Fault automatic detection method based on multi-feature fusion and structural guiding filtering

Abstract

The invention discloses a fault automatic detection method based on multi-feature fusion and structural guiding filtering, and relates to the technical field of geological data detection, and the method comprises the steps: obtaining a three-dimensional coherent body which is corresponding to seismic data of a target region and reflects the stratum discontinuity; processing the fault features in the three-dimensional coherent body through a multi-scale multi-direction filter to generate a fault boundary enhanced image; according to the fault inclination angle in the three-dimensional coherent body, direction diffusion coefficients corresponding to the principal direction gradient and the vertical direction gradient are calculated; based on the directional diffusion coefficient, performing iterative diffusion filtering processing on the three-dimensional coherent body to generate a fault continuity enhanced image; and generating a fault detection result based on a fusion image between the fault boundary enhanced image and the fault continuity enhanced image. According to the invention, the false detection rate during fault identification is reduced, and the identification precision is improved.

Description

Technical Field

[0001] This application relates to the field of geological data detection technology, specifically to an automatic fault detection method based on multi-feature fusion and structural guidance filtering. Background Technology

[0002] A fault is a geological structure in which strata fracture under stress, and the strata on both sides of the fracture surface undergo significant relative displacement. Fault identification research is applied to all stages of oil and gas field exploration, evaluation, and development. Accurate fault identification is of great guiding significance for the rational division of development units, the adjustment and optimization of well placement plans, and the reduction of development risks.

[0003] Currently, fault identification based on seismic data is the foundation of fault research, mainly including the following methods: seismic reflection feature analysis, which determines the location, strike, and displacement of faults by analyzing the characteristics of seismic reflection wave phase axis misalignment, waveform distortion, and amplitude abrupt changes; and frequency division technology, which performs wavelet transform on seismic data to extract fault-sensitive frequency band single-frequency data volumes, analyzes their amplitude anomaly characteristics, and performs fault identification. However, existing methods are subject to noise interference and fault discontinuity during the identification process, and do not take into account the data characteristics in these scenarios, resulting in a high false detection rate and low identification accuracy in fault identification. Summary of the Invention

[0004] To address the technical problems in related technologies, such as noise interference and discontinuous fractures during the identification process, and the failure to consider the data characteristics in these scenarios, which leads to high false detection rates and low identification accuracy in tomography, this application provides an automatic tomography detection method based on multi-feature fusion and constructed guided filtering.

[0005] The specific technical solution adopted is as follows: Obtain the three-dimensional coherence volume reflecting stratigraphic discontinuities corresponding to the seismic data of the target area; Fault features in the three-dimensional coherence volume are processed by multi-scale, multi-directional filters to generate enhanced fault boundary images; Based on the fault dip angle in the three-dimensional coherent volume, the directional diffusion coefficients corresponding to the principal gradient and the vertical gradient are calculated. Based on the directional diffusion coefficient, iterative diffusion filtering is performed on the three-dimensional coherent volume to generate an image with enhanced tomographic continuity. Fault detection results are generated based on the fusion image between the fault boundary enhancement image and the fault continuity enhancement image.

[0006] In one possible implementation of this application, fault features in a three-dimensional coherent volume are processed using a multi-scale, multi-directional filter to generate an enhanced fault boundary image, including: Based on the fault direction features in the three-dimensional coherence volume, a fault image is generated; The tomographic image is convolved using a multi-scale, multi-directional filter to obtain the first enhanced image. By comparing and analyzing the pixel values ​​in the first enhanced image, an enhanced image of the fault boundary is obtained.

[0007] In one possible implementation of this application, generating a tomographic image based on the fault direction features in a three-dimensional coherence volume includes: Two-dimensional planar data are extracted from the three-dimensional coherence volume according to the time depth direction to obtain coherence volume slice data; Based on the horizontal and vertical coordinates of each pixel in the coherence slice data, the slope and aspect of each pixel are calculated. The slope and aspect are used to characterize the fault direction features in the three-dimensional coherence. Based on the slope and aspect, the first pixel value of each pixel after fault enhancement is calculated. Based on the first pixel value, each pixel is updated to generate a tomographic image.

[0008] In one possible implementation of this application, a first enhanced image is obtained by convolution processing of a tomographic image using a multi-scale, multi-directional filter, including: Determine the multiple partition scales and partition directions of the multi-scale, multi-directional filter; Based on different segmentation scales and directions, multi-scale and multi-directional filters are used to perform convolution processing on tomographic images; The maximum value of the convolution result of each pixel at different scales and directions is used as the filtering result of each pixel. Based on the filtering results, the first enhanced image is generated.

[0009] In one possible implementation of this application, a comparative analysis of pixel values ​​in a first enhanced image is performed to obtain a fault boundary enhanced image, including: Determine the global statistical values ​​and the neighborhood median of each pixel in the first enhanced image. The global statistical values ​​include the pixel median, the preset high pixel threshold, and the preset low pixel threshold. Based on the neighborhood median, a preset high pixel threshold, and a preset low pixel threshold, the pixel values ​​of each pixel are compared and analyzed to determine the fault type of each pixel; the fault types include strong faults, undetermined faults, and non-faults. Using pixels in strong faults as seed points, a region growing algorithm is used to determine the connectivity of neighboring pixels of the seed points, thereby obtaining a set of fault pixels within a preset growth range. Based on the fault pixel set, an enhanced image of the fault boundary is generated.

