Intelligent surface wave suppression method and device based on structural constraint and electronic equipment

Abstract

The invention discloses an intelligent surface wave suppression method and device based on structural constraint and electronic equipment, and the method comprises the steps: carrying out the surface wave denoising processing of actual pre-stack common shot gather data, and carrying out the evaluation and screening of data for making a sample label according to a denoising effect; performing regularization processing on the screened pre-stack common shot set data and the corresponding de-noised data to generate sample label data pairs with the same shape and size; establishing a deep neural network of a UNET structure, and introducing a structural similarity regularization function into a loss function of the deep neural network to enhance the effective signal identification capability of the network; training a deep neural network by using the training data set, wherein the trained deep neural network can perform surface wave suppression on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data; and verifying the training effect of the deep neural network by using the test data set. According to the invention, intelligent high-precision suppression of the surface wave noise can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of geophysical exploration, and more specifically, relates to an intelligent surface wave suppression method, device and electronic equipment based on structural constraints. Background Art

[0002] Surface wave noise is the main type of noise in land seismic data. It has the characteristics of strong energy, low frequency, low speed and dispersion. How to suppress surface waves efficiently is the key to seismic data processing. When conventional algorithms suppress surface waves, they mainly perform signal-to-noise separation in the time-space domain or transform domain (such as frequency wavenumber domain or curvelet domain) based on the difference in energy, frequency and apparent velocity between surface waves and effective signals to achieve the purpose of removing surface waves.

[0003] At present, the surface wave suppression technology processes based on traditional algorithms commonly used in actual production are mainly:

[0004] (1) Noise suppression based on signal-to-noise energy and frequency differences through amplitude spectrum editing;

[0005] (2) Noise removal based on the signal-to-noise apparent velocity difference in the frequency space domain or the frequency wave number domain;

[0006] (3) Based on the dispersion characteristics of surface roll, the surface roll model is predicted by picking up the dispersion curve and noise removal is achieved through an adaptive subtraction algorithm.

[0007] Therefore, in the actual seismic data processing process, the corresponding appropriate algorithm process can be selected according to the surface wave characteristics of the work area to suppress the surface wave. Overall, the above technology is relatively mature and has been widely used in actual production. However, in the actual application process of traditional algorithms, processors need to understand the principles of the algorithm and perform manual parameter adjustment testing and quality control. The denoising accuracy and efficiency may be affected due to insufficient processing experience. Summary of the invention

[0008] The purpose of the present invention is to provide an intelligent surface roll suppression method, device and electronic equipment based on structural constraints, so as to realize intelligent high-precision suppression of surface roll noise.

[0009] To achieve the above objectives, in a first aspect, the present invention proposes an intelligent surface roll suppression method based on structural constraints, comprising:

[0010] Perform surface wave denoising on the actual pre-stack common shot gather data, and select the data for making sample labels based on the denoising effect evaluation;

[0011] The screened pre-stack common shot data and their corresponding denoised data are processed in a regularized manner to generate sample label data pairs with the same shape and size to form a training data set, where the samples are pre-stack common shot data and the labels are the corresponding valid shot data after removing surface wave noise.

[0012] Establishing a deep neural network with a UNET structure, and introducing a structural similarity regularization function into the loss function of the deep neural network to enhance the ability of the network to identify valid signals;

[0013] Setting network training hyperparameters, using the training data set to train the deep neural network, so that the trained deep neural network can suppress surface waves on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data;

[0014] The test data set is used to verify the training effect of the deep neural network.

[0015] Optionally, the deep neural network includes an encoding part and a decoding part, the encoding part is used for data downsampling, and the decoding part is used for data upsampling;

[0016] The encoding part includes 5 groups of basic units, each group of basic units includes 2 convolutional layers and 1 maximum pooling layer, the convolution kernel size of the convolutional layer is 3x3, the number of channels of the convolutional layer in each group of basic units is twice the number of channels of the previous group of convolutional layers, and the number of channels of the convolutional layer increases from the initial 32 to 512;

[0017] The decoding part includes 4 groups of basic units, each basic unit includes 1 deconvolution layer and 2 convolution layers, the convolution kernel size of the deconvolution layer is 2x2, and the convolution kernel size of the convolution layer is 3x3.

[0018] Optionally, the decoding part uses a replication method to perform data fusion with the down-sampled data with symmetry of the UNET structure in each up-sampling process, so as to enhance data information in the up-sampling process.

