A method and apparatus for measuring the angle of repose of a granular material

Abstract

The application discloses a kind of granular material angle of repose measurement method and device, promote bottomless forming cylinder pile and obtain side view image, generate multiple candidate masks using multiple segmentation strategies, each candidate is independently cleaned and corrected, and then rescored, and the state after correction is reflected by sorting;Target mask is screened through candidate angle consistency and correction amplitude double-gated screening.Target baseline is determined and normalized to dimensionless relative pile height, left and right slope candidate windows are matched by height, and the best is selected by combining window quality, interval overlap and coverage balance, and the main angle of repose is fitted.Combining global-local slope differences, fitting quality, left and right slope angle difference, and single-image geometric angle interval, etc.A joint quality judgment is performed to reject or save the results.The application can reduce the influence of background, reflection, bearing structure and local boundary anomalies on the measurement results, and improve the measurement reliability.

Description

Technical Field

[0001] This invention relates to the measurement of the physical properties of particulate materials and digital image processing technology, and in particular to a method and apparatus for measuring the angle of repose of particulate materials. Background Technology

[0002] The angle of repose is one of the important parameters characterizing the flowability, packing properties, and stability of granular materials. Common methods for measuring the angle of repose include the funnel stacking method, the discharge method, the tilting method, and the lifting cylinder method. The lifting cylinder method typically involves loading granules into a molded cylinder with an open bottom placed on a supporting surface. By lifting the cylinder, the lateral constraints on the granules are released, allowing them to naturally accumulate on the supporting surface to form a static granule pile. The angle of repose is then determined based on the angle between the slope of the granule pile and the bottom reference. Using an image acquisition device to obtain a side view image of the granule pile and performing digital image analysis can reduce the impact of manual angle measurement, manual point selection, and operator judgment differences on the measurement results.

[0003] In some image-based methods for measuring the angle of repose, threshold segmentation, contour extraction, and geometric fitting can be used to identify the particle pile region and calculate the slope angle. However, in actual image acquisition, factors such as uneven background brightness, low contrast between particles and background, shadows, reflections, the presence of supporting structures in the field of view, and missing local boundaries can all cause the segmented foreground region to deviate from the actual particle pile. Further hole repair, morphological processing, or boundary adjustment of the initial segmentation results may also change the area, contour, and free surface position of the candidate region. Therefore, relying solely on a single segmentation result, the uncorrected candidate evaluation result, or even just the score of a corrected candidate may not be sufficient to fully determine whether the obtained particle pile contour has reliable physical meaning.

[0004] Furthermore, the free surface of an actual particle pile is usually not an ideal straight line, especially near the top and bottom of the pile, where there may be significant curvature changes, local unevenness, or boundary fluctuations caused by particle size. If the slope fitting segment is selected directly according to a fixed pixel range, the corresponding actual particle pile position may change with image magnification and particle size. If the left and right slopes are selected independently with better local fitting results, the actual slopes on the left and right sides may be located at different relative height ranges. At the same time, the direction of the image's horizontal coordinate axis is not necessarily strictly consistent with the direction of the actual bearing surface of the particles. Camera mounting posture, bearing surface tilt, or imaging projection may all cause deviations between the image horizontal reference and the actual particle contact reference. All of these factors may affect the consistency and comparability of the left and right slope angle measurements.

[0005] For particle stack measurements in water or other transparent liquids, imaging conditions are even more complex. The liquid medium, the transparent container walls, and different optical interfaces can introduce imaging position or scale deviations. Specular reflections, uneven illumination, and bubbles can also be misidentified as particle stack boundaries. If the calibration image, background image, and particle stack image differ in terms of liquid type, liquid depth, camera pose, observation interface, or illumination conditions, or if a consistent imaging coordinate reference is not established for different images, the uncertainty in particle stack contour recognition and repose angle measurement will be further increased.

[0006] Several image processing-based methods for measuring the angle of repose of particles have been disclosed in the prior art. For example, CN103278426A discloses an image-based method and apparatus for measuring the angle of repose, which acquires an image of a sample pile using an image acquisition device and calculates the angle of repose based on image feature information; CN108240949A discloses an apparatus and method for measuring the angle of repose of bulk materials, which determines the angle of repose by acquiring an image of the pile, extracting the contour, and performing geometric fitting; CN108596927A discloses a method for measuring the angle of repose based on image processing of a material pile, which obtains the contour of the material pile by performing grayscale processing, contour processing, and binarization on the image of the material pile, and determines the angle of repose on the left and right sides respectively; WO2020014150A1 also discloses a processing method based on particle pile images for extracting the measurement area, fitting the contour, and determining the slope angle by combining with a reference line. The above technologies show that the calculation of the angle of repose of granular materials using image acquisition, contour extraction, and geometric fitting has a corresponding technical foundation.

[0007] However, in actual static particle pile image processing, there may still be situations where multiple competing candidate particle pile regions are generated from the same image, the quality and geometry of candidate regions change after boundary correction, abnormal displacement occurs at local boundaries, the actual measured sections of the left and right slopes are at different relative heights, and there are significant differences between the overall fitted slope angle and the local slope geometry. In these cases, obtaining only a numerical angle of repose is insufficient to reflect the reliability of the measurement result. Therefore, there is still room for improvement in how to comprehensively judge the various factors that may affect the reliability of the angle of repose output during particle pile contour extraction, boundary correction, measurement benchmark determination, and left and right slope fitting, and how to avoid outputting unreliable measurement results in abnormal situations while retaining relevant quality evaluation and result traceability information.

[0008] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0009] The main objective of this invention is to overcome the deficiencies in the aforementioned background technology and to provide a method and apparatus for measuring the angle of repose of granular materials.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for measuring the angle of repose of granular materials includes lifting a bottomless forming cylinder to form a static granule pile on a bearing surface, and further includes the following steps: S1. For the static side view image of the particle stack obtained after the lifting is completed and a predetermined resting time has elapsed, multiple segmentation strategies are used to generate multiple candidate particle stack masks. S2. Perform physical validity judgment on each candidate particle stack mask, and independently perform mask cleaning and free surface boundary correction on each candidate particle stack mask that passes the physical validity judgment. Then, perform physical validity judgment again based on the corrected candidate particle stack mask and recalculate the final mask quality score. Use the final corrected mask quality score of each candidate as the sorting basis. S3. Based on the final mask quality score, determine a set of high-scoring candidates from the candidate particle stack masks that have passed the physical validity judgment after correction and can form an effective double-sided slope geometric representation. Perform candidate geometric consistency judgment based on the dispersion of the candidate angle proxy value of each high-scoring candidate, and perform boundary correction reliability judgment based on the top boundary displacement and mask overlap of the highest-scoring candidate before and after boundary correction. When both the candidate geometric consistency judgment and the boundary correction reliability judgment pass, the highest-scoring candidate is determined as the target mask. S4. Determine the actual bottom baseline between the particle and the bearing surface based on the bottom boundary point of the target mask, define the vertical distance from the top of the pile to the actual bottom baseline as the pile height, and normalize the signed vertical distance from the free surface boundary point to the actual bottom baseline with the pile height to obtain the dimensionless relative pile height. S5. Generate multiple consecutive candidate windows on the left and right slopes respectively. Form a candidate window pair between the left and right candidate windows whose differences in the lower and upper limits of the dimensionless relative stacking height are not greater than the preset height matching tolerance. Calculate the joint evaluation score of the window pair based on the window quality of the left and right candidate windows, the degree of overlap of their dimensionless relative stacking height intervals, and the degree of balance of their coverage lengths. Select the candidate window pair that passes the preset window pair quality threshold and has the highest joint evaluation score as the target window pair. S6. Fit the left and right slope lines within the target window respectively, calculate the angle between each slope line and the actual bottom baseline, and determine the principal rest angle of the current static side view image based on the left and right angles. S7. Perform a slope representativeness judgment based on the global-local angle difference between the principal angle of repose and the local slope representative angle, as well as the local slope angle dispersion. Combine the slope fitting quality, the difference between the left and right slope angles, and at least one independent anomaly information in the single-map geometric angle interval to perform a joint quality judgment. If the preset rejection condition is met, reject the current principal angle of repose; otherwise, save the current principal angle of repose.

[0011] A device for measuring the angle of repose of a static particle pile formed by the lifting cylinder method, comprising: The bearing surface is used to support the static particle stack; A bottomless molding cylinder, the lower end of which can be placed above the bearing surface; A lifting component is used to drive the bottomless forming cylinder to move vertically from the filling position to the stacking position; A clamping component is disposed at the moving end of the lifting component and is used to clamp the bottomless forming cylinder and move it with the lifting component; An image acquisition component is used to acquire at least one static side view image of the particle stack after the bottomless forming cylinder has been lifted and a predetermined settling time has elapsed. And an image processing apparatus, including a processor and a memory, the memory storing program instructions executable by the processor for causing the processor to perform the repose angle measurement method.

[0012] The present invention has the following beneficial effects: This invention proposes a method and apparatus for measuring the angle of repose of granular materials. A static particle stack is formed by lifting a bottomless forming cylinder, and the angle of repose is determined based on the static side view image of the particle stack. Addressing the problem that factors such as uneven background brightness, low contrast between particles and background, shadow reflections, the inclusion of supporting structures in the field of view, and missing local boundaries in actual image acquisition cause the segmented foreground to deviate from the actual particle stack, this invention employs multiple segmentation strategies to generate multiple candidate particle stack masks. After each candidate is independently cleaned and the free surface boundary is corrected, the physical validity judgment and the final mask quality score are recalculated. This ensures that the candidate ranking reflects the geometric state of the particle stack after boundary correction, preventing the evaluation results of the candidates before correction from continuing to dominate the final selection and causing unreliable contours to enter subsequent angle measurements, thereby improving the accuracy of particle stack contour extraction.

[0013] To address issues such as the potential generation of multiple competing candidate particle clusters in the same image, changes in the quality and geometry of candidate regions after boundary correction, and abnormal displacements at local boundaries, this invention judges the slope geometry consistency among high-quality candidates, identifies the divergence between multiple candidates of similar quality corresponding to different slope geometry features by combining the dispersion of candidate angle proxy values, and comprehensively evaluates the reliability of boundary correction by using the average displacement, maximum displacement, and mask overlap degree of the top boundary before and after boundary correction. This effectively identifies situations where the overall mask overlap degree is high but abnormal displacements occur on local free surfaces, reducing the possibility of abnormal candidates directly entering the angle of repose calculation.

[0014] To address the measurement reference deviation caused by inconsistencies between the image's lateral coordinate axis and the actual bearing surface of the particles, as well as camera mounting posture and bearing surface tilt, this invention uses the actual contact baseline between the particles and the bearing surface as the slope angle measurement reference. It also normalizes the relative height of the free surface and the baseline fitting residual using the actual stack height. This reduces the impact of the image's lateral reference deviation from the actual bearing surface of the particles, as well as the influence of different particle stack sizes on the consistency of the slope measurement reference, providing a unified geometric reference for the subsequent selection of corresponding left and right slope measurement areas.

[0015] To address the issue that independently selecting locally fitted sections on the left and right slopes may result in slopes on both sides being measured at different relative heights, this invention forms window pairs between candidate windows on the left and right slopes whose dimensionless relative pile heights both meet matching conditions. The quality of the left and right windows, the degree of overlap in relative height intervals, and the balance of coverage length are jointly evaluated, effectively improving the geometric comparability of the actual measurement areas on the left and right slopes and ensuring the consistency and comparability of the left and right slope angle measurements. After obtaining the effective particle pile outline, the equivalent area repose angle can be determined based on the effective two-dimensional projected area and effective projected base width of the same particle pile, providing auxiliary characterization information of the particle pile from the perspective of overall projected geometry. This results in a particle pile geometric measurement result that considers both local effective slope characteristics and overall pile shape characteristics.

[0016] To address the issue that the slope angle obtained from the overall fitting differs significantly from the local slope geometry, and that obtaining only the numerical angle of repose is insufficient to reflect the reliability of the measurement results, this invention performs slope representativeness judgment based on the global-local angle difference between the principal angle of repose and the local slope representative angle, as well as the degree of dispersion of the local slope angle. It also performs joint quality judgment by combining slope fitting quality, left-right slope angle differences, and at least one independent anomaly from the geometric angle interval of a single image. When a preset rejection condition is met, the current principal angle of repose is rejected. This establishes a quality judgment mechanism for candidate particle pack contours, boundary correction results, actual bottom baselines, and slope measurement areas. This mechanism determines whether the current static side view image is sufficient to support reliable angle of repose measurement and avoids outputting unreliable measurement results when the predetermined quality conditions are not met, while retaining corresponding quality evaluation and result traceability information.

[0017] Furthermore, this invention can also be extended to transparent liquid media conditions. Addressing issues such as imaging position or scale deviations introduced by the liquid medium, transparent container walls, and optical interfaces, as well as problems like specular reflection, uneven illumination, and bubbles potentially being misidentified as particle pile boundaries, this invention establishes geometric correction relationships under consistent medium and imaging conditions. It then uses a background image of an empty tank corrected by the same geometric mapping to normalize the background of the static side view image of the particle pile, while simultaneously eliminating reflective and bubble areas. This reduces interference from transparent liquid, container walls, optical interfaces, background, and local bubbles on particle pile contour recognition, allowing image preprocessing under transparent liquid conditions to seamlessly integrate with the aforementioned particle pile contour extraction and static angle of repose measurement processes. This invention is primarily applicable to the measurement of the angle of repose of static particle piles formed by the lifting cylinder method and can be extended to the measurement of particle pile images from different shooting orientations, providing a reliable technical means for the measurement of particulate material properties.

