Seismic data-based tight gas microcrack detection method and system

Abstract

The invention provides a tight gas micro-crack detection method and system based on seismic data, and relates to the technical field of tight gas micro-crack detection, and the method comprises the steps: obtaining seismic wave data, and obtaining the preprocessed seismic wave data through wavelet transform denoising, compressed sensing reconstruction and median filtering smoothing processing. Using a preset multi-dimensional feature extraction model to extract micro-crack related features from the preprocessed data, including crack edge, curvature, frequency and similarity features, and constructing a multi-dimensional feature matrix; and based on the feature matrix, modeling and training are carried out by adopting a convolutional neural network model of a spatial attention mechanism added with a channel, and then the crack probability of each to-be-detected position is predicted. And carrying out smoothing, denoising and connectivity analysis on the crack probability image through image conversion and optimization processing to obtain a final crack detection result. According to the invention, the precision of micro-crack detection can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tight gas microfracture detection. Specifically, it relates to a method and system for tight gas microfracture detection based on seismic data. Background Art

[0002] Currently, during the development of tight gas reservoirs, fractures are the key factors affecting the permeability and gas production capacity of gas reservoirs. Therefore, the accurate detection and quantitative analysis of fractures are crucial. Traditional tight gas microfracture detection techniques mainly rely on the processing and analysis of seismic wave data. However, the effect of traditional methods is limited by noise interference and the ability to identify tiny fractures. Especially under complex geological conditions, it is difficult for traditional techniques to effectively distinguish microfractures from other geological anomalies, resulting in low detection accuracy. Existing technologies detect fractures through the frequency and amplitude characteristics of seismic waves, but this method cannot fully consider the geometric characteristics and spatial distribution of fractures, nor can it effectively integrate information at different scales, resulting in large deviations in fracture positioning and aperture estimation.

[0003] Therefore, there is an urgent need for a method and system for tight gas microfracture detection based on seismic data to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for tight gas microfracture detection based on seismic data to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] In the first aspect, the present application provides a method for tight gas microfracture detection based on seismic data, including:

[0006] Obtaining seismic wave data based on seismic exploration equipment, where the seismic wave data includes amplitude data, frequency data, and phase data reflecting the characteristics of underground rock formations;

[0007] Performing wavelet transform denoising, compressive sensing reconstruction, and median filtering smoothing processing on the seismic wave data in sequence to obtain preprocessed seismic wave data;

[0008] Performing feature extraction on the preprocessed seismic wave data, where microfracture feature information is extracted through a preset multi-dimensional feature extraction model to obtain a multi-dimensional feature matrix, and the microfracture feature information includes fracture edge features, curvature features corresponding to fractures, frequency features corresponding to fractures, and similarity features corresponding to fractures;

[0009] Based on the multi-dimensional feature matrix, a convolutional neural network model with a spatial attention mechanism is used for modeling and training, and the probability of fractures occurring at each position to be detected is predicted based on the trained convolutional neural network model;

[0010] Perform image conversion and optimization processing according to the probabilities of cracks appearing at all positions to be detected, obtain the optimized crack image, and use it as the detection result of micro-cracks in tight gas.

[0011] In a second aspect, the present application also provides a tight gas micro-crack detection system based on seismic data, including:

[0012] An acquisition unit for acquiring seismic wave data based on seismic exploration equipment, where the seismic wave data includes amplitude data, frequency data, and phase data reflecting the characteristics of underground rock formations;

[0013] A processing unit for sequentially performing wavelet transform denoising, compressive sensing reconstruction, and median filter smoothing on the seismic wave data to obtain preprocessed seismic wave data;

[0014] An extraction unit for extracting features based on the preprocessed seismic wave data. Among them, micro-crack feature information is extracted through a preset multi-dimensional feature extraction model to obtain a multi-dimensional feature matrix. The micro-crack feature information includes crack edge features, curvature features corresponding to cracks, frequency features corresponding to cracks, and similarity features corresponding to cracks;

[0015] A prediction unit for modeling and training according to the multi-dimensional feature matrix using a convolutional neural network model with a spatial attention mechanism, and predicting the probability of cracks appearing at each position to be detected based on the trained convolutional neural network model;

[0016] A detection unit for performing image conversion and optimization processing according to the probabilities of cracks appearing at all positions to be detected, obtaining the optimized crack image, and using it as the detection result of micro-cracks in tight gas.

[0017] The beneficial effects of the present invention are as follows:

[0018] The present invention combines deep learning technology with multiple signal processing algorithms, significantly improving the accuracy and reliability of crack detection. The present invention first denoises and reconstructs the original seismic data, then extracts multi-level information of cracks using a multi-dimensional feature extraction model, and then performs modeling and training through a convolutional neural network combined with a spatial attention mechanism, finally realizing high-precision micro-crack probability prediction. Compared with the prior art, the technology of the present application not only improves the ability to identify micro-cracks, but also can quantify the density, aperture, and connectivity of cracks, thereby providing accurate support for the permeability analysis of reservoirs. Through this method, the present invention effectively overcomes the limitations of traditional technologies in dealing with complex geological conditions and provides a more comprehensive and efficient tight gas reservoir evaluation tool.

[0019] Other features and advantages of the present invention will be described in the subsequent specification, and in part will be obvious from the specification, or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification, claims, and drawings. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as a limitation of the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 Schematic diagram of the process for the tight gas microfracture detection method based on seismic data described in the embodiments of the present invention;

[0022] Figure 2 Schematic diagram of the structure of the tight gas microfracture detection system based on seismic data described in the embodiments of the present invention.

[0023] In the figure: 701, acquisition unit; 702, processing unit; 703, extraction unit; 704, prediction unit; 705, detection unit; 706, analysis unit; 7021, first processing subunit; 7022, second processing subunit; 7023, third processing subunit; 7024, fourth processing subunit; 7031, first extraction subunit; 7032, second extraction subunit; 7033, third extraction subunit; 7034, fourth extraction subunit; 7035, fifth extraction subunit; 7041, first prediction subunit; 7042, second prediction subunit; 7043, third prediction subunit; 7044, fourth prediction subunit; 7051, first detection subunit; 7052, second detection subunit; 7053, third detection subunit; 7054, fourth detection subunit; 7061, first analysis subunit; 7062, second analysis subunit; 7063, third analysis subunit; 7064, fourth analysis subunit. Detailed Embodiments

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0025] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0026] Embodiment 1:

[0027] This embodiment provides a method for detecting microfractures in tight gas based on seismic data.

[0028] See Figure 1 , which shows that this method includes steps S1, S2, S3, S4, and S5.

[0029] Step S1: Obtain seismic wave data based on seismic exploration equipment. The seismic wave data includes amplitude data, frequency data, and phase data that reflect the characteristics of underground rock formations.

