Visual estimation method for three-dimensional size of rockfill material

Abstract

The invention belongs to the field of rockfill material grading detection, and relates to a visual estimation method for the three-dimensional size of a rockfill material, which comprises the following steps: 1) acquiring an image of a mixed grading rockfill material, and carrying out two-dimensional contour segmentation on rockfill material particles on the image of the mixed grading rockfill material; 2) extracting an estimated value of the three-dimensional size of the rockfill material particles based on a result obtained in the step 1); and 3) estimating the three-dimensional size of the rockfill material based on the estimated value of the three-dimensional size of the rockfill material particles obtained in the step 2). The invention provides the visual estimation method for the three-dimensional size of the stone for grading analysis, which is high in identification accuracy and capable of effectively improving the detection efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of aggregate gradation detection of rockfill materials, and relates to a visual estimation method for three-dimensional dimensions of rockfill materials, and particularly to a visual estimation method for three-dimensional dimensions of rockfill materials for gradation analysis based on computer vision and deep learning technologies. Background Art

[0002] The gradation of aggregate (the weight percentage of aggregates in different particle size ranges in the overall aggregate) is a key index for the quality control of building material production. The gradation of rockfill materials is an important link for the quality control of rockfill dam filling. Traditional methods for detecting the gradation of rockfill materials mainly rely on sieving tests. This method usually requires a lot of manual operations, not only with low detection efficiency and high cost, but also with limited representativeness of the detection results, making it difficult to meet the requirements of rapid and refined construction of projects.

[0003] With the rapid development of information technology, computer vision technology has gradually been introduced into the field of rockfill material gradation detection. Detection methods based on two-dimensional visual perception have become a research hotspot due to their advantages of high efficiency and low cost. These methods infer the particle size and gradation by analyzing the two-dimensional images of rockfill materials, reducing the limitations of traditional methods to a certain extent. However, the conversion from the two-dimensional image of the aggregate to the three-dimensional dimension is still a key technical difficulty. In current research, the common method is to use ellipse fitting technology to process the two-dimensional contour of the aggregate, calculate the equivalent particle size, and combine the ellipsoid volume formula to deduce the volume and gradation of the particle. Although initial results have been achieved in improving detection efficiency and realizing automation, on the one hand, it fails to effectively solve the problem of loss of contour information caused by particle stacking and occlusion during the imaging process of rockfill materials; on the other hand, it lacks adaptability to the irregular shape of aggregate particles and their stacking postures in three-dimensional space, resulting in limited accuracy of gradation characterization.

[0004] Therefore, how to overcome the influence of stacking angle and stacking occlusion on particle size estimation has become the core problem to be solved urgently in the improvement of the two-dimensional image-based rockfill material gradation detection technology. Summary of the Invention

[0005] In order to solve the above technical problems existing in the background art, the present invention provides a visual estimation method for three-dimensional dimensions of aggregate for gradation analysis with high recognition accuracy and capable of effectively improving detection efficiency.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A visual estimation method for three-dimensional dimensions of rockfill materials, characterized in that: the visual estimation method for three-dimensional dimensions of rockfill materials includes the following steps:

[0008] 1) Obtain an image of the mixed-graded rockfill material and perform two-dimensional contour segmentation of the rockfill material particles on the image of the mixed-graded rockfill material;

[0009] 2) Based on the result obtained in step 1), extract the estimated values of the three-dimensional dimensions of the rockfill material particles;

[0010] 3) Estimate the three-dimensional dimensions of the rockfill material based on the estimated values of the three-dimensional dimensions of the rockfill material particles obtained in step 2).

[0011] Preferably, the specific implementation manner of step 1) is:

[0012] 1.1) Collect image samples of the mixed-graded rockfill material under different imaging conditions;

[0013] 1.2) Train the image samples obtained in step 1.1) based on YOLOv8m-seg to obtain a two-dimensional contour segmentation model of the rockfill material particles;

[0014] 1.3) Segment the two-dimensional contour of the rockfill material particles based on the two-dimensional contour segmentation model of the rockfill material particles obtained in step 1.2).

