A visual detection method for pre-fracture stage of hard roof working face mining in a coal mine

Abstract

The present application relates to a kind of coal mine hard roof working face mining's fracture pre-period visual detection method, belong to image processing technical field.The method is directly in the original gray annular image of coal mine hard roof borehole wall completes the positioning and parameter extraction of crack.Specifically, by constructing multi-scale gradient response function and crack morphology constraint model in annular domain, gradient analysis is carried out along radial and angle two directions respectively, and cross-ring zone weighted voting mechanism is introduced to re-estimate the confidence of detection result, so that it is completed before crack information is interpolated and degraded.Carry out local unfolding and parameter precision measurement to the identified crack region after detection, instead of whole image unfolding, can fundamentally avoid information loss under crack identification mode after original awakening image is unfolded, significantly improve the recall rate and positioning accuracy of crack detection in borehole peep image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a visual detection method for the early stage of fracture in coal mine working faces with hard roofs. Background Technology

[0002] In evaluating the fracturing effect of caving holes in hard coal mine roofs, borehole inspection is the core method for obtaining information on borehole wall fracture development. Borehole inspection systems utilize rotating prisms or other methods... Figure 1 The panoramic probe with circumferential opening shown acquires a 360° annular image of the borehole wall. This annular image fully records the rock strata structure, natural fractures, and newly formed fractures after artificial fracturing along the entire circumference of the borehole wall. Due to the poor intuitiveness of the annular image, existing technologies generally unfold the annular image into a rectangular panoramic view before performing manual interpretation or automatic identification of fractures.

[0003] However, when unfolding a ring-shaped image, existing technologies exhibit severe blurring and pixelation in the area near the inner circle of the borehole panoramic image, making it difficult to discern fracture details. Specifically, in the same unfolded image, fracture boundaries are clearly identifiable in the outer circle, while the edges of the same fracture in the inner circle are blurred, jagged, or even broken, leading to significant errors in fracture length measurement and trajectory tracking. For micro-cracks and fine fissures distributed in the inner circle, their characteristic details are severely degraded during the unfolding process, directly affecting the accuracy of fracture identification and quantitative analysis.

[0004] The root cause of the aforementioned defects lies in the fact that the process of unfolding a ring image into a rectangle is essentially a non-uniform sampling mapping process. The pixels of the ring image are arranged in polar coordinates—distance information along the radial direction and circumferential information along the angular direction. The inner circle has a smaller radius and a much smaller circumference than the outer circle, but in the unfolded rectangular image, the inner and outer circles occupy the same pixel width. This means that the limited information of the inner circle is forcibly "stretched" to fill the same spatial width as the outer circle. The information originally carried by one pixel needs to be expressed by multiple pixels through interpolation after unfolding. The smoothing effect produced by this interpolation causes the loss of high-frequency details at the crack edges, resulting in a significant decrease in the accuracy of crack detection.

[0005] In other words, the current detection of cracks in the hard roof walls of coal mines suffers from insufficient accuracy. Summary of the Invention

[0006] In view of this, the present invention provides a visual detection method for the early stage of fracture in the mining of hard roof working faces in coal mines, so as to solve the technical problem of insufficient detection accuracy of borehole wall cracks in hard roof working faces in coal mines.

[0007] The present invention provides a visual detection method for the early stage of fracture in coal mine hard roof working face mining, comprising:

[0008] A grayscale annular image of the borehole wall of a hard roof in a coal mine is obtained. The annular image is divided into a preset number of annular bands along the radial direction. At the same position in each annular band, the grayscale value of each pixel is obtained from the same angle to obtain the grayscale sequence of the corresponding annular band.

[0009] In any grayscale sequence, obtain the large-scale difference value and the small-scale difference value of each grayscale value. Based on the large-scale difference value and the sign difference between the large-scale difference value and the small-scale difference value, determine the crack response intensity of each grayscale value in any grayscale sequence. Based on the crack response intensity, determine the candidate pixel points in any grayscale sequence.

[0010] Each candidate pixel with a positive small-scale difference value in any grayscale sequence is paired with each candidate pixel with a negative small-scale difference value within a preset angle change range under a preset angle change direction. The pairing confidence of any pair is determined based on the angle difference between the two pixels in any pair and the magnitude difference of the small-scale difference value. The effective pairing in any grayscale sequence is determined based on the pairing confidence.

[0011] Construct a standardized angle grid, and determine the cross-ring support for each standardized angle based on the mapping value of the undetermined fracture center angle corresponding to each effective matching pair in each gray-scale sequence on the standardized angle grid and the pairing confidence. Then, determine the candidate standardized angle based on the cross-ring support.

[0012] Based on the magnitude of any candidate standardized angle and the number of valid matching pairs within a preset angle deviation range of any candidate standardized angle, the true crack confidence of any candidate standardized angle is determined, and crack detection is completed based on the true crack confidence of each candidate standardized angle.

[0013] Furthermore, the large-scale difference value is the step size difference value, and the small-scale difference value is the center difference value.

[0014] Further, determining the crack response intensity of each gray value in any of the gray-scale sequences includes:

[0015] When the sign of the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence are different, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence shall be taken as the crack response intensity of any gray value in any gray-scale sequence.

[0016] When the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence have the same sign, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence is amplified by a preset factor and used as the crack response intensity of any gray value in any gray-scale sequence.

[0017] Further, determining candidate pixels in any of the grayscale sequences based on the crack response intensity includes:

[0018] The crack response threshold of any grayscale sequence is determined based on the median and absolute deviation of the crack response intensity of all grayscale values ​​in any grayscale sequence, wherein the crack response threshold is proportional to both the median and the absolute deviation of the median.

[0019] The pixel corresponding to the gray value in any gray-scale sequence whose crack response intensity is greater than the crack response threshold is recorded as a candidate pixel in any gray-scale sequence.

[0020] Furthermore, the preset angle change direction refers to the clockwise angle change direction.