[0030] It can be understood that in this step, the seismic wave data obtained based on seismic exploration equipment is the basic data for microfracture detection and reservoir analysis. These seismic wave data include amplitude, frequency, and phase data, and each type of data carries different information about the underground rock formations. The amplitude data reflects the intensity of the interaction between seismic waves and the formation during propagation and is usually used to identify underground reflection interfaces and structures such as fractures. The frequency data provides the elastic characteristics of the formation and the location of the reflection interface. Higher frequencies are usually associated with smaller fractures or detailed structures, while lower frequencies help identify larger-scale structures and reservoir characteristics. The phase data further reveals the propagation path of the wave and the geometric characteristics of the reflection interface, which is important for accurately positioning the orientation and depth of fractures.

[0031] These seismic wave data are obtained through seismic exploration equipment such as seismographs and seismic sources. The accuracy and quality of the data directly affect the results of subsequent processing and analysis. Since seismic wave data usually contain noise and errors, the initially obtained seismic wave data need to be further processed before being used for micro-fracture detection. In this step, the obtained data serves as the basis for subsequent denoising, reconstruction, and feature extraction. The data obtained through this step can provide multi-dimensional and high-quality data support for subsequent fracture feature extraction and neural network model training, ensuring the accuracy and reliability of the final micro-fracture detection.

[0032] Step S2: Perform wavelet transform denoising, compressive sensing reconstruction, and median filter smoothing on the seismic wave data in sequence to obtain preprocessed seismic wave data.

[0033] It can be understood that this step not only improves the reliability of micro-fracture detection but also ensures that the seismic wave data can effectively reflect the true characteristics of the underground structure, laying a foundation for the high-precision detection and permeability evaluation of tight gas micro-fractures. In this step, Step S2 includes Step S21, Step S22, Step S23, and Step S24.

[0034] Step S21: Perform wavelet transform denoising on the seismic wave data. By using the Daubechies wavelet basis to perform multi-scale decomposition on the seismic data, the signal is decomposed into low-frequency and high-frequency parts, and soft threshold filtering is performed on the high-frequency noise based on a preset threshold to obtain denoised seismic wave data.

[0035] It can be understood that in this step, first, multi-scale decomposition is performed on the seismic wave data using the Daubechies wavelet basis. The Daubechies wavelet basis has good localization characteristics in both the time domain and the frequency domain, enabling efficient processing of seismic wave signals. Especially when dealing with signals with fast-changing or sudden change characteristics, it shows good adaptability. Through multi-scale decomposition, the seismic wave signal is decomposed into a low-frequency part and a high-frequency part. The low-frequency part contains the main information of the signal, while the high-frequency part usually contains noise.

[0036] Then, soft threshold filtering is performed on the high-frequency part based on a preset threshold. Soft threshold filtering is a commonly used wavelet denoising method. By performing threshold processing on the coefficients of the high-frequency part, it can effectively suppress the noise component while retaining as much important detail information in the signal as possible. Specifically, the soft threshold method sets the high-frequency coefficients below the set threshold to zero and retains the part above the threshold, reducing the influence of noise.

[0037] After denoising by wavelet transform, the obtained seismic wave data after denoising is relatively smooth, the noise components in the signal are greatly reduced, and the effective seismic signals are retained. This denoising process significantly improves the quality of seismic data and ensures that the subsequent feature extraction and model training stages are not interfered by noise.

[0038] Step S22: Perform compressive sensing reconstruction processing on the seismic wave data after denoising. By applying the Basis Pursuit algorithm based on L1-norm minimization, perform sparse representation and reconstruction on the seismic wave data to obtain the reconstructed seismic wave data.

[0039] It can be understood that in this step, the seismic wave data after denoising is first used, and the Basis Pursuit algorithm based on L1-norm minimization is applied. The goal of L1-norm minimization is to find a sparse solution so that the signal has the fewest non-zero coefficients under Fourier transform. This is because seismic wave signals are usually sparse, that is, most of the information is concentrated in a few important coefficients, while other coefficients are close to zero or very small. By minimizing the L1-norm, the most informative part can be effectively selected from a large amount of measurement data and unnecessary redundancy can be suppressed.

[0040] Specifically, the Basis Pursuit algorithm solves an optimization problem to minimize the L1-norm of the signal while ensuring that the measurement matrix of the signal conforms to the observed data as much as possible. This process can be regarded as an optimization problem, that is, under given constraints, minimizing the sparsity of the signal to achieve efficient reconstruction of the signal.

[0041] In compressive sensing, the goal is to recover the original signal from the observed data and the measurement matrix. Since the signal is sparse, we can recover it by minimizing its sparsity (i.e., the L1-norm), and achieve this goal by solving the following optimization problem:

[0042]

[0043] where, ||x|| 1 is the L1-norm of the signal x, which represents the signal sparsity; Φx = y is the measurement constraint, indicating recovering the original signal x from the observed data y and the measurement matrix Φ.

[0044] Step S23: Perform median filtering and smoothing processing on the seismic wave data after compressive sensing reconstruction. By using a sliding window of a preset size, sort each data point, take the median value, and remove local outliers to obtain the smoothed seismic wave data.

[0045] It is understandable that in this step, the seismic wave data is first locally processed through a sliding window. Specifically, the size of the sliding window is preset. For example, a fixed window length (such as 3, 5, 7, etc.) is selected, and this window slides over each point of the seismic wave data. The window contains a certain number of adjacent data points around this point. For the data points within each window, by sorting these points, the median value is selected as the smoothed value of this point. This method can remove the high-frequency spikes caused by data noise or sudden anomalies, and avoid the problem of excessive signal smoothing that may be caused by traditional mean filtering.

[0046] From a technical effect perspective, median filtering can effectively preserve the edge features of the seismic wave data and the overall structure of the signal. Especially when removing local noise, it will not have an adverse impact on the main features of the signal (such as cracks, fluctuations, etc.). Compared with other filtering methods, median filtering is more robust in dealing with outliers and can better retain the key geological features in the seismic wave data, such as rock layer interfaces and crack features. Through this step, the smoothness of the data is improved and the noise influence is reduced, thus providing a clearer and more reliable data input for subsequent steps such as crack detection and feature extraction.

[0047] Step S24: Perform normalization processing on the smoothed seismic wave data, and process the amplitude of each data point through Z-score standardization to obtain the standardized seismic wave data.

[0048] It is understandable that this step processes the amplitude of the seismic wave data through Z-score standardization, aiming to eliminate the dimensional differences and scale inconsistencies in the data, so as to ensure that the subsequent processing steps can be carried out efficiently and accurately. This step can make the numerical ranges of different features or different data sources consistent, and avoid the unnecessary influence of some larger or smaller amplitude values on the subsequent analysis.

