CN109242968A - A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes - Google Patents
A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes Download PDFInfo
- Publication number
- CN109242968A CN109242968A CN201810971143.5A CN201810971143A CN109242968A CN 109242968 A CN109242968 A CN 109242968A CN 201810971143 A CN201810971143 A CN 201810971143A CN 109242968 A CN109242968 A CN 109242968A
- Authority
- CN
- China
- Prior art keywords
- geological
- attribute
- river
- distance
- river channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 230000011218 segmentation Effects 0.000 claims abstract description 35
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 19
- 239000011159 matrix material Substances 0.000 claims description 45
- 239000000203 mixture Substances 0.000 claims description 24
- 238000003064 k means clustering Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 5
- 238000007476 Maximum Likelihood Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract description 8
- 238000007500 overflow downdraw method Methods 0.000 abstract description 3
- 230000007812 deficiency Effects 0.000 abstract 1
- 150000003839 salts Chemical class 0.000 description 10
- 238000001514 detection method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 238000003708 edge detection Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 235000019994 cava Nutrition 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Graphics (AREA)
- Remote Sensing (AREA)
- Probability & Statistics with Applications (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The present invention provides a kind of river three-dimensional modeling methods cut based on the super voxel figure of more attributes, belong to geology body three-dimensional models building field.The deficiency that river is portrayed for single seismic properties, the invention proposes a kind of based on the improved more attribute fusion methods being locally linear embedding into, preferred a variety of attribute fusions are become into new attribute by ISOLLE algorithm, in view of non-linear relation existing between seismic attributes data, using a kind of nonlinear fusion method, fused attribute be better than before a variety of attributes, it is all improved to the accuracy that river edge and region are portrayed, lays a good foundation for the segmentation and reconstruction of next step；The present invention is based on the river dividing methods that super voxel and figure are cut, three-dimensional super voxel is generated by simple linear iterative algorithm, the super voxel generated has been bonded the edge in river well, and there is good homogeney, frame is cut in conjunction with figure and obtains final segmentation result, and the mode for extracting contour surface obtains the threedimensional model on river surface.
Description
Technical Field
The invention belongs to the field of geologic body three-dimensional model construction, and particularly relates to a river channel three-dimensional modeling method based on multi-attribute supermixel graph cut.
Background
Three-dimensional modeling of geologic bodies (channels) is one of the most important tasks in seismic data interpretation, as most important reservoirs exist around geologic bodies. The three-dimensional model of the geologic body not only can intuitively express the structural form of the geologic body and the distribution condition in a three-dimensional space, so that interpreters can perform quantitative analysis on the geologic body, but also provides important basis for numerical reservoir simulation and well placement of reserve calculation.
The edge detection method is a common method for geologic body detection, and the salt dome boundary is identified by adopting the edge detection-based method, but the method cannot achieve an ideal effect because of larger noise in seismic data. In the prior art, a method combining a 3D edge detector and inclination guiding is used for detection of a geologic body, so that the signal-to-noise ratio and the edge continuity are improved, and the boundary shape of the geologic body is clearer. However, the edge-based method has strong dependency on the change of the amplitude, and cannot obtain a good detection effect when the instantaneous amplitude change of the edge is not obvious. To overcome these problems, a geologic body identification method based on texture attributes is introduced. A set of texture attributes may be used to predict the probability that each pixel in the 3D cube belongs to a salt dome, which is then segmented to find the salt dome boundary. The prior art uses texture gradients to detect the boundaries of salt domes for the feature that the pixels at the edges vary greatly. However, for the texture-based method, the size of the window has a great influence on the recognition result, and the method has no universality for different work areas. The combined edge and texture attribute mixed classification method has the advantages of both an edge method and a texture method and can better detect salt domes. And further combining three attributes of edge detection, geometry and texture, and performing semi-automatic fault detection by a Support Vector Machine (SVM), wherein the detection result is matched with the original image.
However, unlike geologic bodies such as salt dunes and karst caves, the river channel itself has no obvious and definite boundary, and the river channel often changes and crosses, and the continuity is poor, so the above geologic body identification and segmentation method cannot be well applied to river channel segmentation.
Many scholars segment the river from the perspective of seismic attributes. The seismic attributes are data volumes extracted from original three-dimensional seismic data through a certain algorithm, the seismic attributes reflect the characteristics of the seismic data from different angles, and different geological structures show different characteristics on attribute values. Firstly, earthquake attributes are used for assisting in analyzing a river channel, and river channel boundaries are comprehensively compared and divided through comparative analysis of a plurality of earthquake attribute slice images. 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 do not help in dividing the river. The RGB multi-attribute fusion technology is applied to the recognition of the river channel, so that the edge of the river channel is clearer, and the resolution is improved. The seismic attributes are decomposed into three frequency bands through coherence and spectrum decomposition, the three frequency bands are respectively expressed by different colors, visualization is carried out through an RGB (red, green and blue) mixing technology, and the river channel can be rapidly and effectively identified. However, the RGB fusion technique can only fuse three attributes, has certain limitations, and is not applicable when more attributes need to be fused, and most of the studies are only to use the visualization technique to assist subjective judgment, and the river channel boundary is not qualitatively segmented.
