A low-quality seismic data fault identification method and system

Abstract

The application provides a low-quality seismic data fault identification method and system, and belongs to the field of seismic data processing.The technical scheme is as follows: step S1, performing filtering processing on original seismic data to obtain preprocessed seismic data; step S2, performing non-subsampled shear wave transformation on the preprocessed seismic data to obtain enhanced seismic data; step S3, performing parallel calculation on the enhanced seismic data to obtain a fusion feature body; and step S4, extracting an initial fault body from the fusion feature body by adopting an Otsu dynamic threshold algorithm with a directional constraint to obtain a final three-dimensional fault model.The application has the beneficial effect that by means of adaptive filtering, multi-scale feature enhancement and an intelligent fusion mechanism, the precision and continuity of fault identification in low signal-to-noise ratio data are effectively improved without sample training.

Description

Technical Field

[0001] This invention relates to the field of seismic data processing, and in particular to a method and system for fault identification of low-quality seismic data. Background Technology

[0002] The fundamental structural interpretation work for oil and gas field exploration and development relies on seismic data, and the quality of the seismic data directly affects reservoir evaluation, well location deployment, and development plan formulation. In recent years, deep learning technology has shown potential in automatic fault detection, but its performance is highly dependent on the quality of the seismic data. Existing low-quality seismic data identification technologies face the following bottlenecks: (1) Traditional edge detection algorithms (such as Canny and Sobel) are sensitive to low signal-to-noise ratio seismic data and are prone to generating false faults; (2) Conventional coherent volume algorithms (such as C3 and gradient structure tensor) have poor continuity in low-resolution data interruption layers; (3) Deep learning relies on a large number of labeled samples, but low-quality data labeling is costly and lacks generalization. Existing technologies are unable to simultaneously characterize the multi-scale features of faults in terms of longitudinal resolution and lateral extension, which affects the model's accuracy in identifying micro-faults and complex fracture systems. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for identifying faults in low-quality seismic data, which effectively improves the accuracy and continuity of fault identification in low signal-to-noise ratio data without the need for sample training by using adaptive filtering, multi-scale feature enhancement and intelligent fusion mechanisms.

[0004] This invention is achieved through the following measures: A method for fault identification using low-quality seismic data, characterized by the following steps: Step S1: Filtering the original seismic data using the original seismic data itself as a guide field, and dynamically adjusting the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data; Step S2: Performing non-subsampled shear wave transformation on the preprocessed seismic data, decomposing it into sub-bands of multiple scales and directions, and performing nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data; Step S3: Parallel calculation of coherence volume features and phase symmetry features on the enhanced seismic data, and adaptively fusing the coherence volume features and phase symmetry features using a local entropy weighted model to obtain a fused feature body; Step S4: Extracting the initial fault body from the fused feature body using the Otsu dynamic thresholding algorithm with directional constraints, and performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model.

[0005] The invention also has the following specific features: The filtering process of the original seismic data using the original seismic data itself as the guide field includes: using the amplitude distribution of the original seismic data itself as the guide field, calculating the amplitude similarity between the center point and the neighboring points within a local window to obtain the filtering weight, performing a weighted average calculation on the seismic data within the window based on the filtering weight, and obtaining the basic filtered seismic data volume by traversing the entire data volume.

[0006] The filtering process dynamically adjusts the filtering parameters based on the local signal-to-noise ratio of the data to obtain preprocessed seismic data. This includes: calculating the signal-to-noise ratio estimate in real time within the local window; adaptively selecting a combination of parameters for the filtering window size and regularization coefficient based on the different threshold ranges of the signal-to-noise ratio estimate; applying the selected parameter combination to the weight calculation and weighted averaging process in the guided filtering process; optimizing the parameters of the basic filtered seismic data volume; and outputting the preprocessed seismic data.

[0007] Performing a non-downsampled shear wave transform on the preprocessed seismic data to decompose it into subbands of multiple scales and directions includes: performing multi-scale decomposition on the preprocessed seismic data to obtain frequency components at different scales; performing multi-directional decomposition on the frequency components at each scale to obtain subband components in different directions; and obtaining a subband set containing low-frequency subbands and multiple high-frequency subbands through the multi-scale decomposition and multi-directional decomposition process.

[0008] The high-frequency sub-bands obtained after decomposition are subjected to nonlinear enhancement processing based on gradient amplitude to obtain enhanced seismic data. This includes: calculating the gradient amplitude of each data point in the high-frequency sub-band, processing the gradient amplitude using a nonlinear enhancement function, fusing the processed gradient amplitude with the corresponding high-frequency sub-band data, and reconstructing all fused sub-bands through inverse transformation to obtain enhanced seismic data.

[0009] Parallel calculation of coherence volume features and phase symmetry features of the enhanced seismic data includes: taking the enhanced seismic data as input, and simultaneously executing two independent processes: coherence volume feature calculation and phase symmetry feature calculation; The coherence volume feature calculation obtains characteristic data representing stratigraphic discontinuities by analyzing the waveform similarity between adjacent seismic traces, while the phase symmetry feature calculation obtains characteristic data representing edge symmetry by analyzing the local phase distribution through a symmetric filter bank. Through these two parallel calculation processes, complete coherence volume feature data and phase symmetry feature data are output respectively.

