A texture segmentation and enhancement method and system for seismic profile images

By employing multi-level texture segmentation and enhancement methods, the problem of texture blurring in seismic profile images is solved, improving image clarity and the accuracy of geological interpretation. This method is applicable to the interpretation of stratigraphic profile information such as geological faults.

CN118711207BActive Publication Date: 2024-11-15XI'AN PETROLEUM UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411180011.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-15
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

Seismic profile images are affected by external noise during the generation process, resulting in blurred and interrupted textures, which affects the accuracy of subsequent geological interpretation.

Method used

A multi-level texture segmentation method is adopted, including preprocessing, multiple texture segmentation algorithms and machine learning models. Cluster analysis is performed by combining intrinsic mode functions and residual features, and an appropriate texture enhancement model is selected to enhance the segmented sub-images.

Benefits of technology

It improves the visual quality and texture recognition of seismic profile images, enhances the clarity of geological structures, improves the accuracy and reliability of geological interpretation, and shortens processing time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118711207B_ABST
    Figure CN118711207B_ABST
Patent Text Reader

Abstract

The present invention relates to a texture segmentation and enhancement method and system for a seismic profile image, and relates to the technical field of image processing, wherein the method comprises: preprocessing the seismic profile image to obtain the preprocessed seismic profile image; performing a first texture segmentation on the preprocessed seismic profile image through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results; performing a second texture segmentation on the preprocessed seismic profile image based on the plurality of first texture segmentation results through a texture segmentation model to obtain a second texture segmentation result; obtaining a plurality of segmentation sub-images based on the second texture segmentation results; for each segmentation sub-image, determining a texture enhancement model corresponding to the segmentation sub-image based on the texture features of the segmentation sub-image, and performing texture enhancement on the segmentation sub-image through the texture enhancement model corresponding to the segmentation sub-image, which has the advantage of improving the clarity and recognizability of texture information in the seismic profile image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a texture segmentation and enhancement method and system for seismic profile images. Background Art

[0002] Seismic profile images have obvious texture characteristics. Different textures represent different geological bodies, which are of great significance for the interpretation of stratigraphic profile information such as geological faults. However, in the process of generating seismic profile images, the collected seismic data is affected by external noise, and Gaussian random noise is generated during the seismic data processing, which makes the texture of the seismic profile image unclear, and even produces texture interruptions, which interferes with the subsequent interpretation of seismic images.

[0003] Therefore, there is a need to provide a texture segmentation and enhancement method and system for seismic profile images, so as to improve the clarity and recognizability of texture information in seismic profile images. Summary of the invention

[0004] The present invention provides a texture segmentation and enhancement method for a seismic profile image, comprising: preprocessing the seismic profile image to obtain a preprocessed seismic profile image; performing a first texture segmentation on the preprocessed seismic profile image using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results; performing a second texture segmentation on the preprocessed seismic profile image based on the plurality of first texture segmentation results using a texture segmentation model to obtain a second texture segmentation result; obtaining a plurality of segmented sub-images based on the second texture segmentation results; for each of the segmented sub-images, determining a texture enhancement model corresponding to the segmented sub-image based on the texture features of the segmented sub-image, and performing texture enhancement on the segmented sub-image using the texture enhancement model corresponding to the segmented sub-image.

[0005] Furthermore, the preprocessed seismic profile image is subjected to a first texture segmentation through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: the preprocessed seismic profile image is evenly divided into a plurality of first sub-images; for each of the first sub-images, the grayscale value variance of the first sub-image is calculated based on the grayscale value of each pixel included in the first sub-image; when the grayscale value variance of the first sub-image is greater than a grayscale value variance threshold, the secondary segmentation size corresponding to the first sub-image is calculated based on the grayscale value variance of the first sub-image; based on the secondary segmentation size corresponding to the first sub-image, the first sub-image is segmented into a plurality of second sub-images; when the grayscale value variance of the first sub-image is less than or equal to the grayscale value variance threshold, the first sub-image is used as a second sub-image; for each of the second sub-images, the grayscale value variance and the grayscale value mean of the second sub-image are calculated based on the grayscale value of each pixel included in the second sub-image; and the plurality of second sub-images are clustered based on the grayscale value variance and the grayscale value mean of each second sub-image through a clustering algorithm to obtain a first texture segmentation result.

[0006] Furthermore, the preprocessed seismic profile image is subjected to a first texture segmentation through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: for each pixel included in the preprocessed seismic profile image, the adjacent pixels of the pixel are determined, and based on the grayscale values ​​of the adjacent pixels, the adjacent pixel grayscale value variance and the adjacent pixel grayscale value mean of the pixel are calculated; the pixels included in the preprocessed seismic profile image are clustered based on the adjacent pixel grayscale value variance and the adjacent pixel grayscale value mean of each pixel through a clustering algorithm to obtain a first texture segmentation result.