[0018] Other beneficial effects of the embodiments of the present invention will be further described below. Attached Figure Description

[0019] Figure 1A This is a schematic diagram of the mechanical main structure of the particle repose angle measuring device according to an embodiment of the present invention.

[0020] Figure 1B This is a schematic diagram of the image acquisition arrangement of the particle repose angle measuring device according to an embodiment of the present invention.

[0021] Figure 2A This is a schematic diagram of the overall and partial clamping structure of the clamping assembly according to an embodiment of the present invention.

[0022] Figure 2B This is a schematic diagram of the clamping relationship between the clamping part of the clamping assembly and the bottomless forming cylinder in an embodiment of the present invention.

[0023] Figure 3This is a schematic diagram of the process of generating a multi-candidate particle stack mask, independently correcting candidates, re-scoring after correction, judging the consistency of candidate angles, judging the reliability of boundary correction, and measuring the subsequent angle of repose in an embodiment of the present invention.

[0024] Figure 4 This is a geometric schematic diagram of the actual contact baseline, dimensionless relative stack height, left and right matching height windows, and slope fitting relationship in an embodiment of the present invention.

[0025] Figure 5 This is a schematic diagram illustrating the boundary correction reliability judgment principle of an embodiment of the present invention.

[0026] Figure 6 This is a schematic diagram illustrating the principle of high-quality candidate angle consistency judgment in an embodiment of the present invention.

[0027] Figure 7A This is a schematic diagram showing the relationship between imaging in a transparent liquid medium and calibration with the same medium.

[0028] Figure 7B A schematic diagram for image correction and access to a universal angle measurement link for transparent liquid media.

[0029] Figure 8 This is a flowchart illustrating the overall process of the method for measuring the angle of repose of particulate materials according to the present invention. Detailed Implementation

[0030] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.

[0031] It should be noted that when a component is referred to as "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as "connected to" another component, it can be directly connected to or indirectly connected to that other component. Furthermore, a connection can be used for fixing, coupling, or communication.

[0032] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0033] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0034] This invention aims to overcome the problems of insufficient reliability in existing image-based particle repose angle measurement due to factors such as non-unique segmentation results, abnormal boundary correction, bottom baseline deviation, and mismatch between measurement sections on the left and right slopes. It proposes a particle repose angle measurement method and apparatus. This method involves generating multiple candidates and re-evaluating them after boundary correction; combining candidate geometric consistency and boundary correction magnitude for dual gating to select reliable target masks; and using actual bottom baseline and normalized stacking height to achieve joint slope selection based on left and right slope height matching. It also incorporates global-local slope representativeness judgment and multi-factor joint quality gating to perform particle repose angle measurement. This allows for reliable rejection of unreliable measurement results and retention of quality traceability information even in abnormal situations.

[0035] This invention provides a method for measuring the angle of repose of granular materials. This method can employ methods such as... Figure 1A , Figure 1B , Figure 2A and Figure 2B The measurement device shown includes: a bearing surface 141, a bottomless forming cylinder 130, a lifting assembly (including a Z-axis lifting assembly 111), a clamping assembly 120, an image acquisition assembly (including a first image acquisition unit 150 and an optional second image acquisition unit 151), and an image processing device. The bearing surface 141 supports a static particle pile 170. The lower end of the bottomless forming cylinder 130 can be positioned above the bearing surface 141. The lifting assembly drives the bottomless forming cylinder 130 vertically from a filling position to a pile position. The clamping assembly 120 is located at the moving end of the lifting assembly to clamp the bottomless forming cylinder 130, causing it to move with the lifting assembly. The image acquisition assembly acquires at least one static side view image of the particle pile 170 after the bottomless forming cylinder 130 has been lifted and a predetermined settling time has elapsed. Image processing is performed by the image processing device. The device may further include a motion and acquisition controller, used to control the bottomless forming cylinder 130 to complete the lifting according to a preset lifting speed and lifting distance, and to determine the image acquisition timing according to a preset resting time; the controller or image processing device may also associate and save the lifting parameters, resting time and measurement results to form corresponding measurement records. (See also...) Figure 8 The method specifically includes the following steps: Step S1. For the static side view image of the particle stack obtained after the lifting is completed and a predetermined resting time has elapsed, multiple candidate particle stack masks are generated using various segmentation strategies.

[0036] In some embodiments, in step S1, the multiple segmentation strategies include global threshold segmentation, adaptive threshold segmentation, and background correction threshold segmentation, and positive polarity candidate particle stack masks and anti-polarity candidate particle stack masks are generated for each segmentation result, so that each candidate particle stack mask can independently enter the subsequent physical validity judgment and mask quality score calculation.

[0037] Step S2. Perform physical validity judgment on each candidate particle stack mask, and independently perform mask cleaning and free surface boundary correction on each candidate particle stack mask that passes the physical validity judgment. Then, perform physical validity judgment again based on the corrected candidate particle stack masks and recalculate the final mask quality score. Use the final corrected mask quality score of each candidate as the sorting basis.

[0038] In some embodiments, step S2, the physical validity judgment includes: calculating the ratio of the foreground area to the image area, the ratio of the width of the enclosing region to the image width, the ratio of the height of the enclosing region to the image height, the bottom support contact degree, the top boundary coverage degree, the apparent height from the top to the bottom of the stack, and the degree of overlap with a predetermined interference region for each candidate particle stack mask, and comparing them with the corresponding preset thresholds. When any of the aforementioned indicators does not meet the preset threshold conditions, it is determined that the corresponding candidate particle stack mask has not passed the physical validity judgment. The final mask quality score is obtained by weighting the sum of multiple positive quality components including the foreground area score, bottom support contact score, main connected component ratio, outer contour continuity score, top boundary validity score, top boundary coverage score, bottom flatness score, enclosing region aspect ratio score, top to bottom topology score, and enclosing region scale score, and subtracting the penalty term including internal hole penalty and interference region overlap penalty.

[0039] In some embodiments, in step S2, mask cleaning includes: retaining the main connected region that corresponds to the spatial position of the particle stack, repairing internal holes that meet the preset area conditions, and performing morphological cleaning to remove discrete noise and local defects; the free surface boundary correction is achieved by graph cut processing based on foreground and background seed constraints or boundary alignment processing based on image edges.

[0040] Step S3. Based on the final mask quality score, determine a high-scoring candidate set from the candidate particle stack masks that have passed the physical validity judgment after correction and can form an effective bilateral slope geometry representation. Perform a candidate geometric consistency judgment based on the dispersion of the candidate angle surrogate values ​​of each high-scoring candidate. Perform a boundary correction amplitude gating reliability judgment based on the top boundary displacement and mask overlap degree of the highest-scoring candidate before and after boundary correction, normalized to the image height using Himg. When both the candidate geometric consistency judgment and the boundary correction reliability judgment pass, the highest-scoring candidate is determined as the target mask. Record the final mask quality score difference between the highest-scoring candidate and the second-highest-scoring candidate as the candidate score difference. This candidate score difference is saved as diagnostic information of the candidate competition state and does not constitute a rejection condition on its own.

[0041] In some embodiments, step S3, the boundary correction reliability judgment includes: after boundary correction is completed, recording the top boundary ordinates of the mask before and after correction at a common effective abscissa; calculating the normalized average displacement and normalized maximum displacement based on the displacement of the top boundary before and after correction, wherein the normalized average displacement and normalized maximum displacement are normalized using the image height of the entire static side view image; simultaneously calculating the cross-union ratio of the masks before and after correction; when the normalized average displacement exceeds a preset average displacement threshold, or the normalized maximum displacement exceeds a preset maximum displacement threshold and the cross-union ratio of the masks before and after correction exceeds a preset overlap threshold, the boundary correction reliability judgment is deemed to have failed. Figure 5 As shown, 801 represents the top free surface before boundary correction, 802 represents the top free surface after boundary correction, 803 represents the normalized average displacement, 804 represents the normalized maximum displacement, 805 represents the cross-union ratio of the masks before and after correction, 806 represents the state where the boundary correction magnitude meets the preset requirements, and 807 represents the state where the boundary correction reliability rejection condition is triggered.

[0042] In some embodiments, in step S3, the high-scoring candidate set is determined as follows: Candidate particle stack masks that have passed physical validity assessment after correction and can form effective bilateral slope candidate angle proxy values ​​constitute a measurable candidate set. These are sorted from highest to lowest final mask quality score. Using the highest final mask quality score as a benchmark, candidate particle stack masks with a final mask quality score not lower than the highest score minus a preset candidate consensus zone are determined as high-scoring candidates. The candidate angle proxy value is determined as follows: Bottom boundary points are extracted from the corrected mask of the corresponding high-scoring candidate, and robust straight-line fitting is performed to determine the direction and position of the temporary baseline. The temporary stack top and temporary stack height are determined based on the temporary baseline. The signed vertical distance from the free surface boundary point to the temporary baseline is normalized using the temporary stack height. Left and right slope straight lines are fitted within a preset temporary dimensionless relative stack height interval, and the arithmetic mean of the left and right proxy angles is calculated as the candidate angle proxy value. The candidate angle proxy value is only used for angle consistency judgment among high-scoring candidates and is not used as the official rest angle output of the current image. The candidate geometric consistency judgment is as follows: when there are at least two valid candidate angle surrogate values ​​in the high-scoring candidate set, the difference between the maximum and minimum values ​​of the valid candidate angle surrogate values ​​is calculated. If the difference exceeds a preset candidate angle consistency threshold, the candidate geometric consistency judgment is deemed to have failed. Figure 6 As shown, 906 represents the high-scoring candidate set, 907 represents the candidate angle range, 908 represents the geometrically consistent state and continues subsequent angle measurement, and 909 represents the geometrically inconsistent state and rejects the current image. When there are no measurable candidates, or the highest final mask quality score is lower than the preset candidate selection quality threshold, or the candidate geometric consistency judgment fails, or the boundary correction reliability judgment fails, the current static side view image is marked as invalid and its subsequent repose angle calculation is terminated. Furthermore, after the target mask is determined, if the final mask quality score of the target mask is lower than the preset result mask quality threshold, the current repose angle is rejected.

[0043] Step S4. Determine the actual bottom baseline between the particle and the bearing surface based on the bottom boundary point of the target mask. Define the vertical distance from the top of the stack to the actual bottom baseline as the stack height. Normalize the signed vertical distance from the free surface boundary point to the actual bottom baseline with the stack height to obtain the dimensionless relative stack height.

[0044] In some embodiments, in step S4, the actual bottom baseline is determined in two stages according to direction and position: first, a robust straight line fit is performed based on the bottom boundary points of the target mask to determine the baseline direction; then, under the determined baseline direction, the baseline position is determined based on the intercept distribution of the free surface boundary points; the stack height is the vertical distance from the stack apex to the actual bottom baseline; based on the signed vertical distance from the free surface boundary points to the actual bottom baseline, a continuous free surface segment containing the stack apex is identified, and the adjacent slope trends at both ends of the continuous free surface segment are extrapolated to the actual bottom baseline to determine the left and right stack feet. A pure stack mask is formed by the continuous free surface segment, the left stack foot, the right stack foot, and the actual bottom baseline. When the lateral span of the continuous free surface segment is lower than a preset span threshold, the stack height is lower than a preset stack height threshold, or the left and right stack feet cannot be determined, the pure stack mask construction is deemed a failure, and the rest angle of the current static side view image is rejected. The baseline fitting residual is calculated based on the free surface boundary points near the actual bottom baseline, and the baseline fitting residual is normalized using the stack height. When the normalized baseline fitting residual exceeds a preset baseline residual threshold, the rest angle of the current static side view image is rejected. Figure 4 As shown in (a), 701 represents the free surface, 702 represents the actual bottom baseline, and 703 represents the heap vertex; as Figure 4 As shown in (a) to (c), 704 represents the left stack foot, 705 represents the right stack foot, and W L and W R These represent the left and right candidate windows, respectively.

[0045] Step S5. Generate multiple consecutive candidate windows on the left and right slopes respectively. Form a candidate window pair by combining the left and right candidate windows whose differences in the lower and upper limits of dimensionless relative stacking height are not greater than the preset height matching tolerance. Calculate the joint evaluation score of the window pair based on the window quality of each candidate window, the degree of overlap of their dimensionless relative stacking height intervals, and the balance of their coverage lengths. Select the candidate window pair that passes the preset window pair quality threshold and has the highest joint evaluation score as the target window pair.

[0046] In some embodiments, in step S5, the window quality of each candidate window is calculated based on the fitting linearity of the slope points within the window, the coverage length, the dimensionless relative stack height span, the root mean square residual of the fitting normalized by the stack height, and the boundary roughness characterized by the standard deviation of the boundary residual normalized by the stack height. For candidate window pairs that meet the height matching condition, the joint evaluation score of the window pair is determined by the weighted sum of the average window quality of the left and right candidate windows, the crossover ratio of the left and right dimensionless relative stack height intervals, and the balance of the left and right coverage lengths. When there is no candidate window pair that meets the height matching condition, or when the highest joint evaluation score is lower than the preset window pair quality threshold, the rest angle of the current static side view image is rejected.

[0047] Step S6. Fit the left and right slope lines within the target window respectively, calculate the angle between each slope line and the actual bottom baseline, and determine the principal rest angle of the current static side view image based on the left and right angles.

