Adaptive denoising and intelligent optimization processing system and method for geophysical exploration data

Abstract

The application discloses a kind of geophysical exploration data adaptive denoising and intelligent optimization processing system and method, it is related to geophysical exploration data processing technical field.This method first collects geophysical exploration data and carries out pretreatment;Geophysical exploration data is carried out multi-scale morphological decomposition;Hankel embedding is executed to multi-scale morphological matrix and low rank and sparse decomposition are solved to remove impulse noise;Adaptive denoising is carried out to joint low rank matrix and morphological component by constructing gated denoising network;Optimization is carried out to window length, structural element scale, regularization parameter and network weight using swarm intelligence algorithm.System includes data acquisition module, morphological decomposition module, low rank decomposition module, gated denoising network module, intelligent optimization module and control and storage module.The application can automatically select processing parameters under complex noise background and effectively suppress random noise and exogenous interference, improve the signal-to-noise ratio of geophysical exploration data.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration data processing technology, specifically to an adaptive denoising and intelligent optimization processing system and method for geophysical exploration data. Background Technology

[0002] Signals acquired during geophysical exploration are often accompanied by natural noise, human interference, and instrument noise. Noise reduces the signal-to-noise ratio and affects subsequent interpretation and positioning accuracy. For example, existing technologies use filters to suppress noise, but filtering performance significantly decreases when noise and effective signals are in the same frequency band, and filter parameters are difficult to select. Another approach involves transforming the data to the time-frequency domain and thresholding the coefficients to separate the signal from the noise, but finding the optimal threshold is difficult, and inappropriate thresholds can lead to signal attenuation. With the development of deep learning, denoising methods based on deep supervised networks train models on cloud platforms to remove noise sharing the frequency band with the signal, and use sliding windows to convert time-domain sequences into two-dimensional time-frequency sequences to train deep learning networks. However, these methods still rely on fixed windows and pre-defined network architectures, do not fully consider the morphological characteristics of seismic signals at multiple scales, and require large amounts of data and expensive hardware for training.

[0003] To address high-noise data, researchers have proposed an improved ResUnet network, enhancing network stability through residual modules, spectral normalization, and compression-excitation modules. Other methods combine diffusion models with self-attention, achieving seismic data denoising through forward noise addition and inverse denoising. These methods offer advantages in handling complex noise, but still suffer from drawbacks such as high computational cost, strong parameter dependence, and difficulty in preserving morphological integrity. Furthermore, traditional variational mode decomposition (VMD) suppresses random noise by converting two-dimensional noisy signals to the frequency domain and then performing mode decomposition. While this method preserves amplitude, it is inefficient for processing high-dimensional, large-scale data and lacks intelligent parameter optimization.

[0004] Furthermore, mathematical morphology has been introduced into marine seismic data processing to suppress external interference waves. It works by estimating the location of the interference source using time-distance curves, smoothing out external interference, and performing morphological filtering along the interference direction. Although morphological filtering can preserve linear or geometric features, this method is only effective against external interference and is not suitable for random noise or other complex noise types.

[0005] In summary, existing technologies still have shortcomings in terms of parameter adaptability, multi-scale feature extraction, and processing efficiency. There is an urgent need for a denoising and processing system that can adaptively identify different types of noise, take into account both local structure and global trends, and has parameter self-optimization capabilities. Summary of the Invention

[0006] Technical Objective: To address the shortcomings of existing technologies, such as difficulty in selecting filtering parameters, reliance on fixed windows for model training, difficulty in simultaneously protecting local morphology and global trends, and lack of parameter self-optimization capabilities, this invention discloses an adaptive denoising and intelligent optimization processing system and method for geophysical exploration data. This system enables adaptive denoising and intelligent optimization processing of geophysical exploration data, improving the signal-to-noise ratio while preserving the morphological characteristics of seismic signals.

