A quantitative SEM analysis method for the microstructure of moraine after freeze-thaw cycles

By using SEM image scale normalization and watershed segmentation techniques, the instability of glacial till image processing and insufficient characterization of pore connectivity after freeze-thaw cycles were solved, enabling quantitative analysis of glacial till strength parameters and slope stability assessment, and providing a graded early warning mechanism.

CN123118734APending Publication Date: 2026-11-17SOUTHWEST JIAOTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611618961.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

Smart Images

  • Figure CN123118734A_ABST
    Figure CN123118734A_ABST
Patent Text Reader

Abstract

This invention discloses a SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles, belonging to the field of geotechnical engineering microstructure testing and image processing technology. The method specifically includes: acquiring SEM images, scales, and three-dimensional porosity of glacial till under different freeze-thaw cycle cycles; generating candidate threshold segmentation maps after scale normalization and image preprocessing, and determining initial segmentation maps for granular and porous phases; performing labeling control watershed segmentation and oversegmentation merging on cohesive particle components, extracting microstructure parameters, and determining pore connectivity; inputting the freeze-thaw cycle number and microstructure parameter vectors into a strength parameter mapping model to obtain cohesion and internal friction angle, calculating the slope safety factor, and outputting graded early warning signals. This invention improves the stability of SEM image segmentation and microstructure parameter extraction, and enables the application of microstructure analysis results in slope stability early warning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering microstructure testing and image processing technology, specifically to a SEM quantitative analysis method for the microstructure of moraine after freeze-thaw cycles. Background Technology

[0002] The statements in this section are provided only as background information in connection with this disclosure and may not constitute prior art.

[0003] Glacial till is composed of soil particles of different sizes and shapes, as well as interparticle pores. During freeze-thaw cycles, water within the soil undergoes phase changes and migration, leading to changes in particle arrangement, particle contact relationships, and pore structure. Obtaining microscopic images of glacial till under different freeze-thaw cycles using scanning electron microscopy and performing quantitative analysis of particle and pore characteristics is an important method for studying the evolution of the microstructure and changes in the mechanical properties of glacial till.

[0004] Current SEM quantitative analysis typically includes image acquisition, grayscale processing, granular and porous phase segmentation, boundary extraction, and feature parameter calculation. Different samples or different effective fields of view may have different pixel sizes, grayscale distributions, local brightness, and noise levels, resulting in different segmentation effects for the same image processing parameters in different freeze-thaw cycle groups. When determining the granular and porous phases based solely on a single grayscale threshold, the segmentation results may deviate from the overall porosity characteristics of the sample, making it difficult to simultaneously consider the porosity ratio, particle boundary integrity, and the stability of the solidified interconnected structure.

[0005] Meanwhile, the contours of glacial till particles are typically irregular, and adjacent particles may form adhered regions in SEM images due to contact, occlusion, or similar local gray levels. When extracting particles directly based on connected regions, multiple adjacent particles may be identified as a single entity, and some particle boundaries may be broken or not closed, thus affecting the calculation of particle size, morphology, and orientation characteristics. Manually checking and correcting particle boundaries is susceptible to the influence of operator judgment differences on the processing results, hindering the formation of a unified and repeatable quantitative analysis process.

[0006] Furthermore, in the process of pore structure analysis, parameters such as pore area, pore equivalent diameter, and pore fractal dimension can reflect the geometric characteristics of pores from different aspects. However, a single parameter is difficult to simultaneously characterize the internal width of pores, the overall connectivity of pore regions, the degree of cross-scale preservation of pore structure, and the local aggregation state of pore pixels. Therefore, the comprehensive characterization of pore connectivity is still insufficient.

[0007] On the other hand, existing microstructure analysis is mainly used to describe the changes in particles and pores with freeze-thaw cycles. However, a continuous data correlation still needs to be established between microstructure parameters, macroscopic strength parameters of glacial till, and slope stability calculation results in order to further apply the quantitative analysis results of SEM microstructure to the determination of glacial till strength parameters and the assessment of slope stability.

[0008] Therefore, how to process SEM images of glacial till under different freeze-thaw cycles at a unified physical scale, improve the stability of the granular phase and porous phase segmentation results, reduce the influence of cohesive particles and over-segmentation on the extraction results of microstructure parameters, comprehensively and quantitatively characterize the pore connectivity state, and further establish the correspondence between microstructure parameters, macro strength parameters and slope stability calculation results are the technical problems that need to be solved. Summary of the Invention

[0009] The purpose of this invention is to address the problems of inconsistent SEM image processing scales, insufficient stability of particle and pore phase segmentation results, difficulty in repeatedly identifying the boundaries of cohesive particles, inadequate characterization of pore connectivity, and difficulty in continuously using microstructure analysis results for strength parameter determination and slope stability assessment under different freeze-thaw cycles. This invention provides a quantitative SEM analysis method for the microstructure of glacial till after freeze-thaw cycles. This method is based on SEM image scale normalization and preprocessing, joint evaluation of candidate thresholds through the fusion of global and local grayscale information, watershed segmentation using markers of cohesive particle components, fusion of pore connectivity indices, and mapping of strength parameters calibrated from glacial till mechanical tests. This enables quantitative analysis of the microstructure of glacial till under different freeze-thaw cycles, and the microstructure analysis results are used for determining cohesion and internal friction angle, calculating slope stability, and providing graded early warning.

[0010] The technical solution of the present invention is as follows: A quantitative SEM analysis method for the microstructure of glacial till after freeze-thaw cycles includes: Step S1: Obtain SEM images, scales, and three-dimensional porosity of glacial till under different freeze-thaw cycles, and establish a mapping between pixel size and actual size; Step S2: Based on the mapping between the pixel size and the actual size, the SEM image of the glacial till is scaled and preprocessed to obtain a standardized grayscale image; Step S3: Generate multiple candidate threshold segmentation maps based on the global and local grayscale information of the standardized grayscale image. Jointly evaluate each candidate threshold segmentation map according to the difference between the two-dimensional pore area ratio and the corresponding three-dimensional porosity, the integrity of the particle boundary, and the change of the solid interconnection structure with the candidate threshold to determine the initial segmentation map of the particle phase and the porous phase. Step S4: Identify the adhering particle components based on the initial segmentation map, perform mark control watershed segmentation on the adhering particle components, and merge the segmented sub-regions according to the area of ​​the segmented sub-regions and the grayscale features of the shared boundaries of adjacent segmented sub-regions to obtain a particle and pore boundary map. Step S5: Based on the particle and pore boundary map, extract the microstructural parameters that characterize particle morphology, pore structure, solid phase ratio and particle orientation, and determine the pore connectivity that characterizes the pore connectivity state to form a microstructural parameter vector. Step S6: Input the number of freeze-thaw cycles and the micro-parameter vector into the strength parameter mapping model calibrated by the glacial till mechanics test to obtain cohesion and internal friction angle. Use the cohesion and internal friction angle for slope stability calculation and determine the graded early warning signal based on the calculated safety factor.

[0011] Further, step S1 includes: Prepare glacial till samples of uniform size, set up multiple freeze-thaw cycle groups including a non-freeze-thaw control group, and perform freeze-thaw cycles according to preset freezing and thawing conditions; After the freeze-thaw cycle is completed, the sample is dried to constant mass and then subjected to conductive treatment. Multiple effective fields of view of each sample are acquired under the same imaging parameters. The three-dimensional porosity was determined based on the sample's mass, volume, and soil particle specific gravity, and the sample identification, freeze-thaw cycle count, SEM image, and three-dimensional porosity were correlated accordingly.

[0012] Further, step S2 includes: The pixel size is determined by the ratio of the actual length represented by the ruler to the number of pixels corresponding to the ruler. The ruler and text areas in the SEM image are then cropped, and the cropped SEM image is converted into an 8-bit grayscale image. A baseline pixel size is determined based on the pixel size of each SEM image, and the 8-bit grayscale image is resampled based on the difference between the pixel size and the baseline pixel size to achieve scale normalization. Gray-scale enhancement is achieved by using local histogram equalization with limited contrast, followed by edge-preserving noise reduction using median filtering and bilateral filtering in sequence. Based on the pixel size, the actual length and area parameters used in subsequent image processing are converted into the corresponding pixel scale.

[0013] Further, step S3 includes: The global baseline threshold is determined based on the global grayscale information of the standardized grayscale image using the maximum inter-class variance method. Multiple global candidate thresholds are set around the global baseline threshold, and local thresholds are determined based on the local grayscale information within the sliding window. Each global candidate threshold and the local threshold are fused according to a preset fusion weight to form a fusion threshold map corresponding to each global candidate threshold. For each fusion threshold map, the pixel gray value of the standardized grayscale image is compared with the fusion threshold at the corresponding position pixel by pixel. Pixels with a pixel gray value lower than the fusion threshold at the corresponding position are marked as pore phase, and pixels with a pixel gray value not lower than the fusion threshold at the corresponding position are marked as granular phase, generating multiple candidate threshold segmentation maps. For each candidate threshold segmentation map, the two-dimensional pore area ratio is calculated, the particle boundary is extracted, and the boundary non-closure degree is determined according to the ratio of the endpoint pixels of the particle boundary to the total number of particle boundary pixels. The particle phase is labeled with eight neighboring connected components to obtain the number of solid connected components, and the normalized rate of change of the number of solid connected components relative to the adjacent candidate threshold segmentation maps is calculated. The difference between the two-dimensional pore area ratio and the corresponding three-dimensional porosity, the boundary non-closure degree, and the normalized rate of change are processed with the same dimensions and weighted to form a joint evaluation value. The candidate threshold segmentation map with the smallest joint evaluation value is determined as the initial segmentation map.

[0014] Further, step S4 includes: The particle phases in the initial segmentation diagram are marked with connected components to obtain solidly connected components, and the area and convex hull area of ​​each solidly connected component are calculated. A fixedly connected component whose area exceeds a preset multiple of the median area of ​​all fixedly connected components, or whose area of ​​the fixedly connected component and its convex hull area are in proportion to the convex hull area exceeding a preset concavity threshold, is identified as the adhesive particle component. The Euclidean distance transformation is performed on the adhesive particle component to obtain a distance transformation map. The local maxima in the distance transformation map are used as foreground markers. The median of the equivalent diameter of the particles is determined based on the fixed connected components in the initial segmentation map. The minimum spacing between adjacent foreground markers is determined based on the median of the equivalent diameter of the particles, the pixel size, and the preset scale coefficient. Marker-controlled watershed segmentation is then performed. The preset multiple, preset concavity threshold, and preset scale coefficient are predetermined based on the calibration sample.

