Intelligent recognition method for rock mass structural surface occurrence of drilling image

Abstract

The invention discloses a drilling image rock mass structural surface occurrence intelligent identification method. The method comprises the following steps: detecting a rock mass structural plane of a drilling image based on a sliding window dynamic adjustment method and a signal characteristic value analysis method, and extracting a rock mass structural plane area image; a scale selection module and an adaptive weight mechanism are introduced into a bidirectional cascade network to construct a rock mass structural surface multi-scale edge detection model so as to obtain a rock mass structural surface edge extraction result; and obtaining edge pixel coordinates of the rock mass structural plane based on the edge extraction result of the rock mass structural plane, fitting the edge pixel coordinates of the rock mass structural plane by using a curve fitting method to obtain a fitting result, and inverting the fitting result by using a three-dimensional inversion method to intelligently identify the occurrence of the rock mass structural plane. According to the method, for the rock mass structural surface of the drilling image, the occurrence identification precision and the automation degree can be improved, and the problems of identification errors, calculation efficiency and data dependence of an existing method in a complex drilling environment are solved.

Description

Technical Field

[0001] The present invention relates to the field of geological engineering and computer vision technology, and in particular to a method for intelligently identifying the occurrence of rock mass structural surfaces in borehole images. Background Art

[0002] The rock mass structural surface is a key factor in determining the engineering geological characteristics of the rock mass. The accurate acquisition of its occurrence parameters (dip, inclination) is crucial for the stability assessment of underground engineering.

[0003] Although some progress has been made in the existing methods for obtaining rock mass structural surface attitude parameters, significant challenges still exist: traditional methods rely on manual operation, such as the line survey method and the window survey method, which are easily affected by subjective factors; although borehole imaging technology has improved accuracy, it is difficult to meet the rapid data processing needs of large-scale projects; traditional methods perform poorly in complex geological environments, such as noise interference and illumination changes; automated methods such as Hough transform have high computational complexity and are sensitive to noise; deep learning methods rely on large-scale data sets and high-performance computing resources, which limits their application in actual engineering.

[0004] Therefore, there is an urgent need in this field for an intelligent identification method for the occurrence of rock mass structural surfaces in borehole images that can overcome the above-mentioned defects. Summary of the Invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for intelligently identifying the occurrence of rock mass structural surfaces in borehole images.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A method for intelligently identifying the occurrence of rock mass structural surfaces in borehole images comprises the following steps:

[0008] S1. Detect rock mass structural surfaces in borehole images based on a sliding window dynamic adjustment method and a signal eigenvalue analysis method, and extract rock mass structural surface area images;

[0009] S2. A multi-scale edge detection model for rock mass structural surfaces is constructed by introducing a scale selection module and an adaptive weight mechanism into a bidirectional cascade network, the multi-scale edge detection model for rock mass structural surfaces is trained, and the trained multi-scale edge detection model for rock mass structural surfaces is used to extract edges from images of rock mass structural surfaces to obtain edge extraction results for rock mass structural surfaces.

[0010] S3. Obtain the pixel coordinates of the edge of the rock structure surface based on the rock structure surface edge extraction result, use the curve fitting method to fit the pixel coordinates of the edge of the rock structure surface to obtain the fitting result, and use the three-dimensional inversion method to invert the fitting result to perform intelligent identification of the rock structure surface attitude.

[0011] Furthermore, step S1 includes the following steps:

[0012] S11, setting the sliding window width to be consistent with the drilling image width based on the sliding window dynamic adjustment method, and setting the sliding window height to a dynamic adjustment mechanism;

[0013] S12, calculating the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window, and calculating the characteristic value of the rock mass structural surface positioning signal based on the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window;

[0014] S13. Draw a rock mass structural surface positioning signal characteristic value curve based on the rock mass structural surface positioning signal characteristic value curve, use a Savitzky-Golay filter to smooth the rock mass structural surface positioning signal characteristic value curve, and detect the smoothed rock mass structural surface positioning signal characteristic value curve based on the signal characteristic value analysis method to extract the rock mass structural surface area.

[0015] Further, in step S11, the dynamic adjustment mechanism is specifically as follows: setting the first window height, the second window height and the third window height in descending order of size, and setting the first preset threshold and the second preset threshold in descending order of size, calculating the grayscale variance in the sliding window, and judging whether the grayscale variance in the sliding window is greater than the first preset threshold. If so, the sliding window height is set to the first window height; otherwise, continue to judge whether the grayscale variance in the sliding window is less than the second preset threshold. If so, the window height is set to the third window height; otherwise, the window height is set to the second window height.

