Seismic data high-precision imaging method and device

Abstract

The invention provides a seismic data high-precision imaging method and device, and relates to the technical field of seismic exploration, and the method comprises the steps: obtaining a seismic pre-stack gather of a to-be-imaged work area, carrying out the multidirectional pre-stack migration processing of the seismic pre-stack gather at a fixed migration aperture for each imaging line in an imaging space, and obtaining a seismic pre-stack gather; and obtaining a multi-azimuth migration imaging section set of each imaging line, and processing the multi-azimuth migration imaging section set of the target imaging line by using the target neural network model to obtain a high-precision migration imaging section of the target imaging line. A target neural network model is a lightweight neural network formed based on Transform and CNN, and a training target is that an error between an output migration imaging section prediction result and an actual high-precision migration imaging section is smaller than a preset threshold value, so that the output migration imaging section can be ensured to be a high-precision imaging result, and the accuracy of the migration imaging section prediction result is improved. The technical problem that an existing seismic data imaging method is low in imaging precision is effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration technology, and in particular to a method and device for high-precision imaging of seismic data. Background Art

[0002] In seismic exploration, geophysicists have always been striving to obtain information about complex underground structures by performing migration processing on the seismic observation data on the surface. This process is generally called seismic data imaging. For most migration processing methods, the selection of migration aperture is an important task. A smaller migration aperture can reduce the amount of migration calculation, but there is a risk of not being able to correctly image steep-angle structures; an overly large migration aperture brings about migration noise and a larger amount of migration calculation. A better migration aperture can suppress migration noise and improve the computational efficiency of migration processing.

[0003] Due to the limitation of the implementation method of the migration algorithm, a uniform migration aperture is usually always selected in a migration operation. However, because the dip of the underground structure changes with the spatial position, the appropriate migration aperture for a certain imaging point may be too large or too small for other imaging points. In addition, since it is difficult to have an accurate estimate of the dip of the simulated imaging structure before the migration process, the selection of the migration aperture can only adopt a conservative method, that is, to select a larger aperture. This results in the failure to suppress the migration noise well, which in turn leads to unclear imaging sections.

[0004] In summary, the existing seismic data imaging methods have the technical problem of low imaging accuracy. Summary of the invention

[0005] The object of the present invention is to provide a method and device for high-precision imaging of seismic data, so as to alleviate the technical problem of low imaging precision existing in the existing seismic data imaging methods.

[0006] In a first aspect, the present invention provides a method for high-precision imaging of seismic data, comprising: obtaining a seismic pre-stack gather of a work area to be imaged; for each imaging line in the imaging space of the work area to be imaged, performing multi-azimuth pre-stack migration processing on the seismic pre-stack gather with a fixed offset aperture to obtain a multi-azimuth offset imaging profile set for each imaging line; using a target neural network model to process the multi-azimuth offset imaging profile set of the target imaging line to obtain a high-precision offset imaging profile of the target imaging line; wherein the target neural network model is a lightweight neural network based on a Transformer neural network and a CNN neural network, and the Transformer neural network is located between an encoding module and a decoding module of the CNN neural network; a training target of the target neural network model is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold; the target imaging line represents any imaging line in the imaging space; based on the high-precision offset imaging profiles of all imaging lines, a high-precision imaging result of the work area to be imaged is determined.

[0007] Optionally, it also includes: obtaining seismic pre-stack gathers of a training area; determining high-precision offset imaging sections and corresponding multi-azimuth offset imaging section sets of multiple typical imaging lines based on the seismic pre-stack gathers of the training area; wherein the high-precision offset imaging section is an imaging section obtained after offset processing using the Fresnel zone of the imaging point as the offset aperture; using the multi-azimuth offset imaging section set of each typical imaging line as input data of the initial neural network model, and using the corresponding high-precision offset imaging section as a training label, training the initial neural network model until the training goal is achieved, thereby obtaining a target neural network model.

[0008] Optionally, a high-precision offset imaging profile of multiple typical imaging lines is determined based on the seismic pre-stack gathers of the training work area, including: calculating a multi-azimuth dip domain offset gather of a target typical imaging line based on the seismic pre-stack gathers of the training work area; wherein the target typical imaging line represents any one of the multiple typical imaging lines; based on the multi-azimuth dip domain offset gathers, human-computer interaction is used to pick up Fresnel zones of multiple typical imaging points; using the Fresnel zones of multiple typical imaging points as the offset aperture, pre-stack migration processing is performed on the seismic pre-stack gathers of the training work area, and a high-precision offset imaging profile of the target typical imaging line is obtained.

[0009] Optionally, the initial neural network model also includes: a normalization layer and a denormalization layer; taking the multi-azimuth offset imaging section set of each typical imaging line as the input data of the initial neural network model, and taking the corresponding high-precision offset imaging section as the training label, the initial neural network model is trained, including: the normalization layer normalizes the multi-azimuth offset imaging section set of the target typical imaging line and the corresponding high-precision offset imaging section, respectively, to obtain the amplitude-normalized multi-azimuth offset imaging section set and the amplitude-normalized high-precision offset imaging section; the encoding module encodes each local time window data in the amplitude-normalized multi-azimuth offset imaging section set to extract the local features and section position information of each local time window data; all local time window data are encoded. The local features of the data are expanded to construct a long sequence of local features; the Transformer neural network extracts the global features and effective features of the long sequence of local features to obtain the target feature sequence; the target feature sequence is feature folded, and the decoding module decodes the folded features and profile position information to obtain the reconstructed offset imaging section; based on the reconstructed offset imaging section and the high-precision offset imaging section after amplitude normalization, the loss function value is calculated to train the initial neural network model based on the loss function value; the denormalization layer denormalizes the reconstructed offset imaging section based on the multi-azimuth offset imaging section set of the target typical imaging line and the corresponding high-precision offset imaging section to obtain the high-precision offset imaging section after amplitude restoration.

[0010] Optionally, the normalization layer performs normalization processing on the multi-azimuth offset imaging section set and the corresponding high-precision offset imaging section of the target typical imaging line, respectively, including: determining the amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line to obtain an amplitude value set corresponding to each offset imaging section; superimposing the amplitude values ​​of the same sequence number in all amplitude value sets, and taking the absolute value after superposition to obtain a first target amplitude value set; using the maximum amplitude value in the first target amplitude value set as the normalization parameter of the multi-azimuth offset imaging section set to normalize the multi-azimuth offset imaging section set; determining the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, and taking the absolute value to obtain a second target amplitude value set; using the maximum amplitude value in the second target amplitude value set as the normalization parameter of the high-precision offset imaging section to normalize the high-precision offset imaging section.

