A method for identifying and fusion modeling of multivariate engineering geological information on open-pit mine slopes

Abstract

The present invention belongs to the field of multivariate geological information modeling for mining engineering and proposes a method for multivariate engineering geological information identification and fusion modeling of open-pit mine slopes. The method comprises collecting core images from different areas of the open-pit mine, performing lithology identification based on the core images, and obtaining a core lithology dataset; acquiring surface rock layer images of the open-pit mine through drone aerial survey, performing lithology identification based on the surface rock layer images, and obtaining a surface lithology dataset; identifying slope water outlet points on the surface rock layer images of the open-pit mine, and obtaining a slope water outlet point dataset; and quantifying the spatial coordinates of the core lithology dataset, the surface lithology dataset, and the slope water outlet point dataset, and fusion modeling to generate an open-pit mine slope geological model. The present invention can effectively integrate multiple engineering geological parameters such as lithology, slope characteristics, groundwater conditions, and water outlet point distribution, achieving precise quantification of the spatial coordinates of engineering geological characteristics, and significantly improving the efficiency and accuracy of geological modeling.

Description

Technical Field

[0001] The present invention belongs to the field of multivariate geological information modeling for mining engineering, and specifically discloses a method for identifying and fusion modeling of multivariate engineering geological information for open-pit mine slopes. Background Art

[0002] Slope stability in open-pit mining is a key issue for ensuring mining safety and mining area productivity. As mining depths increase, the geological factors affecting mine slope stability become increasingly complex and variable. Traditional slope stability analysis methods often rely on a single source of geological data, such as drill hole data or geological exploration results. However, in practice, this data is often incomplete, unevenly distributed, or difficult to collect. Therefore, effectively integrating multi-source geological data to develop more accurate and comprehensive rock mechanics models has become a major challenge in open-pit mining engineering.

[0003] At present, multivariate geological information identification and fusion technology has been applied to engineering geology and rock mechanics research to a certain extent, but most methods are still based on a single data source and are difficult to deal with slope stability issues in complex geological environments. In mining, it is often necessary to combine multiple geological parameters such as lithology, slope characteristics, groundwater, and adopt different data analysis methods to improve the accuracy and operability of the model. However, existing geological information fusion models generally have problems such as low data processing efficiency, insufficient fusion accuracy, and neglect of spatial correlation between data, resulting in poor results in slope stability analysis and prediction. Therefore, it is very necessary to study and design a new multivariate engineering geological information identification and fusion modeling method for open-pit mine slopes. This method is based on multi-source geological parameters and uses data fusion technology and modeling algorithms to comprehensively consider the multiple influences of geological parameters to solve the problems existing in existing open-pit mine modeling. Summary of the Invention

[0004] In order to solve the problems of low data processing efficiency, insufficient fusion accuracy and neglect of spatial correlation between data in existing open-pit mine modeling, the present invention proposes a method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes.

[0005] The present invention provides a method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes, comprising the following steps:

[0006] S1. Collecting core images from different areas of the open-pit mine, performing lithology identification based on the core images, and obtaining a core lithology dataset;

[0007] S2. Obtaining surface rock images of the open-pit mine through drone aerial surveying, and performing lithology identification based on the surface rock images of the open-pit mine to obtain a surface lithology dataset;

[0008] S3 obtained in step S2 of the surface rock image of the open pit slope water point identification, obtain a slope water point data set;

[0009] S4. quantifying the spatial coordinates of the core lithology dataset obtained in step S1, the surface lithology dataset obtained in step S2, and the slope water outlet point dataset obtained in step S3, and integrating them into a model to generate a geological model of the open pit mine slope.

[0010] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, step S1 includes:

[0011] S1.1. Collect core photographs from different areas of the open-pit mine, perform image processing on the core photographs to generate core images, and record the drill hole number, burial depth, and core placement order corresponding to each core image.

[0012] S1.2. Perform lithologic labeling on the processed core images to obtain a core image dataset. The core image dataset is divided into a training set and a test set for training and testing the Mask R-CNN model.

[0013] S1.3. Pre-train the Mask R-CNN model using the training set, and optimize the network parameters of the Mask R-CNN model until the preset target is achieved.

[0014] S1.4. Input the test set into the trained Mask R-CNN model to generate a target mask and output the lithology recognition results, thereby obtaining the core lithology dataset.