In recent years, many scholars combine geological problems with images to achieve accurate segmentation. The improved PRC segmentation algorithm is applied to two-dimensional and three-dimensional seismic images, and salt dome boundaries can be segmented semi-automatically and accurately. Salt dome boundaries are extracted in a new image using an optimal path picking algorithm that can be quickly updated to obtain salt boundaries by selecting the optimal path with the global maximum envelope value to track highly discontinuous salt dome boundaries.
The image-based approach is also applicable to the problem of river geology. In order to solve the problems of complex structure, poor continuity and the like of the river channel, the local linear characteristic of the river channel is enhanced by a Steerable pyramid method in image processing. There is a confidence and curvature guided level set method to segment the river course from the three-dimensional seismic data. On the basis of fusing various seismic attributes, a level set method is used for carrying out segmentation modeling on various geological bodies such as karst caves, riverways and the like. But the level set based approach is more dependent on the initial shape and is less computationally efficient.
In general, most of the existing researches on the geologic body of the river channel are on the recognition and detection of a two-dimensional plane, and the researches on the establishment of a three-dimensional model of the river channel are insufficient.
Disclosure of Invention
Because geological information carried by the single attribute data is incomplete, the river channel interpretation on the single attribute data is often not accurate enough. Therefore, several attributes which can complement information need to be reasonably selected, and a nonlinear dimension reduction method is adopted, so that the data information is not lost while the dimension is reduced. When the river channel is segmented, if the segmentation is based on the pixel level, the continuity of the edge of the segmented river channel is poor, and the river channel cannot be well explained. In order to solve the problems in the prior art, the invention provides a river channel three-dimensional modeling method based on multi-attribute supermixel graph cut.
A river channel three-dimensional modeling method based on multi-attribute supermixel graph cut comprises the following steps:
step 1, fusing river channel seismic attributes by adopting a nonlinear dimension reduction algorithm of ISOLLE to obtain an attribute data body of a river channel;
step 2, generating geological hyper-voxels through a simple linear iterative clustering algorithm (SLIC) according to the data volume;
and 3, establishing a Gaussian mixture model of the target region and the non-target region by a k-means clustering method, constructing a network diagram and an energy function, segmenting geological data based on a minimum segmentation criterion to obtain a binary segmentation result, and obtaining a three-dimensional model of the river geologic body by a method of extracting an isosurface.
Further, the step 1 comprises the following steps:
step 11, searching k samples which are close to the seismic data sample points according to the geodesic distance;
searching k data points which are closest to the geodetic distance of each seismic data sample point i in the three-dimensional data according to the geodetic distance between the two points
dG(xi,xj)＝min{LG(xi,xj)}
Wherein L isGLength of a path between two points, dEIs the Euclidean distance, dGThe geodesic distance between two points;
step 12, constructing a local optimization reconstruction weight matrix;
an error function is introduced to measure the reconstruction error magnitude, which is
Wherein x isij(j 1, 2.. k.) is k neighbors of seismic data point i, wijIs xiAnd xijWeight value between, wijConform to
For each seismic data point, the error is
Constructing a local covariance matrix
Combining the local covariance matrix sumObtaining a local optimal reconstruction weight matrix by a Lagrange multiplier method
When the local optimization reconstruction weight matrix is a singular matrix, regularization processing is carried out
Qi＝Qi+r·I
Wherein r is a regularization parameter, and I is a k × k unit matrix;
step 13, mapping all seismic data points from a high-dimensional space to a low-dimensional space, and calculating the value of the seismic data point in the low-dimensional space through a local reconstruction weight matrix of the seismic data point and k adjacent points of the local reconstruction weight matrix;
the mapping condition is satisfied
Wherein ε (Y) is the value of the loss function, YiIs xiOutput vector, yij(j ═ 1,2, …, k) are k neighbors found by geodesic distance finding, while also satisfying
Wherein I is an m × m identity matrix; a loss function of
Where M is an NxN array matrix, expressed as
M＝(1-W)T·(I-W)。
Further, the step 2 comprises the following steps:
step 21, initializing a clustering center;
setting the initial number of geological hyper-voxels to be K, and initializing the label of a seismic data point i in a river work area to be-1, namely labeliWhen l is-1, the distance between the point i and the cluster center j is initialized to infinity, i.e., distij＝+∞；
Step 22, in the clustering center CjIn the neighborhood of (2), calculate each point to CjThe distance of (d);
a distance of
Wherein,as seismic data point i and cluster center CjThe distance over the fused attribute values,is a seismic data point i and a cluster center CjThe space distance in the three-dimensional space of the river channel, wherein m is a parameter for adjusting the attribute distance and the space distance weight;
when d isij＜distiUpdate disti＝dij，labeli＝j；
Step 23, updating the clustering centerWherein N isjThe number of seismic data points belonging to a class II geological voxel;
step 24, calculating the residual error
Step 25, updating the clustering center, and ordering Cj＝C'jIf E is smaller than a preset threshold value or exceeds the maximum iteration number, entering the next process; if E does not satisfy the condition, the flow returns to the step 22;
step 26, for the divided geological superpixel, establishing an adjacency matrix A by traversing a connected region of seismic data points in a three-dimensional space, and taking an attribute histogram as a geological superpixel feature;
for geological super voxel i with the volume not larger than the threshold value, calculating the Pasteur distance between the geological super voxel i and the adjacent super voxel j
Wherein M is the dimension of the geological superpixel gray attribute histogram, and the geological superpixel i is merged into the adjacent geological superpixel with the minimum distance T;
and 27, after all discrete geological voxels smaller than the threshold are combined, updating the attribute histograms of the adjacent matrix and the geological voxels.