[0048] In some embodiments, in step S6, the left and right slope lines are fitted using a robust line fitting method, which can be any one of random sampling consistency fitting, Theil-Sen fitting, or Huber fitting. The angle between the fitted left slope line and the actual bottom baseline is calculated as the left slope angle, and the angle between the fitted right slope line and the actual bottom baseline is calculated as the right slope angle. The arithmetic mean of the left and right slope angles is used as the principal angle of repose for the current static side view image. When the fitted slope line is nearly perpendicular to the actual bottom baseline, the minimum angle calculated based on the direction angles of the two lines is used instead of the slope-based angle calculation. Furthermore, within a preset fixed dimensionless relative stacking height range, line fitting is performed on the left and right slope points respectively, and the arithmetic mean of the resulting left and right angles is used as the standardized angle of repose. This standardized angle of repose and the aforementioned principal angle of repose are recorded separately for comparison between different slope selection methods.

[0049] After obtaining the final effective particle stack boundary and actual bottom baseline of the current static side view image, an equivalent isosceles triangle with a corresponding projected area and bottom width is constructed based on the effective two-dimensional projected area of ​​the same particle stack outline and the effective projected bottom width between the left and right stack feet, and the equivalent area repose angle is determined accordingly. The equivalent area repose angle is used to characterize the overall projected geometry of the particle stack in the current imaging direction, and together with the principal repose angle obtained based on the fitting of the left and right effective slopes, constitutes the particle stack geometry measurement result.

[0050] Step S7. Perform slope representativeness judgment based on the global-local angle difference between the principal angle of repose and the local slope representative angle, as well as the local slope angle dispersion. Combine this with slope fitting quality, left-right slope angle differences, and at least one independent anomaly from the single-map geometric angle interval to perform joint quality judgment. If the preset rejection condition is met, reject the current principal angle of repose; otherwise, save the current principal angle of repose. When the current angle of repose is rejected, do not write the corresponding angle into the formal valid result set, but save the obtained quality diagnostic quantity and rejection reason to achieve result rejection and quality traceability.

[0051] In some embodiments, in step S7, the method for determining the local slope representative angle is as follows: multiple local slope segments are formed along the free surface on the left and right slopes respectively; a straight line is fitted to each local slope segment; the local slope angle of each local slope segment relative to the actual bottom baseline is calculated; and the arithmetic mean of the median of the left local slope angle and the median of the right local slope angle is taken as the local representative angle; the global-local angle difference is the absolute difference between the principal angle of repose and the local representative angle; the local slope angle dispersion is the interquartile range of the combined set of the left and right local slope angles; when the global-local angle difference exceeds the preset global-local... When the warning threshold is reached and the local slope angle dispersion exceeds a preset local slope angle dispersion threshold, a slope representative difference indicator is formed. This indicator itself does not constitute a rejection condition on its own. In the joint quality judgment, when the principal angle of repose exceeds a preset extreme angle threshold and the global-local angle difference exceeds a preset auxiliary angle difference threshold or the local slope angle dispersion exceeds the preset local slope angle dispersion threshold, the current angle of repose is rejected. Alternatively, when the left-right slope angle difference exceeds a preset left-right slope angle difference threshold, and the global-local angle difference exceeds the preset auxiliary angle difference threshold and the local slope angle dispersion exceeds the preset local slope angle dispersion threshold, the current angle of repose is rejected.

[0052] In some embodiments, in step S7, the determination of the geometric angle interval of the single image is as follows: Multiple resampling with replacement is performed on the left slope point, right slope point, and baseline point used to fit the actual bottom baseline in the target window pair. For each set of resampling points, the left slope line, right slope line, and actual bottom baseline are refitted, and the left and right included angles and their arithmetic mean are recalculated. The geometric angle interval of the current static side view image is determined based on the preset lower quantile and preset upper quantile values ​​of the arithmetic mean obtained from multiple resampling. When the smaller value of the left and right slope fitting linearity is lower than a preset linearity threshold and the half-width of the geometric angle interval exceeds a preset interval half-width threshold, the current rest angle is rejected. When the global-local angle difference exceeds a preset global-local warning threshold and the local slope angle dispersion exceeds a preset local slope angle dispersion threshold, and the smaller value of the left and right slope fitting linearity is lower than a preset linearity threshold or the half-width of the geometric angle interval exceeds a preset interval half-width threshold, the current rest angle is rejected.

[0053] In some embodiments, the particle stack is located in a transparent liquid. Before lifting the bottomless forming cylinder 130, the particles are immersed in the transparent liquid, and after lifting, a preset defoaming and settling time is waited. When the projected area of ​​the bubble region in a single static side view image does not exceed a preset defoaming threshold, the image is allowed to proceed to subsequent processing. Under the condition that the liquid type, liquid depth, camera pose, optical interface position, and lighting conditions remain consistent, a same-medium calibration image, a cavity background image, and the static side view image located in the particle stack side profile measurement plane are acquired respectively. A geometric mapping is established based on the same-medium calibration image, and the cavity background image and the static side view image are corrected using the geometric mapping respectively. The corrected static side view image is normalized based on the corrected cavity background image, and a reflective area mask and a bubble region mask are generated. The two are combined into an external exclusion mask, and the region corresponding to the external exclusion mask is excluded when generating the candidate particle stack mask. Figure 7A As shown, the image acquisition unit 1001 is disposed on one side of the transparent water tank 140, and the image acquisition direction passes through the transparent tank wall 1002 to acquire calibration images or particle pile images in the transparent liquid medium 1003. The same medium calibration component 1004 is disposed in the transparent liquid medium 1003; as shown... Figure 7B As shown, 1005 represents the same medium calibration image, 1006 represents the empty tank background image, 1007 represents the static particle stack image, 1008 represents geometric mapping and correction, 1009 represents background normalization, 1010 represents the exclusion of reflective / bubble areas, and 1011 represents the corrected particle stack image.

[0054] like Figure 3As shown, the image processing flow mainly includes: inputting the static side view image into the multi-candidate mask generation module, generating multiple candidate particle stack masks using various segmentation strategies; performing physical validity judgment on each candidate particle stack mask; independently performing candidate cleaning and free surface boundary correction on each candidate that passes the physical validity judgment; after completing the boundary correction, performing physical validity judgment on each candidate again and recalculating the corrected final mask quality score; determining the high-scoring candidate set based on the corrected final mask quality score, performing candidate angle consistency judgment and boundary correction reliability judgment, and determining the target mask after passing the judgment; then proceeding to the actual bottom baseline determination and subsequent slope measurement process.

[0055] This invention also provides a device for measuring the angle of repose of a static particle pile formed by the lifting cylinder method, such as... Figure 1A , Figure 1B , Figure 2A and Figure 2B As shown, the device includes: a bearing surface 141 for bearing a static pellet pile 170; a bottomless forming cylinder 130, the lower end of which can be placed above the bearing surface 141; a lifting assembly (including a Z-axis lifting assembly 111) for driving the bottomless forming cylinder 130 to move vertically from the filling position to the stacking position; a clamping assembly 120, disposed at the moving end of the lifting assembly, for clamping the bottomless forming cylinder 130 and moving it with the lifting assembly; an image acquisition assembly (including a first image acquisition unit 150 and an optional second image acquisition unit 151) for acquiring at least one static side view image of the pellet pile 170 after the bottomless forming cylinder 130 has been lifted and a predetermined settling time has elapsed; and an image processing device, including a processor and a memory, the memory storing program instructions executable by the processor, the program instructions being used to cause the processor to execute the repose angle measurement method. The clamping assembly 120 may employ a pneumatic, electric, or other adjustable clamping mechanism capable of stably clamping the bottomless forming cylinder 130, and may include multiple adjustable clamping parts arranged circumferentially along the bottomless forming cylinder 130; the position where the clamping part contacts the bottomless forming cylinder 130 may also be provided with an anti-slip or buffer contact structure. The image processing device further includes a measurement recording and storage unit, used to associate and save image identifiers, enhancement parameters and resting time, each candidate particle stack mask and its indicators before and after correction, final mask quality score, candidate score difference, candidate angle surrogate value and the difference between its maximum and minimum values, average and maximum displacement of the top boundary normalized to image height, target mask, actual bottom baseline parameters, stack height, baseline fitting residual normalized to stack height, dimensionless relative stack height interval and joint evaluation score of the target window pair, angle of repose, and valid or rejected state, and save the corresponding rejection conditions when rejected.

[0056] like Figure 1AAs shown, the particle repose angle measuring device also includes a gantry 100, a motion assembly 110, an illumination assembly 160, and an optional transparent water tank 140. The gantry 100 supports the motion assembly 110 and the Z-axis lifting assembly 111. The motion assembly 110 is mounted on the gantry 100. The Z-axis lifting assembly 111 is connected to the clamping assembly 120 and is used to move the clamping assembly 120 and the clamped bottomless forming cylinder 130 vertically. Figure 1B As shown, in a two-sided image acquisition embodiment, the first image acquisition unit 150 and the second image acquisition unit 151 observe the particle pile 170 from different sides. The shooting directions of the two units are basically orthogonal, and static side view images of the same stationary particle pile 170 in two observation directions are obtained respectively.

[0057] like Figure 2A As shown, the clamping assembly 120 is connected to the Z-axis lifting assembly 111 via an upper connector 301, and includes a clamping body 302, a plurality of clamping arms 304 arranged around the clamping body 302, and corresponding connecting structures for connecting or adjusting the clamping arms 304. The clamping arms 304 are connected to the main structure of the clamping assembly 120 via connecting portions 303, and can be provided with multiple adjustment positions 305 to change the position of the clamping arm 304 or its end clamping structure relative to the clamping body 302, thereby adapting to bottomless forming cylinders 130 with different outer diameters. Figure 2A Local magnified structure and Figure 2B As shown, a clamping contact structure 306 is provided at one end of the clamping arm 304 near the bottomless forming cylinder 130, and an anti-slip or buffer contact layer 307 can be provided between the clamping contact structure 306 and the bottomless forming cylinder 130. The clamping contact structure 306 can be connected to the clamping arm 304 via fasteners 309. Adjacent components can be installed or movably connected via connectors 308.

[0058] The following further describes the implementation methods, algorithm examples, and advantages of specific embodiments of the present invention.

[0059] The following processing steps, calculation relationships, and parameter configurations are used to illustrate one possible implementation of the present invention, wherein each preset threshold, interval, and non-negative weight can be set or adjusted according to image resolution, particle characteristics under test, imaging conditions, and predetermined verification results.

[0060] In this embodiment, after the bottomless forming cylinder is lifted, a predetermined settling time is waited before acquiring a static side view image of the particle stack. In one embodiment, the predetermined settling time can be approximately 30 seconds to allow the particle stack to reach a predetermined settling state. When using a single shooting orientation, at least one static side view image can be acquired; when using two shooting orientations, static side view images of the same particle stack in a settling state from different shooting directions can be acquired, and subsequent repose angle measurements and quality judgments can be performed separately.

[0061] Taking the top left corner of the static side view image as the origin of the image coordinate system, and horizontally to the right as... The positive direction of the axis, vertically downwards is Positive direction of the axis. Free surface boundary points are denoted as... The actual bottom baseline is denoted as The heap vertex is denoted as The angle of repose is determined based on the angle between the fitted straight line of the slope and the actual bottom baseline. The horizontal coordinate axis of the image is not directly used as the default slope angle measurement reference.

[0062] This represents the image height of the entire static side view image; This represents the vertical distance from the top of the pile to the actual bottom baseline, which is the actual height of the pile after it is stacked. This indicates the initial filling height of the particles inside the bottomless forming cylinder; This indicates the diameter of the bottomless forming cylinder. The displacement of the top boundary before and after boundary correction is expressed as the image height. Normalization was performed; the dimensionless relative stack height of the free surface, the slope fitting residual, and the actual bottom baseline fitting residual were taken as the actual stack height. Normalize.

[0063] An example of executable parameter configuration

[0064] Table 1 provides a set of key parameter configurations for this embodiment. The listed values ​​are used to illustrate a specific implementation of the processing flow of this invention and do not constitute a fixed limitation on the corresponding threshold, interval, or weight values. The weights in the candidate mask quality score, single-sided window quality score, and joint evaluation score of window pairs can be determined using a pre-established calibration image set before formal measurement. The calibration image set may include static particle pile images obtained under different particle scales, particle morphologies, and predetermined imaging conditions, and the measurement results of the reference particle pile contour, reference free surface, and reference slope are obtained through manual verification or fine annotation. In one embodiment, under the condition that each positive weight and penalty weight satisfies a predetermined non-negative constraint, the weight combinations can be changed using grid search, random search, coordinate search, or other parameter optimization methods. The weights are determined based on at least one evaluation index among the degree of overlap between the selected candidate and the reference particle pile contour, free surface position deviation, slope angle error, and the degree of correspondence between the relative heights of the left and right measurement areas. This ensures that candidates with higher mask quality scores have higher consistency with the reference particle pile contour, and that the optimal results of single-sided windows and window pairs correspond to smaller slope measurement errors and better left-right relative height correspondence. Once the weights are determined, they can be used consistently under the same or predetermined allowable particle conditions and imaging configurations. When changes occur in image resolution, imaging magnification, illumination conditions, particle type, or media conditions that exceed the predetermined applicable range, the above calibration process can be re-executed. The weights and thresholds in Table 1 are a set of executable example values ​​set according to the above principles.

[0065] Table 1. Example of an executable parameter configuration.