[0011] Optionally, the denormalization layer performs denormalization processing on the reconstructed offset imaging section based on the multi-azimuth offset imaging section set of the target typical imaging line and the corresponding high-precision offset imaging section, including: determining the amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line, and determining the energy average value of the multi-azimuth offset imaging section set based on the amplitude value to obtain a first energy; determining the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, and determining the energy average value of the high-precision offset imaging section based on the amplitude value to obtain a second energy, and determining the maximum amplitude value in the high-precision offset imaging section; determining the denormalization parameters of the reconstructed offset imaging section based on the first energy, the second energy and the maximum amplitude value; and performing denormalization processing on the reconstructed offset imaging section based on the denormalization parameters to obtain the high-precision offset imaging section after amplitude restoration.

[0012] Optionally, the selection criteria of the typical imaging lines are: in a location with drastic structural changes, the imaging lines are selected based on a first interval; in a location with a gentle structural change, the imaging lines are selected based on a second interval, and the first interval is smaller than the second interval.

[0013] In a second aspect, the present invention provides a high-precision imaging device for seismic data, including a first acquisition module for acquiring seismic pre-stack gathers of a work area to be imaged; an offset module for performing multi-azimuth pre-stack migration processing on the seismic pre-stack gathers with a fixed offset aperture for each imaging line in the imaging space of the work area to be imaged, so as to obtain a multi-azimuth offset imaging profile set for each imaging line; a processing module for processing the multi-azimuth offset imaging profile set of the target imaging line using a target neural network model, so as to obtain a high-precision offset imaging profile of the target imaging line; wherein the target neural network model is a lightweight neural network composed of a Transformer neural network and a CNN neural network, and the Transformer neural network is located between an encoding module and a decoding module of the CNN neural network; a training target of the target neural network model is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold; the target imaging line represents any imaging line in the imaging space; a first determination module is used to determine the high-precision imaging result of the work area to be imaged based on the high-precision offset imaging profiles of all imaging lines.

[0014] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and when the processor executes the computer program, the method for high-precision imaging of seismic data described in any one of the aforementioned embodiments is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method for high-precision imaging of seismic data described in any one of the aforementioned embodiments is implemented.

[0016] The present invention provides a method for high-precision imaging of seismic data. After obtaining the seismic pre-stack gathers of the work area to be imaged, firstly, for each imaging line in the imaging space, the seismic pre-stack gathers are subjected to multi-azimuth pre-stack migration processing with a fixed offset aperture to obtain a multi-azimuth offset imaging profile set for each imaging line, and then the multi-azimuth offset imaging profile set of the target imaging line is processed using a target neural network model to obtain a high-precision offset imaging profile of the target imaging line. The target neural network model is a lightweight neural network based on a Transformer neural network and a CNN neural network, and the training goal is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold, therefore, it can ensure that the output offset imaging profile is a high-precision imaging result, effectively alleviating the technical problem of low imaging accuracy in the existing seismic data imaging method. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a method for high-precision imaging of seismic data provided by an embodiment of the present invention; Figure 2 A schematic diagram of a data processing flow of an initial neural network model provided by an embodiment of the present invention; Figure 3 A schematic diagram of a process for converting output data of an encoding module to input data of a decoding module provided by an embodiment of the present invention; Figure 4 A schematic diagram of a multi-azimuth offset imaging section set provided by an embodiment of the present invention; Figure 5 A comparison diagram of imaging results of various seismic data imaging methods provided by embodiments of the present invention; Figure 6 A functional module diagram of a high-precision seismic data imaging device provided by an embodiment of the present invention; Figure 7 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0022] Embodiment 1 In the prior art, if a conservative method is used to select a larger offset aperture before seismic migration imaging, the offset noise cannot be suppressed well, resulting in unclear imaging profiles. Studies have shown that if the Fresnel zone of the imaging point is used as the offset aperture for offset calculation during the migration imaging process, both correct imaging of steep dips and suppression of offset noise can be considered. Strictly speaking, for each imaging point, its Fresnel zone needs to be determined based on the full (multiple) azimuth dip domain offset gathers. However, the number of imaging points in the entire work area is huge, and determining the Fresnel zones of all imaging points will consume huge computing and storage resources.

[0023] In order to reduce the amount of calculation and the requirements for storage resources, the migration gathers in the dip domain along the survey line (Inline direction) and perpendicular to the survey line (Crossline direction) are generally constructed during the migration process, and the Fresnel zones are directly and vividly displayed in the above migration gathers; secondly, the dip values ​​corresponding to the Fresnel zone boundaries in the Inline direction and the Crossline direction picked up by human-computer interaction based on the dip domain migration gathers for each imaging point are regarded as the long and short sides of the rectangle to construct a rectangle, and the constructed rectangle is approximately regarded as the Fresnel zone area of ​​the imaging point, that is, the migration aperture; then, the obtained migration aperture is applied to the migration process for reprocessing to obtain the migration imaging result that takes into account both steep dip imaging and migration noise suppression.

[0024] However, in the above-mentioned pre-stack migration technology based on the Fresnel zone of the imaging point as the migration aperture, on the one hand, the rectangle constructed by the inclination value corresponding to the Fresnel zone boundary in the Inline direction and the Crossline direction picked by human-computer interaction is approximated as the Fresnel zone area of ​​the imaging point. This approach is based on the basic assumption that the underground three-dimensional structure changes gently except in the Inline and Crossline directions. This approximation is relatively complex when the underground three-dimensional structure changes. For example, when there are drastic changes in multiple directions, the accuracy of the set migration aperture is not high, which causes the steep-angle structure of some underground directions to be lost in the migration imaging results. On the other hand, as a key link in the implementation process of this existing technology, even if the migration aperture is determined based on the Fresnel zone boundary in the Inline and Crossline directions of the imaging point picked by human-computer interaction in the inclination domain migration gather, the time spent by the data processing personnel is huge, which greatly limits the promotion and application of this technology. In other words, the above-mentioned pre-stack migration technology based on the Fresnel zone in the Inline and Crossline directions of the imaging point as the migration aperture still has the problems of low imaging accuracy and huge time consumption of data processing personnel.