Further, the step 3 includes the following steps:
step 31, establishing a Gaussian mixture model of a target region and a non-target region by a k-means clustering method;
labeling superpixels α belonging to a riveri1, the remaining voxels αi＝0；
Performing k-means clustering on geological hyper-voxels by taking four attributes of GLCM as features;
initializing the parameter of each Gaussian component in the Gaussian mixture model according to the clustering result, wherein the weight is omegakMean value u of samplekSum covariance ∑kThe density function of the Gaussian mixture model is
Wherein,
calculating posterior probability
Computing maximum likelihood estimates of gaussian component parameters
Obtaining a river channel Gaussian mixture model and a non-river channel Gaussian mixture model through iteration to likelihood function convergence;
step 32, constructing a grid graph and an energy function;
a gray-level co-occurrence matrix is employed to compute the texture attributes for each super-geological voxel. GLCM is the joint probability distribution of two gray values in an image, three attributes of entropy, dissimilarity and energy of GLCM are selected as characteristic values of hyper-voxels, and an energy function integrating geological hyper-voxel region information and edge information is
Wherein, C is the division of the grid graph; erThe weight of the t-link edge is summarized for the area item by the representative grid diagram, and the area information of the geological hyper-voxel is reflected; ebThe boundary item represents the weight of the n-link edge in the network graph and embodies the boundary attribute of the segmentation; by constructing an energy function, the division of the river channel is converted into the problem of minimization of the energy function;
step 33, segmenting the geological data based on the minimum segmentation criteria and updating the seismic data points and data point markers α in the Gaussian mixture modeliAnd outputting a segmentation result after iteration is finished to obtain a binary segmentation result of the river channel geologic body, and obtaining a three-dimensional model of the river channel geologic body by a method of extracting an isosurface.
The invention has the beneficial effects that: the invention provides a river channel three-dimensional modeling method based on multi-attribute super-voxel graph cutting. The ISOLLE fusion method can effectively fuse various seismic attributes, reasonable seismic attribute selection is beneficial to information complementation, the riverway geologic body can be more comprehensively depicted, the nonlinear relation in riverway seismic data can be kept by a multi-attribute fusion mode, and the obtained new attribute body lays a good foundation for next segmentation and reconstruction; the river channel geological hyper-voxel has good edge retentivity, the geological hyper-voxel can be accurately and quickly segmented by combining with graph segmentation, the conversion of a three-dimensional seismic data river channel geological body three-dimensional model is realized through isosurface extraction, and the spatial characteristics of the river channel geological body are more visually displayed; the method based on the multi-attribute fusion hypergsomal graph cut realizes the accurate depiction of the river channel geologic body, the finally obtained three-dimensional model accords with the geological rule, the result is basically consistent with the result obtained by geologists, and the foundation is laid for the follow-up work.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a flow chart of step 1 in FIG. 1.
Fig. 3 is a flowchart of step 2 in fig. 1.
Fig. 4 is an initialized cluster center diagram.
Fig. 5 is a flowchart of step 3 in fig. 1.
Detailed Description
The embodiments of the present invention will be further described with reference to the accompanying drawings.
Referring to fig. 1, the invention provides a channel three-dimensional modeling method based on multi-attribute superminiature graph cutting, which is implemented by the following steps:
step 1, fusing the seismic attributes of the river channel by adopting a nonlinear dimension reduction algorithm of ISOLLE to obtain an attribute data body of the river channel.
In the embodiment, different attribute data describe the geological structure from different angles in the seismic interpretation process, and the geological structure of the reservoir can be more accurately interpreted by fusing different attributes. Because the relationship between seismic attributes and geological features is usually nonlinear, the linear transformation-based PCA method cannot sufficiently reflect the nonlinear relationship, and the accuracy of prediction identification is reduced. The nonlinear LLE dimension reduction algorithm adopts the distance between the Euclidean distance calculation point and the point, cannot reflect the real structure between the points, and is very sensitive to the selection of the number of the adjacent points.