[0015] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, in step S1.2, the core image dataset includes core images collected from different areas of the open-pit mine and lithologic annotations corresponding to the core images.

[0016] According to a method for identifying and fusion modeling of multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, in step S1.3, pre-training a Mask R-CNN model using the training set includes:

[0017] S1.3.1. Set initial parameters and adjust learning rate.

[0018] S1.3.2. Input the training set into the Mask R-CNN network and perform forward propagation to obtain the prediction results.

[0019] S1.3.3. Compare the prediction results with the data labels to obtain the validation loss value.

[0020] S1.3.4. Perform backpropagation and adjust network parameters using mini-batch gradient descent.

[0021] S1.3.5. Repeat S1.3.1 to S1.3.4 until the loss value reaches the preset target.

[0022] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, step S2 includes:

[0023] S2.1. Obtaining surface images of open-pit mines using drone-based oblique photography.

[0024] S2.2. Based on the surface photos of the open-pit mine, an image of the open-pit mine surface rock formations is generated and lithology annotation is performed to obtain an image dataset of the open-pit mine surface rock formations;

[0025] S2.3. Adapt the trained Mask R-CNN model to the surface lithology recognition task through transfer learning. Input the open-pit mine surface rock layer image dataset into the trained Mask R-CNN model, output the surface lithology recognition results, and obtain a surface lithology dataset.

[0026] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, in step S2.2, the open-pit mine surface rock layer image dataset includes open-pit mine surface rock layer images, rock types, rock textures, rock structures, and lithology annotations.

[0027] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, step S3 includes:

[0028] S3.1. Based on the surface rock layer images of the open-pit mine, water point images were generated and annotated. Slope water point identification was transformed into a target detection problem. A water point image dataset was generated. This dataset included water point images and corresponding water point interface annotations for different slope environments. The dataset was divided into a training set and a test set.

[0029] S3.2. Construct a single-stage target detection algorithm, pre-train the single-stage target detection algorithm using the water outlet training set, and test the trained single-stage target detection algorithm using the water outlet test set. The single-stage target detection algorithm includes:

[0030] Backbone network, used to extract multi-scale features from the input water outlet image;

[0031] Neck feature fusion module, used to fuse feature information at different levels;

[0032] The head detection head is used to output the category probability and bounding box position of the water outlet to obtain the prediction result;

[0033] Loss Function is used to calculate the difference between the predicted result and the water outlet interface annotation;

[0034] S3.3. Detect water points in the water point image dataset using the tested single-stage object detection algorithm. Each surface rock layer image of the open-pit mine is gridded, and multiple bounding boxes are predicted for each grid. Output includes water point category probabilities and bounding box offsets, resulting in water point detection results.

[0035] S3.4. Associate the water outlet point detection results and the positions of the water outlet points in the image coordinate system with the geographic coordinate information obtained by the drone aerial survey to obtain the slope water outlet point dataset containing spatial position information.

[0036] According to a method for identifying and fusion modeling multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, step S4 includes:

[0037] S4.1. Quantify the three-dimensional coordinates of the data in the core lithology dataset, find the actual three-dimensional coordinate position based on the drill hole number, quantify the longitude, latitude, elevation, drill hole depth, burial depth, and corresponding lithology of the core image, and obtain a drill hole database;

[0038] S4.2. Quantify the three-dimensional coordinates of the data in the surface lithology dataset and the slope water outlet point dataset. Based on the drone aerial survey imagery, extract the target mask using an image recognition algorithm. Combined with the drone attitude angle and surface elevation point cloud data, quantify the three-dimensional coordinates of the exposed lithology interface and the three-dimensional coordinates of the slope water outlet points to obtain the slope lithology database and slope water outlet point database.

[0039] S4.3. Integrate the borehole database, slope lithology database, and slope water point database obtained after coordinate quantification to construct a layered slope model database with three-dimensional coordinate information;

[0040] S4.4. Based on the slope layered model database, a lithologic layered model is generated by 3Dmine software. The lithologic layered model includes the surface, soil, rock, and coal layers before and after excavation.

[0041] S4.5. Based on the contour coordinates of the lithologic boundary lines in the lithologic layered model, add surface entities and assign lithologic attributes in the GIS platform to implement the slope lithologic model.

