CN109242968A - A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes - Google Patents

Abstract

The invention provides a three-dimensional modeling method of a river channel based on a multi-attribute super-voxel graph cut, which belongs to the field of three-dimensional model construction of geological bodies. Aiming at the insufficiency of a single seismic attribute to describe the river channel, the present invention proposes a multi-attribute fusion method based on an improved local linear embedding. The ISOLLE algorithm fuses the preferred attributes into a new attribute. Considering the existence of existing seismic attribute data A nonlinear fusion method is adopted. The fused attributes are better than the previous attributes, and the accuracy of the river edge and region characterization has been improved, laying a good foundation for the next segmentation and reconstruction. The present invention is based on the super-voxel and graph-cut channel segmentation method, and generates three-dimensional super-voxels through a simple linear iterative algorithm. The generated super-voxels fit the edge of the channel well and have good homogeneity. , and then combined with the graph cutting framework to obtain the final segmentation result, and extract the isosurface to obtain the three-dimensional model of the river surface.

Description

A 3D modeling method of river channel based on multi-attribute hypervoxel graph cut

technical field

The invention belongs to the field of three-dimensional model construction of geological bodies, and in particular relates to a three-dimensional modeling method of a river channel based on multi-attribute super-voxel map cuts.

Background technique

Building a 3D model of a geological body (channel) is one of the most important tasks in seismic data interpretation because most important hydrocarbon reservoirs exist around the geological body. The 3D model of the geological body can not only intuitively express the structural form of the geological body and its distribution in the 3D space, so that the interpreter can quantitatively analyze the geological body, but also provide important information for the numerical simulation of the reservoir and the deployment of the well location for reserves calculation. in accordance with.

The edge detection method is a common method for geological body detection. The edge detection-based method is used to identify salt dome boundaries, but due to the large noise in seismic data, this method cannot achieve ideal results. In the prior art, there is also a method combining a 3D edge detector and dip angle steering for the detection of geological bodies, which improves the signal-to-noise ratio and edge continuity, and makes the boundary shape of the geological body clearer. However, the edge-based method has a strong dependence on the change of the amplitude, and cannot obtain a good detection effect when the instantaneous amplitude change of the edge is not obvious. In order to overcome these problems, a geological body identification method based on texture attributes is introduced. A set of texture attributes can be used to predict the probability that each pixel in the 3D cube belongs to a salt dome, and then segmentation can be used to find the salt dome boundary. The prior art uses texture gradients to detect the boundaries of salt domes in view of the large pixel variation at the edge. But for texture-based methods, the size of the window has a great influence on the recognition results, and it is not universal for different work areas. There is a hybrid classification method combining edge and texture attributes, which has the advantages of edge method and texture method, and can better detect salt domes. The three attributes of edge detection, geometry and texture are further combined, and then semi-automatic tomographic detection is carried out through support vector machine (SVM), and the detection results match the original image.

However, unlike geological bodies such as salt domes and karst caves, the river itself does not have an obvious and definite boundary, and at the same time, the river often changes, intersects, and has poor continuity. Therefore, the above methods of identifying and segmenting geological bodies cannot be well applied to river segmentation. middle.

Many scholars divide river channels from the perspective of seismic attributes. Seismic attributes are data volumes extracted from the original 3D seismic data through certain algorithms. Seismic attributes reflect the characteristics of seismic data from different angles, and different geological structures show different characteristics in attribute values. Firstly, the seismic attribute is used to assist in the analysis of the river channel. Through the comparative analysis of multiple seismic attribute slice maps, the river channel boundary is comprehensively compared and compared. However, in this method, the signal-to-noise ratio of the data greatly affects the resolution and quality of seismic attributes, and some seismic attributes are not helpful for channel division. The RGB multi-attribute fusion technology is applied to the identification of the river channel, so that the edge of the river channel is clearer and the resolution is improved. Seismic attributes are decomposed into three frequency bands through coherence and spectral decomposition, which are represented by different colors, and then visualized through RGB mixing technology, which can quickly and effectively identify river channels. However, RGB fusion technology can only fuse three attributes, which has certain limitations. It is not applicable when more attributes need to be fused, and most studies only use visualization technology to assist subjective judgment, and do not qualitatively segment the river boundary.