In the embodiment, the ISOLLE algorithm is adopted, the concept of geodesic distance is introduced, the Euclidean distance is replaced by the geodesic distance, the advantage of processing high-dimensional manifold data by the LLE algorithm is kept, and meanwhile the compactness of the data in a corresponding low-dimensional space and the linear characteristics of local neighborhood data are improved. The ISOLLE method is applied to multi-attribute fusion of three-dimensional river channel seismic data for the first time, and several attributes are selected to be fused into a new attribute body which is expanded to a three-dimensional space.
In the embodiment, the multi-attribute fusion problem of a certain earthquake work area is researched by selecting the root mean square amplitude attribute, the energy attribute, the texture homogeneity, the frequency spectrum attribute and the instantaneous frequency attribute. And obtaining slice images with different attributes after normalization processing and Fourier transform denoising. The table with the same attribute on different time slices has good or bad performance, and the different attributes can make up for the defects of each other.
Referring to fig. 2, step 1 is implemented by the following process:
step 11, searching k samples which are close to the seismic data sample points according to the geodesic distance;
searching k data points which are closest to the geodetic distance of each seismic data sample point i in the three-dimensional data according to the geodetic distance between the two points
dG(xi,xj)＝min{LG(xi,xj)}
Wherein L isGLength of a path between two points, dEIs the Euclidean distance, dGThe geodesic distance between two points;
and step 12, obtaining a local reconstruction matrix of the seismic data points. The local reconstruction matrix represents a local linear relationship among multiple attributes, and the linear relationship can be maintained among the fused seismic data points.
An error function is introduced to measure the reconstruction error magnitude, which is
Wherein x isij(j 1, 2.. k.) is k neighbors of seismic data point i, wijIs xiAnd xijWeight value between, wijConform to
For each seismic data point, the error is
Constructing a local covariance matrix
Combining local covariance matrix sumsObtaining a local optimization reconstruction weight matrix by a Lagrange multiplier method
When the local optimization reconstruction weight matrix is a singular matrix, regularization processing is carried out
Qi＝Qi+r·I
Wherein r is a regularization parameter, and I is a k × k unit matrix;
step 13, mapping all seismic data points from a high-dimensional space to a low-dimensional space, and calculating the value of the seismic data point in the low-dimensional space through a local reconstruction weight matrix of the seismic data point and k adjacent points of the local reconstruction weight matrix;
the mapping condition is satisfied
Wherein ε (Y) is the value of the loss function, YiIs xiOutput vector, yij(j ═ 1,2, …, k) are k neighbor points found by geodesic distance finding, and satisfy the following two conditions
Wherein I is an m × m identity matrix. Herein, theUsually stored in an N sparse matrix W, when xjIs xiWhen there is a close neighborhood of points, wij＝wjIf the two are not equal, then wij0. These two conditions have their own meanings, the first one ensuresTranslational invariance to Y; the second one guarantees the reconstruction error in the low dimensional space by different coordinatesThe same measurement standards may also be applied. Thereby preventing the occurrence of a degenerate solution in which Y is 0. A loss function of
Where M is an NxN array matrix, expressed as
M＝(1-W)T·(I-W)。
The matrix M has the characteristics of sparseness and semidefinite. If the loss function is to be minimized, then the eigenvector corresponding to the smallest non-zero eigenvalue should be taken. Meanwhile, the characteristic values are arranged according to the order from small to large. The first eigenvalue is approximately equal to zero and is therefore discarded.
And 2, generating geological hyper-voxels by a simple linear iterative clustering algorithm (SLIC) according to the data volume.
In this embodiment, the pixel-level based segmentation does not fully utilize the local relationship between pixels, and thus, an error is caused in the process of discretizing the image. Superpixels use the similarity of features between pixels to group pixels, and a small number of superpixels are used for replacing a large number of pixels to express picture features, so that compared with pixel-level segmentation, the edge features of an original image can be maintained. A superpixel is an extension of a superpixel in three-dimensional space. When three-dimensional seismic data are processed, the geological hyper-voxel method overcomes the problems of unsmooth and the like when a three-dimensional model is finally synthesized by using a hyper-pixel in a two-dimensional slice, and simultaneously reduces the number of blocks and the calculated amount.