[0010] The coherence volume features and phase symmetry features are adaptively fused using a local entropy weighted model to obtain a fused feature volume, which includes: taking the coherence volume feature data volume and the phase symmetry feature data volume as input, and calculating the information entropy values ​​of the two feature data volumes respectively within a local spatial window; Based on the calculated information entropy value, the fusion weights of the coherent volume features and the phase symmetry features are dynamically determined according to the preset weight allocation rules. The determined fusion weights are used to perform a weighted summation calculation on the two feature data volumes to generate an initial fused feature volume. The initial fused feature volume is then subjected to selective filtering based on information entropy analysis, and the filtering operation is applied only to regions where the information entropy value is higher than a preset threshold. Finally, the fused feature volume is output.

[0011] The initial tomographic body is extracted from the fused feature body using the Otsu dynamic thresholding algorithm with directional constraints. This includes: based on the local structural trend data of the fused feature body, introducing directional constraints during the Otsu thresholding process, calculating the optimal segmentation threshold along the structural trend direction, and using the optimal segmentation threshold to perform binarization segmentation on the fused feature body to generate the initial tomographic body.

[0012] The process of performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model includes: determining the orientation of the structural elements for the morphological operations based on the local tectonic dip data; performing erosion and dilation operations on the initial fault body sequentially using the orientation-adaptive structural elements; connecting fault segments and eliminating noise through the directional morphological operations; and outputting the three-dimensional fault model.

[0013] A fault identification system for low-quality seismic data, characterized by comprising: a data preprocessing module: using the original seismic data itself as a guide field to filter the original seismic data, and dynamically adjusting the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data; multi-scale edge enhancement: performing non-subsampled shear wave transformation on the preprocessed seismic data to decompose it into sub-bands of multiple scales and directions, and performing nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data; feature extraction and fusion: calculating coherence volume features and phase symmetry features in parallel on the enhanced seismic data, and adaptively fusing the coherence volume features and phase symmetry features using a local entropy weighted model to obtain a fused feature body; fault extraction and optimization: extracting initial fault bodies from the fused feature body using the Otsu dynamic threshold algorithm with directional constraints, and performing directional morphological operations adapted to the local tectonic dip on the initial fault bodies to obtain the final three-dimensional fault model.

[0014] The beneficial effects of this invention are as follows: This invention accurately preserves fault edges while suppressing noise through adaptive guided filtering, effectively overcoming the blurring effect of traditional filtering; it significantly improves the detection capability of weak faults and discontinuous features in low signal-to-noise ratio data by utilizing a multi-scale edge enhancement mechanism; it adopts a multi-attribute fusion strategy based on local entropy weighting, which enhances the spatial continuity of real faults while suppressing false anomalies. The entire method requires no training samples, has high computational efficiency, and is particularly suitable for complex geological conditions with a signal-to-noise ratio of less than 2dB, solving the key problem of insufficient fault identification accuracy in low-quality seismic data. Attached Figure Description

[0015] Figure 1 The above is an overall flowchart of a low-quality seismic data fault identification method provided in an embodiment of the present invention. Detailed Implementation

[0016] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0017] Example 1 See Figure 1 A method for fault identification using low-quality seismic data includes the following steps: Step S1: Filtering the original seismic data using the original seismic data itself as a guide field, and dynamically adjusting the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data; Step S2: Performing non-subsampled shear wave transformation on the preprocessed seismic data, decomposing it into sub-bands of multiple scales and directions, and performing nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data; Step S3: Parallel calculation of coherence volume features and phase symmetry features on the enhanced seismic data, and adaptively fusing the coherence volume features and phase symmetry features using a local entropy weighted model to obtain a fused feature body; Step S4: Extracting the initial fault body from the fused feature body using the Otsu dynamic threshold algorithm with directional constraints, and performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model.

[0018] The filtering process of the original seismic data using the original seismic data itself as the guide field includes: using the amplitude distribution of the original seismic data itself as the guide field, calculating the amplitude similarity between the center point and the neighboring points within a local window to obtain the filtering weight, performing a weighted average calculation on the seismic data within the window based on the filtering weight, and obtaining the basic filtered seismic data volume by traversing the entire data volume.

[0019] During the filtering process, the filtering parameters are dynamically adjusted based on the local signal-to-noise ratio of the data to obtain the preprocessed seismic data, which includes: calculating the signal-to-noise ratio estimate in real time within the local window; adaptively selecting the parameter combination of the filtering window size and regularization coefficient according to the different threshold ranges of the signal-to-noise ratio estimate; applying the selected parameter combination to the weight calculation and weighted averaging process in the guided filtering process; performing parameter optimization processing on the basic filtered seismic data volume; and outputting the preprocessed seismic data.