[0042] S4.6. Based on the slope water point coordinates in the slope layered model database, dynamically draw a water point range model through WebGIS;

[0043] S4.7. Integrate the lithologic layered model, slope lithologic model, and water point range model into a multivariate information database;

[0044] S4.8. Dynamically construct and update the open pit mine slope geological model using Kriging interpolation based on the multivariate information database.

[0045] According to a method for identifying and fusion modeling of multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, in step S4.6, a solid surface is added through Cesium's WebGIS to realize dynamic drawing of a water outlet range model.

[0046] According to a method for identifying and fusion modeling of multivariate engineering geological information of an open-pit mine slope in some embodiments of the present application, the dynamic update of the open-pit mine slope geological model in step S4.8 is based on real-time data input of the slope layered model database.

[0047] The present invention provides a multi-dimensional engineering geological information identification and fusion modeling method for open-pit mine slopes, which can effectively integrate multiple engineering geological parameters such as lithology, slope characteristics, groundwater conditions, and water point distribution. By constructing a deep learning model, the intelligent identification of drill core lithology, slope rock layers, and slope water points is realized, which greatly improves the automation level of geological information acquisition. The innovative combination of drone aerial survey technology and image recognition algorithm realizes the precise quantification of the spatial coordinates of engineering geological characteristics. The multi-dimensional data fusion method is used to organically integrate core data, slope lithology data, and water point data to establish a unified engineering geological information database, overcome the limitations of a single data source, and make the engineering geological model more accurate. Based on geostatistical methods and geographic information technology, efficient lithology three-dimensional distribution modeling and water point range characterization technology have been developed, which significantly improves the efficiency and accuracy of geological modeling. The three-dimensional visualization display of the engineering geological model is realized through the WebGIS platform, supporting the dynamic update and maintenance of the model. This method effectively solves the problems of low efficiency, high dependence on manual labor, and delayed model updates of traditional geological exploration methods. It provides a complete set of intelligent technical solutions for open-pit mine slope engineering geological survey and monitoring, and effectively enhances the ability to assess mine mining safety under complex geological conditions. It is of great significance to improving the level of mine safety production. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of a method for intelligent identification and fusion modeling of multi-dimensional engineering geological information of open-pit mine slopes provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0050] Example 1, in response to the problems existing in current geological modeling, this embodiment proposes a method for intelligent recognition and fusion modeling of multi-dimensional engineering geological information of open-pit mine slopes based on multi-dimensional engineering geological information recognition, combined with consideration of drone aerial survey image information extraction and multi-dimensional data fusion. This embodiment provides a method for intelligent recognition and fusion modeling of multi-dimensional engineering geological information of open-pit mine slopes, such as Figure 1 As shown, the following steps are included:

[0051] S1. Collect core images from different areas of an open-pit mine and perform lithology identification based on the core images to generate a core lithology dataset.

[0052] S2. Use drone aerial surveys to obtain images of the surface rock formations at the open-pit mine. Lithology identification is performed based on these images to generate a surface lithology dataset.

[0053] S3. Identify slope water outlet points on the surface rock layer image of the open-pit mine obtained in step S2 to obtain a slope water outlet point dataset.

[0054] S4. Quantify the spatial coordinates of the core lithology dataset obtained in step S1, the surface lithology dataset obtained in step S2, and the slope water outlet point dataset obtained in step S3, and fuse them into a model to generate a geological model of the open pit mine slope.

[0055] A multi-dimensional engineering geological information identification and fusion modeling method for open-pit mine slopes in this embodiment realizes the construction of a three-dimensional spatial distribution model of lithology and slope water outlet range by intelligently identifying and quantifying the lithology of open-pit mine cores, surface lithology, and water outlet points of open-pit mine slopes, and combining geostatistical methods and geographic information technology. This improves the intelligent level of engineering geological information acquisition while enhancing the efficiency and reliability of model construction.

[0056] Example 2: This example provides a method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes, comprising the following steps:

[0057] S1. Collect core images from different areas of an open-pit mine and perform lithology identification based on the core images to generate a core lithology dataset.

[0058] As a preference of this embodiment, specifically, step S1 includes:

[0059] S1.1. Collect core photographs from different areas of the open-pit mine, perform image processing on the core photographs to generate core images, and record the drill hole number, burial depth, and core placement order corresponding to each core image.