[0007] Furthermore, the preprocessed seismic profile image is subjected to a first texture segmentation through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: for each of the second sub-images, based on the grayscale value of each pixel included in the second sub-image, a grayscale value matrix corresponding to the second sub-image is constructed; based on the grayscale value matrix corresponding to the second sub-image, a one-dimensional row-direction grayscale value vector and a one-dimensional column-direction grayscale value vector are generated; empirical mode decomposition is performed on the one-dimensional row-direction grayscale value vector to generate a row-direction intrinsic mode function and a row-direction residual; empirical mode decomposition is performed on the one-dimensional column-direction grayscale value vector to generate a column-direction intrinsic mode function and a column-direction residual; and a plurality of second sub-images are clustered through a clustering algorithm based on the row-direction intrinsic mode function, the row-direction residual, the column-direction intrinsic mode function and the column-direction residual of each second sub-image to obtain a first texture segmentation result.

[0008] Furthermore, the texture segmentation model includes a weight calculation unit and an image segmentation unit, wherein the weight calculation unit is used to calculate the weight of each of the first texture segmentation results based on the image features of the preprocessed seismic profile image, and the image segmentation unit is used to perform a second texture segmentation on the preprocessed seismic profile image based on the multiple first texture segmentation results and the weight of each first texture segmentation result to obtain a second texture segmentation result.

[0009] Furthermore, the weight calculation unit includes an input layer, multiple feature extraction layers, a multi-head self-attention mechanism layer, a residual connection, a normalization layer, a fully connected layer and an output layer.

[0010] Furthermore, the method also includes: obtaining multiple sample sub-images; determining the texture features of each of the sample sub-images; clustering the multiple sample sub-images based on the texture features of each of the sample sub-images to determine multiple sub-image clusters; for each sub-image cluster, establishing a texture enhancement model corresponding to the sub-image cluster, and training the texture enhancement model corresponding to the sub-image cluster based on the multiple sample sub-images included in the sub-image cluster.

[0011] Furthermore, based on the texture features of the segmented sub-image, a texture enhancement model corresponding to the segmented sub-image is determined, including: based on the texture features of the segmented sub-image, a target sub-image cluster is determined; and the texture enhancement model corresponding to the target sub-image cluster is used as the texture enhancement model corresponding to the segmented sub-image.

[0012] Furthermore, based on the texture features of the segmented sub-images, a target sub-image cluster is determined, including: for each sub-image cluster, based on a plurality of sample sub-images included in the sub-image cluster, a cluster texture feature is determined; similarity between the texture features of the segmented sub-image and each of the cluster texture features is calculated; based on the similarity between the texture features of the segmented sub-image and each of the cluster texture features, a target sub-image cluster is determined from the plurality of sub-image clusters.

[0013] The present invention provides a texture segmentation and enhancement system for seismic profile images, which is used to execute a texture segmentation and enhancement method for seismic profile images, and includes: a preprocessing module, which is used to preprocess the seismic profile image to obtain a preprocessed seismic profile image; a texture segmentation module, which is used to perform a first texture segmentation on the preprocessed seismic profile image through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, and is also used to perform a second texture segmentation on the preprocessed seismic profile image based on the plurality of first texture segmentation results through a texture segmentation model to obtain a second texture segmentation result, and obtain a plurality of segmented sub-images based on the second texture segmentation results; a texture enhancement module, which is used to determine, for each segmented sub-image, a texture enhancement model corresponding to the segmented sub-image based on the texture features of the segmented sub-image, and perform texture enhancement on the segmented sub-image through the texture enhancement model corresponding to the segmented sub-image.

[0014] Compared with the prior art, the texture segmentation and enhancement method of seismic profile images provided in this specification has at least the following beneficial effects:

[0015] 1. Through the preprocessing steps, the visual quality of the seismic profile image can be improved, providing a clearer data basis for subsequent analysis. The image is segmented initially through a variety of texture segmentation algorithms, and the machine learning model is used for secondary segmentation, which can more accurately identify and separate different texture areas in the image. This multi-level segmentation method helps to improve the accuracy of subsequent geological interpretation. Selecting a suitable texture enhancement model for each segmented sub-image can further enhance the texture details in the image, making the geological structures such as seismic reflection layers and faults more clear and discernible. This targeted enhancement method can maximize the retention and highlighting of useful geological information. For multiple segmented sub-images, different texture enhancement models can be applied in parallel for processing, thereby shortening the overall processing time. After multi-step processing, the geological structure and texture features of the seismic profile image are clearer, which helps to improve the accuracy and reliability of geological interpretation. This is crucial for subsequent underground resource assessment, disaster prediction and other work.