[0007] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution:

[0008] An adaptive denoising and intelligent optimization method for geophysical exploration data includes the following steps:

[0009] Collect geophysical exploration data and perform baseline correction and amplitude normalization;

[0010] The geophysical exploration data is windowed according to adaptive parameters, and the signals of each window are decomposed into multi-scale morphological components by using mathematical morphological erosion, dilation and opening and closing operations and variational mode decomposition.

[0011] The multi-scale morphological components are slidingly embedded to form a Hankel matrix. The Hankel matrix is ​​then subjected to low-rank sparse decomposition to obtain a low-rank matrix and a sparse matrix.

[0012] The low-rank matrix and multi-scale morphological components are input into the gated denoising network. The gated denoising network generates gate weights through the gated layer to adjust the contribution of features at different scales and outputs the denoised signal.

[0013] A swarm intelligence optimization algorithm is used to optimize the window length, morphological structuring element scale, low-rank decomposition regularization parameter, and gated denoising network parameters. The parameters are updated based on the denoised signal-to-noise ratio and structural similarity as the fitness function until the fitness converges.

[0014] The denoised signals from each window are recombined and output as the denoised result.

[0015] Preferably, the low-rank sparse decomposition is achieved by solving an optimization problem. The matrix is ​​obtained, where L is a low-rank matrix, S is a sparse matrix, and M is the original geophysical exploration data matrix. For inverse vectorization operators, The nuclear norm causes L to be low-rank. for This leads to the sparsity of S. It is the Frobenius norm. and These are the low-rank regularization weight and the data fidelity weight, respectively.

[0016] Preferably, the loss function of the gated denoising network is:

[0017] ,

[0018] in For gated denoising networks The complete set of learnable parameters For the morphological components segmented at the k-th scale, The denoising result predicted by the gated denoising network. This is the error weight matrix corresponding to the k-th scale. For Hadamard element-wise multiplication, It is the Frobenius norm. The gating weight for the nth channel is... is the gating entropy regularization coefficient.

[0019] Preferably, the swarm intelligence optimization algorithm is an improved particle swarm optimization algorithm, and its parameter update formula is:

[0020] ,

[0021] in Let be the parameter vector of the i-th particle at time t+1. Let be the parameter vector of the i-th particle at iteration time t. This represents the optimal parameter position achieved by particle i in all iterations so far. This represents the globally optimal parameter position in the entire history of the current population. and As a learning factor, and is the perturbation factor randomly sampled in the interval [0,1].

[0022] This invention also discloses an adaptive denoising and intelligent optimization processing system for geophysical exploration data, used to implement the adaptive denoising and intelligent optimization processing method for geophysical exploration data as described above, including:

[0023] The data acquisition module is used to acquire geophysical exploration data and perform baseline correction and amplitude normalization;

[0024] The morphological decomposition module is used to segment geophysical exploration data according to window parameters and obtain multi-scale morphological components using mathematical morphology operations and variational mode decomposition.

[0025] The low-rank decomposition module is used to convert multi-scale morphological components into Hankel matrices and solve for low-rank and sparse matrices.

[0026] The gated denoising network module is used to receive multi-scale morphological components and low-rank matrices, generate gate weights through a gated layer, and output the denoised signal.

[0027] The intelligent optimization module is used to optimize the window length, morphological structure element scale, low-rank decomposition regularization parameters, and gated denoising network parameters using swarm intelligence optimization algorithms.

[0028] The control and storage module includes a processor and a memory. The processor is used to execute the control logic of the above module, and the memory is used to store program code and intermediate data.

[0029] Preferably, the gated denoising network module includes several convolutional layers, normalization layers, gated layers, and an output layer. The gated layers generate weights through the sigmoid function to control the contribution of features at different scales.

[0030] Preferably, the intelligent optimization module evaluates the signal-to-noise ratio and structural similarity of the denoising result through the fitness evaluation unit, and uses the evaluation result as the fitness function of the swarm intelligence optimization algorithm.