In this embodiment, in the seismic data volume, the probability that two seismic data points that are closer to each other belong to the same geologic body is also higher. According to the characteristic, similar seismic data points are clustered based on a SLIC algorithm to generate geological hyper-voxels. The generated geological voxel can keep the interior uniform and compact, can also keep the edge characteristics of the geological body, and reduces the calculation complexity of subsequent segmentation. In the SLIC algorithm, the main step is to compare each point with the clustering center CjThe distance of the cluster is classified, and then the cluster center is updated. Assuming a three-dimensional data volume of Crossline Inlinc Time size after fusion, the set of all seismic data points V ═ V { (V } V1,v2,...,vNIn which v isi＝{gi,xi,yi,zi}，giIs the fused attribute value, x, of the seismic data point ii， yi，ziThe coordinates in space of grid point i.
Referring to fig. 3, step 2 is implemented by the following process:
step 21, initializing a clustering center;
setting the initial number of geological voxels to be K, and then setting the initial size of the voxels to be KThe river channel area is divided into K small areas. In the three-dimensional space of the river channel, the seed points are selected in a staggered mode as shown in fig. 4, so that the seed points can be distributed in the space as uniformly as possible. Then the initial cluster center C ═ C1,C2,...,CKThe distance between super voxel cluster centers of adjacent river channels is
In order to avoid the situation that the generated seed points may fall on the edge of the river channel, gradient values of all grid points in a 3 × 3 × 3 neighborhood of the current seed point are calculated, and the position of the seed point is moved to the position where the gradient of the neighborhood is minimum. I.e., the cluster center is moved to G (x, y, z) ═ minG (x, y, z).
And calculating the attribute gradient of each seismic data point in the river channel in a 3 multiplied by 3 window taking the seed point as the center. The calculation method is 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 an attribute value when the coordinates in the channel work area are (x, y, z). G (x, y, z) represents a gradient value calculated based on the attribute value at coordinates (x, y, z) within the channel region.
Initializing the label of a seismic data point i in a river work area to be-1, namely labeli＝-1；
The distance between point i and cluster center j is initialized to infinity, i.e., distij＝+∞。
Step 22, in the clustering center CjIn the neighborhood of (2), calculate each point to CjThe distance of (d);
for all cluster centers CjWithin a 2S x 2S neighborhood, each point reaches the cluster center CjAt a distance of
Wherein,as seismic data point i and cluster center CjThe distance over the fused attribute values,is a seismic data point i and a cluster center CjThe space distance in the three-dimensional space of the river channel, wherein m is a parameter for adjusting the attribute distance and the space distance weight;
when d isij＜distiUpdate disti＝dij，labeli＝j；
Step 23, updating the clustering centerWherein N isjThe number of seismic data points belonging to a class II geological voxel;
step 24, calculating the residual error
Step 25, updating the clustering center, and ordering Cj＝C'jIf E is smaller than a preset threshold value or exceeds the maximum iteration number, entering the next process; if E does not satisfy the condition, the flow returns to the step 22;
step 26, for the divided geological superpixel, establishing an adjacency matrix A by traversing a connected region of seismic data points in a three-dimensional space, and taking an attribute histogram as a geological superpixel feature;
for geological super voxel i with the volume not larger than the threshold value, calculating the Pasteur distance between the geological super voxel i and the adjacent super voxel j
Wherein M is the dimension of the geological superpixel gray attribute histogram, and the geological superpixel i is merged into the adjacent geological superpixel with the minimum distance T;
and 27, after all discrete geological voxels smaller than the threshold are combined, updating the attribute histograms of the adjacent matrix and the geological voxels.
In this embodiment, similar data points are considered to be clustered and considered as the minimum unit of the river channel for segmentation. And (4) performing three-dimensional voxel segmentation on the fused river channel data volume, and setting different geological voxel numbers K to obtain different results. And when the number K of the geological voxels is set to be small, the partitioned geological voxels are not attached to the edge of the river channel sufficiently. When K is large, the distinction between the inside and the outside of the river channel is fuzzy, and the calculated amount of graph segmentation cannot be effectively reduced. Therefore, the number K of the super voxels is set to be 150, the segmented super voxels can be well attached to the edge of a river channel, and the homogeneity of the geological super voxels is better. However, K may be set to other values.
And 3, establishing a Gaussian mixture model of the target region and the non-target region by a k-means clustering method, constructing a network diagram and an energy function, segmenting geological data based on a minimum segmentation criterion to obtain a binary segmentation result, and obtaining a three-dimensional model of the river geologic body by a method of extracting an isosurface.