[0019] Optionally, the structural similarity regularization function is:

[0020]

[0021] Among them, ssim(Y label ,Y pred ) represents the label data Y label And the input prediction data Y pred The structural similarity of Y represents the mean of data Y, and σ Y represents the standard deviation of data Y, and Represents the costandard deviation of two data, where C1 and C2 are constants.

[0022] Optionally, the loss function of the deep neural network is:

[0023]

[0024] Among them, Q represents the loss function used by the deep neural network, α and β represent weight factors, Nwin represents the number of sub-windows used to calculate the similarity of two data structures in order to improve the accuracy of similarity calculation; α||Y label -Y pred || 2 is the bi-norm of the label data and the predicted data, which is used to measure the energy similarity of the two data. It is a structural similarity function, which is used to measure the structural similarity of two data.

[0025] Optionally, when applying the structural similarity regularization function, the input label data and prediction data are divided into two-dimensional time windows, calculated in each local time window, and finally the mean is taken as the similarity measure of the label data and the prediction data, where the time window size used is 11x11 and the time window step size is 2x2.

[0026] Optionally, the deep neural network is trained using an Adam iterative optimizer.

[0027] In a second aspect, the present invention provides an electronic device, the electronic device comprising:

[0028] at least one processor; and,

[0029] a memory communicatively connected to the at least one processor; wherein,

[0030] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any intelligent surface roll suppression method based on structural constraints described in the first aspect.

[0031] In a third aspect, the present invention proposes a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute any intelligent surface roll suppression method based on structural constraints as described in the first aspect.

[0032] In a fourth aspect, the present invention proposes an intelligent surface wave suppression device based on structural constraints, comprising:

[0033] The data screening module is used to perform surface wave denoising on the actual pre-stack common shot gather data and screen out the data for making sample labels based on the denoising effect evaluation;

[0034] The sample label preparation module is used to perform regularization processing on the screened pre-stack common shot data and its corresponding denoised data to generate sample label data pairs with the same shape and size to form a training data set, where the sample is the pre-stack common shot data and the label is the corresponding effective shot data after removing the surface wave noise;

[0035] A network building module is used to build a deep neural network with a UNET structure and introduce a structural similarity regularization function into the loss function of the deep neural network to enhance the ability of the network to identify valid signals;

[0036] A training module is used to set network training hyperparameters, and train the deep neural network using the training data set, so that the trained deep neural network can suppress surface waves on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data;

[0037] The test module is used to verify the training effect of the deep neural network using the test data set.

[0038] The beneficial effects of the present invention are:

[0039] The present invention establishes a high-precision UNET network architecture and introduces a structural similarity regularization function in the loss function to enhance the network's ability to identify effective signals. Through network training, the structural characteristics of the signal-noise are fully explored, and the intelligent and high-precision suppression of surface wave noise is achieved. It can better serve the noise suppression link of seismic data preprocessing, and provide new technical ideas for future high-precision and efficient seismic data processing, providing strong technical support for actual production applications.

[0040] The system of the present invention has other characteristics and advantages, which will be apparent from the drawings incorporated herein and the following detailed description, or will be described in detail in the drawings incorporated herein and the following detailed description, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, in which like reference numerals generally represent like components.

[0042] Figure 1 A step diagram of an intelligent surface roll suppression method based on structural constraints according to the present invention is shown.

[0043] Figure 2 A schematic diagram of the deep neural network structure used in an intelligent surface roll suppression method based on structural constraints according to the present invention is shown.

[0044] Figure 3a A schematic diagram of full-field wave data before processing in one embodiment is shown.

[0045] Figure 3b A schematic diagram of surface roll suppression results obtained by processing using only the two-norm loss function in one embodiment is shown.

[0046] Figure 3c A schematic diagram of surface roll suppression results obtained by using the intelligent surface roll suppression method based on structural constraints of the present invention in one embodiment is shown. DETAILED DESCRIPTION

[0047] The development of intelligent deep learning has provided new ideas for seismic data processing. The intelligent processing flow based on deep learning can effectively reduce manual intervention and improve processing efficiency. In recent years, the intelligent denoising technology of seismic data has also been widely studied and gradually applied in the actual data processing process. Overall, the current intelligent denoising algorithm is mainly based on random noise in the application of actual data, and the research on surface wave noise suppression has also achieved certain results. However, in the identification of signal-to-noise features, it mainly uses the energy matching information of the prediction results and the label data, and fails to make full use of the structural information of the signal. Therefore, there is still room for improvement in the denoising accuracy.

