Calculation method for reserves of rockfill yard

Abstract

The invention belongs to the field of rockfill dam stockyard mining process management and control, and relates to a rockfill stockyard reserve calculation method, which comprises the following steps of: 1) shooting a stockyard image by adopting an aerial survey unmanned aerial vehicle, obtaining a stockyard image, and obtaining stockyard point cloud data according to the aerial survey image; 2) establishing a stock ground three-dimensional model according to the point cloud data; 3) constructing a rockfill field blasting area image recognition model; 4) utilizing the rockfill field blasting area image recognition model to recognize a blasted area from the material field image; 5) converting the pixel coordinates of the identified blasting area in the image into earth rectangular coordinates; 6) superposing the space range of the blasting area into the three-dimensional model of the stockyard, and marking the blasting area; and 7) performing three-dimensional model construction on the blasting area in the stockyard three-dimensional model, and calculating the loose volume of the blasting area through rasterization operation after the construction is completed. The rockfill yard reserve calculation method provided by the invention is high in measurement precision and high in automation degree.

Description

Technical Field

[0001] The present invention belongs to the field of rockfill dam material field mining process control, relates to a rockfill material field reserve calculation method, and in particular to a rockfill material field reserve calculation method based on unmanned aerial vehicle aerial photography and image recognition. Background Art

[0002] In the construction of modern rockfill dams, the rockfill material field is the main source of dam filling materials, and its management efficiency and accuracy are directly related to the progress and quality of the entire project. As the mining progress of the material field advances, the blasting area in the rockfill material field will change in position and size. How to quickly and accurately obtain the stone reserves in the blasting area is one of the problems that need to be solved urgently in the rockfill material field. The traditional method of calculating the reserves of the blasting area mainly relies on manual measurement, that is, real-time dynamic carrier phase differential technology (RTK) measurement, but the manual measurement method is inefficient, lacks real-time performance, and is easily affected by the observer's personal experience and subjective judgment, resulting in inaccurate calculation results. In addition, manual observation requires a lot of manpower, material resources and time, is costly, is easily disturbed by external environmental factors, and there are certain personnel safety risks.

[0003] In recent years, with the development of UAV technology and computer vision technology, new ideas have been provided to solve the above problems. UAVs have the advantages of flexibility and wide coverage, and can quickly obtain image data of rockfill sites. Image recognition technology can extract the required data from these images to realize the automatic calculation of the volume of the pile. However, the current method of combining UAV aerial photography with image recognition technology for the calculation of rockfill reserves is still immature, with problems such as insufficient measurement accuracy, poor adaptability, and low degree of automation. Summary of the invention

[0004] In order to solve the above technical problems existing in the background technology, the present invention provides a method for calculating the reserves of a rockfill site with high measurement accuracy and high degree of automation.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for calculating the reserves of a rockfill site, characterized in that the method comprises the following steps:

[0007] 1) Use an aerial survey drone to shoot images of the material yard, obtain images of the material yard, and obtain point cloud data of the material yard based on the aerial survey images;

[0008] 2) Establishing a three-dimensional model of the material yard based on the point cloud data obtained in step 1);

[0009] 3) Construct an image recognition model for the blasting area of ​​a rockfill site;

[0010] 4) using the rockfill blasting area image recognition model obtained in step 3) to identify the blasted area in the material yard image obtained in step 1); the blasted area is the area of ​​all loose rocks within the material yard;

[0011] 5) Convert the pixel coordinates of the identified blasting area in the image into earth rectangular coordinates;

[0012] 6) superimposing the spatial range of the blasting area onto the three-dimensional model of the material field constructed in step 2), and marking the blasting area;

[0013] 7) Construct a three-dimensional model of the blasting area in the three-dimensional model of the material yard, and after the construction is completed, calculate the loose volume of the blasting area through rasterization operation.

[0014] Preferably, the method for acquiring the material yard image in the step 1) is: selecting a drone with GPS and RTK functions, carrying a shooting camera on the drone, and setting the drone along the navigation track of the material yard at a height of 350-450m above the ground to acquire the material yard image; preferably, the resolution of the shooting camera is not less than 5472×3648; the method for acquiring the point cloud data in the step 1) is: stitching, correcting and three-dimensionally reconstructing the material yard image taken by the drone to generate point cloud data.

[0015] Preferably, the step 2) is to use the point cloud data to create an image surface model of the material yard using CloudCompare modeling software, crop the material yard area, and then use CATIA software to generate a three-dimensional model of the material yard.

