Sparse time-frequency spectrum analysis method, model, device and medium based on autoencoder

Abstract

The present application provides a sparse time-frequency spectrum analysis method, model, device and medium based on an autoencoder, wherein the method is based on an encoder of a sparse time-frequency analysis network model of a Unet network, extracts the seismic signal characteristics of the seismic signal, and converts the seismic signal characteristics into a seismic time-frequency spectrum matrix; based on a mixed norm constraint, the seismic time-frequency spectrum matrix is ​​subjected to sparse constraints and smooth constraints; a decoder of a sparse time-frequency analysis network model based on a Unet network decodes the seismic time-frequency spectrum matrix, generates a seismic time-frequency spectrum corresponding to the seismic signal, and performs time-frequency analysis based on the seismic time-frequency spectrum. In the above manner, the present application introduces a deep learning network model to calculate the sparse time-frequency spectrum of seismic data, avoids a large number of manual parameter selection experiments in the traditional calculation process, and improves the efficiency of time-frequency analysis; at the same time, by utilizing the unsupervised idea, it no longer relies on data labels, breaking through the problem that actual seismic data labels are difficult to obtain.

Description

Technical Field

[0001] The present application relates to the field of geophysical exploration technology, and in particular to a sparse time-frequency spectrum analysis method, model, computer device, and computer-readable storage medium based on an autoencoder. Background Art

[0002] Time-spectral analysis (TSA) maps one-dimensional seismic signals into a two-dimensional time-spectral space, capturing localized features of the signals. It has been widely used in seismic signal processing and interpretation. Common TSA methods include short-time Fourier transforms, S-transforms, and wavelet transforms. However, due to the uncertainty principle, these commonly used TSA methods have limited time-frequency resolution, making them incapable of accurately characterizing reservoir heterogeneity.

[0003] In recent years, scholars have proposed many improved time-frequency spectrum analysis methods to improve the resolution of time-frequency spectrum analysis methods. These methods include synchronous squeezing transform, sparse time-frequency transform, etc. Among them, sparse time-frequency transform is based on the principle of sparse representation to express the solution of time-frequency coefficients as an inverse problem, and solves it through an optimized iterative algorithm. This method can effectively improve the resolution of time-frequency spectrum analysis methods. However, traditional sparse time-frequency spectrum analysis methods need to rely on known mathematical models, resulting in the inability to obtain accurate time-frequency spectrum analysis in complex geological environments; at the same time, traditional sparse time-frequency spectrum analysis methods require a lot of computing time, making it impossible to apply to large-scale three-dimensional seismic data. Summary of the Invention

[0004] The present application provides a sparse time-frequency spectrum analysis method, model, computer device and computer-readable storage medium based on an autoencoder to solve the technical problem of low accuracy of time-frequency spectrum analysis.

[0005] In a first aspect, an embodiment of the present application provides a sparse time-frequency spectrum analysis method based on an autoencoder, comprising:

[0006] An encoder based on a sparse time-frequency analysis network model of a Unet network extracts seismic signal features of a seismic signal and converts the seismic signal features into a seismic time-frequency spectrum matrix;

[0007] Based on the mixed norm constraint, the seismic time-frequency spectrum matrix is ​​subjected to sparse constraint and smooth constraint;

[0008] A decoder of a sparse time-frequency analysis network model based on a Unet network decodes the earthquake time-frequency spectrum matrix to generate an earthquake time-frequency spectrum corresponding to the earthquake signal, so as to perform time-frequency analysis based on the earthquake time-frequency spectrum.

[0009] Furthermore, the encoder of the sparse time-frequency analysis network model based on the Unet network extracts the seismic signal features of the seismic signal and converts the seismic signal features into a seismic time-frequency spectrum matrix, including:

[0010] The deep learning framework is used to construct the sparse time-frequency analysis network model based on the Unet network.

[0011] Furthermore, the use of a deep learning framework to construct the Unet network-based sparse time-frequency analysis network model includes:

[0012] Based on the Unet network, the encoder for mapping the one-dimensional seismic signal to the two-dimensional time-frequency spectrum is constructed, and the decoder is constructed according to the inverse transform formula of the short-time Fourier transform.