In recent years, many scholars have combined geological problems with images to achieve accurate segmentation. Applying the improved PRC segmentation algorithm to 2D and 3D seismic images, the salt dome boundary can be accurately segmented semi-automatically. Salt dome boundaries are extracted in new images using an optimal path picking algorithm that tracks highly discontinuous salt dome boundaries by selecting the optimal path with the global maximum envelope value, which can be quickly updated to obtain salt boundaries.

Image-based methods are also applicable to the problem of channel geological bodies. In order to solve the problems of complex river channel structure and poor continuity, the local linear features of the river channel are enhanced by the method of steerable pyramid in image processing. There is a confidence- and curvature-guided level-set method for segmenting channels from 3D seismic data. Based on the fusion of various seismic attributes, the method of level set is used to segment and model various geological bodies such as karst caves and river channels. However, the method based on the level set is more dependent on the initial shape, and the computational efficiency is relatively low.

To sum up, most of the existing researches on river channel geological bodies remain on the identification and detection of the two-dimensional plane, and the research on the construction of the three-dimensional model of the river channel is somewhat insufficient.

SUMMARY OF THE INVENTION

Because the geological information carried by single-attribute data is incomplete, it is often inaccurate to interpret river channels only on single-attribute data. Therefore, it is necessary to reasonably select several attributes that can complement information, and adopt a nonlinear dimensionality reduction method, so as not to lose data information while reducing dimensionality. When segmenting the river channel, if it is only based on pixel-level segmentation, the segmented channel edge has poor continuity and cannot explain the channel well. In order to solve the problems in the prior art, the present invention provides a three-dimensional modeling method of a river channel based on multi-attribute super-voxel graph cuts.

A three-dimensional modeling method of river channel based on multi-attribute hypervoxel graph cut, comprising the following steps:

Step 1, using the nonlinear dimension reduction algorithm of ISOLLE to fuse the seismic attributes of the river channel to obtain the attribute data body of the river channel;

Step 2, according to the data body, by simple linear iterative clustering algorithm (SLIC), generate geological supervoxel;

Step 3, through the k-means clustering method, establish the Gaussian mixture model of the target area and the non-target area, construct the network graph and energy function, and segment the geological data based on the minimum cut criterion to obtain the binarized segmentation result, and pass The method of extracting the isosurface can obtain the three-dimensional model of the channel geological body.

Further, the step 1 includes the following processes:

Step 11, according to the geodetic distance, search for k samples that are close to the seismic data sample point;

According to the geodesic distance between two points, search the k data points with the closest geodetic distance to each seismic data sample point i in the 3D data.

d _G (x _i ,x _j )=min{L _G (x _i ,x _j )}

Among them, L _G is the length of a path between two points, d _E is the Euclidean distance, and d _G is the geodesic distance between the two points;

Step 12, constructing a local optimization reconstruction weight matrix;

An error function is introduced to measure the reconstruction error, which is

Among them, x _ij (j=1,2,...,k) is the k nearest neighbors of seismic data point i, w _ij is the weight between x _i and x _ij , and w _ij conforms to

For each seismic data point, the error is

Construct the local covariance matrix

Combine the local covariance matrix and Through the Lagrange multiplier method, the locally optimized reconstruction weight matrix is obtained

When the local optimization reconstruction weight matrix is a singular matrix, perform regularization processing

Q ⁱ =Q ⁱ +r·I

Among them, r is the regularization parameter, and I is the k×k unit matrix;

Step 13, map all seismic data points from high-dimensional to low-dimensional space, calculate the value of this seismic data point in low-dimensional space by the local reconstruction weight matrix of this seismic data point and its k neighbors;

Mapping conditions are met

Among them, ε(Y) is the value of the loss function, y _i is the output vector of x _i , and y _ij (j=1,2,...,k) is the k nearest neighbors found by the geodesic distance.

