A method for detecting railway waste slag particles

Abstract

The present invention discloses a detection method for railway waste slag particles, which relates to the field of geological disaster identification technology, and includes the following steps: S1: using a drone to photograph and measure the slag field to be tested to obtain field data; S2: performing internal processing on the field data, and establishing a three-dimensional structural model of slag accumulation particles; S3: exporting point cloud data of the three-dimensional structural model of slag accumulation particles; S4: using the point cloud data to calculate the point cloud normal vector; S5: moving the point cloud data along the point cloud normal vector by a set value, performing point cloud data separation, and obtaining separated point cloud data; S6: performing slag particle identification on the separated point cloud data; S7: extracting the morphological data of the identified slag particles to complete the detection. The present invention collects high-definition images of waste slag accumulation particles by drone photogrammetry, establishes a three-dimensional structural model of slag accumulation particles by aerial triangulation, and quickly obtains information such as the number, radius, uniformity coefficient, and curvature coefficient of slag particles.

Description

Technical Field

[0001] The invention relates to the technical field of geological disaster identification, and in particular to a method for detecting railway waste slag particles. Background Art

[0002] In the process of railway construction, it is inevitable that many waste dumps will be generated. If a disaster occurs in the waste dump, it will pose a huge threat to the existing structures and people's lives and property downstream, so it is necessary to study the stability of the waste dump.

[0003] A waste dump is essentially a loose accumulation formed by excavation and accumulation, which is different from a natural soil slope. During the accumulation of the slag, slag particles of different sizes fall randomly under the action of gravity, and the final accumulation has uneven particle size and unstable internal structure. Based on the geometric and distribution characteristics of the slag particles, the physical, mechanical, and permeability characteristics of the slag can be obtained. Statistical analysis of the geometric and distribution characteristics of the particles in the waste dump is of great significance for understanding the physical, mechanical, and permeability characteristics of the slag, and can also provide scientific parameters for stability analysis, prevention and reinforcement of the waste dump.

[0004] At present, the commonly used method for particle analysis is screening, but the screening method is usually applicable to particles with a diameter of less than 2 cm, and is not applicable to larger slag particles. In addition, the area of ​​the abandoned slag field is large, and the efficiency of using the screening method for analysis is low, which is not conducive to the advancement of the project. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for detecting railway slag particles which is intelligent, fast, and has a wide coverage of measuring point areas.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A method for detecting railway slag particles is provided, which comprises the following steps:

[0008] S1: Use a drone to photograph and measure the slag field to obtain field data;

[0009] S2: Process the external data internally and establish a three-dimensional structural model of the slag accumulation particles;

[0010] S3: Export the point cloud data of the three-dimensional structure model of slag accumulation particles;

[0011] S4: Use the point cloud data to calculate the point cloud normal vector;

[0012] S5: moving the point cloud data along the point cloud normal vector by a set value to separate the point cloud data and obtain separated point cloud data;

[0013] S6: Identify slag particles based on the separated point cloud data;

[0014] S7: extract the morphological data of the identified slag particles and complete the detection.

[0015] Furthermore, before the drone takes pictures, a drone control network is set up on the ground, and image coordinates are established through the drone control network; the drone control network includes a number of drone ground control points;

[0016] When taking photos with a drone, the heading overlap rate of the photos is 80%, the lateral overlap rate is 60%, and the altitude of the drone route is Among them, f is the focal length of the drone lens, a is the pixel size, and GSD is the ground resolution of the aerial image.

[0017] By setting up several ground control points to form a UAV control network before shooting, on the one hand, the control of the UAV can be strengthened so that the UAV can fly according to the set route, so that the field data obtained from the shooting can meet the subsequent requirements; on the other hand, the ground control points can also provide coordinates for the captured images, thereby helping to obtain high-precision slag dump image data.

[0018] Furthermore, in-house processing includes data preprocessing, aerial triangulation, image matching, and image fusion;

[0019] Data preprocessing includes correcting the distortion of images captured in the field data; using the image POS data in the field data to create thumbnails of flight strip images;

[0020] Aerial triangulation includes building a route model through overlapping aerial photographs in field data to determine the exterior orientation elements of all aerial photographs in the shooting area.

[0021] Image matching includes matching and identifying the same-name points between two or more images, and establishing the spatial coordinate relationship between the images through the identified same-name points;

[0022] Image fusion involves seamlessly fusing the overlapping areas of all captured images and generating the final stitched image.

[0023] Furthermore, in step S3, before exporting the point cloud data, the point cloud data is refined and screened to remove noise caused by vegetation or mapping errors; the refined and screened point cloud data is brought into step S4.