[0015] Furthermore, step S4 also includes: Sub-regions whose area obtained by dividing the watershed by the marker control is smaller than the preset minimum retention area will be merged into the adjacent sub-region with the longest shared boundary. A gradient merging threshold is determined based on the average gray-level gradient distribution of the shared boundary between adjacent sub-regions generated by watershed segmentation. Adjacent sub-regions whose average gray-level gradient of the shared boundary is less than the gradient merging threshold are then merged. The radius of the disk structure element is determined based on the ratio of the median of the equivalent diameter of the particles to the pixel size. The disk structure element is then used to perform morphological opening and closing operations on each merged particle region. After determining the contour fitting tolerance based on the pixel size, contour polygon fitting is performed to obtain the particle and pore boundary map.

[0016] Further, step S5 includes: The diameter of a circle with the same area as the particle or pore is defined as the equivalent diameter of the corresponding particle or pore. The particle morphology irregularity is characterized by the ratio of the square of the particle perimeter to the product of four times pi and the particle area. The box counting method was used to determine the pore fractal dimension, which characterizes the pore structure, based on the number of grids covering the pore boundaries at different grid scales. The projected solid area ratio, which characterizes the proportion of solid phase, is determined based on the ratio of the difference between the effective field of view area and the porosity area to the effective field of view area. The particle orientation angle is determined based on the principal eigenvector of the particle pixel coordinate covariance matrix, and the standard deviation of the particle orientation angle, which characterizes particle orientation, is calculated.

[0017] Furthermore, in step S5: Taking the pore phase in the particle and pore boundary map as the analysis object, distance transformation index, connected region index, multi-scale morphological index and local neighborhood index are determined respectively. The distance transformation index is used to characterize the width distribution inside the pore, the connected region index is used to characterize the degree of connection and dispersion of the eight-neighbor pore connected region, the multi-scale morphological index is used to characterize the degree of preservation of pore structure at different morphological scales, and the local neighborhood index is used to characterize the degree of local aggregation of pore pixels in the Gaussian weighted neighborhood. The two-dimensional pore area ratio is determined based on the pore phase in the particle and pore boundary diagram, and the fusion weights of the distance transformation index, connected region index, multi-scale morphology index and local neighborhood index are determined based on the interval to which the two-dimensional pore area ratio belongs. All fusion weights are non-negative and the sum of the weights is 1. When the two-dimensional pore area ratio is less than a preset pore area ratio threshold, the fusion weight of the local neighborhood index is made higher than the fusion weight of the distance transformation index, the connected region index, and the multi-scale morphology index respectively; when the two-dimensional pore area ratio is not less than the preset pore area ratio threshold, the fusion weight of the distance transformation index and the connected region index is made higher than the fusion weight of the multi-scale morphology index and the local neighborhood index respectively. The distance transformation index, connected region index, multi-scale morphology index, and local neighborhood index are weighted and fused according to the fusion weight to obtain the porosity connectivity.

[0018] Further, step S6 includes: The microscopic parameter vector is composed of particle morphological irregularity, pore fractal dimension, projected solid area ratio, standard deviation of particle orientation angle, and pore connectivity. A calibration sample set is constructed, which consists of glacial till samples with known freeze-thaw cycles, microscopic parameter vectors, cohesion, and internal friction angle; Using the average vector of micro-parameters of the zero-freeze-thaw sample as the reference vector, the mean and standard deviation of each dimension of the micro-parameters in the calibration sample set are used to standardize the micro-parameter vector and the reference vector on the same scale. Using the least squares method with regularization terms, and taking the cohesion and internal friction angle obtained from direct shear tests under different freeze-thaw cycles as calibration data, a strength parameter mapping model was established, consisting of a decay term of freeze-thaw cycle number and a correction term for the change of the standardized micro-parameter vector relative to the standardized reference vector.

[0019] Furthermore, step S6 also includes: The cohesion and internal friction angle output by the strength parameter mapping model are used as strength parameters to input into the simplified Bishop slice model. The simplified Bishop slice model also receives slope cross-sectional geometric parameters, soil strip weight, slip surface dip angle, and pore water pressure. The safety factor is iteratively calculated until the preset convergence condition is met. The safety factor is then compared with the pre-set grading threshold based on the slope grade and applicable engineering standards to determine the warning level. Generate an early warning data packet that includes at least intensity parameters, safety factor, and early warning level, and send the early warning data packet to the early warning terminal; The early warning terminal performs at least one of the following operations based on the early warning level: displaying the result, recording and storing it, providing audible and visual alerts, or sending an alarm message to the engineering monitoring platform.

[0020] Compared with existing technologies, the advantages of this invention are: 1. This invention establishes a mapping between pixel size and actual size using a scale bar in SEM images, and performs scale normalization, grayscale enhancement, and edge-preserving noise reduction on SEM images under different freeze-thaw cycles, enabling subsequent analysis of different SEM images under a unified actual scale and a unified grayscale quantization range. Furthermore, it generates multiple candidate threshold segmentation maps by combining global and local grayscale information, and conducts joint evaluation based on the difference between two-dimensional pore area ratio and three-dimensional porosity, particle boundary integrity, and the change of solid-connected structures with candidate thresholds. This helps to reduce the influence of different image scales, grayscale distributions, and single threshold selection on the segmentation results of granular and porous phases, and improves the comparability of microstructural parameters between samples from different freeze-thaw cycles.

[0021] 2. This invention identifies adhered particle components by measuring the area and concavity of the solidified interconnected components, and determines the foreground marker based on Euclidean distance transformation to perform marker-controlled watershed segmentation on the adhered particle components; furthermore, it merges oversegmented sub-regions based on the area of ​​the segmented sub-regions and the grayscale features of the shared boundaries of adjacent segmented sub-regions, and then combines morphological opening and closing operations and contour polygon fitting to obtain particle and pore boundary maps, which helps to reduce the impact of particle adhesion and oversegmentation on the extraction results of particle size, particle morphology and particle orientation features.

[0022] 3. This invention extracts particle morphology irregularity, pore fractal dimension, projected solid area ratio, and particle orientation angle standard deviation, and calculates distance transformation index, connectivity index, multi-scale morphology index, and local neighborhood index from pore internal width, overall connectivity of pore region, cross-scale preservation of pore structure, and local pixel aggregation state of pore respectively. The four indices are then fused to obtain pore connectivity, enabling a comprehensive quantitative characterization of particle morphology, pore structure, solid phase ratio, particle orientation, and pore connectivity state.

[0023] 4. This invention uses a strength parameter mapping model calibrated by glacial till mechanical tests to map the number of freeze-thaw cycles and microscopic parameter vectors into cohesion and internal friction angle. The obtained cohesion and internal friction angle are then used for slope stability calculation. Based on the safety factor, the warning level is determined and graded warning signals are output to the warning terminal. This establishes a continuous data correlation between the number of freeze-thaw cycles, SEM microstructural parameters, macroscopic strength parameters of glacial till, slope safety factor, and graded warning results, providing data basis for judging the stability of glacial till slopes under freeze-thaw action. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0025] Figure 1 This is a flowchart of the SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to the present invention. Figure 2 A schematic diagram of sample preparation for a bottomless frame-shaped compression container; Figure 3 This is a flowchart of the sample drying, conductivity treatment, and SEM imaging process. Figure 4 Flowchart for SEM image standardization and joint evaluation of candidate thresholds; Figure 5 SEM images of glacial till under different freeze-thaw cycles are shown, among which, Figure 5 (a)- Figure 5 (h) shows SEM images of glacial till under different freeze-thaw cycles; Figure 6 This is a schematic diagram illustrating the identification of adhering particles, labeling and controlling watershed segmentation, and boundary correction. Figure 6 (a) is a schematic diagram of the initial fixed connection components. Figure 6 (b) is a schematic diagram for area and concavity identification. Figure 6 (c) is a schematic diagram of Euclidean distance transformation. Figure 6 (d) is a schematic diagram of the foreground marking of local maxima. Figure 6 (e) is a schematic diagram of the watershed segmentation controlled by markers. Figure 6 (f) is a schematic diagram of merging small-area sub-regions. Figure 6 (g) is a schematic diagram of low-gradient shared boundary merging. Figure 6 (h) is a schematic diagram of morphological opening and closing operations and contour polygon fitting; Figure 7 The graph shows the distribution curves of equivalent particle diameter under different freeze-thaw cycles; Figure 8 The distribution of equivalent pore diameter under different freeze-thaw cycles; Figure 9 The graph shows the variation of particle morphology irregularity and pore fractal dimension. Figure 9 (a) is a graph showing the variation in particle morphology irregularity. Figure 9 (b) is a graph showing the variation of the fractal dimension of the pores; Figure 10 A flowchart for the integration of four indicators of pore connectivity; Figure 11 This is a comparison chart of the output values ​​and calibration values ​​of cohesion and internal friction angle. Figure 11 (a) is a comparison chart of the output value and the calibration value of the internal friction angle. Figure 11 (b) is a comparison chart of the cohesion output value and the calibration value; Figure 12This is a flowchart for strength parameter mapping, slope stability calculation, and graded early warning. Detailed Implementation

[0026] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0028] Example 1 This embodiment provides a SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles. Sample preparation, freeze-thaw cycles, drying, conductive treatment, and SEM imaging are performed by the corresponding experimental equipment. SEM image reading, data correlation, scale normalization, image preprocessing, granular and porous phase segmentation, cohesive particle identification and segmentation, microstructure parameter calculation, strength parameter mapping, slope stability calculation, and early warning output are executed by the processor.

[0029] like Figure 1 As shown in this embodiment, a SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles is proposed, which specifically includes the following steps: Step S1: Obtain SEM images, scales, and three-dimensional porosity of glacial till under different freeze-thaw cycles, and establish a mapping between pixel size and actual size.

[0030] In this embodiment, specifically, step S1 further includes: Step S11: Prepare glacial till samples and perform freeze-thaw cycles.

[0031] Specifically, glacial till samples with unique sample identifiers are used as data association units. The SEM image is a two-dimensional grayscale image of the sample surface acquired by a scanning electron microscope; the scale bar is a scale marker in the SEM image representing the actual length; and the three-dimensional porosity is the proportion of pore volume to the total volume of the sample.

[0032] In this embodiment, the following is adopted: Figure 2The bottomless frame-shaped sample container shown is used to prepare 15 mm × 15 mm × 15 mm cubic samples from the same batch of glacial till. The initial moisture content, dry density, and pressing direction of each sample are controlled to be consistent. In order to reduce the impact of differences in the original materials and sample preparation errors on the features of subsequent SEM images and three-dimensional porosity, no less than three parallel samples are set up for each freeze-thaw cycle group.

[0033] Furthermore, this embodiment sets 0, 1, 4, 8, 10, 12, 15, and 20 freeze-thaw cycles, with the 0-cycle group serving as a non-freeze-thaw control group. Each freeze-thaw cycle includes, in sequence, the following: The sample was frozen at 15 °C for 12 h and thawed at 15 °C for 12 h. The freezing phase started from the time the sample reached the preset freezing temperature, and the thawing phase started from the time the sample was transferred to the preset thawing environment. The freezing and thawing cycles were performed continuously in the order of freezing first and then thawing until the corresponding number of freeze-thaw cycles were reached.