Among them, I is the identity matrix of m×m; the loss function is

Among them, M is an N×N alignment matrix, which is expressed as

M=(1-W) ^T ·(IW).

Further, the step 2 includes the following processes:

Step 21, initialize the cluster center;

Set the initial number of geological supervoxels as K, initialize the label of seismic data point i in the river channel area to -1, that is, label _i l=-1, and initialize the distance between point i and cluster center j to infinity, That is, dist _ij = +∞;

Step 22, in the neighborhood of the cluster center C _j , calculate the distance from each point to C _j ;

distance is

in, is the distance between the seismic data point i and the cluster center C _j on the fused attribute value, is the spatial distance between the seismic data point i and the cluster center C _j in the three-dimensional space of the river channel, and m is the parameter for adjusting the attribute distance and the weight of the spatial distance;

When d _ij < dist _i , update dist _i =d _ij , label _i =j;

Step 23, update the cluster center where N _j is the number of seismic data points belonging to the first type of geological supervoxel;

Step 24, calculate the residuals

Step 25, update the cluster center, set C _j =C' _j , if E is less than the preset threshold or exceeds the maximum number of iterations, enter the next process; if E does not meet the conditions, the process returns to the step 22;

Step 26, for the segmented geological supervoxel, by traversing the connected region of the seismic data points in three-dimensional space, an adjacency matrix A is established, and the attribute histogram is used as the geological supervoxel feature;

For the geological super-voxel i whose volume is not greater than the threshold, calculate the Babbitt distance between the geological super-voxel i and the adjacent super-voxel j

Among them, M is the dimension of the gray attribute histogram of the geological supervoxel, and the geological supervoxel i is merged into the adjacent geological supervoxel with the smallest distance T;

Step 27: After merging all discrete geological supervoxels smaller than the threshold, update the adjacency matrix and the attribute histogram of the geological supervoxels.

Further, the step 3 includes the following processes:

Step 31, by k-means clustering method, sets up the Gaussian mixture model of target area and non-target area;

Mark the supervoxel α _i = 1 that belongs to the channel, and the remaining supervoxels α _i = 0;

Perform k-means clustering on geological supervoxels by using four attributes of GLCM as features;

Initialize the parameters of each Gaussian component in the Gaussian mixture model according to the clustering results, the weight ω _k , the sample mean u _k and the covariance Σ _k , the density function of the Gaussian mixture model is

in,

Calculate the posterior probability

Compute maximum likelihood estimates of Gaussian component parameters

By iterating until the likelihood function converges, the channel Gaussian mixture model and the non-channel Gaussian mixture model are obtained;

Step 32, constructing a grid diagram and an energy function;

The gray-scale co-occurrence matrix is used to calculate the texture properties of each metageological voxel. GLCM is the joint probability distribution of two grayscale values in the image. The entropy, dissimilarity, and energy of GLCM are selected as the eigenvalues of the supervoxel. The energy function that combines the geological supervoxel region information and edge information is:

Among them, C is the division of the grid map; E _r is the regional item, which represents the weights of the t-link edges in the grid map summary, reflecting the regional information of the geological supervoxel; E _b is the boundary term, representing the network map. The weight of the n-link edge reflects the boundary attribute of the segmentation; by constructing an energy function, the segmentation of the river channel is converted into an energy function minimization problem;

Step 33: Segment the geological data based on the minimum cut criterion, update the seismic data points and the data point markers α _i in the Gaussian mixture model, and output the segmentation results after the iteration to obtain the binarized segmentation results of the channel geological body. The method of isosurface obtains the three-dimensional model of the channel geological body.

