seismic data sample, and interpolate the pixel values ​​at the blind spot according to the adjacent pixel values ​​to obtain a global mask body of the seismic data sample consisting of two mask units.

[0107] Specifically, the odd / even channels in the seismic data sample y can be masked separately based on the global masker Ω(·) to create a mask consisting of two mask units: The global mask volume Ω(y) is composed of. For the blind spots of the seismic data samples, the pixel values ​​at the blind spots are interpolated from the adjacent pixels.

[0108] Among them, Figure 4 The figure shows the schematic diagram of an exemplary global masker provided by the embodiment of the present application. The specific masking method is: blind spot interpolation from adjacent pixels. The workflow can be divided into the following three steps: (1) The mask is generated by masking the odd / even channels in the image y respectively. Convolve the 3×3 convolution kernel with the image y to generate y c (The step size is 1 and the edge pixels are filled with zeros). Then, y c Calculate the Hadamard product with the mask to get y m (2) Calculate the Hadamard product ⊙ between y and (1-mask) to obtain y inv (3) y m and inv Sum them up and finally get the mask image (global mask volume).

[0109] Specifically, in one embodiment, determining blind spot mapping data of seismic data samples according to the global mask volume includes:

[0110] Step 2031, inputting the global mask volume into the denoising network to obtain the corresponding denoised volume;

[0111] Step 2032, input the denoised volume to a preset global mask mapper to collect blind spots in the denoised volume based on the preset global mask mapper, and project the collected blind spots to the same channel to obtain blind spot mapping data of the seismic data sample.

[0112] Specifically, Figure 3 As shown, the denoising volume f can be mapped based on the preset global mask mapper h(·) θ (Ω(y)) is sampled at the blind point, and the result is projected onto the same channel to obtain h(f θ (Ω(y))), which is obtained by the mapper according to the sampled position and pixel value. Among them, the preset global mask mapper allows the loss function to optimize all blind spots at the same time, speed up network training, and achieve global optimization.

[0113] Based on the above embodiments, Figure 5 As shown, it is an overall schematic diagram of the random noise processing method for seismic data provided by the embodiment of the present application. As an implementable method, based on the above embodiment, in one embodiment, according to the blind spot mapping data and the sample denoising results, the parameters of the denoising network are optimized, including:

[0114] Step 2051, weighted combination of blind spot mapping data and sample denoising results to obtain a combined denoising result;

[0115] Step 2052, based on a preset loss function, according to the combined denoising result and the sample denoising result, determine the current loss value of the denoising network;

[0116] Step 2053, optimizing the parameters of the denoising network according to the current loss value of the denoising network.

[0117] Specifically, in one embodiment, the current loss value of the denoising network may be determined based on the following preset loss function (minimizing loss function) according to the combined denoising result and the sample denoising result:

[0118]

[0119] To facilitate the solution, it is converted into the following form:

[0120]

[0121] Among them, Loss represents the current loss value of the denoising network, L rec represents the reconstruction loss, L rev Indicates that you can see the loss again, L ort represents the orthogonal convolution regularization term, h(f θ(Ω(y))) represents the sample denoising result, (h(f θ (Ω(y)))+λ·f θ (y)) / (λ+1) represents the combined denoising result, λ is a hyperparameter used to measure the reproducibility of the denoising result, η represents the weight of the orthogonal convolution regularization term, and y represents the seismic data sample containing random noise. In addition to the reconstruction loss, the loss function also includes the reproducibility loss and the orthogonal convolution regularization constraint. The reproducibility loss can make the blind spots visible, and the denoising network can learn directly from the original noisy image without losing information or falling into the identity mapping. The orthogonal convolution regularization term can avoid the high correlation or even redundancy of deep convolution filters.

[0122] It should be noted that after the denoising network training is completed, the sample denoising result f of the original noisy data (seismic data sample) is θ (y) is the final denoising result, which is compared with the result of the global mask mapper h(f θ (Ω(y))) and the weighted combination result (h(f θ (Ω(y)))+λ·f θ (y)) / (λ+1) is better. Figure 6 As shown, it is the ELU activation function image provided by the embodiment of the present application. The denoising network can use the ELU activation function. Compared with ReLU, the gradient of ELU is non-zero for all negative values, there is no problem of neuron death, and the output distribution is zero mean, which can speed up the training speed. Compared with LeakyReLU, ELU is unilaterally saturated, can converge better, and can also make neurons more robust to noise interference.