[0034] Step S12: Dry, conduct electricity, and perform SEM imaging on the sample that has completed the freeze-thaw cycle.

[0035] Specifically, the sample that has completed the corresponding number of freeze-thaw cycles is placed in an environment of 70 °C for drying. During the drying process, the sample is weighed every 2 hours. When the relative change rate of the mass obtained from two consecutive weighings is not greater than 0.1%, the mass of the sample is determined to be constant.

[0036] like Figure 3 As shown, after the sample reaches a constant mass, an observation surface that has not been cut, squeezed, or disturbed is selected from the middle of the sample. The observation surface is then cleaned of dust and coated with a conductive layer of uniform thickness to ensure that the sample surface meets the conductivity requirements for scanning electron microscopy imaging.

[0037] Furthermore, this embodiment uses the same scanning electron microscope and fixes imaging parameters such as accelerating voltage, working distance, detection mode, image resolution, brightness, contrast and magnification, and collects 3 to 5 effective fields of view for each sample at 250x magnification.

[0038] In this embodiment, it should be noted that the effective field of view is an observation area with clear imaging, complete grayscale distribution, no obvious charging bright spots, no large-area cutting damage, and not obstructed by scales or text areas. The effective fields of view for the same sample are selected at intervals at different positions on the sample observation surface to avoid overlapping coverage. For example... Figure 5 As shown, SEM images of glacial till under different freeze-thaw cycles can reflect the microstructural features of the corresponding samples, such as particle arrangement, pore size, and cracks. Among them, Figure 5 (a)- Figure 5(h) shows SEM images of glacial till collected under different freeze-thaw cycles, which are used for subsequent segmentation of granular and porous phases and extraction of microstructural parameters.

[0039] Step S13: Establish a mapping between pixel size and actual size based on the scale in the SEM image.

[0040] Specifically, for each SEM image, the actual length represented by the scale is read, and the number of pixels corresponding to the scale in the image is determined. The actual length corresponding to a single pixel is determined based on the ratio of the actual length of the scale to the number of pixels of the scale, thereby establishing the mapping between pixel length and actual length, as well as the conversion relationship between pixel area and actual area.

[0041] Furthermore, the actual length of the scale, the number of scale pixels, the pixel size, and the area conversion relationship are recorded as the scale parameters of the current SEM image for subsequent scale normalization, particle size conversion, pore size conversion, and morphological processing.

[0042] Step S14: Determine the three-dimensional porosity of the sample.

[0043] Specifically, the three-dimensional porosity is determined based on the sample's mass, volume, and soil particle specific gravity. First, the dry density of the sample is determined based on the sample's mass and volume after drying to a constant mass. Then, the soil particle density is determined based on the soil particle specific gravity and the density of water at the test temperature. Further, the three-dimensional porosity is determined based on the sample's dry density and the soil particle density.

[0044] Among them, soil particle density is according to It is determined that the three-dimensional porosity is determined according to It is confirmed that, among them, Indicates soil particle density. Indicates the specific gravity of soil particles. This indicates the density of water at the test temperature. Indicates the dry density of the sample. This represents the three-dimensional porosity.

[0045] It should be noted that when there are multiple parallel samples in the same freeze-thaw cycle group, the three-dimensional porosity of each parallel sample is saved separately, and the average value and dispersion of the three-dimensional porosity of the freeze-thaw cycle group are calculated. For each SEM image, the three-dimensional porosity independently measured by the sample to which it belongs is preferentially associated; when the mass or volume data for calculating the three-dimensional porosity of the same sample cannot be obtained independently, the average value of the three-dimensional porosity of the parallel sample group to which the sample belongs is used as the corresponding three-dimensional porosity, and the substitution association is recorded.

[0046] Step S15: Establish the data correlation between SEM images and sample information.

[0047] Specifically, a data record is established for each SEM image. The data record includes at least the sample identification, sampling location, soil layer identification, slope zoning identification, number of freeze-thaw cycles, effective field of view number, image acquisition parameters, actual length of the scale, number of scale pixels, pixel size, and three-dimensional porosity.

[0048] Among them, the slope zoning identifier is used in step S6 to associate the cohesion and internal friction angle output by the strength parameter mapping model with the corresponding slope stability calculation area.

[0049] It should be noted that if a SEM image lacks a valid scale, the image acquisition parameters are inconsistent with those of other images in the same group, or the specimen identification cannot be matched, the SEM image should be marked as invalid data and reacquired. After completing the above data association, a SEM image dataset organized according to specimen identification and freeze-thaw cycle number is formed.

[0050] Step S2: Based on the mapping between the pixel size and the actual size, the SEM image is scale-normalized and preprocessed to obtain a standardized grayscale image. Specifically, using the SEM image dataset and scale parameters formed in step S1 as input, the SEM images are processed one by one according to the sample identifier and effective field of view number. Images with uniform actual scale, uniform grayscale measurement range, clear particle edges, and controlled noise level are determined as standardized grayscale images, so that subsequent comparisons of different candidate threshold segmentation images are based on a consistent data foundation. Figure 4 As shown, in this embodiment, the SEM image is sequentially scaled, enhanced in grayscale, and denoised with edge preservation to obtain a standardized grayscale image. Based on the standardized grayscale image, a global candidate threshold and a local threshold are determined to generate multiple candidate threshold segmentation maps. The initial segmentation map of the granular phase and the porous phase is determined through joint evaluation.

[0051] In this embodiment, specifically, step S2 further includes: Step S21: Verify the scale of the SEM image and obtain a valid grayscale image.

[0052] Specifically, the actual length of the ruler associated with the current SEM image is read, the number of pixels between the two endpoints in the extension direction of the ruler body is determined, and the pixel size is determined based on the ratio between the two. Specifically, the actual length represented by the ruler is evenly distributed to the corresponding number of ruler pixels to obtain the actual length represented by a single pixel, and the obtained pixel size is associated with and stored in the current SEM image.

[0053] Further, the relative deviation between the automatically read scale pixel count and the scale pixel count recorded in step S1 is calculated. When the relative deviation is no greater than 1%, the automatic reading result is adopted; when the relative deviation is greater than 1%, the scale endpoints are re-detected; if the deviation is still greater than 1% after re-detection, the scale parameters recorded in step S1 are called, and the current SEM image is marked as a scale verification abnormal image.

[0054] After completing the ruler verification, crop the ruler and text areas in the SEM image. For SEM images in the same batch where the ruler and text areas are in fixed positions, set the cropping boundary according to the fixed positions; for annotation areas where the positions are not fixed, determine the cropping boundary according to the annotation area formed by continuous highlighted characters and line rulers, and delete the annotation area while preserving the complete and effective field of view.

[0055] Furthermore, the effective field of view is converted into an 8-bit grayscale image. When the original SEM image is a single-channel grayscale image, its grayscale channel is read directly; when the original SEM image contains multiple color channels, grayscale conversion is performed according to the contribution of each color channel to the brightness; when the quantization depth of the original image is higher than 8 bits, the effective grayscale range is mapped to an integer grayscale range of 0 to 255.

[0056] Step S22: Scale normalize the SEM image.

[0057] Specifically, the median pixel size of all valid SEM images in the same batch is determined as the reference pixel size, and the relative deviation between the current image pixel size and the reference pixel size is calculated.

[0058] When the relative deviation is no greater than 1%, the pixel arrangement of the current image is retained; when the relative deviation is greater than 1%, bicubic interpolation is used to resample the effective field of view according to the scale ratio of the two, the pixel size of the resampled image is updated to the reference pixel size, and the effective field of view area is updated synchronously.

[0059] Through the above scale normalization process, the same pixel distance in SEM images of different freeze-thaw cycle groups corresponds to the same or within the allowable deviation range of actual distance, thus providing a consistent physical scale basis for the unified quantification of subsequent microstructural parameters.

[0060] Step S23: Perform grayscale enhancement and edge-preserving noise reduction on the scale-normalized SEM image.

[0061] Specifically, grayscale enhancement is achieved using local histogram equalization with contrast-limited parameters. The 8-bit grayscale image is divided into eight equal parts horizontally and vertically, forming... For each local region, the number of pixels in each local region is divided by 256 to obtain the average gray-scale frequency, and twice the average gray-scale frequency is used as the contrast limit value.

[0062] The grayscale frequencies exceeding the contrast limit are truncated, and the truncated portion is evenly distributed to the remaining grayscale levels to form a local grayscale mapping relationship. Then, bilinear interpolation is used to connect the grayscale mapping results of adjacent local areas, thereby enhancing the local grayscale difference between the granular phase and the porous phase.

[0063] Furthermore, median filtering and bilateral filtering are applied sequentially for edge-preserving noise reduction. First, median filtering of a 3×3 pixel neighborhood is used to remove isolated noise. Then, bilateral filtering is applied to the median-filtered image. The standard deviation of the spatial weights is set to 2 pixels, and the standard deviation of the grayscale weights is set to 0.1 times the standard deviation of the grayscale in the current effective field of view. This bilateral filtering allows pixels with similar grayscale values ​​and spatial proximity to participate in smoothing, and reduces the mutual influence between pixels with large grayscale differences at the boundary between the granular and porous phases. This reduces grayscale fluctuations within the granules and pores while preserving the grayscale gradient required for subsequent boundary recognition.

[0064] It should be noted that when the effective field of view grayscale standard deviation is zero, bilateral filtering is skipped and the current image is marked as a grayscale aberration image.

[0065] Step S24: Complete the physical scale conversion of image processing parameters and output a standardized grayscale image.

[0066] Specifically, based on the pixel size, the actual length and area parameters used in subsequent image processing are converted into the corresponding pixel scale.

[0067] For the local window side length, morphological structural element radius, minimum foreground marker spacing, and contour fitting tolerance, the corresponding actual lengths are converted to pixel lengths according to the pixel size and then taken as integer values ​​that meet the computational requirements. For the minimum retained area and connected component area threshold, the corresponding actual areas are converted to the number of pixels according to the actual area corresponding to a single pixel. In this way, the parameters used for subsequent image processing in SEM images of different freeze-thaw cycle groups have a consistent actual physical scale.

[0068] After completing the above processing, check the effective field size, grayscale range, scale parameters and sample association information, and output the standardized grayscale image and its associated sample identifier, freeze-thaw cycle number, effective field number, three-dimensional porosity and pixel size to step S3.

[0069] Step S3: Based on the global and local grayscale information of the standardized grayscale image, multiple candidate threshold segmentation maps are generated. The candidate threshold segmentation maps are jointly evaluated based on the difference between the two-dimensional pore area ratio and the corresponding three-dimensional porosity, particle boundary integrity, and the change in the solid-state interconnected structure with respect to the candidate threshold, to determine the initial segmentation map between the granular phase and the porous phase. Specifically, using the standardized grayscale image output from step S2 and its associated sample identifier, freeze-thaw cycle number, pixel size, and three-dimensional porosity as input, candidate threshold segmentation maps are generated for each effective field of view and jointly evaluated. This ensures that the selection of candidate thresholds is simultaneously constrained by the sample volume porosity characteristics, particle boundary integrity, and the stability of the solid-state interconnected structure.