[0016] Furthermore, in step S12, the characteristic value of the rock mass structural surface positioning signal is calculated based on the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window, which is expressed as:

[0017] T=M-α×N+γ×V

[0018] Where: T is the characteristic value of the rock structure surface positioning signal, M is the contrast parameter in the sliding window, α is the weight parameter of the grayscale difference between the first and last rows, N is the grayscale difference between the first and last rows in the sliding window, γ is the weight parameter of the grayscale variance, and V is the grayscale variance in the sliding window.

[0019] Furthermore, the specific process of calculating the contrast parameter within the sliding window is as follows: extract the grayscale average value of each column of pixels in the sliding window; then calculate the square of the difference between the maximum grayscale value and the minimum grayscale value of each column of pixels in the sliding window, select the larger value between the grayscale average value and the square of the difference as the eigenvalue of the column of pixels, and sum the eigenvalues of each column in the sliding window in turn to obtain the contrast parameter within the sliding window.

[0020] Furthermore, the specific process of calculating the grayscale difference between the first and last rows in the sliding window is: calculating the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window, and calculating the absolute difference between the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window to obtain the grayscale difference between the first and last rows in the sliding window.

[0021] Furthermore, in step S2, a multi-scale edge detection model of rock structure surface is constructed by introducing a scale selection module and an adaptive weight mechanism into the bidirectional cascade network, specifically: the bidirectional cascade network includes multiple incremental detection blocks of different scales connected layer by layer through a bidirectional cascade structure, the incremental detection block includes an input convolution module, a scale enhancement module and an output convolution module, and the scale selection module and the adaptive weight mechanism are sequentially connected after the input convolution module in the incremental detection block and between different incremental detection blocks; the scale selection module is used to select and extract shallow features in the feature maps of different incremental detection blocks; the adaptive weight mechanism is used to dynamically weight the scale features according to the importance of the scale features in different incremental detection blocks.

[0022] Furthermore, in step S2, the multi-scale edge detection model of the rock mass structural surface is trained. The specific process is as follows:

[0023] Get the training set sample data, expressed as (X, Y), where X is the input rock mass structural surface area image and Y is the corresponding edge annotation map; decompose Y into S edge annotation maps of different scales:

[0024]

[0025] Where: Y s is the edge annotation map at the s-th scale;

[0026] The rock mass structural surface regional image is input into the rock mass structural surface multi-scale edge detection model, and the edge prediction results of different scales are generated using incremental detection blocks of different scales;

[0027] Two complementary supervisions are used during training to determine Y s , these two types of supervision are determined at scale s as:

[0028]

[0029] in: is the edge annotation map from shallow to deep at the s-th scale, is the edge annotation map from deep to shallow at the s-th scale, is the edge prediction result from shallow to deep layer generated at the i-th scale, is the edge prediction result from deep to shallow layer generated at the i-th scale;

[0030] The edge prediction results from the shallow to deep layer generated at the s-th scale And the deep to shallow edge prediction results generated at the sth scale Determined as and The edges of different scales are suppressed, and the combination of the two supervisions is determined as Y s :

[0031]

[0032] The loss function during training is:

[0033]

[0034] Where: L s is the loss function at the sth scale, L(·) is the loss function between the edge prediction result and the true edge annotation map;

[0035] The adaptive weight mechanism is used to assign learnable weight coefficients to features of different scales, and the edge prediction result is:

[0036] p s =w s ⊙f s

[0037] Where: p s is the edge prediction result at the sth scale, w s is the weight parameter learned by the adaptive weight mechanism at the sth scale, ⊙ is the element-by-element multiplication symbol, f s is the input rock mass structural surface area image of the sth scale;

[0038] The weight coefficients that can be learned during training are dynamically adjusted through back propagation together with other parameters of the multi-scale edge detection model of rock structure surfaces to minimize the overall loss function:

[0039]

[0040] Where: L total is the overall loss function of the multi-scale edge detection model for rock mass structural surfaces.

