A System and Method for Single-Pile Interface Positioning in Geotechnical Investigation Based on Borehole Wall Image Measurement

Abstract

This invention relates to the field of image measurement technology and discloses a system and method for locating the interface of a single pile in geotechnical exploration based on borehole wall image measurement. The method first acquires a panoramic image of the borehole inner wall cylindrical surface. After coordinate mapping and equalization processing, a set of closed polygonal contours is extracted through a two-level geometric validity cascade filtering, and further fitted to generate a set of polarized long axis segments. Subsequently, an analysis window sliding along the depth direction is defined. By constructing a histogram of the angular frequency distribution of polarized line segments and calculating the geometric envelope area between it and the uniformly distributed baseline, a depth variation curve reflecting the spatial distribution characteristics is generated, thereby calibrating the manifold distribution transition interface. This invention achieves quantitative identification of the interface between pile tip sediment and bearing layer, and the interface of the soil disturbance zone around the pile, by quantitatively characterizing the statistical laws of the spatial orientation of soil fragments. This provides core geometric control parameters for the accurate calculation of the bearing capacity of a single pile.

Description

Technical Field

[0001] This invention relates to the field of image measurement technology, and more specifically, to a system and method for locating the interface of a single pile in geotechnical exploration based on borehole wall image measurement. Background Technology

[0002] In the fields of geotechnical engineering investigation and pile foundation engineering, accurate assessment of the ultimate vertical bearing capacity of a single pile is a core prerequisite for ensuring the safety of the superstructure. With the popularization of borehole imaging technology, analyzing the characteristics of the bearing stratum at the pile tip, the thickness of sediment, and the degree of soil disturbance around the pile through borehole imaging has become an important auxiliary testing method. However, the complex drilling conditions (such as mud adhering to the borehole walls and non-uniform illumination) and the diversity of the soil and rock masses themselves make the extraction of objective and quantitative mechanical characteristic parameters from high-noise image data a key bottleneck in the current application of image processing technology in geotechnical engineering.

[0003] Existing technologies have explored various approaches to image recognition and region segmentation. For example, patent application CN114757276A discloses a deep learning-based flow field region recognition system and method. This system samples flow field data into regions and calculates the distance matrix of sampling point groups. It then uses a feature network to generate latent vectors representing flow patterns, thereby achieving the recognition of local flow field patterns. Patent CN115035130B discloses a level set image segmentation method, storage medium, and image processing terminal. This technology extracts various image features and initializes an evolution curve. It then uses the RPCA method to reduce the dimensionality of regions inside and outside the curve to extract feature vectors, thereby driving the curve to evolve to the target boundary. This aims to improve segmentation accuracy under sparse noise interference.

[0004] However, in the scenario of identifying sediment at the bottom of pile foundation holes and in-situ fractured rock, the aforementioned segmentation methods based on latent vector learning or grayscale / edge evolution face the problem of "feature entanglement." Because sediment and fractured rock at the weathering interface have extremely high geometric similarity in color, texture, and grayscale gradient, traditional methods struggle to distinguish their physical properties through local pixel features. Sediment, as a discrete body formed by free gravity dispersion, exhibits macroscopic disorder in the orientation of its internal fragments in three-dimensional space; while in-situ fractured rock, although subjected to tectonic stress fracturing, still macroscopically retains the parallel trend of bedding formed by geological deposition or a specific spatial manifold distribution. If this structural feature constituted by "spatial orientation consistency" cannot be captured, the algorithm will misjudge the disordered sediment as the bearing layer, resulting in an inflated pile tip embedment depth, which in turn leads to a calculated single pile bearing capacity far exceeding its actual bearing capacity, causing engineering risks such as excessive pile settlement or even structural instability. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a single-pile interface positioning system and method for geotechnical investigation based on borehole wall image measurement. By extracting the polygonal polarized long axis segments of discrete fragments in the borehole image, constructing an angular frequency distribution histogram using a sliding window, and calculating the geometric envelope area between the histogram and the uniformly distributed baseline, a quantitative characterization of the spatial manifold consistency of the geotechnical mass is achieved. This scheme can effectively distinguish, from a statistical perspective, spatially disordered sediment from bearing strata with preserved original bedding orientation, eliminating interpretation ambiguities caused by image grayscale and texture similarity, and improving the accuracy of acquiring key geometric control parameters for pile foundation engineering and the physical interpretability of test results.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement includes:

[0008] A panoramic image of the cylindrical surface inside the borehole is acquired and the cylindrical surface is unfolded, mapped, and equalized to generate an equalized rectangular unfolded image. Through a two-level cascaded filtering of geometric validity, a set of effective closed polygon contours is extracted from the equalized rectangular unfolded image. The set of effective closed polygon contours is then subjected to minimum bounding rectangle fitting and major-minor axis ratio filtering to generate a set of polarized major axis segments.

[0009] Define an analysis window that slides along the depth direction. Calculate the absolute angle between each polarization long axis segment within the analysis window and the horizontal baseline from the set of polarization long axis segments, and construct an angle frequency distribution histogram. Establish a uniformly distributed baseline horizontal line. Calculate the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line, generate a depth variation curve of the geometric envelope area, and mark the manifold distribution transition interface on the depth variation curve of the geometric envelope area.

[0010] The method for performing cylindrical expansion coordinate mapping includes:

[0011] With the borehole axis as the rotation axis, the unfolding radius is the radial distance from the optical center of the camera device to the borehole wall. Each pixel in the cylindrical panoramic image has two coordinate dimensions: a circumferential angle component and a depth component. Multiplying the circumferential angle component by the unfolding radius yields the horizontal coordinate of the pixel on the rectangular unfolded image. The depth component is directly mapped to the vertical coordinate on the rectangular unfolded image. After completing the coordinate transformation pixel by pixel, the rectangular unfolded image is generated.

[0012] The two-level geometric validity cascaded screening includes a first-level area screening and a second-level boundary integrity screening.

[0013] The method for extracting a set of valid closed polygon contours from an equalized rectangular unfolded image includes:

[0014] Edge detection and morphological closure operations are performed on the equalized rectangular unfolded map to generate a binary edge contour map. Closed polygon contours are extracted from the binary edge contour map. First-level area filtering and second-level boundary integrity filtering are performed on the closed polygon contours to generate a set of valid closed polygon contours.

[0015] The method for performing the area filtering is as follows:

[0016] Calculate the area of ​​the closed polygon contour, compare the contour area with a preset minimum area threshold, delete closed polygon contours whose contour area is less than or equal to the minimum area threshold, and retain closed polygon contours whose contour area is greater than the minimum area threshold as contours that pass the area screening.

[0017] The execution method for the boundary integrity screening is as follows:

[0018] The area filtering is determined by the intersection relationship between the outline and the boundary of the equalized rectangular unfolded diagram. Closed polygon outlines that intersect with any one of the four boundaries of the equalized rectangular unfolded diagram are deleted, and areas that do not intersect with the boundary of the equalized rectangular unfolded diagram are filtered by the outline and marked as valid closed polygon outlines.

[0019] The method for generating the set of polarized long axis segments includes:

[0020] Calculate the ratio of the major axis to the minor axis of each minimum bounding rectangle. For the minimum bounding rectangle whose ratio is greater than the threshold, extract the line connecting the midpoints of the two short sides as the polarization major axis segment. The set of polarization major axis segments is composed of all polarization major axis segments.

[0021] The method for constructing the angular frequency distribution histogram includes:

[0022] Extract the start and end coordinates of each polarization long axis segment from the set of polarization long axis segments one by one, calculate the absolute angle between the polarization long axis segment and the horizontal baseline based on the start and end coordinates, and form an absolute angle set from all the absolute angles in the current analysis window;

[0023] The absolute angles in the set of absolute angles are grouped according to the angle step from 0° to 90°. The number of polarization major axis segments falling into each angle interval is counted as the frequency value. The frequency values ​​of all angle intervals are used to construct an angle frequency distribution histogram.

[0024] The method for generating the set of absolute included angles includes:

[0025] The horizontal projection length is obtained by subtracting the horizontal component of the starting point coordinate from the horizontal component of the endpoint coordinate of the polarization long axis segment, and the vertical projection length is obtained by subtracting the vertical component of the starting point coordinate from the vertical component of the endpoint coordinate. The arctangent function value of the angle, which is the ratio of the absolute value of the vertical projection length to the absolute value of the horizontal projection length, is used as the absolute angle between the polarization long axis segment and the horizontal baseline.

[0026] The total number of polarized long axis segments within the current analysis window is counted. The total number of polarized long axis segments is compared with a preset minimum segment number threshold. When the total number of polarized long axis segments is greater than the minimum segment number threshold, the set of absolute angles is formed by all absolute angles within the current analysis window.

[0027] The method for calculating the geometric envelope area includes:

[0028] Divide the total number of polarization long axis segments by the total number of angle intervals to obtain the uniformly distributed reference frequency value. Plot a horizontal line covering 0° to 90° with the uniformly distributed reference frequency value as the ordinate as the uniformly distributed reference horizontal line. Calculate the absolute value of the difference between the frequency value of each angle interval and the uniformly distributed reference frequency value. Multiply the absolute value of the difference of each angle interval by the angle step size to obtain the local area element. Summate all the local area elements to obtain the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed reference horizontal line.

[0029] The method for generating the depth variation curve of the geometric envelope area includes:

[0030] The depth span and sliding step size of the analysis window are set. Polarized long axis segments whose midpoint longitudinal coordinates fall within the depth range of the current analysis window are extracted from the set of polarized long axis segments to form a subset of local polarized long axis segments. The midpoint of the polarized long axis segment refers to the point corresponding to the average coordinate of its start and end points. The geometric envelope area of ​​the current analysis window position is obtained based on the subset of local polarized long axis segments. The analysis window is gradually slid along the depth direction according to the sliding step size and the calculation is repeated until the traversal is completed. The depth variation curve of the geometric envelope area is generated by arranging the geometric envelope area values ​​of all analysis window positions according to the depth coordinates.

[0031] The method for calibrating the manifold distribution transition interface on the depth variation curve of the geometric envelope area includes:

[0032] Step detection is performed by scanning the depth change curve of the geometric envelope area along the depth direction. The change of manifold distribution type is determined by jointly judging the step transition amplitude threshold and the continuous thickness threshold. The depth coordinates that meet the joint judgment conditions are calibrated as the depth coordinates of the manifold distribution transition interface.