[0021] Further, determining the pairing confidence of any of the said matching pairs includes:

[0022] The difference between the pixel distance corresponding to the angle difference between two pixels in any of the matching pairs and the typical width of the crack in the borehole wall of the hard roof of the coal mine is recorded as the first pairing characterization term.

[0023] The smaller of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair is divided by the sum of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair, and the result is recorded as the second pairing characterization term.

[0024] The pairing confidence of any matching pair is determined based on the first pairing characteristic and the second pairing characteristic, wherein the pairing confidence is inversely proportional to the first pairing characteristic and directly proportional to the second pairing characteristic.

[0025] Furthermore, the cross-ring support under each standardized perspective is determined, including:

[0026] ,

[0027] in, From a standardization perspective The support of the transannular band below, This indicates the total number of the rings. Indicates the first The total number of valid matching pairs corresponding to each ring band. Indicates the first The first ring band The pairing confidence of each of the valid matching pairs. Represents the Gaussian kernel function. Indicates the first The first ring band The mapping value of the undetermined fracture center angle corresponding to each valid matching pair in the normalized angle grid. This indicates the preset angle tolerance bandwidth.

[0028] Furthermore, determining candidate normalized angles based on the cross-ring support includes:

[0029] The cross-ring support under all standardized angles is used to form a cross-ring support sequence. Non-maximum suppression is used to extract the local peaks of the cross-ring support sequence, and the standardized angles corresponding to the local peaks are recorded as the candidate standardized angles.

[0030] Furthermore, determining the true crack confidence level for any of the candidate standardized angles includes:

[0031] The ratio of the cross-ring support at any of the candidate normalized angles to the maximum cross-ring support at all the candidate normalized angles is denoted as the first true fracture characterization term.

[0032] The ratio of the number of rings with valid matching pairs within a preset angle deviation range of any candidate standardized angle to the total number of rings is denoted as the second true fracture characterization term.

[0033] The product of the first true fracture characterization term and the second true fracture characterization term is denoted as the true fracture confidence level for any of the candidate standardized angles.

[0034] Furthermore, crack detection is performed based on the true crack confidence level of each candidate standardized angle, including:

[0035] When the confidence level of the true crack at any candidate standardized angle is greater than a preset true crack confidence threshold, it is considered that a true crack exists at any candidate standardized angle. If the standard deviation between the center angles of the undetermined cracks corresponding to all valid matching pairs within the preset angle deviation range of any candidate standardized angle and any candidate standardized angle is less than a preset standard deviation threshold, and the second true crack characterization term corresponding to any candidate standardized angle is greater than a preset circumferential proportion threshold, the true crack at any candidate standardized angle is determined to be a radially penetrating crack; otherwise, the true crack at any candidate standardized angle is determined to be a circumferential crack.

[0036] The advantages of this invention compared to the prior art are:

[0037] The embodiments of the present invention directly locate the real cracks based on the original grayscale annular image of the borehole wall of the hard roof of the coal mine. Then, based on the location results, only the real crack location is locally expanded and the parameters are precisely measured. This can avoid the crack identification error caused by information loss in the existing technology mode of first expanding the annular image and then identifying it, and significantly improve the detection accuracy of cracks in the borehole wall of the hard roof of the coal mine. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a picture of the panoramic probe of the borehole inspection system;

[0040] Figure 2 This is a schematic diagram of a ring image after binarization processing and the corresponding optical center during the optical center calibration process provided in Embodiment 1 of the present invention. Detailed Implementation

[0041] The overall concept of this invention is as follows:

[0042] To avoid excessive stretching of inner pixels during annular image unfolding, which leads to loss of crack details and a significant decrease in crack detection accuracy, this invention directly locates and extracts parameters of the crack within the original annular domain (polar coordinate space). Specifically, by constructing a multi-scale gradient response function and a crack morphology constraint model within the annular domain, gradient analysis is performed along both radial and angular (circumferential) directions. A cross-annular zone weighted voting mechanism is introduced to re-evaluate the confidence level of the detection results, thus capturing crack information before it is degraded by interpolation. After detection, only the identified crack region is locally unfolded and its parameters are precisely measured, rather than the entire image is unfolded. This fundamentally avoids the dilemma of information loss and subsequent recovery in the "unfold first, then identify" mode, significantly improving the recall rate and localization accuracy of crack detection in borehole peep images.

[0043] To further illustrate the technical solution of the present invention, specific embodiments are described below.

[0044] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a particular feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. Furthermore, a particular feature, structure, or characteristic in one or more embodiments may be combined in any suitable form, and the terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically emphasized.

[0045] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0046] Method Implementation Examples:

[0047] Embodiment 1 of the present invention provides a visual detection method for the early stage of fracture in the mining of hard roof coal faces, the method including the following steps:

[0048] S101, acquire a grayscale annular image of the borehole wall of the hard roof of the coal mine, divide the annular image into a preset number of annular bands along the radial direction, and acquire the grayscale value of each pixel at the same position and from the same angle in each annular band to obtain the grayscale sequence of the corresponding annular band.

[0049] Firstly, through, as... Figure 1 The panoramic probe in the borehole inspection system shown acquires a raw grayscale annular image of the borehole wall to be inspected. Specifically, after the caving hole construction is completed, the panoramic probe of the borehole inspection system is pushed to the section of the borehole to be inspected. The probe's front end is equipped with a 360° side window and an LED auxiliary light source. After the light emitted by the light source illuminates the borehole wall, the reflected light is collected by the CMOS image sensor built into the probe and processed into grayscale, forming an annular image containing circumferential information of the borehole wall. During acquisition, the depth position of the probe is recorded by a depth encoder to ensure accurate depth coordinates for subsequent fracture location.

