A method for outcrop fracture identification based on lidar technology

Abstract

The present invention relates to the technical field of oil and gas exploration, and discloses a method for identifying outcrop fractures based on lidar technology, which includes the following steps: Acquisition of three-dimensional point cloud data: Scanning the outcrop of the experimental area with a lidar system to obtain the three-dimensional point cloud data of the outcrop of the experimental area; Data grid processing: Performing data grid processing on the processed point cloud data to obtain the two-dimensional cloud quality data and two-dimensional cloud density data of the experimental area; Tensor voting: Performing tensor voting on the obtained two-dimensional cloud quality data to remove the background noise of the image; Optimal circumscribed rectangle processing: Performing optimal circumscribed rectangle recognition on the denoised image to identify the optimal circumscribed rectangle of each fracture and obtain the fracture attribute parameters. This fracture detection and identification method can more efficiently and accurately identify some high-angle fractures that are impossible for humans to measure, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and particularly relates to a method for identifying outcrop fractures based on lidar technology. Background Art

[0002] Due to the continuous increase in the demand for oil and gas resources and the continuous exploration and utilization of oil and gas resources, the reserves of conventional pore-type oil and gas reservoirs are decreasing day by day, and the difficulty of oil and gas exploration and development is increasing. The direction of oil and gas exploration is gradually shifting from shallow to deep, and from conventional oil and gas reservoirs to special oil and gas reservoirs. The research on reservoirs has also gradually developed from the research on conventional pore-type reservoirs to the research on various other types of reservoirs. Fractured reservoirs refer to reservoirs with fractures as the main storage space and seepage channels, which have the characteristics of high permeability. They are the main type of current oil and gas reservoir reservoirs and also a research hotspot. However, due to the complex distribution and poor regularity of fractures, and at the same time being restricted by observation means and research methods, so far, the fine description and prediction of fracture grids are still an unsolved problem. At present, the main methods for detecting fractures are on-site measurement, image recognition, etc. The image recognition method using on-site photography has problems of low accuracy and large errors. On-site measurement is the currently commonly used method, which has higher accuracy, but this measurement method takes a long time, has a high risk factor, and is greatly limited in the measurement range. A new method for efficiently, safely, and accurately extracting fracture-related information is very necessary. In view of these situations, researchers have proposed a method that can use lidar scanning to obtain the three-dimensional point cloud data of rocks for fracture detection.

[0003] Three-dimensional lidar scanning technology is a fully automatic and high-precision three-dimensional scanning method, which has been applied to many research fields such as geophysics in recent years. Currently, the common lidar scanning technology is widely used in the detection of regular plane fractures such as road surfaces and city walls. The main processing flow of conventional point clouds is polynomial surface fitting and principal component analysis for denoising and dimensionality reduction. However, the outcrop in the wild is more complex than the road surface situation, the fracture development law is poor, and the application of the conventional point cloud processing flow surface fitting to the outcrop in the wild will cover up many fracture details. The principal component analysis for denoising and dimensionality reduction cannot enhance the fracture characteristics, but only remove noise points, and the effect is not good. Experimental research also proves that the denoising effect of the conventional point cloud processing flow for extracting outcrop fractures in the wild is poor, and the fracture smoothing effect is not good. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for identifying outcrop fractures based on lidar technology, so as to solve the problems that the denoising effect of the conventional point cloud processing flow for extracting outcrop fractures in the wild is poor and the fracture smoothing effect is not good.

[0005] To achieve the above-mentioned invention purpose, the technical solution adopted by the present invention is: A method for identifying outcrop fractures based on lidar technology, comprising the following steps:

[0006] a. Acquisition of three-dimensional point cloud data: Use a lidar system to scan the outcrop in the experimental area to obtain the three-dimensional point cloud data of the outcrop in the experimental area;

[0007] b. Data gridding processing: Perform data gridding processing on the processed point cloud data to obtain the two-dimensional cloud quality data and two-dimensional cloud density data of the experimental area;

[0008] c. Tensor voting: Perform tensor voting on the obtained two-dimensional cloud quality data to remove the background noise of the image;

[0009] d. Optimal circumscribed rectangle processing: Perform optimal circumscribed rectangle recognition on the denoised image to identify the optimal circumscribed rectangle of each crack and obtain the crack attribute parameters.

