Method for processing electrical digital data based on hybrid expert model

Abstract

The disclosure provides a kind of electric digital data processing method based on mixed expert model, comprising: inputting the image gray of surrounding rock borehole, sharpening pretreatment and entering the mixed expert enhancement model;Task identifier extracts four categories of statistical features of image to screen enhanced task, and task arranger selects continuous front task in fixed data flow order and obtains preliminary fissure binary graph by serial processing;After the completion of regional division by line level self-adaptive fissure saliency evaluation, four-dimensional hough transform is used to fit sinusoidal fissure, four parameters of amplitude, period, phase and center line are solved, and pure fissure binary graph is obtained by combining mask operation.The present application does not need deep learning data set, can inhibit various image artifacts, improve fissure recognition accuracy and parameter extraction stability, and is suitable for surrounding rock stability quantitative evaluation.

Description

Technical Field

[0001] This invention belongs to the field of image processing and fracture recognition technology, and can be applied to borehole image analysis, surrounding rock stability assessment and geological structure feature extraction, etc., and falls under the category of electronic digital data processing technology. Background Technology

[0002] Coal is my country's primary energy source, and its safe and efficient mining is crucial to national energy security and economic development. The fracture structure of the surrounding rock in roadways is a key factor affecting rock mass stability and inducing disasters such as roof falls, water inrushes, and gas outbursts.

[0003] While traditional core sampling methods can obtain information about rock masses, they suffer from drawbacks such as long processing time, high cost, and limited accuracy. Borehole imaging technology can visualize the internal structure of rock strata by scanning the borehole wall, and analyze the images using electro-digital data processing techniques, providing a new technical approach for fracture identification.

[0004] However, existing methods for processing electrical digital data of borehole images in surrounding rock still face challenges such as mud residue, geological texture interference, and human identification errors, resulting in insufficient accuracy and efficiency in fracture identification. At the same time, the lack of relevant datasets makes it impossible to use deep learning models to complete automated data parsing.

[0005] Therefore, it is urgent to develop an automated electro-digital data processing method for fracture extraction based on image processing and intelligent recognition algorithms, so as to achieve accurate identification and quantitative analysis of fractures in borehole images and provide technical support for the evaluation of roadway surrounding rock stability and disaster prevention and control. Summary of the Invention

[0006] To address the shortcomings of existing methods for processing electrical digital data of borehole images in surrounding rock, such as significant interference from noise, texture, and splicing artifacts, high errors in human identification, difficulty in using deep learning due to lack of datasets, insufficient accuracy and efficiency in fracture extraction, and unstable extraction of fracture structure parameters, this disclosure provides an electrical digital data processing method based on a hybrid expert model. This method employs a hybrid expert model-like adaptive image enhancement, row-level adaptive fracture partitioning, and four-dimensional Hough sinusoidal fracture fitting for electrical digital data processing.

[0007] The technical solution disclosed herein is:

[0008] A method for processing electrical digital data based on a hybrid expert model, comprising:

[0009] Acquire surrounding rock borehole images, and perform grayscale processing and sharpening preprocessing on the surrounding rock borehole images;

[0010] The preprocessed borehole image of the surrounding rock is input into a hybrid expert image enhancement model. The hybrid expert image enhancement model includes a task recognizer, a task arranger, and four processing units with fixed data flow dependencies: noise removal, gradient processing, morphological processing, and image stitching removal. The task recognizer extracts four statistical features of image noise, gradient, morphology, and stitching, and filters the image enhancement tasks to form a set of tasks to be scheduled. The task arranger calls the processing units corresponding to the tasks to be scheduled in a fixed order of noise removal, gradient processing, morphological processing, and image stitching removal, and outputs a preliminary binary image of the fracture after processing.

[0011] The initial binary image of the fracture is divided into fracture regions to obtain a continuous and stable image of the fracture region;

[0012] The image of the fracture region is processed by four-dimensional Hough transform to extract four structural parameters corresponding to the fracture: amplitude R, period P, phase β, and centerline position C, and output a pure binary image of the fracture.