[0041] Furthermore, in step S3, the fitting results are inverted using a three-dimensional inversion method to perform intelligent identification of the rock mass structural plane occurrence, including the following steps:

[0042] Step A1: Projecting the rock mass structural surface onto a horizontal coordinate system to obtain a two-dimensional rock mass structural surface, and determining the perimeter and radius of the two-dimensional rock mass structural surface based on the fitting results;

[0043] Step A2, calculating the central angle of each point on the two-dimensional rock mass structural surface;

[0044] Step A3: Based on the radius of the two-dimensional rock mass structural surface and the central angle of each point on the two-dimensional rock mass structural surface, the fitting result is converted into a three-dimensional coordinate system using a three-dimensional inversion method to obtain the three-dimensional rock mass structural surface;

[0045] Step A4: performing principal component analysis on the point coordinates of the three-dimensional rock mass structural surface to determine the first three components of the rock mass structural surface characteristic vector;

[0046] Step A5: Calculate the inclination and dip angle based on the first three components of the rock mass structural surface eigenvector.

[0047] The present invention has the following beneficial effects:

[0048] (1) The present invention detects the rock mass structural surface of the borehole image based on the sliding window dynamic adjustment method and the signal characteristic value analysis method, extracts the rock mass structural surface area image, suppresses noise interference, and improves the recognizability of edge features;

[0049] (2) The present invention introduces a scale selection module and an adaptive weight mechanism into a bidirectional cascade network to construct a multi-scale edge detection model for rock mass structural surfaces. This model can adaptively adjust the contribution weights of features at different scales, thereby improving the accuracy and robustness of structural surface recognition.

[0050] (3) The present invention can realize intelligent identification of geometric parameters of rock mass structural surfaces by combining curve fitting and three-dimensional inversion technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of an intelligent identification method for rock mass structural surface occurrence from borehole images;

[0052] Figure 2 is the drilling image and drilling grayscale image;

[0053] Figure 3 Schematic diagram of the rock mass structural surface area for the original and smoothed T value curves;

[0054] Figure 4 This is a schematic diagram of the structure of the multi-scale edge detection model for rock mass structural surfaces;

[0055] Figure 5 Schematic diagram of the rock mass structural surface edge extraction results of the original bidirectional cascade network;

[0056] Figure 6 This is a schematic diagram of the rock mass structural surface edge extraction results of the rock mass structural surface multi-scale edge detection model;

[0057] Figure 7 It is the fitting curve diagram of the edge of the rock mass structural surface;

[0058] Figure 8 Projecting the rock mass structural surface onto the horizontal coordinate system to obtain a two-dimensional rock mass structural surface process diagram;

[0059] Figure 9 Schematic diagram of the coordinates of each point in the fitting result in the three-dimensional coordinate system;

[0060] Figure 10 Schematic diagram of the three-dimensional rock structure surface obtained by inversion fitting of the rock structure surface. DETAILED DESCRIPTION

[0061] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0062] like Figure 1 As shown, a method for intelligently identifying the occurrence of rock mass structural surfaces in borehole images is characterized by comprising steps S1-S3, specifically as follows:

[0063] S1. Based on the sliding window dynamic adjustment method and signal eigenvalue analysis method, the rock structure surface of the drilling image is detected and the rock structure surface area image is extracted.

[0064] In an optional embodiment of the present invention, step S1 includes the following steps:

[0065] S11, setting the sliding window width to be consistent with the drilling image width based on the sliding window dynamic adjustment method, and setting the sliding window height to a dynamic adjustment mechanism;

[0066] The dynamic adjustment mechanism is specifically as follows: setting the first window height, the second window height and the third window height in descending order of size, and setting the first preset threshold and the second preset threshold in descending order of size, calculating the grayscale variance in the sliding window, and judging whether the grayscale variance in the sliding window is greater than the first preset threshold. If so, the sliding window height is set to the first window height; otherwise, continue to judge whether the grayscale variance in the sliding window is less than the second preset threshold. If so, the window height is set to the third window height; otherwise, the window height is set to the second window height.

[0067] S12. Calculate the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window, and calculate the characteristic value of the rock mass structural surface positioning signal based on the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window.

[0068] like Figure 2 As shown, the present invention first Figure 2 The drill hole image on the left was grayscaled to obtain Figure 2 The grayscale image of the borehole on the right. Then, the present invention calculates the characteristic value of the rock structure surface positioning signal based on the grayscale variance, the grayscale difference between the first and last rows and the contrast parameter in the sliding window, which is expressed as:

[0069] T=M-α×N+γ×V

[0070] Where: T is the characteristic value of the rock structure surface positioning signal, M is the contrast parameter in the sliding window, α is the weight parameter of the grayscale difference between the first and last rows, N is the grayscale difference between the first and last rows in the sliding window, γ is the weight parameter of the grayscale variance, and V is the grayscale variance in the sliding window.