[0050] After acquiring the annular image, the geometric reference parameters of the borehole imaging system need to be extracted from the image. Due to assembly errors between the probe optical system and the image sensor, as well as lens distortion, the center point of the annular image usually does not completely coincide with the geometric center of the image pixel matrix. Therefore, calibration is required to obtain the actual optical center coordinates. The specific calibration method is as follows: Under calibration conditions without borehole wall obstruction, a blank annular image illuminated by a uniform light source is acquired. The blank image is then binarized to extract the geometric reference parameters of the borehole imaging system. Figure 2 The annular bright band region shown is fitted with an ellipse, and the center point of the fitted ellipse is taken as the optical center of the annular image. Let the minor semi-axis of the ellipse be denoted as the inner diameter of the annular image. The major semi-axis is denoted as the outer diameter. Optical Center The inner diameter serves as the rotation reference point for subsequent polar coordinate transformations. and outer diameter This limits the range of the effective annular area. The central dark area within the inner diameter and the invalid area outside the outer diameter do not contain hole wall information and are excluded in subsequent processing.

[0051] In a ring-shaped image, the circumference length varies at different radii. The inner ring contains only a few pixels, while the outer ring contains a large number of pixels. If crack detection is performed directly using the original pixels as the unit, the number of available sample points within each angular interval of the inner ring is far fewer than that of the outer ring, resulting in insufficient statistical significance and making the detection results susceptible to noise interference. Therefore, this embodiment divides the ring-shaped image radially into several concentric rings, analyzing each ring independently. This ensures that each ring has enough sampling points to support statistical judgment while allowing for the setting of appropriate detection sensitivities for different radii.

[0052] At the optical center of the aforementioned borehole imaging system , inner diameter and outer diameter Then, with the optical center With the pole as the pole and the vertically upward direction as the polar axis, establish a polar coordinate system for the annular region. Define the annular effective region on the annular image. Divided radially into Concentric rings , ,in That is, the total number of divided rings, and the radial width of each ring is... The formula for its calculation is:

[0053]

[0054] in, The radial width of a single loop, in pixels. and These represent the outer and inner diameters of the effective annular region on the annular image, respectively; the total number of annular zones is [value missing]. In this embodiment, the width of each ring is preferably set to approximately 4 pixels. Round down to the nearest integer.

[0055] Among them, the total number of rings The selection criteria are as follows: when the annular band width is 4 pixels, even in the innermost circle (the smallest radius) of the annular borehole detection image, the number of circumferential pixels in each annular band is approximately In a typical borehole image (with an inner diameter of about 150 pixels), this value is about 235 pixels, which is sufficient to support reliable statistical inference; at the same time, the width of 4 pixels is much smaller than the width of a typical radial crack (usually 10-50 pixels), ensuring that the positioning accuracy of the radial crack in the radial direction is not affected by the annular partition.

[0056] No. Annular band The radius covered is defined as follows:

[0057]

[0058] in, These are the radial coordinates in the polar coordinate system. Angular coordinates; For ring index, Corresponding to the innermost ring (closest to the inner diameter) ), Corresponding to the outermost ring (closest to the outer diameter) ).

[0059] This partitioning method divides the entire annular effective region radially into... Each concentric ring has an equal radial width. This uniform partitioning method makes the analysis results between different rings comparable in the radial direction, which facilitates the subsequent judgment of fracture connectivity across rings and radial tracing.

[0060] After dividing the area into individual rings, the grayscale values ​​of each pixel on the circumference can be obtained by traversing the circumference from the same position and angle within each ring to obtain the grayscale sequence of the corresponding ring, thus representing the grayscale changes of the rings in a one-dimensional sequence form. In this embodiment, the "same position" preferably refers to the radial median position of the ring, and the "same angle" refers to the vertically upward polar axis angle. Therefore, for the... Annular band Unfold the circumferential pixels at the radial median position (or the center line position of the ring) to form a length of One-dimensional grayscale sequence , ,in For the first The number of pixels around the center line of the ring (the number of sampling points around the circumference is equal to the number of pixels along the circumference). , The radial width of a single ring is represented by the corresponding... This represents the radius at the center line of the ring. Round down to the nearest integer.

[0061] S102, in any grayscale sequence, obtain the large-scale difference value and the small-scale difference value of each grayscale value, determine the crack response intensity of each grayscale value in any grayscale sequence based on the large-scale difference value and the sign difference between the large-scale difference value and the small-scale difference value, and determine the candidate pixel points in any grayscale sequence based on the crack response intensity.

[0062] Based on the grayscale sequences of each annulus obtained in the previous step, the fracture characteristics can be analyzed according to the grayscale sequences. Each annulus can be regarded as a one-dimensional grayscale sequence unfolding along the angular direction. The physical essence of the fracture in this grayscale sequence is not a single grayscale abrupt change, but a pair of edge structures with opposite directions and similar amplitudes. The front edge is a descending edge (negative gradient) from the background (i.e., the normal rock wall without fractures) to the interior of the fracture, and the rear edge is an ascending edge (positive gradient) from the interior of the fracture to the background. The fracture center is located at the zero-crossing point between the positive and negative gradients. However, detecting only the gradient on one side will misjudge the rock wall texture boundary as a fracture and cannot accurately locate the fracture center. Therefore, this embodiment proposes to construct two progressive steps: "multi-scale response enhancement - morphological constraint pairing".

[0063] This step first completes the construction of the "multi-scale response enhancement" component: simultaneously calculating large-scale difference and small-scale difference, where the large-scale difference is preferably a step-size difference and the small-scale difference is preferably a center difference. The large-scale difference captures the overall contour of the fracture and suppresses high-frequency noise, while the small-scale difference accurately locates the precise position of the edge. When the gradient directions of the two are consistent, the response at that point is enhanced; when the directions are inconsistent, only the basic response is retained without enhancement, thereby actively suppressing non-fracture noise while enhancing the fracture signal.

[0064] Specifically, first consider the one-dimensional grayscale sequence. Perform multi-scale differencing operations. First, calculate the large-scale differencing. A step-size difference method is used to capture the overall profile of the fracture and suppress high-frequency noise interference.