[0123] Based on the above embodiments, Figure 7 As shown, it is a schematic diagram of the denoising network structure provided in an embodiment of the present application. As an implementable manner, on the basis of the above embodiment, in one embodiment, based on the target denoising network, the target seismic data is denoised to obtain the target denoising result of the target seismic data, including:

[0124] Step 2061, normalizing the target seismic data to obtain normalized target seismic data;

[0125] Step 2062, inputting the normalized target seismic data into the target denoising network, extracting seismic data features from the normalized target seismic data based on the target denoising network, and integrating seismic data features of each channel to obtain a target denoising volume;

[0126] Step 2063, performing inverse normalization processing on the target denoising volume to obtain a target denoising result of the target seismic data.

[0127] Specifically, the denoising network can use unsupervised learning to optimize parameters.

[0128] Specifically, Figure 7 As shown in the figure, the denoising network includes an encoder, a decoder and a skip connection block; both the encoder and the decoder include a random dropout layer. The network depth is 5 layers. As the network depth increases, the number of convolution filters is 8, 16, 32, 64, and 128, respectively. The encoder contains five encoding blocks, which extract features through maximum pooling and convolution modules, and add a random dropout layer to prevent overfitting. The decoder contains five decoding blocks, which reconstruct spatial information through upsampling operations and convolution modules, and add a random dropout layer to prevent overfitting. Skip connection blocks are added to the fourth and fifth layers of the network for feature fusion, so that the network can better extract meaningful seismic data features. Finally, a 1×1 convolution layer and a sigmoid activation function are used to integrate different channel features and output the denoised volume.

[0129] Among them, Figure 8 As shown, there is a topological diagram of a standard neural network provided in an embodiment of the present application and a neural network after applying a random deactivation layer. The random deactivation layer is a layer that temporarily discards a part of the neural network units from the network according to a certain probability during the training process of the deep learning network, which can make the model more robust and reduce overfitting.

[0130] In order to facilitate those skilled in the art to better understand the random noise processing method for seismic data provided in the embodiment of the present application, the random noise attenuation processing method for unsupervised seismic data based on the improved blind spot strategy provided in the embodiment of the present application is compared with the traditional and deep learning-based denoising methods (i.e., fx deconvolution, blind spot scheme, and blind path scheme) to verify the denoising performance of the method provided by the implementation of the present application. For the blind spot and blind path schemes, the embodiment of the present application randomly generates pixel-by-pixel masks and path-by-path masks with a missing rate of 0.7, respectively, and uses the same network structure, and adds an orthogonal convolution regularization term to the loss function.

[0131] Among them, Fig. 9 (a)~ Fig. 9 (o) is a comparison chart of the results of different methods on sigmoid synthetic seismic data, showing the denoising results, residual graphs, and error graphs by column. The subgraphs in the same row correspond to the same algorithm. (a) is synthetic seismic data; (b) is Gaussian white noise with a mean of 0 and a variance of 1; (c) is seismic data after noise addition. Fig. 9 (d), (g), (j), and (m) show that fx deconvolution can only remove part of the noise; the denoising results of the blind spot solution and the blind path solution are relatively clear; while the denoising results obtained by the method provided in the embodiment of the present application do not contain noise, and the effective signal is clear and continuous. Fig. 9(e), (h), (k), and (n) show that the denoising result obtained by the method provided in the embodiment of the present application has the least effective signal leakage. Fig. 9 (b) is the most similar. At the same time, from the error Fig. 9 (f), (i), (l), and (o) show that the difference between the denoising result obtained by the method provided in the embodiment of the present application and the clean data is minimal.

[0132] like Fig.10 As shown, the loss and signal-to-noise ratio change curves of the method provided in the embodiment of the present application for 5000 rounds of training on synthetic data. The loss shows a downward trend with the increase of the number of iterations and gradually stabilizes. Observing the signal-to-noise ratio curve, it can be seen that the signal-to-noise ratio reaches a peak of 14.68dB at around 3200 rounds, and then slowly decreases. When the signal-to-noise ratio is trained to 5000 rounds, it is 14.24dB. The overfitting phenomenon is not serious, so the method provided by the implementation of the present application can be applied to blind denoising.