[0070] In this embodiment, specifically, step S3 further includes: Step S31: Determine the global candidate threshold and the local threshold.

[0071] Specifically, the global baseline threshold is determined using the Otsu's method. The number of pixels corresponding to each gray level in the standardized grayscale image is statistically analyzed to form a global grayscale histogram. The segmentation locations that divide the image into two grayscale categories are traversed, and the segmentation location with the greatest dispersion between the two grayscale categories is determined as the global baseline threshold. Thus, the global baseline threshold serves as a candidate threshold center that is compatible with the overall brightness of the current effective field of view and the grayscale differences between the granular and porous phases.

[0072] Furthermore, following the order from low grayscale to high grayscale, and centering on the global baseline threshold, three candidate thresholds are set at intervals of five grayscale levels, one towards the low grayscale direction and the other towards the high grayscale direction, forming seven global candidate thresholds together with the global baseline threshold. Candidate thresholds exceeding the grayscale range of 0 to 255 are discarded, and duplicate candidate thresholds are deleted.

[0073] For local thresholds, the sliding window side length is taken as one-eighth of the number of pixels on the shorter side of the effective field of view, and adjusted to an odd number not less than 31. The window movement interval is taken as half of the window side length. The local threshold is determined by the Otsu's method within each sliding window.

[0074] It should be noted that when there are fewer than 16 effective gray levels in the window or the proportion of pixels of any gray level category is less than 5%, the bilinear interpolation result of the local threshold of the adjacent effective window is used as the local threshold of the window.

[0075] Step S32: Generate multiple candidate threshold segmentation maps.

[0076] Specifically, the global candidate thresholds and local thresholds are fused according to preset fusion weights. For the first... A global candidate threshold, according to Generate a fusion threshold map, where, Indicates the first A fusion threshold map at pixel location The fusion threshold at that point Indicates the first One global candidate threshold, Indicates pixel position Local threshold at that location The fusion weights represent the global candidate thresholds. The fusion weights represent the local thresholds. This indicates the candidate threshold number arranged in ascending order of grayscale value.

[0077] Furthermore, the pixel grayscale values ​​of the standardized grayscale image are compared pixel by pixel with the corresponding fusion threshold. Under the SEM imaging polarity of this embodiment, pixels with grayscale values ​​below the fusion threshold are marked as porous phases, and pixels with grayscale values ​​not lower than the fusion threshold are marked as granular phases, thereby generating candidate threshold segmentation maps from each fusion threshold map. Simultaneously, for each candidate threshold segmentation map, the corresponding candidate threshold number, global candidate threshold, fusion weight, and the three-dimensional porosity corresponding to the current sample are retained as evaluation data for subsequent joint evaluation.

[0078] Step S33: Calculate the joint evaluation basic index corresponding to each candidate threshold segmentation map.

[0079] First, calculate the difference between the two-dimensional pore area ratio and the three-dimensional porosity.

[0080] For the same sample at the first The two-dimensional pore area ratio is calculated for each effective field of view formed under each candidate threshold, and the average value of the two-dimensional pore area ratio of each effective field of view is taken as the average value. When the same sample has only one effective field of view, the two-dimensional pore area ratio of that effective field of view is directly used as... .

[0081] according to Calculate the porosity difference, where, Indicates the first The porosity difference corresponding to each candidate threshold This represents the corresponding two-dimensional pore area ratio. This indicates the three-dimensional porosity of the current sample.

[0082] Then, the boundary non-closure degree is calculated. From each candidate threshold segmentation map, single-pixel width grain boundaries are extracted. Grain boundary pixels with only one adjacent grain boundary pixel in the eight-neighborhood are identified as grain boundary endpoint pixels, and boundary pixels formed by truncation of effective outside-view edges are excluded.

[0083] according to Calculate the boundary non-closure degree, where, Indicates the first The degree of boundary non-closure of candidate threshold segmentation maps. Indicates the number of pixels at the grain boundary endpoints. This represents the total number of pixels at the grain boundary. When... When it is zero, the corresponding boundary non-closure degree is... Set to 1.

[0084] Next, calculate the rate of change in the number of firmly connected components. For each particle phase in the candidate threshold segmentation image, label it with eight-neighbor connected components. Traverse the particle phase pixels in the scanning order from the upper left to the lower right of the image, grouping horizontally, vertically, or diagonally adjacent particle phase pixels into the same firmly connected component. Assign different component labels to sets of unconnected particle phase pixels. After completing the merging of equivalent labels, count the total number of firmly connected components. The total number of solidly connected components is used to characterize the merging, breakage, and increase or decrease of isolated components in the particle phase under different candidate thresholds.

[0085] For the internal candidate thresholds in the candidate threshold sequence, according to Calculate the rate of change of the number of fixed connected components, where, Indicates the first Rate of change of the number of solidly connected components in a candidate threshold segmentation map. Indicates the first The number of solid connected components in a candidate threshold segmentation map. This indicates that the gray values ​​are adjacent and lower than the first gray value. The number of solid connected components in the candidate threshold segmentation graph for each candidate threshold. This indicates that the gray values ​​are adjacent and higher than the first gray value. The number of solid connected components in the candidate threshold segmentation graph for each candidate threshold. This indicates taking the maximum value of each value within the parentheses. The constant 2 is used to average the changes in two adjacent directions, and the constant 1 is used to limit the minimum value of the denominator.

[0086] For candidate thresholds located at both ends of the candidate threshold sequence, the corresponding rate of change is calculated by the change in the number of fixed connected components between the candidate threshold and its only adjacent candidate threshold.

[0087] Step S34: Determine the joint evaluation value based on the aforementioned basic indicators.

[0088] Specifically, the porosity difference, boundary non-closure degree, and rate of change of the number of solidly connected components are respectively calculated according to... Normalization is performed, where, Pick ;in, Indicates the first The corresponding evaluation indexes of each candidate threshold segmentation map are processed to the same dimension. This represents one of the following raw evaluation values: porosity difference, boundary non-closure degree, or rate of change in the number of solidly connected components. and These represent the minimum and maximum values ​​of the same type of original evaluation value in the segmentation map of all candidate thresholds, thereby converting the three types of evaluation indicators to the same numerical range.

[0089] Furthermore, according to Calculate the joint evaluation value and satisfy the following conditions: ,in, , and These represent the evaluation weights for porosity difference, boundary non-closure degree, and the rate of change in the number of solidly connected components, respectively, and all are non-negative numbers; Indicates the first The joint evaluation value of the candidate threshold segmentation maps. , and These represent the values ​​of porosity difference, boundary non-closure degree, and the rate of change of the number of solidly connected components, respectively, after being processed to the same dimensions.

[0090] Step S35: Determine the fusion weights, evaluation weights, and initial segmentation map.

[0091] Specifically, SEM images of the unfrozen control group and at least two different freeze-thaw cycle groups with associated three-dimensional porosity are selected from the SEM image dataset formed in step S1 as calibration samples.

[0092] The global candidate thresholds are then fused with weights sequentially. Set the values ​​to 0.3, 0.4, 0.5, 0.6 and 0.7, and iterate through the three evaluation weight combinations that satisfy non-negativity and have a total weight sum of 1 at intervals of 0.1.

[0093] For each set of parameters, candidate threshold selection is performed. The coefficient of variation of the average porosity difference, average boundary non-closure degree, and number of solid connected components in different effective fields of view of the same sample are calculated. The three results are normalized and summed. The parameter combination with the smallest value is determined as the preset parameter.

[0094] When multiple parameter combinations have the same sum, select the parameter combination with the smaller average porosity difference, the smaller average boundary non-closure, and the global candidate threshold fusion weight that is closer to 0.5, and keep the preset parameters unchanged in the same batch of SEM images to be tested.

[0095] Furthermore, the joint evaluation values ​​of all candidate threshold segmentation maps are compared, and the candidate threshold segmentation map with the smallest joint evaluation value is determined as the initial segmentation map.

[0096] When two or more candidate threshold segmentation maps have the same minimum joint evaluation value, the candidate threshold segmentation map with the smaller porosity difference, smaller boundary non-closure degree, and global candidate threshold that is closer to the global benchmark threshold is selected in sequence.

[0097] It should be noted that when the minimum joint evaluation value appears at the lowest or highest end of the candidate threshold sequence, candidate thresholds are added along the corresponding grayscale direction and re-evaluated until the candidate threshold corresponding to the minimum joint evaluation value is within the effective candidate range or reaches the 8-bit grayscale value boundary.

[0098] After the initial segmentation map is determined, the granular phase and porous phase in the final selected candidate threshold segmentation map are assigned fixed binary labels respectively. The two-dimensional pore area ratio, porosity difference, boundary non-closure degree, number of solid-connected components, rate of change of number of solid-connected components, joint evaluation value and fusion threshold parameters are recorded. The correlation between the initial segmentation map and the sample identification, number of freeze-thaw cycles, three-dimensional porosity and pixel size are maintained.

[0099] Step S4: Identify the adherent particle components based on the initial segmentation map, perform labeling and control watershed segmentation on the adherent particle components, and merge the segmented sub-regions according to the area of ​​the segmented sub-regions and the grayscale features of the shared boundaries of adjacent segmented sub-regions to obtain a particle and pore boundary map. For example... Figure 6 As shown, this embodiment sequentially performs identification, distance transformation, foreground marking, marker-controlled watershed segmentation, over-segmentation sub-region merging, and boundary correction on the adhered particle component. Among these steps... Figure 6 (a) shows the initial fixed connection assembly. Figure 6 (b) shows the results of adhering particle identification based on area and concavity. Figure 6 (c) shows the result of the Euclidean distance transformation. Figure 6 (d) shows the results of determining the foreground label based on local maxima. Figure 6 (e) shows the results of the marker-controlled watershed segmentation. Figure 6 (f) shows the results of merging small-area sub-regions. Figure 6 (g) shows the results of low-gradient shared boundary merging. Figure 6 (h) shows the particle boundary processing results after morphological opening, closing operations and contour polygon fitting.

[0100] In this embodiment, step S4 further includes: Step S41: Identify Adhesive Particle Components. Specifically, using the initial segmentation image output from step S3, the normalized grayscale image output from step S2, and their associated pixel dimensions as input, the particle phases in the initial segmentation image are labeled with eight-neighbor connected components, and the pixel set, boundary coordinates, and adjacent pore regions of each firmly connected component are recorded. Thus, each firmly connected component and its corresponding normalized grayscale and scale information form a processing object for adhesive particle identification, automatic segmentation, and boundary correction.

