VSP missing data reconstruction method and device based on self-supervised deep learning model

Abstract

The invention relates to a VSP missing data reconstruction method and device based on a self-supervised deep learning model, and the method comprises the steps: inputting a VSP common detection point gather containing shot missing data into a deep convolutional neural network model; reconstructing the shot-lacking data by using a deep convolutional neural network model, and outputting a reconstructed complete VSP common detection point gather; wherein the deep convolutional neural network model is obtained by training in a self-supervised learning mode by adopting an original complete VSP common detection point gather in a target work area; according to the method, a self-supervised learning normal form is adopted, an existing complete gather data training model in a target work area is directly utilized, additional labeling or complex speed model construction is not needed, and the data preparation cost and dependence on prior knowledge are remarkably reduced; the trained model can perform end-to-end intelligent reconstruction on the input incomplete gather, the complete gather is quickly output, and the problems of complex iteration and large calculation amount in a traditional method based on a wave equation or matrix completion are avoided.

Description

Technical Field

[0001] This invention relates to the field of seismic data processing in petroleum exploration, and in particular to a method and apparatus for reconstructing missing VSP data based on a self-supervised deep learning model. Background Technology

[0002] During seismic data acquisition, natural conditions or construction factors often lead to the loss of some seismic traces. Such omissions can affect the accuracy of subsequent seismic imaging and tectonic interpretation, so interpolation and other methods are usually used to reconstruct the missing seismic traces.

[0003] Currently, most common seismic data interpolation methods belong to the model-driven category, mainly including: seismic data reconstruction methods based on wave equations, which require high data sampling rates, have large computational loads, and whose reconstruction results depend on the accuracy of the subsurface medium velocity model; reconstruction methods based on signal analysis, such as FK domain interpolation and Radon domain interpolation, the former generally only applicable to regularly sampled data, while the latter, although capable of handling irregularly sampled data, still has high requirements for data sampling rates; and reconstruction methods based on complete low-rank matrices, such as the Cadzow method and MSSA method, which typically involve large-scale matrix operations and have large computational and storage overheads.

[0004] Vertical seismic profiling (VSP) is a seismic observation method where shot points are located on the ground and receivers are placed in wells. While this acquisition method results in fewer missing seismic traces, it often leads to incomplete shot records for certain shot points, affecting the completeness of the common receiver gathers. Currently, interpolation reconstruction for incomplete shot situations in VSP data still faces challenges such as strong reliance on prior information, low computational efficiency, or high requirements for data regularity. Summary of the Invention

[0005] This invention provides a method and apparatus for reconstructing missing VSP data based on a self-supervised deep learning model, in order to solve the technical problems of traditional methods, such as strong dependence on prior information, low computational efficiency, and high requirements for data regularity.

[0006] In a first aspect, the present invention provides a method for reconstructing missing VSP data based on a self-supervised deep learning model, comprising: inputting a VSP common receiver point gather containing missing shot data into a deep convolutional neural network model; reconstructing the missing shot data by the deep convolutional neural network model and outputting the reconstructed complete VSP common receiver point gather; wherein, the deep convolutional neural network model is obtained by training the original complete VSP common receiver point gather in the target work area through a self-supervised learning method.

[0007] In some embodiments, the deep convolutional neural network model is trained using the following self-supervised learning method: determining the number of missing shots in the VSP co-receiver point set containing missing shot data; slicing the original complete VSP co-receiver point set and setting a mask in the central region of each slice, the width of the mask being consistent with the number of missing shots, to form a masked region and simulate a missing shot state, constructing a training sample set; and training based on the training sample set with the objective of minimizing the difference between the deep convolutional neural network model's prediction of the masked region and the corresponding original data of the masked region before the mask is applied.

[0008] In some embodiments, the deep convolutional neural network model is an encoder-decoder structure, and the encoder and the decoder perform feature fusion through skip connections.

[0009] In some embodiments, the encoder includes four downsampling stages, each of which includes two convolutional layers; the decoder includes four upsampling stages, each of which includes a transposed convolutional layer and two convolutional layers; and the output of each downsampling stage in the encoder is fused with the input of the corresponding upsampling stage in the decoder via skip connections.

[0010] In some embodiments, the convolutional layer and the transposed convolutional layer used for feature extraction in the encoder and the decoder both employ 5×5 convolutional kernels.