[0031] Beneficial Effects: The adaptive denoising and intelligent optimization processing system and method for geophysical exploration data provided by this invention has the following beneficial effects:

[0032] This invention combines multi-scale morphological operations with variational mode decomposition to extract morphological features of signals at different scales, which can suppress external interference and random noise while taking into account both local details and global trends.

[0033] The improved low-rank sparse decomposition combined with a gated denoising network not only separates impulse noise, but also utilizes a learnable gating mechanism to achieve adaptive feature selection, solving the problems of traditional filtering and deep learning methods requiring preset thresholds or being prone to unstable training.

[0034] By automatically adjusting processing parameters through swarm intelligence optimization algorithms, the system can adaptively adjust window length, morphological structural element scale, and regularization parameters according to the noise type of different exploration areas, thereby improving the denoising effect and reducing reliance on human experience.

[0035] The entire system can be deployed on a local server or in the cloud, and it can efficiently process large-scale 3D seismic data, with good versatility and scalability. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0037] Figure 1This is a flowchart of the adaptive denoising and intelligent optimization processing method for geophysical exploration data according to the present invention;

[0038] Figure 2 This is a diagram illustrating the overall architecture of the adaptive denoising and intelligent optimization processing system for geophysical exploration data of this invention.

[0039] Figure 3 This is a schematic diagram of the adaptive window segmentation of the present invention;

[0040] Figure 4 This is a comparison chart of the low-rank sparse decomposition results of this invention;

[0041] Figure 5 This is a diagram of the gating denoising network architecture of the present invention. Detailed Implementation

[0042] The present invention will be described more clearly and completely below by way of preferred embodiments and in conjunction with the accompanying drawings, but the invention is not limited to the scope of the embodiments described herein.

[0043] like Figure 1 As shown, this invention discloses an adaptive denoising and intelligent optimization processing method for geophysical exploration data, specifically including the following steps:

[0044] Collect geophysical exploration data and perform baseline correction and amplitude normalization;

[0045] The geophysical exploration data is windowed according to adaptive parameters, and the signals of each window are decomposed into multi-scale morphological components by using mathematical morphological erosion, dilation and opening and closing operations and variational mode decomposition.

[0046] The multi-scale morphological components are slidingly embedded to form a Hankel matrix. The Hankel matrix is ​​then subjected to low-rank sparse decomposition to obtain a low-rank matrix and a sparse matrix.

[0047] The low-rank matrix and multi-scale morphological components are input into the gated denoising network. The gated denoising network generates gate weights through the gated layer to adjust the contribution of features at different scales and outputs the denoised signal.

[0048] A swarm intelligence optimization algorithm is used to optimize the window length, morphological structuring element scale, low-rank decomposition regularization parameter, and gated denoising network parameters. The parameters are updated based on the denoised signal-to-noise ratio and structural similarity as the fitness function until the fitness converges.

[0049] The denoised signals from each window are recombined and output as the denoised result.

[0050] This invention also discloses an adaptive denoising and intelligent optimization processing system for geophysical exploration data, used to implement the adaptive denoising and intelligent optimization processing method for geophysical exploration data as described above, including:

[0051] The data acquisition module is used to acquire geophysical exploration data and perform baseline correction and amplitude normalization;

[0052] The morphological decomposition module is used to segment geophysical exploration data according to window parameters and obtain multi-scale morphological components using mathematical morphology operations and variational mode decomposition.

[0053] The low-rank decomposition module is used to convert multi-scale morphological components into Hankel matrices and solve for low-rank and sparse matrices.

[0054] The gated denoising network module is used to receive multi-scale morphological components and low-rank matrices, generate gate weights through a gated layer, and output the denoised signal.

[0055] The intelligent optimization module is used to optimize the window length, morphological structure element scale, low-rank decomposition regularization parameters, and gated denoising network parameters using swarm intelligence optimization algorithms.