[0015] Preferably, the specific implementation manner of step 2) is:

[0016] 2.1) According to the two-dimensional contour of the rockfill material particles segmented in step 1), use the convex hull algorithm to extract the key points of the two-dimensional contour of the rockfill material particles;

[0017] 2.2) Fit the result obtained in step 2.1) into an equivalent ellipse, approximate the particle as an ellipsoid, the equivalent ellipse is the projection of the approximate ellipsoid of the particle on the plane, and obtain the equation of the projection relationship;

[0018] 2.3) Solve the equation of the projection relationship obtained in step 2.2) to obtain the estimated values of the three-dimensional dimensions of the rockfill material particles.

[0019] Preferably, the specific implementation manner of step 2.2) is:

[0020] The equation of the ellipsoid is:

[0021]

[0022] The equation of the equivalent ellipse is:

[0023] a 11 X 2 +2a 12 XY+a 22 Y 2 +a 33 =0(2)

[0024] Where:

[0025]

[0026] wherein:

[0027] θ and are respectively the azimuth angle and the pitch angle between the projection plane and the principal plane of the ellipsoid;

[0028] a, b, and c are respectively the three semi-major axes of the ellipsoid;

[0029] The equation for the projection relationship between the major axis A, minor axis B of the ellipse with formula (2) and the three semi-major axes a, b, c of the ellipsoid is:

[0030]

[0031] wherein:

[0032] I = a 11 + a 22 ;

[0033]

[0034] Preferably, the specific implementation manner of step 2.3) is:

[0035] Introduce the aspect ratio λ 1 , thickness-to-width ratio λ 2 and the probability distributions of the stacking azimuth angle θ and pitch angle :

[0036] λ 1 = b / a, λ 1 ~ P 1 ; λ 2 = c / b, λ 2 ~ P 2 ; θ ~ P 3 ;

[0037] Use the Monte Carlo method to solve the feasible set of particle three-dimensional sizes, and use the expectation of the solution set as the representative solution. The conditions for the feasible solution are:

[0038]

[0039] wherein:

[0040] n is the particle number of the rockfill;

[0041] ε A and ε B are respectively the error thresholds of the major and minor axes;

[0042] A *is the dimension of the major semi-axis of the equivalent ellipse;

[0043] B * is the dimension of the minor semi-axis of the equivalent ellipse;

[0044] According to the feasible solution space, the estimated values of the three semi-axes a, b, and c of the ellipse are obtained respectively; the expressions of the estimated values of the three semi-axes of the ellipse are respectively:

[0045]

[0046] where N is the number of feasible solutions;

[0047] is the estimated value of the semi-axis length a;

[0048] is the estimated value of the semi-axis length b;

[0049] is the estimated value of the semi-axis length c.

[0050] Preferably, in the step 2.3), the aspect ratio λ 1 and the thickness-to-width ratio λ 2 of the particle shape characteristic index, as well as the stacking azimuth angle θ and the pitch angle The specific implementation method of the probability distribution is:

[0051] Obtain the aspect ratio λ 1 and the thickness-to-width ratio λ 2 of the particle shape characteristic index. The steps of the probability distribution are as follows:

[0052] Randomly collect samples of rockfill materials, measure the sizes of rockfill particles in each particle size range, and obtain the aspect ratio λ 1 and the thickness-to-width ratio λ 2 of the particle shape characteristic index, and respectively fit the probability distribution models P 1 and P 2 ;

[0053] Obtain the probability distribution of the stacking azimuth angle θ and the pitch angle of the rockfill particles. The steps are as follows:

[0054] a.1) Prepare a sample of stone materials arranged in a dispersed manner, take photos for sampling, and obtain the grading GIR through standard screening;

[0055] a.2) Use the two-dimensional contour segmentation model of rockfill particles to obtain the two-dimensional contours of the stone material particles in step a.1), and extract the lengths of the major and minor semi-axes A * and B * ;