[0065]

[0066] In the formula, For the first On the center line of the first ring, along the direction of angular change... The large-scale difference value at the nth pixel, that is, the nth pixel... In the gray-scale sequence of the nth ring band Large-scale difference values ​​at each pixel; The preferred value for the difference step size is [value to be filled in]. Pixels, 3 pixels correspond to an actual borehole wall angle range of about 15°. This scale is larger than the influence range of single pixel noise (usually 1~2 pixels), which can effectively suppress high frequency noise; at the same time, it is much smaller than the extension range of the typical fracture in the angular direction (usually greater than 30°), ensuring that the overall contour features of the fracture are preserved. and The first In the gray-scale sequence of the nth ring band The and the first The grayscale value at each pixel is determined by the fact that the ring is a continuous loop, meaning that 0° and 360° are physically located at the same point. Therefore, when data needs to be retrieved from the end of the sequence (close to 360°), it is not considered an out-of-bounds error; instead, data is retrieved from the beginning of the sequence (0°), and vice versa. In other words, boundary positions are handled in a circular loop. It's easy to understand that the position of each pixel in the grayscale sequence of any ring is different, so each pixel corresponds to a different angle value in polar coordinates. Therefore, when representing a specific pixel, the corresponding angle value can be considered implicitly represented.

[0067] The reason for using step-length difference instead of adjacent pixel difference is that the crack edge occupies a certain width range in the angular direction. Adjacent pixel difference is too sensitive to single-point noise inside the edge, while step-length difference measures the gray-scale difference on both sides of the crack edge as a whole, which can more stably reflect the existence of the crack.

[0068] Simultaneously calculate small-scale differences The central difference method is used to accurately locate the precise position of the crack edge.

[0069]

[0070] In the formula, For the first On the center line of the first ring, along the direction of angular change... The small-scale difference value at the nth pixel, that is, the nth pixel... In the gray-scale sequence of the nth ring band Small-scale difference values ​​at each pixel. The response peak of the central difference can be precisely aligned with the location of the mutation center, providing sub-pixel-level localization correction for large-scale detection results.

[0071] After obtaining the large-scale difference value and small-scale difference value at each pixel in the grayscale sequence of the current ring band, since the edge pixels of the real crack exhibit a consistent grayscale change trend regardless of the large-scale difference, the crack response intensity of each grayscale value in any grayscale sequence can be determined based on the magnitude of the large-scale difference value and the sign difference between the large-scale difference value and the small-scale difference value, including:

[0072] When the sign of the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence are different, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence shall be taken as the crack response intensity of any gray value in any gray-scale sequence.

[0073] When the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence have the same sign, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence is amplified by a preset factor and used as the crack response intensity of any gray value in any gray-scale sequence.

[0074] Furthermore, as a preferred approach, multi-scale differential responses are integrated to define the crack response intensity of each gray value in the gray-scale sequence of the ring band. for:

[0075]

[0076] In the formula, For the first On the center line of the first ring, along the direction of angular change... The crack response intensity at the nth pixel, i.e., the nth pixel In the gray-scale sequence of the nth ring band Crack response intensity at each pixel For the first In the gray-scale sequence of the nth ring band The absolute value of the large-scale difference at each pixel point reflects the overall contrast of the crack. This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. This is a sign function that returns the sign of the input value. This is the multi-scale consistency enhancement coefficient, with a value of [value missing]. When the directions of the large-scale and small-scale gradients are consistent, i.e., when the signs of the large-scale and small-scale difference values ​​are consistent, the crack signal is amplified by 1.5 times; when the directions are inconsistent, the original value is maintained; 0.5 achieves the best balance between crack signal enhancement and noise suppression. A value that is too large (e.g., >1.0) will over-amplify residual noise, while a value that is too small (e.g., <0.2) will not have a significant enhancement effect. Specifically, when the directions of the large-scale gradient and the small-scale gradient are consistent, it indicates that the crack has a consistent gray-level change trend across the scale range (both large and small scales), and this location is more likely to be a pixel at the edge of a real crack. In this case, the response intensity at this location is enhanced; when the directions are inconsistent, it indicates that the gradient response may originate from noise or non-crack texture, and it is not enhanced. This mechanism effectively amplifies the crack signal response while relatively suppressing the background noise response, achieving an active improvement in the signal-to-noise ratio. The large-scale difference directly outputs the gray-level difference value to characterize the overall contrast of the crack; the small-scale difference outputs the gray-level change rate, used only for determining directional consistency.

[0077] In obtaining the crack response intensity Then, it is necessary to determine the current number. The crack response threshold corresponding to each ring zone This method filters candidate pixels that are most likely to be the edge of a real crack from all pixels corresponding to the grayscale sequence. Unlike traditional methods that calculate the threshold based on the global mean and standard deviation of the crack response intensity corresponding to each grayscale value in the grayscale sequence, this embodiment uses robust statistics (i.e., statistical indicators that are insensitive to outliers) to reduce the contamination of the threshold calculation by potential crack regions and obtain the crack response threshold. And based on the crack response intensity Determine the first Candidate pixels in the grayscale sequence of each ring include:

[0078] The crack response threshold of any grayscale sequence is determined based on the median and absolute deviation of the crack response intensity of all grayscale values ​​in any grayscale sequence, wherein the crack response threshold is proportional to both the median and the absolute deviation of the median.

[0079] The pixel corresponding to the gray value in any gray-scale sequence whose crack response intensity is greater than the crack response threshold is recorded as a candidate pixel in any gray-scale sequence.

[0080] Furthermore, as a preferred option, the first The crack response threshold corresponding to each ring zone for:

[0081]

[0082] In the formula, For the first The crack response threshold corresponding to each annular zone; for exist The median within the range; for The absolute deviation of the median, ; This is the threshold scaling factor, with a value of [value missing]. The threshold is set at 3 times the median absolute deviation, derived from the 3σ criterion in statistics. This approach ensures both the recall rate of subtle fractures and effectively controls the false detection rate of rock wall texture noise. The median and median absolute deviation are used instead of the conventional mean and standard deviation because the median is inherently robust to outliers (i.e., potential fracture responses). Even when significant fractures exist within the annulus, the threshold estimation will not be skewed by the high response values ​​of the fractures, ensuring the stability and adaptability of the threshold.