[0016] 2. Using intrinsic mode functions and residuals as features for cluster analysis can significantly increase the feature dimensions used for segmentation, thereby improving the accuracy and reliability of texture segmentation. Since empirical mode decomposition is adaptive, it can automatically select the level and scale of decomposition according to the actual texture features of the image, which makes the segmentation results more consistent with the actual situation of the image. By performing empirical mode decomposition on each second sub-image and extracting features, the amount of data that needs to be processed can be reduced during the clustering process, thereby improving the execution efficiency of the clustering algorithm. As feature vectors, intrinsic mode functions and residuals can more accurately describe the texture characteristics of each second sub-image, thereby helping the clustering algorithm to more accurately identify image areas with similar texture characteristics.

[0017] 3. The seismic profile image is evenly divided into multiple first sub-images, and secondary segmentation is performed based on the gray value variance, which can ensure that different degrees of refinement are performed on areas with different texture complexity during the processing. This adaptive segmentation method helps to better retain and highlight the detail information of the image. Gray value variance is an important indicator to measure the local texture complexity of an image. By calculating the gray value variance of each sub-image, the complexity of its texture can be judged, and then it can be determined whether further segmentation is needed. This method can avoid over-segmentation of simple texture areas while ensuring sufficient analysis of complex texture areas. Using a clustering algorithm to cluster the second sub-image, image areas with similar texture characteristics can be divided into the same category based on features such as gray value variance and mean. This method can reduce the influence of subjective factors and improve the accuracy and objectivity of texture segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0019] Figure 1 is a flow chart of a texture segmentation and enhancement method for a seismic profile image shown in an embodiment of the present application;

[0020] Figure 2 is a flow chart of obtaining the second texture segmentation result shown in an embodiment of the present application;

[0021] Figure 3 is a structural diagram of a weight calculation unit shown in an embodiment of the present application;

[0022] Figure 4 It is a module diagram of a texture segmentation and enhancement system for a seismic profile image shown in one embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly introduces the drawings required for describing the embodiments.

[0024] Figure 1 is a flow chart of a texture segmentation and enhancement method for a seismic profile image shown in an embodiment of the present application, such as Figure 1 As shown, a texture segmentation and enhancement method for a seismic profile image may include the following steps.

[0025] Step 110, preprocessing the seismic profile image to obtain a preprocessed seismic profile image.

[0026] Specifically, preprocessing may include noise reduction, removal of stray energy, adaptive filtering, data resampling, static correction, energy compensation, suppression of noise interference, etc.

[0027] The purpose of noise reduction is to remove the interference caused by factors such as instrument noise and environmental conditions to make the seismic data cleaner. Filtering technology (for example, adaptive filtering, median filtering, wavelet transform, etc.) can be used to reduce the noise of seismic profile images.

[0028] The purpose of removing stray energy is to weaken or remove seismic energy that is not related to underground targets and improve the quality of data. Filters or signal processing techniques, such as spectral analysis and time-frequency domain filtering, can be used to identify and remove these stray energies.

[0029] The purpose of adaptive filtering is to filter seismic data according to the characteristics of the underground medium to eliminate non-geological reflection energy caused by wave propagation and underground medium heterogeneity. Using the adaptive filtering algorithm, the parameters of the filter are dynamically adjusted according to the statistical characteristics of the seismic data to achieve the best filtering effect.

[0030] The purpose of data resampling is to increase the sampling rate of the data to obtain a higher frequency resolution. Through interpolation or resampling technology, the number of time sampling points of seismic data is increased to make the data have better performance in a higher frequency range.

[0031] The purpose of static correction is to solve the static correction problems caused by factors such as surface elevation changes, low-velocity zone thickness and velocity changes, and to improve the imaging accuracy of seismic profiles. Static correction includes elevation static correction, refraction static correction and tomographic static correction. Among them, refraction static correction is a method of combining terrain correction with refraction static correction by obtaining the velocity of the refraction layer through refraction waves. Tomographic static correction is to obtain a more accurate surface model through forward modeling and inversion iteration.

[0032] The purpose of energy compensation is to compensate for the energy loss caused by geometric diffusion and absorption attenuation during the propagation of seismic waves. Energy compensation includes geometric diffusion compensation and absorption attenuation compensation. Geometric diffusion compensation mainly eliminates the influence of propagation distance on the energy of seismic records; absorption attenuation compensation is to restore the energy attenuation of seismic waves during propagation through methods such as inverse Q filtering and time-frequency analysis.

[0033] The purpose of noise suppression is to improve the signal-to-noise ratio of seismic data and make the reflection signal of underground structure more prominent. According to the type and characteristics of noise, appropriate noise suppression technology is used to suppress noise interference on seismic profile images, such as surface wave suppression, random noise suppression, etc.

[0034] Step 120, performing a first texture segmentation on the preprocessed seismic profile image using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results.

