Seismic data reconstruction method based on VMD filtering and U-net image segmentation

Abstract

The application discloses a seismic data reconstruction method based on VMD filtering and U-net image segmentation, and relates to the technical field of oil exploration and development.The application decomposes seismic data through a VMD method, analyzes the difference between effective signal modal components and noise modal components in a time-frequency domain, optimizes the effective signal modal components to reconstruct seismic signals on the basis, finely suppresses local noise, and can effectively improve the signal-to-noise ratio of the seismic data, thereby providing high-quality training and test data for a subsequent U-net convolution network model.The application can effectively find out bad channels existing in seismic data and accurately reconstruct a missing part of the data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil exploration and development, in particular to the technical field of seismic data processing, and more particularly to a seismic data reconstruction method based on VMD filtering and U-net image segmentation. BACKGROUND

[0002] At present, due to the limitations of the acquisition environment such as surface topography and prohibited mining areas, the collected seismic data will inevitably have problems such as bad channels, missing channels, and insufficient sampling, which makes the seismic data poor in continuity and cannot meet the requirements of subsequent seismic interpretation work on data accuracy, so bad channel rejection and data reconstruction are the basic steps of seismic data processing.

[0003] Traditional seismic data interpolation methods can be divided into three categories: the first category is a data reconstruction method based on seismic wave field migration and de-migration, but the complexity of the stratum has a great influence on the data reconstruction effect; the second category is a seismic data reconstruction method based on a prediction filter, mainly suitable for regularly sampled seismic data; the third category is a data reconstruction method based on sparse transformation, which reconstructs seismic data using sparse constraints.

[0004] With the progress of computer technology, artificial intelligence machine learning technology has made great progress, Hinton et al. first demonstrated the superiority of deep neural networks in image processing, and subsequent scholars have successively proposed more efficient network models, and the convolutional neural network is one of the most widely used neural network models at present.

[0005] U-Net is a convolutional neural network proposed by Ronneberger et al. for medical image segmentation, which can extract multi-scale features of images and does not require manual design of feature extractors, automatically learns the features of data in a data-driven manner, and has high generalization and robustness. Essentially, seismic data also belong to images, and U-net convolutional neural network can be constructed and trained to realize pixel-level prediction of seismic images and high-precision seismic data reconstruction.

[0006] As the disclosure date is November 22, 2019, the publication number is CN110490219A, and the name is "a method for seismic data reconstruction based on texture constraint U-net network", the invention patent includes: S1, the training data set is segmented by using K-means algorithm, and the texture label of the training data is obtained; S2, the U-net network is trained as a texture extractor through the training data set and the texture label, and the optimized texture extractor parameters are obtained; S3, the trained texture extractor and the reconstruction network are connected in series, the optimized texture extractor parameters are adopted, the texture information of the reconstruction label and the reconstruction seismic data is extracted, and the texture loss is obtained; S4, the reconstruction network is optimized through the texture loss, and the seismic data reconstruction is carried out. The invention combines the characteristics of rich texture information of seismic data, and can improve the reconstruction accuracy and continuity of seismic events in the case of limited samples by strengthening the learning of texture.

[0007] The above prior art based on texture information for seismic data reconstruction cannot eliminate the bad channel existing in the seismic data, cannot finely suppress the local noise, and affects the accuracy of the U-net network model for seismic data reconstruction. SUMMARY

[0008] In order to overcome the defects and deficiencies in the above prior art, the present application provides a seismic data reconstruction method based on VMD filtering and U-net image segmentation. The present application decomposes the seismic data by VMD method, analyzes the differences of effective signal modal components and noise modal components in time-frequency domain, and optimizes the effective signal modal components to reconstruct the seismic signal. The local noise is finely suppressed, which can effectively improve the signal-to-noise ratio of the seismic data and provide high-quality training and test data for the subsequent U-net convolutional network model. The present application can effectively eliminate the bad channel existing in the seismic data and accurately reconstruct the missing data.

[0009] In order to solve the problems in the above prior art, the present application is realized by the following technical scheme.