[0033] A single-pile interface positioning system for geotechnical investigation based on borehole wall image measurement is used to implement the aforementioned single-pile interface positioning method for geotechnical investigation based on borehole wall image measurement. The system includes:

[0034] Polarization line segment generation module: used to acquire panoramic images of the cylindrical surface inside the borehole and perform coordinate mapping and equalization processing to generate an equalized rectangular unfolded image. Through two-level geometric validity cascade filtering, it extracts a set of valid closed polygon contours from the equalized rectangular unfolded image. It then performs minimum bounding rectangle fitting and major-minor axis ratio filtering on the set of valid closed polygon contours to generate a set of polarized major axis segments.

[0035] Manifold Interface Calibration Module: This module defines an analysis window that slides along the depth direction. It calculates the absolute angle between each polarization long axis segment within the analysis window and the horizontal baseline from the set of polarization long axis segments and constructs an angle frequency distribution histogram. It establishes a uniformly distributed baseline horizontal line, calculates the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line, generates a depth variation curve of the geometric envelope area, and calibrates the manifold distribution transition interface on the depth variation curve of the geometric envelope area.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] This method transforms the grayscale features of two-dimensional borehole walls into polarized geometric statistical features reflecting the spatial distribution of soil and rock masses, overcoming the limitations of relying on color and texture for interface identification in soil and rock exploration, which is susceptible to mud contamination and uneven lighting. By extracting effective closed polygons and constructing a set of polarized long axis segments, unstructured noise on the borehole walls can be effectively eliminated, accurately restoring the spatial arrangement framework of soil and rock fragments around the borehole. By analyzing the geometric envelope area between the frequency distribution of angles within the analysis window and the baseline of uniform distribution, the spatial manifold consistency of discrete polygons within the image is quantitatively described, revealing the essential difference in spatial orientation statistical features between disordered pile bottom sediment (or disturbed soil) and bearing strata with original geological bedding. This feature discrimination model based on statistical distribution achieves objective calibration of manifold distribution transition interfaces, providing precise physical basis for determining the thickness of the effective bearing stratum at the pile tip resistance, controlling the thickness of pile bottom sediment, and stratified reduction of pile side friction in the calculation of single pile bearing capacity, eliminating subjective errors from manual experience interpretation. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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 flowchart of a method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement, provided in an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of a panoramic image acquisition scenario of the borehole inner wall cylindrical surface provided in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the coordinate mapping of a cylindrical surface unfolding according to an embodiment of the present invention;

[0042] Figure 4 This is a flowchart of two-level screening of closed polygonal contours and extraction of polarization long axis segments provided in an embodiment of the present invention;

[0043] Figure 5 A comparative illustration of disordered and ordered manifold distributions provided for embodiments of the present invention;

[0044] Figure 6 An angular frequency distribution histogram of a disordered manifold distribution region provided in an embodiment of the present invention;

[0045] Figure 7 An angular frequency distribution histogram of the ordered manifold distribution region provided in this embodiment of the invention;

[0046] Figure 8 This is a functional module diagram of a single-pile interface positioning system for geotechnical investigation based on borehole wall image measurement, provided in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1:

[0049] Please see Figure 1 As shown, this embodiment provides a method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement, including:

[0050] Step S10: Acquire panoramic images of the cylindrical surface inside the borehole and perform coordinate mapping and equalization processing to generate an equalized rectangular unfolded image. Through two-level geometric validity cascade filtering, extract a set of valid closed polygon contours from the equalized rectangular unfolded image. Perform minimum bounding rectangle fitting and major-minor axis ratio filtering on the set of valid closed polygon contours to generate a set of polarized major axis segments.

[0051] Specifically, step S10 starts with a panoramic image of the cylindrical surface of the borehole wall acquired by the borehole camera along the borehole depth direction. This image is acquired after the single-pile drilling operation in geotechnical investigation. The process involves sequentially performing cylindrical surface unfolding coordinate mapping, adaptive local grayscale equalization processing, edge detection and morphological closure operation, two-level geometric validity cascade filtering, and minimum bounding rectangle fitting and major-minor axis ratio filtering, ultimately generating a set of polarized major axis segments. The panoramic image of the borehole wall cylindrical surface refers to a continuous image recorded by a panoramic imaging array camera sensor after the borehole camera, with the borehole axis as the optical axis center and a ring-shaped illumination source providing circumferential full coverage of the cylindrical surface of the borehole wall, provides this image. The horizontal dimension of the image corresponds to the circumferential angular position of the borehole wall cylindrical surface, and the vertical dimension corresponds to the depth position of the borehole. The polarization long axis segment is the center line of the long side extracted from the smallest bounding rectangle of the closed polygonal contour of the soil and rock fragments. The center line of the long side refers to the line connecting the midpoints of the two short sides of the smallest bounding rectangle. "Polarization" represents the direction of the line segment, which carries the spatial orientation information of the soil and rock fragments within the borehole wall, distinguishing it from scalar features such as area and perimeter that do not have directionality. The direction of the line segment is quantified based on the horizontal baseline (the direction parallel to the horizontal coordinate axis in the rectangular unfolded coordinate system, corresponding to the circumferential direction of the borehole wall). Subsequently, the spatial orientation of the soil and rock fragments is represented by the absolute angle between the line segment and the horizontal baseline. The set of polarization long axis segments output in step S10 serves as the sole geometric data source for the construction of the angle frequency distribution histogram and the calculation of the geometric envelope area in step S20, ensuring that all subsequent statistical calculations are based on the spatial orientation unfolding of the soil and rock fragments rather than relying on pixel grayscale values ​​or color features.

[0052] Further, step S10 includes:

[0053] Step S11: Acquire panoramic images of the cylindrical surface of the borehole inner wall along the borehole depth direction, perform cylindrical surface unfolding coordinate mapping on the panoramic images of the cylindrical surface of the borehole inner wall to generate a rectangular unfolded image, and perform adaptive local grayscale equalization processing on the rectangular unfolded image to generate an equalized rectangular unfolded image.

[0054] Specifically, see Figure 2 This is a schematic diagram of a panoramic image acquisition scene of the borehole inner wall cylindrical surface provided in the embodiments of this application. Figure 2The diagram illustrates the descent path of the borehole camera along the borehole axis, the borehole depth direction, and the operational mode of acquiring circumferential, full-coverage images with the optical sensor facing the borehole wall, centered on the borehole axis. The panoramic image of the cylindrical surface of the borehole wall is acquired by the borehole camera installed inside the borehole. The borehole camera is continuously lowered along the borehole axis, and its optical sensor continuously acquires image data facing the borehole wall surface, outputting a panoramic cylindrical image data stream with circumferential angle and axial depth as coordinate dimensions. The panoramic cylindrical image covers the surface morphology information of the borehole wall within a 360° range, including both the original image data of the disordered manifold distribution area of ​​discrete polygons within the field of view and the original image data of the ordered manifold distribution area of ​​discrete polygons within the field of view. The disordered manifold distribution region of discrete polygons within the field of view refers to the region of loose fragments accumulated by gravity free fall. In a single pile hole, this typically corresponds to sediment at the bottom of the hole or severely disturbed sidewall soil. There is no directional continuity between the fragments, and their spatial orientation is randomly distributed in three-dimensional space. This is referred to as the disordered manifold distribution region. The ordered manifold distribution region of discrete polygons within the field of view refers to the in-situ fractured rock mass region that, although broken, still retains the original sedimentary bedding orientation characteristics. This typically corresponds to the effective soil and rock bearing layer of a single pile. The long axis of the fragments shows a statistically significant concentration trend along the bedding plane. This is referred to as the ordered manifold distribution region. Both types of manifold distribution regions present a grayish-white fragmented appearance in terms of pixel grayscale values ​​and color characteristics. It is difficult for the human eye and algorithms based on pixel color gradients to directly distinguish between the two from a two-dimensional image.

[0055] See Figure 3 This is a schematic diagram of the cylindrical surface unfolding coordinate mapping provided in the embodiments of this application. Figure 3The diagram illustrates the transformation process from a 3D cylindrical panoramic image to a 2D rectangular unfolded image. It clarifies that the borehole depth dimension of the cylindrical panoramic image is mapped to the longitudinal coordinate of the rectangular unfolded image, and the circumferential position dimension is mapped to the lateral coordinate of the rectangular unfolded image. The cylindrical unfolded coordinate mapping restores the cylindrical panoramic image from a 3D cylindrical projection to a 2D equidistant plane: with the borehole axis as the rotation axis, the unfolding radius is taken as the radial distance from the optical center of the camera device to the borehole wall. Each pixel on the cylindrical panoramic image has two coordinate dimensions: a circumferential angle component and a depth component. The circumferential angle component, in radians, is multiplied by the unfolding radius to obtain the horizontal coordinate of the pixel on the rectangular unfolded image. The depth component is directly mapped to the longitudinal coordinate on the rectangular unfolded image. After completing the coordinate transformation pixel by pixel, a rectangular unfolded image is generated, which represents the circumferential position of the borehole wall in the horizontal direction and the borehole depth in the longitudinal direction. The rectangular unfolded diagram uses a fixed coordinate system: the origin of the coordinate system is the upper left corner of the rectangular unfolded diagram; the horizontal coordinate axis is positive to the right, corresponding to the direction of increasing circumferential angle of the borehole wall; the vertical coordinate axis is positive downwards, corresponding to the direction of increasing borehole depth; the upper boundary of the rectangular unfolded diagram corresponds to the starting position of the shallow borehole, and the lower boundary corresponds to the ending position of the deep borehole; the subsequent equalized rectangular unfolded diagram shares the same coordinate system and boundary range as the rectangular unfolded diagram. The unfolding radius can be obtained in two ways: one is by approximating it by subtracting half of the outer diameter of the camera device from half of the borehole design diameter; the other is by real-time measurement using the distance sensor built into the camera device. The reason for using cylindrical unfolded coordinate mapping instead of performing subsequent analysis directly on the cylindrical image is that the discrete polygons located on both sides of the optical axis in the cylindrical panoramic image undergo circumferential compression due to the tilt angle between the imaging optical path and the borehole wall surface. There is an angular deviation related to the circumferential position between the projected contour of the same discrete polygon in the cylindrical image and its actual geometric contour on the borehole wall surface. The deviation increases with the circumferential distance of the discrete polygon from the center of the optical axis. If we skip the cylindrical expansion coordinate mapping and directly extract the closed polygon contour from the cylindrical image and perform minimum bounding rectangle fitting, the long side direction of the fitted rectangle will deviate from the true spatial orientation of the discrete polygon. This angular deviation is directly transmitted to the angle frequency distribution histogram when calculating the absolute angle in step S21, causing the value of the geometric envelope area in step S22 to deviate from the true statistical characteristics of the fragment orientation distribution. In the ordered manifold distribution region, a false directional dispersion effect is generated, which lowers the frequency peak. In the disordered manifold distribution region, a false directional concentration effect is generated, which causes a false peak in the histogram. The difference in geometric envelope area between the two types of manifold distribution regions is compressed, and the calibration position of the manifold distribution transition interface in step S23 is thus systematically shifted.After the cylindrical unfolding coordinate mapping, the outline shape of each discrete polygon in the rectangular unfolding diagram maintains equidistant geometric consistency with its real spatial shape on the hole wall surface. The polarization long axis segment direction extracted in step S13 can accurately reflect the real spatial orientation of the fragment on the hole wall, providing a distortion-free geometric coordinate basis for angle statistics and geometric envelope area calculation in step S20.