[0083] S103, each candidate pixel with a positive small-scale difference value in any grayscale sequence is paired with each candidate pixel with a negative small-scale difference value within a preset angle change range under a preset angle change direction. The pairing confidence of any pair is determined based on the angle difference between the two pixels in any pair and the magnitude difference of the small-scale difference value. The valid pairing in any grayscale sequence is determined based on the pairing confidence.

[0084] After completing the "multi-scale response enhancement" step in the previous step, this step continues to complete the "morphological constraint pairing" step: an adaptive threshold is calculated using the median and the median absolute deviation, and its natural anti-outlier characteristic is used to ensure that the threshold is not contaminated by high response of the crack, thus screening out candidate high response points; then, positive and negative gradient morphological pairing is performed on the candidate points. Only when a pair of positive and negative gradients simultaneously meet the triple constraints of "opposite direction, similar amplitude, and spacing that conforms to the typical crack width" are they confirmed as the two edges of the same crack, and the crack center position and pairing confidence are output; finally, each ring outputs the crack center position set and pairing confidence set for use in the subsequent cross-ring weighted voting verification step, further realizing the crack identification from "single-point gradient detection" to "complete edge structure matching".

[0085] Specifically, the first The candidate pixels in the grayscale sequence of each ring are denoted as the candidate pixel set. traversal Candidate pixels for each positive gradient in (Right now This corresponds to the rising edge of grayscale from dark to bright, and is the preset angle change range behind its angle (that is, the clockwise angle change direction in polar coordinates, or the negative angle change direction). ( In essence, it refers to the number of pixels. Since different numbers of pixels along the center line of the ring correspond to different ranges of angle variation, the number of pixels at that point is essentially equivalent to the range of angle variation. Desirable This approach can cover the width distribution range of normal cracks while avoiding mismatches caused by over-searching. and (See below for the meaning) Search for candidate pixels with negative gradients. (Right now (corresponding to the falling edge of grayscale from bright to dark), thus determining the first Candidate pixel set in the grayscale sequence of each ring The matching pairs in the data.

[0086] Since the crack width has a typical width value, and the gradient change amplitude or small-scale difference value of the left and right edges of the real crack is approximately equal, the probability that the two pixels in any matching pair are pixels on the left and right edges of the real crack can be measured based on the difference between the angle difference of the two pixels in any matching pair and the typical crack width value, as well as the difference in the amplitude of the small-scale difference value corresponding to the two pixels. This allows for the determination of the pairing confidence of any matching pair, including:

[0087] The difference between the pixel distance corresponding to the angle difference between two pixels in any of the matching pairs and the typical width of the crack in the borehole wall of the hard roof of the coal mine is recorded as the first pairing characterization term.

[0088] The smaller of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair is divided by the sum of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair, and the result is recorded as the second pairing characterization term.

[0089] The pairing confidence of any matching pair is determined based on the first pairing characteristic and the second pairing characteristic, wherein the pairing confidence is inversely proportional to the first pairing characteristic and directly proportional to the second pairing characteristic.

[0090] Furthermore, as a preferred embodiment, the pairing confidence of any of the said matching pairs for:

[0091]

[0092] In the formula, To concentrate candidate pixels in the current grayscale sequence into candidate pixels with positive gradients Candidate pixels with negative gradient The pairing confidence score, which constitutes a matching pair, represents the confidence that the two candidate pixels belong to the two edges of the same crack, and its value ranges from [value missing]. ; Candidate pixels With candidate pixels The pixel distance between them is essentially equivalent to the angle difference; The typical width of the crack, i.e., the expected center of the crack width, is taken as a value. Pixels (corresponding to approximately the actual borehole wall thickness) The angle range covers the width of most cracks. Width tolerance, also known as the variance parameter characterizing how much the width is allowed to deviate from the desired center, takes a value of A tolerance of 2 pixels covers the width distribution range of most crack samples, while avoiding misjudging wide textures that are not cracks as cracks due to excessive tolerance (e.g., >4 pixels). and These represent the operations of taking the minimum value and the absolute value, respectively. The preferred value for the numerical stability constant is [value to be filled in]. To prevent the denominator from being zero.

[0093] The pairwise confidence score consists of a product of two terms: the first term is the Gaussian distance weight, which is the distance between the positive and negative gradients. Approximately the typical crack width The first term has a value close to 1, and decays according to a Gaussian function when it deviates. This reflects the physical constraint of the distance between the edges of the fracture. The edges of a real fracture cannot be touching (i.e., the distance is 0), nor can they be too far apart (the distance is much greater than the typical fracture width). The second term is a measure of gradient magnitude symmetry. When the small-scale difference magnitudes of the candidate pixels of positive and negative gradients are close, the value is close to 1, and it approaches 0 when the difference increases. This reflects the physical constraint of the contrast symmetry between the edges of the fracture. In a real fracture against a uniform rock wall background, the grayscale change amplitudes of the edges on both sides are basically the same. Only when both terms are satisfied simultaneously is a high confidence score assigned, effectively excluding the unilateral edge response of non-fractures.

[0094] when Greater than the paired confidence threshold (This value can be adaptively adjusted according to user needs; a larger value indicates a stricter testing standard, and a smaller value indicates a more lenient testing standard.) This is confirmed as a valid match pair, and the location of the undetermined fracture center corresponding to this valid match pair is... The calculation is as follows:

[0095]

[0096] In the formula, To effectively match pairs The corresponding position index (in pixels) of the undetermined fracture center in the grayscale sequence of the current ring is taken as the pixel index of the candidate pixel with the positive gradient. The pixel index of the candidate pixel with negative gradient The arithmetic mean of the values. The center location corresponds to the undetermined fracture in the [number]th [year]. Annular band The angular coordinates in the middle, that is, by This will determine the valid matching pairs. The corresponding undetermined fracture center angle will be used as the basis for subsequent fracture cross-ring tracking.