[0010] The seismic data reconstruction method based on VMD filtering and U-net image segmentation includes the following steps:

[0011] S1, obtaining the post-stack seismic data with bad channels and missing channels;

[0012] S2, performing spectral analysis on the post-stack seismic data to be processed to determine the amplitude and frequency characteristics of the bad channel data;

[0013] S3, counting the root mean square amplitude energy of each channel of the seismic data, and comparing it with the set threshold value. If it is greater than the set threshold value, it is defined as a bad channel, and the bad channel is replaced with a blank channel, and the position of the bad channel is recorded;

[0014] S4, filtering the seismic data after removing the bad trace by using variational mode decomposition to improve the signal-to-noise ratio of the seismic data;

[0015] S5, selecting the filtered seismic data to construct a training sample set and a test sample set;

[0016] S6, constructing a U-net network structure model and initializing the parameters of each convolutional layer;

[0017] S7, training the U-net network model by using the training sample set constructed by the filtered seismic data, testing the model effect by using the test sample set constructed by the filtered seismic data, and making the model parameters converge through multiple iterations;

[0018] S8, taking the filtered seismic data obtained after the steps S1 to S4 as input and inputting it into the U-net network model trained in the step S7 to reconstruct the stacked seismic data.

[0019] Further, in the step S2, the spectrum of the stacked seismic data to be processed is analyzed, specifically, the original seismic data set without bad trace and the same seismic data set with bad trace obtained in the step S1 are respectively analyzed to determine the frequency and amplitude characteristics of the bad trace.

[0020] Further, in the step S3, the root mean square amplitude energy of each trace in the seismic data is calculated, and the calculation formula is:

[0021] wherein, is the i-th sampling point of a single seismic trace, and N is the total number of sampling points of a single seismic trace.

[0022] Further, in the step S3, according to the spectrum analysis result in the step S2, a set threshold of the bad trace is defined, and when the root mean square energy value of a trace is greater than the set threshold, the trace is determined as a bad trace, the bad trace is replaced by a blank trace, and the position of the bad trace is recorded.

[0023] In the step S3, the set threshold of the bad trace is defined as 1x10 20 .

[0024] Further, in the step S4, the seismic signal is decomposed into k mode components with limited bandwidth by using variational mode decomposition, and each mode component and its center frequency are calculated by the following formula:

[0025] ,

[0026] wherein, is the seismic signal, is the mode component, is a Lagrange multiplier, , , are their corresponding Fourier transforms, respectively, and a is a quadratic multiplication factor, denotes the center frequency of the decomposed seismic signal component, denotes the center frequency of the kth decomposed seismic signal component.

[0027] Further, in the S4 step, the number of decomposition modal components of the variational modal decomposition is determined by the mean of the instantaneous frequencies of all modal components.

[0028] Specifically, the seismic data is decomposed, the mean of the instantaneous frequencies of all modal components is calculated, and the curvature of the mean instantaneous frequency curve is calculated. If the curvature is greater than a set curvature threshold, k-1 is the number of optimal modal components, otherwise k is increased and the decomposition is repeated until the curvature is greater than the set curvature threshold.

[0029] The set curvature threshold is 0.5.

[0030] Further, in the S4 step, after determining the optimal k value, the seismic trace is decomposed, and the signal-to-noise ratio of all modal components is calculated If , it is an effective modal, otherwise it is a noise modal.

[0031] Further, in the S4 step, according to the bad trace position recorded in the S3 step, the corresponding trace of the seismic data filtered by the variational modal decomposition is replaced with a blank trace.

[0032] Further, in the S5 step, for the filtered seismic data obtained in the S4 step, a regular region not containing a blank trace is selected to construct a seismic sub-volume, and 33% of blank traces are randomly inserted into the seismic sub-volume to obtain data of artificially missing seismic traces and corresponding complete seismic data.

[0033] Further, in the S5 step, the seismic sub-volume data set with randomly inserted blank traces is divided into a training set and a test set of a U-net convolutional neural network according to a ratio of 3:1.

[0034] Further, in the S6 step, a U-net convolutional neural network model is constructed using a Keras deep learning framework.

[0035] The constructed U-net convolutional neural network model includes two parts of encoding and decoding, corresponding to the up-sampling and down-sampling processes respectively. Each down-sampling operation includes two convolution layers with a convolution kernel size of 3*3 and a maximum pooling layer with a convolution kernel size of 2*2, and each up-sampling operation includes an up-sampling layer with a convolution kernel of 2*2 and two convolution layers with a convolution kernel size of 3*3.

[0036] Further, in the S7 step, the training seismic data set and the test seismic data set input into the U-net network model include Inline line direction amplitude slices, CrossLine line direction amplitude slices and time amplitude slices.