[0056] a.3) Randomly initialize the parameters of the azimuth angle θ and elevation angle probability distribution model: θ ~ U(θ min , θ max ), The upper and lower limit value ranges are both [0, π / 2]; the U is a uniform distribution;

[0057] a.4) Based on the equivalent major and minor axis length values A * and B * of the ellipse, the aspect ratio λ 1 ~P 1 and the thickness-to-width ratio λ 2 ~P 2 as well as calculate the gradation GIR * of the stone sample prepared in step a.1);

[0058] a.5) Take the absolute error between the gradation GIR * obtained in step a.4) and the gradation GIR in step a.1) as the optimization objective, and adjust the parameters of the azimuth angle θ and elevation angle probability distribution model through the grid optimization method until the absolute error is as small as possible, then determine that the error converges, and obtain the probability distribution of the azimuth angle θ and elevation angle.

[0059] Preferably, the calculation method of the gradation GIR * of the stone sample in step a.4) is:

[0060]

[0061] Where:

[0062] M is the mass of each particle of the rockfill in the stone sample;

[0063] n represents the total number of rockfill particles in this particle size range;

[0064] N' represents the total number of rockfill particles in (1);

[0065] The expression of the said M is:

[0066]

[0067] Preferably, the specific implementation method of step 3) is:

[0068] The density of the rockfill is ρ, calculate the mass M of each particle of the rockfill and the gradation GIR * :

[0069]

[0070]

[0071] Wherein:

[0072] n represents the total number of rockfill particles within the particle size range;

[0073] N' represents the total number of rockfill particles in (1).

[0074] The advantages of the present invention are as follows:

[0075] The present invention provides a visual estimation method for the three-dimensional dimensions of rockfill, including the following steps: 1) Obtain an image of the mixed-graded rockfill and perform two-dimensional contour segmentation of the rockfill particles on the image of the mixed-graded rockfill; 2) Extract the estimated values of the three-dimensional dimensions of the rockfill particles based on the result obtained in step 1); 3) Estimate the three-dimensional dimensions of the rockfill based on the estimated values of the three-dimensional dimensions of the rockfill particles obtained in step 2). The visual estimation method for the three-dimensional dimensions of the rockfill provided by the present invention, combining statistical analysis, deep learning, and Monte Carlo estimation, can efficiently and accurately estimate the three-dimensional dimensions of rockfill particles, improve the accuracy of gradation characterization, and is applicable to the quality control of rockfill dam filling and other geotechnical engineering fields. In combination with statistical analysis and deep learning models, the present method can efficiently process large-scale rockfill samples, significantly improving the efficiency of gradation analysis; at the same time, through the optimal equivalent ellipse and ellipsoid geometric models, the accuracy of particle three-dimensional dimension estimation is greatly improved. This method uses the Monte Carlo method to consider the randomness of the imaging angle and particle morphology, adapts to complex rockfill conditions, realizes the statistical approximation of particle three-dimensional dimensions, and provides accurate data for gradation analysis and engineering design. The influence of the imaging angle on the recognition result is particularly emphasized, and by adjusting the azimuth angle and pitch angle distribution, the recognition accuracy of irregular particle morphologies is improved, and high efficiency in recognition can be maintained even under complex imaging conditions, featuring high detection efficiency, strong adaptability, and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flowchart of the visual estimation method for the three-dimensional dimensions of the rockfill provided by the present invention.