[0097] Therefore, we can obtain the first... Each valid matching pair in the gray-scale sequence of each ring corresponds to the undetermined fracture center angle, and forms the set of undetermined fracture center angles. and the The pairing confidence scores corresponding to each valid matching pair in the gray-scale sequence of each ring band constitute the pairing confidence score set. ,in For the first The total number of valid matching pairs corresponding to the grayscale sequence of each ring zone, and the undetermined crack represented by each valid matching pair carries dual information: its center angle. The precise coordinates of the undetermined fracture in the angular direction were recorded, and the paired confidence scores were... This quantifies the degree to which the undetermined crack satisfies the morphological constraints, providing a reliable prior measure for subsequent cross-ring zone verification.

[0098] S104, construct a standardized angle grid, determine the cross-ring support under each standardized angle based on the mapping value of the undetermined fracture center angle corresponding to each effective matching pair in each gray-scale sequence in the standardized angle grid and the pairing confidence, and determine the candidate standardized angle based on the cross-ring support.

[0099] The undetermined fractures corresponding to the valid matching pairs detected independently by each annulus come from two sources: one is the edge pairing response generated when a real fracture passes through the annulus. These undetermined fractures will appear in multiple radially adjacent annulus within a similar angular deviation range, forming a "fracture trajectory" extending in the radial direction; the other is false edge pairing caused by local texture changes or minor noise in the rock wall. These undetermined fractures often only appear in a single annulus and have no angular correspondence with other annulus. Existing technologies usually use a simple "continuous annulus counting" method for screening, judging a fracture as a fracture if more than a certain number of consecutive annulus appear. However, this method has two drawbacks: first, it does not consider the difference in confidence level of the detection results of each annulus, treating high-confidence and low-confidence detections equally; second, using a hard threshold (judging when a certain number is reached and discarding when the number is not reached) is prone to missing detections when the fracture is weak or false detections when there is concentrated noise. To address this, this embodiment introduces a weighted voting mechanism, using the paired confidence scores of each annular zone detection result as voting weights. Weighted support from adjacent annular zones is accumulated for each angular position, upgrading the crack judgment from "counting" to "weighted integration." Strong-response annular zones cast high-weight votes, while weak-response annular zones cast low-weight votes. Finally, a comprehensive confidence score is determined based on the product of the total weighted score and the radial coverage rate, which preserves the detection possibility of weak signal cracks while effectively suppressing the interference of isolated noise.

[0100] To eliminate the difference in angle sampling intervals between different annular zones due to their varying radii, a standardized angle grid is first established with a grid resolution of [resolution value missing]. Standardized angle grids map the angular positions of different rings to a common 0° to 360° coordinate system. Because the number of pixels on the inner and outer rings differs, a pixel in the inner ring corresponds to a larger angular step size, while a pixel in the outer ring corresponds to a smaller angular step size. If the center angle of a certain undetermined fracture in the inner ring is at pixel index 100, and the center angle of a certain undetermined fracture in the outer ring is also at pixel index 100, their corresponding actual azimuth angles are completely different. Therefore, they must first be uniformly converted to actual angle values ​​(e.g., 45°) before comparisons are made across rings.

[0101] After constructing the standardized angle grid, the undetermined fracture center angle corresponding to each valid matching pair in the gray-level sequence of each annulus is mapped onto the standardized angle grid, forming a set of mapping values ​​for the undetermined fracture center angle on the standardized angle grid. Then, based on the mapping value of the undetermined fracture center angle corresponding to each valid matching pair in each grayscale sequence on the normalized angle grid and the pairing confidence, the cross-ring support under each normalized angle is determined, including:

[0102]

[0103] in, From a standardization perspective The support of the transannular band below, This indicates the total number of the rings. Indicates the first The total number of valid matching pairs corresponding to each ring band. Indicates the first The first ring band The pairing confidence (as voting weight) of each of the valid matching pairs. Represents the Gaussian kernel function. , Indicates the first The first ring band The mapping value of the undetermined fracture center angle corresponding to each valid matching pair in the normalized angle grid. This represents the preset angle tolerance bandwidth, which is also the preset angle deviation range. In this embodiment, the preferred value is [value missing]. The purpose of this angular tolerance bandwidth is to allow the actual crack to drift slightly in the radial direction due to the slight curvature of the borehole wall or the eccentricity of the probe.

[0104] The Gaussian kernel serves to smoothly diffuse discrete votes within the angular neighborhood, allowing minor deviations caused by angular drift in adjacent zones of the same crack to converge at the same peak position in the voting space. Simultaneously, the limited bandwidth of the Gaussian kernel ensures that cracks with significantly different angles are not incorrectly merged. During the cross-zone support calculation, each undetermined crack detected in a zone casts a vote near its actual angular position, weighted by its paired confidence score. Higher paired confidence scores result in greater voting weights; the closer the angle is to the currently examined standardized angular position, the greater the voting contribution (controlled by the Gaussian kernel). Finally, the weighted support votes from each zone for each standardized angle θ are calculated.

[0105] In acquiring all standardized perspectives support of the trans-ring belt After that, A continuous curve can be formed on the angular coordinates, or in other words, a cross-ring support sequence can be constructed. Since a real fracture accumulates a local peak near its normalized angle, while noise does not form a significant peak, non-maximum suppression is used to extract the local peak of the cross-ring support sequence. The normalized angle corresponding to the local peak is recorded as the candidate normalized angle corresponding to the candidate fracture trajectory. , , This represents the number of candidate fracture trajectories. It is important to emphasize that the candidate fracture trajectories here are actually a further screening process towards the actual fractures based on the undetermined fractures mentioned above, so they are different from the undetermined fractures mentioned above and should not be confused.