[0013] The grayscale processing and sharpening preprocessing of the surrounding rock borehole image includes:

[0014] Based on the red of each pixel in the surrounding rock borehole image ,green ,blue Components, according to the formula The grayscale value corresponding to the pixel is obtained until a grayscale image is obtained.

[0015] Gaussian blur is applied to the grayscale image to obtain a low-frequency image;

[0016] Subtracting the grayscale image from the low-frequency image yields the detail-enhanced image;

[0017] The enhanced detail image is superimposed on the surrounding rock borehole image according to the set weight coefficient to generate a sharpened image, thus completing the preprocessing.

[0018] The unit used to perform white noise removal is the noise suppression expert E1: it performs white top hat operation on the preprocessed surrounding rock borehole image to generate a bright noise set, removes white noise from the image while preserving the complete shape of sinusoidal cracks in the image, and outputs a denoised image.

[0019] The unit used to perform gradient processing is the gradient enhancement and binarization expert E2: the Sobel operator is used to solve the image gradient of the denoised image output by E1, highlighting high gradient regions as foreground and suppressing low gradient noise regions as background, to obtain a gradient enhancement image, which improves the saliency and robustness of the crack edge structure.

[0020] The unit used to perform morphological processing is the morphological structure optimization expert E3: it performs asymmetric opening operation on the gradient enhancement image output by E2, and maintains the continuity and integrity of the main crack edges and lines of the image while filtering out high-frequency small artifacts, and outputs the morphologically optimized image.

[0021] The unit used to perform image splicing line removal processing is the splicing artifact removal expert E4: the morphologically optimized image output by E3 is processed by black hat operation and opening operation to extract the splicing line mask, and then the image splicing horizontal line and mud dark line artifacts are removed by using the Telea repair algorithm to obtain a preliminary crack binary image.

[0022] The formula for calculating the bright noise mask used in the noise suppression expert E1 is as follows:

[0023] ;in, , respectively, are the mean and standard deviation of the white cap; k is the threshold adjustment coefficient, used to control the sensitivity of bright noise detection; τ is the bright noise judgment threshold, and pixels with a white cap response value greater than τ are judged as bright noise pixels.

[0024] The gradient enhancement and binarization expert E2 uses the denoised image output by E1 as the input image I, and uses the Sobel operator to calculate the horizontal and vertical gradients of the input image I respectively; and solves for the gradient magnitude.

[0025] The gradient magnitude is normalized to the gray range of [0, 255] to obtain the normalized gradient map G;

[0026] In the gradient enhancement and binarization expert E2:

[0027] A quantile threshold strategy is used to filter high gradient regions. The normalized gradient map G is segmented according to a preset threshold. High gradient regions are identified as foreground, and low gradient and noise regions are identified as background, thus completing the image binarization process.

[0028] The morphological structure optimization expert E3 performs asymmetric morphological opening operations on the gradient enhancement image output by E2, including:

[0029] Small artifacts were removed by erosion using a 5×5 rectangular structural element;

[0030] The crack edges are restored by using a 3×3 rectangular structural element to expand, which balances noise suppression and crack line integrity.

[0031] The stitching artifact removal expert E4 receives the morphologically optimized image;

[0032] The black hat operation is performed by rectangular structuring elements to enhance the thin, dark line splicing artifacts, and the opening operation is combined to identify splicing lines and generate line masks.

[0033] The Telea patching algorithm is used to fill in the mask-covered area, eliminating image stitching lines and mud artifacts.

[0034] The step of dividing the preliminary binary image of the fracture into fracture regions to obtain a continuous and stable image of the fracture region includes:

[0035] For each row of pixels in the preliminary binary image of the crack, extract the maximum and minimum gray values ​​of that row, and calculate the gray value variation of that row.

[0036] A crack significance evaluation metric is constructed using the grayscale variation amplitude, the maximum row gradient value, and the dynamic adaptive weighting coefficient.

[0037] Set a global threshold and a local sliding window threshold. Only when the significance evaluation value of the row is greater than both the global threshold and the local sliding window threshold, is the row determined to belong to the crack candidate region and an initial crack candidate binary sequence is generated.

[0038] Obtain the interval length, gray value of the interval region, and gray value of the two sides of the crack segment from adjacent candidate segments. When the segment interval and gray value similarity meet the preset constraints, merge the adjacent candidate segments.