[0101] Further, the area and convex hull area of ​​each fixedly connected component are calculated. The number of pixels of each particle in the same component is taken as the pixel area of ​​the fixedly connected component. Ordered boundary points are extracted along the outer boundary of the component, and a minimum convex polygon that can enclose all boundary points is constructed. The pixel area covered by the minimum convex polygon is taken as the convex hull area. Here, the convex hull is used to represent the outer shape of the corresponding fixedly connected component after filling the boundary depression. The area difference between the fixedly connected component and its convex hull is used to reflect the degree of contour depression formed at the contact point of adjacent particles.

[0102] according to Calculate the concavity of the component, where, This indicates the serial number of the fixed connection component. Indicates the first The concavity of a fixedly connected component Indicates the first The convex hull area of ​​a fixedly connected component. Indicates the first The area of ​​a single solidly connected component. The adhesive particle component refers to a single solidly connected component formed by multiple contacting particles in the initial segmentation diagram.

[0103] When satisfied or At that time, the corresponding fixed connection component is determined as the adhesive particle component, wherein, This represents a preset multiple of the median area. This represents the median area of ​​all fixed-connected components. This indicates the preset concavity threshold.

[0104] Step S42: Determine the preset parameters used for segmenting adhering particles based on the calibration sample.

[0105] Specifically, the preset multiplier, preset concavity threshold, foreground marker spacing scale coefficient, and morphological structural element scale coefficient are determined based on calibration samples before processing the image to be tested.

[0106] The calibration samples are SEM images obtained using the same material source and SEM imaging conditions as the glacial till soil to be tested. In this embodiment, valid SEM images from the dataset formed in step S1, representing 0, 8, and 20 freeze-thaw cycles, are selected as calibration samples.

[0107] Furthermore, the system sequentially iterates through preset multiples of 1.5 to 3.0 with an interval of 0.25, preset concavity thresholds of 0.05 to 0.30 with an interval of 0.05, foreground marker spacing scale coefficients of 0.20 to 0.40 with an interval of 0.05, and morphological structural element scale coefficients of 0.02 to 0.05 with an interval of 0.01.

[0108] For each parameter combination, the identification and segmentation of adhering particles are performed. The average distance between the segmentation boundary and the edge pixels with a gray-level gradient of not less than the 75th percentile in the standardized gray-level image, the proportion of sub-regions with an area smaller than the minimum retention area, and the relative change rate of the number of particle regions when the initial segmentation threshold changes by one candidate threshold interval are normalized and summed. The parameter combination with the smallest value is determined as the preset parameter and kept fixed in the same batch of SEM images to be tested.

[0109] Step S43: Determine the foreground markers for watershed segmentation.

[0110] Specifically, an Euclidean distance transformation is performed on the adhered particle assembly. For each particle pixel within the assembly, the Euclidean distance to the nearest boundary pixel is calculated, and this distance is used as the distance transformation value for the corresponding pixel, forming a distance transformation map. Regions with larger distance transformation values ​​correspond to particle interior regions far from the assembly boundary. The distance transformation values ​​gradually decrease from the particle interior towards the particle contact area and the outer boundary of the assembly. Thus, local maxima in the distance transformation map are used to characterize the internal positions of potential independent particles.

[0111] Furthermore, local maxima in the distance transformation map are used as foreground markers. The diameter of the equal-area circle for each component is determined based on the area of ​​the solidly connected components in the initial segmentation map. Solidly connected components not identified as bonded particle components are preferentially converted to equal-area circle diameters, and their medians are taken as the median of the particle's equivalent diameter. .

[0112] When there are fewer than three solidly connected components that are not identified as bonded particle components, the median of the equivalent particle diameter is calculated from all solidly connected components whose area is not greater than the median of the component area.

[0113] according to Determine the minimum spacing between adjacent foreground markers, where, This represents the minimum spacing between adjacent foreground markers, expressed in pixels. This represents the foreground marker spacing scale factor. This represents the median of the particle's equivalent diameter in units of actual length. This represents the actual length corresponding to a single pixel.

[0114] For multiple local maxima with a spacing smaller than the minimum spacing, only the local maxima with a larger distance transformation value are retained; for local maximum platforms with the same distance transformation value, the center position of the platform is determined as a single foreground marker.

[0115] Step S44: Perform marker-controlled watershed segmentation and merge the segmented sub-regions. Specifically, each foreground marker is used as the starting region inside the corresponding particle, the external region of the bonded particle component is set as the background constraint region, and a watershed terrain is constructed using the inverse transformation relationship of the distance transformation map, so that the regions controlled by different foreground markers expand outward along the direction of decreasing distance transformation value; when the expanded regions from different foreground markers meet, a candidate watershed boundary is formed, thereby dividing a bonded particle component into multiple segmented sub-regions with independent region markers. Fixed connected components not identified as bonded particle components retain their original region markers.

[0116] Furthermore, 5% of the median area of ​​the solid connected components that were not identified as adhesive particle components was converted into the number of pixels and compared with 9 pixels. The larger of the two values ​​was determined as the preset minimum reserved area.

[0117] The process iterates through each segmented sub-region, identifying those with an area smaller than the preset minimum retention area as area anomaly sub-regions, and merging them into the adjacent sub-region with the longest shared boundary. When two or more adjacent sub-regions share the same longest shared boundary, the area anomaly sub-regions are merged into adjacent sub-regions with smaller average grayscale differences. It should be noted that the over-segmented sub-regions refer to multiple sub-regions formed by the label control watershed dividing the same actual particle. This embodiment first merges small sub-regions formed by local noise or unstable local maxima based on the preset minimum retention area, and then determines whether the remaining candidate watershed boundaries have grayscale changes that support independent particle segmentation based on the average grayscale gradient of the shared boundary. This preserves particle segmentation results with boundary grayscale support and reduces the overestimation of particle count caused by an excessive number of local maxima.

[0118] Based on this, the grayscale changes on both sides of each candidate watershed boundary are read on the standardized grayscale image output in step S2. The grayscale gradient corresponding to the boundary normal is calculated pixel by pixel along the shared boundary and the average value is obtained. The average grayscale gradients of the shared boundary of all adjacent segmented sub-regions are used to form a boundary gradient sample set.

[0119] The maximum inter-class variance method is used to divide the boundary gradient sample set into a low-gradient boundary set and a high-gradient boundary set, and the gray-level gradient value that maximizes the inter-class variance between the two boundary sets is determined as the gradient merging threshold.

[0120] It should be noted that when there are fewer than two boundary gradient samples or all boundary gradients are the same, sub-region merging based on grayscale gradients will not be performed.

[0121] Adjacent segmented regions with an average gray-level gradient of shared boundaries less than the gradient merging threshold are merged, while shared boundaries with an average gray-level gradient greater than or equal to the gradient merging threshold are retained. When a merge causes changes to other adjacency relationships, the average gray-level gradient of the affected shared boundaries is recalculated until no adjacent segmented regions satisfy the merging conditions are found.

[0122] Step S45: Perform morphological correction and contour polygon fitting on the particle boundaries.

[0123] Specifically, according to Determine the radius of the disk structural element and convert the result into a positive integer that satisfies the requirements of morphological operations, where, Represents the radius of the disk structure element, in pixels. This represents the scale coefficient of the morphological structural element. In this embodiment, it should be noted that the scale coefficient of the morphological structural element adopts the parameter value determined in step S42 based on the calibration sample, so that the disk structural elements in SEM images of different pixel sizes correspond to similar actual processing scales.

[0124] The disk-shaped structural element is used to perform morphological opening and closing operations on each merged particle region. Specifically, a morphological opening operation is first performed on a single particle region to remove protrusions, spikes, and residual noise whose edge width is smaller than the scale of the disk-shaped structural element; then a morphological closing operation is performed on the same particle region to fill gaps and narrow depressions near the boundary whose width is smaller than the scale of the disk-shaped structural element.

[0125] After each particle region independently completes morphological operations, it is recombined according to its original region label. When overlapping pixels are generated after different particle regions undergo closing operations, the Euclidean distance from the overlapping pixels to the foreground label of each particle region is calculated, and the overlapping pixels are assigned to the nearest particle region; when the distances are the same, the original region label is maintained along the preserved watershed boundary.

[0126] Furthermore, 1% of the median of the particle's equivalent diameter is divided by the pixel size and rounded to an integer pixel. The resulting value is compared with 1 pixel, and the larger of the two values ​​is taken as the contour fitting tolerance.

[0127] Ordered contour points are extracted from each particle region after morphological processing. The first and last points of the contour point sequence are connected to form an initial line segment. The vertical distance from the intermediate contour point to the line segment is calculated. When the maximum vertical distance exceeds the contour fitting tolerance, the current contour segment is split with the corresponding contour point, and the same judgment is performed on the split contour segments until the maximum vertical distance of each contour segment does not exceed the contour fitting tolerance, thereby obtaining a particle contour polygon composed of finite vertices and line segments.

[0128] Finally, the particle regions that have undergone oversegmentation and merging, morphological processing, and contour polygon fitting are written into the particle boundary map according to the region labels. The complementary regions that are not occupied by particle regions within the effective field of view are identified as the porous phase and the pore boundaries are extracted to obtain the particle and pore boundary map.

[0129] Simultaneously, the marker number, pixel area, actual area calculated based on pixel size, contour coordinates, and adjacent pore regions of each particle region are recorded, while maintaining the correspondence between the particle and pore boundary map and the sample identification, freeze-thaw cycle number, three-dimensional porosity, and pixel size. The particle and pore boundary map and its associated data are output to step S5 for extracting microstructural parameters such as particle and pore size, morphology, orientation features, and pore connectivity.

[0130] Step S5: Extract mesoscopic structural parameters characterizing particle morphology, pore structure, solid phase ratio, and particle orientation based on the particle and pore boundary map, and determine the pore connectivity characterizing the pore connectivity state to form a mesoscopic parameter vector. Specifically, using the particle and pore boundary map, particle region markers, pore phase markers, and pixel dimensions output from step S4 as input, calculate the particle area, particle perimeter, and contour coordinates for each particle region marker, and calculate the pore area and pore boundary for each of the eight neighboring pore connected regions. Convert the length and area in pixels to actual length and actual area, thereby forming a mesoscopic structural parameter record corresponding to the sample identification, freeze-thaw cycle number, and effective field of view number.

[0131] In this embodiment, step S5 further includes: Step S51: Extract the size parameters of particles and pores, as well as the irregularity of particle morphology. Specifically, according to the particle region markers, the particle area, particle perimeter, and contour coordinates are counted one by one. According to the eight neighboring pore connected regions, the pore area and pore boundary are counted one by one. The length and area in pixels are then converted into actual length and actual area.