[0049] Step S3: Extract features based on the preprocessed seismic wave data. Among them, micro-crack feature information is extracted through a preset multi-dimensional feature extraction model to obtain a multi-dimensional feature matrix. The micro-crack feature information includes crack edge features, curvature features corresponding to the cracks, frequency features corresponding to the cracks, and similarity features corresponding to the cracks.

[0050] It is understandable that this step extracts features highly sensitive to microcracks from the preprocessed seismic wave data, providing key data support for subsequent crack identification and analysis. By adopting a preset multi-dimensional feature extraction model, this step can extract the geometric features, physical properties of the cracks and their manifestations in the seismic wave data, which are closely related to the spatial distribution and morphology of the cracks. Feature extraction is not only to enhance the identifiability of crack signals, but also to effectively reduce noise interference and improve detection accuracy. In this step, step S3 includes step S31, step S32, step S33, step S34 and step S35.

[0051] Step S31: Extract crack edge features based on the preprocessed seismic wave data. By using the Sobel operator to calculate the gradient field of the seismic data, calculate the local gradient information in the three directions of X, Y, and Z to obtain the crack edge features.

[0052] It is understandable that this step captures the edge features of cracks in the seismic image by utilizing the gradient information of the seismic wave data. Cracks usually form significant faults or interfaces in underground rock formations, and there are significant differences in physical properties (such as density, wave velocity, etc.) between these interfaces and the surrounding areas. Therefore, they appear as areas with sharp changes in seismic data, especially at the crack edges. By extracting these areas with sharp changes, the location and morphology of the cracks can be effectively identified. First, the Sobel operator uses a preset convolution kernel to perform a convolution operation on the seismic wave data, calculating the gradient values of the data points in the X direction (horizontal direction), Y direction (vertical direction), and Z direction (depth direction). The gradient value reflects the degree of change of the data point relative to its neighborhood and can effectively capture the edge information in the seismic wave data. The larger the gradient, the more obvious the change, and the more likely it is the area where the crack is located. Second, by calculating the gradients of the neighborhoods of each data point, local gradient information is obtained. Finally, by combining the gradient information in the X, Y, and Z directions, the edge position of the crack can be determined by maximizing the gradient magnitude. Usually, the area where the crack is located will show a significant increase in the gradient value, which is also where the edge of the crack is located.

[0053] Step S32: Extract curvature features based on the crack edge features. By calculating the second derivative of the seismic wave data, obtain the maximum curvature value of the local area and use it as the curvature feature corresponding to the crack.

[0054] It can be understood that in this step, by calculating the second-order spatial derivative of the seismic wave data, the curvature information of the data can be obtained. The second derivative reflects the acceleration of signal change and can sensitively capture the drastic changes in the area where the fractures are located. For seismic wave data, the calculation of the second derivative is usually carried out in the spatial domain, that is, calculating the second partial derivatives in the X, Y, and Z directions. The larger the second derivative, the higher the curvature of the area, indicating that there may be obvious fractures there. By calculating the second derivative of each data point, the curvature values of the local area can be obtained. Fractures usually show drastic changes in local areas of seismic data, and such drastic changes are usually accompanied by large curvature values. Therefore, the curvature values in the areas where the fracture edges are located are much higher than those in the surrounding areas. In each local area, the calculated curvature values may vary, so it is necessary to extract the maximum curvature value as the representative curvature feature of this area. The areas where fractures are located are usually the maximum value points of the curvature values, and the areas corresponding to these points are the possible fracture edges. Finally, the extracted maximum curvature value is used as the curvature feature corresponding to the fracture, which reflects the degree of bending in the area where the fracture is located. This feature can not only help identify the shape and location of the fractures but also provide effective information for subsequent quantitative analysis of fractures (such as the aperture and length of the fractures).

[0055] Step S33: Extract frequency features based on the preprocessed seismic wave data. By performing a fast Fourier transform on the preprocessed seismic wave data, obtain the spectral information of the seismic signal, and extract the high-frequency components related to fractures from it to obtain the frequency features corresponding to the fractures;

[0056] It can be understood that in this step, by performing a fast Fourier transform on the preprocessed seismic wave data, the time-domain signal is converted into a frequency-domain signal. The fast Fourier transform is an efficient algorithm that can complete the frequency-domain conversion of a large amount of data in a short time, greatly improving the calculation speed. The result of the transformation is a spectrum, where each frequency point represents the intensity information of the seismic wave signal at that frequency. The output of the fast Fourier transform includes amplitude and phase information. The amplitude information reflects the intensity of each frequency component, while the phase information describes the fluctuation pattern of the signal. By analyzing the spectrum, the high-frequency components related to fractures can be extracted. These high-frequency components usually represent the detailed information of the fractures, including the shape and width of the fractures and the minor deformations of the rock formations. The high-frequency components related to fractures have a stronger frequency response compared to other geological structures (such as large-scale rock formations or dikes). By extracting the high-frequency part from the spectrum, the frequency features of the fractures can be obtained. This frequency feature can reveal the microscopic structural information of the fractures, especially the changes at a smaller scale.

[0057] Step S34: Extract similarity features based on the crack frequency characteristics. Calculate the similarity between adjacent gathers by using the C3 coherence algorithm to highlight the low-coherence regions where the cracks are located, and use them as the similarity features corresponding to the cracks.

[0058] It can be understood that this step is collected through multiple detection points (gathers), and each gather contains seismic wave data at the corresponding position. When extracting similarity features, a group of adjacent gathers need to be selected first, and the data of these gathers should reflect the seismic wave characteristics at different positions. The C3 coherence algorithm extracts similarity features by calculating the coherence between adjacent gathers. During the propagation of seismic waves, due to the different properties of underground rock formations and the existence of structures such as cracks, the propagation speed, amplitude, etc. of seismic waves will change. This change will cause different degrees of offset and attenuation of the waveforms between adjacent gathers. The C3 algorithm measures the waveform similarity between adjacent gathers through a mathematical model and generates a coherence coefficient, which represents the matching degree between different gathers. When the waveform similarity is high, the coherence coefficient is large; while where there are cracks, the waveform similarity is low, so the coherence coefficient is small. Finally, the C3 coherence algorithm will generate a similarity map, in which the low-coherence regions will be highlighted, and these low-coherence regions are the regions where the cracks are located. These regions are used as the similarity features corresponding to the cracks.

[0059] Among them, the calculation formula for the waveform similarity between adjacent gathers is as follows:

[0060]

[0061] Among them, C ij is the coherence coefficient, representing the waveform similarity between adjacent gathers, x i (t) is the seismic signal of gather i, representing the seismic waveform data at gather i at time t, x j (t) is the seismic signal of gather j, representing the seismic waveform data at gather j at time t, t represents the time dimension of the signal, which is the sampling time point of the measured data, |x i (t)| 2 is the square modulus of the signal x i (t), representing the signal intensity of gather i at time t, |x j (t)| 2 is the square modulus of the signal x j (t), representing the signal intensity of gather j at time t, |∫ T x i (t)·x j (t)dt| is the absolute value of the signal inner product.