Beneficial effects of the present invention: The present invention provides a three-dimensional modeling method of a river channel based on multi-attribute super-voxel map cuts, obtains a new data volume by fusing several preferred seismic attributes, and then performs the super-voxel map cut method to obtain a new data volume. Binarized segmentation, and finally obtained the channel surface through isosurface extraction. The fusion method of ISOLLE in the present invention can effectively fuse various seismic attributes, and reasonable selection of seismic attributes is helpful for information complementation, more comprehensive description of the channel geological body, and the multi-attribute fusion method can maintain the nonlinearity in the channel seismic data. The obtained new attribute volume lays a good foundation for the next segmentation and reconstruction; the channel geological supervoxel has good edge retention, and the combination of the geological supervoxel and the map cut can accurately and quickly segment the channel. The isosurface extraction realizes the transformation of the three-dimensional model of the channel geological body from the three-dimensional seismic data, and more intuitively displays the spatial characteristics of the channel geological body; the method based on the multi-attribute fusion super-voxel map cut proposed by the present invention realizes the analysis of the channel geological body. The final three-dimensional model conforms to the geological laws and is basically consistent with the results obtained by the geologists, laying a foundation for the follow-up work.

Description of drawings

FIG. 1 is a flow chart of the present invention.

FIG. 2 is a flowchart of step 1 in FIG. 1 .

FIG. 3 is a flowchart of step 2 in FIG. 1 .

Figure 4 is the initialized cluster center map.

FIG. 5 is a flowchart of step 3 in FIG. 1 .

Detailed ways

The embodiments of the present invention will be further described below with reference to the accompanying drawings.

Please refer to Fig. 1, the present invention provides a kind of river three-dimensional modeling method based on multi-attribute super-voxel graph cut, realizes through the following steps:

Step 1, using the nonlinear dimension reduction algorithm of ISOLLE to fuse the seismic attributes of the river channel to obtain the attribute data volume of the river channel.

In this embodiment, in the process of seismic interpretation, different attribute data describe the geological structure from different angles, and by fusing different attributes, the geological structure of the reservoir can be interpreted more accurately. Because the relationship between seismic attributes and geological features is usually nonlinear, the PCA method based on linear transformation cannot fully reflect this nonlinear relationship, which reduces the accuracy of prediction and identification. The nonlinear LLE dimensionality reduction algorithm uses the Euclidean distance to calculate the distance between points, which cannot reflect the real structure between points, and is very sensitive to the choice of the number of neighbors.

In this embodiment, the ISOLLE algorithm is used, the concept of geodesic distance is introduced, and the Euclidean distance is replaced by the geodesic distance. While maintaining the advantages of the LLE algorithm in processing high-dimensional manifold data, the data in the corresponding low-dimensional space is improved. The tightness and linearity of local neighborhood data. The present invention applies the ISOLLE method to the multi-attribute fusion of the three-dimensional channel seismic data for the first time, and selects several attributes for fusion to form a new attribute body, which is extended to the three-dimensional space.

In this embodiment, the root mean square amplitude attribute, energy attribute, texture homogeneity, spectrum attribute, and instantaneous frequency attribute are selected to study the multi-attribute fusion problem of a certain seismic work area. After normalization and Fourier transform denoising, slice images with different attributes are obtained. The performance of the same attribute on different time slices is good or bad, and different attributes can make up for each other's shortcomings.

Referring to Figure 2, step 1 is achieved through the following process:

Step 11, according to the geodetic distance, search for k samples that are close to the seismic data sample point;

According to the geodesic distance between two points, search the k data points with the closest geodetic distance to each seismic data sample point i in the 3D data.

d _G (x _i ,x _j )=min{L _G (x _i ,x _j )}

Among them, L _G is the length of a path between two points, d _E is the Euclidean distance, and d _G is the geodesic distance between the two points;

Step 12, obtaining a local reconstruction matrix of the seismic data points. The local reconstruction matrix represents the local linear relationship between multiple attributes, and the fused seismic data points should also be able to maintain this linear relationship.