[0025] In view of this, an embodiment of the present invention provides a high-precision imaging method for seismic data, which is used to solve the technical problems involved in the above. Figure 1 A flowchart of a method for high-precision imaging of seismic data provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method specifically comprises the following steps: Step S102, obtaining the seismic pre-stack gathers of the area to be imaged.

[0026] Step S104, for each imaging line in the imaging space of the work area to be imaged, multi-azimuth pre-stack migration processing is performed on the seismic pre-stack gathers with a fixed migration aperture to obtain a multi-azimuth migration imaging section set for each imaging line.

[0027] In an embodiment of the present invention, the source data used for high-precision imaging of the imaging area is the seismic pre-stack gathers of the imaging area, which can be obtained through professional seismic data acquisition equipment. Next, low-precision imaging is first performed on each imaging line in the imaging area. Specifically, multi-azimuth pre-stack migration processing is performed on the seismic pre-stack gathers with a fixed offset aperture to obtain a set of multi-azimuth offset imaging profiles for each imaging line. In other words, for each imaging line, multiple offset imaging profiles are output, and each offset imaging profile is obtained by performing an offset process on the seismic data input in the azimuth. In an embodiment of the present invention, the fixed offset aperture is not less than the maximum inclination value (known parameter) of the underground structure to be imaged.

[0028] If the imaging space contains imaging lines, each imaging line contains CDP (Common Depth Point), each CDP contains imaging points, where the coordinates of any imaging point are , then the multi-azimuth migration imaging result is expressed as: , .in, Indicates the number of azimuth intervals, Indicates the preset azimuth interval length (interval), It means the two-way travel time of earthquake waves. and Represents seismic waves from the shot point Propagate to the imaging point and propagates from the imaging point to the receiving point When traveling, Indicates the first-order derivative of the seismic data sequence of the jth seismic trace, k represents the number of seismic traces in all input pre-stack seismic trace sets, i represents the azimuth number of the seismic trace, , where θ represents the azimuth of the seismic trace and is calculated as follows: .

[0029] is a weight function, which means that a fixed migration aperture is applied during the migration process. Its expression is: ,in, , They represent the maximum formation dip angles in the x and y directions of the preset work area (input offset parameter, which is a constant). , , represents the RMS velocity of the image point (also known as the offset parameter).

[0030] based on From the expression of , we can see that is a four-dimensional data body with dimensions , , , Among them, if , , Unchanged, only The change, that is, the single azimuth imaging result at the same CDP, is called a single azimuth imaging seismic trace; , constant, , The change, that is, the single azimuth imaging result at the same imaging line, is called a single azimuth imaging profile. constant, , , The multi-azimuth imaging results on the same imaging line are called the multi-azimuth migration imaging profile set.

[0031] Step S106, using the target neural network model to process the multi-azimuth offset imaging section set of the target imaging line to obtain a high-precision offset imaging section of the target imaging line.

[0032] It is known that the imaging accuracy of the offset imaging section obtained after pre-stack migration processing with a fixed offset aperture is low. Therefore, an embodiment of the present invention pre-trains a target neural network model, wherein the target neural network model is a lightweight neural network composed of a Transformer neural network and a CNN neural network, and the Transformer neural network is located between the encoding module and the decoding module of the CNN neural network; the training goal of the target neural network model is that the error between the output offset imaging section prediction result and the actual high-precision offset imaging section is less than a preset threshold; optionally, the actual high-precision offset imaging section is an imaging result obtained after the Fresnel zone of the imaging point is determined based on a multi-azimuth inclination domain offset gather, and the Fresnel zone is used as the offset aperture for offset calculation.

[0033] That is to say, after the multi-directional offset imaging profile set of the target imaging line is input into the target neural network model, the offset imaging profile prediction result output by the target neural network model is the high-precision offset imaging profile of the target imaging line, where the target imaging line represents any imaging line in the imaging space.

[0034] The target neural network model combines the advantages of CNN neural network and Transformer neural network, which greatly reduces the memory requirements and training time of neural network training, thereby improving computing efficiency. In addition, the application of the target neural network model in the pre-stack migration imaging process allows seismic data processing personnel to only use a fixed migration aperture to perform multi-azimuth migration processing on the input seismic data (i.e., seismic pre-stack gathers) to achieve correct imaging of underground steep-angle structures and migration noise suppression, which not only avoids the huge workload of determining the spatially variable migration aperture of the imaging space, but also improves the accuracy of migration imaging.

[0035] Step S108 : determining a high-precision imaging result of the work area to be imaged based on the high-precision offset imaging profiles of all imaging lines.

[0036] The embodiment of the present invention uses the set of high-precision offset imaging sections of all imaging lines as the high-precision imaging result of the work area to be imaged.

[0037] The embodiment of the present invention provides a method for high-precision imaging of seismic data. After obtaining the seismic pre-stack gathers of the work area to be imaged, firstly, for each imaging line in the imaging space, the seismic pre-stack gathers are subjected to multi-azimuth pre-stack migration processing with a fixed offset aperture to obtain a multi-azimuth offset imaging profile set for each imaging line, and then the multi-azimuth offset imaging profile set of the target imaging line is processed using a target neural network model to obtain a high-precision offset imaging profile of the target imaging line. The target neural network model is a lightweight neural network based on a Transformer neural network and a CNN neural network, and the training goal is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold, therefore, it can ensure that the output offset imaging profile is a high-precision imaging result, effectively alleviating the technical problem of low imaging accuracy in the existing seismic data imaging method.

[0038] The above describes that the embodiment of the present invention uses a trained target neural network model when performing high-precision imaging of seismic data. The following specifically describes how to obtain the target neural network model through training. In an optional implementation, the embodiment of the present invention further includes the following steps: Step S201, obtaining the seismic pre-stack gathers of the training area.

[0039] Step S202, based on the seismic pre-stack gathers of the training area, high-precision migration imaging sections of multiple typical imaging lines and a corresponding set of multi-azimuth migration imaging sections are determined; wherein the high-precision migration imaging section is an imaging section obtained after migration processing using the Fresnel zone of the imaging point as the migration aperture.

[0040] Step S203, using the multi-azimuth offset imaging profile set of each typical imaging line as input data of the initial neural network model, and using the corresponding high-precision offset imaging profile as a training label, the initial neural network model is trained until the training goal is reached to obtain a target neural network model.