[0037] Compared with the prior art, the beneficial technical effects brought by the present application are as follows:

[0038] 1. The VMD method is used for decomposing the seismic data, analyzing the differences of the effective signal modal components and the noise modal components in the time-frequency domain, and optimizing the effective signal modal components to reconstruct the seismic signal on this basis, so that the local noise can be finely suppressed, the signal-to-noise ratio of the seismic data can be effectively improved, and the subsequent U-net convolutional network model can provide high-quality training and test data.

[0039] 2. The U-net convolutional neural network is used to construct a deep learning model, the advantages of seismic data redundancy are fully utilized, the multi-dimensional features of the seismic data are extracted and fused through dense convolutional layers and pooling layers, the data features are automatically learned in a data-driven manner, the pixel-level prediction of the seismic image is realized, and then the high-precision reconstruction of the seismic data is completed. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used for the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0041] Figure 1 The flowchart of the method of the present application;

[0042] Figure 2 The spectrum analysis diagram of the original non-bad channel seismic data set in the embodiment of the present application;

[0043] Figure 3 The spectrum analysis diagram of the same seismic data set with bad channels in the embodiment of the present application;

[0044] Figure 4 The effect diagram of the original seismic data with bad channels in the embodiment of the present application;

[0045] Figure 5 Effect diagram of bad trace removal for an embodiment of the present application;

[0046] Figure 6 Modal component number k versus average instantaneous frequency relationship diagram for VMD decomposition of an embodiment of the present application;

[0047] Figure 7 Seismic signal signal-to-noise ratio diagram before and after VMD filtering for an embodiment of the present application;

[0048] Figure 8 Inline direction slice in a training data set and a test data set instance for an embodiment of the present application;

[0049] Figure 9 Crossline direction slice in a training data set and a test data set instance for an embodiment of the present application;

[0050] Figure 10 Time slice in a training data set and a test data set instance for an embodiment of the present application;

[0051] Figure 11 U-net convolutional neural network structure diagram for an embodiment of the present application;

[0052] Figure 12 U-net model training error for an embodiment of the present application;

[0053] Figure 13 Original seismic profile containing bad traces for an embodiment of the present application;

[0054] Figure 14 Effect comparison seismic profile after bad trace removal and data reconstruction for an embodiment of the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0056] Embodiment 1

[0057] As a preferred embodiment of the present application, with reference to the drawings in the specification of the present application, the embodiment discloses a seismic data reconstruction method based on VMD filtering and U-net image segmentation, which comprises the following steps: Figure 1

[0058] S1, obtaining post-stack seismic data containing bad traces and missing traces;

[0059] ​S2, performing spectrum analysis on the post-stack seismic data to be processed to determine the amplitude and frequency characteristics of the bad trace data;

[0060] S3, counting the root mean square amplitude energy of each trace of the seismic data and comparing it with a set threshold value, if it is greater than the set threshold value, it is defined as a bad trace, and the bad trace is replaced with a blank trace, and the position of the bad trace is recorded;

[0061] S4, filtering the seismic data after removing the bad trace by using variational mode decomposition to improve the signal-to-noise ratio of the seismic data;

[0062] S5, selecting the filtered seismic data to construct a training sample set and a test sample set;

[0063] S6, constructing a U-net network structure model and initializing the parameters of each convolutional layer;

[0064] S7, training the U-net network model using the training sample set constructed from the filtered seismic data, testing the model effect using the test sample set constructed from the filtered seismic data, and converging the model parameters through multiple iterations;

[0065] S8, taking the filtered seismic data obtained after the steps S1 to S4 as input, inputting it into the U-net network model trained in the step S7, and reconstructing the post-stack seismic data.

[0066] Embodiment 2

[0067] As another preferred embodiment of the present application, with reference to the accompanying drawings Figure 1 , the accompanying drawings Figure 2 , the accompanying drawings Figure 3 , the accompanying drawings Figure 4 and the accompanying drawings Figure 5 , the present embodiment discloses a seismic data reconstruction method based on VMD filtering and U-net image segmentation, which comprises the following steps:

[0068] S1, obtaining post-stack seismic data with bad traces and missing traces;

[0069] S2, performing spectrum analysis on the post-stack seismic data to be processed to determine the amplitude and frequency characteristics of the bad trace data;

[0070] Among them, the spectrum analysis on the post-stack seismic data to be processed is as shown in the accompanying drawings Figure 2 and the accompanying drawings Figure 3 , specifically, the spectrum analysis is performed on the original bad trace-free seismic data set and the same seismic data set with bad traces obtained in the step S1 to determine the frequency and amplitude characteristics of the bad traces.