[0013] Furthermore, the encoder of the sparse time-frequency analysis network model based on the Unet network, before extracting the seismic signal features of the seismic signal and converting the seismic signal features into a seismic time-frequency spectrum matrix, further includes:

[0014] The initial model is iteratively trained based on a training set of unlabeled seismic data, and the trained initial model is verified based on a validation set of seismic signals until the learning rate, number of iterations, and / or loss function value of the verified initial model reach a preset threshold, and training is stopped.

[0015] Furthermore, based on the mixed norm constraint, before the sparse constraint and the smooth constraint are applied to the earthquake time-frequency spectrum matrix, the following steps are included:

[0016] The mixed norm constraint is constructed according to a mixed norm formula, wherein the mixed norm formula is:

[0017]

[0018] Among them, ||·||1 and are sparse constraints and smooth constraints, respectively, ||·||1 and are l1 norm and l2 norm respectively, f is the earthquake signal, Ψ θ It is an encoder with parameters, λ1 and λ2 are used to control the weights of sparsity and smoothness, respectively, acting on the l1 norm and l2 norm.

[0019] Furthermore, the calculation formula of the loss function value is:

[0020]

[0021] in, is the reconstruction loss function.

[0022] Furthermore, after a decoder of a sparse time-frequency analysis network model based on a Unet network decodes the earthquake time-frequency spectrum matrix to generate an earthquake time-frequency spectrum corresponding to the earthquake signal, and performs time-frequency analysis based on the earthquake time-frequency spectrum, the method further includes:

[0023] Calculate and extract earthquake time-frequency spectra of different frequencies, and use RGB tools to analyze and fuse the earthquake time-frequency spectra of different frequencies into multiple components to obtain RGB fusion results, so as to analyze the geological structure corresponding to the earthquake time-frequency spectra according to the RGB fusion results.

[0024] In a second aspect, the present application further provides a sparse time-frequency spectrum analysis model based on an autoencoder, wherein the sparse time-frequency spectrum analysis model based on an autoencoder comprises:

[0025] An encoding module, configured to extract seismic signal features from a seismic signal using an encoder based on a sparse time-frequency analysis network model of a Unet network, and convert the seismic signal features into a seismic time-frequency spectrum matrix;

[0026] A constraint module, used for performing sparse constraint and smooth constraint on the earthquake time-frequency spectrum matrix based on mixed norm constraint;

[0027] A decoding module is used for a decoder of a sparse time-frequency analysis network model based on a Unet network to decode the earthquake time-frequency spectrum matrix and generate an earthquake time-frequency spectrum corresponding to the earthquake signal so as to perform time-frequency analysis based on the earthquake time-frequency spectrum.

[0028] In a third aspect, the present application further provides a computer device comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the sparse time-frequency spectrum analysis method described above when executing the computer program.

[0029] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the sparse time-frequency spectrum analysis method as described above.

[0030] Compared with the prior art, the sparse time-frequency spectrum analysis method based on the autoencoder provided in the embodiment of the present application is based on the encoder of the sparse time-frequency analysis network model of the Unet network, extracts the seismic signal characteristics of the seismic signal, and converts the seismic signal characteristics into a seismic time-frequency spectrum matrix; based on the mixed norm constraint, the seismic time-frequency spectrum matrix is ​​sparsely constrained and smoothed; the decoder of the sparse time-frequency analysis network model based on the Unet network decodes the seismic time-frequency spectrum matrix, generates the seismic time-frequency spectrum corresponding to the seismic signal, and performs time-frequency analysis based on the seismic time-frequency spectrum. In the above manner, the present application introduces a deep learning network model to calculate the sparse time-frequency spectrum of seismic data, avoids a large number of manual parameter selection experiments in the traditional calculation process, and improves the efficiency of time-frequency analysis; at the same time, it uses the idea of ​​unsupervised learning and no longer relies on data labels, breaking through the problem that actual seismic data labels are difficult to obtain, and successfully applies it to identifying the structure of underground reservoirs.