[0132] The diameter of a circle with the same area as the particle or pore is defined as the equivalent diameter of the corresponding particle or pore. The mean, median, and distribution range of the equivalent diameters of particles and pores within the same effective field of view are then statistically analyzed. This equivalent diameter is used to characterize the particle and pore sizes of different contour shapes using a uniform circular scale, thus facilitating comparison of changes in particle and pore sizes under different freeze-thaw cycles. Figure 7 As shown, by statistically analyzing the equivalent particle diameters in each effective field of view under different freeze-thaw cycles, the distribution curves of the equivalent particle diameters corresponding to different freeze-thaw cycles can be obtained. Figure 8 As shown, by statistically analyzing the equivalent pore diameters in each effective field of view under different freeze-thaw cycles, the distribution of equivalent pore diameters corresponding to different freeze-thaw cycles can be obtained, which can be used to characterize the distribution of pore size under different freeze-thaw cycles.

[0133] Furthermore, the irregularity of particle shape is determined by the ratio of the square of the particle's perimeter to the product of four times pi and the particle's area. ,in, Indicates the irregularity of particle morphology. Indicates the perimeter of the particle. This represents the particle area. The particle perimeter is obtained by summing the actual distances between adjacent vertices in the particle outline polygon. The closer the particle outline is to a circle, the closer the particle morphology irregularity is to its theoretical lower limit. The more protrusions, depressions, and bends in the particle outline, the larger the particle perimeter becomes when the particle area is the same or similar, thus increasing the particle morphology irregularity.

[0134] Step S52: Determine the particle orientation angle and the standard deviation of the particle orientation angle.

[0135] Specifically, the particle orientation angle is determined based on the principal eigenvector of the particle pixel coordinate covariance matrix. Using the centroid of each particle region as the coordinate center, all pixel coordinates within the particle region are converted to coordinates relative to the centroid, constructing the particle pixel coordinate covariance matrix. The principal eigenvector corresponding to the largest eigenvalue is then determined as the particle's major axis direction, and its angle relative to the horizontal direction of the image is transformed to... to The range is used as the particle orientation angle. Specifically, the particle pixel coordinate covariance matrix is ​​constructed based on the degree of coordinate dispersion of pixels in the horizontal direction and the degree of coordinate dispersion in the vertical direction within the particle region, as well as the common variation relationship between the coordinates in the two directions.

[0136] Furthermore, axial angle statistics are performed on all particle orientation angles within the same effective field of view. Each particle orientation angle is then doubled before calculating the circumferential average direction. The angle difference between the doubled particle orientation angle and the circumferential average direction is then converted to... to Within the specified range, the angle difference is further reduced to half and the standard deviation is calculated to obtain the standard deviation of the particle orientation angle. A smaller standard deviation indicates a more concentrated particle orientation direction, while a larger standard deviation indicates a more dispersed particle orientation direction.

[0137] Step S53: Determine the pore fractal dimension and projected solid area fraction.

[0138] Specifically, the box counting method is used to determine the pore fractal dimension. Square grids with different side lengths are sequentially superimposed on the pore boundary map. The number of grids intersecting the pore boundary at each grid scale is counted. A linear fit is performed on the logarithm of the grid side length and the logarithm of the corresponding number of covered grids, and the absolute value of the slope of the fitted line is determined as the pore fractal dimension. For example... Figure 9 As shown, the particle morphology irregularity and pore fractal dimension were statistically analyzed under different freeze-thaw cycles. Figure 9 (a) shows how the irregularity of particle morphology changes with the number of freeze-thaw cycles. Figure 9 (b) shows how the pore fractal dimension changes with the number of freeze-thaw cycles.

[0139] The grid side lengths are increased sequentially by 2 pixels, 4 pixels, 8 pixels, and 16 pixels, and continue to increase until they do not exceed one-eighth of the shorter side length of the effective field of view. Linear fitting is performed using no fewer than four grid scales.

[0140] When the coefficient of determination is not less than 0.95, the absolute value of the slope of the fitted line is determined as the pore fractal dimension; when the coefficient of determination is less than 0.95 for all available grid scales, the current effective field of view is marked as the fractal parameter anomalous field of view and is not included in the calculation of the average pore fractal dimension of the same sample.

[0141] Furthermore, the projected solid phase area ratio is determined based on the ratio of the difference between the effective field of view area and the porosity phase area to the effective field of view area. The porosity phase area is obtained by summing the actual surfaces corresponding to all porosity phase pixels within the effective field of view. The projected area of ​​the particle phase on the observation plane is obtained by subtracting the porosity phase area from the effective field of view area, and then the ratio of this projected area to the effective field of view area is calculated.

[0142] It should be noted that the projected solid area ratio is used to characterize the area occupied by the particulate phase on the SEM observation plane. It is a parameter in a different spatial dimension than the three-dimensional porosity obtained in step S1. The former characterizes the solid phase ratio in the two-dimensional observation plane, while the latter characterizes the porosity ratio in the sample volume scale.

[0143] Specifically, the pore phase obtained in step S4 is used as a unified analysis object. Distance transformation index, connectivity index, multi-scale morphology index, and local neighborhood index are calculated from four aspects: pore internal width, overall connectivity of pore region, cross-scale preservation of pore structure, and local aggregation of pore pixels. The corresponding fusion weight is selected according to the interval to which the two-dimensional pore area ratio belongs, so that the pore connectivity reflects the structural characteristics of large-scale connected pores and locally aggregated pores.

[0144] Step S54: Determine the pore connectivity. For example... Figure 10 As shown, this embodiment determines the distance transformation index, connectivity index, multi-scale morphology index, and local neighborhood index respectively, and determines the fusion weight of the four indices according to the interval to which the two-dimensional pore area ratio belongs. The four indices are then weighted and fused to obtain the pore connectivity.

[0145] Specifically, the distance transformation index, connected region index, multi-scale morphology index, and local neighborhood index are calculated separately, and the four indices are weighted and fused.

[0146] First, perform a Euclidean distance transformation on the porous phase, and then... Calculate the distance transformation index, wherein the distance transformation index is used to characterize the width distribution inside the aperture, and the summation operation is used to accumulate the normalized distance transformation values ​​of all aperture phase pixels; This represents the total number of aperture phase pixels within the effective field of view. Indicates the first The Euclidean distance transformation value from each aperture phase pixel to the nearest aperture boundary. This represents the maximum Euclidean distance transformation value of all pore phase pixels.

[0147] It should be noted that when zero or When it is zero, Set to 0.

[0148] Eight-neighbor connectivity component labeling is performed on the porous phase, and according to... Calculate the connectivity index, where, This represents the number of pixels contained in the largest eight-neighbor interconnected region. This indicates the number of interconnected regions in the eight neighboring pores.

[0149] when When it is zero, Set to 0.

[0150] Morphological opening operations were performed on the porous phase using disk structure elements with radii of 1 pixel, 3 pixels, and 5 pixels, respectively, and then... Calculate multi-scale morphological indices, among which, Indicates the use of radius. The number of aperture phase pixels retained after performing morphological opening operations on the disk structural element; the constant 3 indicates the number of morphological scales used. Represents the pixel radius of a disk structure element, set The radius of the three disk structural elements is represented by the summation operation, which is used to accumulate the pore area retention rate at three morphological scales to characterize the degree of preservation of the pore structure at different morphological scales.

[0151] For example, when there are no aperture phase pixels within the effective field of view, i.e. When the value is zero, multi-scale morphological opening operations based on the porous phase are not performed, and the multi-scale morphological indices are... Set it to 0 to avoid calculation errors caused by a denominator of zero.

[0152] Furthermore, a 5×5 pixel neighborhood is constructed centered on each aperture phase pixel, and then... Calculate the local neighborhood index, where, Indicates the first A pore phase marker for each neighboring pixel, with a value of 1 for pore phase and a value of 0 for particle phase; Indicates the first A 5×5 pixel neighborhood centered on each aperture pixel and located within the effective field of view. This represents the neighboring pixels in the neighborhood. Indicates the first The central aperture phase pixel and the first Gaussian weights between neighboring pixels and These represent the coordinate differences between the two in the horizontal and vertical directions, respectively. The local neighborhood index is used to characterize the degree of local clustering of aperture pixels in the Gaussian weighted neighborhood.

[0153] according to Determine the Gaussian weights, where, Take a pixel of 1.25, and assign weights to pixels corresponding to the central aperture. Set to 0.

[0154] Aperture phase pixels located at the edge of the effective field of view only use pixels that actually exist in their neighborhood and are normalized according to the sum of actual Gaussian weights. When If the sum of Gaussian weights in a local neighborhood is zero, the corresponding local neighborhood calculation result is set to 0.

[0155] Finally, the two-dimensional pore area ratio is determined based on the pore phase obtained in step S4, and the fusion weight of the four indicators is determined based on the interval to which the two-dimensional pore area ratio belongs.

[0156] according to Calculate the pore connectivity, where, Indicates pore connectivity. Indicates the distance transformation index. Indicates the connectivity index. This represents a multi-scale morphological index. Indicates local neighborhood index, This represents the fusion weights corresponding to the distance transformation index. This represents the fusion weight corresponding to the connectivity region index. This represents the fusion weights corresponding to multi-scale morphological indicators. The fusion weights corresponding to the local neighborhood indicators; , , and All are non-negative numbers, and the sum of the four fusion weights is 1.

[0157] When the two-dimensional pore area ratio is small, the fusion weight of local neighborhood indicators is increased to enhance the contribution of local pore aggregation features to pore connectivity. When the two-dimensional pore area ratio is large, the fusion weight of distance transformation indicators and connected region indicators is increased to enhance the contribution of pore width and overall connectivity to pore connectivity.

[0158] In this embodiment, the preset pore area ratio threshold is set to 0.1. When the two-dimensional pore area ratio is less than 0.1, the fusion weights of the distance transformation index, connected region index, multi-scale morphology index, and local neighborhood index are set to 0.2, 0.2, 0.2, and 0.4 respectively; when the two-dimensional pore area ratio is not less than 0.1, the fusion weights of the four indices are set to 0.3, 0.3, 0.2, and 0.2 respectively.

[0159] Step S55: Form the micro-parameter vector.

[0160] Specifically, microstructural parameters are recorded for multiple effective fields of view of the same sample, and the average value and dispersion of each parameter are calculated according to the sample identification and the number of freeze-thaw cycles.

[0161] Furthermore, the particle morphology irregularity, pore fractal dimension, projected solid area ratio, particle orientation angle standard deviation, and pore connectivity are arranged in a fixed order to form a micro-parameter vector. This micro-parameter vector is output to step S6 and used to input the strength parameter mapping model calibrated by the glacial till mechanical test together with the number of freeze-thaw cycles. The dimensional parameters such as the equivalent diameter of particles and pores are retained as auxiliary parameters to characterize the evolution of the microstructure of glacial till under different freeze-thaw cycles.