[0066]

[0067] In this embodiment, when the global-local angle difference exceeds And the interquartile range of the local slope angle exceeds When the current result is marked as having a global-local slope representativeness difference, this condition alone does not cause the angle of repose to be rejected. The current angle of repose can be rejected when the representativeness difference is accompanied by at least one independent anomaly; the independent anomaly includes the principal angle of repose exceeding [a certain threshold]. The difference between the left and right slope angles exceeds The smaller of the determination coefficients for the left and right slopes is below 0.75, or the half-width of the 95% interval of the geometric angle of a single image exceeds [a certain value]. .

[0068] In a specific joint quality judgment method, when the principal angle of repose exceeds... And the global-local angle difference exceeds Or the interquartile range of the local slope angle exceeds When the angle of repose is less than 1, the current angle of repose can be rejected; when the difference between the left and right slope angles exceeds 100°, the angle of repose can be rejected. The global-local angle difference exceeds And the interquartile range of the local slope angle exceeds When the smaller of the determination coefficients for the left and right slopes is less than 0.75 and the half-width of the 95% interval of the geometric angle of a single image exceeds a certain value, the current angle of repose can be rejected. The current rest angle can also be rejected. The current rest angle is also rejected when the global-local angle difference exceeds 5.0°, the local slope angle interquartile range exceeds 6.0°, and the smaller of the left and right slope fitting determination coefficients is less than 0.75 or the half-width of the 95% interval of the geometric angle of a single image exceeds 3.0°.

[0069] Device Structure

[0070] like Figure 1A and Figure 1B As shown, the particle repose angle measuring device includes a gantry frame 100, a motion component 110, a Z-axis lifting component 111, a clamping component 120, a bottomless forming cylinder 130, an optional transparent water tank 140, a bearing surface or measuring platform 141, a first image acquisition unit 150, an optional second image acquisition unit 151, an illumination component 160, and an image processing device. After the bottomless forming cylinder 130 is lifted, the particles form a static particle pile 170 on the bearing surface or measuring platform 141. Figure 1A The diagram illustrates the relationship between the main mechanical body of the device and the acquisition of data from a single frontal view. Figure 1B The image acquisition setup is shown under different shooting positions, wherein the first image acquisition unit 150 and the second image acquisition unit 151 are respectively facing different lateral observation directions of the particle stack 170.

[0071] The gantry frame 100 supports the motion assembly 110 and the Z-axis lifting assembly 111. The motion assembly 110 is mounted on the gantry frame 100 and provides a base for the installation and position adjustment of the Z-axis lifting assembly 111. The Z-axis lifting assembly 111 is connected to the clamping assembly 120 and is used to drive the clamping assembly 120 and the clamped bottomless forming cylinder 130 to move vertically. The Z-axis lifting assembly 111 can be a lead screw linear module, a guide rail linear module, a synchronous belt linear module, or other linear motion mechanisms capable of achieving controlled vertical movement.

[0072] During the particle filling stage, the lower end of the bottomless forming cylinder 130 is placed on the bearing surface or measuring platform 141, and the particles are filled inside the bottomless forming cylinder 130. During the stacking stage, the Z-axis lifting assembly 111 drives the clamping assembly 120 and the bottomless forming cylinder 130 to move upward, so that the bottomless forming cylinder 130 gradually releases the lateral constraint on the particles. Under the action of gravity, the particles expand outward and form a static particle pile 170 on the bearing surface or measuring platform 141. The lifting speed and lifting distance can be set according to the particle size, medium conditions, bottomless forming cylinder size, and the working range of the device. In one specific embodiment, the lifting speed is about 2 mm / s and the lifting distance is about 30 cm.

[0073] like Figure 2A As shown, the clamping assembly 120 is connected to the Z-axis lifting assembly 111 via an upper connector 301, and includes a clamping body 302, a plurality of clamping arms 304 arranged around the clamping body 302, and corresponding connecting structures for connecting or adjusting the clamping arms 304. The clamping assembly 120 is used to clamp the bottomless forming cylinder 130 from the outside, allowing the bottomless forming cylinder 130 to move vertically with the Z-axis lifting assembly 111. Figure 2A and Figure 2B In one specific embodiment shown, the clamping assembly 120 includes four clamping arms 304 arranged circumferentially around the bottomless forming cylinder 130.

[0074] The clamping arm 304 is connected to the main structure of the clamping assembly 120 via the connecting part 303, and can be provided with multiple adjustment positions 305 to change the position of the clamping arm 304 or its end clamping structure relative to the clamping body 302, thereby adapting to bottomless forming cylinders 130 with different outer diameters. Adjacent components can be installed or movably connected through the connector 308. Figure 2A The specific number and geometry of the clamping arms 304, connecting parts 303, and adjusting positions 305 shown are used to illustrate an adjustable clamping structure. In other embodiments, other numbers or forms of adjustable clamping parts may be used depending on the diameter range of the bottomless forming cylinder 130.

[0075] like Figure 2A As shown in the enlarged view, a clamping contact structure 306 is provided at one end of the clamping arm 304 near the bottomless forming cylinder 130. An anti-slip or buffering contact layer 307 can be provided between the clamping contact structure 306 and the bottomless forming cylinder 130 to increase contact stability and reduce the impact on the surface of the bottomless forming cylinder 130 during clamping. The clamping contact structure 306 can be connected to the clamping arm 304 by fasteners 309 and can be replaced or repositioned as needed.

[0076] like Figure 2BAs shown, multiple clamping arms 304 are arranged circumferentially along the bottomless forming cylinder 130, and apply clamping action from the outside of the bottomless forming cylinder 130 toward the cylinder body. In the specific embodiment shown, the four clamping arms 304 are distributed approximately along two intersecting directions to form multi-point clamping of the bottomless forming cylinder 130. The multiple clamping arms 304 can be adjusted independently or in a linked manner to adapt to bottomless forming cylinders 130 of different sizes. The four-clamping arm structure is a specific embodiment and does not constitute the only structural limitation of the clamping assembly 120.

[0077] The bottomless forming cylinder 130 is a cylindrical component with an open lower end, which maintains a predetermined shape during particle filling and lifting. The diameter (D) of the bottomless forming cylinder 130 can be selected according to the particle size and measurement requirements. In one embodiment, the diameter (D) can be selected in the range of 2 cm to 20 cm, and the initial particle filling height (Hfill) can be set to approximately 1.5D. For bottomless forming cylinders 130 of different diameters, adaptation can be achieved by adjusting the position of the clamping arm 304, changing the position of the clamping contact structure 306, or selecting a clamping assembly 120 of appropriate size.

[0078] When using a transparent liquid medium for measurement, the bearing surface or measuring stage 141, the bottomless forming cylinder 130, and the particle pile 170 can be placed inside the transparent water tank 140, so that the particle pile 170 is immersed in water or other transparent liquid. A transparent observation area can be formed on the side of the transparent water tank 140 facing the image acquisition unit, allowing the first image acquisition unit 150 or the second image acquisition unit 151 to acquire a static side view image of the particle pile 170 from the side. The transparent water tank 140 is an optional component; it can be omitted in the measurement of dry particles in air.

[0079] The image acquisition assembly includes at least a first image acquisition unit 150. The first image acquisition unit 150 may be an industrial camera or an industrial high-speed camera, and is positioned to the side of the particle stack 170. Its imaging range covers the free surface of the particle stack 170, the left and right slopes, and the contact area between the particle stack 170 and the supporting surface or measuring table 141. In one specific embodiment, the working distance between the first image acquisition unit 150 and the particle stack 170 is approximately 30 cm. During the same set of measurements, the position and imaging direction of the image acquisition unit relative to the supporting surface or measuring table 141 remain substantially unchanged.

[0080] like Figure 1BAs shown, in one embodiment of bidirectional image acquisition, a second image acquisition unit 151 can be further provided. The first image acquisition unit 150 and the second image acquisition unit 151 observe the particle pile 170 from different sides. Preferably, the shooting directions of the two units are substantially orthogonal, thereby obtaining static side view images of the same stationary particle pile 170 in two observation directions. The images from the two shooting directions are subjected to independent particle pile contour extraction, repose angle calculation, and result quality judgment. When effective repose angles are obtained from both directions, the difference between the repose angles of the two directions can be further recorded to characterize the geometric differences of the particle pile 170 in different observation directions.

[0081] like Figure 1A As shown, the illumination assembly 160 can be disposed on both sides of the particle stack 170. In one embodiment, the illumination assembly 160 employs a surface light source to improve the image contrast between the particle stack 170 and the background and reduce obvious unilateral shadows. During the same set of measurements, the position of the illumination assembly 160 and the main illumination conditions remain substantially consistent; in measurements with transparent liquid media, the position of the illumination assembly 160 can also be adjusted according to the optical characteristics of the transparent water tank 140 and the liquid medium.

[0082] After the bottomless forming cylinder 130 is lifted, the particle pile 170 is allowed to stand for a predetermined time before image acquisition. In one specific embodiment, the standing time is approximately 30 seconds. In this specific configuration, particles are placed in the bottomless forming cylinder 130 at a filling ratio of approximately Hfill / D = 1.5. The bottomless forming cylinder 130 is held by the clamping assembly 120 and vertically lifted approximately 30 cm at a speed of approximately 2 mm / s by the Z-axis lifting assembly 111. After the particle pile 170 is formed, it is left to stand for approximately 30 seconds, and then a static side view image is acquired by an industrial camera positioned to the side of the particle pile 170 at a working distance of approximately 30 cm. The above parameters illustrate one feasible configuration and can be adjusted according to actual measurement conditions.

[0083] The image processing device includes a processor and a memory. The memory stores program instructions for executing the particle repose angle measurement method and can save image data and quality evaluation information generated during the measurement process. For each static side view image, the device can associate and save image identifier, lifting parameters, settling time, image acquisition conditions, candidate particle stack mask and its boundary correction evaluation indicators, final mask quality score, candidate angle surrogate value and its dispersion, candidate competition status information, target mask, actual bottom baseline parameters, stack height (H), target window pair and its evaluation information, left slope angle, right slope angle, principal repose angle, equivalent area repose angle, and result quality status. When the current image does not meet the predetermined quality conditions, the device can also associate and save the corresponding abnormal reasons or rejection reasons.

[0084] Static Image Analysis Process

[0085] After static image acquisition, the static image analysis process begins. This process includes particle loading, bottomless forming cylinder lifting, predetermined settling time, static side view image acquisition, static image analysis, and result saving or rejection. Each static side view image is treated as an independent analysis unit; it is only saved as a valid angle of repose result if the corresponding image processing result meets predetermined quality conditions. Even if the numerical angle can be calculated from the image, it is rejected as a valid measurement result if the quality conditions are not met. The core image processing relationships are as follows: Figure 3 As shown.

[0086] like Figure 3 As shown, a static side view image is input into a multi-candidate mask generation module, which generates multiple candidate particle stack masks using various segmentation strategies. These segmentation strategies may include global threshold segmentation, adaptive threshold segmentation, and background correction threshold segmentation. For cases where the grayscale polarity of the particle region relative to the background cannot be predetermined, positive polarity candidates and negative polarity candidates can be generated separately for the same segmentation result. Other image segmentation methods capable of forming candidate particle stack regions can also be used as additional candidate generation methods, with each candidate independently entering the subsequent physical validity judgment and quality evaluation process.

[0087] Physical validity is assessed for each candidate particle stack mask. This assessment includes at least the proportion of the candidate's foreground area to the image area, the width and height of the candidate's enclosing region, the degree of support contact between the candidate and the predetermined bottom region, the effective coverage of the top free surface, the apparent height and connectivity between the top and bottom of the stack, and the degree of overlap between the candidate and a predetermined interference region. If the candidate's foreground area proportion exceeds a preset range, the enclosing region size is lower than a preset size requirement, the bottom support contact is insufficient, the top free surface coverage is insufficient, an effective top-to-bottom geometric relationship is not formed, or the overlap with the interference region exceeds a preset range, the corresponding candidate is deemed physically invalid and is not included in the target mask selection.

[0088] The candidate masks are scored based on their physical validity. Positive quality components include foreground area score, bottom support contact score, main connected region ratio, outer contour continuity score, top boundary validity score, top boundary coverage score, bottom flatness score, aspect ratio of the enclosed region, top-to-bottom topology score, and enclosed region scale score. Penalty components include internal hole penalty and interference region overlap penalty. Quality score of each candidate particle stack mask It can be expressed as follows:

[0089] in, For the first A normalized positive mass component For the first A normalized penalty component, and These are the corresponding non-negative weights. This indicates that the obtained quality score will be limited to the range of 0 to 1. Used for candidate comparison and quality judgment, the numerical value does not represent the probability that the corresponding rest angle is the correct result. The weights corresponding to each positive quality component and penalty component can be determined according to the aforementioned parameter calibration method, based on the geometric consistency between the candidate mask and the reference particle pack contour in the calibration image set. In one embodiment, different weight combinations can be changed and the quality score of each candidate can be calculated separately. The degree of consistency between the quality score ranking result and the reference particle pack contour quality ranking result is used as the parameter selection criterion, and the weights used for formal measurement are determined from the weight combinations that meet the predetermined consistency requirements.

[0090] Candidate cleaning is performed independently on each candidate that passes the physical validity assessment. Candidate cleaning may include preserving the main connected regions corresponding to the spatial location of the particle stack, repairing pores within the main body that meet preset area conditions, and removing discrete noise and local defects using morphological opening and / or closing operations. The cleaning process for each candidate is independent of each other; no candidate is pre-selected for subsequent processing based solely on the initial quality score.

[0091] Free surface boundary correction is performed on each cleaned candidate. Free surface boundary correction can be achieved through graph cut processing that constructs constraints based on candidate foreground and background information, or by redetermining the free surface position within a limited search area near the initial free surface based on image edges or grayscale changes. After boundary correction, physical validity is assessed again for each candidate, and the corrected final mask quality score is recalculated. Subsequent candidate ranking uses the final mask quality score after boundary correction, rather than the candidate ranking result before boundary correction.