[0020] The process of performing non-downsampled shear wave transform on the preprocessed seismic data to decompose it into subbands of multiple scales and directions includes: performing multi-scale decomposition on the preprocessed seismic data to obtain frequency components at different scales; performing multi-directional decomposition on the frequency components at each scale to obtain subband components in different directions; and obtaining a set of subbands containing low-frequency subbands and multiple high-frequency subbands through the multi-scale decomposition and multi-directional decomposition process.

[0021] The high-frequency sub-bands obtained after decomposition are subjected to nonlinear enhancement processing based on gradient amplitude to obtain enhanced seismic data. This process includes: calculating the gradient amplitude of each data point in the high-frequency sub-band, processing the gradient amplitude using a nonlinear enhancement function, fusing the processed gradient amplitude with the corresponding high-frequency sub-band data, and reconstructing all fused sub-bands using inverse transformation to obtain enhanced seismic data.

[0022] Parallel computation of coherence volume features and phase symmetry features on enhanced seismic data includes: using enhanced seismic data as input, simultaneously executing two independent processes: coherence volume feature computation and phase symmetry feature computation; The coherence volume feature calculation obtains characteristic data representing stratigraphic discontinuities by analyzing the waveform similarity between adjacent seismic traces, while the phase symmetry feature calculation obtains characteristic data representing edge symmetry by analyzing the local phase distribution through a symmetric filter bank. Through these two parallel calculation processes, complete coherence volume feature data and phase symmetry feature data are output respectively.

[0023] Furthermore, a local entropy weighted model is used to adaptively fuse coherent volume features and phase symmetry features to obtain a fused feature body, which includes: taking coherent volume feature data body and phase symmetry feature data body as input, calculating the information entropy values ​​of the two feature data bodies respectively within a local spatial window; Based on the calculated information entropy value, the fusion weights of coherent volume features and phase symmetry features are dynamically determined according to the preset weight allocation rules. The determined fusion weights are used to perform a weighted summation calculation on the two feature data volumes to generate an initial fused feature volume. The initial fused feature volume is then subjected to selective filtering based on information entropy analysis, and the filtering operation is applied only to regions where the information entropy value is higher than a preset threshold. Finally, the fused feature volume is output.

[0024] The Otsu dynamic thresholding algorithm with directional constraints is used to extract the initial fault body from the fused feature body. This includes: based on the local structural trend data of the fused feature body, introducing directional constraints in the Otsu thresholding process, calculating the optimal segmentation threshold along the structural trend direction, and using the optimal segmentation threshold to perform binarization segmentation of the fused feature body to generate the initial fault body.

[0025] The process involves performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model. This includes: determining the orientation of the structural elements for the morphological operations based on the local tectonic dip data; performing erosion and dilation operations on the initial fault body sequentially using the orientation-adaptive structural elements; connecting fault segments and eliminating noise through directional morphological operations; and outputting the three-dimensional fault model.

[0026] A fault identification system for low-quality seismic data, characterized by comprising: a data preprocessing module: using the original seismic data itself as a guide field to filter the original seismic data, and dynamically adjusting the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data; multi-scale edge enhancement: performing non-subsampled shear wave transformation on the preprocessed seismic data to decompose it into sub-bands of multiple scales and directions, and performing nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data; feature extraction and fusion: calculating coherence volume features and phase symmetry features in parallel on the enhanced seismic data, and using a local entropy weighted model to adaptively fuse the coherence volume features and phase symmetry features to obtain a fused feature body; fault extraction and optimization: using the Otsu dynamic threshold algorithm with directional constraints to extract the initial fault body from the fused feature body, and performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model.

[0027] Example 2 See Figure 1 A method for fault identification of low-quality seismic data includes the following steps: Step S1: Filter the original seismic data using the original seismic data itself as a guide field, and dynamically adjust the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain pre-processed seismic data. This includes: using the amplitude distribution of the original seismic data itself as a guide field, calculating the amplitude similarity between the center point and neighboring points within a local window to obtain filtering weights, performing a weighted average calculation on the seismic data within the window based on the filtering weights, and obtaining the basic filtered seismic data volume by traversing the entire data volume. The signal-to-noise ratio (SNR) estimate is calculated in real time within a local window. Based on the different threshold ranges of the SNR estimate, the combination of parameters for the filter window size and regularization coefficient is adaptively selected. The selected parameter combination is applied to the weight calculation and weighted averaging process in the guided filtering process to optimize the parameters of the basic filtered seismic data volume and output the preprocessed seismic data.