[0056] Adaptive local grayscale equalization divides the rectangular unfolded image into several equal-height local regions along the vertical direction. Within each local region, grayscale histogram equalization is performed independently, remapping the grayscale values ​​within that region to the full grayscale range. The division height of the local regions is determined based on the illumination characteristics of the light source within the aperture: the light source is fixedly installed at the end of the camera device, and the illumination intensity decreases with increasing depth, resulting in a grayscale gradient from bright to dark along the vertical direction of the rectangular unfolded image. The principle for selecting the division height is to ensure that the variation in illumination intensity within each local region is sufficiently small, so that the effect of grayscale histogram equalization within the region approximates contrast stretching under uniform illumination conditions. For example, when the depth span corresponding to the total vertical length of the rectangular unfolded image is H, the division height of the local regions can be set to a value between 1 / 8 and 1 / 12 of H. Global grayscale histogram equalization (referred to as global equalization) performs grayscale remapping based on the grayscale histogram of the entire rectangular unfolded image. In overexposed areas near the light source, it disperses pixel values ​​densely distributed in the high grayscale range to the full grayscale range, causing the already subtle grayscale differences between the discrete polygon edges and the background area to be further compressed, resulting in the loss of contour details. In darker areas far from the light source, it overstretches pixel values ​​in the low grayscale range, amplifying noise in dark areas and causing a large number of false edge responses in step S12. The aforementioned adaptive local grayscale equalization processing (referred to as partitioned local equalization) ensures that the grayscale differences between the discrete polygon contours and the background in each depth local region are independently stretched to a level that can be recognized by the edge detection operator, ensuring that step S12 obtains uniform contour extraction quality across the entire depth range. The equalized rectangular unfolded map has a uniform contrast distribution across the entire depth range. The edge detection operator can obtain a uniform edge contour output by using the same gradient threshold at different depth positions. This avoids the chain reaction problem of missing discrete polygon contours in dark areas due to insufficient contrast, resulting in insufficient polarization long axis segments, which would cause the corresponding depth area in step S21 to be marked as a region with insufficient statistical samples, and thus cause unusable broken intervals to appear on the depth change curve of the geometric envelope area in step S23.

[0057] Step S12: Perform edge detection and morphological closure operations on the equalized rectangular unfolded map to generate a binary edge contour map. Extract closed polygon contours from the binary edge contour map. Perform first-level area filtering and second-level boundary integrity filtering on the closed polygon contours to generate a set of valid closed polygon contours.

[0058] The method for area filtering is as follows: calculate the contour area of ​​the closed polygon contour, compare the contour area with the preset minimum area threshold, delete the closed polygon contours whose contour area is less than or equal to the minimum area threshold, and retain the closed polygon contours whose contour area is greater than the minimum area threshold as the contours that pass the area filtering.

[0059] The execution method for boundary integrity filtering is as follows: determine the intersection relationship between the area filtering profile and the boundary of the equalized rectangular unfolded diagram, delete the closed polygon profile that intersects with any one of the four boundaries of the rectangular unfolded diagram, and mark the area filtering profile that does not intersect with the boundary of the rectangular unfolded diagram as a valid closed polygon profile.

[0060] In step S12, see Figure 4The edge detection operator traverses every pixel position in the equalized rectangular unfolded image, calculates the gradient components of grayscale values ​​along the horizontal and vertical directions, and compares the magnitude of the gradient components with a preset gradient threshold. Pixel positions with gradient magnitudes greater than the threshold are assigned white and marked as edge pixels, while other pixel positions are assigned black and marked as non-edge pixels. The binarized labels of all pixels constitute a binary edge contour map. The gradient threshold is determined based on the statistical distribution of grayscale gradients in the equalized rectangular unfolded image. For example, the gradient value corresponding to 85% to 90% of the cumulative distribution function of the grayscale gradient magnitudes of all pixels in the entire equalized rectangular unfolded image, sorted from smallest to largest, can be selected as the gradient threshold. This ensures that edge pixels cover the contour boundaries of most discrete polygons without mislabeling excessive texture details as edges. Edge segments in the binary edge contour map may have break points due to interference from surface cracks, mud residue, or local shadows. Morphological closure operations are used to repair these breaks. The morphological closure operation described is a standard closure operation in mathematical morphology within the field of digital image processing. It involves performing a sequence of dilation followed by erosion on a binary edge contour map using a structuring element of a preset size. The dilation operation is a standard mathematical morphological dilation operation, causing each edge pixel to expand into its neighborhood, with the edge endpoints on either side of the break contacting each other after dilation to form a connection. The erosion operation is a standard mathematical morphological erosion operation, subsequently recovering the excess pixels generated by the dilation, restoring the edge width to its pre-dilation level, while preserving the connecting bridges formed at the break. The size of the structuring element determines the maximum width of the break that can be bridged: if the size is too small, it cannot bridge wider break gaps; if the size is too large, it will incorrectly merge adjacent independent discrete polygon contours into a single contour, resulting in shape distortion. The method for determining the size of the structuring element is to statistically analyze the width distribution of typical break gaps on discrete polygon contours on several representative equalized rectangular unfolded images, selecting a width value that can cover more than 90% of the break gaps. For example, the structural element may be a square with a side length of 5% to 10% of the average equivalent diameter of the discrete polygon; the average equivalent diameter of the discrete polygon refers to the arithmetic mean of the diameters of circles whose outlines are equal to the area of ​​all closed polygons in the current equalized rectangular unfolded diagram.

[0061] After morphological closure operation, connected component labeling operation is performed on the binarized edge contour map: starting from the starting pixel position at the upper left corner of the equalized rectangular unfolded map, the map is scanned row by row and column by column. When an unlabeled white edge pixel is scanned, it is used as the seed point to recursively search all connected white edge pixels along the eight-neighbor direction, and all connected pixels are assigned the same connected component number. When the Euclidean distance between the starting point coordinates and the ending point coordinates of the ordered sequence of edge pixels of the connected component is less than or equal to 1 pixel, it is determined that the curve formed by the edge pixels of the connected component is connected end to end in space to form a closed loop. The area enclosed by the closed loop is identified as a closed polygon contour. The ordered sequence of all pixel coordinates on the edge curve in a clockwise direction is used as the geometric description of the closed polygon contour.

[0062] All closed polygonal contours extracted from the binarized edge contour map are then subjected to two levels of geometric validity filtering. Area filtering serves as the first level, calculating the pixel area of ​​the region enclosed by each closed polygonal contour and comparing it to a minimum area threshold. Closed polygonal contours with areas not exceeding the minimum area threshold are deleted. Closed regions with areas below the minimum area threshold originate from tiny closed regions generated by mud splash spots, bubble traces, or image noise. These regions can be several to tens of times more numerous than the effective discrete polygons, with approximately equiaxial shapes and completely random major axis directions, carrying no directional information related to the spatial orientation of fragments. If these tiny regions are retained until step S13 to extract the polarization major axis segments, a large number of randomly oriented short line segments will make the frequency values ​​of the angle frequency distribution histogram tend to be uniform across angle intervals in step S21. The frequency peaks originally concentrated in specific angle intervals in the ordered manifold distribution region are diluted and lowered by random line segments. Therefore, the difference in geometric envelope area between the two types of manifold distribution regions in step S22 is reduced, and the detection sensitivity of the manifold distribution transition interface in step S23 is consequently reduced. The minimum area threshold is determined as follows: The lower limit distribution of the area of ​​discrete polygons with directional analysis value is statistically analyzed on several representative equalized rectangular unfolded maps. The area value that retains more than 95% of the effective discrete polygons is selected as the minimum area threshold. For example, the minimum area threshold can be set to 1 / 10000 to 5 / 10000 of the total area of ​​the rectangular unfolded map. Boundary integrity screening serves as the second level of screening. It determines whether each area screening profile intersects with any of the four boundaries of the equalized rectangular unfolded map. Profiles intersecting with the boundary are deleted, while non-intersecting profiles are marked as valid closed polygon profiles. Intersecting with the boundary means that the corresponding discrete polygon only partially appears within the field of view of the current equalized rectangular unfolded map; the complete profile shape is unavailable. When performing minimum bounding rectangle fitting on incomplete profiles, the long side direction of the rectangle only reflects the shape characteristics of the visible part of the discrete polygon, not the true spatial orientation of the complete shape. The resulting directional error is a deterministic error related to the clipping position and cannot be statistically eliminated by increasing the sample size. The logic of the two-stage filtering performed in series is as follows: Area filtering is performed first, quickly eliminating a large number of tiny noise regions with extremely low computational cost through pixel-by-pixel accumulation, thus compressing the number of computational objects for subsequent boundary integrity filtering to a small proportion of the total number of original closed polygon contours; Boundary integrity filtering is performed second, applying the reduced number of area filtering operations one contour at a time with equalized rectangle expansion. Figure 4 The geometric intersection of the boundaries is determined, and incomplete contours caused by field clipping are eliminated. Each contour in the set of valid closed polygon contours retained after two levels of screening has a complete closed geometry and an area sufficient to carry meaningful directional information, providing reliable and sized input data for the minimum bounding rectangle fitting in step S13.