[0041] Specifically, the embodiment of the present invention does not specifically limit the number and position of training work areas. If the number of training work areas is 1, the training work area is the work area to be imaged; if the number of training work areas is multiple, it is not limited to include the above-mentioned work area to be imaged. In theory, if the number of training work areas is more, the training samples of the model have better diversity, thereby making the generalization ability of the target neural network stronger.

[0042] After obtaining the seismic pre-stack gathers of the training area, the method in the above step S104 can be referred to to determine a set of multi-azimuth migration imaging profiles of multiple typical imaging lines based on the seismic pre-stack gathers of the training area, and the high-precision migration imaging profile of each typical imaging line is an imaging profile obtained after the Fresnel zone of the imaging point is used as the migration aperture for migration processing. In order to ensure the accuracy of the training label, the Fresnel zone of the imaging point is determined based on the multi-azimuth dip domain migration gathers. In the embodiment of the present invention, the number of azimuths included in the multi-azimuth must be greater than 2, and the number of azimuths included in the multi-azimuth is , , Indicates the preset azimuth interval length.

[0043] In an optional embodiment, the selection criteria of typical imaging lines are: at locations with drastic structural changes, imaging lines are selected based on a first interval; at locations with gentle structural changes, imaging lines are selected based on a second interval, and the first interval is smaller than the second interval.

[0044] Specifically, in locations with drastic structural changes, the geological structure is complex, and more dense imaging lines (first intervals) are required to capture detailed geological information and ensure the accuracy and reliability of the imaging results. In locations with gentle structures, the geological structure is relatively simple, and sparser imaging lines (second intervals) can be used to reduce the amount of data and processing time, while still being able to obtain sufficient geological information. In addition, by selecting imaging lines using the first interval in areas with drastic structural changes, limited resources can be used more effectively to ensure detailed imaging of key areas. Selecting imaging lines using the second interval in areas with gentle structures can reduce unnecessary data processing, saving time and costs. The above flexible typical imaging line selection criteria can be adjusted according to different geological conditions to ensure high-quality imaging results in various geological environments and optimize resource utilization.

[0045] After obtaining high-precision offset imaging sections of multiple typical imaging lines and the corresponding multi-azimuth offset imaging section sets, all multi-azimuth offset imaging section sets are used as input data sets for initial neural network training, and all high-precision offset imaging sections are used as label data sets for initial neural network training. Based on the correspondence between the input data and the label data, the initial neural network model is trained until the error between the offset imaging section prediction result output by the model and the training label is less than a preset threshold. The training is then terminated to obtain the target neural network model.

[0046] In an optional implementation manner, in the above step S202, determining high-precision migration imaging profiles of multiple typical imaging lines based on the seismic pre-stack gathers in the training work area specifically includes the following steps: Step S2021, calculating a multi-azimuth dip domain migration gather of a target typical imaging line based on the seismic pre-stack gathers of the training work area; wherein the target typical imaging line represents any one of the multiple typical imaging lines.

[0047] Step S2022, based on the multi-azimuth dip domain migration gathers, human-computer interaction picks the Fresnel zones of multiple typical imaging points.

[0048] Step S2023, using the Fresnel zones of multiple typical imaging points as the migration aperture, performing pre-stack migration processing on the seismic pre-stack gathers in the training area, and obtaining a high-precision migration imaging profile of the target typical imaging line.

[0049] In the existing pre-stack migration technology based on the Fresnel zone of the imaging point as the migration aperture, the Fresnel zones in the Inline direction and the Crossline direction picked by human-computer interaction are used as the migration aperture. Compared with the prior art, when determining the training label, the embodiment of the present invention constructs a "multi-azimuth (i.e., omnidirectional)" dip domain migration gather, and then picks the Fresnel zone of the typical imaging point by human-computer interaction. Compared with the method of picking the dip domain migration gather based on two azimuths (i.e., the Inline direction and the Crossline direction) in the prior art, the obtained Fresnel zone is more accurate, thereby ensuring the accuracy of the pre-stack migration processing result. The embodiment of the present invention does not specifically limit the selection method of the typical imaging point, and can refer to the selection criteria of the typical imaging line, that is, at the location where the structure changes violently, the selected imaging points are relatively dense, and at the location where the structure is gentle, the selected imaging points are relatively sparse.

[0050] In an optional implementation, the initial neural network model further includes: a normalization layer and a denormalization layer; the above step S203, using the multi-azimuth migration imaging section set of each typical imaging line as input data of the initial neural network model, and using the corresponding high-precision migration imaging section as a training label, to train the initial neural network model, specifically includes the following steps: Step S2031: The normalization layer normalizes the multi-azimuth migration imaging section set and the corresponding high-precision migration imaging section of the target typical imaging line to obtain an amplitude-normalized multi-azimuth migration imaging section set and an amplitude-normalized high-precision migration imaging section.

[0051] Specifically, given that the amplitude range of seismic data is too large, which may be as large as millions, it is not conducive to calculation if it is directly input into the neural network. Therefore, the embodiment of the present invention reduces the value range of the input data and the label data through normalization to facilitate the training of the initial neural network model. In addition, considering that the amplitude (value range) of seismic data varies greatly, the method of normalizing the input data and the label data using the same normalization parameters will cause the predicted amplitude and actual error obtained by the trained neural network during reasoning to be large. In view of this, the embodiment of the present invention needs to calculate the normalization parameters of the multi-azimuth offset imaging profile set and the normalization parameters of the high-precision offset imaging profile, and then normalize the two separately.

[0052] Step S2032: the encoding module encodes each local time window data in the multi-azimuth migration imaging section set after amplitude normalization to extract local features and section position information of each local time window data.

[0053] Step S2033, the local features of all local time window data are expanded to construct a long sequence of local features.

[0054] In step S2034, the Transformer neural network extracts global features and effective features of the long sequence of local features to obtain a target feature sequence.

[0055] Step S2035, feature folding is performed on the target feature sequence, and the decoding module decodes the folded features and section position information to obtain the reconstructed offset imaging section.

[0056] Step S2036, calculating the loss function value based on the reconstructed offset imaging section and the high-precision offset imaging section after amplitude normalization, so as to train the initial neural network model based on the loss function value.