[0077] Figure 2 is a projection relationship diagram of the particle ellipsoid and its contour ellipse. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0078] Refer to Figure 1 , the present invention provides a visual estimation method for the three-dimensional dimensions of the rockfill, specifically including the following steps:

[0079] 1) Obtain an image of the mixed-graded rockfill and perform two-dimensional contour segmentation of the rockfill particles on the image of the mixed-graded rockfill. Specifically:

[0080] 1.1) Collect image samples of the mixed-graded rockfill under different imaging conditions;

[0081] 1.2) Based on YOLOv8m-seg, train the image samples obtained in step 1.1) to obtain a two-dimensional contour segmentation model for rockfill particles; establish a rockfill instance segmentation model through a convolutional neural network. In the present invention, the YOLOv8m-seg model is used for the contour segmentation of rockfill particles. On this basis, collect images of the mixed gradation rockfill under different imaging conditions, and use the Labelme annotation tool to annotate the closed polygon contours of the image samples. To improve the generalization ability and robustness of the model, perform enhancement processing on the dataset. Specifically, it includes data enhancement techniques such as brightness transformation, contrast enhancement, random rotation, horizontal flipping, and HSV. Divide the dataset into a training set and a validation set in a ratio of 80% to 20%. Before training the deep learning model, load the weights of the YOLOv8m-seg model pre-trained on the COCO dataset for transfer learning. During the model training process, use the validation set to preliminarily verify the training results, and optimize and adjust the model parameters according to the accuracy performance on the validation set.

[0082] 1.3) Based on the two-dimensional contour segmentation model of rockfill particles obtained in step 1.2), segment the two-dimensional contour of rockfill particles. In the present invention, the contours of the particles are extracted by the findContours function. Then, use the convexHull function to calculate the convex hull of the contour to obtain the key feature points. At the same time, in order to achieve the best equivalent ellipse fitting, the fitEllipse function is adopted. This function is based on the least squares principle and can calculate the optimal fitting ellipse for the given contour point set. This ellipse approximates the overall shape of the contour with the highest accuracy, ensuring the accuracy of the fitting result. Through the parameters returned by the fitEllipse function, the center coordinates, rotation angle, and the dimensions of the major and minor axes of the ellipse can be extracted.

[0083] 2) Based on the results obtained in step 1), extract the estimated values of the three-dimensional dimensions of rockfill particles, specifically:

[0084] 2.1) According to the two-dimensional contour of the rockfill particles segmented in step 1), use the convex hull algorithm to extract the key points of the two-dimensional contour of the rockfill particles;

[0085] 2.2) Fit the results obtained in step 2.1) into an equivalent ellipse, approximate the particles as ellipsoids, and the equivalent ellipse is the projection of the approximate ellipsoids of the particles on the plane. As Figure 2 shown, obtain the equation of the projection relationship, specifically:

[0086] The specific implementation method of step 2.2) is:

[0087] The equation of the ellipsoid is:

[0088]

[0089] The equation of the equivalent ellipse is:

[0090] a 11 X 2 +2a 12 XY + a 22 Y 2 +a 33 =0(2)

[0091] Where:

[0092]

[0093] Where:

[0094] θ and are respectively the azimuth angle and the pitch angle between the projection plane and the principal plane of the ellipsoid, θ ∈ [0, π / 2],

[0095] a, b, and c are respectively the three semi-major axes of the ellipsoid;

[0096] For the equation of the projection relationship between the major axis A, minor axis B of the ellipse with equation (2) and the three semi-major axes a, b, c of the ellipsoid is:

[0097]

[0098]

[0099] Where:

[0100] I = a 11 +a 22 ;

[0101]

[0102] 2.3) Solve the equation of the projection relationship obtained in step 2.2) to obtain the estimated values of the three-dimensional dimensions of the rockfill particles. Specifically: The specific implementation method of step 2.3) is:

[0103] Research in the geotechnical field shows that block stones have an obvious directional distribution law, and this distribution law can be represented by a random normal distribution or a uniform distribution. Based on the measured grain shape characteristic distribution model, it is assumed that the central azimuth angle and pitch angle of particle imaging are discrete uniform distributions. Sample the above length-width ratio, width-thickness ratio, central azimuth angle, and pitch angle. Therefore, when solving the projection relationship equation, the grain shape characteristic index width-length ratio λ 1 , thickness-width ratio λ 2 and the stacking azimuth angle θ and pitch angle are introduced for the probability distribution:

[0104] λ 1 = b / a, λ 1 ~P 1 ; λ 2 = c / b, λ 2 ~P 2 ; θ~P 3 ;

[0105] The aspect ratio λ of the particle shape characteristic index of the particles 1 , the thickness-to-width ratio λ 2 and the probability distributions of the stacking azimuth angle θ and the pitch angle of the particles, the aspect ratio λ 1 , the thickness-to-width ratio λ 2 can be directly obtained by on-site sampling measurement. The method is as follows: Place the particles in the way with the largest horizontal projection area, and measure the length, width and thickness of its minimum circumscribed cuboid as the particle shape characteristic parameters; the stacking azimuth angle θ and the pitch angle cannot be directly measured and will not be discussed for the time being; assume that the probability distributions of the stacking azimuth angle θ and the pitch angle are known; exemplarily, the aspect ratio λ of the particle shape characteristic index adopted in the present invention 1 , the thickness-to-width ratio λ 2 and the probability distributions of the stacking azimuth angle θ and the pitch angle are specifically implemented as follows:

[0106] Obtain the aspect ratio λ of the particle shape characteristic index of the particles 1 , the thickness-to-width ratio λ 2 The steps of the probability distribution are as follows:

[0107] Randomly collect samples of rockfill materials, measure the sizes of the rockfill particles in each particle size range, and obtain the aspect ratio λ 1 and the thickness-to-width ratio λ 2 particle shape characteristic indexes, and respectively fit the probability distribution models P 1 and P 2 ;

[0108] Obtain the probability distributions of the stacking azimuth angle θ and the pitch angle of the rockfill particles The steps are as follows:

[0109] a.1) Prepare a sample of the stone material arranged in a dispersed manner, take photos for sampling, and obtain the grading GIR through standard screening;

[0110] a.2) Use the two-dimensional contour segmentation model of the rockfill particles to obtain the two-dimensional contours of the stone material particles in step a.1), and extract the major and minor semi-axis length values A * and B * ;

[0111] a.3) Randomly initialize the parameters of the azimuth angle θ and pitch angle probability distribution model: θ ~ U(θ min , θ max ), The upper and lower limit value ranges are both [0, π / 2]; U is the uniform distribution;

[0112] a.4) Based on the equivalent major and minor axis length values A * and B * obtained in step a.2), aspect ratio λ 1 ~P 1 and thickness-to-width ratio λ 2 ~P 2 as well as perform gradation GIR * calculation on the aggregate sample prepared in step a.1);

[0113] Among them, the calculation method of the gradation GIR * of the aggregate sample is:

[0114]

[0115] Among them:

[0116] M is the mass of each particle of the rockfill in the aggregate sample;

[0117] n represents the total number of rockfill particles in this particle size range;

[0118] N' represents the total number of rockfill particles in (1);

[0119] The expression of M is:

[0120]

[0121] a.5) Take the absolute error between the gradation GIR * obtained in step a.4) and the gradation GIR in step a.1) as the optimization goal, and adjust the parameters of the azimuth angle θ and pitch angle probability distribution model through the grid optimization method until the absolute error is as small as possible, then determine that the error converges, and obtain the azimuth angle θ and pitch angle probability distribution.

[0122] It should be noted that building material research shows that the natural stacking state of block stones has an obvious directional distribution law, and the stacking angle conditions can be represented by a uniform distribution. Therefore, it can be assumed that the azimuth angle θ and pitch angle obey the uniform distribution. Given the distribution form, only the parameters of the distribution model need to be determined.