In this embodiment, since the geological voxel is a small region, there are hidden texture features in the region. Therefore, on the basis of the traditional image segmentation algorithm, the attribute value and the textural feature of the super voxel are fused to replace the gray attribute to represent the characteristic of each super voxel, and the image segmentation algorithm based on the super voxel is applied to the segmentation of the river geologic body for the first time. Compared with the traditional graph cut algorithm, the method uses a Gaussian Mixture Model (GMM) to replace the attribute histogram so as to accurately express the probability model. The gaussian mixture model is a linear superposition of multiple gaussian models. The core of GMM is to find the mean, variance, and weight ratio of each gaussian component. Firstly, the parameter of each Gaussian component in the GMM is initialized randomly, and the density function, the posterior probability and the maximum likelihood estimation of the parameter of the Gaussian component are calculated. Finally, continuously repeating iteration until the likelihood function converges to obtain a gaussian mixture model of the river geologic body, please refer to fig. 5, which is specifically realized by the following process:
step 31, establishing a Gaussian mixture model of a target region and a non-target region by a k-means clustering method;
labeling superpixels α belonging to a riveri1, the remaining voxels αi＝0；
Setting K to be 5, and performing K-means clustering on geological hyper-voxels by taking four attributes of GLCM as features;
according to the clustering result, randomly determiningInitializing the parameters of each Gaussian component in the Gaussian mixture model, weight omegakSample mean ukSum covariance ∑kThe density function of the Gaussian mixture model is
Wherein,
calculating posterior probability
Computing maximum likelihood estimates of gaussian component parameters
Obtaining a river channel Gaussian mixture model and a non-river channel Gaussian mixture model through iteration to likelihood function convergence;
step 32, constructing a grid graph and an energy function;
a gray-level co-occurrence matrix is employed to compute the texture attributes for each super-geological voxel. GLCM is the joint probability distribution of two gray values in an image, three attributes of entropy, dissimilarity and energy of GLCM are selected as characteristic values of hyper-voxels, and an energy function integrating geological hyper-voxel region information and edge information is
Wherein, C is the division of the grid graph; erThe weight of the t-link edge is summarized for the area item by the representative grid diagram, and the area information of the geological hyper-voxel is reflected; ebThe boundary item represents the weight of the n-link edge in the network graph and embodies the boundary attribute of the segmentation; by constructing an energy function, the division of the river channel is converted into the problem of minimization of the energy function;
step 33, segmenting the geological data based on the minimum segmentation criteria and updating the seismic data points and data point markers α in the Gaussian mixture modeliAnd outputting a segmentation result after iteration is finished to obtain a binary segmentation result of the river channel geologic body, and obtaining a three-dimensional model of the river channel geologic body by a method of extracting an isosurface.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention, and it is to be understood that the scope of the invention is not to be limited to such specific statements and embodiments. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (4)
1. A river channel three-dimensional modeling method based on multi-attribute supermixel graph cut is characterized by comprising the following steps:
step 1, fusing river channel seismic attributes by adopting a nonlinear dimension reduction algorithm of ISOLLE to obtain an attribute data body of a river channel;
step 2, generating geological hyper-voxels through a simple linear iterative clustering algorithm (SLIC) according to the data volume;
and 3, establishing a Gaussian mixture model of the target region and the non-target region by a k-means clustering method, constructing a network map and an energy function, segmenting geological data based on a minimum segmentation criterion to obtain a binary segmentation result, and obtaining a three-dimensional model of the river geologic body by a method of extracting an isosurface.
2. The method for three-dimensional modeling of river channel based on multi-attribute superminiature graph cut according to claim 1, wherein the step 1 comprises the following procedures:
step 11, searching k samples which are close to the seismic data sample points according to the geodesic distance;
searching k data points which are closest to the geodetic distance of each seismic data sample point i in the three-dimensional data according to the geodetic distance between the two points
dG(xi,xj)＝min{LG(xi,xj)}
Wherein L isGLength of a path between two points, dEIs the Euclidean distance, dGIs the geodesic distance between two points;
step 12, constructing a local optimization reconstruction weight matrix;
an error function is introduced to measure the reconstruction error magnitude, which is
Wherein x isij(j 1, 2.. k.) is k neighbors of seismic data point i, wijIs xiAnd xijWeight value of between, wijConform to
For each seismic data point, the error is
Constructing a local covariance matrix
Combining the local covariance matrix sumObtaining a local optimization reconstruction weight matrix by a Lagrange multiplier method
When the local optimization reconstruction weight matrix is a singular matrix, regularization processing is carried out
Qi＝Qi+r·I