An error function is introduced to measure the reconstruction error, which is

Among them, x _ij (j=1,2,...,k) is the k nearest neighbors of seismic data point i, w _ij is the weight between x _i and x _ij , and w _ij conforms to

For each seismic data point, the error is

Construct the local covariance matrix

Combine the local covariance matrix and Through the Lagrange multiplier method, the locally optimized reconstruction weight matrix is obtained

When the local optimization reconstruction weight matrix is a singular matrix, perform regularization processing

Q ⁱ =Q ⁱ +r·I

Among them, r is the regularization parameter, and I is the k×k unit matrix;

Step 13, map all seismic data points from high-dimensional to low-dimensional space, calculate the value of this seismic data point in low-dimensional space by the local reconstruction weight matrix of this seismic data point and its k neighbors;

Mapping conditions are met

Among them, ε(Y) is the value of the loss function, y _i is the output vector of x _i , and y _ij (j=1,2,...,k) is the k nearest neighbors found by the geodesic distance, and also satisfies the following two conditions

where I is an m×m identity matrix. here Usually stored in an N×N sparse matrix W, when x _j is a neighbor of x _i , w _ij =w _j , if the two are not equal, then w _ij =0. These two conditions have their own meaning, the first guarantees Translation invariance to Y; the second guarantees that the same measure can be applied to the reconstruction error generated by different coordinates in low-dimensional space. Thus, the degenerate solution of Y=0 is prevented from appearing. The loss function is

Among them, M is an N×N alignment matrix, which is expressed as

M=(1-W) ^T ·(IW).

The matrix M has the characteristics of sparse and positive semi-definite. To make the loss function reach the minimum value, then should be taken as the eigenvector corresponding to the smallest non-zero eigenvalue. At the same time, the eigenvalues are arranged according to the order from small to large. The first eigenvalue is approximately equal to zero, so the first eigenvalue is discarded.

Step 2: Generate geological supervoxels according to the data volume through a simple linear iterative clustering algorithm (SLIC).

In this embodiment, the segmentation based on the pixel level does not make full use of the local relationship between pixels, which will bring errors in the process of image discretization. Superpixels use the similarity of features between pixels to group pixels, and use a small number of superpixels to replace a large number of pixels to express image features. Compared with pixel-level segmentation, it can maintain the edge features of the original image. A supervoxel is an extension of a superpixel in three-dimensional space. When processing 3D seismic data, the geological supervoxel method overcomes the problem of unsmoothness when synthesizing 3D models using superpixels in 2D slices, and at the same time reduces the number of blocks and reduces the amount of computation.

In this embodiment, in the seismic data volume, the closer the distance between two seismic data points, the higher the probability of belonging to the same geological body. According to this feature, we cluster similar seismic data points based on the SLIC algorithm to generate geological supervoxels. The generated geological supervoxel can not only keep the interior uniform and compact, but also maintain the edge characteristics of the geological body, which reduces the computational complexity of subsequent segmentation. In the SLIC algorithm, the main step is to compare the distance between each point and the cluster center C _j to classify, and then update the cluster center. Assuming that it is a three-dimensional data volume of Crossline*Inlinc*Time size after fusion, the set of all seismic data points V={v ₁ , v ₂ ,...,v _N }, where v _i ={ _gi , _xi , y _i , z _i }, _gi are the attribute values of seismic data point i after fusion, _xi , y _i , _zi are the coordinates of grid point i in space.

Referring to Figure 3, step 2 is achieved through the following process:

Step 21, initialize the cluster center;

Set the initial number of geological supervoxels to K, then the initial size of supervoxels is The river area is divided into K small areas. In the three-dimensional space of the river channel, the seed points are alternately selected as shown in Figure 4, so that the seed points can be distributed as evenly as possible in the space. Then the initial cluster center C={C ₁ ,C ₂ ,...,C _K }, the distance between the adjacent river supervoxel cluster centers is

The generated seed point may fall on the edge of the river. In order to avoid this situation, the gradient value of all grid points in the 3×3×3 neighborhood of the current seed point is calculated, and the position of the seed point is moved to this neighborhood. where the domain gradient is minimal. That is, move the cluster center to G(x,y,z)=minG(x,y,z).