[0071] The original traceless seismic data set represents a normal part of the post-stack seismic data, and the same seismic data set with bad traces is a bad trace in the post-stack seismic data.

[0072] As shown in Figs. 1 and 2, the original traceless seismic data set represents a normal part of the post-stack seismic data, and the same seismic data set with bad traces is a bad trace in the post-stack seismic data. Figure 2 And as shown in Figs. 3 and 4, the original traceless seismic data set represents a normal part of the post-stack seismic data, and the same seismic data set with bad traces is a bad trace in the post-stack seismic data. Figure 3 It can be seen that the amplitude energy of the bad trace can reach 10 37 , far more than the effective signal, and the frequency domain of the bad trace signal overlaps with the effective signal, which cannot be effectively distinguished.

[0073] S3, the root mean square amplitude energy of each trace in the seismic data is counted and compared with a set threshold value, if greater than the set threshold value, the bad trace is defined, and the bad trace is replaced with a blank trace, and the position of the bad trace is recorded;

[0074] The root mean square amplitude energy of each trace in the seismic data is counted, and the calculation formula is:

[0075] , wherein, is the i-th sampling point of a single seismic trace, and N is the total number of sampling points of a single seismic trace;

[0076] At the same time, according to the spectrum analysis result in S2, the set threshold value of the bad trace is defined as 1x10 20 When the root mean square energy value of a trace is greater than the set threshold value, the trace is determined as a bad trace, and the bad trace is replaced with a blank trace, and the position of the bad trace is recorded; as shown in Figs. 5 and 6, the effect comparison chart before and after removing the bad trace is shown. Figure 4 And as shown in Figs. 7 and 8, the effect comparison chart before and after removing the bad trace is shown. Figure 5

[0077] S4, the filtered seismic data after removing the bad trace is filtered by using the variational mode decomposition to improve the signal-to-noise ratio of the seismic data;

[0078] S5, the training sample set and the test sample set are constructed by selecting the filtered seismic data;

[0079] S6, the U-net network structure model is constructed, and the parameters of each convolution layer are initialized;

[0080] S7, the U-net network model is trained by using the training sample set constructed by the filtered seismic data, the model effect is tested by using the test sample set constructed by the filtered seismic data, and the model parameters are converged through multiple iterations;

[0081] S8, the filtered seismic data obtained after the steps S1 to S4 are taken as input, and input into the U-net network model trained in S7, and the post-stack seismic data is reconstructed.

[0082] Embodiment 3

[0083] ​As another preferred embodiment of the present invention, this embodiment is an elaboration of the specific implementation of step S4 in the above-mentioned embodiments 1 and 2.

[0084] As an implementation of this embodiment, in step S4, variational mode decomposition is used to filter the seismic data after bad tracks are removed to improve the signal-to-noise ratio of the seismic data.

[0085] As an implementation method of this embodiment, VMD, or variational mode decomposition, is a nonlinear, non-stationary signal time-frequency analysis method that can adaptively decompose seismic signals into k finite-bandwidth modal components based on the signal's time-scale characteristics, and adaptively determine the center frequency and bandwidth information of each modal component. The modal components and their center frequencies are obtained using the following formula:

[0086] ,

[0087] Where, is the earthquake signal, is the modal component, is the Lagrange multiplication operator, 、 、 are their corresponding Fourier transforms, α is the quadratic multiplication factor, represents the center frequency of the decomposed seismic signal component, Represents the center frequency of the kth seismic signal component obtained by decomposition.

[0088] As another implementation of this embodiment, VMD decomposition is similar to a Togan bandpass filter, where the k value controls the number and bandwidth of the filter and is determined by the mean of the instantaneous frequencies of the components, such as Figure 6 As shown in the figure, when the k value increases to a certain number, the curve shows a clear subtotal phenomenon. This critical number is the optimal k value for VMD decomposition. Initially set k = 2, decompose the seismic data, calculate the instantaneous frequency mean of all modal components, and simultaneously calculate the curvature of the instantaneous frequency mean curve. If the curvature exceeds the set curvature threshold, k-1 is considered the optimal number of modes. Otherwise, increase k and repeat the decomposition until the curvature exceeds the set curvature threshold. This curvature threshold is generally set to 0.5.