[0063] Step S13: See below for further details. Figure 4 Traverse each valid closed polygon contour in the set of valid closed polygon contours, perform minimum bounding rectangle fitting, calculate the major-minor axis ratio of each minimum bounding rectangle, extract the line connecting the midpoints of the two short sides of the minimum bounding rectangle with a major-minor axis ratio greater than the major-minor axis ratio threshold as the polarization major axis segment, and form the polarization major axis segment set by all polarization major axis segments.

[0064] Traverse all edges of each valid closed polygon contour and calculate the area of ​​the bounding rectangle that completely encloses all vertices of the valid closed polygon contour using the direction of each edge as a candidate direction. Select the bounding rectangle with the smallest area as the minimum bounding rectangle of the valid closed polygon contour. In the minimum bounding rectangle, the pair of opposite sides with the larger length is defined as the long side and the pair of opposite sides with the smaller length is defined as the short side. Divide the length of the long side of the minimum bounding rectangle by the length of the short side to obtain the ratio of the major axis to the minor axis.

[0065] In step S13, the process of fitting the minimum bounding rectangle is as follows: traverse all edges of the effective closed polygon contour, determine the direction of the line containing each edge as a side direction of the candidate rectangle, calculate the area of ​​the bounding rectangle that exactly encloses all vertices of the effective closed polygon contour in the candidate direction and its perpendicular direction, and select the bounding rectangle with the smallest area after traversing all edge directions. The minimum bounding rectangle finds the rectangle frame that fits the effective closed polygon contour most closely among all possible rotation angles. The consistency between the long side direction of the rectangle and the overall extension direction of the discrete polygon reaches the highest level under the area minimization constraint. Extract the length of the long side and the length of the short side of the minimum bounding rectangle, and divide the length of the long side by the length of the short side to obtain the ratio of the major axis to the minor axis. The ratio of the major axis to the minor axis quantifies the degree to which the shape of the discrete polygon deviates from the equilateral shape: the larger the ratio, the more elongated the discrete polygon is and the more obvious the directional advantage; the closer the ratio is to 1, the closer the discrete polygon is to a square or circle and there is no obvious directional advantage. The threshold for the ratio of major to minor axes is set based on the following: The direction of the long side of the minimum bounding rectangle of an equiaxed discrete polygon is extremely sensitive to tiny local protrusions or depressions on the contour. Any slight fluctuation on the contour boundary may change the rotation angle of the minimum bounding rectangle. The polarization major axis segment direction of the same equiaxed fragment may differ by tens of degrees in images acquired under different shooting angles or lighting conditions. This directional instability caused by the near-equiaxed shape exists in both disordered and ordered manifold distribution regions and has a similar degree of influence. Including the random directional polarization major axis segment of the equiaxed fragment in the angle statistics of step S21 is equivalent to injecting a set of uniformly distributed noise frequencies into the angle frequency distribution histograms of the two types of manifold distribution regions. The originally prominent directional concentration peak in the ordered manifold distribution region is lowered, while the originally flat uniform distribution in the disordered manifold distribution region remains unchanged, but the difference in the shape of the two types of histograms is reduced. Therefore, the difference in the geometric envelope area of ​​the two types of manifold distribution regions in step S22 is weakened. The method for obtaining the major-minor axis ratio threshold is as follows: Select several sets of calibration images with known manifold distribution types, perform major-minor axis ratio filtering under different candidate thresholds, and calculate the difference in geometric envelope area between the two types of manifold distribution regions. The candidate threshold that makes the difference in geometric envelope area reach a stable level is determined as the major-minor axis ratio threshold. For example, the major-minor axis ratio threshold can be a value between 1.5 and 2.0. Valid closed polygon contours with a major-minor axis ratio greater than the major-minor axis ratio threshold are determined to be elongated discrete polygons with a clear spatial orientation dominance. The line connecting the midpoints of the two short sides of its smallest bounding rectangle is extracted as the polarization major axis segment, and the two endpoints of the line can be arbitrarily marked as the start and end points.The polarization long axis segment compresses the two-dimensional contour shape information of a discrete polygon into a line segment with a definite direction and position. The direction of the line segment represents the dominant spatial orientation direction of the discrete polygon, and the vertical coordinate of the line segment in the rectangular unfolded diagram represents the drilling depth position of the corresponding discrete polygon. "Polarization" describes the purification process of extracting unique directional features from irregular two-dimensional contours. Valid closed polygon contours with a major-minor axis ratio not greater than the major-minor axis ratio threshold are excluded from the polarization long axis segment set. After removing isometric fragments, only long strip-shaped discrete polygon information with stable direction indication is retained in the polarization long axis segment set. Each line segment in the polarization long axis segment set filtered by major-minor axis ratio carries two-dimensional data of direction and position information: the direction information is extracted in step S21 as the absolute angle between the line segment and the horizontal baseline and input into the angle frequency distribution histogram construction process; the position information is used in step S23 to determine the analysis window range to which each polarization long axis segment belongs.

[0066] Step S10 transforms the identification dimension of highly similar two-dimensional appearance regions in the borehole inner wall image from the color texture domain to the geometric shape domain. Step S10 does not rely on any pixel color information; instead, it uses the direction of the major axis of the smallest bounding rectangle of each discrete polygon closed contour as the feature descriptor. This transforms the identification task, which originally required differentiation in color texture space, into a geometric analysis task to determine whether there are macroscopic directional patterns in the spatial orientation of soil and rock fragments. This domain transformation ensures that the angle frequency distribution histogram and geometric envelope area calculations in step S20 obtain pure geometric input data, which no longer contains color information, effectively overcoming color similarity interference caused by mud contamination and uneven lighting at the soil and rock exploration site. The sub-steps in step S10 form an inseparable chain dependency: the cylindrical unfolding coordinate mapping in step S11 provides a two-dimensional planar image input for step S12 to eliminate perspective distortion; the absence of step S11 will cause the closed polygon contour extracted in step S12 to deviate from the true shape, resulting in an incorrect major axis direction in the minimum bounding rectangle fitting in step S13; the adaptive local gray-level equalization in step S11 provides a uniform contrast image input for step S12 across the entire depth range; the absence of equalization will result in insufficient extraction of dark area contours, causing the number of polarized major axis segments in the corresponding depth region in step S21 to fall below the minimum segment number threshold and be marked as a region with insufficient statistical samples, thus affecting the depth variation of the geometric envelope area. The curve breaks in the dark area, and step S23 cannot perform step detection within the broken interval. The two-level screening in step S12 provides step S13 with a valid closed polygonal contour that is complete in shape and meets the size requirements. The lack of area screening will cause step S13 to inject random directional noise into the set of polarization long axis segments. The lack of boundary integrity screening will cause step S13 to fit polarization long axis segments with incorrect orientations to incomplete contours. The major-minor axis ratio screening in step S13 provides step S21 with a set of polarization long axis segments with stable directional indication. The lack of major-minor axis ratio screening will cause the random directional information of equiaxed fragments to dilute the angular distribution differences between the two types of manifold distribution regions, reducing the accuracy of the geometric envelope area in representing the transformation of manifold distribution type. The output of step S10, namely the set of polarization long axis segments, is the only geometric data source for the construction of the angular frequency distribution histogram and the calculation of the geometric envelope area in step S20. The data purity and directional accuracy of the set of polarization long axis segments directly determine the reliability of the manifold distribution transition interface calibration in step S20.Step S10 compresses the two-dimensional contour shape information carried by each discrete polygon into a single-direction angle quantity after fitting with the minimum bounding rectangle and polarization extraction. This compression process discards redundant geometric information such as the area, perimeter, and concavity of the contour that is unrelated to the manifold distribution type, and only retains the directional features directly related to the spatial orientation of the fragments. This allows the angle frequency distribution histogram and geometric envelope area in step S20 to accurately capture the essential difference in macroscopic directional patterns between disordered manifold distributions (such as single pile sediment) and ordered manifold distributions (such as soil bearing layers) with the simplest one-dimensional angle statistics.

[0067] Step S20: Define an analysis window that slides along the depth direction, calculate the absolute angle between each polarization long axis segment in the analysis window and the horizontal baseline from the set of polarization long axis segments, construct an angle frequency distribution histogram, establish a uniformly distributed baseline horizontal line, calculate the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line, generate a depth variation curve of the geometric envelope area, and mark the manifold distribution transition interface on the depth variation curve of the geometric envelope area.

[0068] Specifically, see Figure 5 This is a schematic diagram comparing disordered and ordered manifold distributions provided in the embodiments of this application. Figure 5The diagram illustrates the discrete polygon distribution patterns of two types of manifold distribution regions. The left side represents a disordered manifold distribution region, where the spatial orientation of discrete polygons (rock and soil fragments) within the region is randomly distributed, with no directional continuity between them. The right side represents an ordered manifold distribution region, where the major axes of discrete polygons within the region exhibit a statistically significant concentration trend along the same direction, preserving a clear directional consistency characteristic. Step S20 receives the set of polarization major axis segments output from step S10 as the sole input data. Each polarization major axis segment in the set carries two-dimensional information: one type is directional information, i.e., the spatial orientation of the segment in the rectangular unfolded coordinate system; the other type is positional information, i.e., the longitudinal depth coordinates of the segment. Step S10 has compressed the two-dimensional contour shape of each discrete polygon in the borehole inner wall image into a single directional line segment. However, the direction of a single polarization long axis segment cannot determine whether its region belongs to a disordered or ordered manifold distribution region. This is because even in a disordered manifold distribution region, there are cases where several fragments accidentally point to the same angle, and in an ordered manifold distribution region, there are also individual fragments with random orientations deviating from the bedding direction. Therefore, using the direction angle of a single line segment as a classification criterion will result in a very high probability of misclassification. Step S20 aggregates the direction angles of multiple polarization long axis segments into an angle frequency distribution histogram within an analysis window with a set depth span. The geometric envelope area enclosed between the contour line of the frequency distribution histogram and the horizontal baseline representing a completely random distribution is used as a scalar value. This reduces the discrete directional information of multiple line segments to a single continuous numerical output, thereby quantifying and distinguishing them. Figure 5 The two types of manifold distribution regions are shown. The analysis window slides gradually along the depth direction, so that this scalar value is calculated and output at each depth position. The depth variation curve of the geometric envelope area is formed by arranging the geometric envelope area values ​​of all depth positions according to the depth coordinate. This curve continuously tracks the gradual change trend of the fragment orientation distribution from random disorder to directional order along the depth direction. The precise depth position of the manifold distribution transition interface (i.e., the physical interface between the sediment at the end of the single pile and the bearing layer) is represented on the curve as the positive step transition point of the geometric envelope area value. It is adaptively calibrated after being jointly determined by the step transition amplitude threshold and the continuous thickness threshold.