[0057] Step S2037, the denormalization layer performs denormalization processing on the reconstructed migration imaging section based on the multi-azimuth migration imaging section set of the target typical imaging line and the corresponding high-precision migration imaging section to obtain the high-precision migration imaging section after amplitude restoration.

[0058] In the embodiment of the present invention, the neural network model has multiple inputs and a single output, wherein the input is a set of multi-azimuth migration imaging sections, and the output is a high-precision migration section reconstructed after the neural network extracts and fuses the features of the multi-azimuth migration imaging section set. Figure 2A data processing flow diagram of an initial neural network model provided in an embodiment of the present invention, wherein a convolutional neural network (CNN) includes an encoder block and a decoder block. The embodiment of the present invention does not specifically limit the structure of the encoder block and the decoder block, as long as the above-mentioned encoding function and decoding function can be realized.

[0059] In an optional embodiment, in the encoding module, each layer contains two 3×3 convolution operations for feature extraction and feature mapping of data; its activation layer uses ReLU as the activation function to increase nonlinear mapping; its pooling layer is 2×2 maximum pooling, which is used to extract features and downsample data, reduce neural network parameters, and facilitate the training of subsequent Transformer models. In the decoding module, each layer contains two 3×3 convolution operations, ReLU as the activation function, and an upsampling operation of 2×2 for restoring the sampling size of data and reducing the number of channels. In addition, the last layer adds a 1×1 convolution kernel for fusing and reconstructing high-precision offset result data.

[0060] Figure 3 A schematic diagram of a conversion process of output data of an encoding module to input data of a decoding module provided by an embodiment of the present invention, through Figure 3 It can be seen that the data output by the encoding module is actually multiple two-dimensional feature image blocks. By flattening the elements at the same position in multiple two-dimensional feature image blocks into one dimension (that is, feature expansion), multiple one-dimensional sequences can be obtained. The above multiple one-dimensional sequences are connected head to tail in a specified order to construct a long sequence of local features for input into the Transformer neural network. The Transformer neural network can capture the global features and effective features contained in the input data to obtain the target feature sequence, where the global feature represents the features possessed by the sample data as a whole; the effective feature represents the features possessed by the effective data in the sample excluding interference such as noise. Next, the target feature sequence is folded to construct multiple two-dimensional feature image blocks with the same data structure as the output data of the encoding module. The decoding module can obtain the reconstructed offset imaging section by decoding it.

[0061] Next, the loss function value can be calculated based on the reconstructed offset imaging section and the high-precision offset imaging section after amplitude normalization, and then the initial neural network model can be trained based on the loss function value. The embodiment of the present invention does not specifically calculate the calculation method of the loss function value. The user can design it according to the actual situation, as long as it can ensure that the greater the error between the reconstructed offset imaging section and the high-precision offset imaging section after amplitude normalization, the greater the loss function value. When training the model, batch processing can also be used for training.

[0062] The prediction results of the neural network are generally normalized to [-1, 1] or [0, 1]. Therefore, after the decoding module outputs the prediction results (that is, the reconstructed migration imaging section), it is necessary to perform amplitude recovery (that is, "denormalization") on the prediction results to obtain effective seismic amplitude values. The amplitude recovery function is implemented by the denormalization layer. After the denormalization layer learns the amplitude relationship between the input data (multi-azimuth migration imaging section set) and the label data (high-precision migration imaging section), it can perform denormalization on the reconstructed migration imaging section to obtain a high-precision migration imaging section after amplitude recovery.

[0063] The embodiment of the present invention does not specifically limit the structure of the Transformer neural network, and the user can make a choice according to actual needs. Optionally, the Transformer neural network model, each layer of which is composed of a multi-head self-attention mechanism (efficient multi-head attention, EMHA) and a multi-layer perceptron block (multi-layer perceptron, MLP).

[0064] Among them, in the multi-head self-attention mechanism (EMHA), the input sequence first passes through three different linear transformation layers to obtain Query, Key and Value respectively. Then, these transformed vectors are divided into several "heads", each of which has its own independent Query, Key and Value matrices. For each head, a scaled dot-product attention operation is performed. Finally, the outputs of all heads are spliced ​​together and fused through a linear layer to obtain the final attention output vector. In this way, the multi-head self-attention mechanism (EMHA) can process the input sequence from different angles in parallel, improving the ability of the Transformer neural network model to understand and capture complex dependencies. Alternatively, the multi-head self-attention mechanism (EMHA) obtains the attention distribution of different subspaces of the input sequence by running multiple independent attention mechanisms in parallel, thereby more comprehensively capturing the potential multiple feature associations in the sequence.

[0065] The multi-layer perceptron block (MLP) is a component that further processes information after the multi-head self-attention mechanism. Through two linear transformations and a nonlinear activation function operation, the Transformer neural network model can effectively process and represent complex input data while capturing the dependencies between different positions in the sequence.

[0066] In an optional implementation, in the above step S2031, the normalization layer respectively normalizes the multi-azimuth migration imaging section set of the target typical imaging line and the corresponding high-precision migration imaging section, which specifically includes the following steps: Step S20311, determining the amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line, and obtaining the amplitude value set corresponding to each offset imaging section.

[0067] Step S20312: superimpose the amplitude values ​​of the same sequence number in all amplitude value sets, and take the absolute value after superposition to obtain a first target amplitude value set.

[0068] Step S20313: Using the maximum amplitude value in the first target amplitude value set as a normalization parameter of the multi-azimuth migration imaging section set, so as to perform normalization processing on the multi-azimuth migration imaging section set.

[0069] Step S20314, determining the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, and taking the absolute value to obtain a second target amplitude value set.

[0070] Step S20315: Using the maximum amplitude value in the second target amplitude value set as a normalization parameter of the high-precision migration imaging section to perform normalization processing on the high-precision migration imaging section.

[0071] Specifically, the set of amplitude values ​​corresponding to all the offset imaging sections in the set of multi-azimuth offset imaging sections of the target typical imaging line can be expressed as: ,in, represents the set of amplitude values ​​of all positions on the migration imaging section with azimuth number i, or, It can also be expressed in the form of a two-dimensional array, where the position of the array element corresponds to the position of the pixel point. Based on the above description of the normalization of the multi-azimuth migration imaging section set, the normalization parameter of the multi-azimuth migration imaging section set is: , the normalized processing result of the multi-azimuth migration imaging profile set is expressed as: .