[0031] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the implementation methods of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0033] Figure 1 A schematic diagram of the process of the sparse time-frequency spectrum analysis method provided in the embodiment of the present application;

[0034] Figure 2 This is a schematic diagram of the network structure of the sparse time-frequency analysis network model of the embodiment of the present application;

[0035] Figure 3 This is a schematic diagram of unlabeled seismic data division in an embodiment of the present application;

[0036] Figure 4 This is a diagram showing the effect of unlabeled seismic data in an embodiment of the present application;

[0037] Figure 5 25Hz frequency slice diagram of the spectrum of different algorithms in the embodiments of this application;

[0038] Figure 6 RGB slice diagrams of different algorithms in the embodiments of this application;

[0039] Figure 7 A schematic block diagram of a sparse time-frequency spectrum analysis model provided in an embodiment of the present application;

[0040] Figure 8 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0042] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0043] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0044] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity or execution order, and words such as "first" and "second" do not necessarily mean different.

[0045] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0046] The inventors of this application have discovered that the current sparse time-frequency transform, which expresses the time-frequency coefficient solution as an inverse problem based on the principle of sparse representation and solves it through an optimized iterative algorithm, can effectively improve the resolution of time-frequency spectrum analysis methods. However, traditional sparse time-frequency spectrum analysis methods rely on known mathematical models, which makes it difficult to obtain accurate time-frequency spectrum analysis in complex geological environments. Furthermore, traditional sparse time-frequency spectrum analysis requires a lot of computation time, making it unsuitable for large-scale 3D seismic data.

[0047] In order to solve the above problems, the present application provides a sparse time-frequency spectrum analysis method based on an autoencoder.

[0048] See Figure 1, Figure 1 This is a flow chart of a sparse time-frequency spectrum analysis method provided in an embodiment of the present application. The sparse time-frequency spectrum analysis method is applied to a sparse time-frequency spectrum analysis model based on an autoencoder, and includes steps S101-S103.

[0049] Step S101: An encoder based on a sparse time-frequency analysis network model of a Unet network extracts seismic signal features of a seismic signal and converts the seismic signal features into a seismic time-frequency spectrum matrix.

[0050] In this embodiment, the encoder of the sparse time-frequency analysis network model based on the Unet network can be regarded as a nonlinear mapping from a one-dimensional signal matrix to a two-dimensional time-frequency spectrum matrix. The encoder is used to perform two steps: feature extraction and feature mapping. First, a seismic signal is input to the encoder. The feature extraction structure in the encoder first converts the corresponding seismic signal into a higher-dimensional seismic signal feature through a dimensionality increase operation, that is, a one-dimensional convolutional network with a kernel size of 3 is used to map the input one-dimensional seismic signal to a two-dimensional seismic signal feature matrix, and then a two-dimensional convolutional network with a kernel size of 3x3 is used to extract features from the two-dimensional seismic matrix, that is, Figure 2 After extracting the effective seismic signal features, the seismic signal features are further processed into the real and imaginary parts of the corresponding time-frequency spectrum, that is, Figure 2 The second half of the structure is to reduce the dimension of the extracted seismic signal features through a deconvolution network with a kernel size of 3x3, and map the high-dimensional features into the seismic time-frequency spectrum matrix. Figure 2 The part before the first Concatenate symbol from left to right is called the front part, and the part after it is called the back part; that is, the shortest block in the figure and all the parts to its left are called the front part; the part to its right is called the back half.

[0051] Nonlinear ReLU functions are used in all convolution operations in the encoder, so that the encoder has a stronger ability to fit nonlinear transformations.

[0052] Furthermore, step S101 specifically includes:

[0053] The deep learning framework is used to construct the sparse time-frequency analysis network model based on the Unet network.