[0016] Preferably, the specific implementation of step 2) is:

[0017] 2.1) Select the las point cloud data file and import it into CloudCompare software;

[0018] 2.2) Use the Crop function to crop the discharge area;

[0019] 2.3) Adjust the compass parameters to generate the material field grid model;

[0020] 2.4) Smooth the material yard grid model and export it to obtain a three-dimensional model of the material yard.

[0021] Preferably, the specific implementation of step 3) is:

[0022] 3.1) Collect the overall bird’s-eye view images of the rockfill site taken by drones;

[0023] 3.2) performing sample enhancement on the material field images collected in step 3.1), and dividing the images before and after enhancement into a training set and a test set in proportion;

[0024] 3.3) Configuring image recognition model training parameters, including the number of training rounds and batch size;

[0025] 3.4) Evaluate the performance of the model through the test set images. The evaluation indicators include precision P, recall R and average precision AP, which are calculated as follows:

[0026]

[0027] in:

[0028] TTP is the number of correct predictions;

[0029] FFP is the number of false predictions, including identifying targets that are not in the working area as the working area and missing identification;

[0030] FFN is the number of work areas that are mistakenly identified as other categories;

[0031] 3.5) Based on the evaluation result of step 3.4), the image recognition model of the rockfill blasting area is finally determined. The conditions for determining the image recognition model of the rockfill blasting area are that the recall rate R is not less than 85%, and the accuracy rate and the average accuracy rate are both between 90% and 100%.

[0032] Preferably, the step 4) is to use the rockfill material yard blasting area image recognition model constructed in step 3), import the material yard image with the blasting area marked into the image recognition model for training, call the trained image recognition model to identify the blasting area in the material yard image, and obtain a mask of the blasting area.

[0033] Preferably, the specific implementation of step 4) is:

[0034] 4.1) Use labelme software to mark the blasting area in the material field pictures collected by drone;

[0035] 4.2) Import the labeled images into the image recognition model constructed in step 3) for training to obtain the optimal parameters of the image recognition model;

[0036] 4.3) Using the image recognition model with optimal parameters to identify the blasting area in the stockyard image;

[0037] 4.4) Save the identified material field image with the blasting area mask.

[0038] Preferably, the step 5) is to export the pixel coordinate file of the blasting area mask in the stockyard image identified in step 4), and convert all pixel coordinates covered by the mask into model coordinates, which is specifically implemented as follows:

[0039] 5.1) Obtain the internal and external parameters of the drone camera, the internal parameters of the drone camera include focal length (f x , f y ) and the principal point coordinates (c x , c y );The drone camera external parameters include the rotation matrix K and the translation vector t;

[0040] 5.2) Normalize the pixel coordinates of all pixels of the mask-covered material field image according to the internal and external parameters of the drone camera in step 5.1), specifically:

[0041]

[0042] in:

[0043] (u, v) are the pixel coordinates of all pixels in the mask-covered material field image;

[0044] z is the pixel depth information;

[0045] x, y, z are the three-dimensional coordinates in camera coordinates;

[0046] 5.3) Convert the three-dimensional coordinates in the camera coordinate system to the three-dimensional coordinates (X, Y, Z) in the model coordinate system, specifically:

[0047] [X,Y,Z]=K*[x,y,z]+t.

[0048] Preferably, the specific implementation method of step 6) is: using the coordinates converted in step 5) as the object of convex hull fitting, obtaining the boundary points of each area after calculation by convex hull algorithm, outputting its boundary coordinates, and using polygon tools to draw the blasting area in the three-dimensional model of the material yard according to the boundary coordinates, marking the blasting area, and the marking is to add color, texture or label.

[0049] Preferably, the specific implementation of step 7) is:

[0050] 7.1) Select the established 3D model of the blasted area;

[0051] 7.2) Select the rasterization function and set the rasterization parameters, i.e. the grid size;

[0052] 7.3) Performing a rasterization operation to divide the blasted area model into a series of small grids;

[0053] 7.4) Perform volume accumulation on each grid.

[0054] The advantages of the present invention are:

[0055] The present invention provides a method for calculating the reserve of a rockfill material field, comprising: 1) using an aerial survey drone to shoot images of the material field, obtaining images of the material field, and obtaining point cloud data of the material field according to the aerial survey images; 2) establishing a three-dimensional model of the material field according to the point cloud data obtained in step 1); 3) constructing an image recognition model of a blasting area of ​​the rockfill material field; 4) using the image recognition model of the blasting area of ​​the rockfill material field obtained in step 3) to identify the blasted area in the material field image obtained in step 1); the blasted area is the area of ​​all loose stones within the material field; 5) converting the pixel coordinates of the identified blasting area in the image into earth rectangular coordinates; 6) superimposing the spatial range of the blasting area into the three-dimensional model of the material field constructed in step 2), and marking the blasting area; 7) constructing a three-dimensional model of the blasting area in the three-dimensional model of the material field, and calculating the loose volume of the blasting area through a rasterization operation after the construction is completed. The present invention obtains point cloud data of the material field through aerial photography by a drone, and establishes a three-dimensional model of the material field after registering the coordinate system of the aerial photograph and the material field design model. Next, the image recognition model is trained using a training set of sample photos of the material yard with the blasting area marked. Then, a drone is set to take photos of the material yard at regular intervals. The trained recognition model is directly called to identify the blasting area in the material yard photos in real time, and the identified blasting area mask is superimposed on the material yard three-dimensional model. Taking into account the horizontal expansion factor of the blasting area and its free surface, the blasting area model is corrected, and the rockfill reserves in the blasting area are accurately calculated by professional software. The present invention abandons the manual measurement method and uses image recognition of the blasting area of ​​the rockfill material yard to help managers quickly understand the reserves of the rockfill material yard, provide basic information for material yard mining, blasting planning, resource allocation, etc., make the collaborative work between different work teams more efficient, and improve the level of construction management. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flow chart of the method for calculating the reserve of a rockfill site provided by the present invention;

[0057] Figure 2 This is the effect diagram of the image recognition model identifying the material yard image. DETAILED DESCRIPTION

[0058] See also Figure 1 The present invention provides a method for calculating the reserves of a rockfill site, the method comprising the following steps:

[0059] 1) Use a drone with high-precision GPS and RTK functions to shoot images of the material yard, and obtain point cloud data through photogrammetry software; among them, select a drone with high-precision GPS and RTK functions, and carry a high-resolution camera (exemplarily, the resolution of the shooting camera is not less than 5472×3648), set the drone's flight trajectory according to the terrain and area of ​​the material yard, and follow the material yard's flight trajectory to ensure that the entire material yard area can be covered. The flight trajectory should be a series of parallel scanning paths to avoid omissions or repeated coverage. According to the terrain conditions of the material yard, set the height above the ground to 350-450m, preferably 400m above the ground, obtain a suitable shooting angle and resolution, and scan at an interval of 20Hz, and collect once a day to ensure that changes in the material yard are reflected in a timely manner. Use the photogrammetry function in the software to stitch, correct and reconstruct aerial images in three dimensions to generate high-density point cloud data.

[0060] The specific steps are:

[0061] Import the image files taken by the drone through the "Add Image" function; check the image quality and remove photos with blur, overexposure and other problems; select the specified coordinate system; choose to generate a 3D map, that is, a point cloud surface model;

[0062] 2) Use CATIA software to build a 3D model of the material yard based on the point cloud data. Specifically, use the point cloud data and CloudCompare modeling software to create an image surface model of the material yard, crop the material yard area, and then use CATIA software to generate a 3D model of the material yard.

[0063] When using CATIA, the usage is:

[0064] 2.1) Select the las point cloud data file and import it into CloudCompare software;

[0065] 2.2) Use the Crop function to crop the discharge area;

[0066] 2.3) Adjust the compass parameters to generate the material field grid model;

[0067] 2.4) Smooth the material yard grid model and export it to obtain a three-dimensional model of the material yard.

[0068] 3) Construct an image recognition model for blasting areas in a rockfill material yard, specifically: collect and organize image datasets of the material yard, including images of blasting areas. Then, use image annotation tools to manually annotate the blasting areas and other areas in these images to generate an image dataset containing blasting area annotations. During the annotation process, it is necessary to draw the boundaries of the blasting area to ensure the accuracy and consistency of the annotated data. Import the annotated image dataset into the YOLOv8 model for training. During training, adjust the model's hyperparameters (such as learning rate, batch size, etc.) according to the scale and complexity of the dataset to optimize the training effect. After training, use the validation set to test the model to evaluate its recognition performance in images that are not in the training and test sets. Based on the verification results, further optimize and adjust the model, or expand the training dataset to improve the generalization ability and recognition accuracy of the model. Specifically:

[0069] 3.1) Collect the overall bird’s-eye view images of the rockfill site taken by drones;

[0070] 3.2) performing sample enhancement on the material field images collected in step 3.1), and dividing the images before and after enhancement into a training set and a test set in proportion;

[0071] 3.3) Configure the image recognition model training parameters, including the number of training rounds and batch size;

[0072] 3.4) Evaluate the performance of the model through the test set images. The evaluation indicators include precision P, recall R and average precision AP. The calculation method is:

[0073]

[0074]

[0075] in:

[0076] TTP is the number of correct predictions;

[0077] FFP is the number of false predictions, including identifying targets that are not in the working area as the working area and missing identification;

[0078] FFN is the number of work areas that are mistakenly identified as other categories;

[0079] 3.5) Based on the evaluation results of step 3.4), the image recognition model of the rockfill blasting area is finally determined. The conditions for determining the image recognition model of the rockfill blasting area are that the recall rate R is not less than 85%, and the accuracy rate and average accuracy rate are both between 90% and 100%.