[0024] Screening out noise caused by vegetation or mapping errors can improve the accuracy of point cloud data.

[0025] Further, step S4 includes the following steps:

[0026] S401: Select m adjacent points P in the point cloud data i =[X 1i , X 2i , X 3i ] T , and establish the neighborhood point set matrix:

[0027]

[0028] S402: Calculate the centroid of the domain point set

[0029] S403: Subtract the centroid point K from each point in the neighborhood point set, and re-establish the updated neighborhood point set matrix:

[0030]

[0031] S404: Calculate the correlation covariance matrix of the neighborhood point set The principal component analysis method is used to calculate the minimum eigenvalue of the correlation covariance matrix S;

[0032] S405: Obtain the eigenvalues ​​of points in the neighborhood point set by calculating the characteristic equation of the following covariance matrix S;

[0033] det(λE-S)m=0;

[0034] det represents the determinant, λ is the eigenvalue, E is the unit vector, and S is the covariance matrix;

[0035] Perform eigendecomposition on S, S is a symmetric positive definite matrix, and three real eigenvalues ​​λ0, λ1, λ3 are calculated by the characteristic equation, and λ0≤λ1≤λ3. The eigenvectors corresponding to the three real eigenvalues ​​are v0, v1, v2, and v0≤v1≤v2. The eigenvector v0 corresponding to the minimum eigenvalue λ0 is the point cloud normal vector.

[0036] Furthermore, in step S6, slag particle identification includes the following steps:

[0037] S601: Aggregate the separated point cloud data into each local neighborhood, and use a hill climbing algorithm on the local neighborhood to obtain the peak point of each local neighborhood;

[0038] S602: Bring the peak point of each local neighborhood into the regional growing algorithm to identify slag particles.

[0039] Furthermore, the morphological data of the slag particles include: the number of particles, the particle radius, the uniformity coefficient of the particles and the curvature coefficient of the particles;

[0040] The calculation formula of the particle uniformity coefficient Cu is:

[0041]

[0042] Wherein, D60 is the particle size corresponding to the cumulative particle size distribution percentage reaching 60%; D10 is the particle size corresponding to the cumulative particle size distribution percentage reaching 10%;

[0043] The calculation formula of the curvature coefficient Cc of the particle is:

[0044]

[0045] Wherein, D30 is the particle size corresponding to the cumulative particle size distribution percentage reaching 30%.

[0046] The beneficial effects of the present invention are:

[0047] The present invention uses drone photogrammetry to collect high-definition images of waste slag accumulation particles, establishes a three-dimensional structural model of slag accumulation particles through aerial triangulation, calculates the point cloud normal vector, and moves the point cloud in a short distance along the normal vector direction, and reconstructs the separated point cloud particles through the hill climbing algorithm and the regional growth algorithm, thereby quickly obtaining the number, radius, uniformity coefficient, curvature coefficient and other information of the slag particles. Compared with the traditional method, this method is more intelligent and faster, has a wide coverage of the measurement point area, is more representative, and provides scientific parameter basis for the study of the physical and mechanical properties of railway waste slag particles and the stability analysis of the waste slag field. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the present invention;

[0049] Figure 2 Schematic diagram of screening captured images for real scene modeling software;

[0050] Figure 3 This is a schematic diagram of slag particle identification;

[0051] Figure 4 Schematic diagram of particle size curve. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0053] like Figure 1 As shown, a method for detecting railway slag particles comprises the following steps:

[0054] S1: Use a drone to photograph and measure the slag field to obtain field data;

[0055] Before drone photography, collect information and data related to the research area, understand the terrain, landforms, climate conditions, facilities, etc. of the operation area, and formulate a detailed project implementation plan. Specifically, determine the take-off and landing points, design the route according to the aerial camera parameters, flight altitude, aerial scale, and aerial resolution requirements; set up a drone control network on the ground, and establish image coordinates through the drone control network; the drone control network includes several drone ground control points;

[0056] When taking photos with a drone, the heading overlap rate of the photos is 80%, the lateral overlap rate is 60%, and the altitude of the drone route is Among them, f is the focal length of the drone lens, a is the pixel size, and GSD is the ground resolution of the aerial image.

[0057] S2: Process the external data internally and establish a three-dimensional structural model of the slag accumulation particles;

[0058] The real-scene modeling software ContextCapture is used to process the external data internally and establish a three-dimensional structural model of the slag accumulation particles. The real-scene modeling software ContextCapture performs three-dimensional modeling by using an aerial triangulation algorithm. A large number of connection points are calculated through aerial triangulation encryption, and an irregular triangulated network is constructed using the connection points. The white film of the three-dimensional model and the texture information on the surface are automatically mapped to obtain a high-resolution real-scene three-dimensional model with real and natural textures. It has the advantages of fast computing speed and large processing capacity. Before performing three-dimensional modeling, it is necessary to perform simple manual processing on the collected aerial photographs, and manually remove unqualified photos, such as those with shooting ghosting, severe exposure, and images with unreasonable composition, as well as photos of drone takeoff and landing. Appropriately delete photos with high similarity or duplication, and finally select the following. Figure 2 Carry out modeling operations according to the pictures shown that meet the requirements.