[0056] The control and storage module includes a processor and a memory. The processor is used to execute the control logic of the above module, and the memory is used to store program code and intermediate data.

[0057] Example 1

[0058] Reference Figure 2 The adaptive denoising and intelligent optimization processing system for geophysical exploration data provided in this embodiment includes:

[0059] Data acquisition module: Composed of several sensor arrays and acquisition devices, used to acquire raw data from seismic exploration. The acquired raw data is stored in memory after preliminary correction.

[0060] The morphological decomposition module includes an adaptive window segmentation unit and a morphological operator library. The adaptive window segmentation unit segments the original data according to the window length N provided by the intelligent optimization module. The morphological operator library contains structuring elements of different scales for performing erosion, dilation, opening / closing operations, and variational mode decomposition. The resulting multi-scale morphological components M... k Stored in a buffer.

[0061] The low-rank decomposition module includes Hankel embedding units and low-rank sparse decomposition units. The Hankel embedding unit transforms the morphological component matrix M into a Hankel matrix H, and the low-rank sparse decomposition unit solves for the low-rank matrix L and the sparse matrix S, using the augmented Lagrange multiplier algorithm to solve the optimization problem.

[0062] Gated denoising network module: such as Figure 5 As shown, the network includes convolutional layers, normalization layers, gating layers, and an output layer. The network takes L and M as inputs and outputs a denoised signal. The gating layer generates gating weights using the sigmoid function. It is used to adaptively adjust the feature contributions at different scales.

[0063] The intelligent optimization module includes a parameter encoding unit, a fitness evaluation unit, and a parameter update unit. The parameter encoding unit encodes window length, morphological structuring element scale, low-rank decomposition regularization parameters, and gated denoising network parameters into particle position vectors. The fitness evaluation unit constructs a fitness function using indicators such as the signal-to-noise ratio and structural similarity of the output data. The parameter update unit updates the positions of each particle according to a swarm intelligence algorithm until the fitness meets the stopping condition. The final output of the optimal parameters guides the morphological decomposition module, the low-rank decomposition module, and the gated denoising network module.

[0064] Control and storage module: This includes a processor and a memory. The processor executes the control logic of each module, and the memory is used to store program code, model parameters, and intermediate data.

[0065] The system can be deployed on a ground workstation or a cloud server. The processor can be a general-purpose computer or a graphics processing unit with parallel computing capabilities, and the memory can be a non-volatile storage medium.

[0066] Example 2

[0067] The adaptive denoising and intelligent optimization processing method for geophysical exploration data provided in this embodiment includes the following steps:

[0068] Step S101: Data Acquisition and Preprocessing

[0069] The data acquisition module is used to collect raw geophysical exploration data. The raw data is then baseline-corrected and amplitude-normalized to obtain a preprocessed dataset D.

[0070] Step S102: Multi-scale morphological decomposition and adaptive window segmentation

[0071] like Figure 3 As shown, D is divided into sliding segments based on the window length N provided by the intelligent optimization module, and the signal d of each window is divided into segments. i The morphological components at the k-th scale on window i are obtained using mathematical morphological erosion, dilation, and opening / closing operations. Then, each window signal is converted to the frequency domain and subjected to variational mode decomposition to obtain... , representing the frequency domain modes obtained in window i. Calculate the weights. The two types of components are then fused to form a window fusion matrix. , where K is the total number of scales.

[0072] Step S103: Time-frequency hybrid low-rank embedding

[0073] After fusion Perform Hankel embedding to obtain the Hankel matrix. ,in For column-stacked Hankel embedding operators, m and n are determined by the window length N and the embedding order d. The augmented Lagrange multiplier method is used to solve the low-rank sparse decomposition problem. The low-rank matrix is ​​obtained. Where L is a low-rank matrix representing layered, coherent, or steady-state structural signals in geophysical exploration data, and S is a sparse matrix that captures non-stationary components such as impulse noise, anomalies, and short-period interference. As the inverse vectorization operator, it reduces the vector form vec(L+S) to its original form. Matrix of the same shape The nuclear norm causes L to be low-rank. for This leads to the sparsity of S. The Frobenius norm measures the energy of the reconstruction error. and These are the low-rank regularization weight and the data fidelity weight, respectively, which control the intensity of the sparse noise penalty and the balance between the reconstruction error and the regularization term.