[0010] In one possible implementation of this application, the pixel values ​​of each pixel are compared and analyzed based on the neighborhood median, a preset high pixel threshold, and a preset low pixel threshold to determine the tomographic type of each pixel, including: If the pixel value of a pixel is greater than or equal to the preset pixel height threshold, the pixel will be classified as a strong fault. If a pixel value is greater than or equal to a preset low pixel threshold, less than a preset high pixel threshold, and greater than the median value of its neighborhood, then the pixel is assigned to a layer to be determined. If the pixel value of a pixel is less than or equal to the median of its neighborhood, or less than a preset low pixel threshold, then the pixel is classified as a non-discontinuous layer.

[0011] In one possible embodiment of this application, the directional diffusion coefficient includes a principal directional diffusion coefficient and a vertical directional diffusion coefficient. Based on the fault dip angle in the three-dimensional coherence volume, the directional diffusion coefficients corresponding to the principal directional gradient and the vertical directional gradient are calculated, including: Based on the fault dip angle in the three-dimensional coherence volume, a principal direction gradient operator and a vertical direction gradient operator are constructed. The principal gradient operator and the vertical gradient operator are convolved with the coherence slice data to obtain the principal gradient and the vertical gradient. The coherence slice data is obtained by truncating the three-dimensional coherence. The principal gradient is used to characterize the edge continuity characteristics of the fault along the dip direction, and the vertical gradient is used to characterize noise or non-fault interference signals perpendicular to the fault strike. Based on the main direction gradient, a main direction diffusion coefficient is constructed, and a preset fixed diffusion coefficient is used as the vertical direction diffusion coefficient corresponding to the vertical direction gradient.

[0012] In one possible implementation of this application, an iterative diffusion filtering process is performed on a three-dimensional coherent volume based on the directional diffusion coefficient to generate a tomographic continuity enhanced image, including: Determine the time step in the iterative diffusion process; Based on the time step, the magnitude of the gradient in the principal direction, the magnitude of the gradient in the vertical direction, the diffusion coefficient in the principal direction, and the diffusion coefficient in the vertical direction, an iterative diffusion filtering process is performed on the three-dimensional coherent volume to obtain an image with enhanced tomographic continuity.

[0013] In one possible implementation of this application, a fault detection result is generated based on a fused image between a fault boundary enhancement image and a fault continuity enhancement image, including: The fault boundary enhancement image is fused with the fault continuity enhancement image to obtain a fused image; Based on a pre-defined morphological algorithm, discontinuous faults within the image are connected and fused to generate fault detection results.

[0014] In one possible implementation of this application, based on a preset morphological algorithm, discontinuous faults within an image are connected and fused to generate a fault detection result, including: Based on a pre-defined morphological algorithm, a closing operation is performed on the discontinuous faults inside the fused image to obtain connected fault bands. Edge pixel stripping is performed on the connected fault zone to obtain the fault strike line; The fault strike line is converted into a set of discrete coordinate points, and the set of discrete coordinate points is used as the fault detection result. The fault detection result includes fault location, fault length, and fault strike information.

[0015] This application has, but is not limited to, the following technical effects: By acquiring the three-dimensional coherence volume reflecting the stratigraphic discontinuity corresponding to the seismic data of the target area, and processing the fault features in the three-dimensional coherence volume through the maximum response mechanism of multi-scale and multi-directional filters, a fault boundary enhancement image is generated. This overcomes the problem of insufficient response of single-scale filters to small faults or faults with specific strikes, significantly reducing the false detection rate. Based on the fault dip angle in the three-dimensional coherence volume, the directional diffusion coefficients corresponding to the principal direction gradient and the vertical direction gradient are calculated. Then, based on the directional diffusion coefficients, iterative diffusion filtering is performed on the three-dimensional coherence volume to generate a fault continuity enhancement image. Adaptive coefficients are used to enhance signal continuity in the principal direction and suppress noise diffusion in the vertical direction, effectively solving the problem of fragmented and discontinuous fault identification results and improving noise resistance. Furthermore, the fault boundary enhancement image and the fault continuity enhancement image are fused. By fusing fault boundary features (geometric morphology) and continuity features (spatial extension) and combining morphological closing operations, isolated noise interference points are filtered out, ensuring the high reliability and geological rationality of the fault identification results, reducing the high false detection rate and improving the identification accuracy. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the first embodiment of the automatic fault detection method based on multi-feature fusion and construction-guided filtering in this application. Figure 2 This is a schematic diagram of the overall implementation process of the automatic fault detection method based on multi-feature fusion and construction-guided filtering in this application; Figure 3 This is a schematic diagram of the image corresponding to the fault detection result of the automatic fault detection method based on multi-feature fusion and construction-guided filtering in this application; Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0018] This application provides an automatic fault detection method based on multi-feature fusion and construction-guided filtering. In the first embodiment of the automatic fault detection method based on multi-feature fusion and construction-guided filtering in this application, referring to... Figure 1 ,include: Step S10: Obtain the three-dimensional coherence volume reflecting the stratigraphic discontinuity corresponding to the seismic data of the target area.

[0019] As an example, the automatic fault detection method based on multi-feature fusion and structure-guided filtering can be applied to an automatic fault detection device based on multi-feature fusion and structure-guided filtering. The automatic fault detection device based on multi-feature fusion and structure-guided filtering belongs to the automatic fault detection system based on multi-feature fusion and structure-guided filtering, and the automatic fault detection system based on multi-feature fusion and structure-guided filtering belongs to the automatic fault detection equipment based on multi-feature fusion and structure-guided filtering.