[0059] In-house processing includes data preprocessing, aerial triangulation, image matching, and image fusion;

[0060] Data preprocessing includes correcting the distortion of images captured in the field data; using the image POS data in the field data to create thumbnails of flight strip images;

[0061] Aerial triangulation includes building a route model through overlapping aerial photographs in field data to determine the exterior orientation elements of all aerial photographs in the shooting area.

[0062] Image matching includes matching and identifying the same-name points between two or more images, and establishing the spatial coordinate relationship between the images through the identified same-name points;

[0063] Image fusion involves seamlessly fusing the overlapping areas of all captured images and generating the final stitched image.

[0064] The point cloud encryption parameters can be set in the generated 3D structure model of slag accumulation particles. The point cloud encryption parameters include image scale and point cloud density. Select "original size" for image scale and "optimal" for point cloud density.

[0065] S3: exporting the point cloud data of the three-dimensional structure model of the slag accumulation particles; before exporting the point cloud data, the point cloud data is refined and screened, and the point cloud within the slag particle accumulation range of interest is intercepted using the point cloud registration software Cloucompare, and the noise caused by vegetation or mapping errors is removed; the refined and screened point cloud data is brought into step S4.

[0066] S4: Use the point cloud data to calculate the point cloud normal vector;

[0067] S401: Select m adjacent points P in the point cloud data i =[X 1i , X 2i , X 3i ] T , and establish the neighborhood point set matrix:

[0068]

[0069] S402: Calculate the centroid of the domain point set

[0070] S403: Subtract the centroid point K from each point in the neighborhood point set, and re-establish the updated neighborhood point set matrix:

[0071]

[0072] S404: Calculate the correlation covariance matrix of the neighborhood point set The principal component analysis method is used to calculate the minimum eigenvalue of the correlation covariance matrix S;

[0073] S405: Obtain the eigenvalues ​​of points in the neighborhood point set by calculating the characteristic equation of the following covariance matrix S;

[0074] det(λE-S)m=0;

[0075] det represents the determinant, λ is the eigenvalue, E is the unit vector, and S is the covariance matrix;

[0076] Perform eigendecomposition on S, S is a symmetric positive definite matrix, and three real eigenvalues ​​λ0, λ1, λ3 are calculated by the characteristic equation, and λ0≤λ1≤λ3. The eigenvectors corresponding to the three real eigenvalues ​​are v0, v1, v2, and v0≤v1≤v2. The eigenvector v0 corresponding to the minimum eigenvalue λ0 is the point cloud normal vector.

[0077] S5: Move the point cloud data along the point cloud normal vector by a set value to separate the point cloud data and obtain separated point cloud data; move the point cloud along the opposite direction of the point cloud normal vector, and after completing the movement of each point, the point cloud data is separated. The moving distance of the point determines the shape and size of the particles. A short distance will hinder the accurate recognition between blocks, while a long distance will change the shape and size of the particles. Therefore, it is necessary to select the appropriate number of adjacent point clouds and the moving distance.

[0078] S6: Identify slag particles based on the separated point cloud data;

[0079] S601: Aggregate the separated point cloud data into each local neighborhood, and use a hill climbing algorithm on the local neighborhood to obtain the peak point of each local neighborhood;

[0080] S602: Bring the peak point of each local neighborhood into the region growing algorithm and identify the following Figure 3 Schematic diagram of slag particle identification is shown.

[0081] Furthermore, the morphological data of the slag particles include: the number of particles, the particle radius, the uniformity coefficient of the particles and the curvature coefficient of the particles;

[0082] The calculation formula of the particle uniformity coefficient Cu is:

[0083]

[0084] Wherein, D60 is the particle size corresponding to the cumulative particle size distribution percentage reaching 60%; D10 is the particle size corresponding to the cumulative particle size distribution percentage reaching 10%;

[0085] The calculation formula of the curvature coefficient Cc of the particle is:

[0086]

[0087] Wherein, D30 is the particle size corresponding to the cumulative particle size distribution percentage reaching 30%.

[0088] S7: Extract the morphological data of the identified slag particles and complete the detection. The data table of particle number, particle radius (cm), uniformity coefficient (Cu), curvature coefficient (Cc) and Figure 4 The particle size curve is shown.