[0039] After the stitching is completed, the continuous crack region obtained is extracted from the initial binary crack image, and a continuous and stable crack region image is output.

[0040] The significance evaluation metric for the crack is expressed as: ;

[0041] in, This is a measure of the significance of the crack. For the first The range of grayscale variation in the row; For adaptive weighting coefficients, ; No. Maximum gradient response; For the first Standard deviation of line gray values This is the mean of the standard deviation of the gray levels of the entire image; This is the sensitivity adjustment factor.

[0042] The task recognizer calculates four types of quantitative indicators: image noise level score, gradient distribution dispersion, proportion of morphological fragmentation artifacts, and proportion of splicing line pixels.

[0043] Each quantitative indicator is compared with the four pre-set screening thresholds to determine its magnitude. When the indicator value exceeds the corresponding threshold, the matching task in the corresponding white noise removal, gradient processing, morphological processing, and image stitching removal processing unit is included in the set of tasks to be scheduled.

[0044] The generated set of tasks to be scheduled is sent to the task arranger to complete the subsequent sorting and scheduling, and the inherent execution order of the four types of processing units in the hybrid expert image enhancement model is not adjusted during the sorting process.

[0045] The image of the fracture region is processed using a four-dimensional Hough transform to extract the amplitude of the corresponding fracture. ,cycle Phase and centerline position The four structural parameters include:

[0046] The edge points of the crack region image are mapped to the Hough parameter space, and the sinusoidal crack curve with the highest confidence is determined by voting.

[0047] The Hough parameter space is extended to four-dimensional parameters: amplitude R, period P, phase β, and centerline position C. The sinusoidal fracture of the borehole is fitted and the four structural parameters are solved. The fracture curve obtained by fitting is used to generate a mask. The mask is then subjected to a pixel-by-pixel bitwise AND operation with the preprocessed surrounding rock borehole image to obtain a pure fracture binary image and the four structural parameters.

[0048] The four-dimensional Hough transform is expressed as:

[0049] ;

[0050] in, These are coordinates in a two-dimensional pixel coordinate system. Characterizes the severity of the fracture's tortuosity and undulation; The horizontal pixel span corresponding to the completion of one full sinusoidal morphological fluctuation of the crack; Determine the initial offset position of the crack bending waveform in the transverse direction; The vertical pixel coordinates of the center line of the crack waveform.

[0051] The beneficial effects of this disclosure include at least the following:

[0052] The method disclosed herein first uses a task recognizer to adaptively select image enhancement tasks based on multiple statistical features of the image. Then, a task arranger selects consecutive preceding tasks according to a fixed data flow order to perform noise suppression, gradient enhancement, morphological optimization, and stitching artifact removal collaborative processing. This filters out various image interferences and outputs a preliminary binary image of the cracks without relying on a deep learning dataset. Next, based on a crack saliency evaluation metric with fused adaptive weights combined with global and local dual thresholds, crack segment selection and stitching are completed, resulting in a continuous and stable crack region image. Finally, using four parameters extended to amplitude R, period P, phase β, and centerline position C... The Vehoff parameter space is used to fit sinusoidal fractures through voting. Based on the fitted curve, a mask is generated and a pixel-by-pixel bitwise AND operation is performed on the preprocessed image to obtain a pure binary image of the fracture and stably extract four fracture structure parameters. The entire process is interconnected and optimizes image quality step by step, reducing false and missed fracture detections caused by artifacts from the source. It also weakens the adverse effects of complex imaging conditions such as lighting and mud obstruction, and reduces human intervention throughout the process. This not only improves the accuracy and processing efficiency of fracture identification in surrounding rock boreholes, but also accurately quantifies the sinusoidal morphological geometric characteristics of fractures, providing stable and reliable digital data support for the quantitative analysis of surrounding rock stability under different rock strata and imaging conditions. Attached Figure Description

[0053] Figure 1 This is an overall flowchart of the method described in this invention.

[0054] Figure 2 This shows the task arrangement order and the resulting visuals.

[0055] Figure 3 The process steps and results for the E4 expert are shown in the image.