[0048] In view of the problems existing in the prior art, the present invention proposes an intelligent surface wave suppression method based on structural constraints. By using high-precision denoised data as labels, a high-precision deep neural network is built. The loss function always introduces structural similarity regularization constraints while ensuring that the energy of the prediction result matches the label data, further improving the signal recognition accuracy, achieving effective removal of surface wave noise, and improving denoising efficiency, avoiding manual intervention in denoising applications. The present invention can better serve seismic data processing and lay a foundation for the production, application and promotion of seismic data surface wave noise suppression technology. There is no corresponding research literature on this method.

[0049] The present invention will be described in more detail below with reference to the accompanying drawings. Although preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides an intelligent surface roll suppression method based on structural constraints, including:

[0052] S1: Perform surface wave denoising on the actual pre-stack common shot gather data, and select the data for making sample labels based on the denoising effect evaluation;

[0053] S2: Regularize the selected pre-stack common shot data and its corresponding denoised data to generate sample label data pairs with the same shape and size to form a training data set, where the sample is the pre-stack common shot data and the label is the corresponding effective shot data after removing the surface wave noise;

[0054] S3: Establish a deep neural network with UNET structure and introduce a structural similarity regularization function into the loss function of the deep neural network to enhance the network's ability to identify valid signals;

[0055] like Figure 2 As shown, in this embodiment, the deep neural network includes an encoding part and a decoding part, the encoding part is used for data downsampling, and the decoding part is used for data upsampling;

[0056] The encoding part includes 5 groups of basic units. Each group of basic units includes 2 convolutional layers and 1 maximum pooling layer. The convolution kernel size of the convolutional layer is 3x3. The number of channels of the convolutional layer in each group of basic units is twice the number of channels of the previous group of convolutional layers. The number of channels of the convolutional layer increases from the original 32 to 512.

[0057] The decoding part includes 4 groups of basic units, each of which includes 1 deconvolution layer and 2 convolution layers. The convolution kernel size of the deconvolution layer is 2x2, and the convolution kernel size of the convolution layer is 3x3.

[0058] Preferably, the decoding part uses a replication method to perform data fusion with the down-sampled data with symmetry of the UNET structure in each up-sampling process, so as to achieve data information enhancement in the up-sampling process.

[0059] Among them, the structural similarity regularization function is:

[0060]

[0061] Among them, ssim(Y label ,Y pred ) represents the label data Y label And the input prediction data Y pred The structural similarity of Y represents the mean of data Y, and σ Y represents the standard deviation of data Y, and Represents the costandard deviation of two data, where C1 and C2 are constants.

[0062] The loss function of a deep neural network is:

[0063]

[0064] Among them, Q represents the loss function of the deep neural network, α and β represent weight factors, Nwin represents the number of sub-windows used to calculate the similarity of two data structures in order to improve the accuracy of similarity calculation; α||Y label -Y pred || 2 is the bi-norm of the label data and the predicted data, which is used to measure the energy similarity of the two data. It is a structural similarity function, which is used to measure the structural similarity of two data.

[0065] Preferably, when applying the structural similarity regularization function, the input label data and prediction data are divided into two-dimensional time windows, calculated in each local time window, and finally the mean is taken as the similarity measure of the label data and the prediction data, where the time window size used is 11x11 and the time window step size is 2x2.

[0066] S4: setting network training hyperparameters, and using the training data set to train the deep neural network. The trained deep neural network can suppress the surface wave of the input actual pre-stack common shot gather data and output the corresponding effective wave shot gather data.

[0067] Preferably, this step uses the Adam iterative optimizer to train the deep neural network.

[0068] S5: Use the test data set to verify the training effect of the deep neural network.

[0069] Example 2

[0070] This embodiment proposes an intelligent surface wave suppression method based on structural constraints. This method is aimed at the surface wave suppression needs of pre-stack seismic data. A deep neural network with structural similarity regularization constraints is established based on Tensorflow to form a computationally efficient intelligent surface wave suppression processing flow, and the processing effect of the algorithm is verified through test data.

[0071] This method introduces a structural similarity regularization operator based on the separate L1 or L2 norm of a conventional end-to-end deep neural network to improve the accuracy of the network in identifying effective signals.