[0062] Step S35: Construct a matrix based on the crack edge feature, the curvature feature corresponding to the crack, the frequency feature corresponding to the crack, and the similarity feature corresponding to the crack to obtain a multi-dimensional feature matrix.

[0063] It can be understood that in this step, these features are integrated into a multi-dimensional feature matrix. Each feature (edge feature, curvature feature, frequency feature, similarity feature) will generate a set of numerical values in space, and these numerical values reflect the crack features of the seismic wave data at that position. By arranging these feature values according to the spatial position, a multi-dimensional feature matrix is finally obtained. Each row of this matrix can be regarded as the feature vector of the seismic data at a specific position, and each column represents a certain type of feature (such as crack edge, curvature, etc.). By integrating different types of crack features into a multi-dimensional feature matrix, it can provide comprehensive and efficient input data for subsequent analysis and model training. Each feature represents the changes in different aspects of the seismic wave data in space in the matrix. This multi-angle information fusion helps to capture the detailed features of the cracks. Compared with the single feature extraction method, this multi-dimensional feature matrix can significantly improve the accuracy and robustness of crack identification.

[0064] Step S4: According to the multi-dimensional feature matrix, use a convolutional neural network model with a spatial attention mechanism for modeling and training, and predict the probability of cracks appearing at each position to be detected based on the trained convolutional neural network model;

[0065] It can be understood that in this step, a convolutional neural network model with a spatial attention mechanism is used, which can effectively improve the accuracy and robustness of crack detection. The convolutional neural network can automatically learn the complex patterns and features in the seismic wave data, reducing the dependence on artificial feature design, while the attention mechanism helps the model focus on the most relevant features and regions, avoiding being interfered by noise in the complex seismic data. In this step, Step S4 includes Step S41, Step S42, Step S43, and Step S44.

[0066] Step S41: According to the multi-dimensional feature matrix, perform the initialization input of the convolutional neural network model. By using the multi-dimensional matrix containing crack edge, curvature, frequency, and similarity features as the input of the convolutional neural network, initialize the first-layer convolution kernel of the network to obtain the preliminary feature representation of the convolutional neural network;

[0067] It can be understood that in the network initialization stage, this multi-dimensional feature matrix is used as the input of the convolutional neural network. The convolutional neural network gradually extracts more abstract and high-level features from the original data through a series of convolutional layers. The first layer of convolutional kernels (filters) in the network learn how to extract the initial spatial features through the sliding convolutional operation on the input feature matrix. In this process, the convolutional kernel weights are randomly initialized and gradually adjusted to optimally adapt to the data.

[0068] Step S42: According to the preliminary feature representation of the convolutional neural network, add a spatial attention mechanism. By performing channel attention and spatial attention weighting on the feature map output by the convolutional layer, a weighted feature map is obtained;

[0069] It can be understood that the core idea of this step's channel attention mechanism is to dynamically adjust the weights of each channel according to the importance of each channel. Specifically, the feature map output by the convolutional layer has multiple channels, and each channel corresponds to a different feature dimension. Through the channel attention mechanism, the model can automatically learn which feature channels are more important for crack detection and improve the contribution of these channels through weighting. Channel attention usually generates channel weights through global average pooling or global maximum pooling operations, and then multiplies these weights with the original feature map to achieve channel weighting. By introducing a spatial attention mechanism in the convolutional neural network, it is possible to effectively screen out the most critical feature regions for crack detection, thereby enhancing the detection ability of the model. The channel attention mechanism helps the network identify which feature channels are closely related to crack features and reduces the interference of irrelevant features on the results; while the spatial attention mechanism enhances the attention to the crack region, enabling the network to focus on the crack region in the seismic wave data.

[0070] Step S43: According to the weighted feature map, perform training on the convolutional neural network for a preset number of layers. Through multiple stacks of convolutional, pooling, and fully connected layers, a trained convolutional neural network model is obtained;

[0071] It can be understood that in this step, each convolution layer performs convolution operations on the input feature map through operations with the convolution kernel to extract features at different levels. Low-level convolution layers usually learn basic features, such as edges, textures, etc., while high-level convolution layers can learn more abstract features, such as the shape and position of cracks. Through the superposition of multiple layers of convolution, the network can gradually transition from low-level features to higher-level and more recognizable crack features. The pooling layer is used to reduce the spatial dimension of the feature map output by the convolution layer, thereby reducing the amount of calculation and enhancing the local invariance of the features. Use the maximum pooling operation. After each layer of pooling operation, the size of the feature map will be reduced, but important spatial information and abstract features are retained. After multiple convolutions and pooling, the feature map will be flattened and sent to the fully connected layer for classification or regression prediction. The fully connected layer outputs the probability of cracks at each location to be detected by weighted combination of all features.

[0072] Step S44: perform crack probability prediction based on the trained convolutional neural network model, and output the probability of cracks occurring at each location to be detected by inputting the seismic data to be detected into the trained model.

[0073] It can be understood that this step, with the help of deep learning models, can achieve automatic, efficient and high-precision crack identification, greatly reducing human intervention, and is particularly suitable for large-scale and complex seismic data analysis.

[0074] Step S5: Perform image conversion and optimization processing according to the probability of cracks appearing at all positions to be detected, obtain an optimized crack image, and use it as a dense gas microcrack detection result.

[0075] It can be understood that this step can convert the probability value output by the deep learning model into a specific fracture image, and improve the quality of the image through optimization means, so that the detection result is more accurate and reliable. Through this process, noise can be effectively removed, false detection can be reduced, and the spatial characteristics of the fracture can be strengthened, ensuring that the final fracture detection result can provide accurate geological data support for the development of tight gas reservoirs. In this step, step S5 includes step S51, step S52, step S53 and step S54.

[0076] Step S51, performing image conversion processing according to the probability of cracks appearing at the positions to be detected, by mapping the probability value of each position to be detected to the corresponding image pixel value, converting the probability value into the pixel intensity of the grayscale image, and obtaining a preliminary crack probability grayscale image;

[0077] It can be understood that the crack probability values at each position to be detected in this step need to be mapped to the pixel intensity values of the grayscale image. In a common grayscale image, the pixel value range is usually from 0 (black) to 255 (white). Therefore, the probability values between 0 and 1 need to be linearly mapped to the pixel intensity range of 0 to 255. For example, a point with a probability value of 0.8 will be mapped to a pixel value of 255×0.8 = 204, indicating that there is a relatively significant crack at this position; while a point with a probability value of 0.2 will be mapped to 255×0.2 = 51, indicating a lower probability of crack at this position. By mapping the probability value of each position to be detected to the corresponding pixel intensity value, a grayscale image is finally generated. Each pixel value of this image represents the crack occurrence probability at that position, and the spatial distribution of cracks can be visually presented through the brightness of the image. Generally, the larger the grayscale value, the higher the probability of crack at that position, and vice versa.