[0080] 4) Upload the material field image to the computer and use the image recognition model to identify the blasting area; specifically: after the training is completed, the new material field image is imported into the trained YOLOv8 model to automatically identify the blasting area. The model will detect the blasting area from the image and generate the corresponding mask, such as Figure 2 As shown. This mask is a binary image that marks the location and outline of the blasting area in the material field image. The specific steps are: use labelme annotation software to mark the blasting area in the material field image collected by the drone; import the labeled image into the image recognition model for training to obtain the optimal model parameters; use the model with the optimal parameters to identify the blasting area in the material field image; save the material field image with the blasting area mask;

[0081] 5) Overlay the blasting area coordinates onto the three-dimensional model of the material yard and mark the blasting area. Specifically, export the pixel coordinate file of the blasting area mask in the material yard image identified in step 4), convert all pixel coordinates covered by the mask into model coordinates, use these coordinates as the object of convex hull fitting, calculate the boundary points of each area through the convex hull algorithm, output their boundary coordinates, use the polygon tool to draw the blasting area in the three-dimensional model of the material yard according to the boundary coordinates, and add color, texture or label to the blasting area.

[0082] 5.1) Obtain the internal and external parameters of the drone camera. The internal parameters of the drone camera include focal length (f x , f y ) and the principal point coordinates (c x , c y );UAV camera external parameters include rotation matrix K and translation vector t;

[0083] 5.2) Normalize the pixel coordinates of all pixels of the mask-covered material field image according to the internal and external parameters of the drone camera in step 5.1), specifically:

[0084]

[0085] in:

[0086] (u, v) are the pixel coordinates of all pixels in the mask-covered material field image;

[0087] z is the pixel depth information;

[0088] x, y, z are the three-dimensional coordinates in camera coordinates;

[0089] 5.3) Convert the three-dimensional coordinates in the camera coordinate system to the three-dimensional coordinates (X, Y, Z) in the model coordinate system, specifically:

[0090] [X,Y,Z]=K*[x,y,z]+t.

[0091] 6) The coordinates transformed in step 5) are used as the object of convex hull fitting. The boundary points of each area are obtained after calculation by the convex hull algorithm. The boundary coordinates are output. The blasting area is drawn using a polygon tool in the three-dimensional model of the material yard according to the boundary coordinates. The blasting area is marked by adding color, texture or label.

[0092] 7) Construct a three-dimensional model of the blasting area in the three-dimensional model of the material field, and after the construction is completed, calculate the loose volume of the blasting area through rasterization operation; specifically:

[0093] 7.1) Select the established 3D model of the blasted area;

[0094] 7.2) Select the rasterization function and set the rasterization parameters, i.e. the grid size;

[0095] 7.3) Performing a rasterization operation to divide the blasted area model into a series of small grids;

[0096] 7.4) Perform volume accumulation on each grid.

Claims

1. A method for calculating the reserves of a rockfill site, characterized in that: The rockfill reserve calculation method comprises the following steps: 1) Use an aerial survey drone to shoot images of the material yard, obtain images of the material yard, and obtain point cloud data of the material yard based on the aerial survey images; 2) Establishing a three-dimensional model of the material yard based on the point cloud data obtained in step 1); 3) Construct an image recognition model for the blasting area of ​​a rockfill site; 4) using the rockfill blasting area image recognition model obtained in step 3) to identify the blasted area in the material yard image obtained in step 1); the blasted area is the area of ​​all loose rocks within the material yard; 5) Convert the pixel coordinates of the identified blasting area in the image into earth rectangular coordinates; 6) superimposing the spatial range of the blasting area onto the three-dimensional model of the material field constructed in step 2), and marking the blasting area; 7) Construct a three-dimensional model of the blasting area in the three-dimensional model of the material yard, and after the construction is completed, calculate the loose volume of the blasting area through rasterization operation.