[0054] In this embodiment, the classic Unet model is modified and combined with the sparse time-frequency analysis scenario to build a sparse time-frequency analysis network (SparseTFNet) model. The entire sparse time-frequency analysis network (SparseTFNet) can be divided into two parts: encoder and decoder. The encoder network structure is as follows: Figure 2As shown; wherein the decoder is composed of an inversion formula, and the inversion formula is as follows:

[0055] Sparse time-frequency transform is proposed based on short-time Fourier transform. For the seismic signal f(n), short-time Fourier transform can be defined as follows:

[0056]

[0057] Where m and k are discrete time samples and discrete frequency samples respectively. On this basis, the inverse problem of short-time Fourier transform can be expressed as

[0058]

[0059] Written in matrix form as

[0060] f=Gm (3)

[0061] in, is a vector of f(n), is the dictionary generated by g(n). Therefore, based on the theory of compressed sensing, the unknown time-frequency coefficients can be described as the following inverse problem:

[0062]

[0063] in is the two-norm, and respectively represents the regularization term.

[0064] In order to make the video sparse enough, the sparse regularization term is added to formula (4), that is,

[0065]

[0066] Here ||m||0 represents the l0 norm, but it is actually a non-convex function and very difficult to solve. In order to simplify this model, l1 is widely used.

[0067] Furthermore, step S101 specifically includes:

[0068] The initial model is iteratively trained based on a training set of unlabeled seismic data, and the trained initial model is verified based on a validation set of seismic signals until the learning rate, number of iterations, and / or loss function value of the verified initial model reach a preset threshold, and training is stopped.

[0069] In this embodiment, the unlabeled seismic data is used to construct a dataset for the SparseTFNet model. The dataset is divided into a training set, a validation set, and a test set. The data is then normalized, shuffled, and randomly flipped. The training set uses 5,000 real seismic data. Figure 3 The diagram shows the data set partitioning (the shaded area is divided for training, from which the training set and validation set are extracted, and the remaining area is used as the test set). Another part of the test set is used for evaluation when calculating the sparse time-frequency spectrum of the signal.

[0070] The SparseTFNet model is trained using a training set of actual data and verified using a validation set. The appropriate learning rate, number of iterations, and loss function are set, and an early stopping mechanism is used to prevent overfitting of the model.

[0071] Since the Adam optimization algorithm can accelerate convergence, it is used as the optimizer of the SparseTFNet model. It draws on the unsupervised algorithm to use the encoder to convert the signal into a time-frequency spectrum, and uses the decoder to convert the time-frequency spectrum back into a signal. In this process, the reconstruction error can be obtained. In addition, we need to impose sparse constraints and smooth constraints on the time-frequency spectrum obtained by the model. The loss function of the SparseTFNet model can be summarized as:

[0072]

[0073] The first part is the reconstruction loss, which means that a signal is transformed into a new signal after passing through the encoder and decoder. Through the reconstruction loss, we can ensure that the time-frequency spectrum we obtain is one-to-one corresponding to the signal. The second half is our hybrid constraint term, which is used to impose sparse and smooth constraints on the time-frequency spectrum. Through the loss function formula (6), we obtain a tool that can be used for actual seismic signal analysis and processing.

[0074] After training the SparseTFNet model, select another real data from the actual data set to verify the effect of SparseTFNet. Figure 4 (a) A real seismic data track randomly sampled from a real data set, Figure 4 (b) Time-frequency spectrum results of traditional short-time Fourier transform (STFT), Figure 4 (c) Sparse time-frequency spectrum results of traditional sparse short-time Fourier transform (sparse STFT), Figure 4 (d) The sparse time-frequency spectrum predicted by the Sparse TFNet model. It can be seen that the results calculated by the coefficient time-frequency analysis method are even sparser.

[0075] Furthermore, step S101 specifically includes:

[0076] The mixed norm constraint is constructed according to a mixed norm formula, wherein the mixed norm formula is:

[0077]

[0078] Among them, ||·||1 and are sparse constraints and smooth constraints, respectively, ||·||1 and are l1 norm and l2 norm respectively, f is the earthquake signal, Ψ θ It is an encoder with parameters, λ1 and λ2 are used to control the weights of sparsity and smoothness, respectively, acting on the l1 norm and l2 norm.