[0069] Further, step S20 includes:

[0070] Step S21: Define an analysis window with a set depth span, extract the start coordinates and end coordinates of each polarization long axis segment from the set of polarization long axis segments, calculate the absolute angle between the polarization long axis segment and the horizontal baseline based on the start coordinates and end coordinates, and form an absolute angle set from all the absolute angles in the current analysis window.

[0071] Extract the start and end coordinates of the polarization major axis segment in the rectangular unfolded diagram; subtract the horizontal component of the start coordinate from the horizontal component of the end coordinate to obtain the horizontal projection length, and subtract the vertical component of the start coordinate from the vertical component of the end coordinate to obtain the vertical projection length; use the arctangent function value of the ratio of the absolute value of the vertical projection length to the absolute value of the horizontal projection length as the absolute angle between the polarization major axis segment and the horizontal baseline, and limit the value range of the absolute angle to 0° to 90°; count the total number of polarization major axis segments in the current analysis window, and compare the total number of polarization major axis segments with the preset minimum line segment number threshold. When the total number of polarization major axis segments is greater than the minimum line segment number threshold, the absolute angle set is formed by all the absolute angles in the current analysis window; when the total number of polarization major axis segments is not greater than the preset minimum line segment number threshold, mark the current analysis window as a region with insufficient statistical samples, and skip the calculation in step S22.

[0072] In step S21, the analysis window is a rectangular window with a set depth span along the longitudinal direction of the rectangular unfolded diagram. The horizontal span of the analysis window covers the entire horizontal range of the rectangular unfolded diagram, and the vertical span is determined by the set depth span parameter. The depth span refers to the length of the continuous depth interval covered by the analysis window on the longitudinal coordinate of the rectangular unfolded diagram. Its physical meaning is the thickness of the analysis interval used for statistical analysis of fragment orientation distribution along the borehole axis. The value of the depth span needs to satisfy two constraints: on the one hand, it needs to be large enough to include a sufficient number of polarization long axis segments within the window to support the statistical stability of the frequency distribution histogram; on the other hand, it needs to be small enough to ensure that the spatial resolution near the manifold distribution transition interface is not excessively smoothed. The method for determining the depth span is as follows: On several representative rectangular unfolded diagrams, the density of effective closed polygonal contours within each unit depth is statistically analyzed. The reciprocal of the density is multiplied by a multiple of the minimum line segment number threshold in step S21 to obtain the lower limit of the depth span. The estimated thickness of the transition region near the manifold distribution transition interface is used as the upper limit of the depth span. The depth span parameter is selected between the upper and lower limits. For example, the depth span can be a value between 1 / 10 and 1 / 20 of the total vertical length of the rectangular unfolded diagram.

[0073] From the set of polarized long axis segments, extract the polarized long axis segments whose midpoint longitudinal coordinates fall within the depth range of the current analysis window. The midpoint of a polarized long axis segment refers to the point corresponding to the average coordinates of its start and end points, and the longitudinal coordinate of the midpoint is the borehole depth corresponding to that segment. Each polarized long axis segment has start and end coordinates in the rectangular unfolded coordinate system generated in step S11. The start coordinates contain horizontal and longitudinal components, and the end coordinates also contain horizontal and longitudinal components. Subtract the horizontal component of the start coordinate from the horizontal component of the end coordinate to obtain the horizontal projection length, and subtract the longitudinal component of the start coordinate from the longitudinal component of the end coordinate to obtain the longitudinal projection length. When the absolute value of the horizontal projection length is zero, the absolute angle is directly assigned to 90°; when the absolute value of the horizontal projection length is not zero, the ratio of the absolute value of the longitudinal projection length to the absolute value of the horizontal projection length is used as the independent variable of the arctangent function, and the output value of the arctangent function is the absolute angle between the polarized long axis segment and the horizontal baseline. The horizontal baseline refers to the direction parallel to the horizontal coordinate axis in the rectangular unfolded coordinate system, corresponding to the circumferential direction on the hole wall. The arctangent function is a well-known trigonometric function operation. The input is the ratio of the absolute value of the longitudinal projection length to the absolute value of the horizontal projection length, and the output is an angle value, ranging from 0° to 90°. The reason for taking the absolute values ​​of both the longitudinal and horizontal projection lengths before calculating the arctangent function is that the marking order of the start and end points of each polarized long axis segment is arbitrary. After swapping the start and end points, the signs of the horizontal and longitudinal projection lengths are reversed, but the spatial orientation of the fragment remains unchanged. If the absolute values ​​are not taken and the signed projection lengths are used directly to calculate the arctangent function, the direction angle obtained after swapping the start and end points will differ from the original direction angle by 180°. The distribution position of the same fragment on the angle frequency distribution histogram will shift, and the shape of the angle frequency distribution histogram will no longer simply reflect the orientation distribution characteristics of the fragment, but will be mixed with the artificial factor of the starting and ending point marking order, which has no physical meaning. By taking the absolute value, the output range of the arctangent function is folded from -90° to +90° to a range of 0° to 90°, eliminating the ambiguity of the start and end point marking order. This ensures that each polarization axis segment corresponds to a unique absolute angle within the 0° to 90° range. An absolute angle of 0° indicates that the direction of the polarization axis segment is parallel to the horizontal baseline, i.e., the circumferential direction of the borehole wall. An absolute angle of 90° indicates that the direction of the polarization axis segment is perpendicular to the horizontal baseline, i.e., parallel to the borehole axis.

[0074] After calculating the absolute angles of all polarization long axis segments within the current analysis window, the total number of polarization long axis segments within the current analysis window is counted and compared with a preset minimum segment count threshold. The minimum segment count threshold is set based on the statistical stability requirements of the frequency distribution histogram: the angle frequency distribution histogram consists of several angle intervals, and the frequency value of each angle interval represents the number of polarization long axis segments falling into that interval. When the total number of segments is too small, each angle interval contains only 0 or one segment. The shape of the histogram no longer reflects the statistical regularity of the orientation distribution but is dominated by the angular deviation of individual fragments. In this case, the calculated geometric envelope area may fluctuate drastically between adjacent analysis window positions and is not representative. The minimum segment count threshold is obtained by multiplying the total number of angle intervals by a desired lowest frequency value for each interval. For example, when the angle step size is 10°, resulting in a total of 9 angle intervals, it is expected that each interval contains at least 3 segments, and the minimum segment count threshold can be set to 27 segments. When the total number of polarization long axis segments exceeds the minimum segment number threshold, an absolute angle set is formed by all absolute angles within the current analysis window. This absolute angle set is passed to step S22 to construct an angle frequency distribution histogram. When the total number of polarization long axis segments is not greater than the minimum segment number threshold, the current analysis window is marked as a region with insufficient statistical samples. Step S22 is skipped, and the geometric envelope area value corresponding to the current analysis window position is marked as a missing value. The process returns to step S23, where step S23 slides the analysis window down along the depth direction to the next position by a set sliding step size before re-entering step S21. The region with insufficient statistical samples corresponds to the data missing segment on the depth change curve of the geometric envelope area in step S23. Step S24 skips this data missing segment during the step detection. This mechanism avoids misjudging spurious fluctuations in the geometric envelope area caused by insufficient statistics in fragmented sparse regions as manifold distribution transition interfaces.

[0075] Step S22: Group the set of absolute angles according to the angle step size to generate an angle frequency distribution histogram. Establish a uniform distribution baseline based on the total number of polarization long axis segments and the total number of angle intervals. Calculate the geometric envelope area between the angle frequency distribution histogram and the uniform distribution baseline.

[0076] The absolute angles in the set of absolute angles are grouped according to the angle step from 0° to 90°. The number of polarization long axis segments falling into each angle interval is counted as the frequency value. An angle frequency distribution histogram is constructed from the frequency values ​​of all angle intervals. The total number of polarization long axis segments is divided by the total number of angle intervals to obtain the uniform distribution reference frequency value. A horizontal line covering 0° to 90° is drawn with the uniform distribution reference frequency value as the ordinate as the uniform distribution reference horizontal line. The absolute value of the difference between the frequency value of each angle interval and the uniform distribution reference frequency value is calculated one by one. The absolute value of the difference of each angle interval is multiplied by the angle step to obtain the local area element. All local area elements are summed to obtain the geometric envelope area between the angle frequency distribution histogram and the uniform distribution reference horizontal line.

[0077] Step S22 receives the set of absolute angles output in step S21 and groups all absolute angles in the set from 0° to 90° according to a set angle step size. The angle step size refers to the angular distance between two adjacent angle intervals, and the total number of angle intervals is equal to 90° divided by the angle step size. The value of the angle step size needs to balance the angular resolution of the histogram and the frequency fullness of each interval: if the angle step size is too small, the total number of angle intervals increases, the number of polarized long axis segments in each angle interval decreases accordingly, and the histogram shape is more susceptible to random fluctuations; if the angle step size is too large, the total number of angle intervals decreases, and the angle range of the fragmented long axis concentration distribution in the ordered manifold distribution area may be completely covered by a wide interval, the directional concentration feature is averaged and not sharp enough on the histogram, the geometric envelope area value is correspondingly reduced, and the distinction between the disordered manifold distribution area and the ordered manifold distribution area decreases. The method for determining the angle step size is as follows: Select the smallest possible angle step size while ensuring that the expected frequency of each angle interval is not less than the minimum segment count threshold divided by the total number of angle intervals. For example, the angle step size can be a value between 5° and 15°. When the angle step size is 10°, the total number of angle intervals is 9, covering 0° to 10°, 10° to 20°, 20° to 30°, 30° to 40°, 40° to 50°, 50° to 60°, 60° to 70°, 70° to 80°, and 80° to 90°. The number of polarization major axis segments falling into each angle interval is counted as the frequency value. Polarization major axis segments whose absolute angle is exactly equal to the right endpoint of an interval are assigned to the next interval, and polarization major axis segments whose absolute angle is exactly equal to 90° are assigned to the last interval. An angle frequency distribution histogram is constructed from the frequency values ​​of all angle intervals.