[0035] In some embodiments, a texture segmentation algorithm is:

[0036] Evenly dividing the preprocessed seismic profile image into a plurality of first sub-images;

[0037] For each first sub-image, based on the gray value of each pixel included in the first sub-image, the gray value variance of the first sub-image is calculated; when the gray value variance of the first sub-image is greater than the gray value variance threshold, the secondary segmentation size corresponding to the first sub-image is calculated based on the gray value variance of the first sub-image; based on the secondary segmentation size corresponding to the first sub-image, the first sub-image is segmented into a plurality of second sub-images; when the gray value variance of the first sub-image is less than or equal to the gray value variance threshold, the first sub-image is used as the second sub-image;

[0038] For each second sub-image, based on the gray value of each pixel included in the second sub-image, calculate the gray value variance and the gray value mean of the second sub-image;

[0039] A clustering algorithm is used to cluster multiple second sub-images based on the gray value variance and gray value mean of each second sub-image to obtain a first texture segmentation result.

[0040] Specifically, the grayscale value variance of the first sub-image can be calculated based on the following formula:

[0041] ;

[0042] ;

[0043] in, is the gray value variance of the kth first sub-image, is the gray value of the pixel in the i-th row and j-th column of the k-th first sub-image, is the total number of rows of pixels included in the kth first sub-image, is the total number of columns of pixels included in the kth first sub-image, is the mean gray value of the kth first sub-image.

[0044] The secondary segmentation size corresponding to the first sub-image can be calculated based on the following formula:

[0045] ;

[0046] in, is the secondary segmentation size corresponding to the kth first sub-graph, is the gray value variance of the preset first sub-image, The default secondary split size.

[0047] Multiple second sub-images can be clustered based on the following process to obtain a first texture segmentation result:

[0048] S11, based on the gray value mean of each second sub-image, calculating the gray value mean variance, determining the number of clusters according to the gray value mean variance, and randomly selecting multiple second sub-images as multiple cluster centers based on the number of clusters;

[0049] S12, for each second sub-graph, calculate the first variance difference absolute value between the gray value variance of the second sub-graph and the gray value variance of each cluster center, calculate the first mean difference absolute value between the gray value mean of the second sub-graph and the gray value mean of each cluster center, perform weighted summation of the first variance difference absolute value and the first mean difference absolute value, calculate the difference distance between the second sub-graph and each cluster center, and cluster the second sub-graph to the cluster center with the smallest difference distance;

[0050] S13, determine whether the number of iterations is equal to the number threshold, if so, complete clustering, if not, execute S14;

[0051] S14, for each cluster, average the gray value means of each second sub-image included in the cluster to obtain the gray value mean of the first cluster, average the gray value variances of each second sub-image included in the cluster to obtain the gray value variance mean of the first cluster, calculate the second mean difference absolute value of the gray value mean of the second sub-image and the gray value mean of the first cluster, calculate the second variance difference absolute value of the gray value variance of the second sub-image and the gray value variance mean of the first cluster, perform weighted summation of the second variance difference absolute value and the second mean difference absolute value, calculate the centrality of the second sub-image, and take the second sub-image with the largest centrality as the new cluster center of the cluster;

[0052] S15, judging whether the cluster center of each cluster has not changed, if so, completing clustering, if not, executing S12.

[0053] For example, the gray value mean variance can be calculated according to the following formula:

[0054] ;

[0055] in, is the mean variance of grayscale value, is the mean gray value of the e-th second sub-image, is the total number of the second sub-graphs.

[0056] The number of clusters can be calculated based on the following formula:

[0057] ;

[0058] in, is the number of clusters, is the preset gray value mean variance, is the preset number of clusters.

[0059] The centrality of the second subgraph can be calculated based on the following formula:

[0060] ;

[0061] in, is the centrality of the e-th second subgraph, is the preset parameter, greater than 0, and are weights, and are greater than 0, and , is the mean gray value of the first cluster of the cluster to which the e-th second sub-image belongs, is the gray value variance of the e-th second sub-image, is the mean variance of the grayscale values ​​of the first cluster of the cluster to which the e-th second sub-image belongs.

[0062] In some embodiments, a texture segmentation algorithm is:

[0063] For each pixel included in the preprocessed seismic profile image, determine the adjacent pixels of the pixel, and based on the gray values ​​of the adjacent pixels, calculate the adjacent pixel gray value variance and the adjacent pixel gray value mean of the pixel;

[0064] The pixels included in the preprocessed seismic profile image are clustered based on the grayscale value variance of each pixel and the grayscale value mean of each pixel, and a first texture segmentation result is obtained.

[0065] Specifically, the pixels included in the preprocessed seismic profile image can be clustered based on the following process to obtain a first texture segmentation result:

[0066] S21, based on the mean gray value of the adjacent pixels of each pixel, calculating the mean variance of the gray value of the adjacent pixels, determining the number of clusters according to the mean variance of the gray value of the adjacent pixels, and randomly selecting multiple pixels as multiple cluster centers based on the number of clusters, wherein the method of determining the number of clusters according to the mean variance of the gray value of the adjacent pixels is similar to the method of determining the number of clusters according to the mean variance of the gray value, and will not be repeated here;