[0078] Step S52: According to the preliminary crack probability grayscale image, perform noise removal and smoothing processing. By applying Gaussian filtering, remove the isolated noise points generated by inaccurate crack positioning, and smooth the edges in the image to obtain a denoised and smoothed crack probability image.

[0079] It can be understood that in this step, by applying Gaussian filtering to the crack probability image, the image is smoothed using the Gaussian function to remove the isolated noise points that may be generated in crack positioning. Specifically, a Gaussian kernel with a preset size (such as a 3x3 or 5x5 matrix) is used to convolve the image, and the weighted average value of each pixel and its surrounding neighborhood pixels is calculated. Due to the properties of the Gaussian function, the pixel values around the central pixel will be given different weights according to the distance. The closer the pixel is, the greater the weight, thereby smoothing the image and reducing noise. By removing the isolated noise points and weakening the unnecessary local fluctuations, the processed image is more stable and continuous in representing the spatial distribution of cracks.

[0080] Step S53: According to the denoised and smoothed crack probability image, perform morphological processing. Remove the isolated noise points through opening operation, and connect the discontinuous segments of the cracks through closing operation to obtain an optimized crack connectivity image.

[0081] It can be understood that the opening operation in this step is a commonly used operation in morphological processing, which consists of first performing an erosion operation and then a dilation operation. The erosion operation removes isolated noise points (small individual pixel regions) by comparing each pixel in the image with its neighboring pixels. The dilation operation fills the holes in the image to ensure that the crack regions are completely represented. Through the opening operation, small dots in the image caused by noise or discontinuity are effectively removed, thus eliminating the influence of noise. The closing operation is another operation in morphology, which is to perform dilation first and then erosion. The dilation operation can fill small gaps or missing regions in the crack image, and the erosion operation helps to restore the edge shape of the crack. Through the closing operation, the broken parts between cracks can be connected, the discontinuous segments in the crack image can be filled, and the connectivity of the cracks can be improved, making the cracks appear as a more continuous structure in the image. This step significantly improves the usability of the crack image through the combination of the opening operation and the closing operation, making it more accurately reflect the spatial distribution and structural characteristics of underground cracks.

[0082] Step S54: According to the optimized crack connectivity image, perform crack probability threshold segmentation, binarize the crack probability image through a preset threshold, mark the crack region, and obtain the final crack detection result.

[0083] It can be understood that in this step, by setting a crack probability threshold, the pixel values in the image are classified according to the threshold. Generally speaking, after setting an appropriate threshold, the probability values greater than the threshold are marked as 1 (indicating the crack region), and the probability values less than the threshold are marked as 0 (indicating the non-crack region). The selection of the threshold usually depends on the data characteristics and actual needs, and can be adjusted through experiments or experience to ensure that the crack region can be accurately segmented. Through the threshold segmentation, the optimized crack connectivity image obtains the final crack region marking, making the crack detection result clearer, easier to interpret and use. The technical effect of this step is that it can effectively extract the specific crack region from the continuous probability image, providing clear spatial data for subsequent crack evaluation, analysis and decision-making.

[0084] It can be understood that after step S5, there is also step S6.

[0085] Step S6: Perform crack density calculation, crack aperture estimation and connectivity analysis on the optimized crack image in sequence to obtain the reservoir permeability distribution map.

[0086] It can be understood that through the multi-dimensional analysis of crack characteristics in this step, the fluid flow characteristics of the reservoir can be accurately revealed, and thus provide decision support for optimizing the oil and gas exploitation plan and improving resource utilization rate. In this step, step S6 includes step S61, step S62, step S63 and step S64.

[0087] Step S61: Calculate the fracture density based on the optimized fracture image. By statistically calculating the ratio of the number of voxels in the fracture area to the total number of voxels, obtain the distribution density of fractures per unit volume, and construct it into a fracture density map.

[0088] It can be understood that in this step, each voxel (3D pixel) in the fracture area can be regarded as a part of the fracture. By statistically analyzing the voxels in the fracture area, the volume space occupied by the fractures can be obtained. The ratio between the number of these voxels and the total number of voxels (the total number of voxels in the entire seismic data volume) can reflect the distribution density of fractures per unit volume. Map the relationship between the fracture density value and the coordinates of the corresponding spatial position to generate an image (usually a three-dimensional image) reflecting the fracture distribution density. This image can intuitively display the changes in fracture density in different regions. Usually, regions with higher density will be marked as darker colors or high-intensity regions, while regions with lower density will be shown as lighter colors or low-intensity regions. Through the calculation of fracture density, the generated fracture density map can accurately reveal the spatial distribution of fractures in the reservoir.

[0089] Step S62: Estimate the fracture aperture based on the fracture density map. By establishing a mapping between amplitude and aperture based on the least squares method, fit the correlation between the probability value of fractures and the seismic wave amplitude data to obtain a fracture aperture distribution map.

[0090] It can be understood that in this step, the seismic wave amplitude data is used as the independent variable and the fracture aperture as the dependent variable. By establishing a linear or non-linear regression model (according to the data characteristics) for fitting, the functional relationship between amplitude and aperture can be obtained. By applying the fitted aperture mapping to the amplitude data of the fracture area, the corresponding fracture aperture value can be calculated for each point to be detected. The fracture aperture distribution map is obtained by associating these aperture values with the spatial coordinates of the corresponding positions to obtain an image or three-dimensional model reflecting the changes in fracture aperture.

[0091] Step S63: Analyze the fracture connectivity based on the fracture aperture distribution map. By applying the maximum flow algorithm, analyze the connectivity in the fracture network, identify the connected regions and non-connected regions, and obtain a fracture network connectivity map.

[0092] It can be understood that this step maximizes the flow from the source point to the sink point by finding a feasible path in the network. In the fracture network analysis, the source point represents the inlet area of the fracture, the sink point represents the outlet area of the fracture, and the flow represents the permeability or fluid passing ability between fractures. Regarding the fracture aperture distribution map as a graph structure, where each node represents a fracture segment and each edge represents the connection relationship between fractures, and the weight of the edge is determined by the aperture value of the fracture. Fracture segments with larger apertures will be assigned higher weight values, indicating that fluids can easily pass through these fractures; while fracture segments with smaller apertures or poor connectivity will be assigned lower weight values. In this graph, the purpose of the maximum flow algorithm is to find the maximum flow throughput from the starting point (source point) to the ending point (sink point) of the fracture network, so as to determine which fracture areas are connected and which are not. The algorithm finds potential connection paths between fractures by gradually expanding the flow.

[0093] Step S64: Generate a reservoir permeability distribution map based on the fracture connectivity map. By comprehensively considering the density, aperture, and connectivity information of fractures, apply a weighted summation model to calculate the permeability value of each area and construct a permeability distribution map.