[0078] After constructing the angular frequency distribution histogram, a uniform distribution baseline is established. The uniform distribution baseline frequency value is equal to the total number of polarization long axis segments within the current analysis window divided by the total number of angular intervals. A horizontal line covering the entire angular range from 0° to 90° is drawn with the uniform distribution baseline frequency value as the ordinate; this is the uniform distribution baseline. The physical meaning of the uniform distribution baseline is: assuming that the spatial orientation of all fragments within the current analysis window is completely randomized within the range of 0° to 90° and there is no preference in any angular direction, then the expected frequency value of each angular interval is exactly equal to the uniform distribution baseline frequency value, and the outline of the angular frequency distribution histogram completely coincides with the uniform distribution baseline, with no area difference between the two. The reason for defining the uniform distribution baseline frequency as the total number of polarization long axis segments divided by the total number of angular intervals, rather than using a fixed constant, is that the total number of polarization long axis segments within the analysis window at different depths varies with the spatial variation of fragment density. If a fixed constant is used as the baseline, the frequency bars in the histogram of angular frequency distribution in densely populated areas will generally be higher than the fixed baseline, while in sparsely populated areas, the frequency bars will generally be lower than the fixed baseline. The geometric envelope area will simultaneously contain information on both fragment density variation and orientation distribution variation, failing to reflect orientation distribution characteristics alone. The geometric envelope area is a dimensionless statistical value; its magnitude is only used to characterize the degree of deviation between the spatial orientation distribution of fragments within the current analysis window and a completely random uniform distribution. A larger value indicates a higher degree of orientation concentration. After adopting an adaptive uniform distribution baseline frequency value, regardless of fragment density variations, the uniform distribution baseline horizontal line always lies at the average height of the frequency bars. The geometric envelope area only quantifies the dispersion deviation of the frequency bars between different angular intervals, thus removing the influence of fragment density and making the geometric envelope area a pure measure of fragment orientation orderliness.

[0079] The geometric envelope area is calculated as follows: Iterate through each angular interval, calculating the absolute value of the difference between the frequency value and the uniform distribution reference frequency value. Multiply the absolute value of the difference by the angular step size to obtain the local area element. Summate the local area elements corresponding to all angular intervals to obtain the geometric envelope area. The reason for taking the absolute value of the difference instead of directly summing them is that angular intervals with frequencies higher than the uniform distribution reference frequency value and angular intervals with frequencies lower than the uniform distribution reference frequency value always appear in pairs. An increase in the frequency of some intervals is necessarily accompanied by a decrease in the frequency of other intervals. Directly calculating the algebraic sum of the differences will always result in 0, making it impossible to distinguish between a uniformly oriented distribution and a concentratedly oriented distribution. After taking the absolute value, regardless of the direction of the deviation, the local area element contributed by each angular interval is always positive. The deviations of high-frequency and low-frequency intervals are simultaneously accumulated as the contribution to the geometric envelope area. The local area element is equal to the absolute value of the difference multiplied by the angle step size. Its geometric meaning is the rectangular area segment enclosed by the contour line of the angle frequency distribution histogram within the corresponding angle interval and the uniformly distributed baseline horizontal line. The geometric meaning of summing all local area elements is the total area enclosed by the histogram contour line and the baseline horizontal line.

[0080] See Figure 6 , is the angular frequency distribution histogram of the disordered manifold distribution region provided in the embodiments of this application. Figure 6 The diagram illustrates the frequency distribution histogram with the absolute angle range as the horizontal axis and frequency as the vertical axis, as well as a uniformly distributed baseline horizontal line covering the entire angle range from 0° to 90°. The height of the frequency bars in each angle interval is close to the height of the uniformly distributed baseline horizontal line, with no obvious prominent frequency peaks. In the disordered manifold distribution region, fragments are freely scattered due to gravity and distributed with equal probability in all directions. The frequency values ​​in each interval of the angular frequency distribution histogram approach the uniformly distributed baseline frequency values, and the absolute values ​​of all differences are close to 0, with the geometric envelope area value approaching 0. See also... Figure 7 , is the angular frequency distribution histogram of the ordered manifold distribution region provided in the embodiments of this application. Figure 7The diagram illustrates the frequency distribution histogram with the absolute angle range as the horizontal axis and frequency as the vertical axis, as well as a uniform distribution baseline covering the entire angle range from 0° to 90°. A significant frequency peak is formed near the 20°-30° absolute angle range, while the frequency column heights in other angle ranges are much lower than the uniform distribution baseline. In the ordered manifold distribution region, the long axis of the fragments maintains a dominant orientation along the original sedimentary bedding plane. The frequency distribution histogram forms a frequency peak near the angle range corresponding to the bedding dip angle. The frequency in the peak interval is much higher than the uniform distribution baseline frequency value, while the frequency in other intervals is much lower than the uniform distribution baseline frequency value. The absolute values ​​of both types of deviations are large, and the geometric envelope area values ​​are far from 0. For example, assuming an angle step size of 10°, a total of 9 angle intervals, a total of 45 polarization long axis segments in the current analysis window, and 5 uniform distribution baseline frequency values. In the disordered manifold distribution region, the frequency values ​​of the nine angular intervals may be 4, 6, 5, 5, 4, 6, 5, 5, 5, respectively, with absolute values ​​of the differences being 1, 1, 0, 0, 1, 1, 0, 0, 0, respectively. The local area elements are 10, 10, 0, 0, 10, 10, 0, 0, 0, and the geometric envelope area is 40. In the ordered manifold distribution region, assuming the major axes of the fragments are concentrated near the 20° to 30° interval, the frequency values ​​of the nine angular intervals may be 2, 3, 18, 12, 4, 2, 1, 2, 1, with absolute values ​​of the differences being 3, 2, 13, 7, 1, 3, 4, 3, 4, respectively. The local area elements are 30, 20, 130, 70, 10, 30, 40, 30, 40, and the geometric envelope area is 400. There is an order of magnitude difference in the geometric envelope area of ​​the two types of manifold distribution regions. The physical root of this difference is the essential difference between the degree of order and disorder in the spatial orientation distribution of fragments, rather than the difference in the color, grayscale or texture features of the fragments. Therefore, it is not affected by the condition that the two types of regions are highly similar in color and texture.

[0081] Step S23: Set the depth span and sliding step size of the analysis window. Extract polarization long axis segments whose midpoint longitudinal coordinates fall within the depth range of the current analysis window from the set of polarization long axis segments to form a subset of local polarization long axis segments. The midpoint of the polarization long axis segment refers to the point corresponding to the average coordinate of its start and end points. Using the subset of local polarization long axis segments as input, execute steps S21 and S22 to obtain the geometric envelope area of ​​the current analysis window position. Slide the analysis window along the depth direction step by step and repeat the calculation until the traversal is completed. Generate the depth change curve of the geometric envelope area by arranging the geometric envelope area values ​​of all analysis window positions according to the depth coordinates.

[0082] Step S23 sets the sliding step size of the analysis window. The sliding step size refers to the incremental vertical distance by which the analysis window moves downward along the depth direction each time. The value of the sliding step size needs to balance the depth resolution of the depth variation curve of the geometric envelope area with the total computation: if the sliding step size is too large, the depth spacing between adjacent data points on the curve increases, and the manifold distribution transition interface cannot be accurately located when it falls exactly between two data points; if the sliding step size is too small, the number of data points increases, and the total computation increases accordingly, but the improvement in depth resolution tends to saturate after the sliding step size is less than a certain proportion of the depth span. The sliding step size is determined by multiplying the depth span by an overlap coefficient. The overlap coefficient is less than 1 so that there is partial overlap between two adjacent analysis windows in the depth range. The same set of polarization long axis segments in the overlapping area are shared by the two adjacent windows, so that the geometric envelope area values ​​between adjacent data points on the curve have continuity. For example, the overlap coefficient can be 0.2 to 0.5, corresponding to a sliding step size of 20% to 50% of the depth span. The initial position of the analysis window is set at the beginning of the vertical coordinate of the rectangular unfolded diagram (corresponding to the shallow part of the borehole), and the ending point corresponds to the deep part of the borehole. At the initial position, all polarized long axis segments whose midpoint vertical coordinate falls within the depth range of the analysis window are extracted to form a subset of locally polarized long axis segments. Using this subset as input, steps S21 (absolute angle calculation and minimum segment quantity threshold determination) and S22 (angle frequency distribution histogram construction and geometric envelope area calculation) are executed sequentially… The analysis window is slid down one step along the depth direction to the next position. At the new position, polarized long axis segments whose midpoint vertical coordinate falls within the depth range of the analysis window are extracted to form a new subset of locally polarized long axis segments. Steps S21 and S22 are repeated to obtain the geometric envelope area value at the new position. This process is repeated until the analysis window reaches the end of the vertical coordinate of the rectangular unfolded diagram. The geometric envelope area values ​​at all analysis window positions are arranged according to their corresponding depth coordinates to form a depth variation curve of the geometric envelope area. The horizontal axis of the curve represents the depth coordinate, and the vertical axis represents the geometric envelope area value.

[0083] The depth variation curve of the geometric envelope area exhibits morphological characteristics corresponding to the variation of the discrete polygonal manifold distribution type along the depth direction within the field of view. Within the depth range corresponding to the disordered manifold distribution region (such as the sediment section at the bottom of the borehole), the curve value remains in a low range close to 0, and the fluctuation amplitude is determined by the statistical fluctuations of the random orientation of the fragments, statistically manifested as a narrow-amplitude oscillation around a certain low numerical baseline. Within the depth range corresponding to the ordered manifold distribution region (such as the rock bearing layer section), the curve value jumps to a high range far from 0 and remains there. The depth position transitioning from the disordered to the ordered manifold distribution region is the manifold distribution transition interface. In geotechnical engineering investigation, this interface is the precisely located single-pile bearing layer interface, which is represented on the curve as a positive step transition of the geometric envelope area value from a low range to a high range. This curve morphological characteristic allows the depth location of the single-pile manifold distribution transition interface to be transformed into a step detection operation on a one-dimensional signal, and one-dimensional signal step detection is a well-known basic operation in the field of signal processing.

[0084] Step S24: Perform step detection by scanning the depth change curve of the geometric envelope area along the depth direction. Determine the manifold distribution type change based on the step transition amplitude threshold and the continuous thickness threshold. Mark the depth coordinates that meet the joint determination conditions as the depth coordinates of the manifold distribution transition interface.