[0162] Step S6: Input the number of freeze-thaw cycles and the micro-parameter vector into the strength parameter mapping model calibrated by glacial till mechanics tests to obtain cohesion and internal friction angle. Use the cohesion and internal friction angle for slope stability calculation and determine graded early warning signals based on the safety factor. Figure 12 As shown, in this embodiment, the number of freeze-thaw cycles and the micro-parameter vector are input into the strength parameter mapping model to obtain cohesion and internal friction angle. The cohesion and internal friction angle are then used together with the slope cross-sectional geometric parameters, soil strip weight, slip surface inclination angle, and pore water pressure to calculate slope stability and obtain the slope safety factor. Furthermore, the warning level is determined based on the slope safety factor and the classification threshold, and the warning terminal executes the corresponding warning operation.

[0163] Specifically, using the number of freeze-thaw cycles recorded in step S1, the micro-parameter vector output in step S5, and the strength parameters obtained from mechanical tests of glacial till under the same freeze-thaw cycle conditions as data sources, a correspondence between the evolution of the microstructure of glacial till and the change in macroscopic shear strength is established. The cohesion and internal friction angle output by the strength parameter mapping model are then used for slope stability calculation and graded early warning.

[0164] In this embodiment, step S6 further includes: Step S61: Establish the calibration sample set for the intensity parameter mapping model.

[0165] Specifically, direct shear tests were conducted on till soil samples with the same soil source, initial moisture content, dry density, and freeze-thaw conditions as the SEM samples. Shear failure strength was obtained under multiple preset normal stress conditions, and the cohesion and internal friction angle of each freeze-thaw cycle group were determined based on the relationship between shear failure strength and normal stress. By forming calibration records on a per-freeze-thaw cycle basis, the cohesion and internal friction angle of the same freeze-thaw cycle group were avoided from being repeatedly mapped to multiple effective SEM fields of view, ensuring that microscopic parameters and mechanical parameters corresponded according to the same freeze-thaw cycle conditions.

[0166] For multiple SEM samples in the same freeze-thaw cycle group, the microscopic parameter vectors are calculated and the group average vector is obtained. For multiple direct shear samples in the same freeze-thaw cycle group, the cohesion and internal friction angle are determined and the group average is obtained. The group average vectors of microscopic parameters, the group average value of cohesion, and the group average value of internal friction angle under the same number of freeze-thaw cycles are combined into a calibration record, thereby forming a calibration sample set.

[0167] Step S62: Determine the reference vector and standardize the microscopic parameters.

[0168] Specifically, the average vector of mesoscopic parameters of the zero-freeze-thaw sample is used as the reference vector. The reference vector is used to represent the initial microstructure state of glacial till before it has undergone freeze-thaw cycles, and serves as a unified reference for measuring the direction and degree of change of microstructure parameters under different freeze-thaw cycles.

[0169] The mean and standard deviation of each dimension of the micro-parameters in the calibration sample set are used to standardize the vector of the micro-parameter to be measured and the reference vector on the same scale. For the first... Each parameter is used to calculate its mean in the calibration sample set. and standard deviation ,according to The microscopic parameters to be measured are standardized, among which, Represents the first element in the vector of mesoscopic parameters to be measured. The original values ​​of the microscopic parameters. Indicates the first The standardized values ​​of the microscopic parameters. Indicate the dimension number of the micro-parameter; and use the same mean. and standard deviation The corresponding parameters in the reference vector are standardized so that the microscopic parameter vector to be measured is at the same standardized scale as the reference vector.

[0170] It should be noted that when the following conditions are met... In this case, the standardized result of the micro-parameter in all samples is set to 0, and the change correction coefficient corresponding to the micro-parameter in the intensity parameter mapping model is fixed to 0. This avoids amplifying local measurement errors or numerical fluctuations during the standardization process due to the small actual variation of a certain micro-parameter in the calibration samples.

[0171] Step S63: Establish a strength parameter mapping model using the least squares method with a regularization term. Specifically, the calibration samples are divided into model fitting data and model validation data according to the number of freeze-thaw cycles and the sample identification. The sum of squared residuals between the output value of the strength parameter mapping model and the calibration value of the direct shear test is used as the fitting error, and a regularization term for constraining the amplitude of the model coefficients is added to the fitting error.

[0172] Specifically, the regularization strength is set to... , , , , , 1, 10 and 100.

[0173] A cross-validation approach was adopted, reserving one set of data for each freeze-thaw cycle. One set of freeze-thaw cycles was used as validation data, and the remaining sets were used as fitting data. The mean square error of validation for cohesion and internal friction angle was calculated separately. The two mean square errors were normalized and summed, and the regularization intensity with the smallest sum was determined as the final regularization parameter. This regularization process reduces the correlation between mesoscopic parameters and mitigates the impact of a limited number of calibration samples on the stability of the strength parameter mapping model coefficients.

[0174] Furthermore, an intensity parameter mapping model is established, consisting of a decay term based on the number of freeze-thaw cycles and a correction term for the change of the standardized micro-parameter vector relative to the standardized baseline vector.

[0175] Cohesion according to The internal friction angle is obtained according to... Obtained. Among them, and These represent the reference cohesion and reference internal friction angle of the zero-freeze-thaw sample, respectively. and These represent the attenuation coefficients of cohesion and internal friction angle as a function of the number of freeze-thaw cycles, respectively. Indicates the number of freeze-thaw cycles. and These represent the vectors of correction coefficients for the microscopic parameter variations corresponding to cohesion and internal friction angle, respectively. This represents the standardized vector of the mesoscopic parameters to be measured. This represents the standardized reference vector. Represents the natural constant. This indicates the transpose of a vector and its inner product with the following vector.

[0176] The reference cohesion, reference internal friction angle, two attenuation coefficients, and two micro-parameter variation correction coefficient vectors are determined using a calibration sample set, and the two attenuation coefficients are constrained to be non-negative.

[0177] After determining the regularization parameters based on the cross-validation results, the model parameters are redefined using all calibration records. The resulting model coefficients, regularization parameters, parameter ranges for each freeze-thaw cycle group, and cross-validation errors are then saved as a fixed strength parameter mapping model. Therefore, in subsequent analysis of the test sample, inputting the number of freeze-thaw cycles and the micro-parameter vector of the test sample into the fixed strength parameter mapping model yields the cohesion and internal friction angle corresponding to the current freeze-thaw state and microstructure state. Figure 11 As shown, the strength parameters output by the strength parameter mapping model are compared with the calibration data obtained from direct shear tests. Figure 11 (a) Comparison of the output value and the calibration value of the internal friction angle under different freeze-thaw cycles. Figure 11(b) shows the comparison between the cohesion output value and the calibration value under different freeze-thaw cycles.

[0178] Step S64: Calculate slope stability based on the strength parameter mapping results.

[0179] Specifically, the number of freeze-thaw cycles of the test sample and the micro-parameter vector formed in step S5 are input into the fixed strength parameter mapping model to obtain the cohesion and internal friction angle.

[0180] Based on the sampling location, soil layer identifier, and slope zoning identifier recorded in step S1, cohesion and internal friction angle are associated with the corresponding slope stability calculation area. When multiple valid samples exist for the same slope zoning, the average value of the output results of each sample model is used as the cohesion and internal friction angle for that slope zoning; if no corresponding slope zoning identifier is found, the current stability calculation is stopped and data association anomaly information is output. Thus, the slope stability calculation uses strength parameters determined based on the number of freeze-thaw cycles and the quantitative analysis results of SEM microstructure.

[0181] Furthermore, the cohesion and internal friction angle output from the strength parameter mapping model are used as strength parameter inputs to simplify the Bishop slice model.

[0182] The simplified Bishop slice model also receives slope cross-sectional geometric parameters, soil slice weight, slip surface dip angle, and pore water pressure. The slope cross-sectional geometric parameters include slope height, slope angle, slope crest boundary, slope toe boundary, and the spatial location of candidate slip surfaces, and are obtained from cross-sectional measurement data corresponding to the current slope partition.

[0183] The potential sliding body is divided into multiple vertical soil strips based on the slope cross-section and candidate sliding surfaces. The weight of each soil strip is determined based on the cross-sectional area, calculated width, and unit weight of the soil layer to which it belongs. The dip angle of the sliding surface is determined based on the spatial coordinates of the candidate sliding surface at both ends of the bottom of the corresponding soil strip.

[0184] The pore water pressure is obtained from the pore water pressure monitoring data of the current slope zone; when no pore water pressure monitoring equipment is installed, the pore water pressure is determined based on the head difference between the design groundwater level and the bottom elevation of the soil strip.

[0185] The slope cross-sectional geometric parameters, soil strip weight, slip surface inclination angle, pore water pressure, and the cohesion and internal friction angle output from the strength parameter mapping model are all written into the current stability calculation task according to the same slope zone identifier.

[0186] In this embodiment, the initial value of the safety factor of the current candidate sliding surface is set to 1, and the safety factor is updated iteratively. The iteration ends when the absolute difference between the current safety factor and the previous safety factor is not greater than $10^{-4}$; when the convergence condition is not met after 100 iterations, the current candidate sliding surface is marked as a computational anomaly.

[0187] When there are multiple candidate sliding surfaces, the minimum safety factor among all normally converged candidate sliding surfaces is determined as the current slope safety factor.

[0188] Step S65: Determine the warning level based on the slope safety factor and output the warning result.

[0189] Specifically, the safety factor is compared with the pre-set grading threshold based on the slope grade and applicable engineering standards to determine the warning level.

[0190] The early warning configuration table stores the upper and lower limits of the safety factor corresponding to each early warning level, according to the slope grade, project type, and applicable engineering standard. Each early warning configuration record includes at least the slope grade, project type, applicable engineering standard number, standard validity period, and the safety factor range corresponding to each early warning level.

[0191] The processor reads the corresponding early warning configuration record based on the current slope identifier and verifies the applicable engineering standard and its validity status corresponding to the early warning configuration record. For valid early warning configuration records, the threshold range where the slope safety factor is located is determined as the corresponding early warning level, and when the safety factor is equal to the boundary value of the adjacent range, it is processed according to the early warning level with higher risk.

[0192] It should be noted that when the current slope identifier cannot match a valid early warning configuration record, the determination of the early warning level and the output of threshold configuration anomaly information to the early warning terminal will be stopped, and no arbitrary default safety factor threshold will be used to replace the graded threshold specified in the applicable engineering standard.

[0193] Furthermore, a warning data packet containing at least strength parameters, a safety factor, and a warning level is generated. The cohesion and internal friction angle output by the strength parameter mapping model, along with the sample identifier, number of freeze-thaw cycles, slope identifier, safety factor, warning level, model version, and generation time, are written into the warning data packet according to the preset communication fields and data order, and then sent to the warning terminal via a wired or wireless communication interface.

[0194] After receiving the warning data packet, the warning terminal verifies the data integrity and parses the intensity parameters, safety factor and warning level. Then, based on the warning level, it performs at least one of the following operations: displaying the results, recording and storing them, providing audible and visual prompts, or sending an alarm message to the engineering monitoring platform.