[0094] It can be understood that this step performs a weighted summation of the values of the three characteristics of fracture density, aperture, and connectivity to obtain the comprehensive permeability value of each geological area. Specifically, fractures with higher density, larger aperture, and better connectivity are usually assigned higher weight values, thereby increasing the permeability score. For areas with sparse fractures, small aperture, or poor connectivity, the permeability score is relatively low.

[0095] Among them, the calculation formula for the permeability score is as follows:

[0096] P = w 1 ·D + w 2 ·O + w 3 ·C

[0097] Among them, P is the permeability score, D is the fracture density, O is the aperture, C is the connectivity score, and w 1 、w 2 and w 3 represent the corresponding weight coefficients.

[0098] Example 2:

[0099] As Figure 2 shown, this embodiment provides a tight gas microfracture detection system based on seismic data. Refer to Figure 2 The system includes an acquisition unit 701, a processing unit 702, an extraction unit 703, a prediction unit 704, and a detection unit 705.

[0100] An acquisition unit 701, configured to acquire seismic wave data based on a seismic exploration device, where the seismic wave data includes amplitude data, frequency data, and phase data reflecting the characteristics of underground rock formations;

[0101] A processing unit 702, configured to perform wavelet transform denoising, compressive sensing reconstruction, and median filtering smoothing processing on the seismic wave data in sequence to obtain preprocessed seismic wave data;

[0102] Among them, the processing unit 702 includes a first processing subunit 7021, a second processing subunit 7022, a third processing subunit 7023, and a fourth processing subunit 7024.

[0103] The first processing subunit 7021 is configured to perform wavelet transform denoising processing on the seismic wave data. By using the Daubechies wavelet basis to perform multi-scale decomposition on the seismic data, the signal is decomposed into low-frequency and high-frequency parts, and soft threshold filtering is performed on the high-frequency noise based on a preset threshold to obtain denoised seismic wave data;

[0104] The second processing subunit 7022 is configured to perform compressive sensing reconstruction processing on the denoised seismic wave data. By applying the Basis Pursuit algorithm based on L1 norm minimization, sparse representation and reconstruction are performed on the seismic wave data to obtain reconstructed seismic wave data;

[0105] The third processing subunit 7023 is configured to perform median filtering smoothing processing on the seismic wave data after compressive sensing reconstruction. By adopting a sliding window with a preset size, each data point is sorted, the median is taken, and local outliers are removed to obtain smoothed seismic wave data;

[0106] The fourth processing subunit 7024 is configured to perform normalization processing on the smoothed seismic wave data. By performing Z-score standardization on the amplitude of each data point, standardized seismic wave data is obtained.

[0107] An extraction unit 703, configured to perform feature extraction based on the preprocessed seismic wave data. Among them, micro-fracture feature information is extracted through a preset multi-dimensional feature extraction model to obtain a multi-dimensional feature matrix, and the micro-fracture feature information includes fracture edge features, curvature features corresponding to fractures, frequency features corresponding to fractures, and similarity features corresponding to fractures;

[0108] Among them, the extraction unit 703 includes: a first extraction subunit 7031, a second extraction subunit 7032, a third extraction subunit 7033, a fourth extraction subunit 7034, and a fifth extraction subunit 7035.

[0109] The first extraction subunit 7031 is configured to extract crack edge features based on the preprocessed seismic wave data. By using the Sobel operator to calculate the gradient field of the seismic data, local gradient information is calculated in the three directions of X, Y, and Z to obtain crack edge features;

[0110] The second extraction subunit 7032 is configured to extract curvature features based on the crack edge features. By calculating the second derivative of the seismic wave data, the maximum curvature value of the local area is obtained and used as the curvature feature corresponding to the crack;

[0111] The third extraction subunit 7033 is configured to extract frequency features based on the preprocessed seismic wave data. By performing a fast Fourier transform on the preprocessed seismic wave data, the spectral information of the seismic signal is obtained, and the high-frequency components related to the crack are extracted therefrom to obtain the frequency feature corresponding to the crack;

[0112] The fourth extraction subunit 7034 is configured to extract similarity features based on the crack frequency features. By using the C3 coherence algorithm to calculate the similarity between adjacent gathers, the low-coherence area where the crack is located is highlighted and used as the similarity feature corresponding to the crack;

[0113] The fifth extraction subunit 7035 is configured to construct a matrix based on the crack edge features, the curvature features corresponding to the crack, the frequency features corresponding to the crack, and the similarity features corresponding to the crack to obtain a multi-dimensional feature matrix.

[0114] The prediction unit 704 is configured to perform modeling and training using a convolutional neural network model with a spatial attention mechanism based on the multi-dimensional feature matrix, and predict the probability of cracks occurring at each position to be detected based on the trained convolutional neural network model;

[0115] Among them, the prediction unit 704 includes a first prediction subunit 7041, a second prediction subunit 7042, a third prediction subunit 7043, and a fourth prediction subunit 7044.

[0116] The first prediction subunit 7041 is configured to perform an initial input of the convolutional neural network model based on the multi-dimensional feature matrix. By using the multi-dimensional matrix containing crack edge, curvature, frequency, and similarity features as the input of the convolutional neural network, the first-layer convolutional kernel of the network is initialized to obtain a preliminary feature representation of the convolutional neural network;

[0117] The second prediction subunit 7042 is configured to add a spatial attention mechanism based on the preliminary feature representation of the convolutional neural network. By performing channel attention and spatial attention weighting on the feature map output by the convolutional layer, a weighted feature map is obtained;

[0118] The third prediction subunit 7043 is configured to perform convolutional neural network training for a preset number of layers based on the weighted feature map, and obtain a trained convolutional neural network model through stacking multiple convolutional layers, pooling layers, and fully connected layers;

[0119] The fourth prediction subunit 7044 is configured to perform crack probability prediction based on the trained convolutional neural network model, and output the probability of cracks occurring at each position to be detected by inputting the seismic data to be detected into the trained model.

[0120] The detection unit 705 is configured to perform image conversion and optimization processing based on the probabilities of cracks occurring at all positions to be detected, obtain an optimized crack image, and use it as the detection result of micro-cracks in tight gas.

[0121] Among them, the detection unit 705 includes a first detection subunit 7051, a second detection subunit 7052, a third detection subunit 7053, and a fourth detection subunit 7054.