[0089] As another implementation of this embodiment, the effective mode determination criterion is based on the signal-to-noise ratio of each mode. Is it greater than zero? , it means that the effective signal energy in this mode is stronger than the noise, which is an effective mode, otherwise it is a noise mode. Finally, all effective modes are superimposed to reconstruct the seismic signal. At the same time, the reconstructed signal needs to replace the corresponding position with a blank track according to the bad track position located in step S3, as shown in the attached figure. Figure 7The figure shows the comparison of the signal-to-noise ratio of the seismic signal before and after VMD.

[0090] Embodiment 4

[0091] As another preferred embodiment of the present application, the embodiment is a specific implementation of the S5 step in the above-mentioned embodiments 1 to 3.

[0092] In the embodiment, S5, the filtered seismic data is selected to construct a training sample set and a test sample set.

[0093] As an implementation of the embodiment, in the S5 step, for the filtered seismic data obtained in the S4 step, a regular region excluding blank channels is selected to construct a seismic sub-volume, and 33% of blank channels are randomly inserted into the seismic sub-volume to obtain data of artificially missing seismic channels and corresponding complete seismic data. The data augmentation is realized by 3D stereoscopic time window sliding sampling, and the window size is 16x16x32, i.e., 16 channels in the Inline direction, 16 channels in the Crossline direction, and 32 ms in the time direction. The window is overlapped by 50% in the Inline, Crossline, and time directions, and a total of 83664 seismic sub-volume sets are constructed, as shown in FIG. 1, FIG. 2, and FIG. 3. Figure 8 Figure 9 Figure 10 As shown in FIG. 4, FIG. 5, and FIG. 6, the seismic sub-volume data set is divided into a training set for training the U-net network and a test set according to a ratio of 3:1.

[0094] Embodiment 5

[0095] As another preferred embodiment of the present application, the embodiment is a specific implementation of the S6 step in the above-mentioned embodiments 1 to 4.

[0096] In the embodiment, a U-net network structure model is constructed, and the parameters of each convolution layer are initialized.

[0097] As an implementation of the embodiment, a Keras deep learning framework is selected to construct a U-net convolutional neural network model. The model structure includes two parts of an encoding process and a decoding process.

[0098] The encoding process includes:

[0099] S61, the input seismic image first passes through 2 times (Conv+ReLU) combination to obtain a 64-channel seismic feature map;

[0100] S62, the seismic feature map passes through 4 times [Maxpool+(Conv+ReLU) x 2] combination again to obtain a new 1024-dimensional seismic feature map.

[0101] The decoding process includes:​​

[0102] S63, S62 the 1024-dimensional seismic feature atlas obtained is up-sampled to a 64-dimensional seismic image feature atlas of the original seismic image size through four times of [Up-Conv + (Conv + ReLU) x 2] combination, while the features generated by the encoder and the decoder are spliced and fused by using the skip connection between them;

[0103] S64, the 64-dimensional seismic feature atlas after splicing and fusion is finally generated into a segmentation image through one time of Conv; wherein, Conv represents a convolution layer, ReLU represents an activation layer, Maxpool represents a maximum pooling layer, and Up-Conv represents a transpose convolution layer, as shown in the accompanying Figure 11 .

[0104] S7, the U-net network model is trained by using the training sample set constructed by the seismic data sub-volume, the model parameters are adjusted to optimize the model with the highest precision, the model precision is tested by using the test sample set seismic data, if the precision is higher than the set threshold, the model is used as the final U-Net network model, otherwise the parameters are reselected. The training error of the U-net network model is shown in the accompanying Figure 12 .

[0105] S8, the model inputs the filtered seismic data, and the stacked seismic data is reconstructed by using the trained U-net network model. The input data of the trained U-net network model in this operation is the filtered seismic data obtained in S4. The effect comparison diagram of the seismic data reconstructed by the final U-Net network model is shown in the accompanying Figure 13 and Figure 14 .

[0106] Those skilled in the art will appreciate that the embodiments described herein are intended to help the reader understand the implementation method of the present application and should be understood as the protection scope of the present application is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the protection scope of the present application.