In a 3×3×3 window centered on the seed point, the attribute gradient of each seismic data point in the channel is calculated. It is calculated as follows:

G(x,y,z)=||I(x+1,y,z)-I(x-1,y,z)|| ²

+||I(x,y+1,z)-I(x,y-1,z)|| ²

+||I(x,y,z+1)-I(x,y,z-1)|| ²

In the formula, I(x, y, z) represents the attribute value when the coordinates are (x, y, z) in the river work area. G(x, y, z) represents the gradient value calculated based on the attribute value when the coordinates are (x, y, z) in the river work area.

Initialize the label of seismic data point i in the river work area to -1, that is, label _i = -1;

The distance between point i and cluster center j is initially set to infinity, that is, dist _ij =+∞.

Step 22, in the neighborhood of the cluster center C _j , calculate the distance from each point to C _j ;

For all cluster centers C _j , in the neighborhood of 2S×2S×2S, the distance from each point to the cluster center C _j is

in, is the distance between the seismic data point i and the cluster center C _j on the fused attribute value, is the spatial distance between the seismic data point i and the cluster center C _j in the three-dimensional space of the river channel, and m is the parameter for adjusting the attribute distance and the weight of the spatial distance;

When d _ij < dist _i , update dist _i =d _ij , label _i =j;

Step 23, update the cluster center where N _j is the number of seismic data points belonging to the first type of geological supervoxel;

Step 24, calculate the residuals

Step 25, update the cluster center, set C _j =C' _j , if E is less than the preset threshold or exceeds the maximum number of iterations, enter the next process; if E does not meet the conditions, the process returns to the step 22;

Step 26, for the segmented geological supervoxel, by traversing the connected region of the seismic data points in three-dimensional space, an adjacency matrix A is established, and the attribute histogram is used as the geological supervoxel feature;

For the geological super-voxel i whose volume is not greater than the threshold, calculate the Babbitt distance between the geological super-voxel i and the adjacent super-voxel j

Among them, M is the dimension of the gray attribute histogram of the geological supervoxel, and the geological supervoxel i is merged into the adjacent geological supervoxel with the smallest distance T;

Step 27: After merging all discrete geological supervoxels smaller than the threshold, update the adjacency matrix and the attribute histogram of the geological supervoxels.

In this embodiment, similar data points are considered to be clustered, and the division is performed as the smallest unit of the river channel. The fused channel data volume is divided into three-dimensional supervoxels, and different geological supervoxels K are set, and different results can be obtained. When the number K of geological supervoxels is set to be small, the segmented geological supervoxels do not fit well at the channel edge. When K is large, the distinction between inside and outside the river channel is blurred, and the calculation amount of graph cuts cannot be effectively reduced. Therefore, we set the number K of supervoxels to 150, the segmented supervoxels can fit the edge of the channel well, and the homogeneity of the geological supervoxels is also better. Of course, K can also be set to other values.

Step 3, through the k-means clustering method, establish the Gaussian mixture model of the target area and the non-target area, construct the network graph and energy function, and segment the geological data based on the minimum cut criterion to obtain the binarized segmentation result, and pass The method of extracting the isosurface can obtain the three-dimensional model of the channel geological body.

In this embodiment, since the geological supervoxel is a small area, there are hidden texture features in the area. Therefore, on the basis of the traditional graph-cut algorithm, the characteristics of each super-voxel are represented by fusing the attribute values and texture features of super-voxels instead of gray-scale attributes. on the division. Compared with the traditional graph cut algorithm, the present invention replaces the attribute histogram with a Gaussian mixture model (GMM) to accurately express the probability model. A Gaussian mixture model is a linear superposition of multiple Gaussian models. The core of GMM is to obtain the mean, variance, and weight ratio of each Gaussian component. Firstly, the parameters of each Gaussian component in the GMM are randomly initialized, and its density function, posterior probability and maximum likelihood estimation of the Gaussian component parameters are calculated. Finally, through repeated iterations until the likelihood function converges, the Gaussian mixture model of the channel geological body is obtained, as shown in Figure 5, which is implemented through the following process:

Step 31, by k-means clustering method, sets up the Gaussian mixture model of target area and non-target area;

Mark the supervoxel α _i = 1 that belongs to the channel, and the remaining supervoxels α _i = 0;

Set K=5, and perform k-means clustering on the geological supervoxels by using the four attributes of GLCM as features;

According to the clustering results, randomly initialize the parameters of each Gaussian component in the Gaussian mixture model, the weight ω _k , the sample mean u _k and the covariance Σ _k , the density function of the Gaussian mixture model is

in,

Calculate the posterior probability

Compute maximum likelihood estimates of Gaussian component parameters

By iterating until the likelihood function converges, the channel Gaussian mixture model and the non-channel Gaussian mixture model are obtained;

Step 32, constructing a grid diagram and an energy function;

The gray-scale co-occurrence matrix is used to calculate the texture properties of each metageological voxel. GLCM is the joint probability distribution of two gray values in the image. The entropy, dissimilarity, and energy of GLCM are selected as the eigenvalues of the supervoxel. The energy function that combines the geological supervoxel region information and edge information is:

Among them, C is the division of the grid map; E _r is the regional item, which represents the weights of the t-link edges in the grid map summary, reflecting the regional information of the geological supervoxel; E _b is the boundary term, representing the network map. The weight of the n-link edge reflects the boundary attribute of the segmentation; by constructing an energy function, the segmentation of the river channel is converted into an energy function minimization problem;

Step 33: Segment the geological data based on the minimum cut criterion, update the seismic data points and the data point markers α _i in the Gaussian mixture model, and output the segmentation results after the iteration to obtain the binarized segmentation results of the channel geological body. The method of isosurface obtains the three-dimensional model of the channel geological body.

Those of ordinary skill in the art will appreciate that the embodiments described herein are for the purpose of assisting the reader in understanding the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific modifications and combinations without departing from the essence of the present invention according to these technical teachings disclosed in the present invention, and these modifications and combinations still fall within the protection scope of the present invention.

Claims (4)