[0085] The depth variation curve of the geometric envelope area is scanned point by point along the shallow to deep direction, and the geometric envelope area value is extracted at each depth coordinate. The difference between the geometric envelope area value at the current depth coordinate and the arithmetic mean of the geometric envelope area values ​​within the backtracking depth range set above the current depth coordinate is calculated. The difference is compared with a step transition amplitude threshold. When the difference is greater than the step transition amplitude threshold, the current depth coordinate is marked as a candidate transition depth coordinate. Starting from the candidate transition depth coordinate, the geometric envelope area value is continuously detected to be higher than the candidate transition depth coordinate. For a continuous depth segment corresponding to the sum of the arithmetic mean of the backtracking depth range and the step transition amplitude threshold, calculate the depth span of the continuous depth segment; compare the depth span of the continuous depth segment with the sustained thickness threshold, and when the depth span of the continuous depth segment is greater than the sustained thickness threshold, mark the candidate transition depth coordinate as the manifold distribution transition interface depth coordinate; when no depth coordinate satisfying the joint condition of the step transition amplitude threshold and the sustained thickness threshold is detected from the start end to the end end of the vertical coordinate of the rectangular unfolded diagram, output a prompt message that no manifold distribution transition interface was detected.

[0086] Step S24 involves scanning the depth change curve of the geometric envelope area point by point along the direction from shallow to deep to perform step detection. At each depth coordinate, the geometric envelope area value of the current depth coordinate is extracted, and the geometric envelope area values ​​of all depth coordinates within the set backtracking depth range above the current depth coordinate are also extracted. The arithmetic mean of all geometric envelope area values ​​within the backtracking depth range above is then calculated. The backtracking depth range refers to the length of the depth interval backtracking along the shallow direction from the current depth coordinate. The backtracking depth range is determined by: taking a sufficiently long range to cover multiple analysis window positions to ensure the statistical stability of the arithmetic mean; and taking a value that does not exceed the typical thickness of the disordered manifold distribution region to avoid including the values ​​of the ordered manifold distribution region in the average, which would raise the benchmark. For example, the backtracking depth range can be 2 to 5 times the depth span. The difference between the geometric envelope area value of the current depth coordinate and the above arithmetic mean is calculated, and the difference is compared with the step transition amplitude threshold. The step transition amplitude threshold is used to quantify how much the geometric envelope area must exceed the background level to be considered a statistically significant step rather than a random fluctuation. The method for determining the step transition amplitude threshold is as follows: On a calibrated image containing only disordered manifold distribution regions, calculate the geometric envelope area values ​​at multiple analysis window locations, calculate the standard deviation of these values, and multiply the standard deviation by a confidence factor to obtain the step transition amplitude threshold. The confidence factor is chosen such that the probability of a random fluctuation in a purely disordered region exceeding the step transition amplitude threshold is lower than an acceptable false alarm rate. For example, the confidence factor can be between 2 and 3.

[0087] When the difference is not greater than the step transition amplitude threshold, it is determined that the geometric envelope area value at the current depth coordinate is still within the statistical fluctuation range of the disordered manifold distribution region, and there is no change in manifold distribution type. The process continues scanning to the next depth coordinate. When the difference is greater than the step transition amplitude threshold, the current depth coordinate is marked as a candidate transition depth coordinate, and the continuous thickness verification process begins. Starting from the candidate transition depth coordinate, the geometric envelope area value is checked point by point in the depth direction to see if it is continuously higher than the sum of the arithmetic mean and the step transition amplitude threshold. Continuous depth segments that meet this condition are extracted, and the depth span of the continuous depth segments is calculated. The continuous thickness threshold refers to the minimum continuous depth length at which the geometric envelope area value must remain at a high level after a step transition. The sustained thickness threshold is set based on the following: In the transition zone between disordered and ordered manifold distribution regions, individual larger fragments may occasionally maintain an ordered orientation, causing a pulse-like increase in the geometric envelope area when the analysis window slides to the depth of that fragment. However, this pulse only exists within a very shallow depth range before falling back to a low level. The depth span of the pulse's duration typically does not exceed the major axis length of a single fragment plus the depth span of the analysis window. The sustained thickness threshold must be greater than this pulse duration depth to exclude pulse-like spurious steps. For example, the sustained thickness threshold can be a value between 1.5 and 3 times the depth span of the analysis window.

[0088] When the depth span of a continuous depth segment exceeds the sustained thickness threshold, the candidate transition depth coordinates are designated as manifold distribution transition interface depth coordinates. This represents the starting depth position of the ordered manifold distribution region after removing interference from disordered manifold distribution regions, and the corresponding elevation is the one-dimensional spatial threshold transition interface for the mixed signal. When the depth span of a continuous depth segment does not exceed the sustained thickness threshold, the increase in the geometric envelope area at the candidate transition depth coordinates is judged as a local statistical fluctuation rather than a true change in manifold distribution type. The designation of this candidate transition depth coordinate is abandoned, and the scan continues to the next depth coordinate. When no depth coordinate simultaneously satisfying both the step transition amplitude threshold and the sustained thickness threshold is detected throughout the entire scan from the beginning to the end of the rectangular unfolded diagram's vertical coordinates, a message indicating that no manifold distribution transition interface was detected is output. The simultaneous use of both the step transition amplitude threshold and the duration thickness threshold as joint criteria is necessary because using the step transition amplitude threshold alone triggers calibration whenever the amplitude of any transient pulse exceeds the limit, resulting in a false manifold distribution transition interface in the transition zone. Using the duration thickness threshold alone lacks threshold control over the step amplitude, making it impossible to determine a clear transition start position in cases where the geometric envelope area gradually increases from a low to a high level. When the two thresholds work together, the amplitude condition locks in the start time of the transition, while the duration condition verifies the physical authenticity of the transition; both are indispensable.

[0089] Step S10's cylindrical unfolding coordinate mapping eliminates perspective distortion, ensuring that the absolute angle calculated in step S21 accurately reflects the true spatial orientation of the fragments rather than a projected angle containing distortion bias. The angle frequency distribution histogram thus faithfully depicts the statistical characteristics of the orientation distribution. Step S10's adaptive local grayscale equalization ensures the uniformity of contour extraction quality across the entire depth range, preventing a sharp drop in the number of polarized long axis segments obtained at different depth positions in step S23 due to illumination attenuation. The depth variation curve of the geometric envelope area remains continuous and usable across the entire depth range without missing data segments, allowing step detection in step S24 to be performed at any depth position without being limited by illumination conditions. Step S10's area filtering and boundary integrity filtering eliminate noise. The acoustic profile and incomplete profile ensure that the set of absolute angles in step S21 does not contain spurious angle values ​​introduced by noise or clipping, and the shape of the angle frequency distribution histogram is not disturbed by spurious frequencies. The value of the geometric envelope area in step S22 only reflects the orientation distribution characteristics of the real fragments. The major-minor axis ratio screening in step S10 removes equiaxed fragments, ensuring that the set of absolute angles in step S21 only contains the angle values ​​of long strip fragments with stable direction indication. The direction concentration peak in the ordered manifold distribution region is not diluted by the random angles of equiaxed fragments. The difference between the value of the geometric envelope area of ​​the ordered manifold distribution region and the value of the disordered manifold distribution region in step S22 is widened. The step transition amplitude in step S24 is steeper, thus improving the positioning accuracy of the manifold distribution transition interface.

[0090] Step S20 processes the discrete multi-directional line segment information output from step S10 through a series of steps, including angle frequency statistics, uniform baseline comparison, area quantization, depth curve generation, and joint threshold determination, to output an accurate depth coordinate of the manifold distribution transition interface. All operations involve only arctangent function calculation, frequency statistics, division, absolute difference calculation, multiplication accumulation, and point-by-point comparison, all of which are basic arithmetic and logical operations. They do not include computationally complex operations such as matrix factorization, gradient backpropagation, or iterative optimization. The calculation time for the geometric envelope area of ​​a single analysis window is less than milliseconds, and curve generation and step detection across the entire depth range can be completed on a conventional processor at near real-time speed. Step S20 does not require any labeled training dataset. The construction of the angle frequency distribution histogram and the uniform distribution baseline relies solely on the actual polarization long axis segment data extracted from the current image. The calculation of the geometric envelope area does not depend on the weight parameters of any pre-trained model; therefore, there is no risk of recognition ability degradation when the training sample distribution does not match the actual working conditions. The uniform distribution baseline serves as a reference for the geometric envelope area. Its definition is entirely derived from the mathematical definition of uniform distribution in probability theory, without containing any parameters that require manual calibration or empirical adjustment. This ensures the physical comparability of the geometric envelope area values ​​under different drilling conditions, fragment densities, and depth ranges. Step S20 transforms the original task of distinguishing highly similar two-dimensional appearance regions in color texture space into an arithmetic operation that detects the degree to which the frequency distribution deviates from the uniform baseline in one-dimensional angular statistical space. This domain transformation utilizes the differences in macroscopic statistical regularities of fragment spatial orientation. The fragment orientation of disordered manifold distribution regions is equally probable at all angles, while the fragment orientation of ordered manifold distribution regions is concentrated along the bedding direction, serving as a basis for distinction. This difference in macroscopic statistical regularity originates from the physical essence of the formation mechanism of the two types of regions in three-dimensional space rather than the difference in surface color texture features. Therefore, it is not affected by color texture similarity in principle. The geometric envelope area, as a scalar indicator, also provides an additional information extraction capability: in an ordered manifold distribution region, the peak position of the frequency peak of the angle frequency distribution histogram corresponds to the angle of the concentrated orientation of the long axis of the soil fragments. This angle has a geometric correspondence with the dip angle of the original sedimentary bedding plane. By recording the peak angle interval of the angle frequency distribution histogram within the analysis window after the geometric envelope area value jumps, the estimated value of the bedding plane dip angle within the ordered manifold distribution region can be obtained. This provides additional geometric parameters for subsequent geotechnical engineering single pile bearing capacity and settlement analysis without increasing any computational cost.

[0091] Example 2:

[0092] This embodiment, based on Embodiment 1, provides a single-pile interface positioning system for geotechnical investigation based on borehole wall image measurement, such as... Figure 8 As shown, it includes:

[0093] Polarization line segment generation module: used to acquire panoramic images of the cylindrical surface inside the borehole and perform coordinate mapping and equalization processing to generate an equalized rectangular unfolded image. Through two-level geometric validity cascade filtering, it extracts a set of valid closed polygon contours from the equalized rectangular unfolded image. It then performs minimum bounding rectangle fitting and major-minor axis ratio filtering on the set of valid closed polygon contours to generate a set of polarized major axis segments.