[0072] If the amplitude value set of all points on the high-precision migration imaging section is L, then the normalized parameter of the high-precision migration imaging section is , the normalized processing result of the high-precision migration imaging section is expressed as: .

[0073] In an optional implementation, in step S2037, the denormalization layer performs denormalization processing on the reconstructed migrated imaging section based on the multi-azimuth migrated imaging section set of the target typical imaging line and the corresponding high-precision migrated imaging section, specifically including the following steps: Step S20371, determining the amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line, and determining the energy average value of the multi-azimuth offset imaging section set based on the amplitude value to obtain a first energy.

[0074] Step S20372, determine the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, determine the energy average value of the high-precision offset imaging section based on the amplitude value, obtain the second energy, and determine the maximum amplitude in the high-precision offset imaging section.

[0075] Step S20373, determining the inverse normalization parameters of the reconstructed offset imaging section based on the first energy, the second energy and the maximum amplitude.

[0076] Step S20374, performing denormalization processing on the reconstructed migration imaging section based on the denormalization parameter to obtain a high-precision migration imaging section after amplitude restoration.

[0077] In the embodiment of the present invention, the denormalization parameter is expressed as ,in, , , , represents the first energy, Represents the second energy. The energy of each position point in the imaging section is the square of its amplitude. Indicates the maximum amplitude in the high-precision migration imaging section. Using the inverse normalization parameter The reconstructed migration imaging section is subjected to inverse normalization processing, that is, amplitude recovery processing, which is expressed as: ,in, Represents the set of amplitude values ​​of all positions in the reconstructed migration imaging section, Represents the amplitude value set of all position points in the high-precision migration imaging section after amplitude recovery.

[0078] In one embodiment, when the initial neural network is trained, the azimuth angle interval ∆θ=45 degrees is pre-specified, that is, the multi-azimuth angle is specifically directions. In the neural network, the CNN layer of the network is set to 5 layers, the initial convolution kernel is set to 32, and the number of transformer heads is set to 8. The size of the input data is 128×128×8, the initial learning rate is set to 0.001, and it is reduced by half every 200 training rounds (epochs). The Adam optimizer and smooth L1 loss function are applied to train the neural network model. During the training process, the total loss of the training data set and the validation data set is first calculated, and then they are divided by the storage space occupied by each. After 400 training rounds, the loss tends to be stable. For 8.6G of training data, the training of the neural network can be completed in 4.8 hours.

[0079] In order to verify the performance of the embodiment of the present invention, Figure 4 A schematic diagram of a multi-azimuth offset imaging section set provided by an embodiment of the present invention, Figure 4 Views a to h correspond to the following azimuth ranges: (a) , (b) , (c) , (d) , (e) , (f) , (g) , (h) . Figure 5 A comparison diagram of imaging results of various seismic data imaging methods provided by embodiments of the present invention, Figure 5 The view a in the middle is the result of direct superposition of a set of multi-directional offset sections. Figure 5 The middle b view is the result of migration and superposition based on the dip domain migration gathers in two directions (Inline direction and Crossline direction) after picking the Fresnel zone interactively. Figure 5 View c in the middle is the prediction result of applying the target neural network model in the embodiment of the present invention.

[0080] Figure 5 The a view in the middle is a direct superposition of the multi-directional offset profile set, which is equivalent to the fixed offset aperture offset profile. In the resulting image, the signal-to-noise ratio is very low, and the underground geological structure is not clearly depicted: the continuity of the horizontal strata is destroyed, and the breakpoints of the faults are blurred. Compared with the a view, Figure 5 The signal-to-noise ratio of view b is significantly improved, especially the random noise is significantly suppressed, and the continuity of horizontal strata and the breakpoint characterization of faults are significantly improved compared with view a. However, there is obvious regular noise in the time window of CDP850~1050 and Time1.60S~3.10S, which is not conducive to the accurate characterization of the large faults from top to bottom in this area. Figure 5In the profile shown in the middle c view, both random noise and regular noise are significantly suppressed, and the complex structure of the underground medium is portrayed more clearly: the horizontal stratum imaging has good continuity, the fault breakpoints are crisp, and the portrayal is accurate.

[0081] Embodiment 2 An embodiment of the present invention further provides a high-precision imaging device for seismic data, which is mainly used to execute the high-precision imaging method for seismic data provided in the above-mentioned embodiment 1. The high-precision imaging device for seismic data provided in the embodiment of the present invention is specifically introduced below.

[0082] Figure 6 A functional module diagram of a high-precision seismic data imaging device provided by an embodiment of the present invention, such as Figure 6 As shown, the device mainly includes: a first acquisition module 10, an offset module 20, a processing module 30, and a first determination module 40, wherein: The first acquisition module 10 is used to acquire the seismic pre-stack gathers of the work area to be imaged.

[0083] The migration module 20 is used to perform multi-azimuth pre-stack migration processing on the seismic pre-stack gathers with a fixed migration aperture for each imaging line in the imaging space of the work area to be imaged, so as to obtain a multi-azimuth migration imaging section set for each imaging line.

[0084] The processing module 30 is used to process the multi-azimuth offset imaging profile set of the target imaging line using the target neural network model to obtain a high-precision offset imaging profile of the target imaging line; wherein the target neural network model is a lightweight neural network composed of a Transformer neural network and a CNN neural network, and the Transformer neural network is located between the encoding module and the decoding module of the CNN neural network; the training goal of the target neural network model is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold; the target imaging line represents any imaging line in the imaging space.

[0085] The first determination module 40 is used to determine the high-precision imaging result of the work area to be imaged based on the high-precision offset imaging profiles of all imaging lines.

[0086] The embodiment of the present invention provides a high-precision imaging device for seismic data. After obtaining the seismic pre-stack gathers of the work area to be imaged, firstly, for each imaging line in the imaging space, the seismic pre-stack gathers are subjected to multi-azimuth pre-stack migration processing with a fixed offset aperture to obtain a multi-azimuth offset imaging profile set for each imaging line, and then the multi-azimuth offset imaging profile set of the target imaging line is processed using a target neural network model to obtain a high-precision offset imaging profile of the target imaging line. The target neural network model is a lightweight neural network based on a Transformer neural network and a CNN neural network, and the training goal is that the error between the output offset imaging profile prediction result and the actual high-precision offset imaging profile is less than a preset threshold, therefore, it can ensure that the output offset imaging profile is a high-precision imaging result, effectively alleviating the technical problem of low imaging accuracy in the existing seismic data imaging method.