[0074] like Figure 4 As shown, this embodiment uses an actual seismic profile as the test object. Through the low-rank sparse decomposition algorithm of this invention, the original noisy data is effectively separated into a low-rank matrix L and a sparse matrix S, and a denoised result L+S is obtained based on this. The comparison shows that the continuity of the reflection interface is significantly enhanced after denoising, demonstrating the superiority of this invention under complex noise conditions.

[0075] Step S104: Adaptive denoising using a learnable gating network

[0076] Will and As input, it is sent to the gate control denoising network. The gated denoising network extracts features through convolutional layers, normalization layers improve training stability, and the gate layer determines the features based on the input characteristics. Generate gate weights, where The sigmoid function is used, W and b are linear projection parameters, and h is the convolutional feature, controlling the flow of different feature channels. The network uses... Let be the loss function, where For gated denoising networks The complete set of learnable parameters Let i be the shape component at the k-th scale on window i. This is the error weight matrix corresponding to the k-th scale. For Hadamard element-wise multiplication, It is the Frobenius norm. The gating weight for the nth channel is... These are the gated entropy regularization coefficients. Training is performed using gradient descent, and after training, these coefficients are used to predict new window data, resulting in the denoised signal. .

[0077] Step S105: Intelligent Parameter Optimization

[0078] Multiple particles are initialized using a swarm intelligence optimization algorithm. Each particle includes a window length, a set of structuring element scales, and regularization parameters. For each particle, denoising results are obtained according to steps S102-S104, and the signal-to-noise ratio and structural similarity are calculated to determine the fitness. The update formula of the particle swarm optimization algorithm is then used.

[0079]

[0080] Update particle positions, where Let be the parameter vector of the i-th particle at time t+1. Let be the parameter vector of the i-th particle at iteration time t. This represents the optimal parameter position achieved by particle i in all iterations so far. This represents the globally optimal parameter position in the entire history of the current population. and As a learning factor, and The perturbation factor is randomly sampled in the interval [0,1]; the optimal parameters are output when the overall fitness converges or the preset number of iterations is reached. The parameter is fed back to steps S102-S104 to achieve adaptive adjustment.

[0081] Step S106: Output the result

[0082] The denoised signal output from all windows splicing and reassembling into the final denoised data Signal-to-noise ratio reports and morphological parameters can be output as needed.

[0083] Example 3

[0084] The following uses a 3D seismic exploration dataset as an example to illustrate the specific meaning of the parameters in this invention. Assume a sampling interval of 2ms, a single trace length of 2000 sampling points, and an exploration range containing 500 shot points. The optimal window length N=128 obtained from the intelligent optimization module, with a window overlap rate of 75%. The morphological operator library includes three structural element scales r={3,5,7}; and low-rank decomposition regularization parameters. , The gated network consists of four convolutional layers, each with a 3×3 kernel size. The gated layer has 16 output channels and a sparse regularization coefficient. The swarm intelligence optimization algorithm uses 40 particles and a maximum of 50 iterations. Experimental results show that the signal-to-noise ratio is improved by about 6-8 dB after denoising, and the fault boundaries and weak reflection layers are significantly enhanced. Compared with deep learning denoising methods that only use a fixed window, it has a higher detail preservation capability.