[0011] In some embodiments, a connection module is provided on the jump connection path between the encoder and the decoder. The connection module includes a first submodule and a second submodule. The output of the second submodule is added to the output of the first submodule through a residual connection.

[0012] In some embodiments, the first submodule includes a 5×5 convolutional layer and a ReLU activation function; the second submodule includes two consecutive 5×5 convolutional layers, and each of the 5×5 convolutional layers is followed by a ReLU activation function.

[0013] Secondly, the present invention provides a VSP missing data reconstruction device based on a self-supervised deep learning model, comprising: an input module for inputting a VSP common receiver point gather containing missing shot data into a deep convolutional neural network model; and a reconstruction module for reconstructing the missing shot data by the deep convolutional neural network model and outputting the reconstructed complete VSP common receiver point gather; wherein the deep convolutional neural network model is obtained by training the original complete VSP common receiver point gather in the target work area through a self-supervised learning method.

[0014] Thirdly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model as described in any of the first aspects.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model as described in any of the first aspects.

[0016] This invention provides a method and apparatus for reconstructing missing VSP data based on a self-supervised deep learning model. By employing a self-supervised learning paradigm, it directly trains the model using existing complete gather data from the target work area, eliminating the need for additional annotation or complex velocity model construction. This significantly reduces data preparation costs and reliance on prior knowledge. The trained model can perform end-to-end intelligent reconstruction of incomplete gathers, quickly outputting complete gathers, avoiding the complex iterations and high computational costs of traditional wave equation-based or matrix-complete methods. This invention is particularly suitable for the common missing-bomb situations in actual production, providing a practical and feasible technical solution for the efficient and high-quality reconstruction of VSP data. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a method for reconstructing missing VSP data based on a self-supervised deep learning model, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a deep convolutional neural network model provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the training process of a self-supervised learning method for a deep convolutional neural network model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating how the loss function value changes with the number of training epochs, provided as an embodiment of the present invention. Figure 5 A flowchart illustrating another method for reconstructing missing VSP data based on a self-supervised deep learning model, provided in an embodiment of the present invention; Figure 6a A schematic diagram of a VSP common receiver point gather containing missing shot data provided in an embodiment of the present invention; Figure 6b for Figure 6a A schematic diagram of the corresponding reconstructed complete VSP common detector point gather; Figure 7 A schematic diagram of a VSP missing data reconstruction device based on a self-supervised deep learning model provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Figure 1 This is a flowchart illustrating a VSP missing data reconstruction method based on a self-supervised deep learning model, provided in an embodiment of the present invention. It is applied to a VSP missing data reconstruction device based on a self-supervised deep learning model, or to an electronic device equipped with such a device. Figure 1 As shown, the method includes: Step S101: Input the VSP common receiver point gather containing missing shot data into the deep convolutional neural network model.

[0022] Step S102: The missing shot data is reconstructed by the deep convolutional neural network model, and the reconstructed complete VSP common resonant point gather is output; wherein, the deep convolutional neural network model is obtained by training the original complete VSP common resonant point gather in the target work area through self-supervised learning.

[0023] Specifically, VSP co-receiver gathers containing missing shot data are identified from the target work area and input into a pre-trained deep convolutional neural network model. This model then reconstructs the missing shot data from the input gathers. The model's reconstruction capability stems from the self-supervised learning approach used during its training phase. This approach eliminates the need for manual annotation and directly uses complete, uninterrupted VSP co-receiver gathers from the same target work area as training data. By learning waveform patterns and spatial context relationships from these complete data sets, the model masters the inherent pattern of inferring the whole from partial information and reconstructing missing parts. Therefore, when actual incomplete gathers are input, the model can perform high-precision prediction and filling based on its learned knowledge, ultimately outputting a VSP co-receiver gather that is both visually and physically complete.

[0024] It should be noted that the acquisition methods for VSP common receiver point gathers containing missing shot data and original complete VSP common receiver point gathers are as follows: The original VSP data for the target seismic area is acquired, which consists of seismic records containing multiple receiver points and multiple shot points. Subsequently, this original VSP data is extracted into common receiver point (CRP) gathers. Each trace in the CRP gather corresponds to a seismic signal excited by different shot points received by the same receiver point. Based on this, two types of data are selected from the CRP gathers: one type is CRP gathers containing missing shot data (i.e., the aforementioned VSP common receiver point gathers containing missing shot data), and the other type is completely complete, original CRP gathers without any missing data (i.e., the aforementioned original complete VSP common receiver point gathers).