[0087] Optionally, the device further comprises: The second acquisition module is used to obtain the seismic pre-stack gathers in the training area.

[0088] The second determination module is used to determine the high-precision migration imaging profiles of multiple typical imaging lines and the corresponding multi-azimuth migration imaging profile set based on the seismic pre-stack gathers in the training work area; wherein the high-precision migration imaging profile is an imaging profile obtained after the Fresnel zone of the imaging point is used as the migration aperture for migration processing.

[0089] The training module is used to use the multi-azimuth offset imaging section set of each typical imaging line as the input data of the initial neural network model, and use the corresponding high-precision offset imaging section as the training label to train the initial neural network model until the training goal is achieved to obtain the target neural network model.

[0090] Optionally, the second determining module includes: The calculation unit is used to calculate the multi-azimuth dip domain migration gathers of the target typical imaging line based on the seismic pre-stack gathers in the training area; wherein the target typical imaging line represents any imaging line among the multiple typical imaging lines.

[0091] The picking unit is used to pick the Fresnel zones of multiple typical imaging points through human-computer interaction based on the multi-azimuth dip domain migration gathers.

[0092] The migration unit is used to use the Fresnel zones of multiple typical imaging points as the migration aperture, perform pre-stack migration processing on the seismic pre-stack gathers in the training area, and obtain a high-precision migration imaging profile of the target typical imaging line.

[0093] Optionally, the initial neural network model further includes: a normalization layer and a denormalization layer; the training module includes: The normalization unit is used to normalize the multi-azimuth migration imaging section set and the corresponding high-precision migration imaging section of the target typical imaging line through the normalization layer to obtain the amplitude-normalized multi-azimuth migration imaging section set and the amplitude-normalized high-precision migration imaging section.

[0094] The encoding unit is used to encode each local time window data in the multi-azimuth offset imaging section set after amplitude normalization through the encoding module to extract the local features and section position information of each local time window data.

[0095] The expansion unit is used to expand the local features of all local time window data to construct a long sequence of local features.

[0096] The extraction unit is used to extract the global features and effective features of the long sequence of local features through the Transformer neural network to obtain the target feature sequence.

[0097] The folding unit is used to perform feature folding on the target feature sequence, and the decoding module decodes the folded features and section position information to obtain the reconstructed offset imaging section.

[0098] The calculation unit is used to calculate the loss function value based on the reconstructed offset imaging section and the high-precision offset imaging section after amplitude normalization, so as to train the initial neural network model based on the loss function value.

[0099] The denormalization unit is used to denormalize the reconstructed migration imaging section based on the multi-azimuth migration imaging section set of the target typical imaging line and the corresponding high-precision migration imaging section through the denormalization layer to obtain the high-precision migration imaging section after amplitude restoration.

[0100] Optionally, the normalization unit is specifically used for: The amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line is determined to obtain the amplitude value set corresponding to each offset imaging section.

[0101] The amplitude values ​​of the same sequence number in all the amplitude value sets are superimposed, and the absolute value is taken after superposition to obtain a first target amplitude value set.

[0102] The maximum amplitude value in the first target amplitude value set is used as a normalization parameter of the multi-azimuth migration imaging section set, so as to perform normalization processing on the multi-azimuth migration imaging section set.

[0103] The amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line is determined, and the absolute value is taken to obtain a second target amplitude value set.

[0104] The maximum amplitude value in the second target amplitude value set is used as a normalization parameter of the high-precision migration imaging section to perform normalization processing on the high-precision migration imaging section.

[0105] Optionally, the denormalization unit is specifically used for: The amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line is determined, so as to determine the energy average value of the multi-azimuth offset imaging section set based on the amplitude value to obtain the first energy.

[0106] Determine the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, determine the energy average value of the high-precision offset imaging section based on the amplitude value, obtain the second energy, and determine the maximum amplitude value in the high-precision offset imaging section.

[0107] Based on the first energy, the second energy and the maximum amplitude, an inverse normalization parameter of the reconstructed migration imaging section is determined.

[0108] The reconstructed migration imaging section is denormalized based on the denormalization parameters to obtain a high-precision migration imaging section after amplitude restoration.

[0109] Optionally, the selection criteria of typical imaging lines are: in locations with drastic structural changes, imaging lines are selected based on a first interval; in locations with gentle structural changes, imaging lines are selected based on a second interval, and the first interval is smaller than the second interval.

[0110] Embodiment 3 See also Figure 7 An embodiment of the present invention provides an electronic device, which includes: a processor 60, a memory 61, a bus 62 and a communication interface 63, wherein the processor 60, the communication interface 63 and the memory 61 are connected via the bus 62; the processor 60 is used to execute an executable module stored in the memory 61, such as a computer program.

[0111] The memory 61 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 63 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0112] The bus 62 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0113] Among them, the memory 61 is used to store programs, and the processor 60 executes the program after receiving the execution instruction. The method executed by the device defined by the process disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 60 or implemented by the processor 60.

[0114] The processor 60 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 60. The above processor 60 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 61, and the processor 60 reads the information in the memory 61 and completes the steps of the above method in combination with its hardware.

[0115] A computer program product of a method and apparatus for high-precision imaging of seismic data provided in an embodiment of the present invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0116] In addition, each functional unit in each embodiment of the present invention 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.

[0117] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that can be executed by a processor. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., various media that can store program codes.

[0118] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0119] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0120] In addition, the terms "horizontal", "vertical", "overhanging" and the like do not mean that the components are required to be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0121] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A method for high-precision imaging of seismic data, characterized in that: include: Obtain seismic pre-stack gathers of the area to be imaged; For each imaging line in the imaging space of the work area to be imaged, multi-azimuth pre-stack migration processing is performed on the seismic pre-stack gathers with a fixed migration aperture to obtain a multi-azimuth migration imaging section set for each imaging line; The target neural network model is used to process the multi-azimuth offset imaging section set of the target imaging line to obtain a high-precision offset imaging section of the target imaging line; wherein the target neural network model is a lightweight neural network composed of a Transformer neural network and a CNN neural network, and the Transformer neural network is located between the encoding module and the decoding module of the CNN neural network; the training goal of the target neural network model is that the error between the output offset imaging section prediction result and the actual high-precision offset imaging section is less than a preset threshold; the target imaging line represents any imaging line in the imaging space; Based on the high-precision offset imaging sections of all imaging lines, a high-precision imaging result of the work area to be imaged is determined.