[0092] Record the mask before correction during the candidate boundary correction process. Corrected mask , Correction of top boundary ordinate Corrected top boundary ordinate and the common set of x-coordinates of both having valid top boundaries. When the set If there are not enough common valid boundary locations, the corresponding boundary correction evaluation quantity will be determined as invalid and processed according to the predetermined quality rules, instead of replacing the missing boundary correction evaluation result with zero displacement.

[0093] Correction of front mask With the corrected mask intersection ratio Calculate using the following formula:

[0094] in, This indicates the number of pixels contained in the corresponding region. The Cross-Union Ratio (CUI) is used to characterize the overall overlap of the particle stack mask before and after boundary correction, and is used together with the local boundary displacement to evaluate the reliability of boundary correction. The CUI itself is not interpreted as the accuracy of the angle of repose measurement.

[0095] Normalized average displacement of the top free surface before and after correction and normalized maximum displacement Calculate using the following formula:

[0096]

[0097] in, For the common valid x-coordinate set The number of elements in This is the image height of the entire static side view image. Therefore, and All are dimensionless boundary corrections. This is used here. As a normalization scale; subsequent actual stacking height Normalization of free surface relative stack height, slope fitting residual, and actual bottom baseline residual.

[0098] like Figure 5 As shown, a reliability assessment of boundary correction is performed on the finally selected target candidate, where: (a) shows a reasonable boundary correction, and (b) shows a large local boundary offset. 801 represents the top free surface before boundary correction. 802 indicates the top free surface after boundary correction. 803 represents the normalized average displacement. 804 represents the normalized maximum displacement. 805 indicates the cross-union ratio of the masks before and after boundary correction. 806 indicates that the boundary correction magnitude meets the preset requirements, and 807 indicates that the boundary correction reliability rejection condition is triggered. In one implementation, when Exceeding the preset average displacement threshold When the boundary correction magnitude is abnormal, or when Exceeding the preset maximum displacement threshold Furthermore, the cross-section of the masks before and after the correction is compared. Higher than the preset high overlap threshold If a local free surface aberration is detected while the overall mask still maintains a high degree of overlap, the current candidate mask is rejected from the formal angle of repose calculation. Area changes caused by boundary corrections can be recorded as diagnostic information, but in the above quality judgment of the target mask, they will not be directly rejected simply because the area change exceeds a certain value, so as to avoid excluding reasonable corrections that expand from a locally incomplete coarse mask to a complete particle stack region.

[0099] After correction, candidates that pass the physical validity test and can form valid surrogate values ​​for both sides of the slope will constitute a measurable candidate set. The images are sorted from highest to lowest based on their final mask quality score. If no measurable candidate exists, the current static side view image is rejected. Let the highest final mask quality score among the measurable candidates be denoted as . When a second measurable candidate exists, the second highest final mask quality score is... The highest final mask quality score is used as the benchmark, and the results are determined based on the pre-defined candidate consensus banding. A high-scoring candidate set is determined. This process excludes physically valid candidates that cannot form an effective geometric representation of the two-sided slope from the final target candidate competition.

[0100] like Figure 6 As shown, a proxy value for the candidate angle is determined for each high-scoring candidate. Figure 6 In the text, (a) corresponds to high-quality candidates with consistent geometric meaning, while (b) corresponds to high-quality candidates with inconsistent geometric meaning. 901 and 902 represent two high-scoring candidate correction masks under the condition of consistent candidate geometry, while 903 and 904 represent two high-scoring candidate correction masks under the condition of inconsistent candidate geometry; 906 represents the set of high-scoring candidates. 907 represents the candidate angle range calculated based on the high-scoring candidate angle surrogate value. 908 indicates a state where the candidate geometry is consistent and subsequent angle measurement continues, while 909 indicates a state where the candidate geometry is inconsistent and the current image is rejected. For the j-th high-resolution candidate, the bottom boundary points are first extracted from its corrected candidate mask. For each valid horizontal coordinate within the effective horizontal range of the candidate, the lowest foreground position belonging to the candidate particle cluster region in the corresponding column can be determined as the candidate bottom boundary point, and bottom boundary points located within a pre-determined interference area are excluded. Robust line fitting is applied to the obtained valid bottom boundary points to determine the direction of the temporary baseline; the robust line fitting can employ random sampling consistency fitting, Theil-Sen fitting, Huber fitting, or other fitting methods that can reduce the influence of a small number of abnormal bottom boundary points. After the temporary baseline direction is determined, the temporary baseline position can be determined based on the intercept distribution of the effective bottom boundary fitting inliers in that direction; in one embodiment, the median or a preset quantile position of the intercept can be used to determine the temporary baseline intercept. Thus, the temporary baseline corresponding to the j-th high-resolution candidate is obtained. The signed vertical distance from each point on the candidate top free surface to the temporary baseline is calculated. The free surface position corresponding to the largest effective distance is determined as the temporary top of the stack, and this largest distance is defined as the temporary stack height. The signed vertical distance from each free surface point to the temporary baseline is divided by the temporary stack height to obtain the temporary dimensionless relative stack height used for calculating the candidate angle surrogate value. Left and right slope points are selected within a preset temporary dimensionless relative stack height range, and straight line fitting is performed. The candidate angle surrogate value is determined based on the arithmetic mean of the angles between the left and right fitted straight lines and the temporary baseline. In one specific embodiment, the slope selection range for the candidate angle surrogate value can be 0.22 to 0.85 using a dimensionless relative stack height. The candidate angle surrogate value is only used for geometric consistency comparison between high-resolution candidates and is not used as the formal rest angle output of the current image, nor is it used to replace the subsequent formal slope fitting results obtained based on the target window. When the number of valid bottom boundary points used for temporary baseline fitting is less than a preset number, the lateral coverage of the bottom boundary is less than a preset coverage threshold, or the residual of the temporary baseline fitting exceeds a preset temporary baseline residual threshold after temporary stacking normalization, a valid candidate angle surrogate value for the high-scoring candidate is not generated. The temporary baseline is only used for rapid geometric consistency evaluation of candidate angle surrogate values ​​and is distinguished from the actual bottom baseline determined subsequently for the target mask.

[0101] Candidate score difference High-scoring candidate set M H Extreme difference in candidate angles It can be determined according to the following formula:

[0102]

[0103]

[0104] in, It is used to characterize the scoring advantage of the highest-scoring candidate relative to the second-highest-scoring candidate, and is recorded as a diagnostic measure of the candidate competition state, but is not used as a rejection condition for the current image alone. It is the candidate angle surrogate value determined by the mask, in order to Indicates the corrected number j A candidate particle stack mask, as indicated by the following: a , b Mark two exemplary candidates: This illustrates that multiple high-scoring candidates each correspond to their respective proxy values. When the high-scoring candidate set... There are at least two valid candidate angle proxy values ​​and Exceeding the preset candidate angle consistency threshold When a significant geometric divergence is found among multiple high-quality candidates, the current static side view image is rejected; when No more than If only one valid candidate perspective proxy value exists, a false candidate consensus value is not constructed; instead, the evaluation is continued based on other quality assessment results of that candidate.

[0105] When there is at least one measurable candidate with the highest final mask quality score Not lower than the preset candidate selection quality threshold If the candidate angle consistency judgment is passed and the boundary correction reliability judgment of the highest-scoring candidate is passed, the highest-scoring candidate is determined as the target mask and enters the actual bottom baseline determination and subsequent slope measurement process; if any necessary quality condition is not met, the current static side view image is rejected and the corresponding rejection reason and the obtained quality diagnosis information are saved. Candidate score difference It is always stored as candidate competition status information, and is not used to trigger a valid or invalid determination on its own.

[0106] Actual bottom baseline, relative stacking height and combined left and right slope selection

[0107] like Figure 4 As shown in (a), the free surface 701 and bottom boundary information of the particle stack are extracted from the target mask. First, the contact direction between the particle and the bearing surface is determined based on the bottom boundary points. Then, the contact position of the particle is determined based on the intercept distribution of the free surface boundary points under the determined direction, thus obtaining the actual bottom baseline 702. This processing allows the subsequent slope angle calculation to use the actual contact reference of the particle stack itself, instead of directly using the horizontal coordinate axis of the image or the lowest pixel row of the target mask as the slope angle measurement reference. Figure 4 M in This refers to the target mask determined after the aforementioned candidate quality assessment.

[0108] In one specific implementation, a robust straight-line fit is performed on the bottom boundary points of the target mask to obtain the directional slope of the actual bottom baseline 702. After fixing the direction, for a free surface point Calculate its corresponding intercept The actual bottom baseline intercept is determined based on the intercept distribution near the particle contact point. In one executable implementation, the position of the high quantile of the intercept distribution can be determined based on a preset high quantile position and a predetermined contact plane offset. This yields the actual bottom baseline 702, and under this baseline, the heap vertex 703 and the left and right contact boundaries are further determined.

[0109] Let the actual bottom baseline be 702. The heap vertex is 703. The first on free surface 701 The points are The actual bottom baseline, the signed vertical distance from the free surface point to the baseline. Actual stacking height and dimensionless relative stacking height Determine using the following formula:

[0110]

[0111]

[0112]

[0113] In the coordinate system with the vertical axis pointing downwards, the effective free surface point located above the actual bottom baseline 702 corresponds to... Actual stacking height This represents the vertical distance from heap vertex 703 to the actual bottom baseline 702, relative to the image height of the entire static side view image. They have different meanings.

[0114] Based on the distance distribution of the free surface 701 relative to the actual bottom baseline 702, a continuous main pile free surface segment containing the pile apex 703 is identified. At the left and right ends of this continuous segment, the left pile foot 704 and right pile foot 705 are determined using the intersection of the adjacent slope trend and the actual bottom baseline 702, respectively. The free surface 701, left pile foot 704, right pile foot 705, and actual bottom baseline 702 form an effective region corresponding only to the main particle pile, excluding the influence of the supporting structure and the foreground below the actual bottom baseline on subsequent geometric calculations. If a continuous main pile region with sufficient lateral span and actual pile height cannot be obtained, or if the left and right pile feet cannot be effectively determined, the current static side view image is rejected.

[0115] To evaluate the reliability of the actual bottom baseline 702, the root mean square residual relative to the corresponding contact reference line can be calculated based on the effective free surface points near the particle contact area, and this residual can be divided by the actual stack height. The normalized baseline residuals are obtained. .when Exceeding the preset baseline residual threshold At that time, the current static side view image is rejected. The baseline residual is based on the actual stack height. Normalize instead of using image height .

[0116] To generate comparative results with a unified geometric definition, the left and right slopes can be fitted separately within a preset fixed dimensionless relative stacking height range to obtain standardized left slope angle, standardized right slope angle, and standardized angle of repose within the fixed height range. Furthermore, this embodiment also adaptively selects the target measurement window based on the correspondence between the local mass and relative height of the left and right slopes; the results from the fixed height range and the adaptive window results can be recorded separately for comparison between different slope selection methods.

[0117] Result quality gating and geometric interval

[0118] like Figure 4 As shown in (a) to (c), multiple consecutive candidate windows are generated along the free surface 701 on both the left and right slopes, respectively. Figure 4 The diagram in the middle and These represent the left and right candidate windows, respectively. Each candidate window records its dimensionless lower limit of relative heap height. Relative stacking height limit Relative stacking height span The parameters include: number of slope points, lateral coverage length, linearity of fit, fitting residual, and boundary roughness.

[0119] For each candidate window, a linear fit is first used to obtain the basic geometric evaluation quantity within the window. Let... The coefficient of determination for the straight line fitting within the window. This is the normalized value of the window's lateral coverage length relative to the lateral span of the corresponding complete candidate slope. Let S be the dimensionless relative stacking height span of the window. To fit the root mean square residual to the actual stack height The ratio, To fit the residual standard deviation to the actual stack height The ratio. Define the truncation function.

[0120] The quality score for a single-sided window under an executable configuration. Calculate using the following formula:

[0121] Each coefficient in the formula represents a specific executable configuration. In one implementation, each coefficient can be determined based on a reference slope in the calibration image according to the aforementioned parameter calibration method: for different candidate windows, the fitting linearity, coverage, dimensionless relative height span, normalized fitting residual, and boundary roughness are calculated, and the error between the slope angle obtained by each candidate window and the reference slope angle is calculated; within a preset weight value range, the combination of each weight is changed so that the candidate windows with higher quality scores generally correspond to smaller slope angle errors, thereby determining the corresponding weights. The coefficients listed in the above formula are a set of executable example values ​​that satisfy the above evaluation relationship and do not constitute a fixed limitation on the value of each weight.

[0122] Let the dimensionless relative stack height interval of the left candidate window be... The dimensionless relative stack height range of the candidate window on the right is The preset left and right height matching tolerance is 1. Only when both conditions are met

[0123] and

[0124] Only the left and right windows together form a candidate window pair, ensuring that the actual measurement areas on both sides participating in slope fitting are within the corresponding dimensionless relative height range. Windows that independently achieve the highest single-sided quality score are not directly equivalent to a valid candidate window pair.