[0060] S1.2. Perform lithologic labeling on the processed core images to obtain a core image dataset. This dataset is then divided into a training set and a test set for training and testing the Mask R-CNN model.

[0061] In step S1.2, the core image dataset includes core images collected from different areas of the open pit mine and lithology annotations corresponding to the core images;

[0062] S1.3. Pre-train the Mask R-CNN model using the training set and optimize the network parameters of the Mask R-CNN model until the preset target is achieved.

[0063] In step S1.3, using the training set to pre-train the Mask R-CNN model includes:

[0064] S1.3.1. Set initial parameters and adjust the learning rate. More specifically, when adjusting the learning rate, you can lower it appropriately.

[0065] S1.3.2. The training set is fed into the Mask R-CNN network for forward propagation to obtain the prediction results. More specifically, after the training set is fed into the Mask R-CNN network, it undergoes layer-by-layer forward propagation to obtain the prediction results.

[0066] S1.3.3. Compare the prediction results with the data labels to obtain the validation loss value.

[0067] S1.3.4. Perform backpropagation and adjust network parameters using mini-batch gradient descent.

[0068] S1.3.5. Repeat S1.3.1 through S1.3.4 until the loss reaches the preset target.

[0069] In this embodiment, the training process of the Mask R-CNN model pre-trained on the training set can be similar to an "iterative fitting" process. When the verification loss value or the prediction accuracy does not change much, the iteration is stopped. The trained Mask R-CNN model can be used to identify a single row of cores from the core image;

[0070] S1.4. Input the test set into the trained Mask R-CNN model to generate a target mask and output the lithology recognition results, thus obtaining a core lithology dataset. More specifically, input the test set into the trained Mask R-CNN model, which will use color blocks to fill in the cores in different areas of the core image, thereby distinguishing different lithologies.

[0071] S2. Use drone aerial surveys to obtain images of the surface rock formations at the open-pit mine. Lithology identification is performed based on these images to generate a surface lithology dataset.

[0072] As a preference of this embodiment, specifically, step S2 includes:

[0073] S2.1. Obtain surface images of open-pit mines using drone-based oblique photography. More specifically, these images capture the diverse rock types, textures, and structures found in the mine's diverse geological conditions.

[0074] S2.2. Create surface rock images based on surface mine photos and perform lithology annotation to obtain a dataset of surface rock images.

[0075] In step S2.2, the open-pit mine surface rock stratum image dataset includes open-pit mine surface rock stratum images, rock types, rock textures, rock structures, and lithology annotations; the open-pit mine surface rock stratum image dataset can be divided into an open-pit mine surface rock stratum image training set and an open-pit mine surface rock stratum image test set for training and testing the Mask R-CNN model.

[0076] S2.3. Adapt the trained Mask R-CNN model to the surface lithology recognition task through transfer learning. Input the surface rock layer image dataset of the open-pit mine into the trained Mask R-CNN model, output the surface lithology recognition results, and obtain the surface lithology dataset.

[0077] S3. Identify slope water outlet points on the surface rock layer image of the open-pit mine obtained in step S2 to obtain a slope water outlet point dataset.

[0078] As a preference of this embodiment, specifically, step S3 includes:

[0079] S3.1. Based on images of the open-pit mine surface rock formations, water point images are generated and labeled. Slope water point identification is transformed into a target detection problem, resulting in a water point image dataset. This dataset includes images of water points in different slope environments and corresponding water interface annotations. The dataset is divided into a water point training set and a water point test set. In this embodiment, drone aerial surveys automatically capture high-quality images of the open-pit mine surface rock formations, providing crucial data support for the single-stage target detection algorithm used for image recognition. With this collected open-pit mine surface rock formation image data, slope water point identification is transformed into a target detection problem. Using intelligent algorithms, targets in the open-pit mine surface rock formation images can be automatically identified and classified.