[0195] Specifically, for results that meet the display conditions, the visualization interface is updated; all received results are written to a historical record with a time stamp; for results that reach the audio-visual alert level, indicator lights and sound alarms are activated; and for results that reach the platform alarm level, an alarm message is generated that includes at least the slope identifier, safety factor, warning level, and generation time, and sent to the engineering monitoring platform.

[0196] Finally, the processor stores the sample identifier, sampling location, soil layer identifier, slope zoning identifier, and number of freeze-thaw cycles recorded in step S1, the micro-parameter vector formed in step S5, the cohesion and internal friction angle output by the strength parameter mapping model, the safety factor output by the simplified Bishop slice model, and the early warning operation executed by the early warning terminal according to the corresponding association between the sample identifier and the slope zoning identifier. This forms a continuous engineering application chain from SEM images of glacial till after freeze-thaw cycles, micro-structure parameters, macro-strength parameters, slope stability results to graded early warning signals.

[0197] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

[0198] This background section is provided to generally present the context of the invention. The work of the currently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither expressly nor impliedly acknowledged as prior art to the invention.

Claims

1. A quantitative SEM analysis method for the microstructure of glacial till after freeze-thaw cycles, characterized in that, include: Step S1: Obtain SEM images, scales, and three-dimensional porosity of glacial till under different freeze-thaw cycles, and establish a mapping between pixel size and actual size; Step S2: Based on the mapping between the pixel size and the actual size, the SEM image of the glacial till is scaled and preprocessed to obtain a standardized grayscale image; Step S3: Generate multiple candidate threshold segmentation maps based on the global and local grayscale information of the standardized grayscale image. Jointly evaluate each candidate threshold segmentation map according to the difference between the two-dimensional pore area ratio and the corresponding three-dimensional porosity, the integrity of the particle boundary, and the change of the solid interconnection structure with the candidate threshold, and determine the initial segmentation map of the particle phase and the porous phase. Step S4: Identify the adhering particle components based on the initial segmentation map, perform mark control watershed segmentation on the adhering particle components, and merge the segmented sub-regions according to the area of ​​the segmented sub-regions and the grayscale features of the shared boundaries of adjacent segmented sub-regions to obtain a particle and pore boundary map. Step S5: Based on the particle and pore boundary map, extract the microstructural parameters that characterize particle morphology, pore structure, solid phase ratio and particle orientation, and determine the pore connectivity that characterizes the pore connectivity state to form a microstructural parameter vector. Step S6: Input the number of freeze-thaw cycles and the micro-parameter vector into the strength parameter mapping model calibrated by the glacial till mechanics test to obtain cohesion and internal friction angle. Use the cohesion and internal friction angle for slope stability calculation and determine the graded early warning signal based on the calculated safety factor.

2. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S1 includes: Prepare glacial till samples of uniform size, set up multiple freeze-thaw cycle groups including a non-freeze-thaw control group, and perform freeze-thaw cycles according to preset freezing and thawing conditions; After the freeze-thaw cycle is completed, the sample is dried to constant mass and then subjected to conductive treatment. Multiple effective fields of view of each sample are acquired under the same imaging parameters. The three-dimensional porosity was determined based on the sample's mass, volume, and soil particle specific gravity, and the sample identification, freeze-thaw cycle count, SEM image, and three-dimensional porosity were correlated accordingly.

3. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S2 includes: The pixel size is determined by the ratio of the actual length represented by the ruler to the number of pixels corresponding to the ruler. The ruler and text areas in the SEM image are then cropped, and the cropped SEM image is converted into an 8-bit grayscale image. A baseline pixel size is determined based on the pixel size of each SEM image, and the 8-bit grayscale image is resampled based on the difference between the pixel size and the baseline pixel size to achieve scale normalization. Gray-scale enhancement is achieved by using local histogram equalization with limited contrast, followed by edge-preserving noise reduction using median filtering and bilateral filtering in sequence. Based on the pixel size, the actual length and area parameters used in subsequent image processing are converted into the corresponding pixel scale.

4. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S3 includes: The global baseline threshold is determined based on the global grayscale information of the standardized grayscale image using the maximum inter-class variance method. Multiple global candidate thresholds are set around the global baseline threshold, and local thresholds are determined based on the local grayscale information within the sliding window. Each global candidate threshold and the local threshold are fused according to a preset fusion weight to form a fusion threshold map corresponding to each global candidate threshold. For each fusion threshold map, the pixel gray value of the standardized grayscale image is compared with the fusion threshold at the corresponding position pixel by pixel. Pixels with a pixel gray value lower than the fusion threshold at the corresponding position are marked as pore phase, and pixels with a pixel gray value not lower than the fusion threshold at the corresponding position are marked as granular phase, generating multiple candidate threshold segmentation maps. For each candidate threshold segmentation map, the two-dimensional pore area ratio is calculated, the particle boundary is extracted, and the boundary non-closure degree is determined according to the ratio of the endpoint pixels of the particle boundary to the total number of particle boundary pixels. The particle phase is labeled with eight neighboring connected components to obtain the number of solid connected components, and the normalized rate of change of the number of solid connected components relative to the adjacent candidate threshold segmentation maps is calculated. The difference between the two-dimensional pore area ratio and the corresponding three-dimensional porosity, the boundary non-closure degree, and the normalized rate of change are processed with the same dimensions and weighted to form a joint evaluation value. The candidate threshold segmentation map with the smallest joint evaluation value is determined as the initial segmentation map.

5. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S4 includes: The particle phases in the initial segmentation diagram are marked with connected components to obtain solidly connected components, and the area and convex hull area of ​​each solidly connected component are calculated. A fixedly connected component whose area exceeds a preset multiple of the median area of ​​all fixedly connected components, or whose area of ​​the fixedly connected component and its convex hull area are in proportion to the convex hull area exceeding a preset concavity threshold, is identified as the adhesive particle component. The Euclidean distance transformation is performed on the adhesive particle component to obtain a distance transformation map. The local maxima in the distance transformation map are used as foreground markers. The median of the equivalent diameter of the particles is determined based on the fixed connected components in the initial segmentation map. The minimum spacing between adjacent foreground markers is determined based on the median of the equivalent diameter of the particles, the pixel size, and the preset scale coefficient. Marker-controlled watershed segmentation is then performed. The preset multiple, preset concavity threshold, and preset scale coefficient are predetermined based on the calibration sample.

6. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 5, characterized in that, Step S4 also includes: Sub-regions whose area obtained by dividing the watershed by the marker control is smaller than the preset minimum retention area will be merged into the adjacent sub-region with the longest shared boundary. A gradient merging threshold is determined based on the average gray-level gradient distribution of the shared boundary between adjacent sub-regions generated by watershed segmentation. Adjacent sub-regions whose average gray-level gradient of the shared boundary is less than the gradient merging threshold are then merged. The radius of the disk structure element is determined based on the ratio of the median of the equivalent diameter of the particles to the pixel size. The disk structure element is then used to perform morphological opening and closing operations on each merged particle region. After determining the contour fitting tolerance based on the pixel size, contour polygon fitting is performed to obtain the particle and pore boundary map.

7. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S5 includes: The diameter of a circle with the same area as the particle or pore is defined as the equivalent diameter of the corresponding particle or pore. The particle morphology irregularity is characterized by the ratio of the square of the particle perimeter to the product of four times pi and the particle area. The box counting method was used to determine the pore fractal dimension, which characterizes the pore structure, based on the number of grids covering the pore boundaries at different grid scales. The projected solid area ratio, which characterizes the proportion of solid phase, is determined based on the ratio of the difference between the effective field of view area and the porosity area to the effective field of view area. The particle orientation angle is determined based on the principal eigenvector of the particle pixel coordinate covariance matrix, and the standard deviation of the particle orientation angle, which characterizes particle orientation, is calculated.

8. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, In step S5: Taking the pore phase in the particle and pore boundary map as the analysis object, distance transformation index, connected region index, multi-scale morphological index and local neighborhood index are determined respectively. The distance transformation index is used to characterize the width distribution inside the pore, the connected region index is used to characterize the degree of connection and dispersion of the eight-neighbor pore connected region, the multi-scale morphological index is used to characterize the degree of preservation of pore structure at different morphological scales, and the local neighborhood index is used to characterize the degree of local aggregation of pore pixels in the Gaussian weighted neighborhood. The two-dimensional pore area ratio is determined based on the pore phase in the particle and pore boundary diagram, and the fusion weights of the distance transformation index, connected region index, multi-scale morphology index and local neighborhood index are determined based on the interval to which the two-dimensional pore area ratio belongs. All fusion weights are non-negative and the sum of the weights is 1. When the two-dimensional pore area ratio is less than a preset pore area ratio threshold, the fusion weight of the local neighborhood index is made higher than the fusion weight of the distance transformation index, the connected region index, and the multi-scale morphology index respectively; when the two-dimensional pore area ratio is not less than the preset pore area ratio threshold, the fusion weight of the distance transformation index and the connected region index is made higher than the fusion weight of the multi-scale morphology index and the local neighborhood index respectively. The distance transformation index, connected region index, multi-scale morphology index, and local neighborhood index are weighted and fused according to the fusion weight to obtain the porosity connectivity.

9. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 1, characterized in that, Step S6 includes: The microscopic parameter vector is composed of particle morphological irregularity, pore fractal dimension, projected solid area ratio, standard deviation of particle orientation angle, and pore connectivity. A calibration sample set is constructed, which consists of glacial till samples with known freeze-thaw cycles, microscopic parameter vectors, cohesion, and internal friction angle; Using the average vector of micro-parameters of the zero-freeze-thaw sample as the reference vector, the mean and standard deviation of each dimension of the micro-parameters in the calibration sample set are used to standardize the micro-parameter vector and the reference vector on the same scale. Using the least squares method with regularization terms, and taking the cohesion and internal friction angle obtained from direct shear tests under different freeze-thaw cycles as calibration data, a strength parameter mapping model was established, consisting of a decay term of freeze-thaw cycle number and a correction term for the change of the standardized micro-parameter vector relative to the standardized reference vector.

10. The SEM quantitative analysis method for the microstructure of glacial till after freeze-thaw cycles according to claim 9, characterized in that, Step S6 also includes: The cohesion and internal friction angle output by the strength parameter mapping model are used as strength parameters to input into the simplified Bishop slice model. The simplified Bishop slice model also receives slope cross-sectional geometric parameters, soil strip weight, slip surface dip angle, and pore water pressure. The safety factor is iteratively calculated until the preset convergence condition is met. The safety factor is then compared with the pre-set grading threshold based on the slope grade and applicable engineering standards to determine the warning level. Generate an early warning data packet that includes at least intensity parameters, safety factor, and early warning level, and send the early warning data packet to the early warning terminal; The early warning terminal performs at least one of the following operations based on the early warning level: displaying the result, recording and storing it, providing audible and visual alerts, or sending an alarm message to the engineering monitoring platform.