[0125] For candidate windows that meet the high matching criteria, the degree of overlap between relative stack height intervals is further calculated. and coverage length balance Let the normalized coverage lengths of the left and right candidate windows be respectively... and Then a specific calculation relationship and window are applied to the joint evaluation score. As shown in the following formula:

[0126]

[0127]

[0128] in, and These are the single-sided window quality scores for the left and right candidate windows, respectively. This indicates the balance of window coverage length, to distinguish it from the effective projected bottom width in the calculation of the equivalent area angle of repose later. The weights used in the above formula to characterize the average quality of the left and right windows, the degree of overlap between relative stack height intervals, and the degree of balance in coverage length can also be determined according to the aforementioned parameter calibration method. In one embodiment, the weight combinations can be changed for a calibration image with reference slope measurement results, and at least one of the following—the degree of correspondence between the left and right relative heights of the selected window pairs, the slope angle error, and the success rate of selecting effective window pairs—can be used as evaluation indicators to determine the weight combination that meets the predetermined evaluation requirements. The weights listed in the above formula represent a specific executable configuration.

[0129] For all candidate window pairs that satisfy the left-right height matching condition, calculate the joint evaluation score shown in the above formula, and select the window pair with the highest joint evaluation score. When their joint evaluation score... Not lower than the preset window quality threshold If no candidate window pair meets the high matching condition, or if the highest joint evaluation score is still lower than the quality threshold of the window pair, then the rest angle result of the current static side view image is rejected.

[0130] Robust straight-line fitting is performed on the left and right slope points within the target window pair. The robust straight-line fitting can employ random sample consensus fitting, Theil-Sen fitting, Huber fitting, or other robust fitting methods that can reduce the influence of a small number of outlier boundary points. For each side's fitting results, the fitting slope, intercept, and coefficient of determination are saved. Evaluation information includes root mean square residuals, proportion of interior points, and number of slope points involved in the fitting.

[0131] Let the slopes of the fitted lines for the left and right slopes be respectively and The actual slope of the bottom baseline 702 is The angle between any fitted line on the slope and the actual bottom baseline 702 can be calculated based on the angle relationship between the two lines. Left slope angle. Right side slope angle and the principal angle of repose of the current static side view image Determine using the following formula:

[0132]

[0133]

[0134] Each angle is converted to a degree format for output. When the slope is close to zero and cannot be directly expressed as a slope, the minimum angle between the fitted slope line and the actual bottom baseline can be calculated separately. (Difference between left and right slope angles) It can be used as auxiliary information to judge the degree of asymmetry between the left and right slopes of the particle pile and the quality of the results.

[0135] In addition to the principal angle of repose obtained by fitting the effective slopes on the left and right sides, the overall equivalent area angle of repose can also be calculated based on the same final effective particle pile profile. Let the abscissas of the left pile foot 704 and the right pile foot 705 be respectively... and The actual bottom baseline 702 is represented as

[0136] Free surface on the x-axis The ordinate of the location is represented as Effective two-dimensional projected area Effective projection bottom width and the height of the equivalent isosceles triangle Determine using the following formula:

[0137]

[0138]

[0139] in, This represents the two-dimensional projected area between the final effective free surface and the actual bottom baseline in the current image coordinate system. The portion below the actual bottom baseline is not included in this area. This represents the effective projection span of the left and right stack feet on the horizontal coordinate of the current image. The area and bottom width are determined using the final effective free surface, actual bottom baseline, and left and right stack feet obtained from the principal angle of repose measurement, without using another set of segmentation results.

[0140] In the discrete implementation of digital images, the final effective free surface points located between the left and right stack feet are arranged in ascending order of their horizontal coordinates. Let the effective free surface sampling points be... The corresponding points of the left and right stack feet on the actual bottom baseline are added to the ordered point list. For each sampling point, its effective vertical height relative to the actual bottom baseline is calculated.

[0141] By using trapezoidal integrals to discretize adjacent valid sampling points sorted by their x-coordinates, the effective two-dimensional projected area can be calculated as follows:

[0142] When multiple valid free surface points exist at the same abscissa position, the free surface point on the outer envelope side of the particle stack is selected as the valid free surface point corresponding to that abscissa. In a coordinate system with the ordinate pointing downwards, the free surface point on the outer envelope side is the point with the smaller ordinate. When there are no valid free surface sampling points at some abscissa positions, the trapezoidal integration is performed according to the adjacent existing valid sampling points, without reconstructing free surface points for each integer abscissa. For sampling points located below the actual bottom baseline, their corresponding valid vertical height is set to zero. When the number of valid free surface points is insufficient to form a valid integration interval, or when the obtained valid two-dimensional projected area or valid projected bottom width does not meet the preset validity conditions, the corresponding overall equivalent area repose angle is not output.

[0143] The effective projected area determined by the above formula and effective projection bottom width If the equivalent isosceles triangle has the same area and base width, then the base angle of this equivalent isosceles triangle is defined as the angle of repose of the overall equivalent area. The calculation relationship is determined by the following formula:

[0144] in, Convert to angle-based output.

[0145] The This is used to assist in characterizing the particle stack from the perspective of overall two-dimensional projection geometry, and is compared with the results obtained based on fitting the effective left and right slopes. , and Both should be preserved. The principal angle of repose result remains the same. The equivalent area rest angle does not modify the fitting results for the left and right slopes, nor does it... and Set additional mandatory difference thresholds or independent rejection conditions between them.

[0146] The difference in left and right slope angles is stored as diagnostic information regarding the degree of asymmetry and measurement quality of the left and right slopes of the particle pile. A single difference in left and right slope angles does not necessarily indicate that the current static side view image is invalid. For the left and right slopes covered by the target window, multiple local slope segments are formed along the free surface. Linear fitting is performed on each local slope segment, and its local slope angle relative to the actual bottom baseline is calculated to evaluate the representativeness of the overall slope fitting based on the target window to the local slope geometry.

[0147] Let the set of local slope angles corresponding to each local slope segment on the left be . The set of local slope angles corresponding to each local slope section on the right is as follows: Their respective medians are and Local representative angle Global-local angle difference Local slope angle dispersion and the difference in left and right slope angles Determine using the following formula:

[0148]

[0149]

[0150]

[0151] in, The interquartile range represents the combined set of left and right local slope angles; This indicates the degree of difference between the principal angle of repose and the representative angle of a local slope. It indicates the degree of dispersion of the local slope angle and can be used to reflect the degree of local nonlinearity or irregularity of the slope surface; This indicates the difference in angle between the left and right slopes.

[0152] In a specific quality judgment configuration, when the global-local angle difference... Exceed And the degree of dispersion of local slope angle Exceed When the current result is marked as having a significant difference in overall-local slope representativeness, this condition itself is saved as a slope representativeness quality warning and does not alone cause the current angle of repose to be rejected. When this quality warning is further accompanied by other independent fitting or geometric anomaly information, the decision to reject the current result is then made based on the preset joint quality conditions.

[0153] Let the smaller of the determination coefficients for fitting the left and right slopes be 1.

[0154] The half-width of the geometric angle interval of the single graph described later is In one specific implementation, extreme angle joint anomalies can be defined separately. Left-right asymmetric joint anomaly Fitting-Geometric Joint Anomalies Combined anomalies with nonlinear slopes The quality of the results is judged according to the following formula:

[0155]

[0156]

[0157]

[0158]

[0159] in, Indicates the logical "and". This indicates the logical "OR". The final expression is highlighted in a box, serving as a summary of the overall logical judgment. When If true, reject the principal angle of repose result of the current static side view image; otherwise, save the corresponding quality indicators and warning information, and continue to process the current result as a valid measurement result. The above threshold is used to describe a specific executable configuration, which can be adjusted according to a predetermined verification scheme.

[0160] In the above formula, Used to identify situations where the overall angle of repose is abnormally large and there are also local slope anomalies. Used to identify situations where there is obvious asymmetry between the left and right slopes and abnormalities exist in both the overall and local slope structures. It is used to identify situations where the linearity of slope fitting is low and the geometric uncertainty of a single graph is also large. This is used to determine whether to reject an image when there is already a significant difference between global and local representativeness, combined with low fitting linearity or a large range of single-image geometric angles. Therefore, when a large global-local angle difference or a large local slope angle dispersion exists alone, the true nonlinear particle pile slope surface is not directly judged as an invalid image.

[0161] To evaluate the geometric uncertainty caused by fitting the slope points and the actual bottom baseline in the same static side view image, the left slope points, right slope points, and baseline points used for fitting the actual bottom baseline in the target window pair are resampled with replacement. Each resampling refits the left slope line, right slope line, and actual bottom baseline, and recalculates the left slope angle, right slope angle, and their arithmetic mean. In one specific implementation, the number of geometric resampling operations is... It can be set to 96 times.

[0162] Let the first The average angle of repose obtained from each effective geometric resampling is: The number of valid resampling results is Based on the effective resampling angle set The 2.5% quantile and 97.5% quantile determine the lower and upper limits of the geometric angle interval for a single graph, respectively, and the half-width of the geometric angle interval is determined based on these two values, as expressed by the following formula:

[0163]

[0164]

[0165] in, and These represent the 2.5% quantile and the 97.5% quantile, respectively. The single-image geometric angle interval reflects the geometric fitting changes caused by the resampling of slope points and baseline points in the same static side view image, and is not equivalent to the experimental repeatability interval obtained from multiple independent stacking tests.

[0166] When the number of valid resampling results is insufficient to form a valid geometric angle interval, the zero value or an artificially constructed interval is not used as a substitute. Instead, the current result is judged in conjunction with other already obtained slope fitting quality and geometric quality information. In one specific implementation, when and At that time, it was considered that the current results simultaneously exhibited low slope fitting quality and large single-graph geometric fitting uncertainty, based on the aforementioned... The condition rejects the current angle of rest.

[0167] For static side view images that pass the above quality assessment, the principal angle of repose results should include at least the left slope angle. Right side slope angle and the main angle of repose obtained by the arithmetic mean of the two. Auxiliary geometric information may include the equivalent area repose angle. Actual stacking height Effective projection bottom width Difference between left and right slope angles Global-local angle difference Local slope angle dispersion And the geometric angle range of a single image. Quality traceability information may also include candidate score differences. Extremely poor candidate angles , boundary correction before and after mask intersection and union ratio, with The data includes the normalized top boundary average and maximum displacement, actual bottom baseline parameters, normalized baseline residuals, target window joint evaluation score, and valid, warning, or rejection status. When the current image is rejected, the corresponding angle is not written to the official valid results set, but the obtained quality diagnostics and reasons for rejection are saved.

[0168] Optional transparent liquid medium and two-way implementation method

[0169] like Figure 7A and Figure 7B As shown, in the optional transparent liquid medium embodiment, the particles, bottomless forming cylinder, and bearing surface or measuring platform can be placed in a transparent water tank, and the cylinder can be piled up and static images can be acquired in water or other transparent liquids. Figure 7A The image acquisition unit 1001 is shown to capture images from the same shooting position under both a calibration state and a static particle pile measurement state. The image acquisition unit 1001 is positioned on one side of a transparent water tank, and the image acquisition direction passes through the transparent tank wall 1002 to acquire a calibration image or particle pile image in the transparent liquid medium 1003. Under the calibration state, a calibration element 1004 is placed in the transparent liquid medium 1003. In both the calibration state and the static particle pile measurement state, the liquid type, liquid depth, transparent tank wall and related optical interfaces, the pose of the image acquisition unit 1001, and the main illumination conditions are kept consistent or meet predetermined consistency requirements, so that the calibration image and the actual particle pile image have corresponding imaging geometric references.

[0170] like Figure 7B As shown, for the same shooting position, a calibration image 1005 of the same medium, a background image of an empty tank 1006, and a static particle pile image 1007 are acquired respectively. The calibration image 1005 of the same medium is obtained through a calibration component 1004 of the same medium, which can adopt a checkerboard pattern, a dot array, or other calibration structure that can provide the correspondence of known geometric points; the background image 1006 of the empty tank is a background image obtained under basically the same liquid medium and imaging conditions, when no particle pile to be measured has been formed; the static particle pile image 1007 is a measurement image obtained after the bottomless forming cylinder is lifted and has been left to stand for a predetermined time.

[0171] A geometric mapping relationship is established based on the known correspondence of calibration points in the calibration image 1005 of the same medium. This geometric mapping relationship is then used to perform geometric mapping and correction 1008 on the empty tank background image 1006 and the static particle pile image 1007, transforming the corresponding images to a unified contour measurement coordinate system. Let the first... The target coordinates of the effective calibration points are: The coordinates of the corresponding observation point after the geometric mapping are: The number of valid calibration points is The root mean square error of the reprojection of the geometric mapping is then... It can be calculated using the following formula:

[0172] The geometric mapping can employ homography or other geometric transformations that can convert the observed image to a unified contour measurement coordinate system based on known point correspondences. When a sufficient number of valid calibration points cannot be used to establish a geometric mapping that meets the preset quality requirements, the corresponding processing result is not recorded as a valid result for completing the same-medium geometric correction.

[0173] Background normalization 1009 is performed on the static particle pile image and the empty slot background image after processing with the same geometric mapping and correction 1008. Taking one color channel as an example, let the geometrically corrected static particle pile image be... The background image of the empty slot after the same geometric correction is And define the background image of the empty slot in the first... The reference brightness in each color channel is

[0174] in, Let be the effective image region used to calculate the background reference brightness. Then, a background normalization form can be expressed as follows:

[0175] in, These are the pixel values ​​after background normalization. To avoid generating positive numbers that are divided by zero or abnormally magnified in low-brightness background areas, This represents the maximum pixel value allowed for the corresponding image data type. Background normalization 1009 can also be performed in the luminance channel or other image representations suitable for distinguishing between the particle cluster and the background. The empty slot background image 1006 and the static particle cluster image 1007 should be converted to the same calibration coordinate system before undergoing corresponding pixel-by-pixel background normalization.