[0071] The specific process of calculating the contrast parameter in the sliding window of the present invention is as follows: extracting the grayscale average value of each column of pixels in the sliding window; then calculating the square of the difference between the maximum grayscale value and the minimum grayscale value of each column of pixels in the sliding window, selecting the larger value of the grayscale average value and the square of the difference as the eigenvalue of the pixel column, and summing the eigenvalues of each column in the sliding window in turn to obtain the contrast parameter in the sliding window.

[0072] The specific process of the present invention for calculating the grayscale difference between the first and last rows in the sliding window is as follows: calculating the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window, and calculating the absolute difference between the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window to obtain the grayscale difference between the first and last rows in the sliding window.

[0073] S13. Draw a rock mass structural surface positioning signal characteristic value curve based on the rock mass structural surface positioning signal characteristic value curve, use a Savitzky-Golay filter to smooth the rock mass structural surface positioning signal characteristic value curve, and detect the smoothed rock mass structural surface positioning signal characteristic value curve based on the signal characteristic value analysis method to extract the rock mass structural surface area.

[0074] The present invention detects the characteristic value curve of the smoothed rock structure surface positioning signal based on the signal characteristic value analysis method to extract the rock structure surface area. Specifically, based on the signal characteristic value analysis method, the significant peaks of the characteristic value curve of the smoothed rock structure surface positioning signal are detected by a local peak detection algorithm, and the pseudo peaks caused by noise interference are eliminated in combination with the grayscale variance in the sliding window to extract the rock structure surface area.

[0075] The present invention also incorporates a fallback mechanism. Specifically, if the rock mass structural surface area extracted by the sliding window during the initial scan does not meet the expected characteristics, the first and second preset thresholds are readjusted, and a second scan is performed to optimize the detection results. This fallback mechanism ensures that the sliding window can flexibly adapt to complex structural surface variations under different image backgrounds, thereby improving overall detection accuracy.

[0076] After the sliding window is scanned and the characteristic value T is calculated, several local peaks and valleys usually appear in the curve of the present invention. Local peaks are generally regarded as signs of potential structural surfaces, because the T values at these positions are relatively high, reflecting significant grayscale contrast and local change characteristics within the window. However, in actual applications, some peaks are not real structural surfaces, but local grayscale anomalies caused by interference factors such as uneven illumination, image noise or complex texture background. Although they appear as peaks on the curve, they lack the typical continuity and significant characteristics of the structural surface. Therefore, they need to be further identified by local variance. Relatively speaking, valley values usually correspond to areas with smaller grayscale changes, indicating a relatively uniform background. They usually do not represent structural surface areas, but they are not absolute. For example, when the overall brightness is low or the lighting conditions are poor, the structural surface features may be weakened, resulting in a trough on the T value curve.

[0077] The present invention uses the characteristic signal value T to Figure 2 The grayscale image of the drill hole on the right is used for structural surface detection, and the sliding window height range is set to 120 to 130 pixels. Figure 3 The relationship between the number of pixel rows within a sliding window and the T value is shown. The present invention observed the following: ① When the sliding window does not cover the structural surface, the T value remains low; ② As the structural surface gradually enters the sliding window, the T value shows an upward trend; ③ When the structural surface is completely covered by the sliding window, the T value reaches a local maximum and fluctuates within a small range; ④ As the structural surface gradually exits the sliding window, the T value decreases. Phenomenon ③ indicates that the sliding window can cover multiple areas containing structural surface features, resulting in similar T values in these areas, increasing the difficulty of positioning. However, after using a Savitzky-Golay filter to smooth the characteristic value curve of the volumetric structural surface positioning signal, the present invention uses the smoothed red curve to show two distinct peaks, corresponding to pixel rows 83 and 560, respectively. This allows the location of the rock mass structural surface to be inferred. In contrast, the blue curve, the curve before smoothing, exhibits less peak fluctuation, corresponding to pixel rows 84 and 550, respectively. The rock mass structural surface region determined using this method is less accurate than the smoothed curve. Through this method, the present invention can improve the regional positioning effect of rock structure surface, thereby effectively supporting the subsequent rock structure surface edge extraction.

[0078] S2. By introducing a scale selection module and an adaptive weight mechanism into a bidirectional cascade network to construct a multi-scale edge detection model for rock structural surfaces, the multi-scale edge detection model for rock structural surfaces is trained, and the trained multi-scale edge detection model for rock structural surfaces is used to extract edges from the rock structural surface area image to obtain the rock structural surface edge extraction results.