1. a three-dimensional modeling method based on multi-attribute super-voxel graph cut, is characterized in that, comprises the following steps: Step 1, using the non-linear dimensionality reduction algorithm of ISOLLE to fuse the seismic attributes of the river channel to obtain the attribute data body of the river channel; Step 2, according to the data volume, through a simple linear iterative clustering algorithm (SLIC), generate a geological supervoxel; Step 3, through the k-means clustering method, establish the Gaussian mixture model of the target area and the non-target area, construct the network graph and energy function, and segment the geological data based on the minimum cut criterion to obtain the binarized segmentation result, and pass The method of extracting the isosurface can obtain the three-dimensional model of the channel geological body. 2. the three-dimensional modeling method of river channel based on multi-attribute super-voxel graph cut as claimed in claim 1, is characterized in that, described step 1 comprises following flow process: Step 11, according to the geodetic distance, search for k samples that are close to the seismic data sample point; According to the geodesic distance between two points, search the k data points with the closest geodesic distance to each seismic data sample point i in the 3D data d _G (x _i ,x _j )=min{L _G (x _i ,x _j )} Among them, L _G is the length of a path between two points, d _E is the Euclidean distance, and d _G is the geodesic distance between the two points; Step 12, constructing a local optimization reconstruction weight matrix; An error function is introduced to measure the reconstruction error, which is Among them, x _ij (j=1,2,...,k) is the k nearest neighbors of seismic data point i, w _ij is the weight between x _i and x _ij , and w _ij conforms to For each seismic data point, the error is Construct the local covariance matrix Combine the local covariance matrix and Through the Lagrange multiplier method, the locally optimized reconstruction weight matrix is obtained When the local optimization reconstruction weight matrix is a singular matrix, perform regularization processing Q ⁱ =Q ⁱ +r·I Among them, r is the regularization parameter, and I is the k×k unit matrix; Step 13, mapping all seismic data points from high-dimensional space to low-dimensional space, and calculating the value of the seismic data point in the low-dimensional space through the local reconstruction weight matrix of the seismic data point and its k neighbors; Mapping conditions are met Among them, ε(Y) is the value of the loss function, y _i is the output vector of x _i , and y _ij (j=1,2,...,k) is the k nearest neighbors found by the geodesic distance. Among them, I is the identity matrix of m×m; the loss function is Among them, M is an N×N alignment matrix, which is expressed as M=(1-W) ^T ·(IW). 3. the three-dimensional modeling method of river channel based on multi-attribute super-voxel graph cut as claimed in claim 2, is characterized in that, described step 2 comprises following flow process: Step 21, initialize the cluster center; Set the initial number of geological supervoxels to K, initialize the label of the seismic data point i in the river work area to -1, that is, label _i = -1, and initialize the distance between point i and cluster center j to infinity, that is dist _ij = +∞; Step 22, in the neighborhood of the cluster center C _j , calculate the distance from each point to C _j ; distance is in, is the distance between the seismic data point i and the cluster center C _j on the fused attribute value, is the spatial distance between the seismic data point i and the cluster center C _j in the three-dimensional space of the river channel, and m is the parameter for adjusting the attribute distance and the weight of the spatial distance; When d _ij < dist _i , update dist _i =d _ij , label _i =j; Step 23, update the cluster center where N _j is the number of seismic data points belonging to the first type of geological supervoxel; Step 24, calculate the residuals Step 25, update the cluster center, let C _j =C' _j , if E is less than the preset threshold or exceeds the maximum number of iterations, enter the next process; if E does not meet the conditions, the process returns to the step 22; Step 26, for the segmented geological supervoxel, establish an adjacency matrix A by traversing the connected region of the seismic data points in the three-dimensional space, and use the attribute histogram as the geological supervoxel feature; For the geological super-voxel i whose volume is not greater than the threshold, calculate the Babbitt distance between the geological super-voxel i and the adjacent super-voxel j Among them, M is the dimension of the gray attribute histogram of the geological supervoxel, and the geological supervoxel i is merged into the adjacent geological supervoxel with the smallest distance T; Step 27: After merging all discrete geological supervoxels smaller than the threshold, update the adjacency matrix and the attribute histogram of the geological supervoxels. 4. the three-dimensional modeling method of river channel based on multi-attribute super-voxel graph cut as claimed in claim 3, is characterized in that, described step 3 comprises following process: Step 31, through the k-means clustering method, establish a Gaussian mixture model of the target area and the non-target area; Mark the supervoxel α _i = 1 that belongs to the channel, and the remaining supervoxels α _i = 0; Perform k-means clustering on geological supervoxels by using four attributes of GLCM as features; Initialize the parameters of each Gaussian component in the Gaussian mixture model according to the clustering results, the weight ω _k , the sample mean u _k and the covariance Σ _k , the density function of the Gaussian mixture model is in, Calculate the posterior probability Compute maximum likelihood estimates of Gaussian component parameters By iterating until the likelihood function converges, the channel Gaussian mixture model and the non-channel Gaussian mixture model are obtained; Step 32, constructing a grid diagram and an energy function; The gray-scale co-occurrence matrix is used to calculate the texture properties of each metageological voxel. GLCM is the joint probability distribution of two gray values in the image. The entropy, dissimilarity, and energy properties of GLCM are selected as the eigenvalues of the supervoxel. The energy function that combines the geological supervoxel region information and edge information is Among them, C is the division of the grid map; E _r is the regional item, which represents the weight of the t-link edge of the grid map summary, reflecting the regional information of the geological supervoxel; E _b is the boundary item, representing the network map. The weight of the n-link edge reflects the boundary attribute of the segmentation; by constructing an energy function, the segmentation of the river channel is converted into an energy function minimization problem; Step 33: Segment the geological data based on the minimum cut criterion, update the seismic data points and the data point markers α _i in the Gaussian mixture model, and output the segmentation results after the iteration to obtain the binary segmentation results of the channel geological body. The method of isosurface obtains the three-dimensional model of the channel geological body.