[0010] Preferably, the data gridding processing includes the following sub-steps:

[0011] b01. Set appropriate grid numbers and grid sizes according to the point cloud data;

[0012] b02. Determine the position of each point in the point cloud data in the grid by calculation using the XY coordinates of each point;

[0013] b03. Perform blank pixel detection based on Z difference, subtract the minimum value of the Z value in the grid from the maximum value to obtain the two-dimensional cloud quality distribution, and at the same time export the two-dimensional cloud quality data;

[0014] b04. Retrieve blank pixels from the grid data of the XY coordinates to obtain the two-dimensional cloud density data.

[0015] Preferably, points with the same XY coordinates in the point cloud data are placed in the same grid, and the processing function for this step is:

[0016] i = (X i - X min ) / dx + 1;

[0017] j = (Y i - Y min ) / dy + 1;

[0018] Where: i represents the abscissa of the grid, j represents the ordinate of the grid, X i (i = 1, 2, 3...) represents the abscissa of each point, Y i (i = 1, 2, 3...) represents the ordinate of each point, X min represents the minimum value of X; Y min represents the minimum value of Y, dx is the width of the grid, and dy is the height of the grid.

[0019] Preferably, the tensor voting includes the following processing sub-steps:

[0020] c01. Perform preliminary denoising on the two-dimensional data, where the two-dimensional data includes two-dimensional cloud quality data and two-dimensional cloud density;

[0021] c02. Perform tensor voting on the filtered two-dimensional data. Represent the two-dimensional data as I(x, y), then represents the gradient vector of each pixel point, and use the gradient vector to construct a tensor model;

[0022] c03. Perform eigenvalue analysis, use the main eigenvalue λ1 as the input data for the voting process, and then perform the superposition of the voting fields;

[0023] c04. Perform tensor decomposition and eigenanalysis, place the center of the voting field at the voting point, accumulate the neighborhood votes for each point, calculate the quantity and magnitude of the votes obtained for each pixel, and finally form a new tensor at each pixel point.

[0024] Preferably, the optimal circumscribed rectangle processing includes the following sub-steps:

[0025] d01. Convert the two-dimensional cloud quality map after tensor voting into a binary map, and remove background noise during the conversion process;

[0026] d02. Perform the optimal circumscribed rectangle algorithm recognition on the binary map to obtain the optimal circumscribed rectangle of each crack;

[0027] d03. Screen the rectangles with area, screen out and eliminate the rectangles with too large or too small area, and count the rectangle attribute parameters of the remaining rectangles;

[0028] d04. Convert the rectangle attribute parameters into crack attribute parameters.

[0029] Preferably, the rectangle attribute parameters include at least one of the number, length, width, inclination angle, or area of the rectangle.

[0030] Preferably, the crack attribute parameters include at least one of length information, width information, or inclination angle information.

[0031] The beneficial effects of the present invention are mainly reflected in:

[0032] The present invention uses a lidar scanning system to scan and obtain point cloud data of outcrops in the wild, including the three-dimensional coordinates and RGB information of each point; the scanned three-dimensional point cloud data is processed into a grid to obtain two-dimensional cloud density and two-dimensional cloud quality data, and tensor voting is performed on the obtained two-dimensional data to improve the resolution of the image and remove a large amount of background noise; finally, the optimal circumscribed rectangle of the obtained image is recognized to obtain attributes such as the length, width, and inclination angle of the crack. Therefore, this crack detection and recognition method can more efficiently and accurately identify some high-angle cracks that are inaccessible to humans for measurement, etc. Description of the Drawings

[0033] Figure 1 is the overall flowchart of the recognition method of the present invention;

[0034] Figure 2 is the surface fitting diagram of the experimental area of the recognition of the present invention;

[0035] Figure 3 is the two-dimensional cloud quality map after data grid of the experimental area of the recognition of the present invention;