[0020] As an example, the automatic fault detection method adopted in this application achieves accurate fault identification based on multi-feature fusion and structural guidance filtering. Specifically, an embodiment of this application uses oilfield A as an example to illustrate the implementation process of the method. Currently, fault identification is labor-intensive and inefficient; fault combinations are difficult, and conventional identification methods have low accuracy. Seismic data is used to describe the fault distribution characteristics, laying the foundation for determining the overall fault development characteristics and constructing a three-dimensional structural model. Therefore, taking area A as an example, research on automatic fault identification methods is carried out. The specific analysis process is as follows: Figure 2 As shown.

[0021] As an example, the seismic data includes, but is not limited to, three-dimensional post-stack seismic data volumes (data format is SEGY standard format), survey line coordinate information, seismic wave dominant frequency parameters (dominant frequency range 15~30Hz), and target layer time-depth calibration data. The amplitude data of the post-stack seismic traces in the acquired seismic data are standardized to avoid the influence of amplitude dimension differences on data analysis. After standardization, the amplitude range is [-1,1].

[0022] As an example, a three-dimensional coherence volume can be a three-dimensional data volume reflecting the waveform similarity of seismic traces within a local neighborhood of seismic data. Standardized data can be used as input, and the three-dimensional coherence volume can be calculated using dip scanning and cross-correlation algorithms. The dip angle of faults can also be obtained through dip scanning. Specifically: By quantifying the correlation of reflection signals from adjacent seismic traces, the coherence value range [0, 1] reflects the stratigraphic continuity characteristics; low coherence regions characterize faults or geological discontinuities and are marked as potential faults; the specific dip angle scanning range is 0°~30°, the step size is 5°, and the pixel window size is 3×3 during the coherence value calculation; the specific construction process of the three-dimensional coherence volume is well known to those skilled in the art and will not be described in detail here.

[0023] Step S20: The fault features in the three-dimensional coherent volume are processed by a multi-scale, multi-directional filter to generate an enhanced fault boundary image.

[0024] As an example, a tomographic enhancement image is generated based on the constructed 3D coherence volume. The tomographic enhancement image is then convolved using a multi-scale, multi-directional filter to generate a tomographic boundary enhancement image. The global statistical threshold of the tomographic boundary enhancement image is then compared with the local neighborhood median. The tomographic pixels are labeled using a region growing algorithm. Based on the above processing of the acoustic field tomographic boundary enhancement volume, the purpose is to consider the tomographic features of different scales and orientations, avoid the problem of missing small faults and oblique faults by a single-scale or directional filter, accurately distinguish real tomographic pixels from noise interference, and thus provide clear pixel-level labels for subsequent continuity enhancement.

[0025] As an example, a multi-scale, multi-directional filter can be a Step filter, specifically one that operates along a specified direction. The width on both sides of the detection center pixel is The filter, which measures the difference in the mean of two adjacent windows, performs convolution processing on the fault features in the three-dimensional coherent volume at different directions and scales to obtain the filtering result. Then, based on the filtering result, an enhanced image of the fault boundary is generated.

[0026] As an example, the enhanced image of fault boundary captures the enhanced features of each pixel in the image at the optimal scale and direction, thereby highlighting the significant signals of different types of faults and reducing the interference from non-target directions and scales. This image reflects the precise boundary and distribution range of faults within the target segment.

[0027] The automatic fault detection step S20 based on multi-feature fusion and guided filtering also includes steps S21 to S23, including: Step S21: Generate a fault image based on the fault direction features in the three-dimensional coherence volume.

[0028] As an example, the fault orientation features include the slope and aspect of each pixel. After determining the slope and aspect, the rate of change of each pixel in the slope and aspect is determined by analyzing and calculating these two features. The enhanced pixel value of the fault features is then calculated, and the existing pixel value is updated with the enhanced pixel value to obtain the fault image.

[0029] As an example, a fault image can be a two-dimensional image composed of updated pixel values, where a larger pixel value indicates that the key structural features such as fault edges and inflection points reflected in the area may be more significant; a smaller pixel value indicates that the continuous stratigraphic features or non-fault-interference area features reflected in the corresponding area are more significant.

[0030] Step S21 includes: Two-dimensional planar data are extracted from the three-dimensional coherence volume along the time depth direction to obtain coherence volume slice data.

[0031] As an example, the constructed three-dimensional coherence body is sliced. The specific slicing process is as follows: based on the time-depth calibration data of the target reservoir segment, two-dimensional planar data is extracted from the three-dimensional coherence body along the time-depth direction at intervals of 2ms / slice to obtain a series of coherence body slices covering the target layer segment, that is, coherence body slice data. The purpose of this step is to reduce the three-dimensional data to two-dimensional planar data, thereby accurately analyzing the fault characteristics of the target layer.

[0032] Based on the horizontal and vertical coordinates of each pixel in the coherence slice data, the slope and aspect of each pixel are calculated. The slope and aspect are used to characterize the fault direction features in the three-dimensional coherence.

[0033] As an example, based on three-dimensional coherence volume slices, the slope and aspect of each pixel are calculated. The three-dimensional coherence volume data includes the coherence value corresponding to each spatial location in each slice. Each pixel refers to the smallest spatial unit in the coherence volume slice divided according to the survey line coordinates.

[0034] Specifically, taking the p-th pixel as an example, the slope of the pixel... and slope The calculation formula is as follows:

[0035]

[0036] in, Indicates the first The slope of each pixel; Indicates the first The slope direction of each pixel; Indicating the first coherent slice data The coherence value of each pixel; and These represent the horizontal and vertical coordinates of the survey line for the coherent body slice, respectively. and These represent the horizontal and vertical gradients of the coherent volume slice data, respectively, i.e., the... Pixels in , The rate of change of coherence value in the direction is calculated using the Sobel operator in a 3×3 neighborhood; This represents the single-parameter arctangent function, whose range is... This is used to convert gradient magnitude into slope angle; This represents the two-parameter arctangent function, whose range is... It is used to determine the azimuth angle of the slope direction based on the sign and ratio of the gradient components.