[0025] In some embodiments, the deep convolutional neural network model is an encoder-decoder structure, and the encoder and the decoder perform feature fusion through skip connections.

[0026] Specifically, this embodiment defines the basic architecture of the model. For example... Figure 2 This is a schematic diagram of the structure of a deep convolutional neural network model provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the model employs an encoder-decoder structure. The encoder is responsible for multi-level downsampling of the incomplete input data, progressively extracting and compressing abstract features. The decoder, on the other hand, is responsible for upsampling these abstract features, progressively recovering the details and complete spatial dimensions of the data. Skip connections refer to establishing direct pathways between a certain layer of the encoder and the corresponding layer of the decoder. This allows the decoder to directly utilize the intermediate layer features extracted by the encoder, which contain rich local information, when reconstructing details. This effectively avoids the loss of details, improves reconstruction accuracy, and achieves better feature fusion.

[0027] In some embodiments, the encoder includes four downsampling stages, each of which includes two convolutional layers; the decoder includes four upsampling stages, each of which includes a transposed convolutional layer and two convolutional layers; and the output of each downsampling stage in the encoder is fused with the input of the corresponding upsampling stage in the decoder via skip connections.

[0028] Specifically, this embodiment further refines the specific design of the encoder and decoder. For example... Figure 2 As shown, the encoder expands the receptive field through four downsampling operations to understand the global contextual information of the data. Each downsampling operation uses two convolutional layers to achieve robust feature extraction. The decoder restores the resolution through four upsampling operations. Each upsampling operation first expands the feature map size through a transposed convolutional layer and then refines the features through two convolutional layers. This symmetrical depth design of "four downsampling / upsampling" combined with the tight integration of skip connections (i.e., the output of each downsampling level is directly fed to the upsampling input of the corresponding level) constitutes a hierarchical and efficient reconstruction network, which is particularly suitable for processing signals with complex waveform structures, such as VSP data.

[0029] In some embodiments, the convolutional layer and the transposed convolutional layer used for feature extraction in the encoder and the decoder both employ 5×5 convolutional kernels.

[0030] Specifically, this embodiment clarifies the key parameters of the model. Unlike common image processing methods that often use small 3×3 convolutional kernels, this embodiment uses large 5×5 convolutional kernels in all core convolution operations. This is because missing shots in VSP data often lead to continuous large-area data loss. Larger convolutional kernels can provide a wider "receptive field," enabling the network to "see" information from more distant valid data when processing the boundaries of missing regions or performing long-distance dependency inference, thereby more accurately reconstructing continuous and reasonable waveforms.

[0031] In some embodiments, a connection module is provided on the jump connection path between the encoder and the decoder. The connection module includes a first submodule and a second submodule. The output of the second submodule is added to the output of the first submodule through a residual connection.

[0032] Specifically, this embodiment introduces a Connection Block, which is positioned on the skip connection path to further process the incoming features. Its core structure consists of two parallel sub-modules, with the outputs of the second sub-module and the first sub-module being residually connected through element-wise addition. This design allows the module to learn the residuals of the input features (i.e., the parts that need enhancement or modification), enabling more efficient capture and fusion of multi-scale spatial features from the encoder, effectively mitigating the vanishing gradient problem common in deep network training, and improving the network's training stability and feature representation capabilities.

[0033] In some embodiments, the first submodule includes a 5×5 convolutional layer and a ReLU activation function; the second submodule includes two consecutive 5×5 convolutional layers, and each of the 5×5 convolutional layers is followed by a ReLU activation function.

[0034] Specifically, this embodiment details the internal structure of the two sub-modules within the connection module. The first sub-module employs a lightweight design with a single-layer 5×5 convolution and activation function for initial feature transformation. The second sub-module uses a deeper structure with two consecutive 5×5 convolutions for more complex nonlinear feature extraction. Both sub-modules utilize large 5×5 convolution kernels to ensure the receptive field and introduce nonlinearity using the ReLU activation function. Finally, the features extracted from the deeper layers of the second sub-module are fused with the initial features from the first sub-module through residual addition. This parallel fusion design of "shallow and deep" allows for more refined and powerful enhancement of skip features without excessively increasing complexity, thereby improving the final reconstruction quality.