[0036] Figure 4 is the two-dimensional cloud density map after data grid of the experimental area of the recognition of the present invention;

[0037] Figure 5 is the two-dimensional diagram of denoising and dimensionality reduction by the principal component analysis method in the experimental area of the recognition of the present invention;

[0038] Figure 6 is the two-dimensional cloud quality map after tensor voting in the experimental area of the recognition of the present invention;

[0039] Figure 7 is the schematic diagram of crack recognition by the optimal circumscribed rectangle in the experimental area of the recognition of the present invention. Detailed Embodiments

[0040] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0041] As Figure 1-7 shown, a method for identifying outcrop cracks based on lidar technology includes the following steps:

[0042] a. Acquisition of three-dimensional point cloud data: Use a lidar system to scan the outcrop of the experimental area to obtain the three-dimensional point cloud data of the outcrop of the experimental area, and then process and splice the collected point cloud data to generate a three-dimensional point cloud model of the outcrop profile.

[0043] b. Data grid processing: Since the original point cloud data is three-dimensional, it is not conducive to observing and extracting parameters of cracks in three-dimensional space. It is necessary to reduce its dimension to obtain two-dimensional data for further processing, that is, to obtain the two-dimensional cloud quality data and two-dimensional cloud density data of the experimental area. The data grid technology is the process of reducing the non-uniformly distributed three-dimensional point cloud data into representative values in a regular grid according to their density and quality. By using colors to distinguish the amount of representative values in each grid, the crack development and fault zones can be displayed.

[0044] Specifically, the data grid processing includes the following steps:

[0045] b01. Set appropriate grid numbers and grid sizes according to the point cloud data;

[0046] b02. Use the XY coordinates of each point to calculate and determine the position of each point in the grid in the point cloud data; that is, the points with the same XY coordinates in the point cloud data will be placed in the same grid. The processing function of this step is:

[0047] i = (X i - X min ) / dx + 1;

[0048] j = (Y i - Y min ) / dy + 1;

[0049] Where: i represents the abscissa of the grid, j represents the ordinate of the grid, X i (i = 1, 2, 3...) represents the abscissa of each point, Y i (i = 1, 2, 3...) represents the ordinate of each point, X min represents the minimum value of X; Y min represents the minimum value of Y, dx is the width of the grid, and dy is the height of the grid;

[0050] Whenever a point is placed in the grid, the density of the grid increases by a value. The outcrop model in the field is complex, and surface fitting will cover up many crack details. Determining the position of each point by grid can fully display the crack development.

[0051] b03. Perform blank pixel detection based on Z difference. Since the depth of each point in the point cloud is different, that is, the Z value is different. In the previous step, the grid data distribution based on XY coordinates has been obtained, and the Z values of the points in each grid are different. Subtracting the minimum value of the Z value from the maximum value in the grid can obtain the two-dimensional cloud quality distribution, display the two-dimensional cloud quality distribution map, as Figure 3 shown, and at the same time export the two-dimensional cloud quality data;

[0052] b04. Retrieve blank pixels from the grid data of the XY coordinates. This step is to remove the Z value and retrieve blank pixels from the grid data distribution of the XY coordinates obtained in step b02 on the XY plane. Since the number of points in each grid is different, the two-dimensional cloud density distribution can be obtained, as Figure 4 shown.

[0053] c. Tensor voting: Perform tensor voting on the obtained two-dimensional cloud quality data. On the one hand, during the acquisition process, due to the missing scans of the lidar system, cracks will be discontinuous and there will be a lot of noise. It is necessary to fill in the missing information of the cracks and remove a large amount of background noise. On the other hand, since the point cloud data is discrete points, the data obtained after grid processing is also discrete points, resulting in discontinuous crack information in the display. It is necessary to convert the discrete crack data into continuous crack information;

[0054] Tensor voting is a new method with strong robustness for extracting prominent features of images. Its feature is to use the method of tensor superposition to enhance the features to be extracted, and use the eigenvalue decomposition of the tensor matrix to explain the voting results. It can infer the implicit structural features from the point cloud information with strong noise and outliers;