[0080] S3.2. Construct a single-stage object detection algorithm. Use the water outlet training set to pre-train the single-stage object detection algorithm. Use the water outlet test set to test the trained single-stage object detection algorithm. The single-stage object detection algorithm includes:

[0081] Backbone network, used to extract multi-scale features from the input water outlet image;

[0082] Neck feature fusion module, used to fuse feature information at different levels;

[0083] The head detection head is used to output the category probability and bounding box position of the water outlet to obtain the prediction result;

[0084] Loss Function is used to calculate the difference between the predicted result and the water outlet interface annotation;

[0085] S3.3. Use the tested single-stage object detection algorithm to detect water points in the water point image dataset. Each open-pit mine surface rock layer image is gridded, and multiple bounding boxes are predicted for each grid. Output includes the water point category probability and bounding box offset, resulting in the water point detection result.

[0086] S3.4. Associate the water point detection results and the locations of the water points in the image coordinate system with the geographic coordinate information obtained by the UAV aerial survey to obtain a slope water point dataset containing spatial location information.

[0087] The single-stage target detection algorithm in this embodiment directly extracts features from the input water outlet image, and then outputs the water outlet category probability and bounding box, skipping the process of generating candidate areas. The single-stage target detection algorithm can perform efficient and accurate target detection on the water outlet image in real time. The overall architecture of the single-stage target detection algorithm is a convolutional neural network CNN based on deep learning. Its structure is mainly composed of four modules. The four modules of Backbone backbone network, Neck feature fusion module, Head detection head and Loss Function loss function work together to realize the full process target detection from image input to target output.

[0088] S4. Quantify the spatial coordinates of the core lithology dataset obtained in step S1, the surface lithology dataset obtained in step S2, and the slope water outlet point dataset obtained in step S3, and fuse them into a model to generate a geological model of the open pit mine slope.

[0089] As a preference of this embodiment, specifically, step S4 includes:

[0090] S4.1. Quantify the three-dimensional coordinates of the data in the core lithology dataset, find the actual three-dimensional coordinates based on the borehole number, and quantify the longitude, latitude, elevation, borehole depth, burial depth, and corresponding lithology of the core image to obtain a borehole database. In this embodiment, the core images collected in step S1 each contain the corresponding borehole number, burial depth, and core placement order. Based on the borehole number in the image, the actual location of the borehole or the coordinates recorded in the data are found. Coordinate quantization is then performed on the core images to obtain a borehole database containing the longitude, latitude, elevation, borehole depth, burial depth, and corresponding lithology of the borehole core images.

[0091] S4.2. Quantify the three-dimensional coordinates of the surface lithology dataset and the slope water point dataset. Based on the drone aerial survey imagery, an image recognition algorithm is used to extract a target mask. Combined with the drone's attitude angle and surface elevation point cloud data, the three-dimensional coordinates of the exposed lithology interface and the slope water point are quantified to obtain a slope lithology database and a slope water point database. In this embodiment, steps S2 and S3 are both based on drone aerial survey images of the open-pit mine surface rock formations. Therefore, the methods and principles for quantifying the coordinates of the exposed lithology interface and the slope water point are the same. First, an image recognition algorithm is used to extract a mask of the target in the image. This mask consists of a list of normalized two-dimensional coordinates, representing the target's position in the image coordinate system. Calculations are performed based on the drone's shooting direction and the actual surface elevation point cloud data. The drone's shooting direction is determined by the attitude angles (yaw, pitch, and roll), while the drone's point cloud model provides elevation information for each surface point. By combining the geographic coordinates captured by the drone, namely longitude, latitude and altitude, with the three-dimensional coordinate system of the point cloud model, the location of the target can be accurately calculated in the actual geographic space;

[0092] S4.3. The borehole database, slope lithology database, and slope water point database obtained after coordinate quantization are integrated to construct a slope layered model database with three-dimensional coordinate information. In this embodiment, the multi-dimensional engineering geological information on the surface and interior of the open-pit mine, obtained using different realistic methods, is coordinate-transformed and format-converted, and stored in a unified database under a unified coordinate system.

[0093] S4.4. Based on the slope layer model database, a lithologic layer model was generated using 3Dmine software. The lithologic layer model included the surface, soil, rock, and coal seams before and after excavation. In this embodiment, a 3D engineering geology model was constructed using 3DMine and Rhino modeling software, using a cross-section of the open-pit mine, a surface topographic map including drill hole coordinates, and an open-pit mine boundary map.

[0094] S4.5. Based on the contour coordinates of the lithologic boundary lines in the lithologic layered model, surface entities are added to the GIS platform and assigned lithologic attributes to realize the slope lithologic model. In this embodiment, surface entities close to the ground are added to the GIS platform and assigned different colors to identify lithologic attributes. This enables the integrated functions of automatic identification of slope rock mass lithology, real-time upload of contour coordinates, and intelligent visualization of slope lithology.