[0122] The first detection subunit 7051 is configured to perform image conversion processing based on the probability of cracks occurring at the position to be detected. By mapping the probability value of each position to be detected to the corresponding image pixel value, the probability value is converted into the pixel intensity of a grayscale image, and a preliminary crack probability grayscale image is obtained;

[0123] The second detection subunit 7052 is configured to perform noise removal and smoothing processing based on the preliminary crack probability grayscale image. By applying Gaussian filtering to remove isolated noise points caused by inaccurate crack positioning and smoothing the edges in the image, a denoised and smoothed crack probability image is obtained;

[0124] The third detection subunit 7053 is configured to perform morphological processing based on the denoised and smoothed crack probability image. By performing opening operation to remove isolated noise points and closing operation to connect discontinuous segments of cracks, an optimized crack connectivity image is obtained;

[0125] The fourth detection subunit 7054 is configured to perform crack probability threshold segmentation based on the optimized crack connectivity image. By binarizing the crack probability image with a preset threshold and marking the crack area, the final crack detection result is obtained.

[0126] Among them, an analysis unit 706 is further included after the detection unit 705.

[0127] The analysis unit 706 is configured to calculate the crack density, estimate the crack aperture, and analyze the connectivity of the optimized crack image in sequence to obtain a reservoir permeability distribution map.

[0128] Among them, the analysis unit 706 includes a first analysis subunit 7061, a second analysis subunit 7062, a third analysis subunit 7063, and a fourth analysis subunit 7064.

[0129] The first analysis subunit 7061 is configured to calculate the fracture density based on the optimized fracture image. By statistically calculating the ratio of the number of voxels in the fracture area to the total number of voxels, the distribution density of fractures per unit volume is obtained, and a fracture density map is constructed therefrom.

[0130] The second analysis subunit 7062 is configured to estimate the fracture aperture based on the fracture density map. By establishing a mapping between the amplitude and the aperture based on the least squares method, the correlation between the probability value of the fracture and the seismic wave amplitude data is fitted to obtain a fracture aperture distribution map.

[0131] The third analysis subunit 7063 is configured to analyze the fracture connectivity based on the fracture aperture distribution map. By applying the maximum flow algorithm, the connectivity in the fracture network is analyzed to identify the connected regions and the non-connected regions, and a fracture network connectivity map is obtained.

[0132] The fourth analysis subunit 7064 is configured to generate a reservoir permeability distribution map based on the fracture network connectivity map. By comprehensively considering the density, aperture, and connectivity information of the fractures, a weighted summation model is applied to calculate the permeability value of each region, and a permeability distribution map is constructed.

[0133] It should be noted that regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0134] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0135] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for detecting dense gas microcracks based on seismic data, characterized in that: include: Acquiring seismic wave data based on seismic exploration equipment, wherein the seismic wave data includes amplitude data, frequency data, and phase data reflecting underground rock formation characteristics; According to the seismic wave data, wavelet transform denoising, compressed sensing reconstruction and median filtering smoothing are performed in sequence to obtain pre-processed seismic wave data; Performing feature extraction based on the preprocessed seismic wave data, wherein microcrack feature information is extracted through a preset multidimensional feature extraction model to obtain a multidimensional feature matrix, wherein the microcrack feature information includes crack edge features, curvature features corresponding to the cracks, frequency features corresponding to the cracks, and similarity features corresponding to the cracks; According to the multi-dimensional feature matrix, a convolutional neural network model with a spatial attention mechanism is used for modeling and training, and the probability of cracks occurring at each position to be detected is predicted based on the trained convolutional neural network model; Image conversion and optimization processing are performed according to the probability of cracks appearing at all locations to be detected to obtain an optimized crack image, which is used as the dense gas microcrack detection result.

2. The method for detecting dense gas microcracks based on seismic data according to claim 1, characterized in that , performing wavelet transform denoising, compressed sensing reconstruction and median filtering smoothing processing in sequence according to the seismic wave data to obtain pre-processed seismic wave data, including: Performing wavelet transform denoising processing on the seismic wave data, performing multi-scale decomposition on the seismic data using Daubechies wavelet basis, decomposing the signal into low-frequency and high-frequency parts, and performing soft threshold filtering on the high-frequency noise based on a preset threshold to obtain denoised seismic wave data; Compressed sensing reconstruction is performed on the denoised seismic wave data. The seismic wave data is sparsely represented and reconstructed by applying the Basis Pursuit algorithm based on L1 norm minimization to obtain the reconstructed seismic wave data. Perform median filtering and smoothing processing on the seismic wave data reconstructed by compressed sensing, sort each data point, take the median, and remove local outliers by using a sliding window of a preset size to obtain smoothed seismic wave data; Normalization is performed on the smoothed seismic wave data, and the amplitude of each data point is processed by Z-score standardization to obtain standardized seismic wave data.

3. The method for detecting dense gas microcracks based on seismic data according to claim 1, characterized in that , feature extraction is performed according to the preprocessed seismic wave data, wherein microcrack feature information is extracted through a preset multidimensional feature extraction model to obtain a multidimensional feature matrix, wherein the microcrack feature information includes crack edge features, curvature features corresponding to the cracks, frequency features corresponding to the cracks, and similarity features corresponding to the cracks, including: Extract fracture edge features based on the preprocessed seismic wave data, calculate the gradient field of the seismic data using the Sobel operator, and calculate local gradient information in three directions, namely, X, Y, and Z, to obtain fracture edge features; Extracting curvature features according to the crack edge features, obtaining the maximum curvature value of the local area by calculating the second-order derivative of the seismic wave data, and using it as the curvature feature corresponding to the crack; Extracting frequency features according to the preprocessed seismic wave data, performing fast Fourier transform on the preprocessed seismic wave data to obtain frequency spectrum information of the seismic signal, and extracting high-frequency components related to the cracks therefrom to obtain frequency features corresponding to the cracks; Similarity feature extraction is performed according to the crack frequency feature, and the similarity between adjacent gathers is calculated by using the C3 coherence algorithm to highlight the low coherence area where the crack is located, and use it as the similarity feature corresponding to the crack; A matrix is ​​constructed based on the crack edge features, the curvature features corresponding to the cracks, the frequency features corresponding to the cracks and the similarity features corresponding to the cracks to obtain a multi-dimensional feature matrix.

4. The method for detecting dense gas microcracks based on seismic data according to claim 1, characterized in that: According to the multi-dimensional feature matrix, a convolutional neural network model with a spatial attention mechanism is used for modeling and training, and the probability of cracks occurring at each position to be detected is predicted based on the trained convolutional neural network model, including: According to the multi-dimensional feature matrix, initializing the input of the convolutional neural network model, by using the multi-dimensional matrix containing crack edge, curvature, frequency and similarity features as the input of the convolutional neural network, initializing the first layer of convolution kernel of the network, and obtaining the preliminary feature representation of the convolutional neural network; According to the preliminary feature representation of the convolutional neural network, a spatial attention mechanism is added, and a weighted feature map is obtained by weighting the channel attention and spatial attention on the feature map output by the convolutional layer; According to the weighted feature map, a preset number of layers of convolutional neural network training are performed, and a trained convolutional neural network model is obtained by stacking multiple convolution, pooling and fully connected layers; According to the trained convolutional neural network model, crack probability prediction is performed, and the probability of cracks occurring at each location to be detected is output by inputting the seismic data to be detected into the trained model.