[0028] Specifically, in step S1, the original low-quality seismic data is preprocessed using an adaptive guided filtering method. Its core innovation lies in constructing a filtering mechanism with the original seismic data itself as the guiding field, and realizing dynamic self-optimization of filtering parameters based on the local signal-to-noise ratio, thereby effectively suppressing random noise while accurately preserving structural edge information such as faults. The system first uses the original seismic data itself as the guide field. During the processing, a local sliding window is set up to move point by point on the data volume. For each center point within the window, a weight matrix is ​​constructed by calculating the amplitude difference between it and all its neighboring points. This weight matrix ensures that in uniform regions with gentle amplitude changes, neighboring points contribute significantly to the smoothing result of the center point, achieving effective noise reduction. However, at structural boundaries with significant amplitude differences, the weight of neighboring points crossing the boundary is automatically reduced, thereby suppressing the smoothing operation and protecting the sharpness of the fault edges. The key to step S1 lies in the dynamic optimization mechanism of the filtering parameters. The system calculates the signal-to-noise ratio (SNR) within each local window in real time and automatically selects the most suitable combination of filtering parameters accordingly. The filtering radius (i.e., the window size) is adaptively adjusted within a dynamic range of 3×3 to 9×9 based on the SNR level: in low SNR regions, the algorithm prioritizes using a smaller filtering window; in high SNR regions with uniform construction, a larger filtering window is used. At the same time, the regularization coefficient is also dynamically adjusted in tandem, using a lower coefficient value in low SNR regions to protect weak construction traces, and a higher coefficient value in high SNR regions to achieve sufficient smoothness. Through the filtering based on the data's own guiding field and the dynamic adjustment of parameters according to the local signal-to-noise ratio, step S1 finally outputs a preprocessed seismic data. This data suppresses random noise to the maximum extent while preserving and highlighting key structural boundary information, providing a high-quality data foundation for subsequent multi-scale edge enhancement steps.

[0029] Step S2: Perform non-downsampled shear wave transform on the preprocessed seismic data to decompose it into subbands of multiple scales and directions. Then, perform nonlinear enhancement processing based on gradient amplitude on the high-frequency subbands obtained after decomposition to obtain enhanced seismic data. This includes: performing multi-scale decomposition on the preprocessed seismic data to obtain frequency components at different scales, performing multi-directional decomposition on the frequency components at each scale to obtain subband components in different directions, and obtaining a subband set containing low-frequency subbands and multiple high-frequency subbands through the multi-scale decomposition and multi-directional decomposition process. The gradient magnitude of each data point in the high-frequency sub-band is calculated, and the gradient magnitude is processed using a nonlinear enhancement function. The processed gradient magnitude is then fused with the corresponding high-frequency sub-band data. Finally, an inverse transformation is performed on all fused sub-bands to reconstruct the enhanced seismic data.

[0030] Specifically, in step S2, the system first constructs a non-subsampled shear wave transform with a directional controllable pyramid for the seismic data preprocessed in step S1. Specifically, the input data is decomposed at multiple scales through a non-subsampled pyramid filter bank, for example, set to 4 different scale layers. At each scale layer, directional localization processing is further performed through a shear wave filter. By adjusting the shear matrix parameters, the data at each scale is decomposed into multiple directions, for example, set to 8 or 16 directions. The entire decomposition process maintains the downsampling operation to ensure that all subbands have the same spatial size as the original data. The final output is a set of subbands containing a low-frequency subband and multiple high-frequency subbands of different scales and orientations. The low-frequency subband mainly carries the background structure and stratigraphic information in the seismic data, while the high-frequency subband contains detailed features such as fault edges and discontinuities. Subsequently, the algorithm focuses on targeted enhancement of high-frequency subbands containing fault edge details. For each high-frequency subband, the gradient magnitude of each data point is calculated to quantify the edge strength at that point. Then, a nonlinear enhancement transformation based on an exponential function is applied to process this gradient magnitude. The enhancement intensity of this nonlinear enhancement function is jointly controlled by parameters α and σ, where parameter α mainly determines the magnitude of enhancement, and parameter σ determines the sensitivity range of the enhancement operation. Through this nonlinear transformation, the originally blurred or weak edge response is targeted enhanced, while background noise and irrelevant stratigraphic details are relatively suppressed.

[0031] Finally, all high-frequency subbands after gradient amplitude enhancement processing are input together with the low-frequency subbands that retain their original information, and an inverse non-subsampled shear wave transform is performed. Through this inverse transform process, all processed subbands are accurately reconstructed back into the seismic data space, and a new enhanced seismic data volume with targeted enhancement of edge features at multiple scales is finally output. Step S2 overcomes the shortcomings of traditional single-scale edge detection methods in responding to multi-scale faults in complex geological structures through multi-scale and multi-directional analysis using non-subsampled shear wave transform. Its design, which combines gradient amplitude and nonlinear enhancement, can effectively amplify weak edge signals related to faults in low signal-to-noise ratio data, while suppressing random noise and stratigraphic background, thus solving the key problem of missing weak fault responses in low-quality data by conventional methods.