[0095] S4.6. Dynamically draw a model of the water outlet range using WebGIS based on the coordinates of the slope water outlet points in the slope layer model database.

[0096] In this embodiment, specifically, in step S4.6, an entity surface is added through Cesium's WebGIS to realize the dynamic drawing of the water outlet range model; more specifically, the modeling of the water point range model is based on the automatic recognition of the drone oblique photography model, and the water outlet range coordinates will be generated after the recognition is completed. After storing them in the database, the water outlet range model can be visualized through the front-end code; in the WebGIS based on Cesium, vector surfaces can be directly drawn in the three-dimensional model by adding entity surfaces, and the color, transparency, etc. of the vector surfaces can be changed through material properties; the surface data coordinates here can be given directly in the code, but in order to realize the dynamic update of the water outlet range, the water outlet coordinates are directly stored in the database, and the database is read through the post request function at the front end to realize the dynamic drawing of the water outlet range;

[0097] S4.7. Integrate the lithologic layered model, slope lithologic model, and water point range model into a multivariate information database. In this embodiment, specifically, the spatiotemporal fusion database of multivariate engineering geological information on the surface and within the open-pit mine is combined with geostatistical inversion to form a dynamic, rapid, and precise modeling method for engineering geological models. Based on geological drilling data, intelligent lithologic identification using core images and slope surfaces is used to construct a slope layered model database. After reading the database in 3Dmine software, a lithologic layered model can be directly constructed using commands. The high-precision slope model acquired by the drone is converted into point cloud data, which is then imported into 3Dmine to construct a DTM surface for cropping the layered model. Finally, a lithologic layered model is constructed.

[0098] S4.8. Dynamically construct and update open-pit mine slope geological models using Kriging interpolation based on a multivariate information database;

[0099] More specifically, the dynamic update of the open-pit mine slope geological model in step S4.8 is based on the real-time data input of the slope layered model database. This embodiment uses the multivariate engineering geological information database of the open-pit mine surface and the open-pit mine as input variables, uses Kriging interpolation to quickly construct the engineering geological model, and continuously and dynamically modifies the engineering geological model as the database is updated to maintain its freshness and accuracy.

[0100] The embodiments of the present invention are presented for purposes of illustration and description and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those skilled in the art. The embodiments are chosen and described in order to better illustrate the principles of the invention and its practical application and to enable those skilled in the art to understand the invention and design various embodiments with various modifications as suited for specific applications.

Claims

1. A method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes, characterized by: The steps include: S1. Collecting core images from different areas of the open-pit mine, performing lithology identification based on the core images, and obtaining a core lithology dataset; S2. Obtaining surface rock images of the open-pit mine through drone aerial surveying, and performing lithology identification based on the surface rock images of the open-pit mine to obtain a surface lithology dataset; S3 obtained in step S2 of the surface rock image of the open pit slope water point identification, obtain a slope water point data set; S4. The core lithology dataset obtained in step S1, the surface lithology dataset obtained in step S2, and the slope water point dataset obtained in step S3 are spatially quantified and integrated into a model to generate a geological model of the open pit slope; The S3 includes: S3.

1. Based on the surface rock layer images of the open-pit mine, water point images were generated and annotated. Slope water point identification was transformed into a target detection problem. A water point image dataset was generated. This dataset included water point images and corresponding water point interface annotations for different slope environments. The dataset was divided into a training set and a test set. S3.

2. Construct a single-stage target detection algorithm, pre-train the single-stage target detection algorithm using the water outlet training set, and test the trained single-stage target detection algorithm using the water outlet test set. The single-stage target detection algorithm includes: Backbone network, used to extract multi-scale features from the input water outlet image; Neck feature fusion module, used to fuse feature information at different levels; The head detection head is used to output the category probability and bounding box position of the water outlet to obtain the prediction result; Loss Function is used to calculate the difference between the predicted result and the water outlet interface annotation; S3.

3. Detect water points in the water point image dataset using the tested single-stage object detection algorithm. Each surface rock layer image of the open-pit mine is gridded, and multiple bounding boxes are predicted for each grid. Output includes water point category probabilities and bounding box offsets, resulting in water point detection results. S3.