[0094] Manifold Interface Calibration Module: This module defines an analysis window that slides along the depth direction. It calculates the absolute angle between each polarization long axis segment within the analysis window and the horizontal baseline from the set of polarization long axis segments and constructs an angle frequency distribution histogram. It establishes a uniformly distributed baseline horizontal line, calculates the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line, generates a depth variation curve of the geometric envelope area, and calibrates the manifold distribution transition interface on the depth variation curve of the geometric envelope area.

[0095] Furthermore, in the polarization segment generation module, the method for performing cylindrical expansion coordinate mapping includes:

[0096] With the borehole axis as the rotation axis, the unfolding radius is the radial distance from the optical center of the camera device to the borehole wall. Each pixel in the cylindrical panoramic image has two coordinate dimensions: a circumferential angle component and a depth component. Multiplying the circumferential angle component by the unfolding radius yields the horizontal coordinate of the pixel on the rectangular unfolded image. The depth component is directly mapped to the vertical coordinate on the rectangular unfolded image. After completing the coordinate transformation pixel by pixel, the rectangular unfolded image is generated.

[0097] The two-level geometric validity cascaded screening includes a first-level area screening and a second-level boundary integrity screening.

[0098] The method for extracting a set of valid closed polygon contours from an equalized rectangular unfolded image includes:

[0099] Edge detection and morphological closure operations are performed on the equalized rectangular unfolded map to generate a binary edge contour map. Closed polygon contours are extracted from the binary edge contour map. First-level area filtering and second-level boundary integrity filtering are performed on the closed polygon contours to generate a set of valid closed polygon contours.

[0100] The method for performing the area filtering is as follows:

[0101] Calculate the area of ​​the closed polygon contour, compare the contour area with a preset minimum area threshold, delete closed polygon contours whose contour area is less than or equal to the minimum area threshold, and retain closed polygon contours whose contour area is greater than the minimum area threshold as contours that pass the area screening.

[0102] The execution method for the boundary integrity screening is as follows:

[0103] The area filtering is determined by the intersection relationship between the outline and the boundary of the equalized rectangular unfolded diagram. Closed polygon outlines that intersect with any one of the four boundaries of the equalized rectangular unfolded diagram are deleted, and areas that do not intersect with the boundary of the equalized rectangular unfolded diagram are filtered by the outline and marked as valid closed polygon outlines.

[0104] The method for generating the set of polarized long axis segments includes:

[0105] Calculate the ratio of the major axis to the minor axis of each minimum bounding rectangle. For the minimum bounding rectangle whose ratio is greater than the threshold, extract the line connecting the midpoints of the two short sides as the polarization major axis segment. The set of polarization major axis segments is composed of all polarization major axis segments.

[0106] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0107] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0108] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement, characterized in that, The method includes: A panoramic image of the cylindrical surface inside the borehole is acquired and the cylindrical surface unfolding coordinate mapping and equalization processing are performed to generate an equalized rectangular unfolded image. Through two-level geometric validity cascade filtering, a set of effective closed polygon contours is extracted from the equalized rectangular unfolded image. The set of effective closed polygon contours is then subjected to minimum bounding rectangle fitting and major-minor axis ratio filtering to generate a set of polarized major axis segments. Define an analysis window that slides along the depth direction. Calculate the absolute angle between each polarization long axis segment and the horizontal baseline within the analysis window from the set of polarization long axis segments and construct an angle frequency distribution histogram. Establish a uniformly distributed baseline horizontal line. Calculate the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line. Generate a depth variation curve of the geometric envelope area. Mark the manifold distribution transition interface on the depth variation curve of the geometric envelope area. The method for performing cylindrical unfolded coordinate mapping includes: taking the borehole axis as the rotation axis, taking the unfolding radius as the radial distance from the optical center of the camera device to the borehole wall, each pixel in the cylindrical panoramic image has two coordinate dimensions: circumferential angle component and depth component, multiplying the circumferential angle component by the unfolding radius to obtain the horizontal coordinate of the pixel on the rectangular unfolded image, and directly mapping the depth component to the vertical coordinate on the rectangular unfolded image, and generating the rectangular unfolded image after completing the coordinate transformation pixel by pixel; The two-level geometric validity cascaded filtering includes a first-level area filtering and a second-level boundary integrity filtering; the method for extracting a set of valid closed polygon contours from the equalized rectangular unfolded image includes: performing edge detection and morphological closure operations on the equalized rectangular unfolded image to generate a binary edge contour image, extracting closed polygon contours from the binary edge contour image, and performing a first-level area filtering and a second-level boundary integrity filtering on the closed polygon contours to generate a set of valid closed polygon contours. The method for performing area filtering is as follows: calculate the contour area of ​​the closed polygonal contour, compare the contour area with a preset minimum area threshold, delete closed polygonal contours whose contour area is less than or equal to the minimum area threshold, and retain closed polygonal contours whose contour area is greater than the minimum area threshold as contours that pass the area filtering.

2. The geotechnical investigation single pile interface positioning method based on hole wall image measurement according to claim 1, characterized in that, The execution method for the boundary integrity screening is as follows: The area filtering is determined by the intersection relationship between the outline and the boundary of the equalized rectangular unfolded diagram. Closed polygon outlines that intersect with any one of the four boundaries of the equalized rectangular unfolded diagram are deleted, and areas that do not intersect with the boundary of the equalized rectangular unfolded diagram are filtered by the outline and marked as valid closed polygon outlines.

3. The geotechnical investigation single pile interface positioning method based on hole wall image measurement according to claim 2, characterized in that, The method for generating the set of polarized long axis segments includes: Calculate the ratio of the major axis to the minor axis of each minimum bounding rectangle. For the minimum bounding rectangle whose ratio is greater than the threshold, extract the line connecting the midpoints of the two short sides as the polarization major axis segment. The set of polarization major axis segments is composed of all polarization major axis segments.

4. The method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement according to claim 3, characterized in that, The method for constructing the angular frequency distribution histogram includes: Extract the start and end coordinates of each polarization long axis segment from the set of polarization long axis segments one by one, calculate the absolute angle between the polarization long axis segment and the horizontal baseline based on the start and end coordinates, and form an absolute angle set from all the absolute angles in the current analysis window; The absolute angles in the set of absolute angles are grouped according to the angle step from 0° to 90°. The number of polarization major axis segments falling into each angle interval is counted as the frequency value. The frequency values ​​of all angle intervals are used to construct an angle frequency distribution histogram.

5. The method for locating the interface of a single pile in geotechnical investigation based on borehole wall image measurement according to claim 4, characterized in that, The method for generating the set of absolute included angles includes: The horizontal projection length is obtained by subtracting the horizontal component of the starting point coordinate from the horizontal component of the endpoint coordinate of the polarization long axis segment, and the vertical projection length is obtained by subtracting the vertical component of the starting point coordinate from the vertical component of the endpoint coordinate. The arctangent function value of the angle, which is the ratio of the absolute value of the vertical projection length to the absolute value of the horizontal projection length, is used as the absolute angle between the polarization long axis segment and the horizontal baseline. The total number of polarized long axis segments within the current analysis window is counted. The total number of polarized long axis segments is compared with a preset minimum segment number threshold. When the total number of polarized long axis segments is greater than the minimum segment number threshold, the set of absolute angles is formed by all absolute angles within the current analysis window.

6. The geotechnical investigation single pile interface positioning method based on hole wall image measurement according to claim 5, characterized in that, The method for calculating the geometric envelope area includes: Divide the total number of polarization long axis segments by the total number of angle intervals to obtain the uniformly distributed reference frequency value. Plot a horizontal line covering 0° to 90° with the uniformly distributed reference frequency value as the ordinate as the uniformly distributed reference horizontal line. Calculate the absolute value of the difference between the frequency value of each angle interval and the uniformly distributed reference frequency value. Multiply the absolute value of the difference of each angle interval by the angle step size to obtain the local area element. Summate all the local area elements to obtain the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed reference horizontal line.

7. The geotechnical investigation single pile interface positioning method based on hole wall image measurement according to claim 6, characterized in that, The method for generating the depth variation curve of the geometric envelope area includes: The depth span and sliding step size of the analysis window are set. Polarized long axis segments whose midpoint longitudinal coordinates fall within the depth range of the current analysis window are extracted from the set of polarized long axis segments to form a subset of local polarized long axis segments. The midpoint of the polarized long axis segment refers to the point corresponding to the average coordinate of its start and end points. The geometric envelope area of ​​the current analysis window position is obtained based on the subset of local polarized long axis segments. The analysis window is gradually slid along the depth direction according to the sliding step size and the calculation is repeated until the traversal is completed. The depth variation curve of the geometric envelope area is generated by arranging the geometric envelope area values ​​of all analysis window positions according to the depth coordinates.

8. The geotechnical investigation single pile interface positioning method based on hole wall image measurement according to claim 7, characterized in that, The method for calibrating the manifold distribution transition interface on the depth variation curve of the geometric envelope area includes: Step detection is performed by scanning the depth change curve of the geometric envelope area along the depth direction. The change of manifold distribution type is determined by jointly judging the step transition amplitude threshold and the continuous thickness threshold. The depth coordinates that meet the joint judgment conditions are calibrated as the depth coordinates of the manifold distribution transition interface.

9. A single-pile interface positioning system for geotechnical investigation based on borehole wall image measurement, used to implement the single-pile interface positioning method for geotechnical investigation based on borehole wall image measurement as described in any one of claims 1-8, characterized in that, The system includes: Polarization line segment generation module: used to acquire panoramic images of the cylindrical surface inside the borehole and perform coordinate mapping and equalization processing to generate an equalized rectangular unfolded image. Through two-level geometric validity cascade filtering, it extracts a set of valid closed polygon contours from the equalized rectangular unfolded image. It then performs minimum bounding rectangle fitting and major-minor axis ratio filtering on the set of valid closed polygon contours to generate a set of polarized major axis segments. Manifold Interface Calibration Module: This module defines an analysis window that slides along the depth direction. It calculates the absolute angle between each polarization long axis segment within the analysis window and the horizontal baseline from the set of polarization long axis segments and constructs an angle frequency distribution histogram. It establishes a uniformly distributed baseline horizontal line, calculates the geometric envelope area between the angle frequency distribution histogram and the uniformly distributed baseline horizontal line, generates a depth variation curve of the geometric envelope area, and calibrates the manifold distribution transition interface on the depth variation curve of the geometric envelope area.

Images