[0056] Figure 4 This is a schematic diagram of unfolding a borehole image into a sine wave.

[0057] Figure 5 A flowchart for fitting the crack in a borehole image using the Hough transform. Detailed Implementation

[0058] The following describes specific embodiments of the present disclosure to enable those skilled in the art to understand the disclosure. However, it should be understood that the disclosure is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the disclosure as defined and determined by the appended claims. All disclosed creations utilizing the concept of the present disclosure are protected. Specific Implementation Example 1:

[0060] This embodiment provides an electrical digital data processing method based on a hybrid expert model, which is applied to the automatic identification and parameter extraction of cracks in borehole images. This method aims to solve the problems of low accuracy, poor efficiency, and unstable parameter extraction in existing borehole images due to noise interference, complex textures, and human identification errors. The specific steps are as follows.

[0061] S1: Perform grayscale processing on the original borehole image. Grayscale processing uses the luminance mean method, which calculates the luminance mean value based on the red (R), green (G), and blue (B) components of each pixel in the image, according to the formula... The corresponding grayscale value is calculated. This method can effectively reduce color interference, allowing the image to retain only brightness information, thereby highlighting the contrast between the rock fissures and the background area.

[0062] Subsequently, the grayscale image is sharpened to enhance edge features. This embodiment employs an anti-sharpening mask algorithm for sharpening. This algorithm enhances high-frequency information by generating a blurred copy of the image and overlaying it with the original image, thereby improving the clarity of the crack boundary. Specific steps include: performing Gaussian blur on the grayscale image to obtain a low-frequency image; subtracting the blurred image from the original grayscale image to obtain a detail-enhanced image; and overlaying the detail-enhanced image with the original image according to a set weighting coefficient to generate a sharpened image.

[0063] S2: For the image I input to the model, it is first input to the task recognition module G. This module outputs a weight distribution vector w based on the statistical features of the image and the enhancement requirements, which guides the subsequent expert selection. Four expert models E1, E2, E3, and E4 implement different traditional image enhancement methods: white noise removal, gradient processing, kernel erosion and kernel dilation (morphological processing), and image stitching line removal, respectively. Note: The task arranger only selects consecutive tasks at the beginning of a fixed execution sequence and does not rearrange the order of the four types of expert units.

[0064]

[0065] The task recognizer is a core component that determines the expert assignment strategy. Unlike traditional methods that map single tasks, the recognizer proposed in this embodiment can automatically determine the enhancement requirements of an input image I, i.e., a recognition mechanism that allows one image to correspond to multiple tasks. Its main objective is to generate multiple sets of possible task labels based on the local features and overall statistical distribution of the image. And establish a mapping relationship with the corresponding expert weight distribution w.

[0066] The task recognizer, based on image statistical features and traditional pattern recognition methods, first extracts a series of indicators reflecting the crack characteristics from the input image I, including:

[0067] Noise level estimation: The intensity of background noise is measured by calculating the high-frequency energy, gray-level variance, or local entropy of the image;

[0068] Gradient distribution characteristics: The gradient histogram is calculated using the Sobel and Scharr operators to reflect the clarity and sharpness of the crack edges;

[0069] Morphological complexity: Measuring the connectivity and structural integrity of fractures through binarization and morphological opening and closing operations;

[0070] Splicing line residue detection: Possible splicing artifacts are detected through frequency domain analysis and gray-scale abrupt changes in row and column directions.

[0071] The above eigenvectors are denoted as via rule function This is converted into applicability scores for each enhancement task. The recognizer outputs the following task set:

[0072] in It can be given by logistic regression, linear discriminant analysis, or threshold-based discrimination rules. A threshold is selected for each task. Subsequently, the recognizer maps the task set to the expert model set E1, E2, E3, E4 to obtain the corresponding weight distribution vector w, thereby achieving multi-task recognition and sparse activation.

[0073] The task arranger only filters consecutive tasks at the beginning of a fixed execution sequence and does not rearrange the order of the four types of expert units. Figure 2 It shows the task arrangement order and the resulting effect.