[0072] The network model structure designed by this method is as follows Figure 2As shown in the figure, the network input data is the shot data before the surface wave is removed, and the network output is the corresponding effective wave shot data. The network is a UNET structure, which consists of two parts: encoding and decoding. In the encoding stage, data is downsampled. Its basic unit is 2 convolutional layers (convolution kernel size 3x3) and 1 maximum pooling layer, with a total of 5 groups. The number of convolutional layer channels increases from the initial 32 to 512, and the number of channels in each group of convolutional layers is twice that of the previous group. Data upsampling is achieved in the decoding stage. The basic structural unit is 1 deconvolution layer (convolution kernel size 2x2) and 2 convolution layers (convolution kernel size 3x3). In addition, in each upsampling process, the replication method is used to fuse the downsampled data symmetrically with the UNET structure to enhance data information during the upsampling process.

[0073] In order to improve the accuracy of the network in identifying effective signals, the loss function used in this method is:

[0074]

[0075] Among them, the first term of the loss function is the bi-norm of the label and the predicted data, which measures the energy similarity of the two data; the second term is the structural similarity function, which measures the structural similarity of the two data. When the structural similarity regularization function is actually applied, it is necessary to divide the input label and predicted data into two-dimensional time windows, calculate the average value in each local time window and take it as the similarity measure of the two data. The time window size used in this method is 11x11, and the time window step size is 2x2.

[0076] In addition, Y label ,Y pred are label data and predicted data respectively. α and β represent weight factors.

[0077] The structural similarity regularization function can be specifically expressed as:

[0078]

[0079] Among them, μ Y represents the mean of data Y, and σ Y represents the standard deviation of data Y, and represents the costandard deviation of two data, C 1 and C 2 is a constant.

[0080] Based on the deep neural network structure and loss function established above, the Adam iterative optimizer is used for network training.

[0081] The basic process of the method in this embodiment is:

[0082] (1) Data screening and preparation: The present invention is mainly aimed at surface wave suppression of pre-stack common shot gather data. The main data source is actual data. De-noising effect evaluation is performed on the actual data to select high-quality denoised data as sample labels.

[0083] (2) Sample label set preparation: Data regularization is performed on the screened and prepared data. By counting the maximum number of channels NTR and the number of sampling points NS of the single shot set arrangement of the input data, all data are regularized according to the size of NTR*NS, thereby generating sample label data pairs with the shape and size of (NTR, NS);

[0084] (3) Network training: setting network training hyperparameters and feeding the prepared training data set into the network for training, ultimately obtaining a high-precision intelligent surface roll suppression network;

[0085] (4) Network reasoning: Reload the saved network, perform surface wave suppression on the test data, and verify the effect of the trained network.

[0086] Example 3

[0087] By building the above UNET deep neural network, the learning rate is set to 0.0005, the batch_size is set to 8, and the number of learning rounds is set to 50. In addition, the cluster GPU used in this test is Tesla P100 with a memory size of 16G.

[0088] Based on the above parameters, the network is trained and the trained model is saved. During the test, the data size is 140*2501, and the input pre-processed data is as follows Figure 3a As shown in , the results obtained by using the same sample labels and training hyperparameters and only changing the loss function to a loss function containing only the two norm are as follows Figure 3b As shown, the data after compression by the method of the present invention is as follows Figure 3c As shown in the figure, it can be seen that under the same parameter conditions, the method of the present invention has a better effect on surface roll suppression and removes noise more cleanly.

[0089] Example 4

[0090] This embodiment provides an intelligent surface wave suppression device based on structural constraints, including:

[0091] The data screening module is used to perform surface wave denoising on the actual pre-stack common shot gather data and screen out the data for making sample labels based on the denoising effect evaluation;

[0092] The sample label preparation module is used to perform regularization processing on the screened pre-stack common shot data and its corresponding denoised data to generate sample label data pairs with the same shape and size to form a training data set, where the sample is the pre-stack common shot data and the label is the corresponding effective shot data after removing the surface wave noise;

[0093] A network building module is used to build a deep neural network with a UNET structure and introduce a structural similarity regularization function into the loss function of the deep neural network to enhance the ability of the network to identify valid signals;

[0094] A training module is used to set network training hyperparameters, and train the deep neural network using the training data set, so that the trained deep neural network can suppress surface waves on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data;

[0095] The test module is used to verify the training effect of the deep neural network using the test data set.

[0096] Example 5

[0097] This embodiment provides an electronic device, the electronic device comprising:

[0098] at least one processor; and,

[0099] a memory communicatively connected to the at least one processor; wherein,

[0100] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the intelligent surface roll suppression method based on structural constraints described in any of the above embodiments.

[0101] The electronic device according to an embodiment of the present disclosure includes a memory and a processor, and the memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may, for example, include a read-only memory (ROM), a hard disk, a flash memory, etc.

[0102] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory.

[0103] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0104] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0105] Example 6

[0106] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the intelligent surface roll suppression method based on structural constraints described in any of the above embodiments.