[0106] S105, determine the true crack confidence of any candidate standardized angle based on the size of any candidate standardized angle and the number of valid matching pairs within the preset angle deviation range of any candidate standardized angle, and complete crack detection based on the true crack confidence of each candidate standardized angle.

[0107] For each candidate fracture trajectory, the corresponding candidate normalized angle Its total vote score This characterizes the total weighted support obtained by the candidate fracture trajectory across the annular zone. Further, the number of annular zones actually traversed by the candidate fracture trajectory is statistically analyzed. (i.e., in how many circumferences does it exist with...) The angle deviation is less than the preset angle deviation range The effective matching pairs, that is, limiting each ring to at most one effective matching pair contributing to the participation. To avoid redundant statistics that could lead to errors in the subsequent calculation of the true fracture confidence score, the statistical analysis is performed on the candidate fracture trajectory. :

[0108]

[0109] In the formula, For the first From the perspective of candidate standardization The true fracture confidence score of the corresponding candidate fracture trajectory, with a value range of [value range missing]. ; For the first From the perspective of candidate standardization Support of the transannular band below; The maximum cross-ring support value under all candidate standardization perspectives is used to normalize the voting score; For the first From the perspective of candidate standardization The number of rings crossed by the corresponding candidate fracture trajectory; This represents the total number of annular zones. The confidence level for this true fracture is composed of the product of two terms: the first term... To normalize the voting intensity, characterizing the relative strength of the fracture response relative to the strongest fracture in the entire image; the second term Radial coverage represents the completeness of the fracture's radial extension. Multiplying these two terms means that a true fracture must possess both "strong response" and "wide coverage." A response with only high intensity but confined to a single annulus (such as strong noise) results in a small second term, lowering the confidence level of a true fracture. Conversely, a response with only wide coverage but weak responses in each annulus results in a small first term, also lowering the confidence level of a true fracture. This dual constraint makes the robustness of determining the authenticity of a fracture far superior to a simple threshold counting method.

[0110] when At that time, the first From the perspective of candidate standardization The corresponding candidate trajectories were confirmed as valid fracture targets, among which To determine the threshold, this embodiment preferably sets it to a value of [value to be filled in]. (This value can be adaptively adjusted according to user needs. A larger value results in a stricter detection standard, retaining only crack targets with extremely high confidence in true cracks, suitable for scenarios with stringent accuracy requirements; a smaller value results in a more lenient detection standard, detecting more weak signal cracks, suitable for scenarios with high recall requirements.) The final result is a set of effective crack targets. , ,in This represents the total number of effective fracture targets, and each effective fracture target... Record the following attribute: fracture center angle Radial span range arrive (Representing the annular index from the first appearance of the fracture to its disappearance), overall confidence level and fracture type .

[0111] Regarding the type of fracture The method for determining is: when as well as Within the preset angle deviation range, the standard deviation between the corresponding center angles of the undetermined fractures of the effective matching pairs is less than the preset standard deviation threshold (preferred in this embodiment). ),and The corresponding second real fracture characterization term When the ratio is greater than the preset threshold for the proportion of annular zones (preferably 0.8 in this embodiment) (angle standard deviation < 3° to ensure that the angle position of the crack in each annular zone is consistent); The fracture must span at least 80% of the circumference, i.e., from the inner circle to the outer circle. Only if both conditions are met can it be determined as a radially penetrating fracture (to avoid misclassifying local circumferential fractures as radial fractures). Otherwise, it must be determined as a circumferential fracture.

[0112] After obtaining each valid fracture target, a local unfolding is performed on each valid fracture target. Using the angular center position of the current valid fracture target as the center, an angular width of 20° is taken (i.e., a 10° margin on each side, totaling 20°; the 10° margin on each side of the angular center position of the current valid fracture target is sufficient to cover all fractures and their edge transition zones). Using the radial starting and ending rings of the current valid fracture target as radial boundaries, a margin of two ring widths is extended outwards (i.e., extending outwards by two ring widths to ensure the fracture is completely contained within the unfolded area). All pixel data within this local area are extracted. A polar coordinate to Cartesian coordinate unfolding mapping is performed on this local area. The dimensions of the unfolded local rectangular image are: the number of pixels in the angular direction is obtained by dividing the angular width by the angular sampling step size; the pixel height in the radial direction is obtained by dividing the radial coverage area (the difference between the ending and starting rings plus the expansion margin multiplied by the ring width) by the radial sampling step size, where the radial sampling step size is one pixel. Bicubic interpolation is used during the unfolding process. However, due to the small unfolding area and the pre-determined position of the current effective fracture target, the impact of interpolation degradation on the final parameter measurement results is minimized. On the locally unfolded image, the angular and radial positions of the current effective fracture target in the annular domain are used as prior constraints to automatically measure the fine geometric parameters of the fracture: the width distribution curve of the fracture as a function of radius is measured along the radial direction (i.e., the change in pixel distance between the two edges of the fracture along the radial direction), and its average width and maximum width are calculated; the angular extension range of the fracture is statistically analyzed along the angular direction, and converted into the actual arc length by combining it with the borehole radius, where the actual arc length is equal to the average radius of the fracture location multiplied by the radian value corresponding to the angular extension range; the total area covered by the fracture pixels within the fracture region is statistically analyzed. Finally, a detection report for each fracture target is output, including the fracture type (radially penetrating fracture or circumferential fracture), angular center position (converted to the actual azimuth angle based on the borehole mark azimuth), radial coverage range (converted to the actual depth range along the borehole axis), average width, maximum width, extension arc length, area, and overall confidence level. Simultaneously, a visual annotation map of the locally unfolded image is generated, highlighting the crack boundary on the locally unfolded image for inspection personnel to review and confirm.