[0037] Based on the slope and aspect, the first pixel value of each pixel after fault enhancement is calculated.

[0038] As an example, based on the above calculations, the slope and aspect of each pixel are obtained. The first derivative of the slope and aspect of each pixel is then calculated, and the absolute value of the second derivative of the slope is obtained, denoted as... The absolute value of the calculated second derivative of the slope aspect is denoted as... ; Reflecting the local rate of change of slope, Reflecting the local rate of change of slope aspect, the first pixel value after fault enhancement is calculated based on the slope and slope aspect. :

[0039] in, Indicates the first The first pixel value after tomographic enhancement of each pixel corresponds to a higher value for more significant tomographic features at that location. It is a very small positive number greater than zero; Indicates the first The slope of each pixel is raised to the power of n to enhance the slope feature and highlight the slope difference between fault areas and continuous strata areas. Indicates the first The local rate of change of slope at each pixel reflects the degree of local abrupt change in slope, particularly in the fault edge region. The value will increase significantly; Indicates the first The local rate of change of slope aspect at each pixel reflects the degree of local directional change of the slope aspect. This is used in the calculation of... When calculating, the minimum angle difference is used (i.e., min(∣ab∣, 360-∣ab∣)), where a and b represent the slope aspect values ​​of two pixels, rather than directly subtracting the values, to prevent false high gradients in the due north direction; fault inflection point region. The value will increase significantly; The weighting coefficient represents the slope feature, and its value is an integer from 1 to 3. In this embodiment, the value is 2, which is used to balance the contribution of slope, slope gradient and slope aspect ratio, and avoid a single feature from overdoing the enhancement result.

[0040] Based on the first pixel value, each pixel is updated to generate a tomographic image.

[0041] Step S22: The tomographic image is convolved using a multi-scale, multi-directional filter to obtain the first enhanced image.

[0042] As an example, since a single-scale or single-direction filter cannot cover fault features of different scales and orientations, a multi-scale, multi-directional filter is used to process the fault image to obtain a first enhanced image.

[0043] As an example, the first enhanced image can be filtered to extract the enhanced image data with the best features at different scales and orientations.

[0044] Step S22 includes: Determine the multiple partition scales and partition directions of the multi-scale, multi-directional filter.

[0045] As an example, the filter width is , indicating the first The width value at each scale can be 2, 4, or 6 pixels; the division direction is... , indicating the first In one embodiment of this application, the number of directions is 18, the direction interval is 10°, and the direction coverage range is 0°~180°.

[0046] Based on different segmentation scales and directions, multi-scale and multi-directional filters are used to perform convolution processing on tomographic images; The maximum value of the convolution result of each pixel at different scales and directions is used as the filtering result of each pixel. Based on the filtering results, the first enhanced image is generated.

[0047] As an example, based on the above division, multi-scale and multi-directional convolution processing is performed on each pixel. To maximize the enhancement of features at different scales and with different tomographic orientations, the maximum value of the convolution results at different scales and in different directions for each pixel is used as the filtering result for each pixel; thus, the first enhanced image is obtained. The formula for its calculation is:

[0048] in, This represents all multi-scale, multi-directional images, i.e., tomographic enhancement images processed using different directions. Different widths The series of intermediate images obtained after convolution with the Step filter; among them, Indicates the selected angle value; Indicates the selected width value; This means taking the maximum value among the convolution results for each pixel across all directions and scales; specifically, for each pixel in the multi-scale and multi-directional images, the maximum value among all calculated results is taken. and direction The result with the largest response value is selected as the final filtering result for that pixel. Based on the above processing, the enhanced features of each pixel under the optimal scale and orientation are obtained, thereby highlighting the significant signals of different types of faults and reducing the interference of non-target orientation and scale.

[0049] Step S23: Perform comparative analysis on the pixel values ​​in the first enhanced image to obtain the enhanced image of the fault boundary.

[0050] As an example, considering the problem that a single global threshold can easily lead to misjudgment of isolated noise points or omission of weak fault signals, the pixel values ​​of the first enhanced image are compared with the global statistical values ​​and the local neighborhood median. Based on the comparison results, a fault boundary enhanced image is generated.

[0051] Step S23 includes: Determine the global statistical values ​​and the neighborhood median of each pixel in the first enhanced image. The global statistical values ​​include the pixel median, the preset high pixel threshold, and the preset low pixel threshold.

[0052] As an example, the global statistic is ,in To enhance the image, all pixel values ​​are sorted from largest to smallest. The median of the first 50% of pixel values ​​after sorting is also known as the pixel median. This represents the preset pixel high threshold, which is set to a value greater than the maximum pixel value in the enhanced image. The average pixel value; This represents the preset low pixel threshold, which is set to a value less than the minimum pixel value in the enhanced image. The mean of pixel values; global statistics reflect the overall distribution characteristics of pixel values ​​in the enhanced image, defining the intensity range of the tomographic signal; the median of the neighborhood is... The value is the median of all pixel values ​​in a 3×3 local neighborhood centered on each pixel.

[0053] Based on the neighborhood median, a preset high pixel threshold, and a preset low pixel threshold, the pixel values ​​of each pixel are compared and analyzed to determine the fault type of each pixel; the fault types include strong faults, undetermined faults, and non-faults.

[0054] As an example, the pixel value of each pixel is compared with the neighborhood median, a preset high pixel threshold, and a preset low pixel threshold to determine the fault type of each pixel, such as strong fault, undetermined fault, and non-fault.