[0067] S22, for each pixel, calculate the third difference absolute value of the grayscale value variance of the pixel's adjacent pixels and the grayscale value variance of the adjacent pixels of each cluster center, calculate the third mean difference absolute value of the grayscale value mean of the pixel's adjacent pixels and the grayscale value mean of the adjacent pixels of each cluster center, perform weighted summation on the third difference absolute value and the third mean difference absolute value, calculate the difference distance between the pixel and each cluster center, and locate the pixel to the cluster center with the smallest difference distance;

[0068] S23, determine whether the number of iterations is equal to the number threshold, if so, complete clustering, if not, execute S24;

[0069] S24, for each cluster, average the grayscale value means of the adjacent pixels of each pixel included in the cluster to obtain the grayscale value mean of the second cluster, average the grayscale value variances of the adjacent pixels of each pixel included in the cluster to obtain the grayscale value variance mean of the second cluster, calculate the fourth mean difference absolute value of the grayscale value mean of the adjacent pixels of the pixel and the grayscale value mean of the second cluster, calculate the fourth variance difference absolute value of the grayscale value variance of the adjacent pixels of the pixel and the grayscale value variance mean of the second cluster, perform weighted summation of the fourth mean difference absolute value and the fourth variance difference absolute value, calculate the center of the pixel, and take the pixel with the largest center as the new cluster center of the cluster;

[0070] S25, judging whether the cluster center of each cluster has not changed, if so, completing the clustering, if not, executing S22.

[0071] In some embodiments, a texture segmentation algorithm is:

[0072] For each second sub-image, based on the gray value of each pixel included in the second sub-image, a gray value matrix corresponding to the second sub-image is constructed, based on the gray value matrix corresponding to the second sub-image, a one-dimensional row-direction gray value vector and a one-dimensional column-direction gray value vector are generated, empirical mode decomposition is performed on the one-dimensional row-direction gray value vector to generate a row-direction intrinsic mode function and a row-direction residual, and empirical mode decomposition is performed on the one-dimensional column-direction gray value vector to generate a column-direction intrinsic mode function and a column-direction residual;

[0073] A clustering algorithm is used to cluster multiple second sub-images based on the row-direction intrinsic mode function, row-direction residual, column-direction intrinsic mode function and column-direction residual of each second sub-image to obtain a first texture segmentation result.

[0074] Specifically, the row vector of the gray value matrix may be the gray value of a row of pixels of the second sub-image, and the column vector of the gray value matrix may be the gray value of a column of pixels of the second sub-image.

[0075] For any two second sub-images, the grayscale similarity of the two second sub-images can be calculated based on the row-direction intrinsic mode functions, row-direction residuals, column-direction intrinsic mode functions and column-direction residuals of the two second sub-images through a similarity determination model, wherein the similarity determination model can be a convolutional neural network (CNN) model.

[0076] Multiple second sub-images can be clustered based on the following process to obtain a first texture segmentation result:

[0077] S31, based on the grayscale similarity of any two second sub-graphs, calculating the grayscale similarity variance, determining the number of clusters according to the grayscale similarity variance, and randomly selecting multiple second sub-graphs as multiple cluster centers based on the number of clusters, wherein the method of determining the number of clusters according to the grayscale similarity variance is similar to the method of determining the number of clusters according to the grayscale value mean variance, which will not be repeated here;

[0078] S32, for each second sub-image, clustering the second sub-image to a cluster center with the largest grayscale similarity;

[0079] S33, determine whether the number of iterations is equal to the number threshold, if so, complete clustering, if not, execute S34;

[0080] S34, for each cluster, for each second subgraph included in the cluster, based on the grayscale similarity between the second subgraph and other second subgraphs of the cluster, determine the grayscale similarity mean corresponding to the second subgraph, and use the second subgraph with the largest grayscale similarity mean as the new cluster center of the cluster;

[0081] S35, judging whether the cluster center of each cluster has not changed, if so, completing clustering, if not, executing S32.

[0082] Step 130, performing a second texture segmentation on the preprocessed seismic profile image based on the multiple first texture segmentation results through the texture segmentation model to obtain a second texture segmentation result.

[0083] Figure 2 is a flowchart of obtaining the second texture segmentation result shown in an embodiment of the present application, such as Figure 2As shown, in some embodiments, the texture segmentation model includes a weight calculation unit and an image segmentation unit, wherein the weight calculation unit is used to calculate the weight of each first texture segmentation result based on the image features of the preprocessed seismic profile image, and the image segmentation unit is used to perform a second texture segmentation on the preprocessed seismic profile image based on multiple first texture segmentation results and the weight of each first texture segmentation result to obtain a second texture segmentation result.

[0084] Figure 3 is a structural diagram of a weight calculation unit shown in an embodiment of the present application, such as Figure 3 As shown, in some embodiments, the weight calculation unit includes an input layer, multiple feature extraction layers, a multi-head self-attention mechanism layer, a residual connection, a normalization layer, a fully connected layer and an output layer.