[0074] Next comes the image processing stage. The four experts optimize for different scenarios. Noise suppression expert E1 focuses on removing white noise, i.e., suppressing isolated bright pixels. Gradient enhancement and binarization expert E2 mainly extracts gradients using the Sobel operator and binarizes the parts with the highest intensity. This is the key step in all image binarization, i.e., the gradient processing step. Morphological structure optimization expert E3 focuses on morphological optimization, using 5×5 kernel erosion and 3×3 kernel expansion to connect regions to make the morphology of cracks complete. Finally, stitching artifact removal expert E4 mainly targets the regular horizontal line artifacts generated during image stitching. The specific implementation process and basic principles of each expert will be explained in detail below.

[0075] The noise suppression expert E1 first performs median filtering on the input image to suppress isolated bright pixels. The median filter kernel radius is 3 or 5. Then, a white-hat algorithm is used to extract "small and bright" components, i.e., rectangular structuring elements with a side length of 3, resulting in a bright noise set. Based on the statistical distribution of the bright noise set, an adaptive threshold is used to generate a bright noise mask, as shown in the formula. As shown, where The mean and standard deviation of the white top cap are then used. The boundary is expanded by 2×2 and the original image is repaired with a mask using the Telea method with radius r=2. This removes small bright noise but maintains the integrity of the sinusoidal crack in the original image as much as possible.

[0076] The gradient enhancement and binarization expert E2 primarily extracts gradients using the Sobel operator and binarizes the parts with the highest intensity. This is the key step in all image binarization, namely the gradient processing step. Let the original grayscale image be I, and its gradients in the horizontal and vertical directions be defined as follows:

[0077]

[0078] To ensure comparability between different images, the gradient magnitude results were normalized to the grayscale range of [0, 255]. Subsequently, a quantile thresholding strategy was used to select high-gradient regions: Let the normalized gradient map be G, then the threshold τ is defined as... ; where grad_top_pct=5 means that only the top 5% of high gradient points in terms of intensity are retained.

[0079] This strategy highlights high-gradient regions as foreground and suppresses low-gradient and noisy regions as background, thereby effectively enhancing the salience and robustness of edge structures.

[0080] The morphological structure optimization expert E3 employs a morphological opening operation strategy based on "strong erosion + weak dilation". Specifically, it first uses a 5×5 rectangular structuring element K5×5 to erode the binary image to eliminate small isolated points and narrow spurious responses. Then, it uses a 3×3 rectangular structuring element K3×3 to dilate the eroded result to restore the overly weakened true edges. Its formal expression is as follows:

[0081]

[0082] in, These represent the erosion and dilation operations, respectively. The core idea behind this asymmetric core configuration is that the larger erosion operation ensures effective noise removal, while the smaller dilation operation achieves moderate structural restoration. The final result is able to suppress high-frequency noise while maintaining the continuity and integrity of key edges and lines.

[0083] The E4 stitching artifact removal expert targets regular horizontal line artifacts generated during image stitching. First, it performs a black hat operation on the input using a rectangular structuring element of length L=(15,2)px to explicitly enhance thin, dark lines. Then, it combines this with an opening operation to identify candidate line segments on the geometric structure. The combined result is a line mask. Finally, the Telea patching algorithm is used to fill these areas, achieving the removal of horizontal lines. For special cases, this method can be extended to the vertical direction to further clean up straight line artifacts, and it is particularly effective for artifacts caused by muddy dark lines. The final processing procedure and results are as follows: Figure 3 As shown.

[0084] In the process of automatic crack identification in borehole images, since cracks are usually accompanied by gray-level abrupt changes and edge gradient enhancement in the longitudinal expansion direction, this part constructs a row-level gray-level-gradient adaptive fusion crack saliency evaluation method based on this objective imaging characteristic, which is used to achieve stable partitioning of crack regions.

[0085] Assume the borehole unfolded image is a two-dimensional grayscale function. ,in Represents the row index in the vertical depth direction. Indicates the column index in the horizontal direction.

[0086] For the first For each row of pixels, extract its maximum and minimum grayscale values, and define the grayscale variation range of that row as: ;

[0087] Simultaneously, the gradient magnitude of each pixel in the row is calculated in the horizontal direction, and the maximum value is taken as the row-level maximum gradient response: ;

[0088] in, Used to characterize the strongest abrupt change in the image edge at this depth location.