[0123] Use the Monte Carlo method to solve the feasible set of particle three-dimensional sizes, and take the expectation of the solution set as the representative solution. The conditions for the feasible solution are:

[0124]

[0125] Wherein:

[0126] n is the particle number of the rockfill material;

[0127] ε A and ε B are the error thresholds of the major and minor semi-axes respectively;

[0128] A * is the size of the major semi-axis of the equivalent ellipse;

[0129] B * is the size of the minor semi-axis of the equivalent ellipse;

[0130] Wherein, A * and B * are extracted by determining the conversion relationship between the pixel size and the actual size of the rockfill material through a fixed shooting distance during the shooting stage, so as to determine the actual size of the equivalent ellipse. The function fitEllipse in the opencv library can be used to return the values of the major and minor semi-axes of the ellipse;

[0131] According to the corresponding probability distribution, random sampling is performed on the aspect ratio λ 1 , thickness-to-width ratio λ 2 , stacking azimuth angle θ and pitch angle . At this time, only one of the unknowns a, b, and c exists in the equation system in (1). Assuming it is a, similarly, an initial value of a is set, and a set of A and B can be calculated. Compare the calculated A and B with the measured A * and B * . If the difference between the two exceeds the error threshold, adjust the value of a until it is satisfied (Newton's method); at this time, the value of a that satisfies the error threshold is a feasible solution. Given a, b and c can be further calculated; repeat the above steps multiple times to obtain multiple groups of feasible solutions, which is called a feasible solution space;

[0132] According to the feasible solution space, the estimated values of the three semi-axis lengths a, b, and c of the ellipse are obtained respectively; the expressions of the estimated values of the three semi-axis lengths of the ellipse are respectively:

[0133]

[0134] Wherein, N is the number of feasible solutions;

[0135] is the estimated value of the semi-axis length a;

[0136] is the estimated value of the semi-axis length b;

[0137] It is an estimated value of the semi-major axis length c.

[0138] 3) Estimate the three-dimensional size of the rockfill based on the estimated values of the three-dimensional sizes of the rockfill particles obtained in step 2). Specifically: The density of the rockfill is ρ, calculate the mass M and gradation GIR of each rockfill particle * :

[0139]

[0140] Where:

[0141] n represents the total number of rockfill particles within the particle size range;

[0142] N' represents the total number of rockfill particles in (1).

Claims

1. A method for visually estimating the three-dimensional size of rockfill, characterized by: The method for visually estimating the three-dimensional size of rockfill materials comprises the following steps: 1) acquiring an image of mixed-graded rockfill material, and performing two-dimensional contour segmentation of rockfill material particles on the image of the mixed-graded rockfill material; 2) extracting an estimated value of the three-dimensional size of the rockfill material particles based on the results obtained in step 1); 3) estimating the three-dimensional size of the rockfill material based on the estimated value of the three-dimensional size of the rockfill material particles obtained in step 2).

2. The method for visually estimating the three-dimensional size of rockfill materials according to claim 1, characterized in that: The specific implementation method of step 1) is: 1.1) Collect image samples of mixed graded rockfill under different imaging conditions; 1.2) Training the image samples obtained in step 1.1) based on YOLOv8m-seg to obtain a two-dimensional contour segmentation model of rockfill particles; 1.3) Segmenting the two-dimensional contours of the rockfill material particles based on the two-dimensional contour segmentation model of the rockfill material particles obtained in step 1.2).

3. The method for visually estimating the three-dimensional size of rockfill materials according to claim 2, characterized in that: The specific implementation method of step 2) is: 2.1) according to the two-dimensional contour of the rockfill material particles obtained by segmentation in step 1), a convex hull algorithm is used to extract key points of the two-dimensional contour of the rockfill material particles; 2.2) fitting the result obtained in step 2.1) into an equivalent ellipse, approximating the particle to an ellipsoid, the equivalent ellipse being the projection of the particle approximate ellipsoid on the plane, and obtaining the equation of the projection relationship; 2.3) Solving the equation of the projection relationship obtained in step 2.2) to obtain an estimated value of the three-dimensional size of the rockfill material particles.