2. The method for calculating the reserves of a rockfill site according to claim 1, characterized in that: The method for acquiring the material yard image in the step 1) is: selecting a drone with GPS and RTK functions, carrying a shooting camera on the drone, and setting the drone along the navigation track of the material yard at a height of 350-450m above the ground to acquire the material yard image; preferably, the resolution of the shooting camera is not less than 5472×3648; the method for acquiring the point cloud data in the step 1) is: splicing, correcting and three-dimensionally reconstructing the material yard image taken by the drone to generate point cloud data.

3. The method for calculating the reserves of a rockfill site according to claim 2, characterized in that: The step 2) is to use the point cloud data to create an image surface model of the material yard using CloudCompare modeling software, cut out the material yard area, and then use CATIA software to generate a three-dimensional model of the material yard.

4. The method for calculating the reserves of a rockfill site according to claim 3, characterized in that: The specific implementation method of step 2) is: 2.1) Select the las point cloud data file and import it into CloudCompare software; 2.2) Use the Crop function to crop the discharge area; 2.3) Adjust the compass parameters and generate the material field grid model; 2.4) Smooth the material yard grid model and export it to obtain a three-dimensional model of the material yard.

5. The method for calculating the reserves of a rockfill site according to claim 4, characterized in that: The specific implementation method of step 3) is: 3.1) Collect the overall bird’s-eye view images of the rockfill site taken by drones; 3.2) performing sample enhancement on the material field images collected in step 3.1), and dividing the images before and after enhancement into a training set and a test set in proportion; 3.3) Configuring image recognition model training parameters, including the number of training rounds and batch size; 3.4) Evaluate the performance of the model through the test set images. The evaluation indicators include precision P, recall R and average precision AP, which are calculated as follows: AP=∫0 1 P(P)dR in: TTP is the number of correct predictions; FFP is the number of false predictions, including identifying targets that are not in the working area as the working area and missing identification; FFN is the number of work areas that are mistakenly identified as other categories; 3.5) Based on the evaluation result of step 3.4), the image recognition model of the rockfill blasting area is finally determined. The conditions for determining the image recognition model of the rockfill blasting area are that the recall rate R is not less than 85%, and the accuracy rate and the average accuracy rate are both between 90% and 100%.

6. The method for calculating the reserves of a rockfill site according to claim 5, characterized in that: The step 4) uses the rockfill blasting area image recognition model constructed in step 3), imports the rockfill image with the blasting area marked into the image recognition model for training, calls the trained image recognition model to identify the blasting area in the rockfill image, and obtains a mask of the blasting area.

7. The method for calculating the reserves of a rockfill site according to claim 6, characterized in that: The specific implementation method of step 4) is: 4.1) Use labelme software to mark the blasting area in the material field pictures collected by drone; 4.2) Import the labeled images into the image recognition model constructed in step 3) for training to obtain the optimal parameters of the image recognition model; 4.3) Using the image recognition model with optimal parameters to identify the blasting area in the stockyard image; 4.4) Save the identified material field image with the blasting area mask.

8. The method for calculating the reserves of a rockfill site according to claim 7, characterized in that: The step 5) is to export the pixel coordinate file of the blasting area mask in the stock field image identified in step 4), and convert all pixel coordinates covered by the mask into model coordinates. The specific implementation method is: 5.1) Obtain the internal and external parameters of the drone camera, the internal parameters of the drone camera include focal length (f x , f y ) and the principal point coordinates (c x , c y );The drone camera external parameters include the rotation matrix K and the translation vector t; 5.2) Normalize the pixel coordinates of all pixels of the mask-covered material field image according to the internal and external parameters of the drone camera in step 5.1), specifically: in: (u, v) are the pixel coordinates of all pixels in the mask-covered material field image; z is the pixel depth information; x, y, z are the three-dimensional coordinates in camera coordinates; 5.3) Convert the three-dimensional coordinates in the camera coordinate system to the three-dimensional coordinates (X, Y, Z) in the model coordinate system, specifically: [X,Y,Z]=K*[x,y,z]+t.

9. The method for calculating the reserves of a rockfill site according to claim 8, characterized in that: The specific implementation method of step 6) is: using the coordinates converted in step 5) as the object of convex hull fitting, obtaining the boundary points of each area after calculation by the convex hull algorithm, outputting its boundary coordinates, and using a polygon tool to draw the blasting area in the three-dimensional model of the material yard according to the boundary coordinates, marking the blasting area, and the marking is to add color, texture or label.

10. The method for calculating the reserves of a rockfill site according to claim 9, characterized in that: The specific implementation of step 7) is: 7.1) Select the established 3D model of the blasted area; 7.2) Select the rasterization function and set the rasterization parameters, i.e. the grid size; 7.3) Performing a rasterization operation to divide the blasted area model into a series of small grids; 7.4) Perform volume accumulation on each grid.

Images