[0035] The VSP missing data reconstruction method based on a self-supervised deep learning model provided in this embodiment adopts a self-supervised learning paradigm, directly using existing complete gather data in the target work area to train the model. It does not require additional annotation or complex speed model construction, which significantly reduces the data preparation cost and dependence on prior knowledge. The trained model can perform end-to-end intelligent reconstruction of the input incomplete gather data and quickly output complete gather data, avoiding the problems of complex iteration and large amount of computation in traditional wave equation or matrix completeness methods.

[0036] Furthermore, the deep convolutional neural network model employs an encoder-decoder structure and achieves multi-level feature fusion through skip connections, forming a backbone network that effectively preserves detailed information. Building upon this, four symmetrical downsampling and upsampling operations further enhance the model's ability to capture and reconstruct complex waveform structures and multi-scale features in VSP data. Specifically, the network uses large 5×5 convolutional kernels, significantly expanding the receptive field and enabling the model to reasonably infer continuous missing areas caused by missing shots based on broader contextual information, thus ensuring the spatial continuity and geological plausibility of the reconstructed waveform. In addition, the connection module introduced in the skip connection path further strengthens the feature fusion capability and improves training stability through its residual connection design; the parallel fusion substructure of "one shallow and one deep" in this module achieves more refined feature enhancement while controlling computational complexity. This layered, progressive structural design collectively improves the model's reconstruction accuracy and robustness for VSP missing shot data, especially in cases of continuous large-area missing areas, enabling it to generate high-quality, high-fidelity reconstruction results.

[0037] Based on the foregoing embodiments, Figure 3 This is a schematic diagram illustrating the training process of a self-supervised learning method for a deep convolutional neural network model provided in an embodiment of the present invention, as shown below. Figure 3 As shown, it includes the following steps: Step S301: Determine the number of missing shots in the VSP common receiver point set containing missing shot data.

[0038] Step S302: Slice the original complete VSP common receiver point gather and set a mask in the central region of each slice. The width of the mask is consistent with the number of missing shots to form a mask region and simulate the missing shot state to construct a training sample set.

[0039] Step S303: Train the deep convolutional neural network model based on the training sample set with the goal of minimizing the difference between the predicted value of the masked region and the corresponding original data of the masked region before the mask is applied.

[0040] Specifically, the first step is to determine the number of artillery pieces that need to be addressed. n It originates from the statistics of VSP common receiver point gathers containing missing shot data in the actual work area. It determines the scale of missing data in subsequent simulations, so that the training task is strictly aligned with the actual reconstruction needs.

[0041] Then, based on the number of missing guns n Training samples are constructed using the original, complete VSP co-detector point gather. Specifically, the complete VSP co-detector point gather is divided into multiple segments with a width of [missing information]. W Slice the data, and set a width of [value] in the center area of ​​each slice.n The mask, the masked region in the slice can be represented as: (1) in, =1 indicates the first i One location was obscured. =0 indicates that the image is not occluded. A large number of slices processed in this way constitute the sample set required for self-supervised training, where the occluded parts are the targets that the network needs to learn to predict.

[0042] Finally, the training objective and optimization process are defined. The training objective is for the model to learn to predict the data of the masked portion based on the unmasked context information surrounding the masked region. This objective is quantified as minimizing a reconstruction error, which is the difference between the model's predicted value for the masked region and the original true data corresponding to that region before the mask was applied. This difference is typically calculated using the L1 loss function, as shown below: (2) in, This represents the original data of the masked region before the mask is applied. This represents the predicted value of the masked region. N This represents the number of training samples in a batch. During training, based on this loss value, the gradient of the model parameters is calculated using the backpropagation algorithm, and optimization algorithms such as Adam are used to update the parameters, driving continuous optimization of the model parameters until the loss function tends to be minimized (e.g., ...). Figure 4 This is a schematic diagram showing how the loss function value changes with the number of training rounds, as provided in an embodiment of the present invention, thereby obtaining the optimal training model that can accurately reconstruct the mask region (i.e., simulate missing shots).