[0085] Specifically, the input layer is the starting point of the weight calculation unit and is responsible for receiving the preprocessed seismic profile image data. The feature extraction layer consists of multiple convolutional layers and pooling layers. Their function is to extract useful feature representations from the input image, which may include edges, textures, shapes, etc. As the number of layers increases, the extracted features gradually change from low-level (such as edges) to high-level (such as complex textures or object parts).

[0086] The multi-head self-attention mechanism layer calculates the attention weights between different positions within the image, allowing the model to focus on the important parts of the image. In the weight calculation unit, this layer can help the model better understand and utilize the feature maps output by the feature extraction layer, thereby calculating more accurate weights.

[0087] Residual connections allow the original input information to skip certain layers and be added to the output of subsequent layers. This design helps alleviate the gradient vanishing or gradient exploding problems that may occur during the training of deep neural networks, making it easier for the network to learn useful feature representations.

[0088] The normalization layer is used to normalize the output of the previous layer to make the distribution of data more stable, which helps to accelerate the network training process and improve the generalization ability of the model.

[0089] The fully connected layer is located at the end of the network and is responsible for mapping the features extracted by the previous layer to the sample's label space. In the weight calculation unit, the fully connected layer receives the feature vector from the previous layer and outputs the weight of each first texture segmentation result. These weights will be used in the subsequent image segmentation process to guide the fusion of different segmentation results.

[0090] The output layer is the last layer of the weight calculation unit, which is responsible for outputting the weight of each first texture segmentation result. These weights are usually expressed in the form of a vector, and each element corresponds to the weight value of a segmentation result. These weights will be used in the subsequent image segmentation process to generate the final segmentation result.

[0091] Step 140: obtaining a plurality of segmentation sub-images based on the second texture segmentation result.

[0092] Specifically, there are certain differences in the texture features of any two segmented sub-images.

[0093] Step 150: for each segmented sub-image, determine a texture enhancement model corresponding to the segmented sub-image based on the texture features of the segmented sub-image, and perform texture enhancement on the segmented sub-image using the texture enhancement model corresponding to the segmented sub-image.

[0094] In some embodiments the method further comprises:

[0095] Get multiple sample subgraphs;

[0096] Determine the texture characteristics of each sample sub-image;

[0097] Based on the texture features of each sample sub-image, multiple sample sub-images are clustered to determine multiple sub-image clusters;

[0098] For each sub-image cluster, a texture enhancement model corresponding to the sub-image cluster is established, and based on a plurality of sample sub-images included in the sub-image cluster, the texture enhancement model corresponding to the sub-image cluster is trained.

[0099] Specifically, multiple sample subgraphs can be clustered based on the following process to determine multiple subgraph clusters:

[0100] S41, based on the texture features of any two sample sub-images, calculating the texture feature similarity of any two sample sub-images, based on the texture feature similarity of any two sample sub-images, calculating the texture feature similarity variance, determining the number of clusters according to the texture feature similarity variance, and based on the number of clusters, randomly selecting multiple sample sub-images as multiple clustering centers, wherein the method of determining the number of clusters according to the texture feature similarity variance is similar to the method of determining the number of clusters according to the gray value mean variance, which will not be repeated here;

[0101] S42, for each sample sub-image, clustering the sample sub-image to a cluster center with the largest texture feature similarity;

[0102] S43, determine whether the number of iterations is equal to the number threshold, if so, complete clustering, if not, execute S44;

[0103] S44, for each cluster, for each sample sub-graph included in the cluster, based on the texture feature similarity between the sample sub-graph and other sample sub-graphs of the cluster, determine the mean texture feature similarity corresponding to the sample sub-graph, and use the sample sub-graph with the largest mean texture feature similarity as the new clustering center of the cluster;

[0104] S45, judging whether the cluster center of each cluster has not changed, if so, completing the clustering, if not, executing S42.

[0105] In some embodiments, determining a texture enhancement model corresponding to the segmented sub-image based on the texture features of the segmented sub-image includes:

[0106] Determine the target sub-image cluster based on the texture features of the segmented sub-image;

[0107] The texture enhancement model corresponding to the target sub-image cluster is used as the texture enhancement model corresponding to the segmented sub-image, wherein the texture enhancement model can be a Generative Adversarial Networks (GAN) model.

[0108] In some embodiments, determining a target sub-image cluster based on texture features of the segmented sub-images includes:

[0109] For each sub-image cluster, based on a plurality of sample sub-images included in the sub-image cluster, determining a cluster texture feature;

[0110] Calculate the similarity between the texture features of the segmented sub-image and the texture features of each cluster;

[0111] Based on the similarity between the texture features of the segmented sub-images and the texture features of each cluster, a target sub-image cluster is determined from multiple sub-image clusters. Specifically, the sub-image cluster with the greatest similarity is taken as the target sub-image cluster.

[0112] Figure 4 is a module diagram of a texture segmentation and enhancement system for a seismic profile image shown in an embodiment of the present application, such as Figure 4 As shown, a texture segmentation and enhancement system for seismic profile images may include a preprocessing module, a texture segmentation module and a texture enhancement module.