[0079] In this embodiment, if the signal is directly used to train the encoder and decoder structures, the resulting time-frequency spectrum may not necessarily meet the sparsity requirement. In addition, to ensure the smoothness of the time-frequency spectrum, we introduce the l1 norm and l2 norm to constrain the obtained time-frequency spectrum. The hybrid norm we constructed can be summarized as:

[0080]

[0081] where ||·||1 and represents the sparse constraint and the smooth constraint, that is, the l1 norm and the l2 norm. In addition, f is a signal, Ψ θ It is an encoder with parameters. When the signal is input into the encoder, the time spectrum is obtained; λ1λ2 are two weights used to control sparsity and smoothness, acting on the l1 norm and l2 norm respectively.

[0082] Step S102: Based on the mixed norm constraint, the earthquake time-frequency spectrum matrix is ​​subjected to sparse constraint and smooth constraint.

[0083] In this embodiment, a mixed norm is used to perform sparse constraints and smooth constraints on the earthquake time-frequency spectrum matrix. The l1 norm is used to ensure sparsity, and the l2 norm is used to ensure the smoothness of the time-frequency spectrum. By setting two different weights for the two different norms, the purpose of adjusting sparsity and smoothness is achieved, and the constrained earthquake time-frequency spectrum matrix is ​​obtained.

[0084] Step S103: A decoder of the sparse time-frequency analysis network model based on the Unet network decodes the earthquake time-frequency spectrum matrix to generate an earthquake time-frequency spectrum corresponding to the earthquake signal, so as to perform time-frequency analysis based on the earthquake time-frequency spectrum.

[0085] In this embodiment, the decoder in the sparse time-frequency analysis network (SparseTFNet) model is used to decode the constrained seismic time-frequency spectrum matrix, that is, it is used for time-frequency analysis of three-dimensional post-stack seismic data: the trained sparse time-frequency analysis network (SparseTFNet) model is used to perform an overall prediction of the three-dimensional actual test set seismic data, and frequency slices of 25Hz, 40Hz, and 60Hz are extracted.

[0086] Figure 5 (a) to (c) show: (a) 25 Hz frequency slices calculated using traditional short-time Fourier transform, (b) 25 Hz frequency slices calculated using traditional sparse short-time Fourier transform, and (c) 25 Hz frequency slices predicted using the Sparse TFNet model. As can be seen, the 25 Hz frequency slices calculated using the Sparse TFNet model provide a clearer depiction of the river channel.

[0087] Furthermore, step S103 specifically includes:

[0088] Calculate and extract earthquake time-frequency spectra of different frequencies, and use RGB tools to analyze and fuse the earthquake time-frequency spectra of different frequencies into multiple components to obtain RGB fusion results, so as to analyze the geological structure corresponding to the earthquake time-frequency spectra according to the RGB fusion results.

[0089] In this embodiment, the results of 25Hz, 40Hz, and 60Hz calculated using three methods are used to form RGB channels. The results are as follows: Figure 6 As shown in Figure 6(a), 6(b), and 6(c), respectively, represent RGB slices composed of frequency slices calculated by traditional short-time Fourier transform, RGB slices composed of frequency slices calculated by traditional sparse short-time Fourier transform, and RGB slices composed of frequency slices predicted by the Sparse TFNet model. It can be seen that the Sparse TFNet model method is clearer than the previous two methods and can provide more river channel details.

[0090] Thus, through the above method:

[0091] In the traditional autoencoder principle, a decoder with physical meaning is introduced to make it have strict physical meaning; in the loss function, a mixed norm constraint is introduced to make the time-frequency spectrum coefficients more sparse. The present invention uses a deep learning network model to calculate the time-frequency spectrum of seismic data and perform time-frequency analysis. On the one hand, it reduces the subjectivity and uncertainty of manual parameter selection in the traditional calculation process, and avoids a large number of parameter selection experiments, thereby increasing operating efficiency; on the other hand, by introducing an unsupervised method, the present invention introduces an inverse problem to solve the dependence on labels in deep learning, so that our model can effectively perform time-frequency analysis when facing actual data. When using the network model for time-frequency analysis, our method is obviously clearer than the traditional method, and is more conducive to subsequent time-frequency analysis work, and has high practical significance and practical value.