[0089] To avoid the instability of fracture response caused by fixed weights under different lithological conditions, illumination fluctuations, and mud interference scenarios, this embodiment introduces row-level adaptive weight coefficients. Construct a crack significance evaluation metric: Among them, adaptive weight coefficients The calculation method is as follows: The grayscale statistical distribution of the row is dynamically adjusted. In the formula: For the first Standard deviation of row grayscale values; This is the mean of the standard deviation of the gray levels of the entire image; This is the sensitivity adjustment factor, with a value range of [value range missing]. By employing the above method, when the borehole image is in a low-contrast or high-noise region, the adaptive weighting coefficient will automatically increase the proportion of the gradient component in the fusion evaluation, thereby enhancing the response stability of weak fractures.

[0090] S3: Obtaining the fracture significance evaluation sequence Subsequently, this embodiment employs a joint partitioning strategy with dual constraints of global and local thresholds: ;

[0091] in: For all The mean; For all Standard deviation; The global noise suppression coefficient is preferably taken as follows: .

[0092] In the first row-centered sliding window Within, the local threshold is constructed as follows:

[0093]

[0094] in It is half the width of the sliding window, used to suppress the interference of isolated strong noise points on the partitioning results.

[0095] In line A region is considered a candidate region for fracture if it meets all of the following conditions:

[0096]

[0097] This generates an initial binary sequence of candidate fractures, satisfying the following condition: otherwise Since noise interference during actual borehole imaging may cause a single fracture to be segmented into multiple adjacent short segments, this embodiment further performs automatic stitching processing on the candidate fracture segments: Let the interval length between two adjacent candidate fracture segments be... The average gray level of the corresponding interval segment is The average gray value of adjacent crack segments is , When the following conditions are met: ;and When the two fractured sections are determined to belong to the same continuous fracture, automatic splicing is performed. Specifically: This represents the threshold for crack continuity tolerance. This is the grayscale similarity constraint threshold. This continuity constraint mechanism can effectively prevent "false deletion" of cracks caused by local noise.

[0098] By introducing an adaptive gray-gradient fusion mechanism, global-local dual robust threshold constraints, and fracture continuity splicing rules, this part can significantly improve the stability of weak fracture extraction under complex borehole imaging conditions, and effectively suppress false detection problems caused by mud noise, uneven illumination, and borehole wall reflection, thus demonstrating good engineering applicability.

[0099] S4: Perform the Hough transform. The core idea is to map points in the image space to the parameter space, and then use cumulative statistics to find clusters of points in the parameter space, thereby identifying salient geometric structures in the image. Taking line detection as an example, a planar line can usually be represented by the Cartesian equation y = kx + b.

[0100] But this form is This can lead to instability in the representation. To avoid this problem, the Hough transform uses polar coordinates to represent a straight line: Where ρ represents the distance from point (x, y) to the origin, and θ represents the angle between the perpendicular line and the x-axis. For each edge point (xi, yi) in the image space, a corresponding sine curve can be drawn in the parameter space (ρ, θ). If multiple edge points lie on the same straight line, their corresponding curves will intersect at a point (ρ, θ) in the parameter space, thus completing the line detection.

[0101] In the two-dimensional plane of the borehole imaging unfolded diagram, such as Figure 3 A coordinate system is established as follows: the horizontal axis x represents the circumferential drilling distance, and the vertical axis y represents the axial drilling depth. Based on this coordinate system, the "sinusoidal" fracture can be characterized by the following mathematical model: The complete process of fitting the sinusoidal fracture in the borehole using the four-dimensional Hough transform is shown in Figure 5. The Hough transform can also be used for detection by expanding the parameter space to a higher dimension (R, P, β, C). Fracture edge points will form corresponding clusters in this parameter space, and the local peak value of the accumulator represents the optimal sinusoidal curve parameters.

[0102] Because noise is unavoidable in the binary image of the borehole, i.e. Figure 4 Interfering pixels in the image can lead to misjudgments or curve discontinuities if direct pixel-based geometric detection is used. The Hough transform enhances feature consistency through a voting mechanism, ensuring that the sinusoidal features of the crack remain consistent even in noisy backgrounds. Figure 4 Even the crack structure pixels in the image can still be accurately extracted through significant peaks in the parameter space.