[0085] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for adaptive denoising and intelligent optimization processing of geophysical exploration data, characterized in that, Specifically comprising the following steps: Collecting geophysical exploration data and performing baseline correction and amplitude normalization; According to the window length N in the adaptive parameter, the geophysical exploration data D after the baseline correction and the amplitude normalization is slidingly divided, and each window signal di uses the mathematical morphology erosion, expansion, opening and closing operation to obtain the morphological component of the kth scale on the window i ; Each window signal is converted to the frequency domain, and is decomposed by variational modal decomposition to obtain a frequency domain mode at the kth scale in the window i , and the weights are calculated The two types of components are fused to form a window fusion matrix ; The multi-scale morphological components are slidingly embedded to form a Hankel matrix, and low-rank matrix and sparse matrix are obtained by performing low-rank sparse decomposition on the Hankel matrix; The low-rank matrix and the multi-scale morphological components are input into the gated denoising network, the gated denoising network generates gating weights through the gating layer to adjust the contribution of different scale features, and outputs the denoised signal, and the loss function of the gated denoising network is: ； wherein is the gating denoising network is the whole set of learnable parameters of the gating denoising network, is the morphological component segmented by window i at the k-th scale, is the denoising result predicted by the gating denoising network, is the error weight matrix corresponding to the k-th scale, is the Hadamard element-wise multiplication, is the Frobenius norm, is the gating weight of the n-th channel, is the gating entropy regularization coefficient; The window length, the morphological structure element scale, the low-rank decomposition regularization parameter and the gated denoising network parameter are optimized by using the swarm intelligence optimization algorithm, and the signal-to-noise ratio and the structural similarity of the denoised signal are used as the fitness function to update the parameters until the fitness converges; The denoised signals of the windows are recombined as denoised results and output.

2. The method according to claim 1, wherein, The low-rank sparse decomposition is obtained by solving an optimization problem obtained, where L is a low-rank matrix, S is a sparse matrix, M is the original geophysical survey data matrix, is an inverse vectorization operator, is a nuclear norm, which enforces L to be low-rank, is , which enforces S to be sparse, is a Frobenius norm, and are low-rank regularization weight and data fidelity weight, respectively.

3. The method according to claim 1, wherein, The swarm intelligence optimization algorithm is an improved particle swarm algorithm, and the parameter update formula is: ； where is the parameter vector of the i-th particle at iteration t+1, is the parameter vector of the i-th particle at iteration t, denotes the best parameter position achieved by particle i so far over all previous iterations, is the global best parameter position over the history of the entire population so far, and is the learning factor, and is a perturbation factor randomly sampled in the interval [0, 1].

4. A geophysical exploration data adaptive denoising and intelligent optimization processing system, characterized in that, The method comprises the following steps: A data acquisition module is configured to collect geophysical exploration data and perform baseline correction and amplitude normalization; A morphological decomposition module is configured to split the geophysical exploration data according to window parameters, and obtain multi-scale morphological components by using mathematical morphological operations and variational mode decomposition; A low-rank decomposition module is configured to convert the multi-scale morphological components into a Hankel matrix and solve a low-rank matrix and a sparse matrix; A gated denoising network module is configured to receive the multi-scale morphological components and the low-rank matrix, generate gating weights through a gating layer, and output a denoised signal; An intelligent optimization module is configured to optimize the window length, the morphological structure element scale, the low-rank decomposition regularization parameter and the gated denoising network parameter by using a swarm intelligence optimization algorithm; A control and storage module includes a processor and a memory, the processor is configured to execute the control logic of the above modules, and the memory is configured to store program codes and intermediate data.

5. The adaptive de-noising and intelligent optimization processing system for geophysical prospecting data according to claim 4, characterized in that, The gated denoising network module includes a plurality of convolution layers, normalization layers, gating layers and output layers, the gating layers generate weights through a sigmoid function to control the contribution of different scale features.

6. The adaptive de-noising and intelligent optimization processing system for geophysical prospecting data according to claim 4, characterized in that, The intelligent optimization module evaluates the signal-to-noise ratio and the structural similarity of the denoised results through a fitness evaluation unit, and uses the evaluation results as the fitness function of the swarm intelligence optimization algorithm.

Images