Wherein r is a regularization parameter, and I is a k × k unit matrix;
step 13, mapping all seismic data points from a high-dimensional space to a low-dimensional space, and calculating the value of the seismic data point in the low-dimensional space through a local reconstruction weight matrix of the seismic data point and k adjacent points of the local reconstruction weight matrix;
the mapping condition is satisfied
Wherein ε (Y) is the value of the loss function, YiIs xiOutput vector, yij(j ═ 1,2, …, k) are k neighbors found by geodesic distance finding, while also satisfying
Wherein I is an m × m identity matrix; a loss function of
Where M is an NxN array matrix, expressed as
M＝(1-W)T·(I-W)。
3. The method for three-dimensional modeling of river channel based on multi-attribute superminiature graph cut according to claim 2, wherein said step 2 comprises the following procedures:
step 21, initializing a clustering center;
setting the initial number of geological hyper-voxels to be K, and initializing the label of a seismic data point i in a river work area to be-1, namely labeliThe distance between point i and cluster center j is initialized to infinity, i.e., dist, to-1ij＝+∞；
Step 22, in the clustering center CjIn the neighborhood of (2), calculate each point to CjThe distance of (d);
a distance of
Wherein,as seismic data point i and cluster center CjThe distance over the fused attribute values,is a seismic data point i and a cluster center CjThe space distance in the three-dimensional space of the river channel, m is a parameter for adjusting the attribute distance and the space distance weight;
when d isij＜distiUpdate disti＝dij，labeli＝j；
Step 23, updating the clustering centerWherein N isjThe number of seismic data points belonging to a class II geological voxel;
step 24, calculating the residual error
Step 25, updating the clustering center, and ordering Cj＝C'jIf E is smaller than a preset threshold value or exceeds the maximum iteration number, entering the next process; if E does not satisfy the condition, the flow returns to the step 22;
step 26, for the divided geological superpixel, establishing an adjacency matrix A by traversing a connected region of seismic data points in a three-dimensional space, and taking an attribute histogram as a geological superpixel feature;
for geological super voxel i with the volume not larger than the threshold value, calculating the Pasteur distance between the geological super voxel i and the adjacent super voxel j
Wherein M is the dimension of the geological superpixel gray attribute histogram, and the geological superpixel i is merged into the adjacent geological superpixel with the minimum distance T;
and 27, after all discrete geological voxels smaller than the threshold are combined, updating the attribute histograms of the adjacent matrix and the geological voxels.
4. The method for three-dimensional modeling of river channel based on multi-attribute superminiature graph cut according to claim 3, wherein said step 3 comprises the following procedures:
step 31, establishing a Gaussian mixture model of a target region and a non-target region by a k-means clustering method;
labeling superpixels α belonging to a riveri1, the remaining voxels αi＝0；
Performing k-means clustering on geological hyper-voxels by taking four attributes of GLCM as features;
initializing the parameter of each Gaussian component in the Gaussian mixture model according to the clustering result, wherein the weight is omegakSample mean ukSum covariance ∑kThe density function of the Gaussian mixture model is
Wherein,
calculating posterior probability
Computing maximum likelihood estimates of gaussian component parameters
Obtaining a river channel Gaussian mixture model and a non-river channel Gaussian mixture model through iteration to likelihood function convergence;
step 32, constructing a grid graph and an energy function;
a gray-level co-occurrence matrix is employed to compute the texture attributes for each super-geological voxel. GLCM is the joint probability distribution of two gray values in an image, three attributes of entropy, dissimilarity and energy of GLCM are selected as characteristic values of hyper-voxels, and an energy function integrating geological hyper-voxel region information and edge information is
Wherein, C is the division of the grid graph; erThe weight of the t-link edge is summarized for the area item by the representative grid diagram, and the area information of the geological hyper-voxel is reflected; ebThe boundary item represents the weight of the n-link edge in the network graph and embodies the boundary attribute of the segmentation; the method comprises the steps of converting the division of the river channel into an energy function minimization problem by constructing an energy function;
step 33, segmenting the geological data based on the minimum segmentation criteria and updating the seismic data points and data point markers α in the Gaussian mixture modeliAnd outputting a segmentation result after iteration is finished to obtain a binarization segmentation result of the river channel geologic body, and obtaining a three-dimensional model of the river channel geologic body by a method of extracting an isosurface.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810971143.5A CN109242968A (en) | 2018-08-24 | 2018-08-24 | A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810971143.5A CN109242968A (en) | 2018-08-24 | 2018-08-24 | A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109242968A true CN109242968A (en) | 2019-01-18 |
Family
ID=65069261
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810971143.5A Pending CN109242968A (en) | 2018-08-24 | 2018-08-24 | A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242968A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349225A (en) * | 2019-07-12 | 2019-10-18 | 四川易利数字城市科技有限公司 | A kind of BIM model exterior contour rapid extracting method |