[0055] The specific steps for determining the fault type of each pixel include: If the pixel value of a pixel is greater than or equal to the preset pixel height threshold, the pixel will be classified as a strong fault. If a pixel value is greater than or equal to a preset low pixel threshold, less than a preset high pixel threshold, and greater than the median value of its neighborhood, then the pixel is assigned to a layer to be determined. If the pixel value of a pixel is less than or equal to the median of its neighborhood, or less than a preset low pixel threshold, then the pixel is classified as a non-discontinuous layer.

[0056] Specifically, if the pixel value of a pixel satisfies Then the pixels are divided into strong faults. The pixels; if the pixels satisfy ,and Then the pixels are divided into undetermined layers. The pixels; if the pixels satisfy ,and Then the pixels are divided into non-discontinuous layers. The pixels.

[0057] Using pixels in strong faults as seed points, a region growing algorithm is used to determine the connectivity of neighboring pixels of the seed points, thereby obtaining a set of fault pixels within a preset growth range.

[0058] Based on the fault pixel set, an enhanced image of the fault boundary is generated.

[0059] As an example, when classifying individual pixels into different tomographic types, each pixel is labeled. Based on the classification and labeling results, a region growing algorithm is used to process the image containing each pixel, thus determining the undetermined layers. Medium and strong faults Connected pixels are classified as discontinuous, while disconnected pixels are classified as non-discontinuous. Specifically: strong fault The pixels are used as seed points, and the connectivity of neighboring pixels is determined by the region growing algorithm. Specifically, the growth range is 4 neighborhoods (preset growth range), and then the set of connected fault pixels is obtained to obtain the fault boundary enhancement image, which reflects the accurate boundary and distribution range of the fault in the target segment. The detailed processing of the region growing algorithm is well known to those skilled in the art and will not be described in detail here.

[0060] Step S30: Based on the fault dip angle in the three-dimensional coherent volume, calculate the directional diffusion coefficients corresponding to the principal direction gradient and the vertical direction gradient. The directional diffusion coefficients include the principal direction diffusion coefficient and the vertical direction diffusion coefficient.

[0061] As an example, to enhance the spatial continuity and edge features of faults and suppress noise interference in non-fault areas, this embodiment constructs a directional gradient operator based on the fault dip angle for processing. The main directional gradient kernel is constructed based on the fault dip angle, and the vertical directional gradient kernel is orthogonal to the main direction. Iterative optimization is performed using an adaptive diffusion coefficient and a fixed vertical diffusion coefficient, and edge filling is performed to prevent distortion.

[0062] As an example, the main direction diffusion coefficient and the vertical direction diffusion coefficient can be coefficients used in subsequent weighted calculations of the main direction gradient and the vertical direction gradient for iterative diffusion optimization of the image.

[0063] Step S30 includes: Based on the fault dip angle in the three-dimensional coherent volume, a principal direction gradient operator and a vertical direction gradient operator are constructed.

[0064] As an example, a principal direction gradient kernel operator is generated based on the fault dip angle to enhance the edge response in a specific direction. The vertical gradient kernel is orthogonal to the principal direction gradient kernel. Since random noise and lithological abrupt changes in non-fault regions are mostly distributed along directions perpendicular to the fault strike, it is necessary to suppress gradient components orthogonal to the principal direction gradient kernel to force smoothing of non-edge regions. The aim is to accurately separate fault signals from noise interference, prevent noise from spreading along the vertical direction and disrupting the continuity of the fault, while highlighting the extension characteristics of the fault along the principal direction.

[0065] The principal gradient operator and the vertical gradient operator are convolved with the coherence slice data to obtain the principal gradient and the vertical gradient. The coherence slice data is obtained by truncating the three-dimensional coherence. The principal gradient is used to characterize the edge continuity of the fault along the dip direction, and the vertical gradient is used to characterize noise or non-fault interference signals perpendicular to the fault strike.

[0066] As an example, the principal gradient and vertical gradient are obtained by convolving the generated principal gradient kernel and vertical gradient kernel with coherent volume slices.

[0067] In an embodiment of an oilfield A area, the principal directional gradient kernel operator is generated based on a fault dip angle θ = 45°. Specifically, a 3×3 diagonal gradient kernel is set with the 45° dip angle as the core direction. The weights within the kernel are symmetrically distributed along the 45° direction, with the highest weight values ​​in the core region. For example, the rotation matrix could be... To enhance the gradient response in this direction, the principal directional gradient kernel can accurately capture the strike characteristics of the fault in oilfield area A, amplifying the continuity signal of the fault along the 45° dip direction. The vertical directional gradient kernel is orthogonal to the principal directional gradient kernel, corresponding to a dip angle θ + 90° = 135°. The vertical directional gradient kernel is a 3×3 orthogonal directional gradient kernel, with weights distributed along the 135° direction within the kernel. The core region has lower weight values ​​compared to the principal directional kernel. For example... The non-fault gradient response in this direction is weakened; based on the constructed principal direction kernel gradient operator and vertical direction gradient kernel operator, convolution is performed with each coherent volume slice obtained from seismic data of oilfield A area to obtain the principal direction gradient of the pixel value at each location in each coherent volume slice. and vertical gradient The principal direction gradient reflects the edge continuity characteristics of the fault along the dip direction. High-value areas correspond to the core extension zone of the fault, which can clearly show the strike and extension trend of the fault. The vertical direction gradient reflects noise or non-fault interference signals perpendicular to the fault strike. High-value areas are mostly random noise, lithological changes, or thin-layer reflection interference areas. The principal direction kernel gradient operator and the vertical direction gradient kernel operator can adopt the Sobel operator. The specific process of convolution based on the principal direction kernel gradient operator and the vertical direction gradient kernel operator is well known to those skilled in the art and will not be described in detail here.