4. Associating the water point detection results, the position of the water point in the image coordinate system with the geographic coordinate information obtained by the drone aerial survey to obtain the slope water point dataset containing spatial location information; The S4 includes: S4.

1. Quantify the three-dimensional coordinates of the data in the core lithology dataset, find the actual three-dimensional coordinate position based on the drill hole number, quantify the longitude, latitude, elevation, drill hole depth, burial depth, and corresponding lithology of the core image, and obtain a drill hole database; S4.

2. Quantify the three-dimensional coordinates of the data in the surface lithology dataset and the slope water outlet point dataset. Based on the drone aerial survey imagery, extract the target mask using an image recognition algorithm. Combined with the drone attitude angle and surface elevation point cloud data, quantify the three-dimensional coordinates of the exposed lithology interface and the three-dimensional coordinates of the slope water outlet points to obtain the slope lithology database and slope water outlet point database. S4.

3. Integrate the borehole database, slope lithology database, and slope water point database obtained after coordinate quantification to construct a layered slope model database with three-dimensional coordinate information; S4.

4. Based on the slope layered model database, a lithologic layered model is generated by 3Dmine software. The lithologic layered model includes the surface, soil, rock, and coal layers before and after excavation. S4.

5. Based on the contour coordinates of the lithologic boundary lines in the lithologic layered model, add surface entities and assign lithologic attributes in the GIS platform to implement the slope lithologic model. S4.

6. Based on the slope water point coordinates in the slope layered model database, dynamically draw a water point range model through WebGIS; S4.

7. Integrate the lithologic layered model, slope lithologic model, and water point range model into a multivariate information database; S4.

8. Dynamically construct and update the open pit mine slope geological model using Kriging interpolation based on the multivariate information database.

2. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 1 is characterized in that: The step S1 comprises: S1.

1. Collect core photographs from different areas of the open-pit mine, perform image processing on the core photographs to generate core images, and record the drill hole number, burial depth, and core placement order corresponding to each core image. S1.

2. Perform lithology annotation on the processed core images to obtain a core image dataset, and divide the core image dataset into a training set and a test set; S1.

3. Pre-train the Mask R-CNN model using the training set, and optimize the network parameters of the Mask R-CNN model until the preset target is achieved. S1.

4. Input the test set into the trained Mask R-CNN model to generate a target mask and output the lithology recognition results, thereby obtaining the core lithology dataset.

3. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 2 is characterized in that: In S1.2, the core image dataset includes core images collected from different areas of the open-pit mine and lithology annotations corresponding to the core images.

4. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 2, characterized in that: In S1.3, pre-training the Mask R-CNN model using the training set includes: S1.3.

1. Set initial parameters and adjust learning rate. S1.3.

2. Input the training set into the Mask R-CNN network and perform forward propagation to obtain the prediction results. S1.3.

3. Compare the prediction results with the data labels to obtain the validation loss value. S1.3.

4. Perform backpropagation and adjust network parameters using mini-batch gradient descent. S1.3.

5. Repeat S1.3.1 to S1.3.4 until the loss value reaches the preset target.

5. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 2, characterized in that: The S2 includes: S2.

1. Obtaining surface images of open-pit mines using drone-based oblique photography. S2.

2. Based on the surface photos of the open-pit mine, an image of the open-pit mine surface rock formations is generated and lithology annotation is performed to obtain an image dataset of the open-pit mine surface rock formations; S2.

3. Adapt the trained Mask R-CNN model to the surface lithology recognition task through transfer learning. Input the open-pit mine surface rock layer image dataset into the trained Mask R-CNN model, output the surface lithology recognition results, and obtain a surface lithology dataset.

6. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 5, characterized in that: In S2.2, the open-pit mine surface rock layer image dataset includes open-pit mine surface rock layer images, rock types, rock textures, rock structures, and lithology annotations.

7. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 1, characterized in that: In the S4.6, a physical surface is added through Cesium's WebGIS to realize the dynamic drawing of the water outlet range model.

8. The method for identifying and fusion modeling of multivariate engineering geological information of open-pit mine slopes according to claim 1 is characterized in that: The dynamic update of the open pit mine slope geological model in S4.8 is based on real-time data input of the slope layered model database.

Images