[0092] In addition, an embodiment of the present invention also provides a sparse time-frequency spectrum analysis model based on an autoencoder.

[0093] See also Figure 7 , Figure 7 The embodiment of the present application provides a schematic block diagram of a sparse time-frequency spectrum analysis model based on an autoencoder.

[0094] like Figure 7 As shown, the sparse time-frequency spectrum analysis model based on the autoencoder includes:

[0095] An encoding module 10 is used for extracting seismic signal features of a seismic signal based on a sparse time-frequency analysis network model of a Unet network and converting the seismic signal features into a seismic time-frequency spectrum matrix;

[0096] A constraint module 20 is used to perform sparse constraints and smooth constraints on the earthquake time-frequency spectrum matrix based on mixed norm constraints;

[0097] The decoding module 30 is used as a decoder of the sparse time-frequency analysis network model based on the Unet network to decode the earthquake time-frequency spectrum matrix and generate the earthquake time-frequency spectrum corresponding to the earthquake signal so as to perform time-frequency analysis based on the earthquake time-frequency spectrum.

[0098] Furthermore, the sparse time-frequency spectrum analysis model based on the autoencoder further includes:

[0099] Model building module, used to build the sparse time-frequency analysis network model based on the Unet network using a deep learning framework

[0100] The model training module is used to iteratively train the initial model based on a training set of unlabeled seismic data, and verify the trained initial model based on a verification set of seismic signals until the learning rate, number of iterations and / or loss function value of the verified initial model reaches a preset threshold, and then stop training.

[0101] The geological analysis module is used to calculate and extract the time-frequency spectra of earthquakes of different frequencies, and analyze and fuse the time-frequency spectra of earthquakes of different frequencies through the RGB tool to obtain RGB fusion results, so as to analyze the geological structure corresponding to the time-frequency spectra of earthquakes according to the RGB fusion results.

[0102] Furthermore, the model building module specifically includes:

[0103] The model construction unit is used to construct the encoder for mapping the one-dimensional seismic signal to the two-dimensional time-frequency spectrum based on the Unet network, and to construct the decoder according to the inverse transformation formula of the short-time Fourier transform.

[0104] Furthermore, the constraint module specifically includes:

[0105] A norm editing unit is used to construct the mixed norm constraint according to a mixed norm formula, wherein the mixed norm formula is:

[0106]

[0107] Among them, ||·||1 and are sparse constraints and smooth constraints, respectively, ||·||1 and are l1 norm and l2 norm respectively, f is the earthquake signal, Ψ θ It is an encoder with parameters, λ1 and λ2 are used to control the weights of sparsity and smoothness, respectively, acting on the l1 norm and l2 norm.

[0108] It should be noted that those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned control method embodiments and will not be repeated here.

[0109] See also Figure 8 , Figure 8 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0110] See Figure 8 The computer device includes a processor, a memory and a network interface connected through a model bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0111] The non-volatile storage medium can store an operation model and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any sparse time-frequency spectrum analysis method based on the autoencoder.

[0112] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0113] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any sparse time-frequency spectrum analysis method based on the autoencoder.

[0114] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0115] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0116] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and the computer program includes program instructions. The processor executes the program instructions to implement any one of the sparse time-frequency spectrum analysis methods based on the autoencoder provided in the embodiments of the present application.

[0117] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc., equipped on the computer device.