[0055] Specifically, the tensor voting includes the following processing sub-steps:

[0056] c01. Perform preliminary denoising on the two-dimensional data, that is, set a threshold to filter out some obvious noises. The two-dimensional data includes two-dimensional cloud quality data and two-dimensional cloud density;

[0057] c02. Perform tensor voting on the filtered two-dimensional data. Let I(x,y) represent the two-dimensional data, then represents the gradient vector of each pixel point, and a tensor model is constructed using the gradient vector;

[0058]

[0059] As can be seen from the second-order matrix, any second-order symmetric non-negative definite tensor can be decomposed into:

[0060]

[0061] Where:

[0062] λ i (i = 1, 2) represents the eigenvalue;

[0063] e i (i = 1, 2) represents the eigenvector;

[0064] c03. Then perform eigenvalue analysis, using the main eigenvalue λ1 as the input data for the voting process to improve the accuracy of tensor information through voting; then perform the superposition of the voting domain and define the attenuation function:

[0065]

[0066] Where:

[0067] represents the arc length;

[0068] represents the arc curvature;

[0069] controls the attenuation degree of the curvature;

[0070] σ is the voting scale factor, which determines the size of the voting domain;

[0071] c04. Finally, perform tensor decomposition and eigenanalysis, place the center of the voting domain at the voting point, accumulate the neighborhood votes for each point, calculate the quantity and size of the votes obtained for each pixel, and finally form a new tensor at each pixel point. The expression is as follows:

[0072]

[0073] Where:

[0074] V represents the accumulated votes;

[0075] V s (p) is the received ballot tensor;

[0076] K represents the number of voters within the receiver's neighborhood;

[0077] The tensor after voting can be decomposed into:

[0078]

[0079] Where:

[0080] is the rod tensor;

[0081] λ1 - λ2 represents the significance size of the rod-shaped tensor;

[0082] is the spherical tensor;

[0083] λ2 represents the significance size of the circular tensor;

[0084] If λ1-λ2>λ2, it means that the dominant tensor is a stick tensor, and the point is most likely located on the curve with e1 as the normal; if λ1≈λ2>0, it means that the point is likely to be located at the intersection of multiple curves, or the point is located in a region and all directions are equally likely; when λ1 and λ2 are both extremely small, it means that the point may be an outlier.

[0085] σ is an important parameter in the voting process. When the value of σ is small, the voting domain is small, the smoothness of the curve is weak, and the local information is highlighted. When σ is increased, the voting domain is increased and the smoothness of the curve is increased. After the test, the voting effect is best when σ = 5. After tensor voting, the direction and development trend of the cracks are more clearly displayed, while maintaining the image details, suppressing the background noise of a large area, improving the image resolution, and ensuring the integrity and coherence of the fracture information. The characteristics of the fracture direction change and the connection mode become clearer; finally, the two-dimensional cloud quality after tensor voting is obtained, such as Figure 6 shown.

[0086] d. Optimal bounding rectangle processing: The optimal bounding rectangle algorithm is used to identify cracks in the two-dimensional cloud quality map after tensor voting obtained in step c. The optimal bounding rectangle algorithm is to obtain the minimum bounding rectangle of the connected domain; the minimum bounding rectangle of each crack is obtained, and the length, width, inclination and other information of the rectangle are converted to obtain the length, width, inclination and other information of the crack.

[0087] Specifically, the optimal bounding rectangle processing includes the following sub-steps:

[0088] d01. Convert the two-dimensional cloud quality map after tensor voting into a binary map. In the process of converting into a binary map, the threshold should be adjusted. The Gaussian value greater than the threshold is 1, and the Gaussian value less than the threshold is 0. Select a suitable threshold so that the image retains more crack detail information while removing as much background noise as possible;

[0089] d02. Perform the optimal bounding rectangle algorithm to identify the binary image and obtain the optimal bounding rectangle of each crack, such as Figure 7 As shown;

[0090] d03. Screen rectangles by area, screen and remove rectangles with too large or too small areas, further ensure the accuracy of crack identification, and count the rectangular attribute parameters of the remaining rectangles, wherein the rectangular attribute parameters include at least one of the number, length, width, inclination or area of ​​rectangles, and the number of rectangles is the number of cracks;

[0091] d04. Convert the rectangular attribute parameters into crack attribute parameters, that is, convert the length, width and inclination of the rectangle into the length, width and inclination of the crack to obtain the attribute parameters of the crack. The obtained crack rectangle parameters are stored for further analysis and processing.