[0068] The specific matrix in the embodiment is only a special case when the dip angle θ = 45°. The general generation method is to construct a gradient operator for the reference direction (such as horizontal) and use the rotation matrix R = [cosθ, -sinθ; sinθ, cosθ] to rotate it to the direction of the fault dip angle θ.

[0069] Based on the main direction gradient, a main direction diffusion coefficient is constructed, and a preset fixed diffusion coefficient is used as the vertical direction diffusion coefficient corresponding to the vertical direction gradient.

[0070] As an example, the main direction diffusion coefficient is dynamically adjusted based on the local gradient magnitude, and the formula for its adjustment is:

[0071] in, Indicates the diffusion coefficient in the main direction; As a diffusion sensitivity parameter, diffusion is weaker in edge regions with larger gradients. In one embodiment of this application, The value is 0.5; This represents the magnitude of the gradient in the principal direction.

[0072] As an example, the vertical diffusion coefficient In other embodiments, it can also be set to 0.2~0.4. For example, 0.2 is used for low signal-to-noise ratio data to enhance noise suppression, and 0.4 is used for high signal-to-noise ratio data to retain more details. The purpose of taking a fixed value for the vertical diffusion coefficient is to maintain the weak diffusion intensity perpendicular to the fault strike direction, continuously suppress noise interference in this direction, avoid noise being amplified or propagated during the diffusion process, and thus force a low diffusion coefficient to suppress the influence of noise.

[0073] Step S40: Based on the directional diffusion coefficient, perform iterative diffusion filtering on the three-dimensional coherent volume to generate an image with enhanced tomographic continuity.

[0074] As an example, a fault continuity enhancement image can be image data obtained by iterative diffusion filtering of pixels in different directions in a three-dimensional coherent volume, thereby ensuring the accuracy of fault edge preservation and noise suppression.

[0075] Step S40 includes: Determine the time step in the iterative diffusion process; Based on the time step, the magnitude of the gradient in the principal direction, the magnitude of the gradient in the vertical direction, the diffusion coefficient in the principal direction, and the diffusion coefficient in the vertical direction, an iterative diffusion filtering process is performed on the three-dimensional coherent volume to obtain an image with enhanced tomographic continuity.

[0076] As an example, using time steps Controlling the diffusion process, in which, It can take any value between 0.1 and 0.3 to balance iteration efficiency and optimization effect.

[0077] As an example, image optimization can be achieved by iteratively updating the diffusion coefficient of the main direction using repeated convolution. The number of iterations can be 20. Specifically, a pixel-by-pixel iterative update method can be used. The purpose is to: gradually smooth residual noise in non-edge regions through multiple iterations; enhance the spatial continuity of fault edges and connect fault segments truncated by noise; use a replicate mode for boundary filling to avoid edge distortion; and use the extracted directional feature maps to perform weighted superposition correction on the original data.

[0078] in, Indicates the first The image data corresponding to the three-dimensional coherence volume after the next iteration, after a round of diffusion optimization, highlights the fault continuity features. Indicates the first Image data after the next iteration; This represents the time step, which is 0.25 in one embodiment of this application; and These represent the diffusion coefficients in the principal direction and the vertical direction, respectively; and These are the feature maps that reflect the energy changes in the main direction and vertical direction, respectively. This represents element-wise multiplication, where the diffusion coefficient and gradient of pixels at the same location in two images are multiplied separately. This achieves precise matching and weighting of the diffusion coefficient and gradient, ensuring that diffusion optimization only applies to pixels in the corresponding direction, thus guaranteeing the accuracy of tomographic edge preservation and noise suppression. The formula represents the pixel value of the current image. Add a "correction factor". The correction factor is determined by the second derivative (trend of change) and the diffusion coefficient (intensity); because the second derivative can be positive or negative, the pixel value will be finely adjusted according to the surrounding environment (smoothing noise, connecting breakpoints), rather than monotonically increasing and causing the image to whiten.

[0079] Step S50: Based on the fused image between the fault boundary enhancement image and the fault continuity enhancement image, generate the fault detection result.

[0080] As an example, the fault boundary enhancement image and the fault continuity enhancement image are fused to obtain a fused image that combines the features of the two images, and then the fault detection results are extracted from the fused image.

[0081] Step S50 includes steps S51 to S52: Step S51: The fault boundary enhancement image and the fault continuity enhancement image are fused together to obtain a fused image.

[0082] As an example, a logical AND operation is performed between the fault boundary enhancer and the continuity enhancer to generate a composite attribute image, i.e., a fused image. Before fusion, the fault boundary continuity enhancer image is binarized (e.g., a threshold is set to extract high-response areas) to ensure that it can be logically ANDed with the fault boundary image. The purpose is to accurately select fault regions with clear boundaries and good continuity, reduce the impact of isolated noise points and blurred false fault regions on fault identification, and thus accurately reflect the spatial distribution and complete contour of faults within the target segment.

[0083] Step S52: Based on a preset morphological algorithm, connect and fuse discontinuous faults within the image to generate fault detection results.

[0084] As an example, the preset morphological algorithm can be a mathematical morphology algorithm, which connects discontinuous faults within the fused image to obtain various displayable fault lines, directions, etc., and then generates fault detection results.

[0085] Step S52 includes: Based on a pre-defined morphological algorithm, a closing operation is performed on the discontinuous faults within the fused image to obtain connected fault zones.