[0103] The above disclosure only discloses a few specific implementation scenarios. However, this disclosure is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this disclosure.

Claims

1. A method for processing electrical digital data based on a hybrid expert model, for processing borehole images with fractured surrounding rock, characterized in that, include: Acquire surrounding rock borehole images, and perform grayscale processing and sharpening preprocessing on the surrounding rock borehole images; The preprocessed borehole image of the surrounding rock is input into a hybrid expert image enhancement model. The hybrid expert image enhancement model includes a task recognizer, a task arranger, and four processing units with fixed data flow dependencies: noise removal, gradient processing, morphological processing, and image stitching removal. The task recognizer extracts four statistical features of image noise, gradient, morphology, and stitching, and filters the image enhancement tasks to form a set of tasks to be scheduled. The task arranger calls the processing units corresponding to the tasks to be scheduled in a fixed order of noise removal, gradient processing, morphological processing, and image stitching removal, and outputs a preliminary binary image of the fracture after processing. The initial binary image of the fracture is divided into fracture regions to obtain a continuous and stable image of the fracture region; The image of the fracture region is processed by four-dimensional Hough transform to extract four structural parameters corresponding to the fracture: amplitude R, period P, phase β, and centerline position C, and output a pure binary image of the fracture.

2. The electrical digital data processing method based on a hybrid expert model according to claim 1, characterized in that, The grayscale processing and sharpening preprocessing of the surrounding rock borehole image includes: Based on the red of each pixel in the surrounding rock borehole image ,green ,blue Components, according to the formula The grayscale value corresponding to the pixel is obtained until a grayscale image is obtained. Gaussian blur is applied to the grayscale image to obtain a low-frequency image; Subtracting the grayscale image from the low-frequency image yields the detail-enhanced image; The enhanced detail image is superimposed on the surrounding rock borehole image according to the set weight coefficient to generate a sharpened image, thus completing the preprocessing.

3. The electrical digital data processing method based on a hybrid expert model according to claim 1, characterized in that: The unit used to perform white noise removal is the noise suppression expert E1: it performs white top hat operation on the preprocessed surrounding rock borehole image to generate a bright noise set, removes white noise from the image while preserving the complete shape of sinusoidal cracks in the image, and outputs a denoised image. The unit used to perform gradient processing is the gradient enhancement and binarization expert E2: the Sobel operator is used to solve the image gradient of the denoised image, highlighting high gradient regions as foreground and suppressing low gradient noise regions as background, to obtain a gradient enhancement image, thereby improving the salience and robustness of the crack edge structure. The unit used to perform morphological processing is the morphological structure optimization expert E3: it performs an asymmetric opening operation on the gradient-enhanced image, and maintains the continuity and integrity of the main crack edges and lines of the image while filtering out high-frequency small artifacts, and outputs the morphologically optimized image. The unit used to perform image splicing line removal processing is the splicing artifact removal expert E4: it extracts the splicing line mask from the morphologically optimized image through black hat operation and opening operation, and then removes the image splicing horizontal lines and mud dark line artifacts with the help of the Telea repair algorithm to obtain a preliminary crack binary image.

4. The electrical digital data processing method based on a hybrid expert model according to claim 3, characterized in that: The formula for calculating the bright noise mask used in the noise suppression expert E1 is as follows: ;in, , respectively, represent the mean and standard deviation of the white cap; k is the threshold adjustment coefficient; τ is the bright noise judgment threshold.

5. The electrical digital data processing method based on a hybrid expert model according to claim 3, characterized in that: The gradient enhancement and binarization expert E2 uses the denoised image output by the noise suppression expert E1 as the input image I, and uses the Sobel operator to calculate the horizontal gradient and vertical gradient of the input image I respectively. And solve for the gradient magnitude; The gradient magnitude is normalized to the gray range of [0, 255] to obtain the normalized gradient map G; In the gradient enhancement and binarization expert E2: A quantile threshold strategy is used to filter high gradient regions. The normalized gradient map G is segmented according to a preset threshold. High gradient regions are identified as foreground, and low gradient and noise regions are identified as background, thus completing the image binarization process.