[0113] The preprocessing module can be used to preprocess the seismic profile image to obtain a preprocessed seismic profile image;

[0114] The texture segmentation module can be used to perform a first texture segmentation on the preprocessed seismic profile image through a variety of texture segmentation algorithms to obtain a plurality of first texture segmentation results, and is also used to perform a second texture segmentation on the preprocessed seismic profile image based on the plurality of first texture segmentation results through a texture segmentation model to obtain a second texture segmentation result, and obtain a plurality of segmentation sub-images based on the second texture segmentation results;

[0115] The texture enhancement module can be used to determine the texture enhancement model corresponding to each segmented sub-image based on the texture features of the segmented sub-image, and perform texture enhancement on the segmented sub-image using the texture enhancement model corresponding to the segmented sub-image.

[0116] A texture segmentation and enhancement system for seismic profile images can be used to execute a texture segmentation and enhancement method for seismic profile images. For more descriptions of a texture segmentation and enhancement system for seismic profile images, please refer to the relevant descriptions of a texture segmentation and enhancement method for seismic profile images, which will not be repeated here.

[0117] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A texture segmentation and enhancement method for seismic profile images, characterized in that: include: Preprocessing the seismic profile image to obtain a preprocessed seismic profile image; Performing a first texture segmentation on the preprocessed seismic profile image by using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results; Based on the multiple first texture segmentation results, the texture segmentation model is used to perform a second texture segmentation on the preprocessed seismic profile image to obtain a second texture segmentation result; Based on the second texture segmentation result, a plurality of segmentation sub-images are obtained; For each of the segmented sub-images, based on the texture features of the segmented sub-images, a texture enhancement model corresponding to the segmented sub-images is determined, and texture enhancement is performed on the segmented sub-images using the texture enhancement model corresponding to the segmented sub-images; The preprocessed seismic profile image is subjected to a first texture segmentation by using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: Evenly dividing the preprocessed seismic profile image into a plurality of first sub-images; For each of the first sub-images, based on the gray value of each pixel included in the first sub-image, a gray value variance of the first sub-image is calculated; when the gray value variance of the first sub-image is greater than a gray value variance threshold, a secondary segmentation size corresponding to the first sub-image is calculated based on the gray value variance of the first sub-image; based on the secondary segmentation size corresponding to the first sub-image, the first sub-image is segmented into a plurality of second sub-images; when the gray value variance of the first sub-image is less than or equal to the gray value variance threshold, the first sub-image is used as a second sub-image; For each of the second sub-images, based on the gray value of each pixel included in the second sub-image, calculating the gray value variance and the gray value mean of the second sub-image; Clustering the plurality of second sub-images based on the grayscale value variance and grayscale value mean of each second sub-image by a clustering algorithm to obtain the first texture segmentation result; Cluster multiple second sub-images based on the following process to obtain a first texture segmentation result: S11, based on the gray value mean of each second sub-image, calculating the gray value mean variance, determining the number of clusters according to the gray value mean variance, and randomly selecting multiple second sub-images as multiple cluster centers based on the number of clusters; S12, for each second sub-graph, calculate the first variance difference absolute value between the gray value variance of the second sub-graph and the gray value variance of each cluster center, calculate the first mean difference absolute value between the gray value mean of the second sub-graph and the gray value mean of each cluster center, perform weighted summation of the first variance difference absolute value and the first mean difference absolute value, calculate the difference distance between the second sub-graph and each cluster center, and cluster the second sub-graph to the cluster center with the smallest difference distance; S13, determine whether the number of iterations is equal to the number threshold, if so, complete clustering, if not, execute S14; S14, for each cluster, average the gray value means of each second sub-image included in the cluster to obtain the gray value mean of the first cluster, average the gray value variances of each second sub-image included in the cluster to obtain the gray value variance mean of the first cluster, calculate the second mean difference absolute value of the gray value mean of the second sub-image and the gray value mean of the first cluster, calculate the second variance difference absolute value of the gray value variance of the second sub-image and the gray value variance mean of the first cluster, perform weighted summation of the second variance difference absolute value and the second mean difference absolute value, calculate the centrality of the second sub-image, and take the second sub-image with the largest centrality as the new cluster center of the cluster; S15, judging whether the cluster center of each cluster has not changed, if so, completing clustering, if not, executing S12; The gray value mean variance is calculated according to the following formula: ; in, is the mean variance of grayscale value, is the mean gray value of the e-th second sub-image, is the total number of the second subgraph; The number of clusters is calculated based on the following formula: ; in, is the number of clusters, is the preset gray value mean variance, is the preset number of clusters; The centrality of the second subgraph is calculated based on the following formula: ; in, is the centrality of the e-th second subgraph, is the preset parameter, greater than 0, and are weights, and are greater than 0, and , is the mean gray value of the first cluster of the cluster to which the e-th second sub-image belongs, is the gray value variance of the e-th second sub-image, is the mean variance of the grayscale values ​​of the first cluster of the cluster to which the e-th second sub-image belongs.