[0032] Step S3: Parallel calculation of coherence volume features and phase symmetry features of the enhanced seismic data, and adaptive fusion of coherence volume features and phase symmetry features using a local entropy weighted model to obtain the fused feature volume. This includes: taking the enhanced seismic data as input, and simultaneously executing two independent processes: coherence volume feature calculation and phase symmetry feature calculation. The coherence volume feature calculation obtains characteristic data representing stratigraphic discontinuities by analyzing the waveform similarity between adjacent seismic traces, while the phase symmetry feature calculation obtains characteristic data representing edge symmetry by analyzing the local phase distribution through a symmetric filter bank. Through these two parallel calculation processes, complete coherence volume feature data and phase symmetry feature data are output respectively. Using coherence volume feature data and phase symmetry feature data as input, the information entropy values ​​of the two feature data volumes are calculated separately within a local spatial window; Based on the calculated information entropy value, the fusion weights of coherent volume features and phase symmetry features are dynamically determined according to the preset weight allocation rules. The determined fusion weights are used to perform a weighted summation calculation on the two feature data volumes to generate an initial fused feature volume. The initial fused feature volume is then subjected to selective filtering based on information entropy analysis, and the filtering operation is applied only to regions where the information entropy value is higher than a preset threshold. Finally, the fused feature volume is output.

[0033] Specifically, in step S3, the system first calculates two feature attributes in parallel on the enhanced seismic data; The first parallel process calculates third-generation coherence volume (C3) features based on local waveform similarity. This process defines a three-dimensional analysis window on the data volume and generates a coherence volume feature data volume that is sensitive to waveform discontinuities caused by fault zones by calculating the degree of waveform similarity between adjacent seismic traces within the window. The second parallel process simultaneously calculates phase symmetry features based on phase information. This process constructs even-symmetric and odd-symmetric filter banks through local cosine transform, analyzes the local phase distribution of each data point, calculates its local energy function and phase deviation function, and finally generates a phase symmetry feature data volume that is more robust to illumination changes and signal-to-noise ratio fluctuations.

[0034] Subsequently, an adaptive fusion strategy based on local entropy weighting was adopted to fuse the two features. Within a local spatial window, the information entropy of the coherence volume feature data volume and the information entropy of the phase symmetry feature data volume were calculated separately. and ; Based on the calculated information entropy value, and according to the weight allocation rules... and The fusion weights are dynamically determined, and the two feature data volumes are weighted and summed using the determined fusion weights. Generate the initial fusion feature body.

[0035] Finally, selective filtering based on information entropy analysis is applied to the initial fused feature volume. The system identifies regions in the initial fused feature volume with information entropy higher than a preset threshold as high uncertainty regions, and applies improved nonlocal mean filtering only to these regions, while leaving the original data unchanged for regions with information entropy lower than the preset threshold. Step S3 overcomes the limitation of incomplete response of a single feature in low-quality data interruption layer by parallel computing of two complementary features: coherence volume and phase symmetry. The local entropy weighted fusion strategy innovatively uses information entropy as a measure of feature reliability and realizes adaptive allocation of fusion weights.

[0036] Step S4: The initial fault body is extracted from the fused feature body using the Otsu dynamic thresholding algorithm with directional constraints, and directional morphological operations adapted to the local structural dip are performed on the initial fault body to obtain the final three-dimensional fault model, including: based on the local structural dip data of the fused feature body, directional constraints are introduced in the Otsu thresholding process, the optimal segmentation threshold is calculated along the structural dip direction, and the fused feature body is binarized and segmented using the optimal segmentation threshold to generate the initial fault body; Based on local tectonic tendency data, the orientation of the structuring elements for morphological operations is determined. The erosion and dilation operations are then performed sequentially on the initial fault body using the orientation-adaptive structuring elements. Fault segments are connected and noise is eliminated through directional morphological operations, resulting in a three-dimensional fault model.

[0037] Specifically, step S4 uses an improved Otsu dynamic thresholding algorithm to extract the initial binary tomographic body from the fused feature body. The core innovation of this improved algorithm is that it introduces directional constraints when calculating the inter-class variance. Specifically, based on the fusion feature volume, local structural dip data is obtained, which represents the estimated spatial distribution direction of the fault. When performing threshold segmentation, the algorithm does not uniformly calculate the inter-class variance within the entire local window, but strictly calculates it along the local structural dip direction. This directional constraint mechanism ensures that the segmentation process is consistent with the geological geometric characteristics of the fault, thereby effectively avoiding interference from high-intensity non-fault edges (such as strong reflection phase axes), obtaining more accurate initial segmentation results, and generating the initial fault body.

[0038] Subsequently, a series of post-processing operations are applied to optimize the initial extraction results. The most critical of these is the directional morphological operation. Based on the same local structural trend data, the system determines the orientation of the structural elements for morphological operations (including directional erosion and dilation), ensuring that the long axis direction is strictly adaptively adjusted according to the local structural trend or gradient direction. First, the directional erosion operation is performed, in which the structural elements extend along the fault trend direction, which can intelligently suppress and eliminate false anomalies with inconsistent orientations. Then, the directional dilation operation is performed, using structural elements with the same adaptive orientation to connect fractured fault segments interrupted by noise and fill small voids.

[0039] Through the processing flow that combines threshold segmentation with directional constraints and directional morphological optimization, the final output is a three-dimensional fault model with clear boundaries, good spatial continuity, and the ability to truly reflect the underground geological structure.