[0043] Building upon the aforementioned embodiments, training samples are constructed directly using complete gathers within the same work area through masking operations that simulate actual missing shot situations. This eliminates the need for any manual annotation, significantly reducing data preparation costs and time. Simultaneously, this "mask-prediction" training mechanism forces the model to learn and infer the inherent data patterns of the missing parts from known context, thereby enabling it to achieve accurate reconstruction capabilities for real-world missing shot scenarios. This approach not only improves training efficiency but also enhances the model's adaptability to specific work area data characteristics and the reliability of reconstruction results.

[0044] Figure 5 This is a flowchart illustrating another method for reconstructing missing VSP data based on a self-supervised deep learning model, provided in an embodiment of the present invention. Figure 6a This is a schematic diagram of a VSP common receiver point gather containing missing shot data provided in an embodiment of the present invention. Figure 6b for Figure 6aA schematic diagram of the corresponding reconstructed complete VSP common receiver gather. To more clearly illustrate the embodiments of the present invention, the following is combined with... Figures 5 to 6b A detailed explanation of the VSP missing data reconstruction method based on a self-supervised deep learning model is provided: First, acquire the VSP data for the target work area and extract it into the CRP co-receiver point gather. Filter the CRP gather to separate two types of data: one type is missing trace data (i.e., VSP co-receiver point gathers containing missing shot data, such as...). Figure 6a (As shown), record the number of missing shots. n The other type is complete trace data (i.e., the original complete VSP common detector point gather).

[0045] Then, data slices are created using the complete data stream. Based on these slices, a masking operation is performed, where the mask width is... n (The number of missing channels is consistent with the number of missing channels) and the position is located in the middle of the slice, so as to simulate the missing shot state and build a training sample set; based on the training sample set, a convolutional neural network model is trained through self-supervised learning, and finally a trained network model is obtained.

[0046] Finally, the missing trace data is input into the trained model, and the predicted result, i.e., the reconstructed complete VSP co-detector point gather, is output, such as... Figure 6b As shown.

[0047] In summary, this invention utilizes complete VSP common receiver point gathers from the target work area to construct a training set, and trains it using an optimized convolutional neural network, thereby achieving effective reconstruction of missing shot data. Experimental results show that this method can significantly improve the interpolation reconstruction quality of missing VSP traces.

[0048] Figure 7 A schematic diagram of a VSP missing data reconstruction device based on a self-supervised deep learning model provided in an embodiment of the present invention is shown below. Figure 7 As shown, the VSP missing data reconstruction device based on a self-supervised deep learning model includes: Input module 701 is used to input VSP common resonator point gather containing missing shot data into the deep convolutional neural network model; Reconstruction module 702 is used to reconstruct the missing shot data using the deep convolutional neural network model and output the reconstructed complete VSP common receiver point gather; The deep convolutional neural network model is obtained by training the original complete VSP common detection point gather in the target work area through a self-supervised learning method.

[0049] In some embodiments, the apparatus further includes a training module 703 for training a deep convolutional neural network model through a self-supervised learning method: Determine the number of missing shots in the VSP common receiver point set containing missing shot data; The original complete VSP common receiver point gather is sliced, and a mask is set in the central region of each slice. The width of the mask is consistent with the number of missing shots to form a mask region and simulate the missing shot state, thereby constructing a training sample set. The training is performed based on the training sample set with the goal of minimizing the difference between the predicted value of the deep convolutional neural network model for the masked region and the corresponding original data of the masked region before the mask is applied.

[0050] In some embodiments, the deep convolutional neural network model is an encoder-decoder structure, and the encoder and the decoder perform feature fusion through skip connections.

[0051] In some embodiments, the encoder includes four downsampling stages, each of which includes two convolutional layers; the decoder includes four upsampling stages, each of which includes a transposed convolutional layer and two convolutional layers; and the output of each downsampling stage in the encoder is fused with the input of the corresponding upsampling stage in the decoder via skip connections.

[0052] In some embodiments, the convolutional layer and the transposed convolutional layer used for feature extraction in the encoder and the decoder both employ 5×5 convolutional kernels.

[0053] In some embodiments, a connection module is provided on the jump connection path between the encoder and the decoder. The connection module includes a first submodule and a second submodule. The output of the second submodule is added to the output of the first submodule through a residual connection.