2. The texture segmentation and enhancement method of a seismic profile image according to claim 1, characterized in that: The preprocessed seismic profile image is subjected to a first texture segmentation by using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: For each pixel included in the preprocessed seismic profile image, determine the adjacent pixels of the pixel, and calculate the adjacent pixel grayscale value variance and the adjacent pixel grayscale value mean of the pixel based on the grayscale values ​​of the adjacent pixels; The pixels included in the preprocessed seismic profile image are clustered based on the grayscale value variance of adjacent pixels and the grayscale value mean of adjacent pixels of each pixel by using a clustering algorithm to obtain the first texture segmentation result.

3. The texture segmentation and enhancement method of a seismic profile image according to claim 1, characterized in that: The preprocessed seismic profile image is subjected to a first texture segmentation by using a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, including: For each of the second sub-images, based on the grayscale value of each pixel included in the second sub-image, a grayscale value matrix corresponding to the second sub-image is constructed, based on the grayscale value matrix corresponding to the second sub-image, a one-dimensional row-direction grayscale value vector and a one-dimensional column-direction grayscale value vector are generated, empirical mode decomposition is performed on the one-dimensional row-direction grayscale value vector to generate a row-direction intrinsic mode function and a row-direction residual, and empirical mode decomposition is performed on the one-dimensional column-direction grayscale value vector to generate a column-direction intrinsic mode function and a column-direction residual; A clustering algorithm is used to cluster multiple second sub-images based on the row-direction intrinsic mode function, row-direction residual, column-direction intrinsic mode function and column-direction residual of each second sub-image to obtain the first texture segmentation result.

4. A texture segmentation and enhancement method for seismic profile images according to any one of claims 1 to 3, characterized in that: The texture segmentation model includes a weight calculation unit and an image segmentation unit, wherein the weight calculation unit is used to calculate the weight of each of the first texture segmentation results based on the image features of the preprocessed seismic profile image, and the image segmentation unit is used to perform a second texture segmentation on the preprocessed seismic profile image based on the multiple first texture segmentation results and the weight of each first texture segmentation result to obtain a second texture segmentation result.

5. The texture segmentation and enhancement method of a seismic profile image according to claim 4, characterized in that: The weight calculation unit includes an input layer, multiple feature extraction layers, a multi-head self-attention mechanism layer, a residual connection, a normalization layer, a fully connected layer and an output layer.

6. A texture segmentation and enhancement method for seismic profile images according to any one of claims 1 to 3, characterized in that: Also includes: Get multiple sample subgraphs; Determining the texture features of each of the sample sub-images; Clustering the plurality of sample sub-images based on the texture feature of each of the sample sub-images to determine a plurality of sub-image clusters; For each sub-image cluster, a texture enhancement model corresponding to the sub-image cluster is established, and based on a plurality of sample sub-images included in the sub-image cluster, the texture enhancement model corresponding to the sub-image cluster is trained.

7. The texture segmentation and enhancement method of a seismic profile image according to claim 6, characterized in that: Determining a texture enhancement model corresponding to the segmented sub-image based on the texture feature of the segmented sub-image includes: Determining a target sub-image cluster based on the texture features of the segmented sub-images; The texture enhancement model corresponding to the target sub-image cluster is used as the texture enhancement model corresponding to the segmented sub-image.

8. The texture segmentation and enhancement method of a seismic profile image according to claim 7, characterized in that: Determining a target sub-image cluster based on the texture features of the segmented sub-images includes: For each sub-image cluster, determining a cluster texture feature based on a plurality of sample sub-images included in the sub-image cluster; Calculating the similarity between the texture feature of the segmented sub-image and the texture feature of each cluster; Based on the similarity between the texture feature of the segmented sub-image and the texture feature of each cluster, a target sub-image cluster is determined from the multiple sub-image clusters.

9. A texture segmentation and enhancement system for seismic profile images, used to execute a texture segmentation and enhancement method for seismic profile images as claimed in any one of claims 1 to 8, characterized in that: include: A preprocessing module, used for preprocessing the seismic profile image to obtain a preprocessed seismic profile image; A texture segmentation module, used to perform a first texture segmentation on the preprocessed seismic profile image through a plurality of texture segmentation algorithms to obtain a plurality of first texture segmentation results, and also used to perform a second texture segmentation on the preprocessed seismic profile image based on the plurality of first texture segmentation results through a texture segmentation model to obtain a second texture segmentation result, and obtain a plurality of segmentation sub-images based on the second texture segmentation results; The texture enhancement module is used to determine, for each of the segmented sub-images, a texture enhancement model corresponding to the segmented sub-image based on the texture features of the segmented sub-image, and perform texture enhancement on the segmented sub-image using the texture enhancement model corresponding to the segmented sub-image.