[0040] Step S4 integrates geological structural dip information into the image segmentation and post-processing, solving the key technical challenges of fragmented fault extraction results and numerous false anomalies in conventional algorithms for low-quality data. The directionally constrained Otsu algorithm significantly improves the geological rationality of the initial segmentation, while directional morphological operations effectively improve the spatial continuity of faults while preserving the true fault geometry. This series of targeted optimization measures greatly enhances the accuracy and reliability of fault interpretation in low-quality seismic data, providing a reliable technical means for structural analysis under complex geological conditions.

[0041] Example 3 A fault identification system for low-quality seismic data, characterized by comprising: a data preprocessing module: using the original seismic data itself as a guide field to filter the original seismic data, and dynamically adjusting the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data; multi-scale edge enhancement: performing non-subsampled shear wave transformation on the preprocessed seismic data to decompose it into sub-bands of multiple scales and directions, and performing nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data; feature extraction and fusion: calculating coherence volume features and phase symmetry features in parallel on the enhanced seismic data, and adaptively fusing the coherence volume features and phase symmetry features using a local entropy weighted model to obtain a fused feature body; fault extraction and optimization: extracting initial fault bodies from the fused feature body using the Otsu dynamic threshold algorithm with directional constraints, and performing directional morphological operations adapted to the local tectonic dip on the initial fault bodies to obtain the final three-dimensional fault model.

[0042] Example 4 A Neogene hydrocarbon-rich area in a depression zone of an oilfield in the Bohai Bay Basin has a complex tectonic evolution history, with multiple superimposed tectonic movements, intersecting fault systems, and low signal-to-noise ratio, making conventional fault identification methods ineffective. This experiment uses a typical terrestrial 3D seismic survey area in this region, covering approximately 120 km², targeting the Neogene Guantao Formation. The original seismic data in this area generally have a signal-to-noise ratio below 2 dB, poor phase axis continuity, and severe noise interference, posing a significant challenge to tectonic interpretation. The specific implementation process of fault identification using the method of this invention is as follows: (1) Data preprocessing: After inputting the original seismic data volume, the data is first preprocessed using adaptive guided filtering. In this example, the initial regularization coefficient λ=0.1 is set to emphasize edge protection; the filtering radius is dynamically adjusted according to the local signal-to-noise ratio. In low signal-to-noise ratio areas (such as signal-to-noise ratio below 1.5dB), a 5×5 small-scale window is used, while in relatively high signal-to-noise ratio areas, a 7×7 or 9×9 window is used. This effectively suppresses random noise while maximizing the protection of fault boundaries and fragile structural information. The estimated signal-to-noise ratio of the data is improved by about 35% after preprocessing.

[0043] (2) Multi-scale edge enhancement: Non-subsampled shear wave transform (NSST) is applied to the preprocessed data to decompose it into high- and low-frequency sub-bands with 3 scales and 8 directions. The high-frequency sub-band reflecting the fault boundary information is the focus of nonlinear enhancement based on gradient amplitude. The enhancement coefficient is set to 1.5, which significantly enhances the weak edge and discontinuity information caused by the fault, while suppressing high-frequency noise, making the fault outline clearer and more continuous.

[0044] (3) Feature extraction and fusion: The third-generation coherence volume (C3) and phase symmetry attribute of the data are calculated separately. In this example, based on prior knowledge and comparison of experimental results, the weight of the coherence volume is set to 0.6 and the weight of the phase symmetry attribute is set to 0.4. The two attributes are weighted and fused to generate a fused feature volume. This fusion strategy effectively combines the high sensitivity of the coherence volume to faults and the robustness of the phase symmetry to noise, making the response in the fault area more prominent and suppressing non-fault geological anomalies.

[0045] (4) Fault extraction and optimization: An improved dynamic threshold segmentation algorithm (introducing dip constraints in the Otsu algorithm) was used to initially identify the fused feature volume, obtaining a binarized fault probability volume. Based on this, a 3×3 directional structural element was selected for morphological post-processing. The template direction is consistent with the local structural dip. An erosion-dilation operation was performed, which effectively eliminated isolated noise points with small areas, connected fault segments interrupted by noise, and significantly improved the continuity of fault lines and geological rationality.

[0046] Example 5 For the Ordovician carbonate reservoirs in a certain block of the Tarim Basin, the fault system in this area is complex, with numerous microfractures and steep faults formed by multiple tectonic movements. Affected by desert surface conditions and interference from deep multiples, the seismic data has an extremely low signal-to-noise ratio (approximately 1.5 dB), weak effective signals, and traditional fault identification methods suffer from poor response and numerous artifacts. The specific steps of implementing the method of this invention are as follows: (1) Data preprocessing: After inputting the raw seismic data, adaptive guided filtering was first applied. The initial filtering window was set to 5×5, the regularization coefficient λ=0.15, and the smoothing intensity was dynamically adjusted according to the local signal-to-noise ratio—the window was automatically reduced to 3×3 and the smoothing intensity was reduced in areas with severe noise to protect micro-fracture traces, while the window was expanded to 7×7 in areas with relatively high signal-to-noise ratio to enhance the suppression effect. This step effectively eliminated random noise while significantly improving the clarity of fault boundaries.