[0176] The image after background normalization is further processed to exclude reflective and bubble regions (1010). Reflective regions can be determined based on brightness, saturation, locally highlighted connected regions, or a combination thereof; bubble regions can be determined based on local grayscale or color differences, contour shape, scale, roundness, or a combination thereof. Let the reflective region mask be... The bubble region mask is Then the external mask is excluded. It can be determined by the following formula:

[0177] Or, equivalently, in a binary mask representation:

[0178] The external exclusion mask This is used for subsequent candidate particle stack mask generation and quality evaluation to reduce the possibility of reflective and bubble areas being misidentified as particle stacks or free surfaces of particle stacks. Depending on the actual imaging situation, interfering areas supported by image evidence can be appropriately morphologically expanded, but the identification of true free surfaces of particle stacks should not be replaced by arbitrarily expanding the exclusion area.

[0179] When the proportion of reflective areas, bubble areas, or other externally excluded areas in the current corrected image exceeds the preset liquid medium image quality threshold, the current static particle pile image can be determined as not meeting the preset liquid medium imaging quality requirements, and the image can be reset or re-acquired; when the image quality conditions meet the requirements, the image after completing geometric mapping and correction 1008, background normalization 1009, and exclusion of reflective and bubble areas 1010 is used as the corrected particle pile image 1011.

[0180] like Figure 7B As shown, the corrected particle pile image 1011 is then integrated into the aforementioned general static angle measurement process. The general static angle measurement process includes the following steps: multi-candidate particle pile mask generation, candidate physical validity judgment, independent candidate cleaning and free surface boundary correction, post-correction re-evaluation, candidate angle consistency judgment, boundary correction reliability judgment, actual bottom baseline determination, dimensionless relative pile height calculation, joint selection of left and right slope measurement windows, robust slope fitting, angle of repose calculation, and result quality judgment. Image correction processing under a transparent liquid medium is used to form the corrected particle pile image 1011 suitable for the aforementioned general static angle measurement process, without changing the left slope angle. Right side slope angle and principal angle of repose The definition, where:

[0181] In the dual-position implementation, image processing and result quality judgment can be performed separately for the first and second shooting positions. Under transparent liquid medium conditions, corresponding calibration images and empty tank background images can be acquired for each shooting position, and geometric mapping and correction, background normalization, and elimination of reflective and bubble areas can be performed respectively. Let the first and second shooting positions obtain the principal angle of repose that passes their respective quality judgments. and When two shooting positions correspond to the same static state in the same pile-forming process and both obtain an effective angle of repose, the difference in angle of repose between the two shooting positions can be further calculated using the following formula:

[0182] And The geometric difference information between different observation directions of the particle stack is stored. The candidate quality scores and other quality indicators of the two shooting azimuths are used to determine the validity of their respective images. The quality score of one shooting azimuth is not used to weight, replace or correct the rest angle of the other shooting azimuth. When there are no static particle stack images of both azimuths that meet the pairing conditions, the azimuth difference record is not generated.

[0183] Computer implementation and alternative forms

[0184] The aforementioned static image analysis process can be executed by a general-purpose computer, workstation, industrial computer, embedded processor, or other computing device with image processing capabilities. Corresponding program functions may include image input, multi-candidate particle stack mask generation, candidate physical validity judgment, independent candidate cleaning and free surface boundary correction, post-correction re-scoring, candidate angle consistency judgment, boundary correction reliability judgment, determination of actual bottom baseline and pure stack area, dimensionless relative stack height calculation, joint selection of left and right slope windows, robust slope fitting, angle of repose calculation, auxiliary geometric quantity calculation, result quality judgment, and measurement record output. These functions can be implemented by one or more program modules. Module division is only used to illustrate the data and functional relationships between each processing step and does not limit the specific file structure, programming language, or operating platform of the software.

[0185] This invention can also be implemented using a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium can be a magnetic storage medium, an optical storage medium, a semiconductor memory, flash memory, or other medium capable of storing a computer program. When the computer program is executed by a processor, the processor acquires or receives a static side view image of the particle stack, and performs the following on the static side view image: multi-candidate particle stack mask generation, independent candidate processing and boundary correction, post-correction re-evaluation, candidate geometric consistency judgment, boundary correction reliability judgment, actual bottom baseline determination, dimensionless relative stack height calculation, joint selection of left and right slope measurement windows, robust slope fitting, principal angle of repose calculation, and result quality judgment. It can also calculate the equivalent area angle of repose based on the same effective particle stack profile, and associate and save the corresponding measurement results, quality indicators, and reasons for anomalies or rejections.

[0186] While maintaining the aforementioned core technological relationships, the specific implementation of each processing step can be replaced or adjusted. For example, the multi-candidate particle stack mask can be generated by different threshold segmentation methods, color- or texture-based segmentation methods, or trained image segmentation methods; the free surface boundary can be determined using graph cut, edge alignment, or other methods that can correct candidate boundaries based on image information; the slope can be fitted using random sampling consistency fitting, Theil-Sen fitting, Huber fitting, or other robust linear fitting methods; the lifting component, clamping component, illumination component, and image acquisition component can also adopt other structural forms capable of achieving the corresponding functions. The parameters listed in the aforementioned specific embodiments, such as the bottomless forming cylinder size, initial particle filling height ratio, lifting speed, lifting distance, settling time, and camera working distance, are used to illustrate an implementable configuration that can be adjusted according to particle size, medium conditions, forming cylinder size, imaging conditions, and the device's working range.

[0187] Examples and Comparative Examples

[0188] The following detailed description of the static particle pile repose angle measurement process of the present invention, with reference to specific embodiments, further illustrates this process. Unless otherwise specified, particle pile formation, static side view image acquisition, and image processing are all performed according to the aforementioned embodiments. The specific particle sizes, image processing results, and quality evaluation parameters described below are used to illustrate one practical implementation of the present invention and do not constitute a limitation on particle type, particle size, or the range of values ​​for each quality parameter.

[0189] Example 1: Measurement of the Angle of Repose of Dry Spherical Particles

[0190] Dry spherical particles with a nominal particle size of 1 mm were selected and filled into a bottomless forming cylinder. The cylinder was then vertically lifted by a lifting assembly, forming a static particle pile on the bearing surface. After a predetermined settling time, a static side view image of the particle pile was acquired. The image was then analyzed according to the aforementioned processing flow, which included multi-candidate particle pile mask generation, candidate physical validity judgment, independent candidate cleaning and free surface boundary correction, post-correction re-evaluation, candidate angle consistency judgment, boundary correction reliability judgment, actual bottom baseline determination, and joint selection of left and right slope measurement windows.

[0191] After processing the static side view image, the left slope angle is obtained. right slope angle principal angle of repose The final quality score of the target mask is 0.856098, indicating a poor candidate angle range. The angle of repose is 0.324903°, and the cross-union ratio of the masks before and after boundary correction is 0.976242. This static side view image meets the corresponding quality judgment conditions, therefore the obtained principal angle of repose is saved as a valid measurement result.

[0192] Example 2: Measurement of the Angle of Repose of Dry Non-Spherical Particles

[0193] Cylindrical particles with a nominal size of 1.3 mm × 3 mm were selected and piled up using the same method as in Example 1 to form a static particle pile. After a predetermined settling time, a static side view image of the particle pile was obtained. The static side view image was then subjected to the aforementioned processing steps, including multi-candidate generation, independent candidate processing, boundary correction and re-evaluation, candidate reliability judgment, determination of the actual bottom baseline, and joint selection of left and right slope windows.

[0194] After processing the static side view image, the left slope angle is obtained. right slope angle principal angle of repose The final quality score of the target mask is 0.873794, indicating a poor candidate angle range. The angle of repose is 0.830182°, and the cross-union ratio of the masks before and after boundary correction is 0.888216. This static side view image meets the corresponding quality judgment conditions, therefore the obtained principal angle of repose is saved as a valid measurement result. This embodiment illustrates that the aforementioned static image analysis process can be used for particle piles formed by non-spherical particles, but is not limited to spherical particles.

[0195] Example 3: General static image processing of particle stacks in transparent liquid media

[0196] Spherical particles with a nominal diameter of 4 mm are selected and formed into a static particle pile in a transparent liquid medium using a lifting method. After a predetermined settling time, a static side view image of the particle pile is acquired. This embodiment uses the aforementioned general static particle pile image analysis process to process the image, including multi-candidate particle pile mask generation, independent candidate cleaning and boundary correction, post-correction re-scoring, candidate angle consistency judgment, boundary correction reliability judgment, actual bottom baseline determination, and joint selection of left and right slope windows.

[0197] After processing the static side view image, the left slope angle is obtained. right slope angle principal angle of repose The final quality score of the target mask is 0.931664, with a poor candidate angle range. The value is 0, and the cross-union ratio of the masks before and after boundary correction is 0.890113. This static side view image is judged for quality through a general static angle measurement process.

[0198] In this embodiment, the static side view image has not undergone the aforementioned same-medium geometric calibration and empty tank background normalization processing. Therefore, this embodiment is used to illustrate that the general static image analysis process can handle particle pile images acquired under transparent liquid medium conditions, and is not used to evaluate the independent role of the same-medium geometric calibration, empty tank background normalization, and reflection or bubble area exclusion steps. When the corresponding same-medium calibration image and empty tank background image are available, the corresponding image correction can be completed first according to the aforementioned transparent liquid medium implementation method, and then the corrected particle pile image can be input into the general static angle measurement process.

[0199] Comparative Example 1: Re-evaluation without boundary correction

[0200] To illustrate the role of candidate particle pack masks in re-evaluation after boundary correction, the same static particle pack image was analyzed using both the processing method described in this invention and a control method without boundary correction. In one specific processing result, before boundary correction, the quality score of the first candidate was 0.814, and the quality score of the second candidate was 0.414; after candidate cleaning and free surface boundary correction, the quality score of the first candidate became 0.808, and the quality score of the second candidate became 0.861.

[0201] In the processing method described in this invention, the physical validity judgment and quality evaluation are re-executed based on the candidate particle stack mask after boundary correction. Therefore, the second candidate becomes the candidate with a higher score after correction. If the re-evaluation after boundary correction is not performed and the candidate ranking before boundary correction is continued, the first candidate will still maintain a higher ranking. It can be seen that free surface boundary correction may change the geometry and corresponding quality characteristics of the candidate particle stack mask. Re-evaluating and ranking based on the corrected candidates can enable the selection of subsequent target candidates to be based on the corrected actual particle stack geometry.

[0202] Comparative Example 2: Independent selection of measurement windows for left and right slopes

[0203] To illustrate the effect of jointly selecting measurement windows based on dimensionless relative pile height for the left and right slopes, the slope fitting region was determined using both the joint selection method of left and right windows and the independent selection method of left and right slopes described in this invention for the same set of static particle pile images. The method described in this invention only selects candidate window pairs from those where the difference between the lower and upper limits of the dimensionless relative pile height of the left and right candidate windows both satisfy a preset height matching tolerance. Furthermore, target window pairs are selected based on the quality of the left and right windows, the degree of overlap between the relative pile height intervals, and the balance of coverage length. The control method, on the other hand, independently selects measurement windows with higher evaluation on one side of each slope.

[0204] In a set of 16 static particle pile images used to compare the geometric correspondence between the left and right measurement areas, the average relative height difference of the lower boundary obtained by the joint selection of left and right windows was 0.011, the average relative height difference of the upper boundary was 0.008, and the average overlap of the relative height intervals was 0.925. The corresponding values ​​obtained by the independent selection of left and right slopes were 0.191, 0.158, and 0.313, respectively. This comparison result shows that by jointly matching the left and right measurement windows according to the dimensionless relative pile height upper and lower boundaries, the areas of the left and right slopes that actually participate in the fitting can be in a closer relative height range, thereby improving the geometric comparability of the measurement areas of the left and right slopes.

[0205] Other implementation methods

[0206] Those skilled in the art can make corresponding adjustments or equivalent substitutions to the candidate particle stack mask generation method, free surface boundary correction method, robust slope fitting method, lifting and clamping drive form, image acquisition method, and measurement data storage form without changing the above-mentioned core technical relationships.

[0207] In summary, this invention proposes a method and apparatus for measuring the angle of repose of particles based on controlled lifting cylinders and multi-candidate quality gating. It employs a dual gating system—multi-candidate mask generation and re-evaluation after correction, candidate geometric consistency, and boundary correction magnitude—to screen reliable target masks. Furthermore, it achieves joint slope selection based on actual bottom baseline and normalized pile height matching of left and right slopes. Combined with global-local slope representativeness judgment and multi-factor joint quality gating, it reliably rejects unreliable measurement results and retains quality traceability information even in abnormal situations. Compared with existing technologies, the significant technical advantages of this invention are reflected in the following aspects: Since each candidate particle stack mask undergoes independent cleaning and boundary correction, and the physical validity judgment is re-executed and the final mask quality score is calculated, the candidate ranking can reflect the particle stack geometry after boundary correction. This reduces the possibility that the candidate evaluation results before correction will continue to dominate the final candidate selection, leading to unreliable profiles entering subsequent angle measurements.

[0208] By judging the slope geometry consistency among high-quality candidates and evaluating the reliability of boundary correction by combining the average displacement of the top boundary, the maximum displacement of the top boundary, and the degree of mask overlap before and after boundary correction, it is possible to identify cases where there are significant geometric discrepancies among high-quality candidates, as well as cases where the overall mask still has a high degree of overlap but abnormal displacement occurs on local free surfaces. This reduces the possibility of abnormal candidates directly entering the angle of repose calculation. The relevant quality judgment process is as follows: Figure 3 As shown, the boundary correction reliability judgment and the high-quality candidate angle consistency judgment are respectively as follows: Figure 5 and Figure 6 As shown.