2. The high-precision imaging method for seismic data according to claim 1, characterized in that: Also includes: Obtain the seismic pre-stack gathers in the training area; Determine a high-precision migration imaging section of a plurality of typical imaging lines and a corresponding multi-azimuth migration imaging section set based on the seismic pre-stack gathers of the training work area; wherein the high-precision migration imaging section is an imaging section obtained by performing migration processing by using the Fresnel zone of the imaging point as the migration aperture; The multi-azimuth offset imaging section set of each typical imaging line is used as input data of the initial neural network model, and the corresponding high-precision offset imaging section is used as a training label to train the initial neural network model until the training goal is achieved, thereby obtaining the target neural network model.

3. The high-precision seismic data imaging method according to claim 2, characterized in that: Determine a high-precision migration imaging profile of multiple typical imaging lines based on the seismic pre-stack gathers in the training area, including: Calculating a multi-azimuth dip domain migration gather of a target typical imaging line based on the seismic pre-stack gathers of the training work area; wherein the target typical imaging line represents any one of the multiple typical imaging lines; Based on the multi-azimuth dip-domain migration gathers, human-computer interaction is used to pick up Fresnel zones of multiple typical imaging points; The Fresnel zones of the multiple typical imaging points are used as migration apertures, and pre-stack migration processing is performed on the seismic pre-stack gathers in the training area to obtain a high-precision migration imaging profile of the target typical imaging line.

4. The method for high-precision imaging of seismic data according to claim 3, characterized in that: The initial neural network model further includes: a normalization layer and a denormalization layer; taking the multi-azimuth migration imaging section set of each typical imaging line as input data of the initial neural network model, and taking the corresponding high-precision migration imaging section as a training label, training the initial neural network model, including: The normalization layer normalizes the multi-azimuth migration imaging section set and the corresponding high-precision migration imaging section of the target typical imaging line, respectively, to obtain the amplitude-normalized multi-azimuth migration imaging section set and the amplitude-normalized high-precision migration imaging section; The encoding module encodes each local time window data in the amplitude-normalized multi-azimuth migration imaging section set to extract local features and section position information of each local time window data; Expanding the local features of all the local time window data to construct a long sequence of local features; The Transformer neural network extracts global features and effective features of the long sequence of local features to obtain a target feature sequence; The target feature sequence is feature folded, and the decoding module decodes the folded features and the section position information to obtain a reconstructed offset imaging section; Calculating a loss function value based on the reconstructed migration imaging section and the amplitude-normalized high-precision migration imaging section, so as to train the initial neural network model based on the loss function value; The denormalization layer performs denormalization processing on the reconstructed offset imaging section based on the multi-azimuth offset imaging section set of the target typical imaging line and the corresponding high-precision offset imaging section to obtain the high-precision offset imaging section after amplitude restoration.

5. The method for high-precision imaging of seismic data according to claim 4, characterized in that: The normalization layer performs normalization processing on the multi-azimuth migration imaging section set of the target typical imaging line and the corresponding high-precision migration imaging section, respectively, including: Determine the amplitude value of each position point in the multi-azimuth offset imaging section set of the typical imaging line of the target, and obtain the amplitude value set corresponding to each offset imaging section; The amplitude values ​​of the same sequence number in all the amplitude value sets are superimposed, and the absolute value is taken after superposition to obtain a first target amplitude value set; using the maximum amplitude value in the first target amplitude value set as a normalization parameter of the multi-azimuth migration imaging section set, so as to perform normalization processing on the multi-azimuth migration imaging section set; Determine the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, and take the absolute value to obtain a second target amplitude value set; The maximum amplitude value in the second target amplitude value set is used as a normalization parameter of the high-precision migration imaging section to perform normalization processing on the high-precision migration imaging section.

6. The method for high-precision imaging of seismic data according to claim 4, characterized in that: The denormalization layer performs denormalization processing on the reconstructed migrated imaging section based on the multi-azimuth migrated imaging section set of the target typical imaging line and the corresponding high-precision migrated imaging section, including: Determine the amplitude value of each position point in the multi-azimuth offset imaging section set of the target typical imaging line, and determine the energy average value of the multi-azimuth offset imaging section set based on the amplitude value to obtain a first energy; Determine the amplitude value of each position point in the high-precision offset imaging section of the target typical imaging line, determine the energy average value of the high-precision offset imaging section based on the amplitude value to obtain a second energy, and determine the maximum amplitude value in the high-precision offset imaging section; Determining an inverse normalization parameter of the reconstructed migration imaging section based on the first energy, the second energy and the maximum amplitude; The reconstructed migration imaging section is subjected to a denormalization process based on the denormalization parameter to obtain a high-precision migration imaging section after amplitude restoration.

7. The method for high-precision imaging of seismic data according to claim 2, characterized in that: The selection criteria of the typical imaging lines are: in a location with a drastic structural change, the imaging lines are selected based on a first interval, and in a location with a gentle structural change, the imaging lines are selected based on a second interval, and the first interval is smaller than the second interval.

8. A high-precision imaging device for seismic data, characterized in that: The first acquisition module is used to acquire the seismic pre-stack gathers of the work area to be imaged; A migration module, for performing multi-azimuth pre-stack migration processing on the seismic pre-stack gathers with a fixed migration aperture for each imaging line in the imaging space of the work area to be imaged, so as to obtain a multi-azimuth migration imaging section set for each imaging line; A processing module, used for processing a set of multi-azimuth offset imaging sections of a target imaging line using a target neural network model to obtain a high-precision offset imaging section of the target imaging line; wherein the target neural network model is a lightweight neural network based on a Transformer neural network and a CNN neural network, and the Transformer neural network is located between an encoding module and a decoding module of the CNN neural network; a training goal of the target neural network model is that an error between an output offset imaging section prediction result and an actual high-precision offset imaging section is less than a preset threshold; the target imaging line represents any imaging line in the imaging space; The first determination module is used to determine the high-precision imaging result of the work area to be imaged based on the high-precision offset imaging sections of all imaging lines.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the computer program, the method for high-precision imaging of seismic data according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method for high-precision imaging of seismic data according to any one of claims 1 to 7 is implemented.

Images