[0047] (2) Multi-scale edge enhancement: The data was decomposed into subbands with 4 scales and 16 directions using non-subsampled shear wave transform (NSST). Directional gradient enhancement was applied to the third high-frequency subband, which reflects the characteristics of small and medium-scale faults, with an enhancement coefficient set to 1.8. This process significantly enhanced the weak edge response caused by low-amplitude faults and high-dipping faults, while effectively suppressing coherent interference caused by multiple waves.

[0048] (3) Feature fusion: The improved third-generation coherent volume (C3) and the maximum curvature attribute are extracted and fused based on the local entropy weighted model. According to the test, the best results are achieved when the weight of C3 is set to 0.55 and the weight of curvature is set to 0.45. The entropy weight model makes the fracture more continuous in spatial distribution, especially improving the recognizability of the microfracture system in the background of strong noise.

[0049] (4) Fault extraction and optimization: A dynamic threshold algorithm with dip constraints is used for initial identification, making full use of local structural dip information to avoid misidentifying steeply dipped strata as faults. On this basis, directional morphological operations are performed using 3×3 rhombus structural elements to connect fault segments, repair interruptions, and eliminate discrete noise points, making the final fault system more geologically consistent in space.

[0050] Example 6: Addressing the challenges of deep reservoir exploration in a shale gas block in the Sichuan Basin, where the target layer is buried at depths exceeding 3500 meters, seismic data has a dominant frequency below 25Hz and low resolution, and numerous microfaults with displacements less than 10 meters have developed due to complex tectonic stresses. Accurate identification of these microfaults is crucial for horizontal well trajectory design and fracturing scheme optimization, but traditional methods are ineffective in their detection. The specific implementation process of the method of this invention is as follows: (1) Preprocessing optimization: After inputting low-frequency, high-noise seismic data, a two-parameter adaptive guided filter is used for preprocessing. This filter sets two regularization parameters: λ1=0.1 for strong smoothing in high-noise areas, and λ2=0.3 for enhancing boundary preservation in edge areas. The algorithm automatically switches parameters according to the local gradient magnitude, which significantly enhances the high-frequency edge components related to micro-faults while suppressing low-frequency background noise, providing a high-quality data foundation for subsequent micro-fracture identification.

[0051] (2) Multi-scale analysis: The data was decomposed into 5 scales and 16 directions using non-subsampled shear wave transform (NSST). For the first and second high-frequency subbands containing weak signals from micro-faults, a nonlinear compression enhancement algorithm was adopted. A threshold T = 0.4 × the maximum gradient value of the subband was set. Weak signals below this threshold were selectively enhanced, while strong signals above this threshold were kept unchanged. This method effectively highlighted the weak edge response of the micro-faults, while avoiding the artifacts caused by excessive enhancement of strong reflection phase axes.

[0052] (3) Multi-attribute decision fusion: Three attributes sensitive to micro-faults were extracted: phase symmetry (sensitive to edge symmetry, weight 0.5), variance volume (reflecting local heterogeneity, weight 0.3), and ant body attribute (enhancing linear discontinuity, weight 0.2). A random forest classifier was used for attribute fusion. This classifier uses 200 decision trees and learns the contribution of different attributes in fault and non-fault regions through training, effectively reducing conflicts and redundancy between attributes, generating a probabilistic fault distribution, and significantly improving the confidence of micro-fault identification.

[0053] (4) Fault optimization and extraction: The fault probability volume is refined by multi-scale edge focusing algorithm. First, the initial fault segments are extracted by dynamic threshold segmentation with directional constraints. Then, morphological connection is performed by 3×3 adaptive template based on geological dip. Finally, the sparse constraint algorithm is used to remove isolated outliers to complete the spatial optimization of micro-faults.

[0054] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A method for fault identification in low-quality seismic data, characterized in that, Includes the following steps: Step S1: Filter the original seismic data using the original seismic data itself as the guide field, and dynamically adjust the filtering parameters according to the local signal-to-noise ratio of the data during the filtering process to obtain preprocessed seismic data. Step S2: Perform non-subsampled shear wave transformation on the preprocessed seismic data, decomposing it into sub-bands of multiple scales and directions. Perform nonlinear enhancement processing based on gradient amplitude on the high-frequency sub-bands obtained after decomposition to obtain enhanced seismic data. Step S3: Calculate coherence volume features and phase symmetry features in parallel on the enhanced seismic data, and adaptively fuse the coherence volume features and phase symmetry features using a local entropy weighted model to obtain a fused feature body. Step S4: Extract the initial fault body from the fused feature body using the Otsu dynamic thresholding algorithm with directional constraints, and perform directional morphological operations on the initial fault body that are adapted to the local tectonic dip to obtain the final three-dimensional fault model. The filtering process of the original seismic data using the original seismic data itself as the guide field includes: using the amplitude distribution of the original seismic data itself as the guide field, calculating the amplitude similarity between the center point and the neighboring points within a local window to obtain the filtering weight, performing a weighted average calculation on the seismic data within the window based on the filtering weight, and obtaining the basic filtered seismic data volume by traversing the entire data volume. The filtering process dynamically adjusts the filtering parameters based on the local signal-to-noise ratio of the data to obtain preprocessed seismic data. This includes: calculating the signal-to-noise ratio estimate in real time within the local window; adaptively selecting a combination of parameters for the filtering window size and regularization coefficient based on the different threshold ranges of the signal-to-noise ratio estimate; applying the selected parameter combination to the weight calculation and weighted averaging process in the filtering process; optimizing the parameters of the basic filtered seismic data volume; and outputting the preprocessed seismic data.