4. The method for visually estimating the three-dimensional size of rockfill materials according to claim 3, characterized in that: The specific implementation method of step 2.2) is: The equation of the ellipsoid is: The equation of the equivalent ellipse is: a 11 X 2 +2a 12 XY+a 22 Y 2 +a 33 =0(2) in: in: θ and are the azimuth and elevation angles between the projection plane and the principal plane of the ellipsoid, respectively; a, b, and c are the three semi-major axes of the ellipsoid; The equation for the projection relationship between the ellipse length A, the minor semi-axis B and the three semi-major axes a, b, c of the ellipsoid with formula (2) is: in: I=a 11 +a 22 ; 5. The method for visually estimating the three-dimensional size of rockfill materials according to claim 4, characterized in that: The specific implementation method of step 2.3) is: The particle shape characteristic indicators width-to-length ratio λ1, thickness-to-width ratio λ2, stacking azimuth angle θ and pitch angle are introduced The probability distribution of is: The Monte Carlo method is used to solve the feasible three-dimensional particle size set, and the expectation of the solution set is used as the representative solution. The conditions for the feasible solution are: in: n is the particle number of the rockfill material; ε A and ε B are the error thresholds of the major and minor semi-axes respectively; A * is the dimension of the major semi-axis of the equivalent ellipse; B * is the dimension of the minor semiaxis of the equivalent ellipse; According to the feasible solution space, the estimated values ​​of the three semi-axis lengths a, b, and c of the ellipse are obtained respectively; the expressions of the estimated values ​​of the three semi-axis lengths of the ellipse are respectively: Where N is the number of feasible solutions; is the estimated value of the semi-axis length a; is the estimated value of the semi-axis length b; is an estimate of the semi-axis length c.

6. The method for visually estimating the three-dimensional size of rockfill materials according to claim 5, characterized in that: The particle shape characteristic indicators in step 2.3) include the aspect ratio λ1, the thickness ratio λ2, and the stacking azimuth angle θ and the pitch angle The specific implementation of the probability distribution is: The steps to obtain the probability distribution of particle shape characteristic indicators width-to-length ratio λ1 and thickness-to-width ratio λ2 are as follows: Randomly collect rockfill samples, measure the size of rockfill particles in each particle size range, obtain the particle shape characteristic indexes of width-to-length ratio λ1 and thickness-to-width ratio λ2, and fit the probability distribution models P1 and P2 of the corresponding particle shape characteristic indexes respectively; Obtain the azimuth angle θ and pitch angle of the rockfill particle stack The steps for probability distribution are as follows: a.1) Prepare dispersed stone samples, take photos and obtain the graded GIR through standard sieving; a.2) Use the two-dimensional contour segmentation model of rockfill particles to obtain the two-dimensional contour of the stone sample particles in step a.1) and extract the length value A of the major and minor semi-axis of the equivalent ellipse * and B * ; a.3) Randomly initialize the azimuth angle θ and pitch angle probability distribution model parameters: θ~U(θ min ,θ max ), The upper and lower limits are both in the range of [0,π / 2]; U is uniformly distributed; a.4) Based on the equivalent ellipse major and minor axis length value A obtained in step a.2) * and B * , width-to-length ratio λ1~P1 and thickness-to-width ratio λ2~P2, and GIR grading of the stone sample prepared in step a.1) * calculate; a.5) The graded GIR obtained in step a.4) * The absolute error with the gradation GIR in step a.1) is used as the optimization target, and the probability distribution model parameters of the azimuth angle θ and the pitch angle are adjusted by the grid optimization method until the absolute error is as small as possible. It is determined that the error converges, and the probability distribution of the azimuth angle θ and the pitch angle is obtained.

7. The method for visually estimating the three-dimensional size of rockfill materials according to claim 6, characterized in that: In step a.4), the stone sample is graded GIR * The calculation method is: in: M is the mass of each particle of rockfill material in the rock sample; n represents the total number of rockfill particles within the particle size range; N' represents the total number of rockfill particles in (1); The expression of M is:

8. The method for visually estimating the three-dimensional size of rockfill materials according to claim 7, characterized in that: The specific implementation method of step 3) is: The density of rockfill is ρ, and the mass M and gradation GIR of each particle of rockfill are calculated. * : in: n represents the total number of rockfill particles within the particle size range; N' represents the total number of rockfill particles in (1).

Images