[0054] In some embodiments, the first submodule includes a 5×5 convolutional layer and a ReLU activation function; the second submodule includes two consecutive 5×5 convolutional layers, and each of the 5×5 convolutional layers is followed by a ReLU activation function.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process and corresponding beneficial effects of the VSP missing data reconstruction device based on the self-supervised deep learning model described above can be referred to the corresponding process in the aforementioned method examples, and will not be repeated here.

[0056] like Figure 8 As shown, this embodiment of the invention provides an electronic device, including a processor 801, a communication interface 802, a memory 803, and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other via the communication bus 804. Memory 803 is used to store computer programs; In one embodiment of the present invention, when the processor 801 executes the program stored in the memory 803, it implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model provided in any of the foregoing method embodiments.

[0057] The electronic device provided in this embodiment of the invention has a similar implementation principle and technical effect to the above embodiments, and will not be described again here.

[0058] The aforementioned memory 803 can be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 803 has storage space for program code used to perform any of the method steps described above. For example, the storage space for program code may include individual program codes for implementing the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, optical discs (CDs), memory cards, or floppy disks. Such computer program products are typically portable or fixed storage units. The storage unit may have storage segments or storage spaces arranged similarly to memory 803 in the aforementioned electronic device. The program code may be compressed, for example, in a suitable form. Typically, the storage unit includes programs for performing the method steps according to embodiments of the invention, i.e., code that can be read by a processor such as 801, which, when run by the electronic device, causes the electronic device to perform the various steps in the methods described above.

[0059] Embodiments of the present invention also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model as described above.

[0060] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments; or it may exist independently and not assembled into the device / apparatus. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of the present invention.

[0061] According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0063] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for reconstructing missing VSP data based on a self-supervised deep learning model, characterized in that, include: Input the VSP common receiver point gather containing missing shot data into the deep convolutional neural network model; The missing shot data is reconstructed using the deep convolutional neural network model, and the reconstructed complete VSP common receiver point gather is output. The deep convolutional neural network model is obtained by training the original complete VSP common detection point gather in the target work area through a self-supervised learning method.

2. The method according to claim 1, characterized in that, The deep convolutional neural network model is trained using the following self-supervised learning method: Determine the number of missing shots in the VSP common receiver point set containing missing shot data; The original complete VSP common receiver point gather is sliced, and a mask is set in the central region of each slice. The width of the mask is consistent with the number of missing shots to form a mask region and simulate the missing shot state, thereby constructing a training sample set. The training is performed based on the training sample set with the goal of minimizing the difference between the predicted value of the deep convolutional neural network model for the masked region and the corresponding original data of the masked region before the mask is applied.

3. The method according to claim 1, characterized in that, The deep convolutional neural network model is an encoder-decoder structure, and the encoder and the decoder perform feature fusion through skip connections.

4. The method according to claim 3, characterized in that, The encoder includes four downsampling stages, each of which contains two convolutional layers; the decoder includes four upsampling stages, each of which contains one transposed convolutional layer and two convolutional layers; and the output of each downsampling stage in the encoder is fused with the input of the corresponding upsampling stage in the decoder via skip connections.

5. The method according to claim 4, characterized in that, In both the encoder and the decoder, the convolutional layer and the transposed convolutional layer used for feature extraction both employ 5×5 convolutional kernels.

6. The method according to any one of claims 3-5, characterized in that, A connection module is provided on the jump connection path between the encoder and the decoder. The connection module includes a first submodule and a second submodule. The output of the second submodule is added to the output of the first submodule through a residual connection.

7. The method according to claim 6, characterized in that, The first submodule includes a 5×5 convolutional layer and a ReLU activation function; the second submodule includes two consecutive 5×5 convolutional layers, and each of the 5×5 convolutional layers is followed by a ReLU activation function.

8. A VSP missing data reconstruction device based on a self-supervised deep learning model, characterized in that, include: The input module is used to input the VSP common resonator gather containing missing shot data into the deep convolutional neural network model; The reconstruction module is used to reconstruct the missing shot data using the deep convolutional neural network model and output the reconstructed complete VSP common receiver point gather; The deep convolutional neural network model is obtained by training the original complete VSP common detection point gather in the target work area through a self-supervised learning method.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the VSP missing data reconstruction method based on a self-supervised deep learning model as described in any one of claims 1-7.

Images