| CN110706191A (en) * | 2019-10-14 | 2020-01-17 | 中国空气动力研究与发展中心超高速空气动力研究所 | High-pressure container detection method based on infrared image splicing fusion algorithm |
| CN112346117A (en) * | 2019-08-09 | 2021-02-09 | 中国石油天然气集团有限公司 | Reservoir characteristic prediction method and device based on seismic attribute fusion |
| CN112415582A (en) * | 2020-10-15 | 2021-02-26 | 山东大学 | Three-dimensional speed geological modeling method for random arrangement of structure and wave velocity |
| CN112649863A (en) * | 2019-10-12 | 2021-04-13 | 中国石油化工股份有限公司 | Frequency division seismic attribute data optimization method and system |
| CN115665031A (en) * | 2022-12-27 | 2023-01-31 | 中南大学 | Three-dimensional irregular edge network perception data acquisition method, system and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862698A (en) * | 2017-11-29 | 2018-03-30 | 首都师范大学 | Light field foreground segmentation method and device based on K mean cluster |
| CN108090485A (en) * | 2017-11-28 | 2018-05-29 | 西安电子科技大学昆山创新研究院 | Display foreground extraction method based on various visual angles fusion |
-
2018
- 2018-08-24 CN CN201810971143.5A patent/CN109242968A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090485A (en) * | 2017-11-28 | 2018-05-29 | 西安电子科技大学昆山创新研究院 | Display foreground extraction method based on various visual angles fusion |
| CN107862698A (en) * | 2017-11-29 | 2018-03-30 | 首都师范大学 | Light field foreground segmentation method and device based on K mean cluster |
Non-Patent Citations (2)
| Title |
|---|
| CLAUDIO VARINI等: "ISOLLE: Locally Linear Embedding with Geodesic Distance", 《KNOWLEDGE DISCOVERY IN DATABASES: PKDD 2005》 * |
| 戴庆焰 等: "基于超像素和改进迭代图割算法的图像分割", 《计算机工程》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349225A (en) * | 2019-07-12 | 2019-10-18 | 四川易利数字城市科技有限公司 | A kind of BIM model exterior contour rapid extracting method |
| CN110349225B (en) * | 2019-07-12 | 2023-02-28 | 四川易利数字城市科技有限公司 | BIM model external contour rapid extraction method |
| CN112346117A (en) * | 2019-08-09 | 2021-02-09 | 中国石油天然气集团有限公司 | Reservoir characteristic prediction method and device based on seismic attribute fusion |
| CN112649863A (en) * | 2019-10-12 | 2021-04-13 | 中国石油化工股份有限公司 | Frequency division seismic attribute data optimization method and system |
| CN110706191A (en) * | 2019-10-14 | 2020-01-17 | 中国空气动力研究与发展中心超高速空气动力研究所 | High-pressure container detection method based on infrared image splicing fusion algorithm |
| CN110706191B (en) * | 2019-10-14 | 2022-09-20 | 中国空气动力研究与发展中心超高速空气动力研究所 | High-pressure container detection method based on infrared image splicing fusion algorithm |
| CN112415582A (en) * | 2020-10-15 | 2021-02-26 | 山东大学 | Three-dimensional speed geological modeling method for random arrangement of structure and wave velocity |
| CN115665031A (en) * | 2022-12-27 | 2023-01-31 | 中南大学 | Three-dimensional irregular edge network perception data acquisition method, system and device |
| CN115665031B (en) * | 2022-12-27 | 2023-04-07 | 中南大学 | Three-dimensional irregular edge network perception data acquisition method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109242968A (en) | A kind of river three-dimensional modeling method cut based on the super voxel figure of more attributes | |
| CN109522908B (en) | Image significance detection method based on region label fusion | |
| Sun et al. | Aerial 3D building detection and modeling from airborne LiDAR point clouds | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| WO2019071976A1 (en) | Panoramic image saliency detection method based on regional growth and eye movement model | |
| CN106846322B (en) | The SAR image segmentation method learnt based on curve wave filter and convolutional coding structure | |
| CN110443809A (en) | Structure sensitive property color images super-pixel method with boundary constraint | |
| CN110796667A (en) | Color image segmentation method based on improved wavelet clustering | |
| CN107330875A (en) | Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images | |
| Sun et al. | Classification of MLS point clouds in urban scenes using detrended geometric features from supervoxel-based local contexts | |
| CN113177592B (en) | Image segmentation method and device, computer equipment and storage medium | |
| Chen et al. | A local tangent plane distance-based approach to 3D point cloud segmentation via clustering | |
| CN111091129A (en) | Image salient region extraction method based on multi-color characteristic manifold sorting | |
| Sutha | Object based classification of high resolution remote sensing image using HRSVM-CNN classifier | |
| Yalic et al. | Automatic Object Segmentation on RGB-D Data using Surface Normals and Region Similarity. | |
| Takahashi et al. | Applying manifold learning to plotting approximate contour trees | |
| Srivastava et al. | Drought stress classification using 3D plant models | |
| Xiao et al. | Auto rock detection via sparse-based background modeling for mars rover | |
| CN112884884A (en) | Candidate region generation method and system | |
| Chochia | Image segmentation based on the analysis of distances in an attribute space | |
| Harshit et al. | Geometric features interpretation of photogrammetric point cloud from Unmanned Aerial Vehicle | |
| CN116645499A (en) | Determination method, determination device, determination apparatus, determination device, determination program storage medium, and determination program product | |
| CN108765384B (en) | Significance detection method for joint manifold sequencing and improved convex hull | |
| CN111539966A (en) | Colorimetric sensor array image segmentation method based on fuzzy c-means clustering | |
| Zhang et al. | Automatic superpixel generation algorithm based on a quadric error metric in 3D space |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190118 |