2. The method for fault identification of low-quality seismic data according to claim 1, characterized in that, Performing a non-downsampled shear wave transform on the preprocessed seismic data to decompose it into subbands of multiple scales and directions includes: performing multi-scale decomposition on the preprocessed seismic data to obtain frequency components at different scales; performing multi-directional decomposition on the frequency components at each scale to obtain subband components in different directions; and obtaining a subband set containing low-frequency subbands and multiple high-frequency subbands through the multi-scale decomposition and multi-directional decomposition process.

3. The method for fault identification of low-quality seismic data according to claim 2, characterized in that, The high-frequency sub-bands obtained after decomposition are subjected to nonlinear enhancement processing based on gradient amplitude to obtain enhanced seismic data. This includes: calculating the gradient amplitude of each data point in the high-frequency sub-band, processing the gradient amplitude using a nonlinear enhancement function, fusing the processed gradient amplitude with the corresponding high-frequency sub-band data, and reconstructing all fused sub-bands through inverse transformation to obtain enhanced seismic data.

4. The method for fault identification of low-quality seismic data according to claim 3, characterized in that, Parallel calculation of coherence volume features and phase symmetry features of the enhanced seismic data includes: taking the enhanced seismic data as input, and simultaneously executing two independent processes: coherence volume feature calculation and phase symmetry feature calculation; The coherence volume feature calculation obtains characteristic data representing stratigraphic discontinuities by analyzing the waveform similarity between adjacent seismic traces, while the phase symmetry feature calculation obtains characteristic data representing edge symmetry by analyzing the local phase distribution through a symmetric filter bank. Through these two parallel calculation processes, complete coherence volume feature data and phase symmetry feature data are output respectively.

5. The method for fault identification of low-quality seismic data according to claim 4, characterized in that, The coherence volume features and phase symmetry features are adaptively fused using a local entropy weighted model to obtain a fused feature volume, which includes: taking the coherence volume feature data volume and the phase symmetry feature data volume as input, and calculating the information entropy values ​​of the two feature data volumes respectively within a local spatial window; Based on the calculated information entropy value, the fusion weights of the coherent volume features and the phase symmetry features are dynamically determined according to the preset weight allocation rules. The determined fusion weights are used to perform a weighted summation calculation on the two feature data volumes to generate an initial fused feature volume. The initial fused feature volume is then subjected to selective filtering based on information entropy analysis, and the filtering operation is applied only to regions where the information entropy value is higher than a preset threshold. Finally, the fused feature volume is output.

6. The method for fault identification of low-quality seismic data according to claim 5, characterized in that, The initial tomographic body is extracted from the fused feature body using the Otsu dynamic thresholding algorithm with directional constraints. This includes: based on the local structural trend data of the fused feature body, introducing directional constraints during the Otsu thresholding process, calculating the optimal segmentation threshold along the structural trend direction, and using the optimal segmentation threshold to perform binarization segmentation on the fused feature body to generate the initial tomographic body.

7. The method for fault identification of low-quality seismic data according to claim 6, characterized in that, The process of performing directional morphological operations adapted to the local tectonic dip on the initial fault body to obtain the final three-dimensional fault model includes: determining the orientation of the structural elements for the morphological operations based on the local tectonic dip data; performing erosion and dilation operations on the initial fault body sequentially using the orientation-adaptive structural elements; connecting fault segments and eliminating noise through the directional morphological operations; and outputting the three-dimensional fault model.

8. A fault identification system for low-quality seismic data employing the method described in any one of claims 1-7, characterized in that, include: Data preprocessing module: The original seismic data itself is used as a guide field to filter the original seismic data. During the filtering process, the filtering parameters are dynamically adjusted according to the local signal-to-noise ratio of the data to obtain the preprocessed seismic data. Multi-scale edge enhancement: The preprocessed seismic data is subjected to non-subsampled shear wave transform to decompose it into sub-bands of multiple scales and directions. The high-frequency sub-bands obtained after decomposition are subjected to nonlinear enhancement processing based on gradient amplitude to obtain enhanced seismic data. Feature extraction and fusion: Coherence volume features and phase symmetry features are calculated in parallel for the enhanced seismic data, and the coherence volume features and phase symmetry features are adaptively fused using a local entropy weighted model to obtain a fused feature volume; Fault extraction and optimization: Initial fault bodies are extracted from the fused feature volume using the Otsu dynamic threshold algorithm with directional constraints, and directional morphological operations adapted to the local tectonic dip are performed on the initial fault bodies to obtain the final three-dimensional fault model.

Images