[0113] The embodiments of the present invention directly locate the real cracks based on the original grayscale annular image of the borehole wall of the hard roof of the coal mine. Then, based on the location results, only the real crack location is locally expanded and the parameters are precisely measured. This can avoid the crack identification error caused by information loss in the existing technology mode of first expanding the annular image and then identifying it, and significantly improve the detection accuracy of cracks in the borehole wall of the hard roof of the coal mine.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A visual inspection method for the early stage of fracture detection in coal mine hard roof working faces, characterized in that, The method includes: A grayscale annular image of the borehole wall of a hard roof in a coal mine is obtained. The annular image is divided into a preset number of annular bands along the radial direction. At the same position in each annular band, the grayscale value of each pixel is obtained from the same angle to obtain the grayscale sequence of the corresponding annular band. In any grayscale sequence, obtain the large-scale difference value and the small-scale difference value of each grayscale value. Based on the large-scale difference value and the sign difference between the large-scale difference value and the small-scale difference value, determine the crack response intensity of each grayscale value in any grayscale sequence. Based on the crack response intensity, determine the candidate pixel points in any grayscale sequence. Each candidate pixel with a positive small-scale difference value in any grayscale sequence is paired with each candidate pixel with a negative small-scale difference value within a preset angle change range under a preset angle change direction. The pairing confidence of any pair is determined based on the angle difference between the two pixels in any pair and the magnitude difference of the small-scale difference value. The effective pairing in any grayscale sequence is determined based on the pairing confidence. Construct a standardized angle grid, and determine the cross-ring support for each standardized angle based on the mapping value of the undetermined fracture center angle corresponding to each effective matching pair in each gray-scale sequence on the standardized angle grid and the pairing confidence. Then, determine the candidate standardized angle based on the cross-ring support. Based on the magnitude of any candidate standardized angle and the number of valid matching pairs within a preset angle deviation range of any candidate standardized angle, the true crack confidence of any candidate standardized angle is determined, and crack detection is completed based on the true crack confidence of each candidate standardized angle.

2. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1, characterized in that, The large-scale difference value is the step size difference value, and the small-scale difference value is the center difference value.

3. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1, characterized in that, Determining the crack response intensity of each gray value in any of the aforementioned gray-scale sequences includes: When the sign of the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence are different, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence shall be taken as the crack response intensity of any gray value in any gray-scale sequence. When the large-scale difference value and the small-scale difference value of any gray value in any gray-scale sequence have the same sign, the absolute value of the large-scale difference value of any gray value in any gray-scale sequence is amplified by a preset factor and used as the crack response intensity of any gray value in any gray-scale sequence.

4. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1 or 3, characterized in that, Determining candidate pixels in any of the grayscale sequences based on the crack response intensity includes: The crack response threshold of any grayscale sequence is determined based on the median and absolute deviation of the crack response intensity of all grayscale values ​​in any grayscale sequence, wherein the crack response threshold is proportional to both the median and the absolute deviation of the median. The pixel corresponding to the gray value in any gray-scale sequence whose crack response intensity is greater than the crack response threshold is recorded as a candidate pixel in any gray-scale sequence.

5. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1, characterized in that, The preset angle change direction refers to the clockwise angle change direction.

6. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1 or 5, characterized in that, Determining the pairing confidence of any of the said matching pairs includes: The difference between the pixel distance corresponding to the angle difference between two pixels in any of the matching pairs and the typical width of the crack in the borehole wall of the hard roof of the coal mine is recorded as the first pairing characterization term. The smaller of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair is divided by the sum of the absolute values ​​of the small-scale difference values ​​of the two pixels in any matching pair, and the result is recorded as the second pairing characterization term. The pairing confidence of any matching pair is determined based on the first pairing characteristic and the second pairing characteristic, wherein the pairing confidence is inversely proportional to the first pairing characteristic and directly proportional to the second pairing characteristic.

7. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1, characterized in that, Determine the crossband support under each standardized perspective, including: ， in, From a standardization perspective The support of the transannular band below, This indicates the total number of the rings. Indicates the first The total number of valid matching pairs corresponding to each ring band. Indicates the first The first ring band The pairing confidence of each of the valid matching pairs. Represents the Gaussian kernel function. Indicates the first The first ring band The mapping value of the undetermined fracture center angle corresponding to each valid matching pair in the normalized angle grid. This indicates the preset angle tolerance bandwidth.

8. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claims 1 to 7, characterized in that, Determining candidate normalized angles based on the cross-ring support includes: The cross-ring support under all standardized angles is used to form a cross-ring support sequence. Non-maximum suppression is used to extract the local peaks of the cross-ring support sequence, and the standardized angles corresponding to the local peaks are recorded as the candidate standardized angles.

9. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 1, characterized in that, Determining the true crack confidence of any of the candidate standardized angles includes: The ratio of the cross-ring support at any of the candidate normalized angles to the maximum cross-ring support at all the candidate normalized angles is denoted as the first true fracture characterization term. The ratio of the number of rings with valid matching pairs within a preset angle deviation range of any candidate standardized angle to the total number of rings is denoted as the second true fracture characterization term. The product of the first true fracture characterization term and the second true fracture characterization term is denoted as the true fracture confidence level for any of the candidate standardized angles.

10. The method for early-stage visual inspection of fractures in coal mine hard roof working faces according to claim 9, characterized in that, Crack detection is performed based on the true crack confidence level of each candidate standardized angle, including: When the confidence level of the true crack at any candidate standardized angle is greater than a preset true crack confidence threshold, it is considered that a true crack exists at any candidate standardized angle. If the standard deviation between the center angles of the undetermined cracks corresponding to all valid matching pairs within the preset angle deviation range of any candidate standardized angle and any candidate standardized angle is less than a preset standard deviation threshold, and the second true crack characterization term corresponding to any candidate standardized angle is greater than a preset circumferential proportion threshold, the true crack at any candidate standardized angle is determined to be a radially penetrating crack; otherwise, the true crack at any candidate standardized angle is determined to be a circumferential crack.

Images