[0086] Specifically, the morphological closing operation (connecting fractures) takes a composite attribute image as input, selects a structuring element (such as a circular or square kernel) of a preset size (e.g., 3×3 or 5×5), and first performs a dilation operation on the fault pixel regions in the image to fill in the fine fractures and voids inside the fault lines; then it performs an erosion operation to restore the original thickness of the fault lines. Through the closing operation, the originally discontinuous fault segments are connected into a connected fault zone.

[0087] Edge pixel stripping is performed on the connected fault zone to obtain the fault direction line.

[0088] As an example, a thinning algorithm (such as the Zhang-Suen thinning algorithm) is applied to the tomographic image after the closing operation, successively stripping pixels from the edges of the fault band until the fault band of a certain width is reduced to a central skeleton line retaining only a single pixel width. This step removes the thickness redundancy of the fault, resulting in a regular fault strike line.

[0089] The fault strike line is converted into a set of discrete coordinate points, and the set of discrete coordinate points is used as the fault detection result. The fault detection result includes fault location, fault length, and fault strike information.

[0090] As an example, based on the thinned single-pixel skeleton line, an eight-neighbor chain code tracing algorithm or an edge tracing algorithm is used to traverse all connected pixels in the image, transforming each connected fault skeleton line into a series of discrete coordinate point sets (i.e., vector data). The final output is a vector discretization result containing fault location, length, and orientation information (i.e., the final fault detection result). A schematic diagram of the fault detection result is shown below. Figure 3 As shown, in Figure 3 The white lines in the diagram represent the identified faults.

[0091] In one embodiment of this application, the above method was used to perform automatic fault identification processing in area A. The identification results were as follows: 1,270 faults were initially identified, with the faults mainly trending northwest. Based on the identification results, 12 fracturing and perforation measures were implemented, resulting in a cumulative increase of 11,300 tons of oil. The fault identification accuracy rate reached 96%.

[0092] This application provides an automatic fault detection method based on multi-feature fusion and structurally guided filtering. It acquires a three-dimensional coherence volume reflecting stratigraphic discontinuities corresponding to seismic data of the target area. Through the maximum response mechanism of a multi-scale, multi-directional filter, fault features in the three-dimensional coherence volume are processed to generate enhanced fault boundary images. This overcomes the problem of insufficient response of single-scale filters to small faults or faults with specific strikes, significantly reducing the false negative rate. Based on the fault dip angle in the three-dimensional coherence volume, the directional diffusion coefficients corresponding to the principal direction gradient and the vertical direction gradient are calculated. Then, based on the directional diffusion coefficients, the three-dimensional... The coherent volume undergoes iterative diffusion filtering to generate a fault continuity enhancement image. Adaptive coefficients are used to enhance signal continuity in the main direction and suppress noise diffusion in the vertical direction, effectively solving the problem of fragmented and discontinuous fault identification results and improving noise resistance. Furthermore, the fault boundary enhancement image is fused with the fault continuity enhancement image. By fusing fault boundary features (geometric morphology) and continuity features (spatial extension) and combining morphological closing operations, isolated noise interference points are filtered out, ensuring high reliability and geological rationality of the fault identification results, reducing the high false detection rate and improving identification accuracy.

[0093] Reference Figure 4 , Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0094] like Figure 4 As shown, the automatic tomography detection device based on multi-feature fusion and construction-guided filtering may include: a processor 1001, a memory 1003, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1003.

[0095] Optionally, the automatic tomographic detection device based on multi-feature fusion and guided filtering may also include a user interface, a network interface, a camera, RF (Radio Frequency) circuitry, sensors, a WiFi module, etc. The user interface may include a display screen and an input submodule such as a keyboard; optionally, the user interface may also include standard wired or wireless interfaces. The network interface may include standard wired or wireless interfaces (such as a Wi-Fi interface).

[0096] Those skilled in the art will understand that Figure 4 The structure of the automatic fault detection device based on multi-feature fusion and construction-guided filtering shown in the figure does not constitute a limitation on the automatic fault detection device based on multi-feature fusion and construction-guided filtering. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0097] like Figure 4 As shown, the memory 1003, serving as a storage medium, may include an operating system, a network communication module, and an automatic tomography detection program based on multi-feature fusion and construction-guided filtering. The operating system is a program that manages and controls the hardware and software resources of the automatic tomography detection device based on multi-feature fusion and construction-guided filtering, supporting the operation of the automatic tomography detection program based on multi-feature fusion and construction-guided filtering, as well as other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1003, as well as communication with other hardware and software in the automatic tomography detection system based on multi-feature fusion and construction-guided filtering.

[0098] exist Figure 4 In the automatic fault detection device based on multi-feature fusion and construction-guided filtering shown, the processor 1001 is used to execute the automatic fault detection program based on multi-feature fusion and construction-guided filtering stored in the memory 1003 to implement the steps of the automatic fault detection method based on multi-feature fusion and construction-guided filtering described above.

[0099] The specific implementation of the automatic fault detection device based on multi-feature fusion and structure-guided filtering in this application is basically the same as the embodiments of the automatic fault detection method based on multi-feature fusion and structure-guided filtering described above, and will not be repeated here.

[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0101] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0103] The above are merely preferred embodiments of this application and do not limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

[0104] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0105] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An automatic fault detection method based on multi-feature fusion and guided filtering, characterized in that, The method includes: Obtain the three-dimensional coherence volume reflecting stratigraphic discontinuities corresponding to the seismic data of the target area; The fault features in the three-dimensional coherent volume are processed by a multi-scale, multi-directional filter to generate an enhanced fault boundary image. Based on the fault dip angle in the three-dimensional coherent body, the directional diffusion coefficients corresponding to the principal gradient and the vertical gradient are calculated. Based on the directional diffusion coefficient, the three-dimensional coherent volume is subjected to iterative diffusion filtering to generate a tomographic continuity enhanced image. A fault detection result is generated based on the fused image between the fault boundary enhancement image and the fault continuity enhancement image.

2. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 1, characterized in that, The step of processing the fault features in the three-dimensional coherence volume using a multi-scale, multi-directional filter to generate an enhanced fault boundary image includes: Based on the fault direction features in the three-dimensional coherence volume, a fault image is generated; The tomographic image is convolved using a multi-scale, multi-directional filter to obtain a first enhanced image. The pixel values ​​in the first enhanced image are compared and analyzed to obtain the fault boundary enhanced image.

3. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 2, characterized in that, The step of generating a tomographic image based on the fault direction features in the three-dimensional coherence volume includes: Two-dimensional planar data from the three-dimensional coherence volume are extracted according to the time depth direction to obtain coherence volume slice data; Based on the horizontal and vertical coordinates of each pixel in the coherence slice data, the slope and aspect of each pixel are calculated. The slope and aspect are used to characterize the fault direction features in the three-dimensional coherence. Based on the slope and aspect, the first pixel value of each pixel after fault enhancement is calculated. Based on the first pixel value, each pixel is updated to generate a tomographic image.

4. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 2, characterized in that, The step of convolving the tomographic image using a multi-scale, multi-directional filter to obtain the first enhanced image includes: Determine the multiple partition scales and partition directions of the multi-scale, multi-directional filter; Based on different division scales and division directions, the tomographic image is convolved using a multi-scale, multi-directional filter. The maximum value of the convolution result of each pixel at different scales and directions is used as the filtering result of each pixel. Based on the filtering results, a first enhanced image is generated.

5. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 2, characterized in that, The step of comparing and analyzing the pixel values ​​in the first enhanced image to obtain the fault boundary enhanced image includes: Determine the global statistical values ​​in the first enhanced image and the neighborhood median of each pixel. The global statistical values ​​include the pixel median, a preset high pixel threshold, and a preset low pixel threshold. Based on the neighborhood median, the preset high pixel threshold, and the preset low pixel threshold, the pixel values ​​of each pixel are compared and analyzed to determine the fault type of each pixel; the fault type includes strong fault, undetermined fault, and non-fault. Using the pixels in the strong fault as seed points, the connectivity of the neighboring pixels of the seed points is determined by the region growing algorithm to obtain a set of fault pixels within a preset growth range. Based on the set of fault pixels, an enhanced image of the fault boundary is generated.

6. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 5, characterized in that, The step of comparing and analyzing the pixel values ​​of each pixel based on the neighborhood median, a preset high pixel threshold, and a preset low pixel threshold to determine the fault type of each pixel includes: If the pixel value of the pixel is greater than or equal to a preset pixel height threshold, then the pixel is classified into a strong fault. If the pixel value of the pixel is greater than or equal to a preset low pixel threshold, less than a preset high pixel threshold, and greater than the median value of the neighborhood, then the pixel is assigned to a layer to be determined. If the pixel value of the pixel is less than or equal to the median of the neighborhood, or less than a preset low pixel threshold, then the pixel is classified into a non-discontinuous layer.

7. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 1, characterized in that, The directional diffusion coefficient includes a principal directional diffusion coefficient and a vertical directional diffusion coefficient. The calculation of the directional diffusion coefficients corresponding to the principal directional gradient and the vertical directional gradient based on the fault dip angle in the three-dimensional coherence volume includes: Based on the fault dip angle in the three-dimensional coherence volume, a principal direction gradient operator and a vertical direction gradient operator are constructed. The principal direction gradient operator and the vertical direction gradient operator are convolved with the coherence slice data to obtain the principal direction gradient and the vertical direction gradient. The coherence slice data is obtained by truncating the three-dimensional coherence. The principal direction gradient is used to characterize the edge continuity characteristics of the fault along the dip direction, and the vertical direction gradient is used to characterize noise or non-fault interference signals perpendicular to the fault strike. Based on the main direction gradient, a main direction diffusion coefficient is constructed, and a preset fixed diffusion coefficient is used as the vertical direction diffusion coefficient corresponding to the vertical direction gradient.

8. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 7, characterized in that, The step of performing iterative diffusion filtering on the three-dimensional coherence volume based on the directional diffusion coefficient to generate an enhanced tomographic continuity image includes: Determine the time step in the iterative diffusion process; Based on the time step, the magnitude of the gradient in the main direction, the magnitude of the gradient in the vertical direction, the diffusion coefficient in the main direction, and the diffusion coefficient in the vertical direction, the three-dimensional coherent volume is subjected to iterative diffusion filtering to obtain an image with enhanced tomographic continuity.

9. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 1, characterized in that, The process of generating a fault detection result based on the fused image between the fault boundary enhancement image and the fault continuity enhancement image includes: The enhanced image of the fault boundary is fused with the enhanced image of the fault continuity to obtain a fused image; Based on a preset morphological algorithm, discontinuous faults within the fused image are connected to generate fault detection results.

10. The automatic fault detection method based on multi-feature fusion and construction-guided filtering as described in claim 9, characterized in that, The step of connecting discontinuous faults within the fused image based on a preset morphological algorithm to generate fault detection results includes: Based on a preset morphological algorithm, a closing operation is performed on the discontinuous faults inside the fused image to obtain connected fault bands. Edge pixel stripping is performed on the connected fault zone to obtain the fault direction line; The fault strike line is converted into a set of discrete coordinate points, and the set of discrete coordinate points is used as the fault detection result, wherein the fault detection result includes fault location, fault length, and fault strike information.

Images