[0079] In an optional embodiment of the present invention, the present invention constructs a multi-scale edge detection model for rock structural surfaces by introducing a scale selection module and an adaptive weight mechanism into a bidirectional cascade network. Specifically, the bidirectional cascade network includes multiple incremental detection blocks of different scales connected layer by layer through a bidirectional cascade structure. The incremental detection block includes an input convolution module, a scale enhancement module and an output convolution module. The scale selection module and the adaptive weight mechanism are connected in sequence after the input convolution module in the incremental detection block and between different incremental detection blocks. The scale selection module is used to select and extract shallow features in the feature maps of different incremental detection blocks. The adaptive weight mechanism is used to dynamically weight the scale features according to the importance of the scale features in different incremental detection blocks.

[0080] like Figure 4As shown, the multi-scale edge detection model of rock structure surface of the present invention includes five incremental detection blocks, namely ID Block1, IDBlock2, ID Block3, ID Block4 and ID Block5. In the first incremental detection block IDBlock1, the scale selection module extracts key features from the shallow layer, and the adaptive weight mechanism weights the scale features to highlight the most useful edge information. In the second incremental detection block ID Block2, the scale selection module continues to extract shallow features from feature maps of different scales, and the adaptive weight mechanism further adjusts the response of the features according to the feature importance of the second layer, so that the deep features can focus on the most important information. In the second incremental detection block ID Block3, as the network depth increases, the scale selection module continues to extract and fuse more shallow features, and the adaptive weight mechanism weights the feature maps of each scale, thereby strengthening the extraction of deep edge information. In the fourth incremental detection block, ID Block4, the scale selection module and the adaptive weight mechanism work together to enable the multi-scale edge detection model of rock structure surfaces to utilize feature maps with a larger receptive field, further optimizing edge detection capabilities. The adaptive weight mechanism weights features at different scales to ensure that the multi-scale edge detection model of rock structure surfaces can still highlight important features in complex backgrounds. In the fifth incremental block, ID Block5, the role of the scale selection module and the adaptive weight mechanism are further optimized. The scale selection module ensures that important features can be extracted from feature maps at different scales, while the adaptive weight mechanism ensures that the edge features ultimately output by the multi-scale edge detection model of rock structure surfaces are the most accurate, optimizing the performance of the entire multi-scale edge detection model of rock structure surfaces.

[0081] The present invention trains a multi-scale edge detection model for rock mass structural surfaces. The specific process is as follows:

[0082] The present invention obtains training set sample data, expressed as (X, Y), where X is the input rock mass structural surface area image and Y is the corresponding edge annotation map; Y is decomposed into S edge annotation maps of different scales:

[0083]

[0084] Where: Y s is the edge annotation map at the s-th scale.

[0085] The present invention inputs the rock mass structural surface regional image into the rock mass structural surface multi-scale edge detection model, and uses incremental detection blocks of different scales to generate edge prediction results of different scales.

[0086] The present invention uses two complementary supervisions to determine Y s , these two types of supervision are determined at scale s as:

[0087]

[0088] in: is the edge annotation map from shallow to deep at the s-th scale, is the edge annotation map from deep to shallow at the s-th scale, is the edge prediction result from shallow to deep layer generated at the i-th scale, is the edge prediction result from deep to shallow layer generated at the i-th scale.

[0089] The present invention generates the edge prediction results from the shallow layer to the deep layer at the sth scale. And the deep to shallow edge prediction results generated at the sth scale Determined as and The edges of different scales are suppressed, and the combination of the two supervisions is determined as Y s :

[0090]

[0091] The loss function during training is:

[0092]

[0093] Where: L s is the loss function at the sth scale, and L(·) is the loss function between the edge prediction result and the true edge annotation map.

[0094] The adaptive weight mechanism is used to assign learnable weight coefficients to features of different scales, and the edge prediction result is:

[0095] p s =w s ⊙f s

[0096] Where: p s is the edge prediction result at the sth scale, w s is the weight parameter learned by the adaptive weight mechanism at the sth scale, ⊙ is the element-by-element multiplication symbol, f s is the input rock mass structural surface area image at the sth scale.

[0097] The weight coefficients that can be learned during training are dynamically adjusted through back propagation together with other parameters of the multi-scale edge detection model of rock mass structural surfaces to minimize the overall loss function:

[0098]

[0099] Where: L total is the overall loss function of the multi-scale edge detection model for rock mass structural surfaces.

[0100] During training, since the scales of each incremental detection block are different, different loss gradients can be set to make it possible to train incremental detection blocks of specific scales. The loss gradient is expressed as:

[0101]

[0102] in: is the loss gradient at the s-th scale.