Claims

1. A seismic data reconstruction method based on VMD filtering and U-net image segmentation, characterized in that, The method comprises the following steps: S1, obtaining post-stack seismic data with bad channels and missing channels; S2, performing spectral analysis on the post-stack seismic data to be processed to determine the amplitude and frequency characteristics of the bad channel data; S3, calculating the root mean square amplitude energy of each channel of the seismic data and comparing it with a set threshold value, if it is greater than the set threshold value, it is defined as a bad channel, and the bad channel is replaced with a blank channel, and the position of the bad channel is recorded; S4, filtering the seismic data after removing the bad channels using variational mode decomposition to improve the signal-to-noise ratio of the seismic data; S5, selecting the filtered seismic data to construct a training sample set and a test sample set; S6, constructing a U-net network structure model and initializing the parameters of each convolutional layer; S7, training the U-net network model using the training sample set constructed from the filtered seismic data, testing the model effect using the test sample set constructed from the filtered seismic data, and converging the model parameters through multiple iterations; S8, inputting the filtered seismic data obtained after the steps S1 to S4 into the U-net network model trained in the step S7 to reconstruct the post-stack seismic data.

2. The method of claim 1, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction method is characterized by: In the step S2, the spectral analysis on the post-stack seismic data to be processed is specifically performed on the original seismic data set without bad channels and the same seismic data set with bad channels obtained in the step S1 to determine the frequency and amplitude characteristics of the bad channels.

3. The method of claim 1 or 2, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction method is characterized by: In the step S3, the root mean square amplitude energy of each channel of the seismic data is calculated, and the calculation formula is: wherein, is the ith sample point of a single seismic trace, and N is the total number of sample points of a single seismic trace.

4. The method of claim 3, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction method is characterized by: In the step S3, according to the spectral analysis result in the step S2, a set threshold value of the bad channel is defined, and when the root mean square energy value of a channel is greater than the set threshold value, the channel is determined as a bad channel, and the bad channel is replaced with a blank channel, and the position of the bad channel is recorded.

5. The method of claim 4, wherein: The set threshold value defining a bad track is 1 x 10 20 .

6. The method of claim 1 or 2, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction method is characterized by: In the step S4, the seismic signal is decomposed into k limited bandwidth mode components by variational mode decomposition, and each mode component and its center frequency are calculated by the following formula: ， wherein, is a seismic signal, is a modal component, is a Lagrange multiplier, , , are their corresponding Fourier transforms, respectively, and a is a quadratic multiplication factor, denotes the center frequency of the decomposed seismic signal component, denotes the center frequency of the k-th decomposed seismic signal component.

7. The method of claim 6, wherein: In the step S4, the number of decomposition mode components of the variational mode decomposition is determined by the average instantaneous frequency of all mode components.

8. The method of claim 7, wherein: The seismic data is decomposed, the average instantaneous frequency of all mode components is calculated, and the curvature of the average instantaneous frequency curve is calculated, if the curvature is greater than a set curvature threshold value, k-1 is the optimal number of mode components, otherwise k is increased for repeated decomposition until the curvature is greater than the set curvature threshold value.

9. The method of claim 8, wherein: The set curvature threshold value is 0.

5.

10. The method of claim 8, wherein: In the S4 step, after determining the optimal k value, the seismic trace is decomposed, and the signal-to-noise ratio of all modal components is calculated ; if , it is an effective modal, otherwise it is a noise modal.

11. The method of claim 1, wherein: In the step S4, according to the bad channel position recorded in the step S3, the corresponding channel of the seismic data filtered by the variational mode decomposition is replaced with a blank channel.

12. The method of claim 1 or 11, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction is characterized by: In the step S5, a regular region excluding the blank channel is selected from the filtered seismic data obtained in the step S4 to construct a seismic sub-volume, and 33% of blank channels are randomly inserted into the seismic sub-volume to obtain artificial missing seismic channel data and corresponding complete seismic data.

13. The method of claim 12, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction is characterized by: In the step S5, the seismic sub-volume data set with randomly inserted blank channels is divided into a training set and a test set of the U-net convolutional neural network according to a ratio of 3:

1.

14. The method of claim 1, wherein: In the S6 step, a U-net convolutional neural network model is built by using a Keras deep learning framework.

15. The method of claim 14, wherein the VMD filtering and U-net image segmentation based seismic data reconstruction is characterized by: The built U-net convolutional neural network model comprises an encoding part and a decoding part, which correspond to an up-sampling process and a down-sampling process respectively; each down-sampling operation comprises two convolution layers with a convolution kernel size of 3*3 and a maximum pooling layer with a convolution kernel size of 2*2; and each up-sampling operation comprises an up-sampling layer with a convolution kernel of 2*2 and two convolution layers with a convolution kernel size of 3*3.

16. The method of claim 1, wherein: In the S7 step, the training seismic data set and the test seismic data set input into the U-net network model comprise Inline line direction amplitude slices, CrossLine line direction amplitude slices and time amplitude slices.

Images