[0133] like Fig.11 (a)~ Fig.11 (i) shows the comparison of the results of different methods on actual seismic data, which is a two-dimensional section of a three-dimensional data volume. Fig.11 The denoising results and residual graphs are shown in columns, and the sub-graphs in the same row correspond to the same algorithm. Fig.11 (b), (d), (f), and (h) show that fx deconvolution cannot completely remove noise; the blind spot solution and the blind path solution have similar denoising results to the method provided in the embodiment of the present application. Fig.11 (c), (e), (g), and (i) show that fx deconvolution has serious signal leakage; the blind spot solution and the blind path solution remove some signals as noise; and the method provided in the embodiment of the present application can protect the valid signal while removing the noise.

[0134] like Fig.12 (a)~ Fig.12 (i) shows the comparison of the results of different methods on actual seismic data. Fig.12 The rectangular area in (a) is used for more detailed comparison. This area contains more noise, and some valid signals are submerged by the noise. Fig.12 (b) and (c) show that the denoising result of fx deconvolution still contains some noise and there is a certain degree of signal leakage. Fig.12 As can be seen from the arrows (d), (f), and (h), the blind spot and blind path solutions cannot recover some weak signals well, and the continuity of the event axis is poor; while the method provided in the embodiment of the present application can retain clear event axis information while removing random noise. Fig.12It can be seen from (e), (g), and (i) that the noise removed by the method provided in the embodiment of the present application contains the least effective signal.

[0135] Compared with the existing random noise attenuation method for seismic data, the random noise processing method for seismic data provided by the embodiment of the present application has the following advantages:

[0136] 1. Compared with traditional denoising methods, the method provided by the present application does not require any prior information or assumptions.

[0137] 2. Compared with the supervised denoising method, the method provided by the present application does not need to construct noisy-clean or noisy-noisy training data pairs, saving the economic, time and computational costs required to collect and train a large number of data pairs.

[0138] 3. Compared with the denoising method in the unsupervised denoising method that needs to construct a training set, the method provided by the implementation of this application only uses a single noisy seismic data for training, without the need to collect a large amount of synthetic or actual seismic data. It is convenient and does not need to consider the generalization ability issue.

[0139] 4. Compared with the single image denoising method with sensitive hyperparameters in the unsupervised denoising method, the method provided by the implementation of the present application is suitable for blind denoising. Like deep image prior, its working principle is to reconstruct noisy images by training a convolutional neural network with randomly initialized weights. Since convolutional neural networks are naturally low-impedance to seismic signals and high-impedance to noise, that is, in the fitting process, the model first learns the seismic signal and then learns the noise. However, the algorithm is highly sensitive to the number of iterations. If the number of iterations is too many, it will quickly overfit and the trend will be obvious. Therefore, it is not completely suitable for blind denoising. Using this algorithm to obtain a relatively good denoising result depends on formulating a suitable stopping criterion. If a fixed number of iterations is used to force blind denoising, the denoising result obtained will be much lower than the best denoising result that the algorithm can achieve.

[0140] 5. Compared with the blind spot network that only uses a single image for denoising in the unsupervised denoising method, the method provided by the implementation of this application can use complete noisy data without information loss, thereby improving the performance of the denoising network. The method provided by the implementation of this application can retain clear event information while removing random noise, thereby improving the signal-to-noise ratio of seismic data.

[0141] Through the comparison results of different methods, it can be determined that the method provided by the embodiment of the present application actually solves the performance degradation and overfitting problems of the existing unsupervised method that only uses a single noisy seismic data for training. Its effectiveness has been verified by synthetic and actual data testing. The method provided by the embodiment of the present application can retain clear event information while removing noise, thereby improving the signal-to-noise ratio and visual quality of seismic data.

[0142] The random noise processing method for seismic data provided by the embodiment of the present application is obtained by obtaining seismic data samples containing random noise; inputting the seismic data samples into a preset global masker to obtain a global mask body of the seismic data samples; determining the blind spot mapping data of the seismic data samples according to the global mask body; inputting the seismic data samples into a denoising network to obtain a sample denoising result; optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising result to obtain a target denoising network; based on the target denoising network, denoising the target seismic data to obtain a target denoising result of the target seismic data. The method provided by the above scheme automatically optimizes the parameters of the denoising network according to the blind spot mapping data and the sample denoising result, so that the target denoising network finally obtained is no longer dependent on assumptions / priorities, thereby improving the denoising accuracy of the target denoising network for seismic data. In addition, the method can use complete noisy data without information loss, and update the denoising network parameters with a mixed loss function with orthogonal convolution regularization constraints, thereby improving the performance of the denoising network. At the same time, adding a random inactivation layer to the denoising network enhances the robustness of the model and reduces overfitting.