[0089] Table 1 Slag field particle morphology data table

[0090] parameter Numeric Number of particles 2749 Particle radius (cm) 45.59 Uniformity coefficient (Cu) 1.26 Curvature coefficient (Cc) 0.96

Claims

1. A method for detecting railway waste slag particles, characterized in that: The steps include: S1: Use a drone to photograph and measure the slag field to obtain field data; S2: Process the external data internally and establish a three-dimensional structural model of the slag accumulation particles; The internal processing includes data preprocessing, aerial triangulation, image matching, and image fusion; The data preprocessing includes performing distortion correction on the images captured in the field data; using the image POS data in the field data to create a thumbnail of the flight strip image; The aerial triangulation includes establishing a route model through overlapping aerial photographs in the field data to determine the exterior orientation elements of all aerial photographs in the shooting area; The image matching includes matching and identifying the same-name points between two or more images, and establishing the spatial coordinate relationship between the images through the identified same-name points; The image fusion includes seamlessly fusing the overlapping areas of all captured images to generate a final stitched image; S3: Export the point cloud data of the three-dimensional structure model of slag accumulation particles; S4: Use the point cloud data to calculate the point cloud normal vector; specifically: select m adjacent points P in the point cloud data i =[X 1i , X 2i , X 3i ] T , and establish the neighborhood point set matrix; The correlation covariance matrix of the neighborhood point set is calculated, and the minimum eigenvalue of the correlation covariance matrix S is calculated using the principal component analysis method; the eigenvector v0 corresponding to the minimum eigenvalue λ0 is the point cloud normal vector; S5: moving the point cloud data along the point cloud normal vector by a set value to separate the point cloud data and obtain separated point cloud data; S6: Perform slag particle identification on the separated point cloud data; slag particle identification includes the following steps: S601: Aggregate the separated point cloud data into each local neighborhood, and use a hill climbing algorithm on the local neighborhood to obtain the peak point of each local neighborhood; S602: Bring the peak point of each local neighborhood into the regional growing algorithm to identify slag particles; S7: extracting the morphological data of the identified slag particles to complete the detection; the morphological data includes: the number of particles, the particle radius, the uniformity coefficient of the particles and the curvature coefficient of the particles.

2. The method for detecting railway waste slag particles according to claim 1, characterized in that: Before the drone takes pictures, a drone control network is set up on the ground, and the image coordinates are established through the drone control network; the drone control network includes a number of drone ground control points; when the drone takes pictures, the heading overlap rate of the pictures taken is 80%, the lateral overlap rate is 60%, and the altitude of the drone route is Among them, f is the focal length of the drone lens, a is the pixel size, and GSD is the ground resolution of the aerial image.

3. The method for detecting railway waste slag particles according to claim 1, characterized in that: In the step S3, before exporting the point cloud data, the point cloud data is refined and screened to remove noise caused by vegetation or mapping errors; the refined and screened point cloud data is brought into step S4.

4. The method for detecting railway waste slag particles according to claim 1, characterized in that: The step S4 comprises the following steps: S401: Select m adjacent points P in the point cloud data i =[X 1i , X 2i , X 3i ] T , and establish the neighborhood point set matrix: S402: Calculate the centroid of the domain point set S403: Subtract the centroid point K from each point in the neighborhood point set, and re-establish the updated neighborhood point set matrix: S404: Calculate the correlation covariance matrix of the neighborhood point set The principal component analysis method is used to calculate the minimum eigenvalue of the correlation covariance matrix S; S405: The eigenvalues ​​of the points in the neighborhood point set are calculated by the following characteristic equation of the covariance matrix S: det(λE-S)m=0; det represents the determinant, λ is the eigenvalue, E is the unit vector, and S is the covariance matrix; Perform eigendecomposition on S, S is a symmetric positive definite matrix, and three real eigenvalues ​​λ0, λ1, λ3 are calculated by the characteristic equation, and λ0≤λ1≤λ3. The eigenvectors corresponding to the three real eigenvalues ​​are v0, v1, v2, and v0≤v1≤v2. The eigenvector v0 corresponding to the minimum eigenvalue λ0 is the point cloud normal vector.

5. The method for detecting railway waste slag particles according to claim 1, characterized in that: The calculation formula of the uniformity coefficient Cu of the particles is: Wherein, D60 is the particle size corresponding to the cumulative particle size distribution percentage reaching 60%; D10 is the particle size corresponding to the cumulative particle size distribution percentage reaching 10%; The calculation formula of the curvature coefficient Cc of the particle is: Wherein, D30 is the particle size corresponding to the cumulative particle size distribution percentage reaching 30%.

Images