[0092] It should be noted that, for each of the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should understand that this application is not limited by the described action sequence, because according to this application, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and units involved are not necessarily essential to this application.

Claims

1. A method for identifying outcrop fractures based on lidar technology, characterized in that: It includes the following steps: Acquisition of three-dimensional point cloud data: Scanning the outcrop of the experimental area with a lidar system to obtain the three-dimensional point cloud data of the outcrop of the experimental area; Data gridding processing: Performing data gridding processing on the processed point cloud data to obtain the two-dimensional cloud quality data and two-dimensional cloud density data of the experimental area; Tensor voting: Performing tensor voting on the obtained two-dimensional cloud quality data to remove the background noise of the image; Optimal circumscribed rectangle processing: Identifying the optimal circumscribed rectangle of each crack on the denoised image to obtain the crack attribute parameters; Among them, the data gridding processing includes the following sub-steps: Setting appropriate grid numbers and grid sizes according to the point cloud data; Determining the position of each point in the point cloud data in the grid by calculation using the XY coordinates of each point; Performing blank pixel detection based on Z difference, subtracting the minimum value from the maximum value of the Z value in the grid to obtain the two-dimensional cloud quality distribution, and at the same time exporting the two-dimensional cloud quality data; Retrieving blank pixels from the grid data of the XY coordinates to obtain the two-dimensional cloud density data; Points with the same XY coordinates in the point cloud data will be placed in the same grid, and the processing function of this step is: i = (X i - X min ) / dx + 1; j = (Y i - Y min ) / dy + 1; Where: i represents the abscissa of the grid, j represents the ordinate of the grid, X i (i = 1, 2, 3…) represents the abscissa of each point, Y i (i = 1, 2, 3…) represents the ordinate of each point, X min represents the minimum value of X; Y mi n represents the minimum value of Y, dx is the width of the grid, and dy is the height of the grid; Among them, the tensor voting includes the following processing sub-steps: Performing preliminary denoising on the two-dimensional data, where the two-dimensional data includes two-dimensional cloud quality data and two-dimensional cloud density data; Perform tensor voting on the filtered two-dimensional data. Let I(x, y) represent the two-dimensional data, then represents the gradient vector of each pixel point, and a tensor model is constructed using the gradient vector; Performing eigenvalue analysis, using the main eigenvalue λ1 as the input data for the voting process, and then performing the superposition of the voting domains; Performing tensor decomposition and eigenanalysis, placing the center of the voting domain on the voting point, accumulating the votes of each point for the neighborhood voting, calculating the number and size of the votes obtained by each pixel, and finally forming a new tensor at each pixel point.

2. The outcrop fracture identification method based on lidar technology according to claim 1, wherein: The optimal circumscribed rectangle processing includes the following sub-steps: Converting the two-dimensional cloud quality map after tensor voting into a binary map, and removing the background noise during the conversion process; Performing the optimal circumscribed rectangle algorithm recognition on the binary map to obtain the optimal circumscribed rectangle of each crack; Screening the rectangles with area, screening and removing the rectangles with too large or too small area, and counting the rectangle attribute parameters of the remaining rectangles; Converting the rectangle attribute parameters into crack attribute parameters.

3. The outcrop fracture identification method based on lidar technology according to claim 2, wherein: The rectangle attribute parameters include at least one of the number, length, width, inclination angle or area of the rectangle.

4. The outcrop fracture identification method based on lidar technology according to claim 2, wherein: The crack attribute parameters include at least one of the length information, width information or inclination angle information.

Images