[0209] By using the actual contact baseline between the particles and the bearing surface as the slope angle measurement benchmark, and normalizing the relative height of the free surface and the baseline fitting residual with the actual pile height H, the deviation between the image lateral benchmark and the actual bearing surface direction of the particles, as well as the influence of different particle pile sizes on the consistency of the slope measurement benchmark, can be reduced, providing a unified geometric reference for the subsequent selection of corresponding measurement areas on the left and right slopes.

[0210] Since only candidate windows of the left and right slopes that meet the matching conditions at both the upper and lower boundaries of the dimensionless relative stack height are paired, and the quality of the left and right windows, the degree of overlap of the relative height intervals, and the degree of balance of the coverage length are jointly evaluated, the geometric comparability of the actual measurement areas of the left and right slopes can be improved.

[0211] After obtaining the effective particle pile outline, in addition to obtaining the main repose angles by fitting the left and right effective slopes, the equivalent area repose angles can also be determined based on the effective two-dimensional projected area and effective projected bottom width of the same particle pile. This provides auxiliary characterization information of the particle pile from the perspective of overall projected geometry, thereby forming a particle pile geometric measurement result that takes into account both local effective slope characteristics and overall pile shape characteristics.

[0212] In the transparent liquid medium implementation, by establishing a geometric correction relationship under consistent medium and imaging conditions, and using the empty tank background image corrected by the same geometric mapping to normalize the background of the static side view image of the particle stack, while eliminating reflective areas and bubble areas, the interference of transparent liquid, container wall, optical interface, background and local bubbles on particle stack contour recognition can be reduced, so that image preprocessing under transparent liquid conditions can be connected with the above-mentioned particle stack contour extraction and static repose angle measurement process.

[0213] The above description, in conjunction with specific / preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the inventive concept, and all such substitutions or modifications should be considered within the scope of protection of the present invention.

Claims

1. A method for measuring the angle of repose of granular materials, comprising lifting a bottomless forming cylinder to form a static granule pile on a bearing surface, characterized in that, It also includes the following steps: S1. For the static side view image of the particle stack obtained after the lifting is completed and a predetermined resting time has elapsed, multiple segmentation strategies are used to generate multiple candidate particle stack masks. S2. Perform physical validity judgment on each candidate particle stack mask, and independently perform mask cleaning and free surface boundary correction on each candidate particle stack mask that passes the physical validity judgment. Then, perform physical validity judgment again based on the corrected candidate particle stack mask and recalculate the final mask quality score. Use the final corrected mask quality score of each candidate as the sorting basis. S3. Based on the final mask quality score, determine a set of high-scoring candidates from the candidate particle stack masks that have passed the physical validity judgment after correction and can form an effective double-sided slope geometry representation. Perform candidate geometric consistency judgment based on the dispersion of the candidate angle proxy value of each high-scoring candidate, and perform boundary correction reliability judgment based on the top boundary displacement and mask overlap degree of the highest-scoring candidate before and after boundary correction. When both the candidate geometric consistency judgment and the boundary correction reliability judgment pass, the highest-scoring candidate is determined as the target mask. S4. Determine the actual bottom baseline between the particle and the bearing surface based on the bottom boundary point of the target mask, define the vertical distance from the top of the pile to the actual bottom baseline as the pile height, and normalize the signed vertical distance from the free surface boundary point to the actual bottom baseline with the pile height to obtain the dimensionless relative pile height. S5. Generate multiple consecutive candidate windows on the left and right slopes respectively. Form a candidate window pair between the left and right candidate windows whose differences in the lower and upper limits of the dimensionless relative stacking height are not greater than the preset height matching tolerance. Calculate the joint evaluation score of the window pair based on the window quality of the left and right candidate windows, the degree of overlap of their dimensionless relative stacking height intervals, and the degree of balance of their coverage lengths. Select the candidate window pair that passes the preset window pair quality threshold and has the highest joint evaluation score as the target window pair. S6. Fit the left and right slope lines within the target window respectively, calculate the angle between each slope line and the actual bottom baseline, and determine the principal rest angle of the current static side view image based on the left and right angles. S7. Perform a slope representativeness judgment based on the global-local angle difference between the principal angle of repose and the local slope representative angle, as well as the local slope angle dispersion. Combine the slope fitting quality, the difference between the left and right slope angles, and at least one independent anomaly information in the single-map geometric angle interval to perform a joint quality judgment. If the preset rejection condition is met, reject the current principal angle of repose; otherwise, save the current principal angle of repose.

2. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S1, the multiple segmentation strategies include global threshold segmentation, adaptive threshold segmentation, and background correction threshold segmentation. Positive polarity candidate particle stack mask and anti-polarity candidate particle stack mask are generated for each segmentation result, so that each candidate particle stack mask can independently enter the subsequent physical validity judgment and mask quality score calculation. In step S2, the physical validity judgment includes: calculating the ratio of the foreground area to the image area, the ratio of the width of the enclosing region to the image width, the ratio of the height of the enclosing region to the image height, the bottom support contact degree, the top boundary coverage degree, the apparent height from the top to the bottom of the pile, and the degree of overlap with the predetermined interference area for each candidate particle pile mask, and comparing them with the corresponding preset thresholds. When any of the indicators does not meet the preset threshold conditions, it is determined that the corresponding candidate particle pile mask has not passed the physical validity judgment. The final mask quality score is obtained by weighting multiple positive quality components, including foreground area score, bottom support contact score, main connected region ratio, outer contour continuity score, top boundary effectiveness score, top boundary coverage score, bottom flatness score, surrounding area aspect ratio score, top-to-bottom topology score, and surrounding area scale score, and subtracting a penalty term including internal hole penalty and interference area overlap penalty.

3. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S2, mask cleaning includes: retaining the main connected regions that correspond to the spatial position of the particle stack, repairing internal holes that meet the preset area conditions, and performing morphological cleaning to remove discrete noise and local defects; the free surface boundary correction is achieved by graph cut processing based on foreground and background seed constraints or boundary alignment processing based on image edges; In step S3, the boundary correction reliability judgment includes: after the boundary correction is completed, recording the top boundary ordinates of the mask before correction and the mask after correction at the common effective abscissa, calculating the normalized average displacement and normalized maximum displacement based on the displacement of the top boundary before and after correction, wherein the normalized average displacement and normalized maximum displacement are normalized with the image height of the entire static side view image; and calculating the cross-union ratio of the masks before and after correction. When the normalized average displacement exceeds a preset average displacement threshold, or the normalized maximum displacement exceeds a preset maximum displacement threshold and the crossover ratio of the masks before and after correction exceeds a preset overlap threshold, the boundary correction reliability judgment is deemed to have failed.

4. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S3, the high-scoring candidate set is determined as follows: the candidate particle stack masks that have been corrected and passed the physical validity judgment and can form effective double-sided slope candidate angle proxy values ​​constitute a measurable candidate set, which is sorted from high to low according to the final mask quality score. Based on the highest final mask quality score, the candidate particle stack masks whose final mask quality score is not lower than the highest score minus the preset candidate consensus zone are determined as high-scoring candidates. The candidate angle proxy value is determined as follows: extract the bottom boundary points from the corrected mask of the corresponding high-scoring candidate and perform robust straight line fitting to determine the direction and position of the temporary baseline. Use the temporary baseline as a reference to determine the temporary top and temporary height. Normalize the signed vertical distance from the free surface boundary point to the temporary baseline with the temporary height. Fit the left and right slope lines respectively within the preset temporary dimensionless relative height range and calculate the arithmetic mean of the left and right proxy angles as the candidate angle proxy value. The candidate geometric consistency judgment is as follows: when there are at least two valid candidate angle proxy values ​​in the high-scoring candidate set, the difference between the maximum and minimum values ​​of the valid candidate angle proxy values ​​is calculated. If the difference exceeds the preset candidate angle consistency threshold, the candidate geometric consistency judgment is determined to be unsuccessful.

5. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S4, the actual bottom baseline is determined in two stages according to direction and position: first, a robust straight line fit is performed based on the bottom boundary points of the target mask to determine the baseline direction, and then the baseline position is determined based on the intercept distribution of the free surface boundary points under the determined baseline direction. The stack height is the vertical distance from the stack apex to the actual bottom baseline; based on the signed vertical distance from the free surface boundary point to the actual bottom baseline, a continuous free surface segment containing the stack apex is identified, and the adjacent slope trends at both ends of the continuous free surface segment are extrapolated to the actual bottom baseline to determine the left and right stack feet, and a pure stack mask is formed by the continuous free surface segment, the left stack foot, the right stack foot, and the actual bottom baseline; The baseline fitting residual is calculated based on the free surface boundary point close to the actual bottom baseline, and the baseline fitting residual is normalized with the stack height. When the normalized baseline fitting residual exceeds the preset baseline residual threshold, the rest angle of the current static side view image is rejected.

6. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S5, the window quality of each candidate window is calculated based on the fitting linearity of the slope points within the window, the coverage length, the dimensionless relative pile height span, the root mean square residual of the fitting normalized by the pile height, and the boundary roughness characterized by the standard deviation of the boundary residual normalized by the pile height. For candidate window pairs that meet the high matching condition, the joint evaluation score of the window pair is determined by the weighted sum of the average window quality of the left and right candidate windows, the crossover ratio of the dimensionless relative stack height intervals of the left and right windows, and the balance of the left and right coverage lengths. If there is no candidate window pair that meets the high matching condition, or if the highest joint evaluation score is lower than the preset window pair quality threshold, the rest angle of the current static side view image is rejected.

7. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S7, the method for determining the representative angle of the local slope is as follows: multiple local slope segments are formed along the free surface on the left and right slopes respectively, and a straight line is fitted to each local slope segment to calculate the local slope angle of each local slope segment relative to the actual bottom baseline. The arithmetic mean of the median of the left local slope angle and the median of the right local slope angle is taken as the local representative angle. The global-local angle difference is the absolute difference between the principal angle of repose and the local representative angle. The dispersion of the local slope angle is the interquartile range of the combined set of the left and right local slope angles; When the global-local angle difference exceeds the preset global-local warning threshold and the local slope angle dispersion exceeds the preset local slope angle dispersion threshold, a slope representative difference mark is formed. This mark itself does not constitute a rejection condition on its own. In the joint quality judgment, when the principal angle of repose exceeds a preset extreme angle threshold and the global-local angle difference exceeds a preset auxiliary angle difference threshold or the local slope angle dispersion exceeds a preset local slope angle dispersion threshold, the current angle of repose is rejected; or, when the left-right slope angle difference exceeds a preset left-right slope angle difference threshold, and the global-local angle difference exceeds the preset auxiliary angle difference threshold and the local slope angle dispersion exceeds the preset local slope angle dispersion threshold, the current angle of repose is rejected.

8. The method for measuring the angle of repose as described in claim 1, characterized in that, In step S7, the method for determining the geometric angle interval of the single image is as follows: the left slope point, the right slope point and the baseline point used to fit the actual bottom baseline in the target window pair are resampled multiple times with replacement. For each set of resampled points, the left slope line, the right slope line and the actual bottom baseline are refitted and the left and right included angles and their arithmetic mean are recalculated. The geometric angle interval of the current static side view image is determined based on the preset lower quantile and preset upper quantile of the arithmetic mean obtained from multiple resampling. When the smaller value of the linearity of the left and right slope fitting is lower than the preset linearity threshold and the half-width of the geometric angle interval exceeds the preset half-width threshold, the current rest angle is rejected. When the global-local angle difference exceeds a preset global-local warning threshold and the local slope angle dispersion exceeds a preset local slope angle dispersion threshold, and the smaller of the left and right slope fitting linearity values ​​is lower than a preset linearity threshold or the geometric angle interval half-width exceeds a preset interval half-width threshold, the current rest angle is rejected.

9. The method for measuring the angle of repose as described in claim 1, characterized in that, The particle stack is located in a transparent liquid. Before lifting the bottomless forming cylinder, the particles are immersed in the transparent liquid, and after the lifting is completed, a preset defoaming and settling time is waited. Under the condition that the liquid type, liquid depth, camera pose, optical interface position and lighting conditions are kept consistent, the same medium calibration image, empty tank background image and static side view image located in the side profile measurement plane of the particle stack are acquired respectively. A geometric mapping is established based on the same medium calibration image. The geometric mapping is used to correct the empty slot background image and the static side view image respectively. The corrected static side view image is normalized based on the corrected empty slot background image, and a reflective area mask and a bubble area mask are generated. The two are combined into an external exclusion mask. When generating the candidate particle stack mask, the area corresponding to the external exclusion mask is excluded.

10. A device for measuring the angle of repose of granular materials, characterized in that, include: The bearing surface is used to support the static particle stack; A bottomless molding cylinder, the lower end of which can be placed above the bearing surface; A lifting component is used to drive the bottomless forming cylinder to move vertically from the filling position to the stacking position; A clamping component is disposed at the moving end of the lifting component and is used to clamp the bottomless forming cylinder and move it with the lifting component; An image acquisition component is used to acquire at least one static side view image of the particle stack after the bottomless forming cylinder has been lifted and a predetermined settling time has elapsed. And an image processing apparatus, including a processor and a memory, the memory storing program instructions executable by the processor, the program instructions being used to cause the processor to perform the angle of repose measurement method as described in any one of claims 1 to 9.

Images