[0143] Embodiment three:

[0144] An embodiment of the present application provides a seismic data random noise processing device, which is used to execute the seismic data random noise processing method provided in the above embodiment.

[0145] like Fig.13 , which is a schematic diagram of the structure of the random noise processing device for seismic data provided by the embodiment of the present application. The random noise processing device for seismic data 130 comprises: an acquisition module 1301, a mask module 1302, a blind spot module 1303, a first denoising module 1304, an optimization module 1305 and a second denoising module 1306.

[0146] Among them, the acquisition module is used to obtain seismic data samples containing random noise; the mask module is used to input the seismic data samples into a preset global masker to obtain a global mask body of the seismic data samples; the blind spot module is used to determine the blind spot mapping data of the seismic data samples according to the global mask body; the first denoising module is used to input the seismic data samples into the denoising network to obtain the sample denoising results; the optimization module is used to optimize the parameters of the denoising network according to the blind spot mapping data and the sample denoising results to obtain the target denoising network; the second denoising module is used to denoise the target seismic data based on the target denoising network to obtain the target denoising results of the target seismic data.

[0147] Specifically, in one embodiment, the mask module is specifically used to:

[0148] Based on a preset global masker, the odd and even channels in the seismic data samples are masked, and the pixel values ​​at the blind spots are interpolated according to the adjacent pixel values ​​to obtain a global mask body of the seismic data samples consisting of two mask units.

[0149] Specifically, in one embodiment, the blind spot module is specifically used to:

[0150] Input the global mask volume into the denoising network to obtain the corresponding denoised volume;

[0151] The denoised volume is input into a preset global mask mapper to collect blind spots in the denoised volume based on the preset global mask mapper, and the collected blind spots are projected into the same channel to obtain blind spot mapping data of the seismic data sample.

[0152] Specifically, in one embodiment, the optimization module is specifically used to:

[0153] A weighted combination of the blind spot mapping data and the sample denoising results is performed to obtain a combined denoising result;

[0154] Based on the preset loss function, the current loss value of the denoising network is determined according to the combined denoising result and the sample denoising result;

[0155] According to the current loss value of the denoising network, optimize the parameters of the denoising network.

[0156] Specifically, in one embodiment, the optimization module is specifically used to:

[0157] Based on the following preset loss function, the current loss value of the denoising network is determined according to the combined denoising results and the sample denoising results:

[0158]

[0159] To facilitate the solution, it is converted into the following form:

[0160]

[0161] Among them, Loss represents the current loss value of the denoising network, L rec represents the reconstruction loss, L rev Indicates that you can see the loss again, L ort represents the orthogonal convolution regularization term, h(f θ (Ω(y))) represents the sample denoising result, (h(f θ (Ω(y)))+λ·f θ (y)) / (λ+1) represents the combined denoising result, λ is a hyperparameter used to measure the reproducibility of the denoising result, η represents the weight of the orthogonal convolution regularization term, and y represents the seismic data sample containing random noise.

[0162] Specifically, in one embodiment, the second denoising module is specifically used to:

[0163] Performing normalization processing on the target seismic data to obtain normalized target seismic data;

[0164] The normalized target seismic data is input into the target denoising network, so as to extract seismic data features of the normalized target seismic data based on the target denoising network, and integrate the seismic data features of each channel to obtain a target denoising volume;

[0165] The target denoising volume is denormalized to obtain the target denoising result of the target seismic data.

[0166] Specifically, in one embodiment, the denoising network includes an encoder, a decoder, and a skip connection block;

[0167] Both the encoder and decoder include dropout layers.

[0168] Regarding the seismic data random noise processing device in this 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.

[0169] The random noise processing device for seismic data provided in the embodiment of the present application is used to execute the random noise processing method for seismic data provided in the above embodiment. Its implementation method and principle are the same and will not be repeated here.

[0170] An embodiment of the present application provides an electronic device for executing the random noise processing method for seismic data provided in the above embodiment.

[0171] like Fig.14 1 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 140 includes: at least one processor 1401 and a memory 1402.