6. The electrical digital data processing method based on a hybrid expert model according to claim 3, characterized in that, The morphological structure optimization expert E3 performs asymmetric morphological opening operations on the gradient-enhanced image output by the gradient enhancement and binarization expert E2, including: Small artifacts were removed by erosion using a 5×5 rectangular structural element; The crack edges are restored by using a 3×3 rectangular structural element to expand, which balances noise suppression and crack line integrity.

7. The electrical digital data processing method based on a hybrid expert model according to claim 3, characterized in that: The stitching artifact removal expert E4 receives the morphologically optimized image; The black hat operation is performed by rectangular structuring elements to enhance the thin, dark line splicing artifacts, and the opening operation is combined to identify splicing lines and generate line masks. The Telea patching algorithm is used to fill in the mask-covered area, eliminating image stitching lines and mud artifacts.

8. The electrical digital data processing method based on a hybrid expert model according to claim 1, characterized in that, The step of dividing the preliminary binary image of the fracture into fracture regions to obtain a continuous and stable image of the fracture region includes: For each row of pixels in the preliminary binary image of the crack, extract the maximum and minimum gray values ​​of that row, and calculate the gray value variation of that row. A crack significance evaluation metric is constructed using the grayscale variation amplitude, the maximum row gradient value, and the dynamic adaptive weighting coefficient. Set a global threshold and a local sliding window threshold. Only when the significance evaluation value of the row is greater than both the global threshold and the local sliding window threshold, is the row determined to belong to the crack candidate region and an initial crack candidate binary sequence is generated. Obtain the interval length, gray value of the interval region, and gray value of the two sides of the crack segment from adjacent candidate segments. When the segment interval and gray value similarity meet the preset constraints, merge the adjacent candidate segments. After the stitching is completed, the continuous crack region obtained is extracted from the initial binary crack image, and a continuous and stable crack region image is output. The significance evaluation metric for the crack is expressed as: ; in, This is a measure of the significance of the crack. For the first The range of grayscale variation in the row; For adaptive weighting coefficients, ; No. Maximum gradient response; For the first Standard deviation of line gray values This is the mean of the standard deviation of the gray levels of the entire image; This is the sensitivity adjustment factor.

9. The electrical digital data processing method based on a hybrid expert model according to claim 1, characterized in that: The task recognizer calculates four types of quantitative indicators: image noise level score, gradient distribution dispersion, proportion of morphological fragmentation artifacts, and proportion of splicing line pixels. Each quantitative indicator is compared with the four pre-set screening thresholds to determine its magnitude. When the indicator value exceeds the corresponding threshold, the matching task in the corresponding white noise removal, gradient processing, morphological processing, and image stitching removal processing unit is included in the set of tasks to be scheduled. The generated set of tasks to be scheduled is sent to the task arranger to complete the subsequent sorting and scheduling, and the inherent execution order of the four types of processing units in the hybrid expert image enhancement model is not adjusted during the sorting process.

10. The electrical digital data processing method based on a hybrid expert model according to claim 1, characterized in that, The image of the fracture region is processed using a four-dimensional Hough transform to extract the amplitude of the corresponding fracture. ,cycle Phase and centerline position The four structural parameters include: The edge points of the crack region image are mapped to the Hough parameter space, and the sinusoidal crack curve with the highest confidence is determined by voting. The Hough parameter space is extended to four-dimensional parameters: amplitude R, period P, phase β, and centerline position C. A sinusoidal fracture in the borehole is fitted and the four structural parameters are solved. A mask is generated using the fitted fracture curve. The mask and the binary image of the fracture region are then subjected to a pixel-by-pixel bitwise AND operation to obtain a pure binary image of the fracture and the four structural parameters. The four-dimensional Hough transform is expressed as: ； in, These are coordinates in a two-dimensional pixel coordinate system. Characterizes the severity of the fracture's tortuosity and undulation; The horizontal pixel span corresponding to the completion of one full sinusoidal morphological fluctuation of the crack; Determine the initial offset position of the crack bending waveform in the transverse direction; The vertical pixel coordinates of the center line of the crack waveform.

Images