[0103] After the training is completed, the rock mass structural surface area image is input into the trained rock mass structural surface multi-scale edge detection model, and the edge prediction results of different scales are generated by using incremental detection blocks of different scales, and then the edge prediction results of different scales are fused. s2d And edge prediction result p d2s , and processed by a 1×1 convolution layer to finally generate the rock structure surface edge extraction result.

[0104] like Figure 5 As shown in FIG, the present invention provides a schematic diagram of the rock mass structural surface edge extraction result of the original bidirectional cascade network. Figure 6 As shown in the figure, the present invention provides a schematic diagram of the rock surface edge extraction results of the multi-scale edge detection model for rock surface. A comparison shows that the multi-scale edge detection model for rock surface constructed by the present invention performs better under complex background interference, can effectively capture subtle edge changes of rock surface, and ensure the continuity and integrity of the edge.

[0105] S3. Obtain the pixel coordinates of the edge of the rock structure surface based on the rock structure surface edge extraction result, use the curve fitting method to fit the pixel coordinates of the edge of the rock structure surface to obtain the fitting result, and use the three-dimensional inversion method to invert the fitting result to perform intelligent identification of the rock structure surface attitude.

[0106] In an optional embodiment of the present invention, the present invention obtains the edge pixel coordinates of the rock structural surface based on the rock structural surface edge extraction result. The specific process is: based on the rock structural surface edge extraction result, the threshold is used to extract the framework of the rock structural surface to extract the crack skeleton of the rock structural surface, and then the crack skeleton of the rock structural surface is morphologically refined to obtain the edge pixel coordinates.

[0107] The present invention then uses a curve fitting method to fit the edge pixel coordinates to obtain the fitting results, such as Figure 7 As shown, the fitted model includes:

[0108] Model A: y = a1sin(b1x+c1)+a2sin(b1x+c2);

[0109] Model B: y=a1sin(b1x+c1)+a2sin(b2x+c1)+a3sin(b3x+c3);

[0110] Model C: y = p1x 4 +p1x 3 +p3x 2 +p4x+p5;

[0111] Model D: y = p1x 5 +p2x 4 +p3x 3 +p4x 2 +p5x+p6

[0112] Where: (x, y) is the edge pixel coordinates, a1, b1, c1, a2, b2, c2, a3, b3, c3, p1, p2, p3, p4, p5, and p6 are fitting parameters.

[0113] The present invention compares the fitting effects of different models, and the results show that the matching degree of model C is better, that is, model C can represent Figure 3 Feature information of structural surface 1 extracted from the borehole image.

[0114] The present invention uses a three-dimensional inversion method to invert the fitting results to perform intelligent identification of the rock mass structural surface occurrence, including the following steps:

[0115] Step A1: Project the rock mass structural surface onto the horizontal coordinate system to obtain a two-dimensional rock mass structural surface. Based on the fitting results, determine the perimeter and radius of the two-dimensional rock mass structural surface. The specific process is as follows: Figure 8 As shown in the figure, the rock mass structural surface is projected onto the horizontal coordinate system to obtain the two-dimensional rock mass structural surface, and the starting coordinates (x0, y0) and the end coordinates (x e ,y e ), calculate the starting coordinates (x0, y0) and the end coordinates (x e ,y e ) and determine it as the perimeter A of the two-dimensional rock mass structural surface, expressed as:

[0116] A=x e -x0

[0117] The radius r is calculated by inversely calculating the perimeter of the two-dimensional rock mass structural surface, which is expressed as:

[0118]

[0119] Step A2: Calculate the center angle of each point on the two-dimensional rock mass structure surface. The specific process is: determine the coordinates (xi ,y i ), the horizontal coordinate x of each point i Subtract the starting point's abscissa x0 to get the arc length s i , calculate the central angle θ of each point on the two-dimensional rock mass structural surface according to the arc length formula i , expressed as:

[0120]

[0121] Step A3: Based on the radius of the two-dimensional rock mass structural surface and the center angle of each point on the two-dimensional rock mass structural surface, the fitting result is converted into a three-dimensional coordinate system using a three-dimensional inversion method to obtain the three-dimensional rock mass structural surface. The specific process is as follows: based on the radius of the two-dimensional rock mass structural surface and the center angle of each point on the two-dimensional rock mass structural surface; Figure 9 As shown, determine the coordinates (x i ′,y i ′,z i );like Figure 10 As shown, the three-dimensional inversion method is used to convert the coordinates (x i ′,y i ′,z i ) to perform inversion fitting to obtain the three-dimensional rock mass structural surface, where the y-axis in the two-dimensional rock mass structural surface is the z-axis in the three-dimensional coordinate system, which is expressed as:

[0122] x i ′=r·cosθ i

[0123] y i ′=r·sinθ i

[0124] z i =y i

[0125] Step A4: Perform principal component analysis on the point coordinates of the three-dimensional rock mass structural surface to determine the first three components of the rock mass structural surface eigenvector. The specific process is: perform principal component analysis on the point coordinates of the three-dimensional rock mass structural surface, calculate the eigenvalues and corresponding eigenvectors, obtain the eigenvectors corresponding to the three minimum eigenvalues, and determine them as the first three components Δx, Δy and Δz of the rock mass structural surface eigenvector.

[0126] Step A5: Calculate the inclination θ and the dip angle α based on the first three components of the rock mass structural surface characteristic vector. The specific process is as follows:

[0127] If the component Δy of the rock mass structural plane characteristic vector is 0 and the component Δx of the rock mass structural plane characteristic vector is 0, the inclination is 0 degrees. Otherwise, the sine and cosine values are calculated and expressed as:

[0128]

[0129] The present invention calculates the tendency θ based on the sine value and the cosine value, which is expressed as:

[0130]

[0131] Among them: if Δy≥0 and Δx≥0, then k=0; if Δy≥0 and Δx<0 or Δy<0 and Δx<0, then k=1; if Δy<0 and Δx≥0, then k=2.

[0132] Calculate the inclination angle, expressed as:

[0133]

[0134] Where: Δx, Δy and Δz represent the first three components of the eigenvector respectively, and ∈ is a very small value used to avoid division by zero.

[0135] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0136] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0138] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0139] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for intelligently identifying the occurrence of rock mass structural surfaces in borehole images, characterized in that: The following steps are involved: S1. Detect rock mass structural surfaces in borehole images based on a sliding window dynamic adjustment method and a signal eigenvalue analysis method, and extract rock mass structural surface area images; S2. A multi-scale edge detection model for rock mass structural surfaces is constructed by introducing a scale selection module and an adaptive weight mechanism into a bidirectional cascade network, the multi-scale edge detection model for rock mass structural surfaces is trained, and the trained multi-scale edge detection model for rock mass structural surfaces is used to extract edges from images of rock mass structural surfaces to obtain edge extraction results for rock mass structural surfaces. S3. Obtain the pixel coordinates of the edge of the rock structure surface based on the rock structure surface edge extraction result, use the curve fitting method to fit the pixel coordinates of the edge of the rock structure surface to obtain the fitting result, and use the three-dimensional inversion method to invert the fitting result to perform intelligent identification of the rock structure surface attitude.

2. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 1, characterized in that: Step S1 includes the following steps: S11, setting the sliding window width to be consistent with the drilling image width based on the sliding window dynamic adjustment method, and setting the sliding window height to a dynamic adjustment mechanism; S12, calculating the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window, and calculating the characteristic value of the rock mass structural surface positioning signal based on the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window; S13. Draw a rock mass structural surface positioning signal characteristic value curve based on the rock mass structural surface positioning signal characteristic value curve, use a Savitzky-Golay filter to smooth the rock mass structural surface positioning signal characteristic value curve, and detect the smoothed rock mass structural surface positioning signal characteristic value curve based on the signal characteristic value analysis method to extract the rock mass structural surface area.

3. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 2, characterized in that: In step S11, the dynamic adjustment mechanism is specifically as follows: setting the first window height, the second window height and the third window height in descending order of size, and setting the first preset threshold and the second preset threshold in descending order of size, calculating the grayscale variance in the sliding window, and judging whether the grayscale variance in the sliding window is greater than the first preset threshold. If so, the sliding window height is set to the first window height; otherwise, continue to judge whether the grayscale variance in the sliding window is less than the second preset threshold. If so, the window height is set to the third window height; otherwise, the window height is set to the second window height.

4. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 2, characterized in that: In step S12, the characteristic value of the rock mass structural surface positioning signal is calculated based on the grayscale variance, the grayscale difference between the first and last rows, and the contrast parameter within the sliding window, which is expressed as: T=M-α×N+γ×V Where: T is the characteristic value of the rock structure surface positioning signal, M is the contrast parameter in the sliding window, α is the weight parameter of the grayscale difference between the first and last rows, N is the grayscale difference between the first and last rows in the sliding window, γ is the weight parameter of the grayscale variance, and V is the grayscale variance in the sliding window.

5. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 4, characterized in that: The specific process of calculating the contrast parameter within the sliding window is as follows: extract the grayscale average value of each column of pixels in the sliding window; then calculate the square of the difference between the maximum grayscale value and the minimum grayscale value of each column of pixels in the sliding window, select the larger value of the grayscale average value and the square of the difference as the eigenvalue of the pixel column, and sum the eigenvalues of each column in the sliding window in turn to obtain the contrast parameter within the sliding window.

6. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 4, characterized in that: The specific process of calculating the grayscale difference between the first and last rows in the sliding window is as follows: calculate the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window, and calculate the absolute difference between the grayscale mean of the first row of pixels and the grayscale mean of the last row of pixels in the sliding window to obtain the grayscale difference between the first and last rows in the sliding window.

7. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 1, characterized in that: In step S2, a multi-scale edge detection model for rock mass structural surfaces is constructed by introducing a scale selection module and an adaptive weight mechanism into a bidirectional cascade network. Specifically, the bidirectional cascade network includes multiple incremental detection blocks of different scales connected layer by layer through a bidirectional cascade structure. The incremental detection block includes an input convolution module, a scale enhancement module, and an output convolution module. The scale selection module and the adaptive weight mechanism are sequentially connected after the input convolution module in the incremental detection block and between different incremental detection blocks. The scale selection module is used to select and extract shallow features in the feature maps of different incremental detection blocks. An adaptive weight mechanism is used to dynamically weight scale features according to their importance in different incremental detection blocks.

8. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 1, characterized in that: In step S2, the multi-scale edge detection model of rock mass structural surface is trained. The specific process is as follows: Get the training set sample data, expressed as (X, Y), where X is the input rock mass structural surface area image and Y is the corresponding edge annotation map; decompose Y into S edge annotation maps of different scales: Where: Y s is the edge annotation map at the s-th scale; The rock mass structural surface regional image is input into the rock mass structural surface multi-scale edge detection model, and the edge prediction results of different scales are generated using incremental detection blocks of different scales; Two complementary supervisions are used during training to determine Y s , these two types of supervision are determined at scale s as: in: is the edge annotation map from shallow to deep at the s-th scale, is the edge annotation map from deep to shallow at the s-th scale, is the edge prediction result from shallow to deep layer generated at the i-th scale, is the edge prediction result from deep to shallow layer generated at the i-th scale; The edge prediction results from the shallow to deep layer generated at the s-th scale And the deep to shallow edge prediction results generated at the sth scale Determined as and The edges of different scales are suppressed, and the combination of the two supervisions is determined as Y s : The loss function during training is: Where: L s is the loss function at the sth scale, L(·) is the loss function between the edge prediction result and the true edge annotation map; The adaptive weight mechanism is used to assign learnable weight coefficients to features of different scales, and the edge prediction result is: p s =w s ⊙f s Where: p s is the edge prediction result at the sth scale, w s is the weight parameter learned by the adaptive weight mechanism at the sth scale, ⊙ is the element-by-element multiplication symbol, f s is the input rock mass structural surface area image of the sth scale; The weight coefficients that can be learned during training are dynamically adjusted through back propagation together with other parameters of the multi-scale edge detection model of rock structure surfaces to minimize the overall loss function: Where: L total is the overall loss function of the multi-scale edge detection model for rock mass structural surfaces.

9. The method for intelligently identifying rock mass structural surface occurrence from borehole images according to claim 1, characterized in that: In step S3, the fitting results are inverted using a three-dimensional inversion method to perform intelligent identification of the rock mass structural plane, including the following steps: Step A1: Projecting the rock mass structural surface onto a horizontal coordinate system to obtain a two-dimensional rock mass structural surface, and determining the perimeter and radius of the two-dimensional rock mass structural surface based on the fitting results; Step A2, calculating the central angle of each point on the two-dimensional rock mass structural surface; Step A3: Based on the radius of the two-dimensional rock mass structural surface and the central angle of each point on the two-dimensional rock mass structural surface, the fitting result is converted into a three-dimensional coordinate system using a three-dimensional inversion method to obtain the three-dimensional rock mass structural surface; Step A4: performing principal component analysis on the point coordinates of the three-dimensional rock mass structural surface to determine the first three components of the rock mass structural surface characteristic vector; Step A5: Calculate the inclination and dip angle based on the first three components of the rock mass structural surface eigenvector.

Images