[0172] The memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the random noise processing method for seismic data provided in the above embodiment.

[0173] An electronic device provided in an embodiment of the present application is used to execute the random noise processing method for seismic data provided in the above embodiment. Its implementation method and principle are the same and will not be repeated here.

[0174] An embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the random noise processing method for seismic data provided in any of the above embodiments is implemented.

[0175] The storage medium containing computer executable instructions in the embodiment of the present application can be used to store computer executable instructions of the seismic data random noise processing method provided in the aforementioned embodiment. Its implementation method and principle are the same and will not be repeated here.

[0176] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0177] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0178] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0179] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0180] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for processing random noise in seismic data, characterized in that: include: Obtaining seismic data samples containing random noise; Inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample; Determining blind spot mapping data of the seismic data sample according to the global mask volume; Inputting the seismic data sample into a denoising network to obtain a sample denoising result; Optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising results to obtain a target denoising network; Based on the target denoising network, the target seismic data is denoised to obtain a target denoising result of the target seismic data.

2. The method according to claim 1, characterized in that The step of inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample comprises: Based on the preset global masker, the odd and even channels in the seismic data sample are masked, and the pixel values ​​at the blind spot are interpolated according to the adjacent pixel values ​​to obtain a global mask body of the seismic data sample composed of two mask units.

3. The method according to claim 1, characterized in that Determining blind spot mapping data of the seismic data sample according to the global mask volume includes: Inputting the global mask volume into the denoising network to obtain a corresponding denoised volume; The denoised volume is input into a preset global mask mapper to collect blind spots in the denoised volume based on the preset global mask mapper, and the collected blind spots are projected into the same channel to obtain blind spot mapping data of the seismic data sample.

4. The method according to claim 1, characterized in that Optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising result includes: Performing a weighted combination of the blind spot mapping data and the sample denoising results to obtain a combined denoising result; Based on a preset loss function, determining a current loss value of the denoising network according to the combined denoising result and the sample denoising result; Optimizing parameters of the denoising network according to a current loss value of the denoising network.

5. The method according to claim 4, characterized in that The determining, based on a preset loss function and according to the combined denoising result and the sample denoising result, a current loss value of the denoising network comprises: Based on the following preset loss function, the current loss value of the denoising network is determined according to the combined denoising result and the sample denoising result: Among them, Loss represents the current loss value of the denoising network, L rec represents the reconstruction loss, L rev Indicates that you can see the loss again, L ort represents the orthogonal convolution regularization term, h(f θ (Ω(y))) represents the sample denoising result, (h(f θ (Ω(y)))+λ·f θ (y)) / (λ+1) represents the combined denoising result, λ is a hyperparameter used to measure the reproducibility of the denoising result, η represents the weight of the orthogonal convolution regularization term, and y represents the seismic data sample containing random noise.

6. The method according to claim 1, characterized in that The denoising process is performed on the target seismic data based on the target denoising network to obtain a target denoising result of the target seismic data, including: Normalizing the target seismic data to obtain normalized target seismic data; Inputting the normalized target seismic data into the target denoising network, extracting seismic data features from the normalized target seismic data based on the target denoising network, and integrating seismic data features of each channel to obtain a target denoised volume; The target denoising volume is subjected to inverse normalization processing to obtain a target denoising result of the target seismic data.

7. The method according to claim 1, characterized in that The denoising network includes an encoder, a decoder and a skip connection block; The encoder and decoder both include a random dropout layer.

8. A seismic data random noise processing device, characterized in that: include: An acquisition module, used for acquiring seismic data samples containing random noise; A mask module, used for inputting the seismic data sample into a preset global masker to obtain a global mask volume of the seismic data sample; A blind spot module, used to determine blind spot mapping data of the seismic data sample according to the global mask volume; A first denoising module, used for inputting the seismic data sample into a denoising network to obtain a sample denoising result; An optimization module, used for optimizing the parameters of the denoising network according to the blind spot mapping data and the sample denoising result, so as to obtain a target denoising network; The second denoising module is used to perform denoising processing on the target seismic data based on the target denoising network to obtain a target denoising result of the target seismic data.

9. An electronic device, characterized in that: include: at least one processor and memory; The memory stores computer-executable instructions; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when a processor executes the computer-executable instructions, the method according to any one of claims 1 to 7 is implemented.

Images