[0118] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A sparse time-frequency spectrum analysis method based on an autoencoder, characterized in that: The sparse time-frequency spectrum analysis method based on the autoencoder includes: A deep learning framework is used to build a sparse time-frequency analysis network model based on the Unet network, including: Based on the Unet network, an encoder is constructed to map the one-dimensional seismic signal to a two-dimensional time-frequency spectrum, and a decoder is constructed based on the inverse transform formula of the short-time Fourier transform. The encoder based on the sparse time-frequency analysis network model of the Unet network extracts the seismic signal features of the seismic signal and converts the seismic signal features into a seismic time-frequency spectrum matrix; The mixed norm constraint is constructed according to the mixed norm formula, wherein the mixed norm formula is: Among them, ||·||1 and are sparse constraints and smooth constraints, respectively, ||·||1 and are l1 norm and l2 norm respectively, f is the earthquake signal, Ψ θ It is an encoder with parameters, λ1 and λ2 are used to control the weights of sparsity and smoothness, respectively, acting on the l1 norm and l2 norm; Based on the mixed norm constraint, sparse constraint and smooth constraint are performed on the earthquake time-frequency spectrum matrix; A decoder of a sparse time-frequency analysis network model based on a Unet network decodes the earthquake time-frequency spectrum matrix to generate an earthquake time-frequency spectrum corresponding to the earthquake signal, so as to perform time-frequency analysis based on the earthquake time-frequency spectrum.

2. The sparse time-frequency spectrum analysis method according to claim 1, characterized in that: The encoder of the sparse time-frequency analysis network model based on the Unet network, before extracting the seismic signal features of the seismic signal and converting the seismic signal features into a seismic time-frequency spectrum matrix, further includes: The initial model is iteratively trained based on a training set of unlabeled seismic data, and the trained initial model is verified based on a validation set of seismic signals until the learning rate, number of iterations, and / or loss function value of the verified initial model reach a preset threshold, and training is stopped.

3. The sparse time-frequency spectrum analysis method according to claim 2, characterized in that: The calculation formula of the loss function value is: in, is the reconstruction loss function.

4. The sparse time-frequency spectrum analysis method according to any one of claims 1 to 3, characterized in that: The decoder of the sparse time-frequency analysis network model based on the Unet network decodes the earthquake time-frequency spectrum matrix to generate the earthquake time-frequency spectrum corresponding to the earthquake signal, and performs time-frequency analysis based on the earthquake time-frequency spectrum, further comprising: Calculate and extract earthquake time-frequency spectra of different frequencies, and use RGB tools to analyze and fuse the earthquake time-frequency spectra of different frequencies into multiple components to obtain RGB fusion results, so as to analyze the geological structure corresponding to the earthquake time-frequency spectra according to the RGB fusion results.

5. A sparse time-frequency spectrum analysis model based on an autoencoder, characterized in that: The sparse time-frequency spectrum analysis model based on the autoencoder includes: An encoding module, configured to extract seismic signal features from a seismic signal using an encoder based on a sparse time-frequency analysis network model of a Unet network, and convert the seismic signal features into a seismic time-frequency spectrum matrix; A constraint module, configured to perform sparse constraint and smooth constraint on the earthquake time-frequency spectrum matrix based on the mixed norm constraint; A decoding module is used for decoding the earthquake time-frequency spectrum matrix based on the sparse time-frequency analysis network model of the Unet network to generate the earthquake time-frequency spectrum corresponding to the earthquake signal, so as to perform time-frequency analysis according to the earthquake time-frequency spectrum; Among them, the deep learning framework is used to build a sparse time-frequency analysis network model based on the Unet network, including: Based on the Unet network, an encoder is constructed to map the one-dimensional seismic signal to a two-dimensional time-frequency spectrum, and a decoder is constructed based on the inverse transform formula of the short-time Fourier transform. The mixed norm constraint is constructed according to a mixed norm formula, wherein the mixed norm formula is: Among them, ||·||1 and are sparse constraints and smooth constraints, respectively, ||·||1 and are l1 norm and l2 norm respectively, f is the earthquake signal, Ψ θ It is an encoder with parameters, λ1 and λ2 are used to control the weights of sparsity and smoothness, respectively, acting on the l1 norm and l2 norm.

6. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and implement the sparse time-frequency spectrum analysis method according to any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to implement the sparse time-frequency spectrum analysis method according to any one of claims 1 to 4.

Images