5. The method for detecting dense gas microcracks based on seismic data according to claim 1, characterized in that: According to the probability of cracks appearing at all locations to be detected, image conversion and optimization processing are performed to obtain an optimized crack image, which is used as the dense gas microcrack detection result, including: According to the probability of cracks appearing at the positions to be detected, image conversion processing is performed, by mapping the probability value of each position to be detected to the corresponding image pixel value, converting the probability value into the pixel intensity of the grayscale image, and obtaining a preliminary crack probability grayscale image; According to the preliminary crack probability grayscale image, noise removal and smoothing processing are performed, and isolated noise points generated by inaccurate crack positioning are removed by applying Gaussian filtering, and edges in the image are smoothed to obtain a crack probability image after denoising and smoothing; According to the de-noised and smoothed crack probability image, morphological processing is performed to remove isolated noise points through an opening operation and to connect the crack discontinuity segments through a closing operation to obtain an optimized crack connectivity image; According to the optimized crack connectivity image, crack probability threshold segmentation is performed, the crack probability image is binarized by a preset threshold, the crack area is marked, and the final crack detection result is obtained.

6. A dense gas microcrack detection system based on seismic data, characterized in that: include: An acquisition unit, configured to acquire seismic wave data based on seismic exploration equipment, wherein the seismic wave data includes amplitude data, frequency data, and phase data reflecting underground rock formation characteristics; A processing unit, used for sequentially performing wavelet transform denoising, compressed sensing reconstruction and median filtering smoothing processing on the seismic wave data to obtain pre-processed seismic wave data; An extraction unit is used to perform feature extraction based on the preprocessed seismic wave data, wherein microcrack feature information is extracted through a preset multidimensional feature extraction model to obtain a multidimensional feature matrix, wherein the microcrack feature information includes crack edge features, curvature features corresponding to the cracks, frequency features corresponding to the cracks, and similarity features corresponding to the cracks; A prediction unit, configured to perform modeling and training using a convolutional neural network model with a spatial attention mechanism according to the multidimensional feature matrix, and predict the probability of cracks occurring at each position to be detected based on the trained convolutional neural network model; The detection unit is used to perform image conversion and optimization processing according to the probability of cracks appearing at all positions to be detected, obtain an optimized crack image, and use it as a dense gas microcrack detection result.

7. The dense gas microcrack detection system based on seismic data according to claim 6 is characterized in that: The processing unit comprises: A first processing subunit is used to perform wavelet transform denoising processing on the seismic wave data, perform multi-scale decomposition on the seismic data using Daubechies wavelet basis, decompose the signal into low-frequency and high-frequency parts, and perform soft threshold filtering on the high-frequency noise based on a preset threshold to obtain denoised seismic wave data; The second processing subunit is used to perform compressed sensing reconstruction processing on the denoised seismic wave data, and to obtain reconstructed seismic wave data by sparsely representing and reconstructing the seismic wave data by applying the Basis Pursuit algorithm based on L1 norm minimization; The third processing subunit is used to perform median filtering and smoothing processing on the seismic wave data reconstructed by compressed sensing, and to sort, median and remove local outliers of each data point by using a sliding window of a preset size to obtain smoothed seismic wave data; The fourth processing subunit is used to perform normalization processing on the smoothed seismic wave data, and process the amplitude of each data point through Z-score standardization to obtain standardized seismic wave data.

8. The dense gas microcrack detection system based on seismic data according to claim 6, characterized in that: The extraction unit comprises: A first extraction subunit is used to extract crack edge features according to the preprocessed seismic wave data, and calculate the local gradient information in the three directions of X, Y and Z by using the Sobel operator to calculate the gradient field of the seismic data to obtain the crack edge features; A second extraction subunit is used to extract curvature features according to the fracture edge features, obtain the maximum curvature value of the local area by calculating the second-order derivative of the seismic wave data, and use it as the curvature feature corresponding to the fracture; A third extraction subunit is used to extract frequency features according to the preprocessed seismic wave data, obtain frequency spectrum information of seismic signals by performing fast Fourier transform on the preprocessed seismic wave data, and extract high-frequency components related to cracks therefrom to obtain frequency features corresponding to the cracks; A fourth extraction subunit is used to extract similarity features according to the crack frequency features, and to calculate the similarity between adjacent gathers by using a C3 coherence algorithm, to highlight the low coherence area where the crack is located, and to use it as the similarity feature corresponding to the crack; The fifth extraction subunit is used to construct a matrix based on the crack edge features, the curvature features corresponding to the cracks, the frequency features corresponding to the cracks, and the similarity features corresponding to the cracks to obtain a multi-dimensional feature matrix.

9. The dense gas microcrack detection system based on seismic data according to claim 6, characterized in that: The prediction unit comprises: A first prediction subunit is used to perform initialization input of a convolutional neural network model according to the multidimensional feature matrix, by using the multidimensional matrix containing crack edge, curvature, frequency and similarity features as the input of the convolutional neural network, initializing the first layer of convolution kernel of the network, and obtaining a preliminary feature representation of the convolutional neural network; A second prediction subunit is used to add a spatial attention mechanism according to the preliminary feature representation of the convolutional neural network, and obtain a weighted feature map by weighting the channel attention and the spatial attention on the feature map output by the convolutional layer; A third prediction subunit is used to perform a preset number of convolutional neural network training layers according to the weighted feature map, and obtain a trained convolutional neural network model by stacking multiple convolution, pooling and fully connected layers; The fourth prediction subunit is used to predict the probability of cracks according to the trained convolutional neural network model, and output the probability of cracks appearing at each position to be detected by inputting the seismic data to be detected into the trained model.

10. The dense gas microcrack detection system based on seismic data according to claim 6, characterized in that: The detection unit comprises: The first detection subunit is used to perform image conversion processing according to the probability of cracks appearing at the position to be detected, by mapping the probability value of each position to be detected to the corresponding image pixel value, converting the probability value into the pixel intensity of the grayscale image, and obtaining a preliminary crack probability grayscale image; The second detection subunit is used to perform noise removal and smoothing processing on the preliminary crack probability grayscale image, remove isolated noise points caused by inaccurate crack positioning by applying Gaussian filtering, and smooth the edges in the image to obtain a crack probability image after denoising and smoothing; The third detection subunit is used to perform morphological processing on the crack probability image after denoising and smoothing, remove isolated noise points through opening operation, and connect the crack discontinuity segments through closing operation to obtain an optimized crack connectivity image; The fourth detection subunit is used to perform crack probability threshold segmentation according to the optimized crack connectivity image, binarize the crack probability image through a preset threshold, mark the crack area, and obtain the final crack detection result.

Images