[0107] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of each embodiment of the present disclosure are executed.

[0108] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0109] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. An intelligent surface roll suppression method based on structural constraints, It is characterized in that include: Perform surface wave denoising on the actual pre-stack common shot gather data, and select the data for making sample labels based on the denoising effect evaluation; The screened pre-stack common shot data and their corresponding denoised data are processed in a regularized manner to generate sample label data pairs with the same shape and size to form a training data set, where the samples are pre-stack common shot data and the labels are the corresponding valid shot data after removing surface wave noise. Establishing a deep neural network with a UNET structure, and introducing a structural similarity regularization function into the loss function of the deep neural network to enhance the ability of the network to identify valid signals; Setting network training hyperparameters, using the training data set to train the deep neural network, so that the trained deep neural network can suppress surface waves on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data; The test data set is used to verify the training effect of the deep neural network.

2. According to the intelligent surface wave suppression method based on structural constraints in claim 1, It is characterized in that The deep neural network includes an encoding part and a decoding part, the encoding part is used for data downsampling, and the decoding part is used for data upsampling; The encoding part includes 5 groups of basic units, each group of basic units includes 2 convolutional layers and 1 maximum pooling layer, the convolution kernel size of the convolutional layer is 3x3, the number of channels of the convolutional layer in each group of basic units is twice the number of channels of the previous group of convolutional layers, and the number of channels of the convolutional layer increases from the initial 32 to 512; The decoding part includes 4 groups of basic units, each basic unit includes 1 deconvolution layer and 2 convolution layers, the convolution kernel size of the deconvolution layer is 2x2, and the convolution kernel size of the convolution layer is 3x3.

3. The intelligent surface wave suppression method based on structural constraints according to claim 2, It is characterized in that The decoding part uses a replication method to perform data fusion with the down-sampled data with symmetry of the UNET structure in each up-sampling process, so as to enhance the data information in the up-sampling process.

4. The intelligent surface wave suppression method based on structural constraints according to claim 1, It is characterized in that The structural similarity regularization function is: Among them, ssim(Y label ,Y pred ) represents the label data Y label And the input prediction data Y pred The structural similarity of Y represents the mean of data Y, and σ Y represents the standard deviation of data Y, and Represents the costandard deviation of two data, where C1 and C2 are constants.

5. The intelligent surface wave suppression method based on structural constraints according to claim 4, It is characterized in that The loss function of the deep neural network is: Among them, Q represents the loss function used by the deep neural network, α and β represent weight factors, Nwin represents the number of sub-windows used to calculate the similarity of two data structures; α||Y label -Y pred || 2 is the bi-norm of the label data and the predicted data, which is used to measure the energy similarity of the two data. It is a structural similarity function, which is used to measure the structural similarity of two data.

6. The intelligent surface wave suppression method based on structural constraints according to claim 5, It is characterized in that When applying the structural similarity regularization function, the input label data and prediction data are divided into two-dimensional time windows, calculated in each local time window, and finally the mean is taken as the similarity measure of the label data and the prediction data, wherein the time window size used is 11x11 and the time window step size is 2x2.

7. The intelligent surface wave suppression method based on structural constraints according to claim 1, It is characterized in that The deep neural network is trained using the Adam iterative optimizer.

8. An electronic device, It is characterized in that The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the intelligent surface roll suppression method based on structural constraints as described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium, It is characterized in that The non-transitory computer-readable storage medium stores computer instructions, which are used to enable a computer to execute the intelligent surface roll suppression method based on structural constraints as described in any one of claims 1-7.

10. An intelligent surface wave suppression device based on structural constraints, It is characterized in that include: The data screening module is used to perform surface wave denoising on the actual pre-stack common shot gather data and screen out the data for making sample labels based on the denoising effect evaluation; The sample label preparation module is used to perform regularization processing on the screened pre-stack common shot data and its corresponding denoised data to generate sample label data pairs with the same shape and size to form a training data set, where the sample is the pre-stack common shot data and the label is the corresponding effective shot data after removing the surface wave noise; A network building module is used to build a deep neural network with a UNET structure and introduce a structural similarity regularization function into the loss function of the deep neural network to enhance the ability of the network to identify valid signals; A training module is used to set network training hyperparameters, and train the deep neural network using the training data set, so that the trained deep neural network can suppress surface waves on the input actual pre-stack common shot gather data and output corresponding effective wave shot gather data; The test module is used to verify the training effect of the deep neural network using the test data set.

Images